DOX Solutions 02 DDP 2 Manual
Solutions%20Manual(%EC%8B%A4%EA%B3%84%20%EC%97%B0%EC%8A%B5%EB%AC%B8%EC%A0%9C%20%EC%86%94%EB%A3%A8%EC%85%98)
User Manual: DDP-2
Open the PDF directly: View PDF 
.
Page Count: 488
| Download | |
| Open PDF In Browser | View PDF |
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 2
Simple Comparative Experiments
Solutions
2-1 The breaking strength of a fiber is required to be at least 150 psi. Past experience has indicated that
the standard deviation of breaking strength is V = 3 psi. A random sample of four specimens is tested. The
results are y1=145, y2=153, y3=150 and y4=147.
(a) State the hypotheses that you think should be tested in this experiment.
H0: P = 150
H1: P > 150
(b) Test these hypotheses using D = 0.05. What are your conclusions?
n = 4, V = 3, y = 1/4 (145 + 153 + 150 + 147) = 148.75
zo
y Po
V
n
148.75 150
3
4
1.25
3
2
0.8333
Since z0.05 = 1.645, do not reject.
(c) Find the P-value for the test in part (b).
From the z-table: P # 1 >0.7967 2 3 0.7995 0.7967 @ 0.2014
(d) Construct a 95 percent confidence interval on the mean breaking strength.
The 95% confidence interval is
V
V
d P d y zD 2
n
n
148.75 1.96 3 2 d P d 148.75 1.96 3 2
y zD 2
145. 81 d P d 151. 69
2-2 The viscosity of a liquid detergent is supposed to average 800 centistokes at 25qC. A random
sample of 16 batches of detergent is collected, and the average viscosity is 812. Suppose we know that the
standard deviation of viscosity is V = 25 centistokes.
(a) State the hypotheses that should be tested.
H0: P = 800
H1: P z 800
(b) Test these hypotheses using D = 0.05. What are your conclusions?
2-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
zo
y Po
V
n
812 800
25
16
12
25
4
Since zD/2 = z0.025 = 1.96, do not reject.
1.92
(c) What is the P-value for the test?
P
2(0.0274)
0.0549
(d) Find a 95 percent confidence interval on the mean.
The 95% confidence interval is
y zD 2
V
V
d P d y zD
2
n
n
812 1.96 25 4 d P d 812 1.96 25 4
812 12.25 d P d 812 12.25
799.75 d P d 824.25
2-3 The diameters of steel shafts produced by a certain manufacturing process should have a mean
diameter of 0.255 inches. The diameter is known to have a standard deviation of V = 0.0001 inch. A
random sample of 10 shafts has an average diameter of 0.2545 inches.
(a) Set up the appropriate hypotheses on the mean P.
H0: P = 0.255
H1: P z 0.255
(b) Test these hypotheses using D = 0.05. What are your conclusions?
n = 10, V = 0.0001, y = 0.2545
zo
y Po
V
n
0.2545 0.255
0.0001
10
15.81
Since z0.025 = 1.96, reject H0.
(c) Find the P-value for this test. P=2.6547x10-56
(d) Construct a 95 percent confidence interval on the mean shaft diameter.
The 95% confidence interval is
y zD 2
V
V
d P d y zD
2
n
n
§ 0.0001 ·
§ 0.0001 ·
0.2545 1.96 ¨
¸ d P d 0.2545 1.96 ¨
¸
© 10 ¹
© 10 ¹
0. 254438 d P d 0. 254562
2-4 A normally distributed random variable has an unknown mean P and a known variance V2 = 9. Find
the sample size required to construct a 95 percent confidence interval on the mean, that has total width of
1.0.
2-2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Since y a N(P,9), a 95% two-sided confidence interval on P is
V
y zD 2
d P d y zD 2
n
3
y (196
. )
n
V
n
d P d y (196
. )
3
n
If the total interval is to have width 1.0, then the half-interval is 0.5. Since z
1.96 3
n
n
n
11.76
= z0.025 = 1.96,
0.5
1.96 3 0.5
2
/2
11.76
138.30 # 139
2-5 The shelf life of a carbonated beverage is of interest. Ten bottles are randomly selected and tested,
and the following results are obtained:
Days
108
124
124
106
115
138
163
159
134
139
(a) We would like to demonstrate that the mean shelf life exceeds 120 days. Set up appropriate
hypotheses for investigating this claim.
H0: P = 120
H1: P > 120
(b) Test these hypotheses using D = 0.01. What are your conclusions?
y = 131
s2 = [ (108 - 131)2 + (124 - 131)2 + (124 - 131)2 + (106 - 131)2 + (115 - 131)2 + (138 - 131)2
+ (163 - 131)2 + (159 - 131)2 + (134 - 131)2 + ( 139 - 131)2 ] / (10 - 1)
s2 = 3438 / 9 = 382
s
382 19. 54
to
y Po
s
n
131 120
19. 54
10
1. 78
since t0.01,9 = 2.821; do not reject H0
Minitab Output
T-Test of the Mean
Test of mu = 120.00 vs mu > 120.00
Variable
Shelf Life
N
10
Mean
131.00
StDev
19.54
SE Mean
6.18
T
1.78
T Confidence Intervals
2-3
P
0.054
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Variable
Shelf Life
N
10
Mean
131.00
StDev
19.54
SE Mean
6.18
(
99.0 % CI
110.91, 151.09)
(c) Find the P-value for the test in part (b). P=0.054
(d) Construct a 99 percent confidence interval on the mean shelf life.
s
s
d P d y tD , n 1
The 95% confidence interval is y tD , n 1
2
2
n
n
§ 1954 ·
§ 1954 ·
131 3.250 ¨
¸ d P d 131 3.250 ¨
¸
10
©
¹
© 10 ¹
110.91 d P d 151.09
2-6 Consider the shelf life data in Problem 2-5. Can shelf life be described or modeled adequately by a
normal distribution? What effect would violation of this assumption have on the test procedure you used in
solving Problem 2-5?
A normal probability plot, obtained from Minitab, is shown. There is no reason to doubt the adequacy of
the normality assumption. If shelf life is not normally distributed, then the impact of this on the t-test in
problem 2-5 is not too serious unless the departure from normality is severe.
Normal Probability Plot for Shelf Life
ML Estimates
99
ML Estimates
95
Mean
131
StDev
18.5418
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
1.292
20
10
5
1
86
96
106
116
126
136
146
156
166
176
Data
2-7 The time to repair an electronic instrument is a normally distributed random variable measured in
hours. The repair time for 16 such instruments chosen at random are as follows:
Hours
159
224
222
149
280
379
362
260
101
179
168
485
2-4
212
264
250
170
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) You wish to know if the mean repair time exceeds 225 hours. Set up appropriate hypotheses for
investigating this issue.
H0: P = 225 H1: P > 225
(b) Test the hypotheses you formulated in part (a). What are your conclusions? Use D = 0.05.
y = 247.50
s2 =146202 / (16 - 1) = 9746.80
s
to
9746. 8
y Po
s
n
98. 73
241.50 225
98.73
16
0.67
since t0.05,15 = 1.753; do not reject H0
Minitab Output
T-Test of the Mean
Test of mu = 225.0 vs mu > 225.0
Variable
Hours
N
16
Mean
241.5
StDev
98.7
Mean
241.5
StDev
98.7
SE Mean
24.7
T
0.67
P
0.26
T Confidence Intervals
Variable
Hours
N
16
SE Mean
24.7
95.0 % CI
188.9,
294.1)
(
(c) Find the P-value for this test. P=0.26
(d) Construct a 95 percent confidence interval on mean repair time.
The 95% confidence interval is y t D
s
2
, n 1
n
d P d y tD
s
2
,n 1
n
§ 98.73 ·
§ 98.73 ·
241.50 2.131 ¨
¸ d P d 241.50 2.131 ¨
¸
© 16 ¹
© 16 ¹
188.9 d P d 294.1
2-8 Reconsider the repair time data in Problem 2-7. Can repair time, in your opinion, be adequately
modeled by a normal distribution?
The normal probability plot below does not reveal any serious problem with the normality assumption.
2-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot for Hours
ML Estimates
99
ML Estimates
95
Mean
241.5
StDev
95.5909
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
1.185
20
10
5
1
50
150
250
350
450
Data
2-9 Two machines are used for filling plastic bottles with a net volume of 16.0 ounces. The filling
processes can be assumed to be normal, with standard deviation of V1 = 0.015 and V2 = 0.018. The quality
engineering department suspects that both machines fill to the same net volume, whether or not this volume
is 16.0 ounces. An experiment is performed by taking a random sample from the output of each machine.
Machine 1
16.03
16.01
16.04
15.96
16.05
15.98
16.05
16.02
16.02
15.99
Machine 2
16.02
16.03
15.97
16.04
15.96
16.02
16.01
16.01
15.99
16.00
(a) State the hypotheses that should be tested in this experiment.
H0: P1 = P2
H1: P1 z P2
(b) Test these hypotheses using D=0.05. What are your conclusions?
y1 16. 015
V1 0. 015
n1 10
zo
y2 16. 005
V 2 0. 018
n2 10
y1 y2
16. 015 16. 018
V12
0. 0152 0. 0182
10
10
n1
V 22
n2
1. 35
z0.025 = 1.96; do not reject
(c) What is the P-value for the test? P=0.1770
(d) Find a 95 percent confidence interval on the difference in the mean fill volume for the two machines.
2-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The 95% confidence interval is
y1 y 2 z D 2
(16.015 16.005) (19.6)
V 12 V 22
V 12 V 22
d P 1 P 2 d y1 y 2 z D 2
n1
n2
n1
n2
0.0152 0.0182
0.0152 0.0182
d P1 P 2 d (16.015 16.005) (19.6)
10
10
10
10
0.0045 d P1 P 2 d 0.0245
2-10 Two types of plastic are suitable for use by an electronic calculator manufacturer. The breaking
strength of this plastic is important. It is known that V1 = V2 = 1.0 psi. From random samples of n1 = 10
and n2 = 12 we obtain y 1 = 162.5 and y 2 = 155.0. The company will not adopt plastic 1 unless its
breaking strength exceeds that of plastic 2 by at least 10 psi. Based on the sample information, should they
use plastic 1? In answering this questions, set up and test appropriate hypotheses using D = 0.01.
Construct a 99 percent confidence interval on the true mean difference in breaking strength.
H0: P1 - P2 =10
H1: P1 - P2 >10
y1 162.5
V1 1
y2
V2
n2 10
n1 10
zo
y1 y 2 10
V12
n1
155.0
1
162. 5 155. 0 10
V 22
12 12
10 12
n2
5. 85
z0.01 = 2.225; do not reject
The 99 percent confidence interval is
y1 y 2 z D 2
(162.5 155.0) ( 2.575)
V 12 V 22
V 12 V 22
d P 1 P 2 d y1 y 2 z D
2
n1
n2
n1
n2
12 12
12 12
d P1 P 2 d (162.5 155.0) ( 2.575)
10 12
10 12
6.40 d P 1 P 2 d 8.60
2-11 The following are the burning times of chemical flares of two different formulations. The design
engineers are interested in both the means and variance of the burning times.
Type 1
65
81
57
66
82
Type 2
64
71
83
59
65
82
67
59
75
70
56
69
74
82
79
(a) Test the hypotheses that the two variances are equal. Use D = 0.05.
2-7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
H 0 : V 12
V 22
H1: V 12
V 22
S12
S 22
z
F0
F0.025,9 ,9
4.03
F0.975,9 ,9
S1
S2
8582
.
87.73
1
F0.025,9 ,9
9.264
9.367
0.98
1
4.03
0.248 Do not reject.
(b) Using the results of (a), test the hypotheses that the mean burning times are equal. Use D = 0.05.
What is the P-value for this test?
S 2p
Sp
(n1 1) S12 (n 2 1) S 22
n1 n2 2
9.32
y1 y 2
t0
Sp
1
1
n1 n2
t 0.025,18
156195
.
18
70.4 70.2
1 1
9.32
10 10
86.775
0.048
2.101 Do not reject.
From the computer output, t=0.05; do not reject. Also from the computer output P=0.96
Minitab Output
Two Sample T-Test and Confidence Interval
Two sample T for Type 1 vs Type 2
Type 1
Type 2
N
10
10
Mean
70.40
70.20
StDev
9.26
9.37
SE Mean
2.9
3.0
95% CI for mu Type 1 - mu Type 2: ( -8.6, 9.0)
T-Test mu Type 1 = mu Type 2 (vs not =): T = 0.05
Both use Pooled StDev = 9.32
P = 0.96
DF = 18
(c) Discuss the role of the normality assumption in this problem. Check the assumption of normality for
both types of flares.
The assumption of normality is required in the theoretical development of the t-test. However, moderate
departure from normality has little impact on the performance of the t-test. The normality assumption is
more important for the test on the equality of the two variances. An indication of nonnormality would be
of concern here. The normal probability plots shown below indicate that burning time for both
formulations follow the normal distribution.
2-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot for Type 1
ML Estimates
99
ML Estimates
95
Mean
70.4
StDev
8.78863
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
1.387
20
10
5
1
50
60
70
80
90
Data
Normal Probability Plot for Type 2
ML Estimates
99
ML Estimates
95
Mean
70.2
StDev
8.88594
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
1.227
20
10
5
1
50
60
70
80
90
Data
2-12 An article in Solid State Technology, "Orthogonal Design of Process Optimization and Its
Application to Plasma Etching" by G.Z. Yin and D.W. Jillie (May, 1987) describes an experiment to
determine the effect of C2F6 flow rate on the uniformity of the etch on a silicon wafer used in integrated
circuit manufacturing. Data for two flow rates are as follows:
C2F6
(SCCM)
125
200
1
2.7
4.6
2
4.6
3.4
Uniformity Observation
3
4
5
2.6
3.0
3.2
2.9
3.5
4.1
(a) Does the C2F6 flow rate affect average etch uniformity? Use D = 0.05.
No, C2F6 flow rate does not affect average etch uniformity.
Minitab Output
Two Sample T-Test and Confidence Interval
2-9
6
3.8
5.1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Two sample T for Uniformity
Flow Rat
125
200
N
6
6
Mean
3.317
3.933
StDev
0.760
0.821
SE Mean
0.31
0.34
95% CI for mu (125) - mu (200): ( -1.63, 0.40)
T-Test mu (125) = mu (200) (vs not =): T = -1.35
Both use Pooled StDev = 0.791
P = 0.21
DF = 10
(b) What is the P-value for the test in part (a)? From the computer printout, P=0.21
(c) Does the C2F6 flow rate affect the wafer-to-wafer variability in etch uniformity? Use D = 0.05.
H0 : V12
V 22
H1: V 12 z V 22
F0.05,5,5 5.05
F0
0.5776
0.6724
0.86
Do not reject; C2F6 flow rate does not affect wafer-to-wafer variability.
(d) Draw box plots to assist in the interpretation of the data from this experiment.
The box plots shown below indicate that there is little difference in uniformity at the two gas flow rates.
Any observed difference is not statistically significant. See the t-test in part (a).
Uniformity
5
4
3
125
200
Flow Rate
2-13 A new filtering device is installed in a chemical unit. Before its installation, a random sample
2
yielded the following information about the percentage of impurity: y 1 = 12.5, S1 =101.17, and n1 = 8.
2
After installation, a random sample yielded y 2 = 10.2, S2 = 94.73, n2 = 9.
(a) Can you concluded that the two variances are equal? Use D = 0.05.
2-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
H 0 : V 12
V 22
H1 : V 12 z V 22
F0.025 ,7 ,8 4.53
S12 101.17
94.73
S22
Do Not Reject. Assume that the variances are equal.
F0
1.07
(b) Has the filtering device reduced the percentage of impurity significantly? Use D = 0.05.
H 0 : P1 P 2
H1 : P1 z P 2
S p2
( n1 1 )S12 ( n2 1 )S 22
n1 n2 2
Sp
9.89
y1 y 2
t0
Sp
t 0.05 ,15
( 8 1 )( 101.17 ) ( 9 1 )( 94.73 )
892
12.5 10.2
1
1
n1 n2
1 1
9.89
8 9
97.74
0.479
1.753
Do not reject. There is no evidence to indicate that the new filtering device has affected the mean
2-14 Twenty observations on etch uniformity on silicon wafers are taken during a qualification
experiment for a plasma etcher. The data are as follows:
5.34
6.00
5.97
5.25
6.65
7.55
7.35
6.35
4.76
5.54
5.44
4.61
5.98
5.62
4.39
6.00
7.25
6.21
4.98
5.32
(a) Construct a 95 percent confidence interval estimate of V2.
n 1 S 2
n 1 S 2
2
d
d
V
FD2 ,n 1
F (12 D ),n 1
2
2
20 1 0.88907
2
32.852
0.457 d V 2 d 1.686
dV2 d
20 1 0.88907
2
8.907
(b) Test the hypothesis that V2 = 1.0. Use D = 0.05. What are your conclusions?
H0 : V 2
1
2
H1 : V z 1
F 02
F 02.025,19
SS
V 02
32.852
2-11
15019
.
F 02.975,19
8.907
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Do not reject. There is no evidence to indicate that V 12 z 1
(c) Discuss the normality assumption and its role in this problem.
The normality assumption is much more important when analyzing variances then when analyzing means.
A moderate departure from normality could cause problems with both statistical tests and confidence
intervals. Specifically, it will cause the reported significance levels to be incorrect.
(d) Check normality by constructing a normal probability plot. What are your conclusions?
The normal probability plot indicates that there is not any serious problem with the normality assumption.
Normal Probability Plot for Uniformity
ML Estimates
99
ML Estimates
95
Mean
5.828
StDev
0.866560
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
0.835
20
10
5
1
3.8
4.8
5.8
6.8
7.8
Data
2-15 The diameter of a ball bearing was measured by 12 inspectors, each using two different kinds of
calipers. The results were:
Inspector
1
2
3
4
5
6
7
8
9
10
11
12
Caliper 1
0.265
0.265
0.266
0.267
0.267
0.265
0.267
0.267
0.265
0.268
0.268
0.265
Caliper 2
0.264
0.265
0.264
0.266
0.267
0.268
0.264
0.265
0.265
0.267
0.268
0.269
Difference
.001
.000
.002
.001
.000
-.003
.003
.002
.000
.001
.000
-.004
¦ 0.003
Difference^2
.000001
0
.000004
.000001
0
.000009
.000009
.000004
0
.000001
0
.000016
¦
0.000045
(a) Is there a significant difference between the means of the population of measurements represented by
the two samples? Use D = 0.05.
2-12
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
H 0 : P1 P 2
H 0 : Pd 0
or equivalently
H1 : P1 z P 2
H1 : P d z 0
Minitab Output
Paired T-Test and Confidence Interval
Paired T for Caliper 1 - Caliper 2
N
12
12
12
Caliper
Caliper
Difference
Mean
0.266250
0.266000
0.000250
StDev
0.001215
0.001758
0.002006
SE Mean
0.000351
0.000508
0.000579
95% CI for mean difference: (-0.001024, 0.001524)
T-Test of mean difference = 0 (vs not = 0): T-Value = 0.43
P-Value = 0.674
(b) Find the P-value for the test in part (a). P=0.674
(c) Construct a 95 percent confidence interval on the difference in the mean diameter measurements for
the two types of calipers.
d t
Sd
2 ,n 1
n
d PD
P1 P 2 d d t
0.002
Sd
2 ,n 1
n
0.002
d P d d 0.00025 2.201
12
12
0.00102 d P d d 0.00152
0.00025 2.201
2-16 An article in the Journal of Strain Analysis (vol.18, no. 2, 1983) compares several procedures for
predicting the shear strength for steel plate girders. Data for nine girders in the form of the ratio of
predicted to observed load for two of these procedures, the Karlsruhe and Lehigh methods, are as follows:
Girder
S1/1
S2/1
S3/1
S4/1
S5/1
S2/1
S2/2
S2/3
S2/4
Karlsruhe Method
1.186
1.151
1.322
1.339
1.200
1.402
1.365
1.537
1.559
Lehigh Method
1.061
0.992
1.063
1.062
1.065
1.178
1.037
1.086
1.052
Sum =
Average =
Difference
0.125
0.159
0.259
0.277
0.135
0.224
0.328
0.451
0.507
2.465
0.274
Difference^2
0.015625
0.025281
0.067081
0.076729
0.018225
0.050176
0.107584
0.203401
0.257049
0.821151
(a) Is there any evidence to support a claim that there is a difference in mean performance between the two
methods? Use D = 0.05.
H 0 : Pd 0
H 0 : P1 P 2
or equivalently
H1 : P1 z P 2
H1 : P d z 0
d
1 n
¦d
ni1 i
1
2.465
9
2-13
0.274
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ª n 2 1 § n ·2 º
« ¦ di ¨ ¦ di ¸ »
n© i 1 ¹ »
«i1
«
»
n 1
«
»
«¬
»¼
sd
2
1
ª
2º
« 0.821151 9 (2.465) »
«
»
9 1
«
»
¬
¼
1
2
0.135
d
0.274
6.08
Sd
0.135
n
9
2.306 , reject the null hypothesis.
t0
t D 2 ,n 1
1
t 0.025 ,9
Minitab Output
Paired T-Test and Confidence Interval
Paired T for Karlsruhe - Lehigh
N
Karlsruh
Lehigh
Difference
Mean StDev SE Mean
9 1.3401 0.1460 0.0487
9 1.0662 0.0494 0.0165
9 0.2739 0.1351 0.0450
95% CI for mean difference: (0.1700, 0.3777)
T-Test of mean difference = 0 (vs not = 0): T-Value = 6.08 P-Value = 0.000
(b) What is the P-value for the test in part (a)? P=0.0002
(c) Construct a 95 percent confidence interval for the difference in mean predicted to observed load.
d tD
Sd
,n 1
Sd
,n 1
n
0.135
0.274 2.306
d P d d 0.274 2.306
9
9
0.17023 d P d d 0.37777
2
n
0.135
d P d d d tD
2
(d) Investigate the normality assumption for both samples.
Normal Probability Plot
.999
.99
Probability
.95
.80
.50
.20
.05
.01
.001
1.15
1.25
1.35
1.45
1.55
Karlsruhe
Av erage: 1.34011
StDev : 0.146031
N: 9
Anderson-Darling Normality Test
A-Squared: 0.286
P-Value: 0.537
2-14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot
.999
.99
Probability
.95
.80
.50
.20
.05
.01
.001
1.00
1.05
1.10
1.15
Lehigh
Av erage: 1.06622
StDev : 0.0493806
N: 9
Anders on-Darling Normality Tes t
A-Squared: 0.772
P-Value: 0.028
(e) Investigate the normality assumption for the difference in ratios for the two methods.
Normal Probability Plot
.999
.99
Probability
.95
.80
.50
.20
.05
.01
.001
0.12
0.22
0.32
0.42
0.52
Difference
Av erage: 0.273889
StDev : 0.135099
N: 9
Anderson-Darling Normality Tes t
A-Squared: 0.318
P-Value: 0.464
(f) Discuss the role of the normality assumption in the paired t-test.
As in any t-test, the assumption of normality is of only moderate importance. In the paired t-test, the
assumption of normality applies to the distribution of the differences. That is, the individual sample
measurements do not have to be normally distributed, only their difference.
2-17 The deflection temperature under load for two different formulations of ABS plastic pipe is being
studied. Two samples of 12 observations each are prepared using each formulation, and the deflection
temperatures (in qF) are reported below:
212
194
211
193
Formulation 1
199
213
191
195
198
216
200
184
177
197
206
201
2-15
Formulation 2
176
185
200
197
198
188
189
203
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Construct normal probability plots for both samples. Do these plots support assumptions of normality
and equal variance for both samples?
Normal Probability Plot
.999
.99
Probability
.95
.80
.50
.20
.05
.01
.001
185
195
205
215
Form 1
Av erage: 200.5
StDev : 10.1757
N: 12
Anderson-Darling Normality Tes t
A-Squared: 0.450
P-Value: 0.227
Normal Probability Plot
.999
.99
Probability
.95
.80
.50
.20
.05
.01
.001
175
185
195
205
Form 2
Av erage: 193.083
StDev : 9.94949
N: 12
Anderson-Darling Normality Test
A-Squared: 0.443
P-Value: 0.236
(b) Do the data support the claim that the mean deflection temperature under load for formulation 1
exceeds that of formulation 2? Use D = 0.05.
Minitab Output
Two Sample T-Test and Confidence Interval
Two sample T for Form 1 vs Form 2
Form 1
Form 2
N
12
12
Mean
200.5
193.08
StDev
10.2
9.95
SE Mean
2.9
2.9
95% CI for mu Form 1 - mu Form 2: ( -1.1, 15.9)
T-Test mu Form 1 = mu Form 2 (vs >): T = 1.81 P = 0.042
Both use Pooled StDev = 10.1
(c) What is the P-value for the test in part (a)? P = 0.042
2-16
DF = 22
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2-18 Refer to the data in problem 2-17. Do the data support a claim that the mean deflection temperature
under load for formulation 1 exceeds that of formulation 2 by at least 3 qF?
Yes, formulation 1 exceeds formulation 2 by at least 3 qF.
Minitab Output
Two-Sample T-Test and CI: Form1, Form2
Two-sample T for Form1 vs Form2
N
Mean
StDev
SE Mean
Form1 12
200.5
10.2
2.9
Form2 12
193.08
9.95
2.9r
Difference = mu Form1 - mu Form2
Estimate for difference: 7.42
95% lower bound for difference: 0.36
T-Test of difference = 3 (vs >): T-Value = 1.08
Both use Pooled StDev = 10.1
P-Value = 0.147
DF = 22
2-19 In semiconductor manufacturing, wet chemical etching is often used to remove silicon from the
backs of wafers prior to metalization. The etch rate is an important characteristic of this process. Two
different etching solutionsare being evaluated. Eight randomly selected wafers have been etched in each
solution and the observed etch rates (in mils/min) are shown below:
Solution 1
9.9
10.6
9.4
10.3
10.0
9.3
10.3
9.8
Solution 2
10.2
10.6
10.0
10.2
10.7
10.4
10.5
10.3
(a) Do the data indicate that the claim that both solutions have the same mean etch rate is valid? Use D =
0.05 and assume equal variances.
See the Minitab output below.
Minitab Output
Two Sample T-Test and Confidence Interval
Two sample T for Solution 1 vs Solution 2
Solution
Solution
N
8
8
Mean
9.925
10.362
StDev
0.465
0.233
SE Mean
0.16
0.082
95% CI for mu Solution - mu Solution: ( -0.83, -0.043)
T-Test mu Solution = mu Solution (vs not =):T = -2.38 P = 0.032 DF = 14
Both use Pooled StDev = 0.368
(b) Find a 95% confidence interval on the difference in mean etch rate.
From the Minitab output, -0.83 to –0.043.
(c) Use normal probability plots to investigate the adequacy of the assumptions of normality and equal
variances.
2-17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot
.999
.99
.95
Probability
.80
.50
.20
.05
.01
.001
9.5
10.0
10.5
Solution 1
Av erage: 9.925
StDev : 0.465219
N: 8
Anders on-Darling Normality Tes t
A-Squared: 0.222
P-Value: 0.743
Normal Probability Plot
.999
.99
.95
Probability
.80
.50
.20
.05
.01
.001
10.0
10.1
10.2
10.3
10.4
10.5
10.6
10.7
Solution 2
Av erage: 10.3625
StDev : 0.232609
N: 8
Anderson-Darling Normality Test
A-Squared: 0.158
P-Value: 0.919
Both the normality and equality of variance assumptions are valid.
2-20 Two popular pain medications are being compared on the basis of the speed of absorption by the
body. Specifically, tablet 1 is claimed to be absorbed twice as fast as tablet 2. Assume that V 12 and V 22
are known. Develop a test statistic for
H0: 2P1 = P2
H1: 2P1 z P2
§
4V 2 V 2 ·
2 y1 y2 ~ N ¨ 2 P1 P 2 , 1 2 ¸ , assuming that the data is normally distributed.
n1
n2 ¹
©
2 y1 y2
, reject if zo ! zD
The test statistic is: zo
2
4 V 12 V 22
n1
n2
2-21 Suppose we are testing
H0: P1 = P2
H1: P1 z P2
2-18
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
where V 12 and V 22 are known. Our sampling resources are constrained such that n1 + n2 = N. How should
we allocate the N observations between the two populations to obtain the most powerful test?
The most powerful test is attained by the n1 and n2 that maximize zo for given y1 y 2 .
y1 y 2 , subject to n1 + n2 = N.
Thus, we chose n1 and n2 to
max z o
V 12 V 22
n1 n2
This is equivalent to min L
V12 V 22
n1 n2
V 12
V 22
, subject to n1 + n2 = N.
n1 N n1
V 12
V 22
0 , implies that n1 / n2 = V1 / V2.
2
2
n1
N n1
Thus n1 and n2 are assigned proportionally to the ratio of the standard deviations. This has
intuitive appeal, as it allocates more observations to the population with the greatest variability.
Now dL
dn1
2-22 Develop Equation 2-46 for a 100(1 - D) percent confidence interval for the variance of a normal
distribution.
SS
~ F n21 . Thus, P ® F
V2
¯
2
1 , n1
2
d
SS
dF
V2
2
2
,n1
½
¾ 1 D . Therefore,
¿
½
SS °
° SS
,
P ® 2 d V 2 d 2 ¾ 1 D
F
F
1 , n1 °
2
¯° 2 ,n1
¿
ª
º
so « SS , SS » is the 100(1 - D)% confidence interval on V2.
« F 2 ,n1 F 12 ,n1 »
2
¬ 2
¼
2-23 Develop Equation 2-50 for a 100(1 - D) percent confidence interval for the ratio V12 / V 22 , where V 12
and V 22 are the variances of two normal distributions.
S22 V 22
~ Fn2 1, n1 1
S12 V 12
½
S2 V 2
P ® F1 2 ,n2 1, n1 1 d 22 22 d F ¾ 1 D or
n
n
,
1,
1
2
1
2
S1 V 1
¯
¿
2
2
2
S
½
S
V
P ® 12 F1 2 ,n2 1,n1 1 d 12 d 12 F
¾ 1D
n
n
,
1,
1
V 2 S2 2 2 1 ¿
¯ S2
2-24 Develop an equation for finding a 100(1 - D) percent confidence interval on the difference in the
means of two normal distributions where V 12 z V 22 . Apply your equation to the portland cement
experiment data, and find a 95% confidence interval.
2-19
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
y1 y2 P1 P 2
S12 S22
n1 n2
t
2 ,X
~ tD 2 ,X
S12 S 22
d y1 y2 P1 P 2 d t
n1 n2
y1 y2 t
S12 S 22
n1 n2
2 ,X
S12 S22
d P1 P 2 d y1 y2 t
n1 n2
2 ,X
2 ,X
S12 S 22
n1 n2
2
where
§ S12 S22 ·
¨ ¸
© n1 n2 ¹
2
2
§ S12 · § S22 ·
¨ ¸ ¨ ¸
© n1 ¹ © n2 ¹
n1 1
n2 1
X
Using the data from Table 2-1
n1
y1
S12
16.764 17.343 2.110
10
16.764
0100138
.
n2
y2
10
17.343
S 22
0.0614622
0.100138 0.0614622
d P1 P 2 d
10
10
16.764 17.343 2.110
where X
§ 0.100138 0.0614622 ·
¨
¸
10
© 10
¹
2
0.100138 0.0614622
10
10
2
§ 0.100138 ·
§ 0.0614622 ·
¨
¸
¨
¸
10
10
©
¹
©
¹
10 1
10 1
2
17.024 # 17
1.426 d P1 P 2 d 0.889
This agrees with the result in Table 2-2.
2-25 Construct a data set for which the paired t-test statistic is very large, but for which the usual twosample or pooled t-test statistic is small. In general, describe how you created the data. Does this give you
any insight regarding how the paired t-test works?
A
7.1662
2.3590
19.9977
0.9077
-15.9034
-6.0722
B
8.2416
2.4555
21.1018
2.3401
-15.0013
-5.5941
2-20
delta
1.07541
0.09650
1.10412
1.43239
0.90204
0.47808
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
9.9501
-1.0944
-4.6907
-6.6929
10.6910
-0.1358
-3.3446
-5.9303
0.74085
0.95854
1.34615
0.76256
Minitab Output
Paired T-Test and Confidence Interval
Paired T for A - B
N
10
10
10
A
B
Difference
Mean
0.59
1.48
-0.890
StDev
10.06
10.11
0.398
SE Mean
3.18
3.20
0.126
95% CI for mean difference: (-1.174, -0.605)
T-Test of mean difference = 0 (vs not = 0): T-Value = -7.07
P-Value = 0.000
Two Sample T-Test and Confidence Interval
Two sample T for A vs B
A
B
N
10
10
Mean
0.6
1.5
StDev
10.1
10.1
SE Mean
3.2
3.2
95% CI for mu A - mu B: ( -10.4, 8.6)
T-Test mu A = mu B (vs not =): T = -0.20
Both use Pooled StDev = 10.1
P = 0.85
DF = 18
These two sets of data were created by making the observation for A and B moderately different within
each pair (or block), but making the observations between pairs very different. The fact that the difference
between pairs is large makes the pooled estimate of the standard deviation large and the two-sample t-test
statistic small. Therefore the fairly small difference between the means of the two treatments that is present
when they are applied to the same experimental unit cannot be detected. Generally, if the blocks are very
different, then this will occur. Blocking eliminates the variabiliy associated with the nuisance variable that
they represent.
2-21
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 3
Experiments with a Single Factor: The Analysis of Variance
Solutions
3-1 The tensile strength of portland cement is being studied. Four different mixing techniques can be
used economically. The following data have been collected:
Mixing Technique
1
2
3
4
3129
3200
2800
2600
Tensile Strength (lb/in2)
3000
2865
3300
2975
2900
2985
2700
2600
2890
3150
3050
2765
(a) Test the hypothesis that mixing techniques affect the strength of the cement. Use D = 0.05.
Design Expert Output
Response:
Tensile Strengthin lb/in^2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
4.897E+005
3
1.632E+005
A
4.897E+005
3
1.632E+005
Residual
1.539E+005
12
12825.69
Lack of Fit
0.000
0
Pure Error
1.539E+005
12
12825.69
Cor Total
6.436E+005
15
F
Value
12.73
12.73
Prob > F
0.0005
0.0005
significant
The Model F-value of 12.73 implies the model is significant. There is only
a 0.05% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
2971.00
56.63
2-2
3156.25
56.63
3-3
2933.75
56.63
4-4
2666.25
56.63
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-185.25
37.25
304.75
222.50
490.00
267.50
DF
1
1
1
1
1
1
Standard
Error
80.08
80.08
80.08
80.08
80.08
80.08
t for H0
Coeff=0
-2.31
0.47
3.81
2.78
6.12
3.34
Prob > |t|
0.0392
0.6501
0.0025
0.0167
< 0.0001
0.0059
The F-value is 12.73 with a corresponding P-value of .0005. Mixing technique has an effect.
(b) Construct a graphical display as described in Section 3-5.3 to compare the mean tensile strengths for
the four mixing techniques. What are your conclusions?
S yi .
MS E
n
12825.7
4
3-1
56.625
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a le d t D is t r ib u t io n
(4 )
(3 )
2700
2800
2900
(1 )
3000
(2 )
3100
T e n s ile S t r e n g th
Based on examination of the plot, we would conclude that P1 and P3 are the same; that P 4 differs from
P1 and P3 , that P 2 differs from P1 and P3 , and that P 2 and P 4 are different.
(c) Use the Fisher LSD method with D=0.05 to make comparisons between pairs of means.
LSD
tD
2
,N a
2MSE
n
2( 12825.7 )
4
LSD
t 0.025 ,16 4
LSD
2.179 6412.85
174.495
Treatment 2 vs. Treatment 4 = 3156.250 - 2666.250 = 490.000 > 174.495
Treatment 2 vs. Treatment 3 = 3156.250 - 2933.750 = 222.500 > 174.495
Treatment 2 vs. Treatment 1 = 3156.250 - 2971.000 = 185.250 > 174.495
Treatment 1 vs. Treatment 4 = 2971.000 - 2666.250 = 304.750 > 174.495
Treatment 1 vs. Treatment 3 = 2971.000 - 2933.750 = 37.250 < 174.495
Treatment 3 vs. Treatment 4 = 2933.750 - 2666.250 = 267.500 > 174.495
The Fisher LSD method is also presented in the Design-Expert computer output above. The results agree
with graphical method for this experiment.
(d) Construct a normal probability plot of the residuals. What conclusion would you draw about the
validity of the normality assumption?
There is nothing unusual about the normal probability plot of residuals.
3-2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t o f re sidua ls
99
N orm al % probability
95
90
80
70
50
30
20
10
5
1
-1 8 1 .25
-9 6 .4 37 5
-1 1 .6 25
7 3 .1 87 5
158
R es idu al
(e) Plot the residuals versus the predicted tensile strength. Comment on the plot.
There is nothing unusual about this plot.
Residuals vs. Predicted
158
Residuals
73.1875
-11.625
2
-96.4375
-181.25
2666.25
2788.75
2911.25
3033.75
3156.25
Predicted
(f) Prepare a scatter plot of the results to aid the interpretation of the results of this experiment.
Design-Expert automatically generates the scatter plot. The plot below also shows the sample average for
each treatment and the 95 percent confidence interval on the treatment mean.
3-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One Factor Plot
3300
Tensile Strength
3119.75
2939.51
2759.26
2
2579.01
1
2
3
4
Technique
3-2 Rework part (b) of Problem 3-1 using the Duncan's multiple range test. Does this make any
difference in your conclusions?
S yi .
MS E
n
2 ,12 S yi .
12825.7
56.625
4
3.08 56.625 174.406
R2
r0.05
R3
r0.05 3,12 S yi .
3.23 56.625
182.900
R4
r0.05 4 ,12 S yi .
3.33 56.625
188.562
Treatment 2 vs. Treatment 4 = 3156.250 - 2666.250 = 490.000 > 188.562 (R4)
Treatment 2 vs. Treatment 3 = 3156.250 - 2933.750 = 222.500 > 182.900 (R3)
Treatment 2 vs. Treatment 1 = 3156.250 - 2971.000 = 185.250 > 174.406 (R2)
Treatment 1 vs. Treatment 4 = 2971.000 - 2666.250 = 304.750 > 182.900 (R3)
Treatment 1 vs. Treatment 3 = 2971.000 - 2933.750 = 37.250 < 174.406 (R2)
Treatment 3 vs. Treatment 4 = 2933.750 - 2666.250 = 267.500 > 174.406 (R2)
Treatment 3 and Treatment 1 are not different. All other pairs of means differ. This is the same result
obtained from the Fisher LSD method and the graphical method.
(b) Rework part (b) of Problem 3-1 using Tukey’s test with D = 0.05. Do you get the same conclusions
from Tukey’s test that you did from the graphical procedure and/or Duncan’s multiple range test?
Minitab Output
Tukey's pairwise comparisons
Family error rate = 0.0500
Individual error rate = 0.0117
Critical value = 4.20
Intervals for (column level mean) - (row level mean)
1
2
2
3
-423
3-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
53
3
-201
275
-15
460
4
67
543
252
728
30
505
No, the conclusions are not the same. The mean of Treatment 4 is different than the means of Treatments
1, 2, and 3. However, the mean of Treatment 2 is not different from the means of Treatments 1 and 3
according to the Tukey method, they were found to be different using the graphical method and Duncan’s
multiple range test.
(c) Explain the difference between the Tukey and Duncan procedures.
A single critical value is used for comparison with the Tukey procedure where a – 1 critical values are
used with the Duncan procedure. Tukey’s test has a type I error rate of D for all pairwise comparisons
where Duncan’s test type I error rate varies based on the steps between the means. Tukey’s test is more
conservative and has less power than Duncan’s test.
3-3 Reconsider the experiment in Problem 3-1. Find a 95 percent confidence interval on the mean
tensile strength of the portland cement produced by each of the four mixing techniques. Also find a 95
percent confidence interval on the difference in means for techniques 1 and 3. Does this aid in
interpreting the results of the experiment?
MSE
MSE
yi . tD ,N a
d Pi d yi . tD ,N a
n
n
2
2
Treatment 1: 2971 r 2.179
1282837
4
2971 r 123.387
2847.613 d P1 d 3094.387
Treatment 2: 3156.25r123.387
3032.863 d P 2 d 3279.637
Treatment 3: 2933.75r123.387
2810.363 d P3 d 3057.137
Treatment 4: 2666.25r123.387
2542.863 d P4 d 2789.637
Treatment 1 - Treatment 3: yi . y j . tD
2
,N a
2MS E
2 MSE
d Pi P j d yi . y j . tD ,N a
n
n
2
2 12825.7
4
137.245 d P1 P3 d 211.745
2971.00 2933.75 r 2.179
3-4 An experiment was run to determine whether four specific firing temperatures affect the density of
a certain type of brick. The experiment led to the following data:
Temperature
100
125
150
175
21.8
21.7
21.9
21.9
21.9
21.4
21.8
21.7
Density
21.7
21.5
21.8
21.8
3-5
21.6
21.4
21.6
21.4
21.7
21.5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Does the firing temperature affect the density of the bricks? Use D = 0.05.
No, firing temperature does not affect the density of the bricks. Refer to the Design-Expert output below.
Design Expert Output
Response:
Density
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.16
3
0.052
A
0.16
3
0.052
Residual
0.36
14
0.026
Lack of Fit
0.000
0
Pure Error
0.36
14
0.026
Cor Total
0.52
17
F
Value
2.02
2.02
Prob > F
0.1569
0.1569
not significant
The "Model F-value" of 2.02 implies the model is not significant relative to the noise. There is a
15.69 % chance that a "Model F-value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-100
21.74
0.072
2-125
21.50
0.080
3-150
21.72
0.072
4-175
21.70
0.080
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
0.24
0.020
0.040
-0.22
-0.20
0.020
DF
1
1
1
1
1
1
Standard
Error
0.11
0.10
0.11
0.11
0.11
0.11
t for H0
Coeff=0
2.23
0.20
0.37
-2.05
-1.76
0.19
Prob > |t|
0.0425
0.8465
0.7156
0.0601
0.0996
0.8552
(b) Is it appropriate to compare the means using Duncan’s multiple range test in this experiment?
The analysis of variance tells us that there is no difference in the treatments. There is no need to proceed
with Duncan’s multiple range test to decide which mean is difference.
(c) Analyze the residuals from this experiment. Are the analysis of variance assumptions satisfied?
There is nothing unusual about the residual plots.
3-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
0.2
99
2
0.075
90
80
70
Res iduals
Normal % probability
95
50
30
20
10
5
2
-0.05
2
-0.175
1
-0.3
-0.3
-0.175
-0.05
0.075
0.2
21.50
21.56
Res idual
21.62
21.68
21.74
Predicted
(d) Construct a graphical display of the treatments as described in Section 3-5.3. Does this graph
adequately summarize the results of the analysis of variance in part (b). Yes.
S c a le d t D i s tr ib u ti o n
(1 2 5 )
2 1 .2
2 1 .3
(1 7 5 ,1 5 0 ,1 0 0 )
2 1 .4
2 1 .5
2 1 .6
2 1 .7
2 1 .8
M e a n D e n s it y
3-5 Rework Part (d) of Problem 3-4 using the Fisher LSD method. What conclusions can you draw?
Explain carefully how you modified the procedure to account for unequal sample sizes.
When sample sizes are unequal, the appropriate formula for the LSD is
LSD
Treatment 1
Treatment 1
Treatment 1
Treatment 3
Treatment 4
Treatment 3
tD
2
,N a
vs. Treatment 2
vs. Treatment 3
vs. Treatment 4
vs. Treatment 2
vs. Treatment 2
vs. Treatment 4
§ 1
1 ·¸
MS E ¨
¨ ni n j ¸
¹
©
= 21.74 – 21.50 = 0.24 > 0.2320
= 21.74 – 21.72 = 0.02 < 0.2187
= 21.74 – 21.70 = 0.04 < 0.2320
= 21.72 – 21.50 = 0.22 < 0.2320
= 21.70 – 21.50 = 0.20 < 0.2446
= 21.72 – 21.70 = 0.02 < 0.2320
3-7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Treatment 1, temperature of 100, is different than Treatment 2, temperature of 125. All other pairwise
comparisons do not identify differences. Notice something very interesting has happened here. The
analysis of variance indicated that there were no differences between treatment means, yet the LSD
procedure found a difference; in fact, the Design-Expert output indicates that the P-value if slightly less
that 0.05. This illustrates a danger of using multiple comparison procedures without relying on the results
from the analysis of variance. Because we could not reject the hypothesis of equal means using the
analysis of variance, we should never have performed the Fisher LSD (or any other multiple comparison
procedure, for that matter). If you ignore the analysis of variance results and run multiple comparisons,
you will likely make type I errors.
The LSD calculations utilized Equation 3-32, which accommodates different sample sizes. Equation 3-32
simplifies to Equation 3-33 for a balanced design experiment.
3-6 A manufacturer of television sets is interested in the effect of tube conductivity of four different
types of coating for color picture tubes. The following conductivity data are obtained:
Coating Type
1
2
3
4
Conductivity
141
150
149
137
136
132
127
132
143
152
134
129
146
143
127
129
(a) Is there a difference in conductivity due to coating type? Use D = 0.05.
Yes, there is a difference in means. Refer to the Design-Expert output below..
Design Expert Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
844.69
3
281.56
A
844.69
3
281.56
Residual
236.25
12
19.69
Lack of Fit
0.000
0
Pure Error
236.25
12
19.69
Cor Total
1080.94
15
F
Value
14.30
14.30
Prob > F
0.0003
0.0003
The Model F-value of 14.30 implies the model is significant. There is only
a 0.03% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
145.00
2.22
2-2
145.25
2.22
3-3
132.25
2.22
4-4
129.25
2.22
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-0.25
12.75
15.75
13.00
16.00
3.00
DF
1
1
1
1
1
1
Standard
Error
3.14
3.14
3.14
3.14
3.14
3.14
t for H0
Coeff=0
-0.080
4.06
5.02
4.14
5.10
0.96
(b) Estimate the overall mean and the treatment effects.
3-8
Prob > |t|
0.9378
0.0016
0.0003
0.0014
0.0003
0.3578
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pˆ 2207 / 16
Wˆ 1 y1. y ..
Wˆ 2 y 2. y ..
Wˆ 3 y 3. y ..
Wˆ 4 y 4. y ..
137.9375
145.00 137.9375 7.0625
145.25 137.9375 7.3125
132.25 137.9375 5.6875
129.25 137.9375 8.6875
(c) Compute a 95 percent interval estimate of the mean of coating type 4. Compute a 99 percent interval
estimate of the mean difference between coating types 1 and 4.
19.69
4
124.4155 d P 4 d 134.0845
Treatment 4: 129.25 r 2.179
Treatment 1 - Treatment 4: 145 129.25 r 3.055
2 19.69
4
6.164 d P1 P 4 d 25.336
(d) Test all pairs of means using the Fisher LSD method with D=0.05.
Refer to the Design-Expert output above. The Fisher LSD procedure is automatically included in the
output.
The means of Coating Type 2 and Coating Type 1 are not different. The means of Coating Type 3 and
Coating Type 4 are not different. However, Coating Types 1 and 2 produce higher mean conductivity that
does Coating Types 3 and 4.
(e) Use the graphical method discussed in Section 3-5.3 to compare the means. Which coating produces
the highest conductivity?
S yi .
MS E
n
16.96
4
2.219 Coating types 1 and 2 produce the highest conductivity.
S c a le d t D is t r ib u t io n
(4 )
(3 )
130
(1 )(2 )
135
140
C o n d u c t iv it y
3-9
145
150
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(f) Assuming that coating type 4 is currently in use, what are your recommendations to the
manufacturer? We wish to minimize conductivity.
Since coatings 3 and 4 do not differ, and as they both produce the lowest mean values of conductivity, use
either coating 3 or 4. As type 4 is currently being used, there is probably no need to change.
3-7 Reconsider the experiment in Problem 3-6. Analyze the residuals and draw conclusions about
model adequacy.
There is nothing unusual in the normal probability plot. A funnel shape is seen in the plot of residuals
versus predicted conductivity indicating a possible non-constant variance.
Normal plot of residuals
Residuals vs. Predicted
6.75
95
90
3
80
70
Res iduals
Normal % probability
99
50
30
20
10
-0.75
2
-4.5
5
1
-8.25
-8.25
-4.5
-0.75
3
6.75
129.25
Res idual
133.25
137.25
141.25
145.25
Predicted
Residuals vs. Coating Type
6.75
Res iduals
3
2
-0.75
-4.5
-8.25
1
2
3
4
Coating Type
3-8 An article in the ACI Materials Journal (Vol. 84, 1987. pp. 213-216) describes several experiments
investigating the rodding of concrete to remove entrapped air. A 3” x 6” cylinder was used, and the
3-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
number of times this rod was used is the design variable. The resulting compressive strength of the
concrete specimen is the response. The data are shown in the following table.
Rodding Level
10
15
20
25
Compressive Strength
1530
1530
1610
1650
1560
1730
1500
1490
1440
1500
1530
1510
(a) Is there any difference in compressive strength due to the rodding level? Use D = 0.05.
There are no differences.
Design Expert Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
28633.33
3
9544.44
A
28633.33
3
9544.44
Residual
40933.33
8
5116.67
Lack of Fit
0.000
0
Pure Error
40933.33
8
5116.67
Cor Total
69566.67
11
F
Value
1.87
1.87
Prob > F
0.2138
0.2138
not significant
The "Model F-value" of 1.87 implies the model is not significant relative to the noise. There is a
21.38 % chance that a "Model F-value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated Standard
Mean
Error
1-10
1500.00 41.30
2-15
1586.67 41.30
3-20
1606.67 41.30
4-25
1500.00 41.30
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-86.67
-106.67
0.000
-20.00
86.67
106.67
DF
1
1
1
1
1
1
Standard
Error
58.40
58.40
58.40
58.40
58.40
58.40
t for H0
Coeff=0
-1.48
-1.83
0.000
-0.34
1.48
1.83
Prob > |t|
0.1761
0.1052
1.0000
0.7408
0.1761
0.1052
(b) Find the P-value for the F statistic in part (a). From computer output, P=0.2138.
(c) Analyze the residuals from this experiment. What conclusions can you draw about the underlying
model assumptions?
There is nothing unusual about the residual plots.
3-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
123.333
99
70.8333
90
80
70
Res iduals
Normal % probability
95
50
2
18.3333
30
20
10
5
-34.1667
1
-86.6667
-86.6667
-34.1667
18.3333
70.8333
123.333
1500.00
Res idual
1526.67
1553.33
1580.00
Predicted
Residuals vs. Rodding Level
123.333
Res iduals
70.8333
2
18.3333
-34.1667
-86.6667
1
2
3
4
Rodding Level
(d) Construct a graphical display to compare the treatment means as describe in Section 3-5.3.
3-12
1606.67
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a l e d t D is t r ib u tio n
(1 0 , 2 5 )
(1 5 )
(2 0 )
1418 1459 1500 1541 1582 1623
1664
M e a n C o m p r e s s iv e S tr e n g t h
3-9 An article in Environment International (Vol. 18, No. 4, 1992) describes an experiment in which
the amount of radon released in showers was investigated. Radon enriched water was used in the
experiment and six different orifice diameters were tested in shower heads. The data from the experiment
are shown in the following table.
Orifice Dia.
0.37
0.51
0.71
1.02
1.40
1.99
80
75
74
67
62
60
Radon Released (%)
83
83
75
79
73
76
72
74
62
67
61
64
85
79
77
74
69
66
(a) Does the size of the orifice affect the mean percentage of radon released? Use D = 0.05.
Yes. There is at least one treatment mean that is different.
Design Expert Output
Response: Radon Released in %
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1133.38
5
226.68
A
1133.38
5
226.68
Residual
132.25
18
7.35
Lack of Fit
0.000
0
Pure Error
132.25
18
7.35
Cor Total
1265.63
23
F
Value
30.85
30.85
The Model F-value of 30.85 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
EstimatedStandard
Mean Error
1-0.37
82.75
1.36
2-0.51
77.00
1.36
3-0.71
75.00
1.36
4-1.02
71.75
1.36
5-1.40
65.00
1.36
3-13
Prob > F
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
6-1.99
62.75
1.36
Treatment
1 vs 2
1 vs 3
1 vs 4
1 vs 5
1 vs 6
2 vs 3
2 vs 4
2 vs 5
2 vs 6
3 vs 4
3 vs 5
3 vs 6
4 vs 5
4 vs 6
5 vs 6
Mean
Difference
5.75
7.75
11.00
17.75
20.00
2.00
5.25
12.00
14.25
3.25
10.00
12.25
6.75
9.00
2.25
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Standard
Error
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
1.92
t for H0
Coeff=0
3.00
4.04
5.74
9.26
10.43
1.04
2.74
6.26
7.43
1.70
5.22
6.39
3.52
4.70
1.17
Prob > |t|
0.0077
0.0008
< 0.0001
< 0.0001
< 0.0001
0.3105
0.0135
< 0.0001
< 0.0001
0.1072
< 0.0001
< 0.0001
0.0024
0.0002
0.2557
(b) Find the P-value for the F statistic in part (a). P=3.161 x 10-8
(c) Analyze the residuals from this experiment.
There is nothing unusual about the residuals.
Normal plot of residuals
Residuals vs. Predicted
4
2
95
90
2
1.8125
80
70
Res iduals
Normal % probability
99
50
30
20
2
-0.375
2
10
-2.5625
5
2
1
-4.75
-4.75
-2.5625
-0.375
1.8125
4
62.75
Res idual
67.75
72.75
Predicted
3-14
77.75
82.75
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Orifice Diameter
4
Res iduals
2
2
1.8125
2
-0.375
2
-2.5625
2
-4.75
1
2
3
4
5
6
Orifice Diameter
(d) Find a 95 percent confidence interval on the mean percent radon released when the orifice diameter is
1.40.
7.35
Treatment 5 (Orifice =1.40): 6 r 2.101
4
62.152 d P d 67.848
(e) Construct a graphical display to compare the treatment means as describe in Section 3-5.3. What
conclusions can you draw?
S c a le d t D is t r ib u t io n
(6 )
60
(5 )
65
(4 )
70
(3 )
75
(2 )
(1 )
80
C o n d u c t iv it y
Treatments 5 and 6 as a group differ from the other means; 2, 3, and 4 as a group differ from the other
means, 1 differs from the others.
3-10 The response time in milliseconds was determined for three different types of circuits that could be
used in an automatic valve shutoff mechanism. The results are shown in the following table.
3-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Circuit Type
1
2
3
9
20
6
Response Time
10
8
23
17
8
16
12
21
5
15
30
7
(a) Test the hypothesis that the three circuit types have the same response time. Use D = 0.01.
From the computer printout, F=16.08, so there is at least one circuit type that is different.
Design Expert Output
Response: Response Time in ms
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
543.60
2
271.80
A
543.60
2
271.80
Residual
202.80
12
16.90
Lack of Fit
0.000
0
Pure Error
202.80
12
16.90
Cor Total
746.40
14
F
Value
16.08
16.08
Prob > F
0.0004
0.0004
significant
The Model F-value of 16.08 implies the model is significant. There is only
a 0.04% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
10.80
1.84
2-2
22.20
1.84
3-3
8.40
1.84
Treatment
1 vs 2
1 vs 3
2 vs 3
Mean
Difference
-11.40
2.40
13.80
DF
1
1
1
Standard
Error
2.60
2.60
2.60
t for H0
Coeff=0
-4.38
0.92
5.31
Prob > |t|
0.0009
0.3742
0.0002
(b) Use Tukey’s test to compare pairs of treatment means. Use D = 0.01.
S yi .
MS E
1690
1.8385
n
5
q0.01, 3 ,12 5.04
t 0 1.8385 5.04 9.266
1 vs. 2: ~10.8-22.2~=11.4 > 9.266
1 vs. 3: ~10.8-8.4~=2.4 < 9.266
2 vs. 3: ~22.2-8.4~=13.8 > 9.266
1 and 2 are different. 2 and 3 are different.
Notice that the results indicate that the mean of treatment 2 differs from the means of both treatments 1
and 3, and that the means for treatments 1 and 3 are the same. Notice also that the Fisher LSD procedure
(see the computer output) gives the same results.
(c) Use the graphical procedure in Section 3-5.3 to compare the treatment means. What conclusions can
you draw? How do they compare with the conclusions from part (a).
The scaled-t plot agrees with part (b). In this case, the large difference between the mean of treatment 2
and the other two treatments is very obvious.
3-16
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a le d t D is t r ib u t io n
(3 )
5
(1 )
(2 )
10
15
20
25
T e n s ile S t r e n g th
(d) Construct a set of orthogonal contrasts, assuming that at the outset of the experiment you suspected
the response time of circuit type 2 to be different from the other two.
C1
P1 2P 2 P 3 0
P1 2 P2 P3 z 0
y1. 2 y2. y3.
C1
54 2 111 42 126
H0
H1
SSC1
FC1
126
5 6
529.2
16.9
2
529.2
31.31
Type 2 differs from the average of type 1 and type 3.
(e) If you were a design engineer and you wished to minimize the response time, which circuit type
would you select?
Either type 1 or type 3 as they are not different from each other and have the lowest response time.
(f) Analyze the residuals from this experiment. Are the basic analysis of variance assumptions satisfied?
The normal probability plot has some points that do not lie along the line in the upper region. This may
indicate potential outliers in the data.
3-17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
7.8
99
4.55
90
80
70
Res iduals
Normal % probability
95
50
30
20
10
5
1.3
-1.95
1
-5.2
-5.2
-1.95
1.3
4.55
7.8
8.40
Res idual
11.85
15.30
18.75
22.20
Predicted
Residuals vs. Circuit Type
7.8
Res iduals
4.55
1.3
-1.95
-5.2
1
2
3
Circuit Type
3-11 The effective life of insulating fluids at an accelerated load of 35 kV is being studied. Test data
have been obtained for four types of fluids. The results were as follows:
Fluid Type
1
2
3
4
17.6
16.9
21.4
19.3
18.9
15.3
23.6
21.1
Life (in h) at 35 kV Load
16.3
17.4
20.1
18.6
17.1
19.5
19.4
18.5
20.5
16.9
17.5
18.3
21.6
20.3
22.3
19.8
(a) Is there any indication that the fluids differ? Use D = 0.05.
At D = 0.05 there are no difference, but at since the P-value is just slightly above 0.05, there is probably a
difference in means.
Design Expert Output
Response:
Life
in in h
3-18
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
30.17
3
10.06
A
30.16
3
10.05
Residual
65.99
20
3.30
Lack of Fit
0.000
0
Pure Error
65.99
20
3.30
Cor Total
96.16
23
F
Value
3.05
3.05
Prob > F
0.0525
0.0525
not significant
The Model F-value of 3.05 implies there is a 5.25% chance that a "Model F-Value"
this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
18.65
0.74
2-2
17.95
0.74
3-3
20.95
0.74
4-4
18.82
0.74
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
0.70
-2.30
-0.17
-3.00
-0.87
2.13
DF
1
1
1
1
1
1
Standard
Error
1.05
1.05
1.05
1.05
1.05
1.05
t for H0
Coeff=0
0.67
-2.19
-0.16
-2.86
-0.83
2.03
Prob > |t|
0.5121
0.0403
0.8753
0.0097
0.4183
0.0554
(b) Which fluid would you select, given that the objective is long life?
Treatment 3. The Fisher LSD procedure in the computer output indicates that the fluid 3 is different from
the others, and it’s average life also exceeds the average lives of the other three fluids.
(c) Analyze the residuals from this experiment. Are the basic analysis of variance assumptions satisfied?
There is nothing unusual in the residual plots.
Normal plot of residuals
Residuals vs. Predicted
2.95
99
1.55
90
80
70
Res iduals
Normal % probability
95
50
30
20
10
5
0.15
-1.25
1
-2.65
-2.65
-1.25
0.15
1.55
2.95
17.95
Res idual
18.70
19.45
Predicted
3-19
20.20
20.95
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Fluid Type
2.95
Res iduals
1.55
0.15
-1.25
-2.65
1
2
3
4
Fluid Type
3-12 Four different designs for a digital computer circuit are being studied in order to compare the
amount of noise present. The following data have been obtained:
Circuit Design
1
2
3
4
19
80
47
95
Noise Observed
19
30
73
56
25
35
83
78
20
61
26
46
8
80
50
97
(a) Is the amount of noise present the same for all four designs? Use D = 0.05.
No, at least one treatment mean is different.
Design Expert Output
Response:
Noise
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
12042.00
3
4014.00
A
12042.00
3
4014.00
Residual
2948.80
16
184.30
Lack of Fit
0.000
0
Pure Error
2948.80
16
184.30
Cor Total
14990.80
19
F
Value
21.78
21.78
Prob > F
< 0.0001
< 0.0001
The Model F-value of 21.78 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
19.20
6.07
2-2
70.00
6.07
3-3
36.60
6.07
4-4
79.80
6.07
Treatment
1 vs 2
Mean
Difference
-50.80
DF
1
Standard
Error
8.59
t for H0
Coeff=0
-5.92
3-20
Prob > |t|
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1 vs
1 vs
2 vs
2 vs
3 vs
3
4
3
4
4
-17.40
-60.60
33.40
-9.80
-43.20
1
1
1
1
1
8.59
8.59
8.59
8.59
8.59
-2.03
-7.06
3.89
-1.14
-5.03
0.0597
< 0.0001
0.0013
0.2705
0.0001
(b) Analyze the residuals from this experiment. Are the basic analysis of variance assumptions satisfied?
There is nothing unusual about the residual plots.
Normal plot of residuals
Residuals vs. Predicted
17.2
2
95
90
4.45
2
80
70
Res iduals
Normal % probability
99
50
30
20
10
-8.3
-21.05
5
1
-33.8
-33.8
-21.05
-8.3
4.45
17.2
19.20
Res idual
34.35
49.50
64.65
79.80
Predicted
Residuals vs. Circuit Design
17.2
2
4.45
Res iduals
2
-8.3
-21.05
-33.8
1
2
3
4
Circuit Design
(c) Which circuit design would you select for use? Low noise is best.
From the Design Expert Output, the Fisher LSD procedure comparing the difference in means identifies
Type 1 as having lower noise than Types 2 and 4. Although the LSD procedure comparing Types 1 and 3
has a P-value greater than 0.05, it is less than 0.10. Unless there are other reasons for choosing Type 3,
Type 1 would be selected.
3-21
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
3-13 Four chemists are asked to determine the percentage of methyl alcohol in a certain chemical
compound. Each chemist makes three determinations, and the results are the following:
Chemist
1
2
3
4
Percentage of Methyl Alcohol
84.99
84.04
84.38
85.15
85.13
84.88
84.72
84.48
85.16
84.20
84.10
84.55
(a) Do chemists differ significantly? Use D = 0.05.
There is no significant difference at the 5% level, but chemists differ significantly at the 10% level.
Design Expert Output
Response: Methyl Alcohol in %
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1.04
3
0.35
A
1.04
3
0.35
Residual
0.86
8
0.11
Lack of Fit
0.000
0
Pure Error
0.86
8
0.11
Cor Total
1.90
11
F
Value
3.25
3.25
Prob > F
0.0813
0.0813
The Model F-value of 3.25 implies there is a 8.13% chance that a "Model F-Value"
this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
84.47
0.19
2-2
85.05
0.19
3-3
84.79
0.19
4-4
84.28
0.19
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-0.58
-0.32
0.19
0.27
0.77
0.50
DF
1
1
1
1
1
1
Standard
Error
0.27
0.27
0.27
0.27
0.27
0.27
t for H0
Coeff=0
-2.18
-1.18
0.70
1.00
2.88
1.88
(b) Analyze the residuals from this experiment.
There is nothing unusual about the residual plots.
3-22
Prob > |t|
0.0607
0.2703
0.5049
0.3479
0.0205
0.0966
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
0.52
99
0.2825
90
80
70
Res iduals
Normal % probability
95
50
30
20
10
5
0.045
-0.1925
1
-0.43
-0.43
-0.1925
0.045
0.2825
0.52
84.28
Res idual
84.48
84.67
84.86
85.05
Predicted
Residuals vs. Chemist
0.52
Res iduals
0.2825
0.045
-0.1925
-0.43
1
2
3
4
Chemist
(c) If chemist 2 is a new employee, construct a meaningful set of orthogonal contrasts that might have
been useful at the start of the experiment.
Chemists
1
2
3
4
Total
253.41
255.16
254.36
252.85
Contrast Totals:
3-23
C1
1
-3
1
1
-4.86
C2
-2
0
1
1
0.39
C3
0
0
-1
1
-1.51
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
SS C1
SS C 2
SS C 3
0.656
4.86 2
0.656 FC1
6.115*
3 12
0.10727
0.39 2
0.008
0.008 FC 2
0.075
36
0.10727
0.380
1.51 2
0.380 FC 3
3.54
32
0.10727
Only contrast 1 is significant at 5%.
3-14 Three brands of batteries are under study. It is s suspected that the lives (in weeks) of the three
brands are different. Five batteries of each brand are tested with the following results:
Brand 1
100
96
92
96
92
Weeks of Life
Brand 2
76
80
75
84
82
Brand 3
108
100
96
98
100
(a) Are the lives of these brands of batteries different?
Yes, at least one of the brands is different.
Design Expert Output
Response:
Life
in Weeks
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1196.13
2
598.07
A
1196.13
2
598.07
Residual
187.20
12
15.60
Lack of Fit
0.000
0
Pure Error
187.20
12
15.60
Cor Total
1383.33
14
F
Value
38.34
38.34
Prob > F
< 0.0001
< 0.0001
The Model F-value of 38.34 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
95.20
1.77
2-2
79.40
1.77
3-3
100.40
1.77
Mean
Standard
Treatment Difference
DF
Error
1 vs 2
15.80
1
2.50
1 vs 3
-5.20
1
2.50
2 vs 3
-21.00
1
2.50
t for H0
Coeff=0
6.33
-2.08
-8.41
(b) Analyze the residuals from this experiment.
There is nothing unusual about the residuals.
3-24
Prob > |t|
< 0.0001
0.0594
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
7.6
99
4.6
90
80
70
Res iduals
Normal % probability
95
50
30
20
1.6
2
2
10
5
-1.4
1
2
-4.4
-4.4
-1.4
1.6
4.6
7.6
79.40
84.65
Res idual
89.90
95.15
100.40
Predicted
Residuals vs. Brand
7.6
Res iduals
4.6
1.6
2
2
-1.4
2
-4.4
1
2
3
Brand
(c) Construct a 95 percent interval estimate on the mean life of battery brand 2. Construct a 99 percent
interval estimate on the mean difference between the lives of battery brands 2 and 3.
y i . r tD
2
,N a
MS E
n
Brand 2: 79.4 r 2.179
15.60
5
79.40 r 3.849
75.551 d P2 d 83.249
Brand 2 - Brand 3: y i . y j . r t D
2
,N a
2 15.60
5
28.631 d P2 P3 d 13.369
79.4 100.4 r 3.055
3-25
2MS E
n
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(d) Which brand would you select for use? If the manufacturer will replace without charge any battery
that fails in less than 85 weeks, what percentage would the company expect to replace?
Chose brand 3 for longest life. Mean life of this brand in 100.4 weeks, and the variance of life is
estimated by 15.60 (MSE). Assuming normality, then the probability of failure before 85 weeks is:
§ 85 100.4 ·
¸ ) 3.90
) ¨¨
¸
© 15.60 ¹
0.00005
That is, about 5 out of 100,000 batteries will fail before 85 week.
3-15 Four catalysts that may affect the concentration of one component in a three component liquid
mixture are being investigated. The following concentrations are obtained:
Catalyst
2
3
56.3
50.1
54.5
54.2
57.0
55.4
55.3
1
58.2
57.2
58.4
55.8
54.9
4
52.9
49.9
50.0
51.7
(a) Do the four catalysts have the same effect on concentration?
No, their means are different.
Design Expert Output
Response:
Concentration
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
85.68
3
28.56
A
85.68
3
28.56
Residual
34.56
12
2.88
Lack of Fit
0.000
0
Pure Error
34.56
12
2.88
Cor Total
120.24
15
F
Value
9.92
9.92
Prob > F
0.0014
0.0014
The Model F-value of 9.92 implies the model is significant. There is only
a 0.14% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
56.90
0.76
2-2
55.77
0.85
3-3
53.23
0.98
4-4
51.13
0.85
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
1.13
3.67
5.77
2.54
4.65
2.11
DF
1
1
1
1
1
1
Standard
Error
1.14
1.24
1.14
1.30
1.20
1.30
t for H0
Coeff=0
0.99
2.96
5.07
1.96
3.87
1.63
3-26
Prob > |t|
0.3426
0.0120
0.0003
0.0735
0.0022
0.1298
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(b) Analyze the residuals from this experiment.
There is nothing unusual about the residual plots.
Normal plot of residuals
Residuals vs. Predicted
2.16667
95
90
0.841667
80
70
Res iduals
Normal % probability
99
50
-0.483333
30
20
10
-1.80833
5
1
-3.13333
-3.13333
-1.80833
-0.483333
0.841667
2.16667
51.13
Res idual
52.57
54.01
55.46
Predicted
Residuals vs. Catalyst
2.16667
Res iduals
0.841667
-0.483333
-1.80833
-3.13333
1
2
3
4
Catalyst
(c) Construct a 99 percent confidence interval estimate of the mean response for catalyst 1.
y i . r tD
2
,N a
MS E
n
Catalyst 1: 56.9 r 3.055
2.88
5
56.9 r 2.3186
54.5814 d P1 d 59.2186
3-27
56.90
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
3-16 An experiment was performed to investigate the effectiveness of five insulating materials. Four
samples of each material were tested at an elevated voltage level to accelerate the time to failure. The
failure times (in minutes) is shown below.
Material
1
2
3
4
5
110
1
880
495
7
Failure Time (minutes)
157
194
178
2
4
18
1256
5276
4355
7040
5307
10050
5
29
2
(a) Do all five materials have the same effect on mean failure time?
No, at least one material is different.
Design Expert Output
Response:
Failure Time in Minutes
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1.032E+008 4
2.580E+007
A
1.032E+008 4
2.580E+007
Residual
6.251E+007 15
4.167E+006
Lack of Fit
0.000
0
Pure Error
6.251E+00715
4.167E+006
Cor Total
1.657E+008 19
F
Value
6.19
6.19
Prob > F
0.0038
0.0038
significant
The Model F-value of 6.19 implies the model is significant. There is only
a 0.38% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
159.75
1020.67
2-2
6.25
1020.67
3-3
2941.75
1020.67
4-4
5723.00
1020.67
5-5
10.75
1020.67
Treatment
1 vs 2
1 vs 3
1 vs 4
1 vs 5
2 vs 3
2 vs 4
2 vs 5
3 vs 4
3 vs 5
4 vs 5
Mean
Difference
153.50
-2782.00
-5563.25
149.00
-2935.50
-5716.75
-4.50
-2781.25
2931.00
5712.25
DF
1
1
1
1
1
1
1
1
1
1
Standard
Error
1443.44
1443.44
1443.44
1443.44
1443.44
1443.44
1443.44
1443.44
1443.44
1443.44
t for H0
Coeff=0
0.11
-1.93
-3.85
0.10
-2.03
-3.96
-3.118E-003
-1.93
2.03
3.96
Prob > |t|
0.9167
0.0731
0.0016
0.9192
0.0601
0.0013
0.9976
0.0732
0.0604
0.0013
(b) Plot the residuals versus the predicted response. Construct a normal probability plot of the residuals.
What information do these plots convey?
3-28
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Predicted
Normal plot of residuals
4327
99
95
Normal % probability
Res iduals
1938.25
-450.5
-2839.25
90
80
70
50
30
20
10
5
1
-5228
6.25
1435.44
2864.62
4293.81
5723.00
-5228
-2839.25
Predicted
-450.5
1938.25
4327
Res idual
The plot of residuals versus predicted has a strong outward-opening funnel shape, which indicates the
variance of the original observations is not constant. The residuals plotted in the normal probability plot
also imply that the normality assumption is not valid. A data transformation is recommended.
(c) Based on your answer to part (b) conduct another analysis of the failure time data and draw
appropriate conclusions.
A natural log transformation was applied to the failure time data. The analysis identifies that there exists
at least one difference in treatment means.
Design Expert Output
Response:
Failure Time in Minutes Transform:
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
165.06
4
41.26
A
165.06
4
41.26
Residual
16.44
15
1.10
Lack of Fit
0.000
0
Pure Error
16.44
15
1.10
Cor Total
181.49
19
Natural log
Constant:
F
Value
37.66
37.66
Prob > F
< 0.0001
< 0.0001
The Model F-value of 37.66 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
5.05
0.52
2-2
1.24
0.52
3-3
7.72
0.52
4-4
8.21
0.52
5-5
1.90
0.52
Treatment
1 vs 2
1 vs 3
1 vs 4
1 vs 5
2 vs 3
Mean
Difference
3.81
-2.66
-3.16
3.15
-6.47
DF
1
1
1
1
1
Standard
Error
0.74
0.74
0.74
0.74
0.74
t for H0
Coeff=0
5.15
-3.60
-4.27
4.25
-8.75
3-29
Prob > |t|
0.0001
0.0026
0.0007
0.0007
< 0.0001
0.000
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2 vs
2 vs
3 vs
3 vs
4 vs
4
5
4
5
5
-6.97
-0.66
-0.50
5.81
6.31
1
1
1
1
1
0.74
0.74
0.74
0.74
0.74
-9.42
-0.89
-0.67
7.85
8.52
< 0.0001
0.3856
0.5116
< 0.0001
< 0.0001
There is nothing unusual about the residual plots when the natural log transformation is applied.
Normal plot of residuals
Residuals vs. Predicted
1.64792
95
90
0.733576
80
70
Res iduals
Normal % probability
99
50
-0.180766
30
20
10
-1.09511
5
1
-2.00945
-2.00945
-1.09511
-0.180766
0.733576
1.64792
1.24
Res idual
2.99
4.73
6.47
8.21
Predicted
Residuals vs. Material
1.64792
Res iduals
0.733576
-0.180766
-1.09511
-2.00945
1
2
3
4
5
Material
3-17 A semiconductor manufacturer has developed three different methods for reducing particle counts
on wafers. All three methods are tested on five wafers and the after-treatment particle counts obtained.
The data are shown below.
Method
1
2
3
31
62
58
10
40
27
Count
21
24
120
3-30
4
30
97
1
35
68
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Do all methods have the same effect on mean particle count?
No, at least one method has a different effect on mean particle count.
Design Expert Output
Response:
Count
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
8963.73
2
4481.87
A
8963.73
2
4481.87
Residual
6796.00
12
566.33
Lack of Fit
0.000
0
Pure Error
6796.00
12
566.33
Cor Total
15759.73
14
F
Value
7.91
7.91
Prob > F
0.0064
0.0064
significant
The Model F-value of 7.91 implies the model is significant. There is only
a 0.64% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
13.40
10.64
2-2
38.20
10.64
3-3
73.00
10.64
Treatment
1 vs 2
1 vs 3
2 vs 3
Mean
Difference
-24.80
-59.60
-34.80
DF
1
1
1
Standard
Error
15.05
15.05
15.05
t for H0
Coeff=0
-1.65
-3.96
-2.31
Prob > |t|
0.1253
0.0019
0.0393
(b) Plot the residuals versus the predicted response. Construct a normal probability plot of the residuals.
Are there potential concerns about the validity of the assumptions?
The plot of residuals versus predicted appears to be funnel shaped. This indicates the variance of the
original observations is not constant. The residuals plotted in the normal probability plot do not fall along
a straight line, which suggests that the normality assumption is not valid. A data transformation is
recommended.
Residuals vs. Predicted
Normal plot of residuals
47
99
95
Normal % probability
Res iduals
23.75
0.5
-22.75
90
80
70
50
30
20
10
5
1
-46
13.40
28.30
43.20
58.10
73.00
-46
Predicted
-22.75
0.5
Res idual
3-31
23.75
47
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) Based on your answer to part (b) conduct another analysis of the particle count data and draw
appropriate conclusions.
For count data, a square root transformation is often very effective in resolving problems with inequality
of variance. The analysis of variance for the transformed response is shown below. The difference
between methods is much more apparent after applying the square root transformation.
Design Expert Output
Response:
Count Transform: Square root
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
63.90
2
31.95
A
63.90
2
31.95
Residual
38.96
12
3.25
Lack of Fit
0.000
0
Pure Error
38.96
12
3.25
Cor Total
102.86
14
Constant:
0.000
F
Value
9.84
9.84
Prob > F
0.0030
0.0030
significant
The Model F-value of 9.84 implies the model is significant. There is only
a 0.30% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
3.26
0.81
2-2
6.10
0.81
3-3
8.31
0.81
Treatment
1 vs 2
1 vs 3
2 vs 3
Mean
Difference
-2.84
-5.04
-2.21
Standard
Error
1.14
1.14
1.14
DF
1
1
1
t for H0
Coeff=0
-2.49
-4.42
-1.94
Prob > |t|
0.0285
0.0008
0.0767
3-18 Consider testing the equality of the means of two normal populations, where the variances are
unknown but are assumed to be equal. The appropriate test procedure is the pooled t test. Show that the
pooled t test is equivalent to the single factor analysis of variance.
t0
y1. y 2.
2
n
Sp
n
¦
Sp
Furthermore,
y1 j y1. 2
j 1
§n·
y1. y2. 2 ¨ ¸
©2¹
~ t 2 n 2 assuming n1 = n2 = n
n
¦
y2 j y2.
2
j 1
¦
i 1
n
¦¦ y
ij
y1.
i 1 j 1
2n 2
2
2
2n 2
2
{ MSE for a=2
yi2. y..2
, which is exactly the same as SSTreatments in a one-way
n 2n
classification with a=2. Thus we have shown that t 20
SS Treatments
. In general, we know that t u2
MS E
F1,u
so that t 02 ~ F1,2 n 2 . Thus the square of the test statistic from the pooled t-test is the same test statistic
that results from a single-factor analysis of variance with a=2.
3-32
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a
3-19 Show that the variance of the linear combination
¦c y
i
i.
is V 2
i 1
ª a
º
V « ci yi . »
«¬ i 1
»¼
a
¦
¦
º
ª ni
ci2V «
yij »
«¬ j 1 »¼
1
a
V ci yi .
i 1
¦
i
¦
a
¦c
a
¦n c
2
i i
.
i 1
a
ni
¦ ¦V y
ci2
i 1
ij .
, V yij
V2
j 1
2
2
i niV
i 1
3-20 In a fixed effects experiment, suppose that there are n observations for each of four treatments. Let
Q12 , Q22 , Q32 be single-degree-of-freedom components for the orthogonal contrasts.
Prove that
Q12 Q22 Q32 .
SS Treatments
C1
C2
C3
Q12
Q 22
Q32
3 y1. y 2. y 3. y 4.
2 y 2. y 3. y 4.
y 3. y 4.
Q 22
Q32
Q12
SS C 2
SS C 3
Q 22
Q32
( 3 y1. y 2. y 3. y 4. ) 2
12n
( 2 y 2. y 3. y 4. ) 2
6n
( y 3. y 4. ) 2
2n
4
§
·
9
y i2. 6¨
yi. y j. ¸
¨
¸
i 1
i j
©
¹
and since
12n
¦
Q12
SS C1
¦¦
4
¦¦ y y
i. j .
i j
1 §¨ 2
y..
2 ¨©
4
¦
i 1
·
yi2. ¸ , we have Q12 Q 22 Q32
¸
¹
for a=4.
¦y
12
2
i.
3 y ..2
i 1
12n
4
¦
i 1
y i2. y ..2
n
4n
SS Treatments
3-21 Use Bartlett's test to determine if the assumption of equal variances is satisfied in Problem 3-14.
Use D = 0.05. Did you reach the same conclusion regarding the equality of variance by examining the
residual plots?
F 02
2.3026
q
, where
c
3-33
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
N a log 10 S 2p
q
a
¦n
i
1 log 10 S i2
i 1
c 1
1
3 a 1
a
S p2
S12
S22
S32
11.2
14.8
20.8
¦n
i
§
¨
¨
©i
a
¦n
i
1
1
N a
1
·
¸
¹
1 ¸
1 S i2
i 1
N a
5 1 11.2 5 1 14.8 5 1 20.8
S 2p
15 3
1 14.8 5 1 20.8
5
1
11
2
5
.
S 2p
15.6
15 3
a
·
1 §¨
5 1 1 15 3 1 ¸
c 1
¸
¨
3 3 1 © i 1
¹
1 §3 1 ·
c 1
¨ ¸ 1.1389
3 3 1 © 4 12 ¹
¦
a
¦n
q
N a log 10 S p2
q
15 3 log 10 15.6 4 log 10 11.2 log 10 14.8 log 10 20.8
i
1 log 10 S i2
i 1
q 14.3175 14.150
F 02
2.3026
q
c
0.1675
2.3026
0.1675
1.1389
0.3386
F 02.05 ,4
9.49
Cannot reject null hypothesis; conclude that the variance are equal. This agrees with the residual plots in
Problem 3-16.
3-22 Use the modified Levene test to determine if the assumption of equal variances is satisfied on
Problem 3-14. Use D = 0.05. Did you reach the same conclusion regarding the equality of variances by
examining the residual plots?
The absolute value of Battery Life – brand median is:
yij y i
Brand 1
4
0
4
0
4
Brand 2
4
0
5
4
2
Brand 3
8
0
4
2
0
The analysis of variance indicates that there is not a difference between the different brands and therefore
the assumption of equal variances is satisfired.
Design Expert Output
Response: Mod Levine
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
3-34
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Source
Model
A
Pure Error
Cor Total
Sum of
Squares
0.93
0.93
80.00
80.93
Mean
Square
0.47
0.47
6.67
DF
2
2
12
14
F
Value
0.070
0.070
Prob > F
0.9328
0.9328
3-23 Refer to Problem 3-10. If we wish to detect a maximum difference in mean response times of 10
milliseconds with a probability of at least 0.90, what sample size should be used? How would you obtain a
preliminary estimate of V 2 ?
nD 2
)2
2aV
2
, use MSE from Problem 3-10 to estimate V 2 .
n 10 2
2 3 16.9
)2
Letting D
0.986n
0.05 , P(accept) = 0.1 , X1
a 1 2
Trial and Error yields:
n
5
6
7
X2
12
15
18
)
P(accept)
2.22
2.43
2.62
0.17
0.09
0.04
Choose n t 6, therefore N t 18
Notice that we have used an estimate of the variance obtained from the present experiment. This indicates
that we probably didn’t use a large enough sample (n was 5 in problem 3-10) to satisfy the criteria
specified in this problem. However, the sample size was adequate to detect differences in one of the
circuit types.
When we have no prior estimate of variability, sometimes we will generate sample sizes for a range of
possible variances to see what effect this has on the size of the experiment. Often a knowledgeable expert
will be able to bound the variability in the response, by statements such as “the standard deviation is going
to be at least…” or “the standard deviation shouldn’t be larger than…”.
3-24 Refer to Problem 3-14.
(a) If we wish to detect a maximum difference in mean battery life of 0.5 percent with a probability of at
least 0.90, what sample size should be used? Discuss how you would obtain a preliminary estimate of
V2 for answering this question.
Use the MSE from Problem 3-14.
)2
nD 2
2aV 2
Letting D
n 0.005 u 91.6667 2
0.002244n
2 3 15.60
0.05 , P(accept) = 0.1 , X1 a 1 2
)2
3-35
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Trial and Error yields:
X2
117
132
147
n
40
45
50
)
P(accept)
1.895
2.132
2.369
0.18
0.10
0.05
Choose n t 45, therefore N t 135
See the discussion from the previous problem about the estimate of variance.
(b) If the difference between brands is great enough so that the standard deviation of an observation is
increased by 25 percent, what sample size should be used if we wish to detect this with a probability
of at least 0.90?
X1
O
a 1 2
>
X2
N a 15 3 12
1 n 1 0.01P
2
@
1
>
D
0.05
1 n 1 0.01 25
2
@
1
P( accept ) d 0.1
1 0.5625n
Trial and Error yields:
n
40
45
50
X2
117
132
147
O
P(accept)
4.84
5.13
5.40
0.13
0.11
0.10
Choose n t 50, therefore N t 150
3-25 Consider the experiment in Problem 3-16. If we wish to construct a 95 percent confidence interval
on the difference in two mean battery lives that has an accuracy of r2 weeks, how many batteries of each
brand must be tested?
D
width
MS E
0.05
t 0.025 ,N a
15.6
2MS E
n
Trial and Error yields:
n
5
10
31
32
X2
12
27
90
93
t
width
2.179
2.05
1.99
1.99
5.44
3.62
1.996
1.96
Choose n t 31, therefore N t 93
3-26 Suppose that four normal populations have means of P1=50, P2=60, P3=50, and P4=60. How many
observations should be taken from each population so that the probability or rejecting the null hypothesis
3-36
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
of equal population means is at least 0.90? Assume that D=0.05 and that a reasonable estimate of the
error variance is V 2 =25.
Pi
P W i , i 1,2,3,4
4
P
W1
4
¦
¦P
i
i 1
4
5,W 2
W i2
220
4
5,W 3
)2
55
5,W 4
5
)
¦W
n
aV
2
2
i
100n
4 25
n
n
100
i 1
X1
3,X 2
4 n 1 ,D
0.05 , From the O.C. curves we can construct the following:
n
4
5
)
2.00
2.24
X2
12
16
E
0.18
0.08
1-E
0.82
0.92
Therefore, select n=5
3-27 Refer to Problem 3-26.
(a) How would your answer change if a reasonable estimate of the experimental error variance were V 2 =
36?
)2
)
X1
3,X 2
4 n 1 ,D
n
¦W
aV
2
i
100n
4 36
2
0.6944n
0.6944n
0.05 , From the O.C. curves we can construct the following:
n
5
6
7
)
1.863
2.041
2.205
X2
16
20
24
E
0.24
0.15
0.09
1-E
0.76
0.85
0.91
Therefore, select n=7
(b) How would your answer change if a reasonable estimate of the experimental error variance were V 2 =
49?
)2
)
X1
3,X 2
4 n 1 ,D
¦W
n
aV
2
i
100n
4 49
2
0.5102n
0.5102n
0.05 , From the O.C. curves we can construct the following:
3-37
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
n
7
8
9
)
1.890
2.020
2.142
X2
24
28
32
E
0.16
0.11
0.09
1-E
0.84
0.89
0.91
Therefore, select n=9
(c) Can you draw any conclusions about the sensitivity of your answer in the particular situation about
how your estimate of V affects the decision about sample size?
As our estimate of variability increases the sample size must increase to ensure the same power of the test.
(d) Can you make any recommendations about how we should use this general approach to choosing n in
practice?
When we have no prior estimate of variability, sometimes we will generate sample sizes for a range of
possible variances to see what effect this has on the size of the experiment. Often a knowledgeable expert
will be able to bound the variability in the response, by statements such as “the standard deviation is going
to be at least…” or “the standard deviation shouldn’t be larger than…”.
3-28 Refer to the aluminum smelting experiment described in Section 4-2. Verify that ratio control
methods do not affect average cell voltage. Construct a normal probability plot of residuals. Plot the
residuals versus the predicted values. Is there an indication that any underlying assumptions are violated?
Design Expert Output
Response:
Cell Average
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.746E-003 3
9.153E-004
A
2.746E-003 3
9.153E-004
Residual
0.090
20
4.481E-003
Lack of Fit
0.000
0
Pure Error
0.090
20
4.481E-003
Cor Total
0.092
23
F
Value
0.20
0.20
Prob > F
0.8922
0.8922
The "Model F-value" of 0.20 implies the model is not significant relative to the noise. There is a
89.22 % chance that a "Model F-value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
4.86
0.027
2-2
4.83
0.027
3-3
4.85
0.027
4-4
4.84
0.027
Mean
Treatment Difference
1 vs 2
0.027
1 vs 3
0.013
1 vs 4
0.025
2 vs 3
-0.013
2 vs 4
-1.667E-003
3 vs 4
0.012
DF
1
1
1
1
1
1
Standard
Error
0.039
0.039
0.039
0.039
0.039
0.039
t for H0
Coeff=0
0.69
0.35
0.65
-0.35
-0.043
0.30
The following residual plots are satisfactory.
3-38
Prob > |t|
0.4981
0.7337
0.5251
0.7337
0.9660
0.7659
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
0.105
99
2
0.05125
90
80
70
3
Res iduals
Normal % probability
95
50
-0.0025
30
20
10
5
-0.05625
1
-0.11
-0.11
-0.05625
-0.0025
0.05125
0.105
4.833
Res idual
4.840
4.847
4.853
4.860
Predicted
Residuals vs. Algorithm
0.105
2
0.05125
Res iduals
3
-0.0025
-0.05625
-0.11
1
2
3
4
Algorithm
3-29 Refer to the aluminum smelting experiment in Section 3-8. Verify the analysis of variance for pot
noise summarized in Table 3-13. Examine the usual residual plots and comment on the experimental
validity.
Design Expert Output
Response:
Cell StDev Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
6.17
3
2.06
A
6.17
3
2.06
Residual
1.87
20
0.094
Lack of Fit
0.000
0
Pure Error
1.87
20
0.094
Cor Total
8.04
23
Constant:
0.000
F
Value
21.96
21.96
Prob > F
< 0.0001
< 0.0001
The Model F-value of 21.96 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
3-39
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
-3.09
0.12
2-2
-3.51
0.12
3-3
-2.20
0.12
4-4
-3.36
0.12
Mean
Difference
0.42
-0.89
0.27
-1.31
-0.15
1.16
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Standard
Error
0.18
0.18
0.18
0.18
0.18
0.18
DF
1
1
1
1
1
1
t for H0
Coeff=0
2.38
-5.03
1.52
-7.41
-0.86
6.55
Prob > |t|
0.0272
< 0.0001
0.1445
< 0.0001
0.3975
< 0.0001
The following residual plots identify the residuals to be normally distributed, randomly distributed
through the range of prediction, and uniformly distributed across the different algorithms. This validates
the assumptions for the experiment.
Normal plot of residuals
Residuals vs. Predicted
0.512896
99
2
0.245645
90
80
70
Res iduals
Normal % probability
95
50
-0.0216069
30
20
10
5
3
2
2
-0.288858
2
1
-0.55611
-0.55611
-0.288858 -0.0216069
0.245645
0.512896
-3.51
Res idual
0.512896
2
Res iduals
0.245645
3
2
2
-0.288858
2
-0.55611
1
2
-2.85
Predicted
Residuals vs. Algorithm
-0.0216069
-3.18
3
4
Algorithm
3-40
-2.53
-2.20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
3-30 Four different feed rates were investigated in an experiment on a CNC machine producing a
component part used in an aircraft auxiliary power unit. The manufacturing engineer in charge of the
experiment knows that a critical part dimension of interest may be affected by the feed rate. However,
prior experience has indicated that only dispersion effects are likely to be present. That is, changing the
feed rate does not affect the average dimension, but it could affect dimensional variability. The engineer
makes five production runs at each feed rate and obtains the standard deviation of the critical dimension
(in 10-3 mm). The data are shown below. Assume that all runs were made in random order.
Feed Rate
(in/min)
10
12
14
16
1
0.09
0.06
0.11
0.19
Production
2
0.10
0.09
0.08
0.13
Run
3
0.13
0.12
0.08
0.15
4
0.08
0.07
0.05
0.20
5
0.07
0.12
0.06
0.11
(a) Does feed rate have any effect on the standard deviation of this critical dimension?
Because the residual plots were not acceptable for the non-transformed data, a square root transformation
was applied to the standard deviations of the critical dimension. Based on the computer output below, the
feed rate has an effect on the standard deviation of the critical dimension.
Design Expert Output
Response:
Run StDev Transform: Square root
Constant:
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
0.040
3
0.013
7.05
A
0.040
3
0.013
7.05
Residual
0.030
16
1.903E-003
Lack of Fit
0.000
0
Pure Error
0.030
16
1.903E-003
Cor Total
0.071
19
0.000
Prob > F
0.0031
0.0031
significant
The Model F-value of 7.05 implies the model is significant. There is only
a 0.31% chance that a "Model F-Value" this large could occur due to noise.
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-10
0.30
0.020
2-12
0.30
0.020
3-14
0.27
0.020
4-16
0.39
0.020
Mean
Treatment Difference
1 vs 2
4.371E-003
1 vs 3
0.032
1 vs 4
-0.088
2 vs 3
0.027
2 vs 4
-0.092
3 vs 4
-0.12
DF
1
1
1
1
1
1
Standard
Error
0.028
0.028
0.028
0.028
0.028
0.028
t for H0
Coeff=0
0.16
1.15
-3.18
0.99
-3.34
-4.33
Prob > |t|
0.8761
0.2680
0.0058
0.3373
0.0042
0.0005
(b) Use the residuals from this experiment of investigate model adequacy. Are there any problems with
experimental validity?
The residual plots are satisfactory.
3-41
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
0.0584817
99
2
0.028646
90
80
70
2
Res iduals
Normal % probability
95
50
-0.00118983
30
20
10
5
-0.0310256
1
-0.0608614
-0.0608614 -0.0310256 -0.00118983 0.028646
0.0584817
0.27
0.30
Res idual
0.33
0.36
0.39
Predicted
Residuals vs. Feed Rate
0.0584817
2
0.028646
Res iduals
2
-0.00118983
-0.0310256
-0.0608614
1
2
3
4
Feed Rate
3-31 Consider the data shown in Problem 3-10.
(a) Write out the least squares normal equations for this problem, and solve them for P and Wi , using the
usual constraint §¨
©
¦
3
Ŵ
i 1 i
0 ·¸ . Estimate W 1 W 2 .
¹
15P̂
5P̂
5P̂
5Ŵ 1
5Ŵ 1
5Ŵ 2
5Ŵ 2
15P̂
3
Imposing
¦ Ŵ
i
0 , therefore Pˆ 13.80 , Wˆ 1
5Ŵ 3
=111
5Ŵ 3
3.00 , Wˆ 2
i 1
3-42
=207
=54
=42
8.40 , Wˆ 3
5.40
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Wˆ1 Wˆ 2
3.00 8.40
11.40
(b) Solve the equations in (a) using the constraint Ŵ 3 0 . Are the estimators Ŵ i and P̂ the same as you
found in (a)? Why? Now estimate W 1 W 2 and compare your answer with that for (a). What
statement can you make about estimating contrasts in the W i ?
Imposing the constraint, Ŵ 3 0 we get the following solution to the normal equations: Pˆ 8.40 ,
Wˆ 1 2.40 , Wˆ 2 13.8 , and Ŵ 3 0 . These estimators are not the same as in part (a). However,
Wˆ1 Wˆ 2 2.40 13.80 11.40 , is the same as in part (a). The contrasts are estimable.
(c) Estimate P W 1 , 2W 1 W 2 W 3 and P W 1 W 2 using the two solutions to the normal equations.
Compare the results obtained in each case.
Contrast
P W1
2W 1 W 2 W 3
P W1 W 2
1
2
3
Estimated from Part (a)
10.80
-9.00
19.20
Estimated from Part (b)
10.80
-9.00
24.60
Contrasts 1 and 2 are estimable, 3 is not estimable.
3-32 Apply the general regression significance test to the experiment in Example 3-1. Show that the
procedure yields the same results as the usual analysis of variance.
From Table 3-3:
y ..
376
from Example 3-1, we have:
Pˆ 15.04 Wˆ1 5.24 Wˆ 2 0.36
Wˆ 3 2.56 Wˆ 4 6.56 Wˆ 5 4.24
5
5
¦¦ y
2
ij
6292 , with 25 degrees of freedom.
i 1 j 1
R P ,W
Pˆ y..
5
¦Wˆy
i.
i 1
15.04 376 5.24 49 0.36 77 2.56 88 ) 6.56 108 4.24 54
6,130.80
with 5 degrees of freedom.
5
SS E
5
¦¦ y
2
ij
R P ,W
6292 6130.8 161.20
i 1 j 1
with 25-5 degrees of freedom.
This is identical to the SSE found in Example 3-1.
3-43
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The reduced model:
RP
Pˆ y ..
RW P
15.04 376
R P ,W R P
Note: R W P
5655.04 , with 1 degree of freedom.
6130.8 5655.04
475.76 , with 5-1=4 degrees of freedom.
SSTreatment from Example 3-1.
Finally,
R Wt P
F0
118.94
8.06
4
SS E
20
14.76
which is the same as computed in Example 3-1.
3-33 Use the Kruskal-Wallis test for the experiment in Problem 3-11. Are the results comparable to
those found by the usual analysis of variance?
From Design Expert Output of Problem 3-11
Response:
Life
in in h
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
30.17
3
10.06
A
30.16
3
10.05
3.05
Residual 65.99
20
3.30
Lack of Fit 0.000
0
Pure Error 65.99
20
3.30
Cor Total 96.16
23
H
F
Value
3.05
0.0525
ª a Ri2. º
12
«
» 3 N 1
N N 1 ¬« i 1 ni »¼
Prob > F
0.0525
not significant
12
>4040.5@ 3 24 1
24 24 1
¦
F 02.05 ,3
5.81
7.81
Accept the null hypothesis; the treatments are not different. This agrees with the analysis of variance.
3-34 Use the Kruskal-Wallis test for the experiment in Problem 3-12. Compare conclusions obtained
with those from the usual analysis of variance?
From Design Expert Output of Problem 3-12
Response:
Noise
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
12042.00
3
4014.00
A
12042.00 3
4014.00 21.78
Residual 2948.80 16
184.30
Lack of Fit 0.000
0
F
Value
21.78
< 0.0001
3-44
Prob > F
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pure Error 2948.80 16
Cor Total 14990.80 19
H
184.30
ª a Ri2. º
12
» 3 N 1
«
N N 1 ¬« i 1 ni »¼
12
>2691.6@ 3 20 1
20 20 1
¦
F 02.05 ,4
13.90
12.84
Reject the null hypothesis because the treatments are different. This agrees with the analysis of variance.
3-35 Consider the experiment in Example 3-1. Suppose that the largest observation on tensile strength
is incorrectly recorded as 50. What effect does this have on the usual analysis of variance? What effect
does is have on the Kruskal-Wallis test?
The incorrect observation reduces the analysis of variance F0 from 14.76 to 5.44. It does not change the
value of the Kruskal-Wallis test.
3-45
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 4
Randomized Blocks, Latin Squares, and Related Designs
Solutions
4-1 A chemist wishes to test the effect of four chemical agents on the strength of a particular type of
cloth. Because there might be variability from one bolt to another, the chemist decides to use a randomized
block design, with the bolts of cloth considered as blocks. She selects five bolts and applies all four
chemicals in random order to each bolt. The resulting tensile strengths follow. Analyze the data from this
experiment (use D = 0.05) and draw appropriate conclusions.
Chemical
1
2
3
4
1
73
73
75
73
Design Expert Output
Response:
Strength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
157.00
4
39.25
Model
12.95
3
4.32
A
12.95
3
4.32
Residual
21.80
12
1.82
Cor Total
191.75
19
Bolt
3
74
75
78
75
2
68
67
68
71
4
71
72
73
75
F
Value
Prob > F
2.38
2.38
0.1211
0.1211
5
67
70
68
69
not significant
The "Model F-value" of 2.38 implies the model is not significant relative to the noise. There is a
12.11 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.35
71.75
1.88
60.56
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.3727
0.2158
-0.7426
10.558
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
70.60
0.60
2-2
71.40
0.60
3-3
72.40
0.60
4-4
72.60
0.60
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-0.80
-1.80
-2.00
-1.00
-1.20
-0.20
DF
1
1
1
1
1
1
Standard
Error
0.85
0.85
0.85
0.85
0.85
0.85
t for H0
Coeff=0
-0.94
-2.11
-2.35
-1.17
-1.41
-0.23
Prob > |t|
0.3665
0.0564
0.0370
0.2635
0.1846
0.8185
There is no difference among the chemical types at D = 0.05 level.
4-2 Three different washing solutions are being compared to study their effectiveness in retarding
bacteria growth in five-gallon milk containers. The analysis is done in a laboratory, and only three trials
4-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
can be run on any day. Because days could represent a potential source of variability, the experimenter
decides to use a randomized block design. Observations are taken for four days, and the data are shown
here. Analyze the data from this experiment (use D = 0.05) and draw conclusions.
Solution
1
2
3
Design Expert Output
Response:
Growth
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
1106.92
3
368.97
Model
703.50
2
351.75
A
703.50
2
351.75
Residual
51.83
6
8.64
Cor Total
1862.25
11
1
13
16
5
2
22
24
4
Days
3
18
17
1
4
39
44
22
F
Value
Prob > F
40.72
40.72
0.0003
0.0003
significant
The Model F-value of 40.72 implies the model is significant. There is only
a 0.03% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.94
18.75
15.68
207.33
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.9314
0.9085
0.7255
19.687
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
23.00
1.47
2-2
25.25
1.47
3-3
8.00
1.47
Treatment
1 vs 2
1 vs 3
2 vs 3
Mean
Difference
-2.25
15.00
17.25
DF
1
1
1
Standard
Error
2.08
2.08
2.08
t for H0
Coeff=0
-1.08
7.22
8.30
Prob > |t|
0.3206
0.0004
0.0002
There is a difference between the means of the three solutions. The Fisher LSD procedure indicates that
solution 3 is significantly different than the other two.
4-3 Plot the mean tensile strengths observed for each chemical type in Problem 4-1 and compare them to
a scaled t distribution. What conclusions would you draw from the display?
4-2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a le d t D is tr ib u ti o n
(1 )
7 0 .0
(3 ,4 )
(2 )
7 1 .0
7 2 .0
7 3 .0
M e a n S tre n g t h
S yi .
MS E
b
1.82
5
0.603
There is no obvious difference between the means. This is the same conclusion given by the analysis of
variance.
4-4 Plot the average bacteria counts for each solution in Problem 4-2 and compare them to an
appropriately scaled t distribution. What conclusions can you draw?
S c a le d t D is t r ib u t io n
(3)
5
(1)
10
15
20
(2)
25
B a c t e r ia G r o w t h
S yi .
MS E
b
8.64
4
1.47
There is no difference in mean bacteria growth between solutions 1 and 2. However, solution 3 produces
significantly lower mean bacteria growth. This is the same conclusion reached from the Fisher LSD
procedure in Problem 4-4.
4-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4-5 An article in the Fire Safety Journal (“The Effect of Nozzle Design on the Stability and Performance
of Turbulent Water Jets,” Vol. 4, August 1981) describes an experiment in which a shape factor was
determined for several different nozzle designs at six levels of efflux velocity. Interest focused on potential
differences between nozzle designs, with velocity considered as a nuisance variable. The data are shown
below:
Jet Efflux Velocity (m/s)
Nozzle
Design
1
2
3
4
5
11.73
0.78
0.85
0.93
1.14
0.97
14.37
0.80
0.85
0.92
0.97
0.86
16.59
0.81
0.92
0.95
0.98
0.78
20.43
0.75
0.86
0.89
0.88
0.76
23.46
0.77
0.81
0.89
0.86
0.76
28.74
0.78
0.83
0.83
0.83
0.75
(a) Does nozzle design affect the shape factor? Compare nozzles with a scatter plot and with an analysis
of variance, using D = 0.05.
Design Expert Output
Response:
Shape
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.063
5
0.013
Model
0.10
4
0.026
A
0.10
4
0.026
Residual
0.057
20
2.865E-003
Cor Total
0.22
29
F
Value
Prob > F
8.92
8.92
0.0003
0.0003
The Model F-value of 8.92 implies the model is significant. There is only
a 0.03% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.054
0.86
6.23
0.13
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.6407
0.5688
0.1916
9.438
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
0.78
0.022
2-2
0.85
0.022
3-3
0.90
0.022
4-4
0.94
0.022
5-5
0.81
0.022
Treatment
1 vs 2
1 vs 3
1 vs 4
1 vs 5
2 vs 3
2 vs 4
2 vs 5
3 vs 4
3 vs 5
4 vs 5
Mean
Difference
-0.072
-0.12
-0.16
-0.032
-0.048
-0.090
0.040
-0.042
0.088
0.13
DF
1
1
1
1
1
1
1
1
1
1
Standard
Error
0.031
0.031
0.031
0.031
0.031
0.031
0.031
0.031
0.031
0.031
t for H0
Coeff=0
-2.32
-3.88
-5.23
-1.02
-1.56
-2.91
1.29
-1.35
2.86
4.21
Nozzle design has a significant effect on shape factor.
4-4
Prob > |t|
0.0311
0.0009
< 0.0001
0.3177
0.1335
0.0086
0.2103
0.1926
0.0097
0.0004
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One Factor Plot
1.14
Shape
1.04236
0.944718
2
2
0.847076
2
2
0.749435
1
2
3
4
5
Nozzle Design
(b) Analyze the residual from this experiment.
The plots shown below do not give any indication of serious problems. Thre is some indication of a mild
outlier on the normal probability plot and on the plot of residualks versus the predicted velocity.
Residuals vs. Predicted
Normal plot of residuals
0.121333
99
0.0713333
90
80
70
Res iduals
Normal % probability
95
0.0213333
50
30
20
10
-0.0286667
5
1
-0.0786667
-0.0786667 -0.0286667 0.0213333 0.0713333
0.73
0.121333
Res idual
0.80
0.87
Predicted
4-5
0.95
1.02
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Nozzle Design
0.121333
Residuals
0.0713333
2
0.0213333
2
-0.0286667
-0.0786667
1
2
3
4
5
Nozzle Design
(c) Which nozzle designs are different with respect to shape factor? Draw a graph of average shape factor
for each nozzle type and compare this to a scaled t distribution. Compare the conclusions that you
draw from this plot to those from Duncan’s multiple range test.
S yi .
MS E
b
0.002865
6
0.021852
R2=
r0.05(2,20) S yi . =
(2.95)(0.021852)=
0.06446
R3=
r0.05(3,20) S yi . =
(3.10)(0.021852)=
0.06774
R4=
r0.05(4,20) S yi . =
(3.18)(0.021852)=
0.06949
R5=
r0.05(5,20) S yi . =
(3.25)(0.021852)=
0.07102
1 vs 4
1 vs 3
1 vs 2
1 vs 5
5 vs 4
5 vs 3
5 vs 2
2 vs 4
2 vs 3
3 vs 4
Mean Difference
0.16167
0.12000
0.07167
0.03167
0.13000
0.08833
0.04000
0.09000
0.04833
0.04167
4-6
>
>
>
<
>
>
<
>
<
<
R
0.07102
0.06949
0.06774
0.06446
0.06949
0.06774
0.06446
0.06774
0.06446
0.06446
different
different
different
different
different
different
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a le d t D is t r ib u t io n
(1)
0 .7 5
(5)
(2)
0 .8 0
0 .8 5
(3)
(4 )
0 .9 0
0 .9 5
S h a p e F a c to r
4-6 Consider the ratio control algorithm experiment described in Chapter 3, Section 3-8. The
experiment was actually conducted as a randomized block design, where six time periods were selected as
the blocks, and all four ratio control algorithms were tested in each time period. The average cell voltage
and the standard deviation of voltage (shown in parentheses) for each cell as follows:
Ratio Control
Time Period
Algorithms
1
2
3
4
5
6
1
4.93 (0.05)
4.86 (0.04)
4.75 (0.05)
4.95 (0.06)
4.79 (0.03)
4.88 (0.05)
2
4.85 (0.04)
4.91 (0.02)
4.79 (0.03)
4.85 (0.05)
4.75 (0.03)
4.85 (0.02)
3
4.83 (0.09)
4.88 (0.13)
4.90 (0.11)
4.75 (0.15)
4.82 (0.08)
4.90 (0.12)
4
4.89 (0.03)
4.77 (0.04)
4.94 (0.05)
4.86 (0.05)
4.79 (0.03)
4.76 (0.02)
(a) Analyze the average cell voltage data. (Use D = 0.05.) Does the choice of ratio control algorithm
affect the cell voltage?
Design Expert Output
Response:
Average
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.017
5
3.487E-003
Model
2.746E-003 3
9.153E-004
A
2.746E-003 3
9.153E-004
Residual
0.072
15
4.812E-003
Cor Total
0.092
23
F
Value
Prob > F
0.19
0.19
0.9014
0.9014
The "Model F-value" of 0.19 implies the model is not significant relative to the noise. There is a
90.14 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.069
4.84
1.43
0.18
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.0366
-0.1560
-1.4662
2.688
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
4.86
0.028
4-7
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2-2
3-3
4-4
4.83
4.85
4.84
0.028
0.028
0.028
Mean
Treatment Difference
1 vs 2
0.027
1 vs 3
0.013
1 vs 4
0.025
2 vs 3
-0.013
2 vs 4
-1.667E-003
3 vs 4
0.012
DF
1
1
1
1
1
1
Standard
Error
0.040
0.040
0.040
0.040
0.040
0.040
t for H0
Coeff=0
0.67
0.33
0.62
-0.33
-0.042
0.29
Prob > |t|
0.5156
0.7438
0.5419
0.7438
0.9674
0.7748
The ratio control algorithm does not affect the mean cell voltage.
(b) Perform an appropriate analysis of the standard deviation of voltage. (Recall that this is called “pot
noise.”) Does the choice of ratio control algorithm affect the pot noise?
Design Expert Output
Response:
StDev Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.94
5
0.19
Model
6.17
3
2.06
A
6.17
3
2.06
Residual
0.93
15
0.062
Cor Total
8.04
23
Constant:
0.000
F
Value
Prob > F
33.26
33.26
< 0.0001
< 0.0001
significant
The Model F-value of 33.26 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.25
-3.04
-8.18
2.37
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.8693
0.8432
0.6654
12.446
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-1
-3.09
0.10
2-2
-3.51
0.10
3-3
-2.20
0.10
4-4
-3.36
0.10
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
0.42
-0.89
0.27
-1.31
-0.15
1.16
DF
1
1
1
1
1
1
Standard
Error
0.14
0.14
0.14
0.14
0.14
0.14
t for H0
Coeff=0
2.93
-6.19
1.87
-9.12
-1.06
8.06
Prob > |t|
0.0103
< 0.0001
0.0813
< 0.0001
0.3042
< 0.0001
A natural log transformatio was applied to the pot noise data. The ratio control algorithm does affect the
pot noise.
(c) Conduct any residual analyses that seem appropriate.
4-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
Residuals vs. Predicted
0.288958
99
0.126945
90
80
70
Res iduals
Normal % probability
95
50
-0.0350673
30
20
10
5
-0.19708
1
-0.359093
-0.359093
-0.19708
-0.0350673
0.126945
0.288958
-3.73
Res idual
-3.26
-2.78
-2.31
-1.84
Predicted
Residuals vs. Algorithm
0.288958
Res iduals
0.126945
-0.0350673
-0.19708
-0.359093
1
2
3
4
Algorithm
The normal probability plot shows slight deviations from normality; however, still acceptable.
(d) Which ratio control algorithm would you select if your objective is to reduce both the average cell
voltage and the pot noise?
Since the ratio control algorithm has little effect on average cell voltage, select the algorithm that
minimizes pot noise, that is algorithm #2.
4-7 An aluminum master alloy manufacturer produces grain refiners in ingot form. This company
produces the product in four furnaces. Each furnace is known to have its own unique operating
characteristics, so any experiment run in the foundry that involves more than one furnace will consider
furnace a nuisance variable. The process engineers suspect that stirring rate impacts the grain size of the
product. Each furnace can be run at four different stirring rates. A randomized block design is run for a
particular refiner and the resulting grain size data is shown below.
Furnace
4-9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Stirring Rate
5
10
15
20
1
8
14
14
17
2
4
5
6
9
3
5
6
9
3
4
6
9
2
6
(a) Is there any evidence that stirring rate impacts grain size?
Design Expert Output
Response:
Grain Size
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
165.19
3
55.06
Model
22.19
3
7.40
A
22.19
3
7.40
Residual
78.06
9
8.67
Cor Total
265.44
15
F
Value
Prob > F
0.85
0.85
0.4995
0.4995
not significant
The "Model F-value" of 0.85 implies the model is not significant relative to the noise. There is a
49.95 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.95
7.69
38.31
246.72
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.2213
-0.0382
-1.4610
5.390
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-5
5.75
1.47
2-10
8.50
1.47
3-15
7.75
1.47
4-20
8.75
1.47
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
-2.75
-2.00
-3.00
0.75
-0.25
-1.00
DF
1
1
1
1
1
1
Standard
Error
2.08
2.08
2.08
2.08
2.08
2.08
t for H0
Coeff=0
-1.32
-0.96
-1.44
0.36
-0.12
-0.48
Prob > |t|
0.2193
0.3620
0.1836
0.7270
0.9071
0.6425
The analysis of variance shown above indicates that there is no difference in mean grain size due to the
different stirring rates.
(b) Graph the residuals from this experiment on a normal probability plot. Interpret this plot.
4-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal plot of residuals
99
Normal % probability
95
90
80
70
50
30
20
10
5
1
-3.8125
-2.0625
-0.3125
1.4375
3.1875
Res idual
The plot indicates that normality assumption is valid.
(c) Plot the residuals versus furnace and stirring rate. Does this plot convey any useful information?
Residuals vs. Stirring Rate
3.1875
Res iduals
1.4375
-0.3125
-2.0625
-3.8125
1
2
3
4
Stirring Rate
The variance is consistent at different stirring rates. Not only does this validate the assumption of uniform
variance, it also identifies that the different stirring rates do not affect variance.
(d) What should the process engineers recommend concerning the choice of stirring rate and furnace for
this particular grain refiner if small grain size is desirable?
There really isn’t any effect due to the stirring rate.
4-8
Analyze the data in Problem 4-2 using the general regression significance test.
P:
12P̂
4Ŵ 1
W1 :
4P̂
4Ŵ 1
W2 :
4P̂
4Ŵ 2
4Ŵ 3
3 Ê 1
3 Ê 2
3 Ê 3
3 Ê 4
=225
Ê 1
Ê 2
Ê 3
Ê 4
=92
Ê 1
Ê 2
Ê 3
Ê 4
=101
4Ŵ 2
4-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
W3 :
4P̂
4Ŵ 3
Ê 1
E1 :
3P̂
Ŵ 1
Ŵ 2
Ŵ 3
3 Ê 1
E2 :
3P̂
Ŵ 1
Ŵ 2
Ŵ 3
E3 :
3P̂
Ŵ 1
Ŵ 2
Ŵ 3
E4 :
3P̂
Ŵ 1
Ŵ 2
Ŵ 3
¦Wˆ ¦ Eˆ
Applying the constraints
i
Ê 2
Ê 3
Ê 4
=32
=34
3 Ê 2
=50
3 Ê 3
=36
3 Ê 4
=105
0 , we obtain:
j
129
225
51
78
89
25
81
, Ŵ 1
, Ŵ 2
, Ŵ 3
, Ê1
, Ê 2
, Ê 3
, Ê 4 195
12
12
12
12
12
12
12
12
§ 225 ·
§ 51 ·
§ 78 ·
§ 129 ·
§ 89 ·
§ 25 ·
R P ,W , E ¨
¸ 225 ¨ ¸ 92 ¨ ¸ 101 ¨
¸ 32 ¨
¸ 34 ¨
¸ 50
12
12
12
12
12
©
¹
© ¹
© ¹
©
¹
©
¹
© 12 ¹
§ 81 ·
§ 195 ·
¨
¸ 36 ¨
¸ 105
© 12 ¹
© 12 ¹
P̂
¦¦ y
2
ij
¦¦ y
6081 , SS E
Model Restricted to W i
2
ij
R P ,W , E
51.83
0:
P:
12P̂
3 Ê 1
E1 :
3P̂
3 Ê 1
E2 :
3P̂
E3 :
3P̂
E4 :
3P̂
Applying the constraint
6081 6029.17
¦ Ê
j
3 Ê 2
3 Ê 3
3 Ê 4
=225
=34
3 Ê 2
=50
3 Ê 3
=36
3 Ê 4
=105
0 , we obtain:
225 ˆ
25
81
, E1 89 / 12 , Ê 2
, Ê 3
, Ê 4 195 . Now:
12
12
12
12
§ 195 ·
§ 225 ·
§ 89 ·
§ 25 ·
§ 81 ·
R P,E ¨
¸ 105 5325.67
¸ 225 ¨
¸ 34 ¨
¸ 50 ¨
¸ 36 ¨
© 12 ¹
© 12 ¹
© 12 ¹
© 12 ¹
© 12 ¹
R W P , E R P ,W , E R P , E 6029.17 5325.67 703.50 SS Treatments
P̂
Model Restricted to E j
0:
P:
W1 :
W2 :
W3 :
Applying the constraint
P̂
225
, Ŵ 1
12
¦ Ŵ
51
, Ŵ 2
12
i
12P̂
4P̂
4Ŵ 1
4Ŵ 1
4P̂
4P̂
4Ŵ 2
4Ŵ 2
4Ŵ 3
0 , we obtain:
78
, Ŵ 3
12
4Ŵ 3
129
12
4-12
=225
=92
=101
=32
6029.17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
§ 225 ·
§ 51 ·
§ 78 ·
§ 129 ·
¨
¸ 225 ¨ ¸ 92 ¨ ¸ 101 ¨
¸ 32 4922.25
© 12 ¹
© 12 ¹
© 12 ¹
© 12 ¹
R E P ,W R P ,W , E R P ,W 6029.17 4922.25 1106.92 SS Blocks
R P ,W
4-9 Assuming that chemical types and bolts are fixed, estimate the model parameters Wi and Ej in
Problem 4-1.
Using Equations 4-14, Applying the constraints, we obtain:
35
, Ŵ 1
20
P̂
23
, Ŵ 2
20
7
, Ŵ 3
20
13
, Ŵ 4
20
17
, Ê
20 1
35
, Ê 2
20
65
, Ê 3
20
75
, Ê 4
20
20
, Ê 5
20
65
20
4-10 Draw an operating characteristic curve for the design in Problem 4-2. Does this test seem to be
sensitive to small differences in treatment effects?
Assuming that solution type is a fixed factor, we use the OC curve in appendix V. Calculate
)
¦W
b
2
2
i
aV 2
¦W
4
2
i
3 8.69
using MSE to estimate V2. We have:
X1
If
¦ Wˆ
2
i
V2
¦ Wˆ
i
2V 2
a 1 b 1
2 3
6.
MSE , then:
)
If
X2
a 1 2
4
31
1.15 and E # 0.70
2MS E , then:
)
4
32
1.63 and E # 0.55 , etc.
This test is not very sensitive to small differences.
4-11 Suppose that the observation for chemical type 2 and bolt 3 is missing in Problem 4-1. Analyze the
problem by estimating the missing value. Perform the exact analysis and compare the results.
y 23 is missing. ŷ 23
ay '2. by '.3 y '..
a 1 b 1
4 282 5 227 1360
4 3
Thus, y2.=357.25, y.3=3022.25, and y..=1435.25
Source
Chemicals
SS
12.7844
4-13
DF
3
MS
4.2615
F0
2.154
75.25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Bolts
Error
Total
158.8875
21.7625
193.4344
4
11
18
1.9784
F0.10,3,11=2.66, Chemicals are not significant.
4-12 Two missing values in a randomized block. Suppose that in Problem 4-1 the observations for
chemical type 2 and bolt 3 and chemical type 4 and bolt 4 are missing.
(a) Analyze the design by iteratively estimating the missing values as described in Section 4-1.3.
4 y'2. 5 y'.3 y'..
and ŷ44
12
ŷ23
4 y'4. 5 y'.4 y'..
12
0
Data is coded y-70. As an initial guess, set y23 equal to the average of the observations available for
0
chemical 2. Thus, y 23
2
4
0.5 . Then ,
4 8 5 6 25.5
3.04
12
4 2 5 17 28.04
ŷ 123
5.41
12
4 8 5 6 30.41
ŷ 144
2.63
12
4 2 5 17 27.63
2
ŷ 44
5.44
12
4 8 5 6 30.44
2
ŷ 44
2.63
12
? ŷ23 5.44 ŷ 44 2.63
0
ŷ 44
Design Expert Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
156.83
4
39.21
Model
9.59
3
3.20
A
9.59
3
3.20
Residual
18.41
12
1.53
Cor Total
184.83
19
F
Value
Prob > F
2.08
2.08
0.1560
0.1560
not significant
(b) Differentiate SSE with respect to the two missing values, equate the results to zero, and solve for
estimates of the missing values. Analyze the design using these two estimates of the missing values.
SS E
SS E
From
wSSE
wy23
wSS E
wy44
¦¦ y
2
ij
15
¦y
2
i.
14
¦y
2
.j
1
20
¦y
2
..
2
2
0.6 y44
6.8 y23 3.7 y44 0.1y23 y44 R
0.6 y23
0 , we obtain:
4-14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1.2 ŷ23 0.1ŷ44
6.8
0.1ŷ23 1.2 ŷ44
3. 7
ŷ23
5.45 , ŷ 44
2.63
These quantities are almost identical to those found in part (a). The analysis of variance using these new
data does not differ substantially from part (a).
(c) Derive general formulas for estimating two missing values when the observations are in different
blocks.
2
y iu2 y kv
SS E
From
wSSE
wy23
wSS E
wy44
c
y ic. y iu
2
2
y ck . y kv
y c.u y iu
b
2
y c.v y kv
a
2
y ..c y iu y
2
kv
ab
0 , we obtain:
ay' i . by' . j y' ..
ª a 1 b 1 º
ŷ iu «
»
ab
¬
¼
ab
ª ( a 1 )( b 1 ) º
ŷkv «
»
ab
¬
¼
ŷ kv
ab
ay'k . by'.v y'.. ŷiu
ab
ab
whose simultaneous solution is:
ŷ iu
ŷ kv
>
y' i . a 1 a 1
2
@
>
b 1 2 ab y' .u b 1 a 1
2
>
@
>
b 1 2 ab y' .. 1 ab a 1 2 b 1
a 1 b 1 1 a 1
2
b 1
2
@
2
b 1 2
@
ab>ay' k . by' .v y' .. @
>1 a 1
ay' i . by' .u y' .. b 1 a 1 >ay' k . by' .v y' .. @
>1 a 1
2
2
b 1 2
@
@
(d) Derive general formulas for estimating two missing values when the observations are in the same
block. Suppose that two observations yij and ykj are missing, izk (same block j).
SS E
From
wSS E
wy 23
wSS E
wy 44
yij2
y kj2
yic. yij
2
yck . y kj
2
b
y.cj yij ykj
a
0 , we obtain
ŷij
ŷkj
ayic. by.cj y..c
a 1 b 1
aykc . by.cj y..c
a 1 b 1
ŷ kj a 1 b 1 2
ŷij a 1 b 1 2
whose simultaneous solution is:
4-15
2
yc.. yij y kj
ab
2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ŷij
ayic. by.cj y..c
a 1 b 1
ŷkj
>
b 1 ay kc . by.cj y ..c a 1 b 1 2 ayic. by.cj y ..c
>1 a 1
aykc . by.c j y..c b 1
2
>
2
>
b 1
@
a 1 ayic. by.c j y..c
2
a 1 b 1 1 a 1 b 1
4
@
2
@
@
4-13 An industrial engineer is conducting an experiment on eye focus time. He is interested in the effect
of the distance of the object from the eye on the focus time. Four different distances are of interest. He has
five subjects available for the experiment. Because there may be differences among individuals, he decides
to conduct the experiment in a randomized block design. The data obtained follow. Analyze the data from
this experiment (use D = 0.05) and draw appropriate conclusions.
Distance (ft)
4
6
8
10
1
10
7
5
6
Design Expert Output
Response:
Focus Time
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
36.30
4
9.07
Model
32.95
3
10.98
A
32.95
3
10.98
Residual
15.30
12
1.27
Cor Total
84.55
19
Subject
3
6
6
3
4
2
6
6
3
4
4
6
1
2
2
F
Value
Prob > F
8.61
8.61
0.0025
0.0025
The Model F-value of 8.61 implies the model is significant. There is only
a 0.25% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.13
4.85
23.28
42.50
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.6829
0.6036
0.1192
10.432
Treatment Means (Adjusted, If Necessary)
Estimated
Standard
Mean
Error
1-4
6.80
0.50
2-6
5.20
0.50
3-8
3.60
0.50
4-10
3.80
0.50
Treatment
1 vs 2
1 vs 3
1 vs 4
2 vs 3
2 vs 4
3 vs 4
Mean
Difference
1.60
3.20
3.00
1.60
1.40
-0.20
DF
1
1
1
1
1
1
Standard
Error
0.71
0.71
0.71
0.71
0.71
0.71
t for H0
Coeff=0
2.24
4.48
4.20
2.24
1.96
-0.28
Prob > |t|
0.0448
0.0008
0.0012
0.0448
0.0736
0.7842
Distance has a statistically significant effect on mean focus time.
4-16
5
6
6
5
3
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4-14 The effect of five different ingredients (A, B, C, D, E) on reaction time of a chemical process is
being studied. Each batch of new material is only large enough to permit five runs to be made.
Furthermore, each runs requires approximately 1 1/2 hours, so only five runs can be made in one day. The
experimenter decides to run the experiment as a Latin square so that day and batch effects can be
systematically controlled. She obtains the data that follow. Analyze the data from this experiment (use D =
0.05) and draw conclusions.
Batch
1
2
3
4
5
1
A=8
C=11
B=4
D=6
E=4
2
B=7
E=2
A=9
C=8
D=2
Day
3
D=1
A=7
C=10
E=6
B=3
4
C=7
D=3
E=1
B=6
A=8
5
E=3
B=8
D=5
A=10
C=8
Minitab Output
General Linear Model
Factor
Type Levels Values
Batch
random
5 1 2 3 4 5
Day
random
5 1 2 3 4 5
Catalyst fixed
5 A B C D E
Analysis of Variance for Time, using Adjusted SS for Tests
Source
Catalyst
Batch
Day
Error
Total
DF
4
4
4
12
24
Seq SS
141.440
15.440
12.240
37.520
206.640
Adj SS
141.440
15.440
12.240
37.520
Adj MS
35.360
3.860
3.060
3.127
F
11.31
1.23
0.98
P
0.000
0.348
0.455
4-15 An industrial engineer is investigating the effect of four assembly methods (A, B, C, D) on the
assembly time for a color television component. Four operators are selected for the study. Furthermore,
the engineer knows that each assembly method produces such fatigue that the time required for the last
assembly may be greater than the time required for the first, regardless of the method. That is, a trend
develops in the required assembly time. To account for this source of variability, the engineer uses the
Latin square design shown below. Analyze the data from this experiment (D = 0.05) draw appropriate
conclusions.
Order of
Assembly
1
2
3
4
1
C=10
B=7
A=5
D=10
2
D=14
C=18
B=10
A=10
Operator
3
A=7
D=11
C=11
B=12
4
B=8
A=8
D=9
C=14
Minitab Output
General Linear Model
Factor
Type Levels Values
Order
random
4 1 2 3 4
Operator random
4 1 2 3 4
Method
fixed
4 A B C D
Analysis of Variance for Time, using Adjusted SS for Tests
Source
DF
Seq SS
Adj SS
Adj MS
4-17
F
P
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Method
Order
Operator
Error
Total
3
3
3
6
15
72.500
18.500
51.500
10.500
153.000
72.500
18.500
51.500
10.500
24.167
6.167
17.167
1.750
13.81
3.52
9.81
0.004
0.089
0.010
4-16 Suppose that in Problem 4-14 the observation from batch 3 on day 4 is missing. Estimate the
missing value from Equation 4-24, and perform the analysis using this value.
y 354 is missing. ŷ 354
>
@
5>28 15 24@ 2 146
3 4
p y ic.. y .cj . y ..c k 2 y ...c
p 2 p 1
3.58
Minitab Output
General Linear Model
Factor
Type Levels Values
Batch
random
5 1 2 3 4 5
Day
random
5 1 2 3 4 5
Catalyst fixed
5 A B C D E
Analysis of Variance for Time, using Adjusted SS for Tests
Source
Catalyst
Batch
Day
Error
Total
DF
4
4
4
12
24
Seq SS
128.676
16.092
8.764
34.317
187.849
Adj SS
128.676
16.092
8.764
34.317
Adj MS
32.169
4.023
2.191
2.860
F
11.25
1.41
0.77
P
0.000
0.290
0.567
4-17 Consider a p x p Latin square with rows (Di), columns (Ek), and treatments (Wj) fixed. Obtain least
squares estimates of the model parameters Di, Ek, Wj.
¦
Dˆ i p
¦
¦
Wˆ j p
¦ Eˆ
k
yi .. , i 1,2 ,..., p
p
p
Dˆ i pWˆ j p
p
p
¦
¦ Eˆ
k
y. j . , j 1,2 ,..., p
pEˆ k
y..k , k 1,2 ,..., p
k 1
i 1
E k : pPˆ p
y...
k
k 1
k 1
j 1
¦
¦ Eˆ
p
p
W j : pPˆ p
Wˆ j p
j 1
i 1
D i : pPˆ pDˆ i p
p
p
p
P : p 2 Pˆ p
Dˆ i p
¦Wˆ
j
j 1
i 1
There are 3p+1 equations in 3p+1 unknowns. The rank of the system is 3p-2. Three side conditions are
¦
i 1
Pˆ
Dˆ i
p
p
p
necessary. The usual conditions imposed are:
Dˆ i
¦
j 1
Wˆ j
¦ Eˆ
k 1
y...
y...
p2
yi .. y... , i 1, 2,..., p
4-18
k
0 . The solution is then:
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Wˆ j
y. j . y... , j 1, 2,..., p
Eˆ k
yi.. y... , k 1, 2,..., p
4-18 Derive the missing value formula (Equation 4-24) for the Latin square design.
SS E
¦¦¦
2
yijk
¦
yi2..
p
¦
y.2j .
p
¦
§ y2 ·
y..2k
2¨ ...2 ¸
¨p ¸
p
©
¹
Let yijk be missing. Then
SS E
2
y ijk
2
y ic.. y ijk
y c. j . y ijk
p
p
where R is all terms without yijk.. From
y ijk
2
y ..c k y ijk
wSS E
wy ijk
p 1 p 2
p y' i .. y' . j . y' ..k 2 y' ...
p2
p2
p
2
2 y ...c y ijk
p2
R
0 , we obtain:
, or y ijk
p y' i .. y' . j . y' ..k 2 y' ...
p 1 p 2
4-19 Designs involving several Latin squares. [See Cochran and Cox (1957), John (1971).] The p x p
Latin square contains only p observations for each treatment. To obtain more replications the experimenter
may use several squares, say n. It is immaterial whether the squares used are the same are different. The
appropriate model is
y ijkh
P U h D i( h ) W j E k ( h ) ( WU ) jh H ijkh
i
°
°j
®
°k
°¯ h
1,2,..., p
1,2,..., p
1,2,..., p
1,2,...,n
where yijkh is the observation on treatment j in row i and column k of the hth square. Note that D i ( h ) and
E k ( h ) are row and column effects in the hth square, and Uh is the effect of the hth square, and ( WU ) jh is the
interaction between treatments and squares.
(a) Set up the normal equations for this model, and solve for estimates of the model parameters. Assume
Û h 0 ,
D̂ i h 0 , and
Ê k h 0
that appropriate side conditions on the parameters are
for each h,
¦
j
Ŵ j
0,
¦
¦
0 for each h, and ¦ WˆU
¦
h
j
WˆU
jh
h
4-19
jh
i
0 for each j.
¦
k
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pˆ y ....
Uˆ h y ...h y ....
Wˆ j y. j .. y ....
Dˆ i( h )
yi ..h y ...h
Eˆ k ( h )
y ..kh y...h
§ ·
¨¨WU ¸¸
© ¹ jh
^
y . j .h y . j .. y...h y ....
(b) Write down the analysis of variance table for this design.
Source
SS
DF
y.2j ..
2
y....
2
p-1
2
y...2 h y....
p 2 np 2
n-1
¦ np np
Treatments
¦
Squares
Treatment x Squares
Rows
¦
y.2j .h
¦
yi2..h y...2 h
2
p
np
p
2
y....
SSTreatments SSSquares
np 2
n(p-1)
y..2kh y...2 h
2
p
np
subtraction
¦
Columns
Error
n(p-1)
n(p-1)(p-2)
¦¦¦¦ y
2
ijkh
Total
(p-1)(n-1)
y2
....2
np
np2-1
4-20 Discuss how the operating characteristics curves in the Appendix may be used with the Latin square
design.
For the fixed effects model use:
¦ pW ¦ W
V
pV
2
j
)2
2
2
j
2
, X1
p 1
X2
p 2 p 1
For the random effects model use:
O
1
pV W2
, X1
V2
p 1
X2
p 2 p 1
4-21 Suppose that in Problem 4-14 the data taken on day 5 were incorrectly analyzed and had to be
discarded. Develop an appropriate analysis for the remaining data.
Two methods of analysis exist: (1) Use the general regression significance test, or (2) recognize that the
design is a Youden square. The data can be analyzed as a balanced incomplete block design with a=b=5,
r=k=4 and O=3. Using either approach will yield the same analysis of variance.
4-20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Minitab Output
General Linear Model
Factor
Type Levels Values
Catalyst fixed
5 A B C D E
Batch
random
5 1 2 3 4 5
Day
random
4 1 2 3 4
Analysis of Variance for Time, using Adjusted SS for Tests
Source
Catalyst
Batch
Day
Error
Total
DF
4
4
3
8
19
Seq SS
119.800
11.667
6.950
32.133
170.550
Adj SS
120.167
11.667
6.950
32.133
Adj MS
30.042
2.917
2.317
4.017
F
7.48
0.73
0.58
P
0.008
0.598
0.646
4-22 The yield of a chemical process was measured using five batches of raw material, five acid
concentrations, five standing times, (A, B, C, D, E) and five catalyst concentrations (D, E, J, G, H). The
Graeco-Latin square that follows was used. Analyze the data from this experiment (use D = 0.05) and draw
conclusions.
Batch
1
2
3
4
5
1
AD=26
BJ=18
CH=20
DE=15
EG=10
2
BE=16
CG=21
DD=12
EJ=15
AH=24
Acid
3
CJ=19
DH=18
EE=16
AG=22
BD=17
Concentration
4
DG=16
ED=11
AJ=25
BH=14
CE=17
5
EH=13
AE=21
BG=13
CD=17
DJ=14
General Linear Model
Factor
Type Levels Values
Time
fixed
5 A B C D
Catalyst random
5 a b c d
Batch
random
5 1 2 3 4
Acid
random
5 1 2 3 4
E
e
5
5
Analysis of Variance for Yield, using Adjusted SS for Tests
Source
Time
Catalyst
Batch
Acid
Error
Total
DF
4
4
4
4
8
24
Seq SS
342.800
12.000
10.000
24.400
46.800
436.000
Adj SS
342.800
12.000
10.000
24.400
46.800
Adj MS
85.700
3.000
2.500
6.100
5.850
F
14.65
0.51
0.43
1.04
P
0.001
0.729
0.785
0.443
4-23 Suppose that in Problem 4-15 the engineer suspects that the workplaces used by the four operators
may represent an additional source of variation. A fourth factor, workplace (D, E, J, G) may be introduced
and another experiment conducted, yielding the Graeco-Latin square that follows. Analyze the data from
this experiment (use D = 0.05) and draw conclusions.
Order of
Assembly
1
2
3
1
CE=11
BD=8
AG=9
2
BJ=10
CG=12
DD=11
4-21
Operator
3
DG=14
AJ=10
BE=7
4
AD=8
DE=12
CJ=15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DJ=9
4
AE=8
CD=18
BG=6
Minitab Output
General Linear Model
Factor
Type Levels Values
Method
fixed
4 A B C D
Order
random
4 1 2 3 4
Operator random
4 1 2 3 4
Workplac random
4 a b c d
Analysis of Variance for Time, using Adjusted SS for Tests
Source
Method
Order
Operator
Workplac
Error
Total
DF
3
3
3
3
3
15
Seq SS
95.500
0.500
19.000
7.500
27.500
150.000
Adj SS
95.500
0.500
19.000
7.500
27.500
Adj MS
31.833
0.167
6.333
2.500
9.167
F
3.47
0.02
0.69
0.27
P
0.167
0.996
0.616
0.843
However, there are only three degrees of freedom for error, so the test is not very sensitive.
4-24 Construct a 5 x 5 hypersquare for studying the effects of five factors. Exhibit the analysis of
variance table for this design.
Three 5 x 5 orthogonal Latin Squares are:
ABCDE
BCDEA
CDEAB
DEABC
EABCD
DEJGH
JGHDE
HDEJG
EJGHD
GHDEJ
12345
45123
23451
51234
34512
Let rows = factor 1, columns = factor 2, Latin letters = factor 3, Greek letters = factor 4 and numbers =
factor 5. The analysis of variance table is:
Source
Rows
Columns
Latin Letters
Greek Letters
Numbers
Error
Total
DF
4
4
4
4
4
4
24
4-25 Consider the data in Problems 4-15 and 4-23. Suppressing the Greek letters in 4-23, analyze the data
using the method developed in Problem 4-19.
Batch
1
2
3
4
1
C=10
B=7
A=5
D=10
(32)
Square 1 - Operator
2
3
4
D=14
A=7
B=8
C=18
D=11
A=8
B=10
C=11
D=9
A=10
B=12
C=14
(52)
(41)
(36)
4-22
Row Total
(39)
(44)
(35)
(46)
164=y…1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Square 2 - Operator
2
3
4
B=10
D=14
A=8
C=12
A=10
D=12
D=11
B=7
C=15
A=8
C=18
B=6
(41)
(49)
(41)
1
C=11
B=8
A=9
D=9
(37)
Batch
1
2
3
4
Assembly Methods
A
B
C
D
Source
Assembly Methods
Squares
AxS
Assembly Order (Rows)
Operators (columns)
Error
Total
SS
159.25
0.50
8.75
19.00
70.50
45.50
303.50
Row Total
(43)
(42)
(42)
(41)
168=y…2
Totals
y.1..=65
y.2..=68
y.3..=109
y.4..=90
DF
3
1
3
6
6
12
31
MS
53.08
0.50
2.92
3.17
11.75
3.79
F0
14.00*
0.77
Significant at 1%.
4-26 Consider the randomized block design with one missing value in Table 4-7. Analyze this data by
using the exact analysis of the missing value problem discussed in Section 4-1.4. Compare your results to
the approximate analysis of these data given in Table 4-8.
P:
15P
4W1
W1 :
4 P
3P
4W1
W2 :
4 P
4 P
4 P
4 P
3P
4 P
W3 :
W4 :
E1 :
E2 :
E3 :
E4 :
41
, Ŵ 1
36
4W3
4W4
4 E1
4 E2
3E3
4 E4
=17
E1
E
E3
E4
E
=3
1
E2
E
E2
E
E3
E
E4
E
=-2
4W4
E1
E
W4
4 E1
3W2
4W3
W1
W1
W1
W1
Applying the constraints
P̂
3W2
W2
W2
W3
W3
W3
W3
W2
¦ Wˆ ¦ Eˆ
14
, Ŵ 2
36
i
j
24
, Ŵ 3
36
1
2
=1
4
2
3
=15
4
=-4
3E2
W4
W4
=-3
4 E3
W4
=6
4 E4
0 , we obtain:
59
, Ŵ 4
36
94
, Ê1
36
4-23
77
, Ê 2
36
68
, Ê 3
36
24
, Ê 4
36
121
36
=19
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
R P ,W , E
Pˆ y..
4
4
¦ Wˆ y ¦ Eˆ y
i i.
j .j
i 1
138.78
j 1
With 7 degrees of freedom.
¦¦ y
2
ij
¦¦ y
145.00 , SS E
2
ij
R P ,W , E
145.00 138.78
6.22
which is identical to SSE obtained in the approximate analysis. In general, the SSE in the exact and
approximate analyses will be the same.
To test Ho: W i
P:
E1 :
E2 :
15Pˆ
4 P
4 P
E3 :
E4 :
3P
4 P
Applying the constraint
P̂
P E j H ij . The normal equations used are:
0 the reduced model is yij
19
, Ê1
16
¦ Ê
4 Eˆ1
4 Eˆ
4 Eˆ2
3Eˆ3
4 Eˆ4
=17
=-4
1
4 Eˆ2
=-3
3Eˆ3
=6
4 Eˆ4
=18
0 , we obtain:
j
35
, Ê 2
16
31
, Ê 3
16
13
, Ê 4
16
53
. Now R P , E
16
Pˆ y ..
4
¦ Eˆ
j
y.j
99.25
j 1
with 4 degrees of freedom.
R W P,E
R P ,W , E R P , E
138.78 99.25 39.53
with 7-4=3 degrees of freedom. R W P , E is used to test Ho: W i
SS Treatments
0.
The sum of squares for blocks is found from the reduced model y ij
P W i H ij . The normal equations
used are:
Model Restricted to E j
0:
P:
15P
W1 :
4Pˆ
3Pˆ
4Pˆ
4Pˆ
W2 :
W3 :
W4 :
Applying the constraint
¦Wˆ
i
Pˆ
4W1
4W1
3W2
4W3
4W4
=3
3W2
=1
4W3
=-2
4W4
0 , we obtain:
13
, Wˆ1
12
=17
4
, Wˆ2
12
9
, Wˆ3
12
4-24
19
, Wˆ4
12
32
12
=15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
R P ,W
Pˆ y..
4
¦ Wˆ y
i i.
59.83
i 1
with 4 degrees of freedom.
R E P ,W
R P ,W , E R P ,W
138.78 59.83 78.95
SS Blocks
with 7-4=3 degrees of freedom.
Source
Tips
Blocks
Error
Total
DF
3
3
8
14
SS(exact)
39.53
78.95
6.22
125.74
SS(approximate)
39.98
79.53
6.22
125.73
Note that for the exact analysis, SST z SSTips SS Blocks SSE .
4-27 An engineer is studying the mileage performance characteristics of five types of gasoline additives.
In the road test he wishes to use cars as blocks; however, because of a time constraint, he must use an
incomplete block design. He runs the balanced design with the five blocks that follow. Analyze the data
from this experiment (use D = 0.05) and draw conclusions.
Additive
1
2
3
4
5
1
14
14
13
11
2
17
14
11
12
Car
3
14
13
11
10
4
13
13
14
12
5
12
10
9
8
There are several computer software packages that can analyze the incomplete block designs discussed in
this chapter. The Minitab General Linear Model procedure is a widely available package with this
capability. The output from this routine for Problem 4-27 follows. The adjusted sums of squares are the
appropriate sums of squares to use for testing the difference between the means of the gasoline additives.
Minitab Output
General Linear Model
Factor
Type Levels Values
Additive fixed
5 1 2 3 4 5
Car
random
5 1 2 3 4 5
Analysis of Variance for Mileage, using Adjusted SS for Tests
Source
Additive
Car
Error
Total
DF
4
4
11
19
Seq SS
31.7000
35.2333
10.0167
76.9500
Adj SS
35.7333
35.2333
10.0167
Adj MS
8.9333
8.8083
0.9106
F
9.81
9.67
P
0.001
0.001
4-28 Construct a set of orthogonal contrasts for the data in Problem 4-27. Compute the sum of squares
for each contrast.
4-25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One possible set of orthogonal contrasts is:
H 0 : P 4 P5 P1 P 2
H 0 : P1 P2
H 0 : P 4 P5
H 0 : 4P3 P4 P5 P1 P2
(1)
(2)
(3)
(4)
The sums of squares and F-tests are:
Brand ->
Qi
1
33/4
2
11/4
3
-3/4
4
-14/4
5
-27/4
¦ ci Qi
SS
F0
(1)
(2)
(3)
(4)
-1
1
0
-1
-1
-1
0
-1
0
0
0
4
1
0
-1
-1
1
0
1
-1
-85/4
-22/4
-13/4
-15/4
30.10
4.03
1.41
0.19
39.09
5.23
1.83
0.25
Contrasts (1) and (2) are significant at the 1% and 5% levels, respectively.
4-29 Seven different hardwood concentrations are being studied to determine their effect on the strength
of the paper produced. However the pilot plant can only produce three runs each day. As days may differ,
the analyst uses the balanced incomplete block design that follows. Analyze this experiment (use D = 0.05)
and draw conclusions.
Hardwood
Concentration (%)
2
4
6
8
10
12
14
1
114
126
2
120
137
141
3
Days
4
5
120
6
7
117
119
114
129
145
134
149
150
120
143
118
136
123
130
127
There are several computer software packages that can analyze the incomplete block designs discussed in
this chapter. The Minitab General Linear Model procedure is a widely available package with this
capability. The output from this routine for Problem 4-29 follows. The adjusted sums of squares are the
appropriate sums of squares to use for testing the difference between the means of the hardwood
concentrations.
Minitab Output
General Linear Model
Factor
Type Levels Values
Concentr fixed
7 2 4 6 8 10 12 14
Days
random
7 1 2 3 4 5 6 7
Analysis of Variance for Strength, using Adjusted SS for Tests
Source
Concentr
Days
Error
Total
DF
6
6
8
20
Seq SS
2037.62
394.10
168.57
2600.29
Adj SS
1317.43
394.10
168.57
Adj MS
219.57
65.68
21.07
4-26
F
10.42
3.12
P
0.002
0.070
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4-30 Analyze the data in Example 4-6 using the general regression significance test.
P:
12 P
3W1
W1 :
3P
3P
3W1
W2 :
W3 :
W4 :
E1 :
E2 :
E3 :
E4 :
3P
3P
3P
3P
3P
3P
3W2
3W 3
3W 4
3E1
3E2
E1
E2
E
3W2
3W 4
E1
E
W4
3E1
3W 3
W1
W3
W3
W3
W2
W2
W2
W1
W1
3E4
=870
E3
E
E4
E
=218
3
E2
1
P
870 / 12 , W1
E1
7 / 8 , E2
=207
3E3
9 / 8 , W2
24 / 8 , E4
4 / 8 , W 4
20 / 8 ,
0/ 8
with 7 degrees of freedom.
, .00
¦ ¦ yij2 63156
SS E ¦ ¦ y ij2 R ( P , W , E )
To test Ho: W i
P:
12 P
E1 :
E2 :
E3 :
E4 :
3P
3P
3P
3P
Applying the constraint
P̂
¦ Ê
j
870
, Ê1
12
R P,E
63156.00 63152.75 3.25 .
P E j Hij . The normal equations used are:
0 the reduced model is yij
3E1
3E
3E2
3E3
3E4
3E2
=207
3E3
=224
3E4
0 , we obtain:
Pˆ y ..
21
, Ê 3
6
13
, Ê 4
6
4
¦ Eˆ
j
y. j
=870
=221
1
7
, Ê 2
6
63,130.00
j 1
with 4 degrees of freedom.
4-27
=224
3E4
7 / 8 , W3
1
6
=222
=221
3E2
W4
=214
=216
E4
0 , we obtain:
31 / 8 , E3
4
E3
2
W4
¦ Wi ¦ E j
Applying the constraints
3E3
=218
=218
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
R W P,E
R P ,W , E R P , E
63152.75 63130.00
with 7-4=3 degrees of freedom. R W P , E is used to test Ho: W i
22.75
SS Treatments
0.
P W i H ij . The normal equations
The sum of squares for blocks is found from the reduced model y ij
used are:
Model Restricted to E j
0:
P:
12 P
W1 :
3P
3P
3P
3P
W2 :
W3 :
W4 :
3W1
3W1
3W2
3W 3
3W 4
=870
=218
3W2
=214
3W 3
=216
3W 4
=222
The sum of squares for blocks is found as in Example 4-6. We may use the method shown above to find an
adjusted sum of squares for blocks from the reduced model, y ij P W i H ij .
4-31 Prove that
k
¦
a
i 1
Oa
Qi2
is the adjusted sum of squares for treatments in a BIBD.
We may use the general regression significance test to derive the computational formula for the adjusted
treatment sum of squares. We will need the following:
Wˆ i
b
kQi
, kQi
Oa
R P ,W , E
Pˆ y ..
kyi .
¦n y
ij . j
i 1
a
¦
Wˆ i y i .
i 1
b
¦ Eˆ
j
y. j
j 1
and the sum of squares we need is:
R W P ,E
Pˆ y ..
a
¦
Wˆ i y i .
i 1
b
¦
Eˆ j y . j
j 1
a
¦ n Wˆ
ij i
kEˆ j
y. j
i 1
and from this we have:
ky. j Eˆ j
y.2j ky. j Pˆ y. j
a
¦ n Wˆ
ij i
i 1
4-28
¦
j 1
The normal equation for E is, from equation (4-35),
E : kPˆ
b
y .2j
k
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
therefore,
a
ª
º
y
nij Wˆ i
«
.j
2
2 »
a
b
y.j »
« y . j kPˆ y . j
i 1
Pˆ y .. Wˆ i y i .
« k k
k
k »»
i 1
j 1«
«
»
¬
¼
¦
R W P ,E
§
1
Wˆ i ¨ y i .
¨
k
1
©
a
R( W P ,E )
¦
i
a
¦
i 1
¦
¦
·
nij y . j ¸
¸
¹
§Q2
¨ i
k
¨ Oa
i 1 ©
a
a
§ kQ ·
Qi ¨
¸
© Oa ¹
1
¦
i
¦
·
¸ { SS
Treatments ( adjusted )
¸
¹
4-32 An experimenter wishes to compare four treatments in blocks of two runs. Find a BIBD for this
experiment with six blocks.
Treatment
1
2
3
4
Block 1
X
X
Block 2
X
Block 3
X
Block 4
Block 5
X
X
X
X
X
X
Block 6
X
X
Note that the design is formed by taking all combinations of the 4 treatments 2 at a time. The parameters of
the design are O = 1, a=4, b=6, k=3, and r=2
4-33 An experimenter wishes to compare eight treatments in blocks of four runs. Find a BIBD with 14
blocks and O = 3.
The design has parameters a=8, b=14, O = 3, r=2 and k=4. It may be generated from a 23 factorial design
confounded in two blocks of four observations each, with each main effect and interaction successively
confounded (7 replications) forming the 14 blocks. The design is discussed by John (1971, pg. 222) and
Cochran and Cox (1957, pg. 473). The design follows:
Blocks
1
2
3
4
5
6
7
8
9
10
11
12
13
14
1=(I)
X
2=a
3=b
X
X
X
X
4=ab
5=c
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
X
4-29
8=abc
X
X
X
X
7=bc
X
X
X
X
X
X
6=ac
X
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4-34 Perform the interblock analysis for the design in Problem 4-27.
The interblock analysis for Problem 4-27 uses Vˆ 2
0.77 and Vˆ E2
2.14 . A summary of the interblock,
intrablock and combined estimates is:
Parameter
W1
W2
Intrablock
2.20
0.73
-0.20
-0.93
-1.80
W3
W4
W5
Interblock
-1.80
0.20
-5.80
9.20
-1.80
Combined
2.18
0.73
-0.23
-0.88
-1.80
4-35 Perform the interblock analysis for the design in Problem 4-29. The interblock analysis for problem
MSBlocks( adj ) MSE b 1 >65.68 21.07@ 6
4-29 uses Vˆ 2 21.07 and V E2
19.12 . A summary of
72
a r 1
the interblock, intrablock, and combined estimates is give below
>
@
Parameter
W1
W2
W3
W4
W5
W6
W7
Intrablock
-12.43
-8.57
Interblock
-11.79
-4.29
Combined
-12.38
-7.92
2.57
10.71
-8.79
9.21
1.76
10.61
13.71
-5.14
21.21
-22.29
14.67
-6.36
-0.86
10.71
-0.03
4-36 Verify that a BIBD with the parameters a = 8, r = 8, k = 4, and b = 16 does not exist. These
r ( k 1) 8(3) 24
conditions imply that O
, which is not an integer, so a balanced design with these
a 1
7
7
parameters cannot exist.
4-37 Show that the variance of the intra block estimators { W i } is
Note that Wˆ i
kQi
, and Qi
Oa
yi.
1
k
Oa 2
b
b
¦
k ( a 1 )V 2
n ij y . j , and kQi
kyi .
¦
nij y. j
j 1
j 1
.
§
k 1 yi . ¨
¨
©
b
¦n y
ij . j
j 1
·
yi . ¸
¸
¹
y i. contains r observations, and the quantity in the parenthesis is the sum of r(k-1) observations, not
including treatment i. Therefore,
V kQi
k 2V Qi
r k 1 2V 2 r k 1 V 2
4-30
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
or
V Qi
1
k2
>r k 1 V
2
^ k 1 1`@
r k 1 V 2
k
To find V Wˆ i , note that:
2
V Wˆ i
However, since O a 1
§ k ·
¨ ¸ V Q
© Oa ¹
2
§ k · r k 1 2
V
¨ ¸
k
© Oa ¹
i
kr k 1
Oa
2
V2
r k 1 , we have:
k a 1
V Wˆ i
Oa 2
V2
Furthermore, the ^Ŵ i ` are not independent, this is required to show that V Wˆ i Wˆ j
2k 2
V
Oa
4-38 Extended incomplete block designs. Occasionally the block size obeys the relationship a < k < 2a.
An extended incomplete block design consists of a single replicate or each treatment in each block along
with an incomplete block design with k* = k-a. In the balanced case, the incomplete block design will have
parameters k* = k-a, r* = r-b, and O*. Write out the statistical analysis. (Hint: In the extended incomplete
block design, we have O = 2r-b+O*.)
As an example of an extended incomplete block design, suppose we have a=5 treatments, b=5 blocks and
k=9. A design could be found by running all five treatments in each block, plus a block from the balanced
incomplete block design with k* = k-a=9-5=4 and O*=3. The design is:
Block
1
2
3
4
5
Complete Treatment
1,2,3,4,5
1,2,3,4,5
1,2,3,4,5
1,2,3,4,5
1,2,3,4,5
Incomplete Treatment
2,3,4,5
1,2,4,5
1,3,4,5
1,2,3,4
1,2,3,5
Note that r=9, since the augmenting incomplete block design has r*=4, and r= r* + b = 4+5=9, and O = 2rb+O*=18-5+3=16. Since some treatments are repeated in each block it is possible to compute an error sum
of squares between repeat observations. The difference between this and the residual sum of squares is due
to interaction. The analysis of variance table is shown below:
SS
Source
Treatments
(adjusted)
Blocks
Interaction
Error
Total
k
Qi2
¦ aO
y .2j
y ..2
k
N
Subtraction
[SS between repeat observations]
y2
y ij2 ..
N
¦
¦¦
4-31
DF
a-1
b-1
(a-1)(b-1)
b(k-a)
N-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4-32
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 5
Introduction to Factorial Designs
Solutions
5-1 The yield of a chemical process is being studied. The two most important variables are thought to
be the pressure and the temperature. Three levels of each factor are selected, and a factorial experiment
with two replicates is performed. The yield data follow:
Temperature
150
160
170
200
90.4
90.2
90.1
90.3
90.5
90.7
Pressure
215
90.7
90.6
90.5
90.6
90.8
90.9
230
90.2
90.4
89.9
90.1
90.4
90.1
(a) Analyze the data and draw conclusions. Use D = 0.05.
Both pressure (A) and temperature (B) are significant, the interaction is not.
Design Expert Output
Response:Surface Finish
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
1.14
8
A
0.77
2
B
0.30
2
AB
0.069
4
Residual
0.16
9
Lack of Fit
0.000
0
Pure Error
0.16
9
Cor Total
1.30
17
Mean
Square
0.14
0.38
0.15
0.017
0.018
F
Value
8.00
21.59
8.47
0.97
0.018
The Model F-value of 8.00 implies the model is significant. There is only a 0.26% chance that a
"Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Values greater than 0.1000 indicate the model terms are not significant.
If there are many insignificant model terms (not counting those required to support hierarchy),
model reduction may improve your model.
(b) Prepare appropriate residual plots and comment on the model’s adequacy.
The residuals plot show no serious deviations from the assumptions.
5-1
Prob > F
0.0026
0.0004
0.0085
0.4700
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Predicted
Normal plot of residuals
0.15
99
95
90
Res iduals
Normal % probability
0.075
4.26326E-014
-0.075
80
70
50
30
20
10
5
1
-0.15
90.00
90.21
90.43
90.64
90.85
-0.15
Predicted
-0.075
-4.26326E-014
0.075
0.15
Res idual
Residuals vs. Temperature
Residuals vs. Pressure
0.15
0.15
2
2
0.075
0.075
Res iduals
Res iduals
3
4.26326E-014
4.26326E-014
-0.075
-0.075
2
2
-0.15
2
-0.15
1
2
3
1
Temperature
2
Pres sure
(c) Under what conditions would you operate this process?
5-2
3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
T e m p e ra tu re
Yield
9 1 .0 0 0 8
X = A: Pressure
Y = B: Temperature
Design Points
9 0 .7 1 2 9
2
Y ie ld
B1 150
B2 160
B3 170
9 0 .4 2 5
2
9 0 .1 3 7 1
2
8 9 .8 4 9 2
200
215
230
P re s s u re
Pressure set at 215 and Temperature at the high level, 170 degrees C, give the highest yield.
The standard analysis of variance treats all design factors as if they were qualitative. In this case, both
factors are quantitative, so some further analysis can be performed. In Section 5-5, we show how response
curves and surfaces can be fit to the data from a factorial experiment with at least one quantative factor.
Since both factors in this problem are quantitative and have three levels, we can fit linear and quadratic
effects of both temperature and pressure, exactly as in Example 5-5 in the text. The Design-Expert
output, including the response surface plots, now follows.
Design Expert Output
Response:Surface Finish
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
1.13
5
A
0.10
1
B
0.067
1
A2
0.67
1
B2
0.23
1
AB
0.061
1
Residual
0.17
12
Lack of Fit
7.639E-003
3
Pure Error
0.16
9
Cor Total 1.30
17
Mean
Square
0.23
0.10
0.067
0.67
0.23
0.061
0.014
2.546E-003
0.018
F
Value
16.18
7.22
4.83
47.74
16.72
4.38
Prob > F
< 0.0001
0.0198
0.0483
< 0.0001
0.0015
0.0582
0.14
0.9314
The Model F-value of 16.18 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, A2, B2 are significant model terms.
Values greater than 0.1000 indicate the model terms are not significant.
If there are many insignificant model terms (not counting those required to support hierarchy),
model reduction may improve your model.
Std. Dev.
Mean
C.V.
PRESS
0.12
90.41
0.13
0.42
Coefficient
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.8708
0.8170
0.6794
11.968
Standard
5-3
95% CI
95% CI
significant
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Intercept
A-Pressure
B-Temperature
A2
B2
AB
Estimate
90.52
-0.092
0.075
-0.41
0.24
-0.087
DF
1
1
1
1
1
1
Error
0.062
0.034
0.034
0.059
0.059
0.042
Low
90.39
-0.17
6.594E-004
-0.54
0.11
-0.18
High
90.66
-0.017
0.15
-0.28
0.37
3.548E-003
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Yield
+90.52
-0.092
+0.075
-0.41
+0.24
-0.087
=
*A
*B
* A2
* B2
*A*B
Final Equation in Terms of Actual Factors:
Yield
+48.54630
+0.86759
-0.64042
-1.81481E-003
+2.41667E-003
-5.83333E-004
1 7 0 .0 0
=
* Pressure
* Temperature
* Pressure2
* Temperature2
* Pressure * Temperature
Yie2 ld
2
2
90.8
90.7
90.6
91
90.8
90.2
1 6 0 .0 0
90.5
2
2
90.5
90.4 90.3
90.6
90.12
90.4
90.4
Yie ld
B: Tem perature
1 6 5 .0 0
90.3
90.2
90
1 5 5 .0 0
170.00
90.6
2
2
200.00
2
207.50
1 5 0 .0 0
2 0 0 .0 0
2 0 7 .5 0
2 1 5 .0 0
2 2 2 .5 0
165.00
2 3 0 .0 0
160.00
215.00
B: Tem perature
155.00
222.50
A: Pre s s ure
A: Pre s s ure
230.00
150.00
5-2 An engineer suspects that the surface finish of a metal part is influenced by the feed rate and the
depth of cut. She selects three feed rates and four depths of cut. She then conducts a factorial experiment
and obtains the following data:
Feed Rate (in/min)
0.20
0.15
74
64
60
Depth of
0.18
79
68
73
Cut (in)
0.20
82
88
92
0.25
99
104
96
92
98
99
104
5-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0.25
0.30
86
88
104
88
108
95
110
99
99
98
102
104
99
95
108
110
99
114
111
107
(a) Analyze the data and draw conclusions. Use D = 0.05.
The depth (A) and feed rate (B) are significant, as is the interaction (AB).
Design Expert Output
Response:
Surface Finish
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Squares
DF
5842.67
11
2125.11
3
2
1580.25
6
92.84
689.33
24
0.000
0
689.33
24
6532.00
35
Source
Model
A
B 3160.50
AB 557.06
Residual
Lack of Fit
Pure Error
Cor Total
Mean
Square
531.15
708.37
55.02
3.23
28.72
F
Value
18.49
24.66
< 0.0001
0.0180
Prob > F
< 0.0001
< 0.0001
significant
28.72
The Model F-value of 18.49 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
(b) Prepare appropriate residual plots and comment on the model’s adequacy.
The residual plots shown indicate nothing unusual.
Residuals vs. Predicted
Normal plot of residuals
8
99
95
Res iduals
Normal % probability
3.83333
-0.333333
-4.5
90
80
70
50
30
20
10
5
1
-8.66667
66.00
77.17
88.33
99.50
110.67
-8.66667
Predicted
-4.5
-0.333333
Res idual
5-5
3.83333
8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Feed Rate
8
Residuals vs. Depth of Cut
8
2
3.83333
3.83333
Res iduals
Res iduals
2
-0.333333
-0.333333
-4.5
-4.5
-8.66667
-8.66667
1
2
3
1
2
Feed Rate
3
4
Depth of Cut
(c) Obtain point estimates of the mean surface finish at each feed rate.
Feed Rate
0.20
0.25
0.30
DESIGN-EXPERT Plot
Average
81.58
97.58
103.83
O n e F a c to r P lo t
Surface Finish
114
W a rn in g ! F a c to r in v o lv e d in a n in te ra c tio n .
X = B: Feed Rate
S u rfa c e F in is h
Actual Factor
A: Depth of Cut = Average
1 0 0 .5
87
7 3 .5
60
0 .2 0
0 .2 5
0 .3 0
F e e d R a te
(d) Find P-values for the tests in part (a).
The P-values are given in the computer output in part (a).
5-3 For the data in Problem 5-2, compute a 95 percent interval estimate of the mean difference in
response for feed rates of 0.20 and 0.25 in/min.
5-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
We wish to find a confidence interval on P1 P2 , where P1 is the mean surface finish for 0.20 in/min and
P2 is the mean surface finish for 0.25 in/min.
y1.. y 2.. tD
2,ab n 1
2MS E
d P1 P 2 d y1.. y 2.. tD
n
2,ab n 1 )
2MS E
n
2(28.7222)
16 r 9.032
3
Therefore, the 95% confidence interval for P1 P2 is -16.000 r 9.032.
(81.5833 97.5833) r (2.064)
5-4 An article in Industrial Quality Control (1956, pp. 5-8) describes an experiment to investigate the
effect of the type of glass and the type of phosphor on the brightness of a television tube. The response
variable is the current necessary (in microamps) to obtain a specified brightness level. The data are as
follows:
Glass
Type
1
2
1
280
290
285
Phosphor Type
2
300
310
295
3
290
285
290
230
235
240
260
240
235
220
225
230
(a) Is there any indication that either factor influences brightness? Use D = 0.05.
Both factors, phosphor type (A) and Glass type (B) influence brightness.
Design Expert Output
Response: Current in microamps
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Source
Model
A
B
AB
Residual
Lack of Fit
Pure Error
Cor Total16150.00
Sum of
Squares
15516.67
933.33
14450.00
133.33
633.33
0.000
633.33
17
DF
5
2
1
2
12
0
12
Mean
Square
3103.33
466.67
14450.00
66.67
52.78
52.78
The Model F-value of 58.80 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
(b) Do the two factors interact? Use D = 0.05.
There is no interaction effect.
(c) Analyze the residuals from this experiment.
5-7
F
Value
58.80
8.84
273.79
1.26
Prob > F
< 0.0001
0.0044
< 0.0001
0.3178
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The residual plot of residuals versus phosphor content indicates a very slight inequality of variance. It is
not serious enough to be of concern, however.
Residuals vs. Predicted
Normal plot of residuals
15
99
95
2.5
90
Normal % probability
Res iduals
8.75
2
-3.75
80
70
50
30
20
10
5
1
-10
225.00
244.17
263.33
282.50
301.67
-10
-3.75
Predicted
15
Residuals vs. Phosphor Type
15
15
8.75
8.75
Res iduals
Res iduals
8.75
Res idual
Residuals vs. Glass Type
2.5
2.5
2
2
-3.75
2.5
2
2
-3.75
2
-10
-10
1
2
1
Glas s Type
2
3
Phosphor Type
5-5 Johnson and Leone (Statistics and Experimental Design in Engineering and the Physical Sciences,
Wiley 1977) describe an experiment to investigate the warping of copper plates. The two factors studies
were the temperature and the copper content of the plates. The response variable was a measure of the
amount of warping. The data were as follows:
Temperature (°C)
50
75
100
125
40
17,20
12,9
16,12
21,17
Copper
60
16,21
18,13
18,21
23,21
5-8
Content (%)
80
24,22
17,12
25,23
23,22
100
28,27
27,31
30,23
29,31
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Is there any indication that either factor affects the amount of warping? Is there any interaction
between the factors? Use D = 0.05.
Both factors, copper content (A) and temperature (B) affect warping, the interaction does not.
Design Expert Output
Response: Warping
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
968.22
15
A
698.34
3
B
156.09
3
AB
113.78
9
Residual
108.50
16
Lack of Fit
0.000
0
Pure Error
108.50
16
Cor Total
1076.72
31
Mean
Square
64.55
232.78
52.03
12.64
6.78
F
Value
9.52
34.33
7.67
1.86
Prob > F
< 0.0001
< 0.0001
0.0021
0.1327
significant
6.78
The Model F-value of 9.52 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
(b) Analyze the residuals from this experiment.
There is nothing unusual about the residual plots.
Residuals vs. Predicted
Normal plot of residuals
3.5
99
1.75
90
80
70
Res iduals
Normal % probability
95
1.06581E-014
50
30
20
10
-1.75
5
1
-3.5
-3.5
-1.75
-1.06581E-014
1.75
3.5
10.50
Res idual
15.38
20.25
Predicted
5-9
25.13
30.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Copper Content
Residuals vs. Temperature
3.5
1.75
3.5
1.75
2
Res iduals
Res iduals
2
1.06581E-014
1.06581E-014
2
-1.75
2
-1.75
-3.5
-3.5
1
2
3
4
1
Copper Content
2
3
4
Temperature
(c) Plot the average warping at each level of copper content and compare them to an appropriately scaled
t distribution. Describe the differences in the effects of the different levels of copper content on
warping. If low warping is desirable, what level of copper content would you specify?
Design Expert Output
Factor
Name
A
Copper Content
B
Temperature
Level
40
Average
Low Level
40
50
High Level
100
125
Prediction
Warping15.50
SE Mean
1.84
95% CI low
11.60
95% CI high
19.40
Factor
A
B
Level
60
Average
Low Level
40
50
High Level
100
125
Prediction
Warping18.88
SE Mean
1.84
95% CI low
14.97
95% CI high
22.78
Factor
A
B
Level
80
Average
Low Level
40
50
High Level
100
125
Prediction
Warping21.00
SE Mean
1.84
95% CI low
17.10
95% CI high
24.90
Factor
A
B
Level
100
Average
Low Level
40
50
High Level
100
125
SE Mean
1.84
95% CI low
24.35
95% CI high
32.15
Name
Copper Content
Temperature
Name
Copper Content
Temperature
Name
Copper Content
Temperature
Prediction
Warping28.25
Use a copper content of 40 for the lowest warping.
S
MS E
b
6.78125
4
5-10
1.3
SE Pred
3.19
95% PI low
8.74
95% PI high
22.26
SE Pred
3.19
95% PI low
12.11
95% PI high
25.64
SE Pred
3.19
95% PI low
14.24
95% PI high
27.76
SE Pred
3.19
95% PI low
21.49
95% PI high
35.01
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S c a le d t D is tr ib u ti o n
C u= 4 0
C u= 6 0
1 5 .0
C u= 8 0
1 8 .0
2 1 .0
C u= 1 0 0
2 4 .0
2 7 .0
W a rp i n g
(d) Suppose that temperature cannot be easily controlled in the environment in which the copper plates
are to be used. Does this change your answer for part (c)?
Use a copper of content of 40. This is the same as for part (c).
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
T e m p e ra tu re
Warping
3 2 .7 6 0 2
2
X = A: Copper Content
Y = B: Temperature
Design Points
50
75
100
125
2
2
3
W a rp in g
B1
B2
B3
B4
2
2 6 .5 0 5 1
2 0 .2 5
2
2
1 3 .9 9 4 9
2
7 .7 3 9 7 9
40
60
80
100
C o p p e r C o n te n t
5-6 The factors that influence the breaking strength of a synthetic fiber are being studied. Four
production machines and three operators are chosen and a factorial experiment is run using fiber from the
same production batch. The results are as follows:
Operator
1
1
109
110
2
110
115
Machine
3
108
109
4
110
108
2
110
112
110
111
111
109
114
112
3
116
112
114
120
5-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
114
115
119
117
(a) Analyze the data and draw conclusions. Use D = 0.05.
Only the Operator (A) effect is significant.
Design Expert Output
Response:Stength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
217.46
11
A
160.33
2
B
12.46
3
AB
44.67
6
Residual
45.50
12
Lack of Fit
0.000
0
Pure Error
45.50
12
Cor Total
262.96
23
Mean
Square
19.77
80.17
4.15
7.44
3.79
F
Value
5.21
21.14
1.10
1.96
Prob > F
0.0041
0.0001
0.3888
0.1507
significant
3.79
The Model F-value of 5.21 implies the model is significant.
There is only a 0.41% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms aresignificant.
In this case A are significant model terms.
(b) Prepare appropriate residual plots and comment on the model’s adequacy.
The residual plot of residuals versus predicted shows that variance increases very slightly with strength.
There is no indication of a severe problem.
Residuals vs. Predicted
Normal plot of residuals
2.5
99
95
Res iduals
Normal % probability
1.25
7.81597E-014
-1.25
90
80
70
50
30
20
10
5
1
-2.5
108.50
111.00
113.50
116.00
118.50
-2.5
Predicted
-1.25
-7.81597E-014
Res idual
5-12
1.25
2.5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Operator
2.5
2
1.25
Res iduals
3
7.81597E-014
3
-1.25
2
-2.5
1
2
3
Operator
5-7 A mechanical engineer is studying the thrust force developed by a drill press. He suspects that the
drilling speed and the feed rate of the material are the most important factors. He selects four feed rates
and uses a high and low drill speed chosen to represent the extreme operating conditions. He obtains the
following results. Analyze the data and draw conclusions. Use D = 0.05.
(A)
Drill Speed
125
0.015
2.70
2.78
0.030
2.45
2.49
Rate
(B)
0.045
2.60
2.72
200
2.83
2.86
2.85
2.80
2.86
2.87
Design Expert Output
Response: Force
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
0.28
7
A
0.15
1
B
0.092
3
AB
0.042
3
Residual
0.021
8
Lack of Fit
0.000
0
Pure Error
0.021
8
Cor Total
0.30
15
Feed
Mean
Square
0.040
0.15
0.031
0.014
2.600E-003
F
Value
15.53
57.01
11.86
5.37
2.600E-003
The Model F-value of 15.53 implies the model is significant.
There is only a 0.05% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
The factors speed and feed rate, as well as the interaction is important.
5-13
0.060
2.75
2.86
2.94
2.88
Prob > F
0.0005
< 0.0001
0.0026
0.0256
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
D rill S p e e d
Force
2 .9 6 8 7 9
X = B: Feed Rate
Y = A: Drill Speed
2 .8 2 9 4
Design Points
F o rc e
A1 125
A2 200
2 .6 9
2 .5 5 0 6
2 .4 1 1 2 1
0 .0 1 5
0 .0 3 0
0 .0 4 5
0 .0 6 0
F e e d R a te
The standard analysis of variance treats all design factors as if they were qualitative. In this case, both
factors are quantitative, so some further analysis can be performed. In Section 5-5, we show how response
curves and surfaces can be fit to the data from a factorial experiment with at least one quantative factor.
Since both factors in this problem are quantitative and have three levels, we can fit linear and quadratic
effects of both temperature and pressure, exactly as in Example 5-5 in the text. The Design-Expert
output, including the response surface plots, now follows.
Design Expert Output
Response: Force
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
0.23
4
A
0.15
1
B
0.019
1
B2
0.058
1
AB
1.125E-003
1
Residual
0.077
11
Lack of Fit
0.056
3
Pure Error
0.021
8
Cor Total
0.30
15
Mean
Square
0.057
0.15
0.019
0.058
1.125E-003
7.021E-003
0.019
2.600E-003
F
Value
8.05
21.11
2.74
8.20
0.16
7.23
Prob > F
0.0027
0.0008
0.1262
0.0154
0.6966
0.0115
significant
significant
The Model F-value of 8.05 implies the model is significant. There is only
a 0.27% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B2 are significant model terms.
Values greater than 0.1000 indicate the model terms are not significant.
If there are many insignificant model terms (not counting those required to support hierarchy),
model reduction may improve your model.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
A-Drill Speed
B-Feed Rate
B2
AB
0.084
2.77
3.03
0.16
Coefficient
Estimate
2.69
0.096
0.047
0.13
-0.011
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
0.7455
0.6529
0.4651
7.835
Standard
Error
0.034
0.021
0.028
0.047
0.028
5-14
95% CI
Low
2.62
0.050
-0.015
0.031
-0.073
95% CI
High
2.76
0.14
0.11
0.24
0.051
VIF
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Final Equation in Terms of Coded Factors:
Force
+2.69
+0.096
+0.047
+0.13
-0.011
=
*A
*B
* B2
*A*B
Final Equation in Terms of Actual Factors:
Force
+2.48917
+3.06667E-003
-15.76667
+266.66667
-0.013333
0 .0 6
=
* Drill Speed
* Feed Rate
* Feed Rate2
* Drill Speed * Feed Rate
F o rce
2
2
2.9
2.85
2.8
3
0 .0 5
2
2.9
2.75
2.8
2.7
2.7
0 .0 4
Fo rce
B: Feed R ate
2
2.65
2.6
2
2.6
2.5
2
0 .0 3
2.8
0.06
2.85
2
0 .0 2
1 2 5 .0 0
1 4 3 .7 5
1 6 2 .5 0
1 8 1 .2 5
200.00
0.05
2
181.25
0.04
2 0 0 .0 0
B: Feed R ate
A: D rill Spe ed
162. 50
0.03
143.75
0.02
125.00
A: D rill Spe ed
5-8 An experiment is conducted to study the influence of operating temperature and three types of faceplate glass in the light output of an oscilloscope tube. The following data are collected:
100
580
568
570
Temperature
125
1090
1087
1085
150
1392
1380
1386
2
550
530
579
1070
1035
1000
1328
1312
1299
3
546
575
599
1045
1053
1066
867
904
889
Glass Type
1
5-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Use D = 0.05 in the analysis. Is there a significant interaction effect? Does glass type or temperature
affect the response? What conclusions can you draw? Use the method discussed in the text to partition
the temperature effect into its linear and quadratic components. Break the interaction down into
appropriate components.
Design Expert Output
Response: Light Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
Model
2.412E+006
A
1.509E+005
B
1.970E+006
AB
2.906E+005
Residual
6579.33
Lack of Fit
0.000
Pure Error
6579.33
Cor Total
2.418E+006
Mean
Square
3.015E+005
75432.26
9.852E+005
72637.93
365.52
DF
8
2
2
4
18
0
18
26
F
Value
824.77
206.37
2695.26
198.73
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
365.52
The Model F-value of 824.77 implies the model is significant.
There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
Both factors, Glass Type (A) and Temperature (B) are significant, as well as the interaction (AB). For
glass types 1 and 2 the response is fairly linear, for glass type 3, there is a quadratic effect.
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
G la s s T y p e
Light Output
1 4 0 2 .4
X = B: Temperature
Y = A: Glass Type
1 1 8 4 .3
A1 1
A2 2
A3 3
L ig h t O u tp u t
Design Points
9 6 6 .1 9 9
7 4 8 .0 9 9
530
100
125
150
T e m p e ra tu re
Design Expert Output
Response: Light Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
2.412E+006
8
A
1.509E+005
2
B
1.780E+006
1
B2
1.906E+005
1
AB
2.262E+005
2
AB2
64373.93
2
Pure Error
6579.33
18
Mean
F
Square
Value
3.015E+005 824.77
75432.26
206.37
1.780E+006 4869.13
1.906E+005 521.39
1.131E+005 309.39
32186.96
88.06
365.52
5-16
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Cor Total
2.418E+006
26
The Model F-value of 824.77 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, B2, AB, AB2 are significant model terms.
Values greater than 0.1000 indicate the model terms are not significant.
If there are many insignificant model terms (not counting those required to support hierarchy),
model reduction may improve your model.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
A[1]
A[2]
B-Temperature
B2
A[1]B
A[2]B
A[1]B2
A[2]B2
19.12
940.19
2.03
14803.50
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
1059.00
28.33
-24.00
314.44
-178.22
92.22
65.56
70.22
76.22
DF
1
1
1
1
1
1
1
1
1
0.9973
0.9961
0.9939
75.466
Standard
Error
6.37
9.01
9.01
4.51
7.81
6.37
6.37
11.04
11.04
95% CI
Low
1045.61
9.40
-42.93
304.98
-194.62
78.83
52.17
47.03
53.03
95% CI
High
1072.39
47.27
-5.07
323.91
-161.82
105.61
78.94
93.41
99.41
VIF
1.00
1.00
Final Equation in Terms of Coded Factors:
Light Output
+1059.00
+28.33
-24.00
+314.44
-178.22
+92.22
+65.56
+70.22
+76.22
=
* A[1]
* A[2]
*B
* B2
* A[1]B
* A[2]B
* A[1]B2
* A[2]B2
Final Equation in Terms of Actual Factors:
Glass Type
Light Output
-3646.00000
+59.46667
-0.17280
Glass Type
Light Output
-3415.00000
+56.00000
-0.16320
Glass Type
Light Output
-7845.33333
+136.13333
-0.51947
1
=
* Temperature
* Temperature2
2
=
* Temperature
* Temperature2
3
=
* Temperature
* Temperature2
5-9 Consider the data in Problem 5-1. Use the method described in the text to compute the linear and
quadratic effects of pressure.
See the alternative analysis shown in Problem 5-1 part (c).
5-17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
5-10 Use Duncan’s multiple range test to determine which levels of the pressure factor are significantly
different for the data in Problem 5-1.
y.3.
90.18
S y. j .
r0.01 2,9 4.60
R 2 4.60 0.0543
y.1.
90.37
MSE
an
y.2.
0.01777
3 2
0.2498
90.68
0.0543
r0.01 3,9 4.86
R3 4.86 0.0543
0.2640
2 vs. 3 = 0.50 > 0.2640 (R3)
2 vs. 1 = 0.31 > 0.2498 (R2)
1 vs. 3 = 0.19 < 0.2498 (R2)
Therefore, 2 differs from 1 and 3.
5-11 An experiment was conducted to determine if either firing temperature or furnace position affects
the baked density of a carbon anode. The data are shown below.
Position
1
2
800
570
565
583
Temperature (°C)
825
1063
1080
1043
850
565
510
590
528
547
521
988
1026
1004
526
538
532
Suppose we assume that no interaction exists. Write down the statistical model. Conduct the analysis of
variance and test hypotheses on the main effects. What conclusions can be drawn? Comment on the
model’s adequacy.
The model for the two-factor, no interaction model is yijk
P W i E j H ijk . Both factors, furnace
position (A) and temperature (B) are significant. The residual plots show nothing unusual.
Design Expert Output
Response: Density
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
9.525E+005
3
A
7160.06
1
B
9.453E+005
2
Residual
6188.78
14
Lack of Fit
818.11
2
Pure Error
5370.67
12
Cor Total
9.587E+005
17
Mean
Square
3.175E+005
7160.06
4.727E+005
442.06
409.06
447.56
F
Value
718.24
16.20
1069.26
0.91
The Model F-value of 718.24 implies the model is significant.
There is only a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
5-18
Prob > F
< 0.0001
0.0013
< 0.0001
significant
0.4271 not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Predicted
Residuals vs. Position
26.5556
6.55556
6.55556
Res iduals
Res iduals
26.5556
-13.4444
-13.4444
-33.4444
-33.4444
-53.4444
-53.4444
523.56
656.15
788.75
921.35
1053.94
1
2
Predicted
Position
Residuals vs. Temperature
26.5556
Res iduals
6.55556
-13.4444
-33.4444
-53.4444
1
2
3
Temperature
5-12 Derive the expected mean squares for a two-factor analysis of variance with one observation per
cell, assuming that both factors are fixed.
Degrees of Freedom
E MS A
V2 b
a
¦
i 1
E MS B
V 2 a
b
W i2
a-1
a 1
E 2j
¦ b 1
b-1
j 1
E MS AB
V2
a
b
¦¦
i 1 j 1
WE
2
ij
a 1 b 1
5-19
a 1 b 1
ab 1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
5-13 Consider the following data from a two-factor factorial experiment. Analyze the data and draw
conclusions. Perform a test for nonadditivity. Use D = 0.05.
Row Factor
1
2
3
1
36
18
30
Design Expert Output
Response:
data
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
609.42
5
121.88
A
580.50
2
290.25
B
28.92
3
9.64
Residual
28.83
6
4.81
Cor Total
638.25
11
Column
2
39
20
37
F
Value
25.36
60.40
2.01
Factor
3
36
22
33
4
32
20
34
Prob > F
0.0006
0.0001
0.2147
significant
The Model F-value of 25.36 implies the model is significant. There is only
a 0.06% chance that a "Model F-Value" this large could occur due to noise.
The row factor (A) is significant.
The test for nonadditivity is as follows:
ª a
«
«¬ i 1
b
§
y 2 ·º
yij yi . y. j y.. ¨¨ SS A SS B .. ¸¸»
ab ¹»
©
1
¼
abSS A SS B
2
¦¦
SS N
SS N
SS N
SS Error
j
ª
§
357 2 ·¸º
»
«4010014 357 ¨¨ 580.50 28.91667
4 3 ¸¹»¼
«¬
©
4 3 580.50 28.91667
3.54051
SS Re sidual SS N
2
28.8333 3.54051 25.29279
Source of
Variation
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
Row
Column
580.50
28.91667
2
3
290.25
9.63889
57.3780
1.9054
Nonadditivity
3.54051
1
3.54051
0.6999
Error
25.29279
5
5.058558
Total
638.25
11
5-14 The shear strength of an adhesive is thought to be affected by the application pressure and
temperature. A factorial experiment is performed in which both factors are assumed to be fixed. Analyze
the data and draw conclusions. Perform a test for nonadditivity.
Temperature (°F)
5-20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pressure (lb/in2)
120
130
140
150
250
9.60
9.69
8.43
9.98
260
11.28
10.10
11.01
10.44
Design Expert Output
Response: Strength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
5.24
5
1.05
A
0.58
3
0.19
B
4.66
2
2.33
Residual
2.15
6
0.36
Cor Total
7.39
11
270
9.00
9.57
9.03
9.80
F
Value
2.92
0.54
6.49
Prob > F
0.1124
0.6727
0.0316
not significant
The "Model F-value" of 2.92 implies the model is not significant relative to the noise.
There is a 11.24 % chance that a "Model F-value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B are significant model terms.
Temperature (B) is a significant factor.
ª a
«
«¬ i 1
b
¦¦
SS N
j 1
§
y 2 ·º
yij yi . y. j y.. ¨¨ SS A SSB .. ¸¸»
ab ¹»
©
¼
abSS A SS B
2
ª
§
117.932 ·¸º
»
«415113.777 117.93 ¨¨ 0.5806917 4.65765
4 3 ¸¹»¼
«¬
©
SS N
4 3 0.5806917 4.65765
SS N 0.48948
SS Error SSRe sidual SS N 2.1538833 0.48948 1.66440
Source of
Variation
Sum of
Squares
Row
Column
Nonadditivity
Degrees of
Freedom
Mean
Square
F0
0.5806917
4.65765
3
2
0.1935639
2.328825
0.5815
6.9960
0.48948
1
0.48948
1.4704
0.33288
Error
1.6644
5
Total
7.392225
11
5-15 Consider the three-factor model
yijk
2
P W i E j J k WE
ij
EJ
5-21
jk
H ijk
i 1,2,...,a
°
® j 1,2,...,b
°k 1,2,...,c
¯
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Notice that there is only one replicate. Assuming the factors are fixed, write down the analysis of variance
table, including the expected mean squares. What would you use as the “experimental error” in order to
test hypotheses?
Source
Degrees of Freedom
A
a-1
Expected Mean Square
V 2 bc
a
¦
W i2
a 1
b
E 2j
i 1
B
V 2 ac
b-1
¦ b 1
j 1
C
V 2 ab
c-1
c
¦
k 1
AB
(a-1)(b-1)
V 2 c
a
b
¦¦
i 1 j 1
BC
(b-1)(c-1)
Error (AC + ABC)
Total
b(a-1)(c-1)
abc-1
2
V a
b
c
J k2
c 1
W E
2
ij
a 1 b 1
EJ
2
jk
¦¦ b 1 c 1
j 1 k 1
V2
5-16 The percentage of hardwood concentration in raw pulp, the vat pressure, and the cooking time of
the pulp are being investigated for their effects on the strength of paper. Three levels of hardwood
concentration, three levels of pressure, and two cooking times are selected. A factorial experiment with
two replicates is conducted, and the following data are obtained:
Percentage
of Hardwood
Concentration
2
Cooking
400
196.6
196.0
Time 3.0
Pressure
500
197.7
196.0
4
198.5
197.2
8
197.5
196.6
Hours
Cooking
650
199.8
199.4
400
198.4
198.6
Time 4.0
Pressure
500
199.6
200.4
Hours
196.0
196.9
198.4
197.6
197.5
198.1
198.7
198.0
199.6
199.0
195.6
196.2
197.4
198.1
197.6
198.4
197.0
197.8
198.5
199.8
650
200.6
200.9
(a) Analyze the data and draw conclusions. Use D = 0.05.
Design Expert Output
Response: strength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
59.73
17
3.51
A
7.76
2
3.88
B
20.25
1
20.25
C
19.37
2
9.69
AB
2.08
2
1.04
AC
6.09
4
1.52
F
Value
9.61
10.62
55.40
26.50
2.85
4.17
5-22
Prob > F
< 0.0001
0.0009
< 0.0001
< 0.0001
0.0843
0.0146
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
BC
ABC
Residual
Lack of Fit
Pure Error
Cor Total
2.19
1.97
6.58
0.000
6.58
66.31
2
4
18
0
18
35
1.10
0.49
0.37
3.00
1.35
0.0750
0.2903
0.37
The Model F-value of 9.61 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AC are significant model terms.
All three main effects, concentration (A), pressure (C) and time (B), as well as the concentration x
pressure interaction (AC) are significant at the 5% level. The concentration x time (AB) and pressure x
time interactions (BC) are significant at the 10% level.
(b) Prepare appropriate residual plots and comment on the model’s adequacy.
Residuals vs. Cooking Time
0.85
0.425
0.425
Res iduals
Res iduals
Residuals vs. Pressure
0.85
0
0
-0.425
-0.425
-0.85
-0.85
1
2
2
3
2
1
Pres sure
Cooking Tim e
Residuals vs. Hardwood
0.85
0.425
0.425
Res iduals
Res iduals
Residuals vs. Predicted
0.85
0
-0.425
-0.85
-0.85
197.11
198.33
199.54
200.75
2
0
-0.425
195.90
2
2
1
Predicted
2
Hardwood
There is nothing unusual about the residual plots.
5-23
3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) Under what set of conditions would you run the process? Why?
DESIGN-EXPERT Plot
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
P re s s u re
strength
In te ra c tio n G ra p h
H a rd w o o d
strength
2 0 0 .9
2 0 0 .9
X = B: Cooking Time
Y = A: Hardwood
1 9 9 .5 7 5
C1 400
C2 500
C3 650
Actual Factor
A: Hardwood = Average1 9 8 .2 5
1 9 9 .5 7 5
A1 2
A2 4
A3 8
Actual Factor
C: Pressure = Average 1 9 8 .2 5
1 9 6 .9 2 5
1 9 6 .9 2 5
1 9 5 .6
1 9 5 .6
s tre n g th
s tre n g th
X = B: Cooking Time
Y = C: Pressure
3
4
3
C o o k in g T im e
DESIGN-EXPERT Plot
strength
4
C o o k in g T im e
In te ra c tio n G ra p h
H a rd w o o d
2 0 0 .9
X = C: Pressure
Y = A: Hardwood
s tre n g th
1 9 9 .5 7 5
A1 2
A2 4
A3 8
Actual Factor
B: Cooking Time = Average
1 9 8 .2 5
1 9 6 .9 2 5
1 9 5 .6
400
500
650
P re s s u re
For the highest strength, run the process with the percentage of hardwood at 2, the pressure at 650, and
the time at 4 hours.
The standard analysis of variance treats all design factors as if they were qualitative. In this case, all three
factors are quantitative, so some further analysis can be performed. In Section 5-5, we show how response
curves and surfaces can be fit to the data from a factorial experiment with at least one quantative factor.
Since the factors in this problem are quantitative and two of them have three levels, we can fit linear and
quadratic. The Design-Expert output, including the response surface plots, now follows.
Design Expert Output
Response: Strength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
58.02
13
Mean
Square
4.46
5-24
F
Value
11.85
Prob > F
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A
B
C
A2
C2
AB
AC
BC
A2B
A2C
AC2
BC2
ABC
Residual
Lack of Fit
Pure Error
Cor Total
7.15
3.42
0.22
1.09
4.43
1.06
3.39
0.15
1.30
2.19
1.65
2.18
0.40
8.29
1.71
6.58
66.31
1
1
1
1
1
1
1
1
1
1
1
1
1
22
4
18
35
7.15
3.42
0.22
1.09
4.43
1.06
3.39
0.15
1.30
2.19
1.65
2.18
0.40
0.38
0.43
0.37
18.98
9.08
0.58
2.88
11.77
2.81
9.01
0.40
3.46
5.81
4.38
5.78
1.06
0.0003
0.0064
0.4559
0.1036
0.0024
0.1081
0.0066
0.5350
0.0763
0.0247
0.0482
0.0251
0.3136
1.17
0.3576
not significant
The Model F-value of 11.85 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C2, AC, A2C, AC2, BC2 are significant model terms.
Values greater than 0.1000 indicate the model terms are not significant.
If there are many insignificant model terms (not counting those required to support hierarchy),
model reduction may improve your model.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
A-Hardwood
B-Cooking Time
C-Pressure
A2
C2
AB
AC
BC
A2B
A2C
AC2
BC2
ABC
0.61
198.06
0.31
22.17
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
197.21
-0.98
0.78
0.19
0.42
0.79
-0.21
-0.46
0.080
0.46
0.73
0.57
-0.55
0.15
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0.8750
0.8011
0.6657
14.071
Standard
Error
0.26
0.23
0.26
0.25
0.25
0.23
0.13
0.15
0.13
0.25
0.30
0.27
0.23
0.15
Final Equation in Terms of Coded Factors:
Strength
+197.21
-0.98
+0.78
+0.19
+0.42
+0.79
-0.21
-0.46
+0.080
+0.46
+0.73
+0.57
-0.55
+0.15
=
*A
*B
*C
* A2
* C2
*A*B
*A*C
*B*C
* A2 * B
* A2 * C
* A * C2
* B * C2
*A*B*C
Final Equation in Terms of Actual Factors:
5-25
95% CI
Low
196.67
-1.45
0.24
-0.33
-0.094
0.31
-0.47
-0.78
-0.18
-0.053
0.10
4.979E-003
-1.02
-0.16
95% CI
High
197.74
-0.51
1.31
0.71
0.94
1.26
0.050
-0.14
0.34
0.98
1.36
1.14
-0.075
0.46
VIF
3.36
6.35
4.04
1.04
1.03
1.06
1.08
1.04
3.96
3.97
3.32
3.30
1.02
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Strength
+229.96981
+12.21654
-12.97602
-0.21224
-0.65287
+2.34333E-004
-1.60038
-0.023415
+0.070658
+0.10278
+6.48026E-004
+1.22143E-005
-7.00000E-005
+8.23308E-004
6 5 0 .0 0
2
2
=
* Hardwood
* Cooking Time
* Pressure
* Hardwood2
* Pressure2
* Hardwood * Cooking Time
* Hardwood * Pressure
* Cooking Time * Pressure
* Hardwood2 * Cooking Time
* Hardwood2 * Pressure
* Hardwood * Pressure2
* Cooking Time * Pressure2
* Hardwood * Cooking Time * Pressure
S tre ng th
2
198.5
200.5
6 0 0 .0 0
200
198
5 5 0 .0 0
5 0 0 .0 0
199
198.5
2
2
2
Streng th
C : Pres s ure
199.5
201.5
201
200.5
200
199.5
199
198.5
198
197.5
197
650.00
4 5 0 .0 0
600.00
2
2
550.00
2
2.00
4 0 0 .0 0
2 .0 0
3 .5 0
5 .0 0
6 .5 0
C : Pres s ure
500.00
3.50
8. 0 0
5.00
450.00
6.50
8.00
A: H a rdw oo d
400.00
A: H ardw oo d
Cooking Time: B = 4.00
5-17 The quality control department of a fabric finishing plant is studying the effect of several factors on
the dyeing of cotton-synthetic cloth used to manufacture men’s shirts. Three operators, three cycle times,
and two temperatures were selected, and three small specimens of cloth were dyed under each set of
conditions. The finished cloth was compared to a standard, and a numerical score was assigned. The
results follow. Analyze the data and draw conclusions. Comment on the model’s adequacy.
Temperature
Cycle Time
40
50
60
1
23
24
25
300°
Operator
2
27
28
26
1
24
23
28
350°
Operator
2
38
36
35
3
31
32
29
3
34
36
39
36
35
36
34
38
39
33
34
35
37
39
35
34
38
36
34
36
31
28
24
35
35
26
27
26
29
36
37
28
26
5-26
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
27
34
25
25
34
24
All three main effects, and the AB, AC, and ABC interactions are significant. There is nothing unusual
about the residual plots.
Design Expert Output
Response: Score
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
1239.33
17
A
436.00
2
B
261.33
2
C
50.07
1
AB
355.67
4
AC
78.81
2
BC
11.26
2
ABC
46.19
4
Residual
118.00
36
Lack of Fit
0.000
0
Pure Error
118.00
36
Cor Total
1357.33
53
Mean
Square
72.90
218.00
130.67
50.07
88.92
39.41
5.63
11.55
3.28
F
Value
22.24
66.51
39.86
15.28
27.13
12.02
1.72
3.52
Prob > F
< 0.0001
< 0.0001
< 0.0001
0.0004
< 0.0001
0.0001
0.1939
0.0159
significant
3.28
The Model F-value of 22.24 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB, AC, ABC are significant model terms.
DESIGN-EXPERT Plot
DESIGN-EXPERT Plot
In te ra c tio n G ra p h
T e m p e ra tu re
Score
39
X = A: Cycle Time
Y = C: Temperature
35
31
35
B1 1
B2 2
B3 3
Actual Factor
C: T emperature = Average 3 1
27
27
23
23
S c o re
S c o re
O p e ra to r
39
X = A: Cycle Time
Y = B: Operator
2
C1 300
C2 350
Actual Factor
B: Operator = Average
In te ra c tio n G ra p h
Score
40
50
60
40
C y c le T im e
50
C y c le T im e
5-27
60
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Operator
Residuals vs. Cycle Time
3
3
4.26326E-014
1.5
2
2
2
2
2
2
3
Res iduals
Res iduals
1.5
4.26326E-014
2
-1.5
-3
-3
2
2
2
2
-1.5
1
3
3
2
2
3
1
2
Operator
3
Cycle Tim e
Residuals vs. Predicted
Residuals vs. Temperature
3
3
1.5
1.5
2
Res iduals
Res iduals
2
2
2
2
4.26326E-014
4.26326E-014
2
2
4
3
2
2
-1.5
-1.5
2
-3
-3
24.00
27.25
30.50
33.75
37.00
1
Predicted
2
Temperature
5-18 In Problem 5-1, suppose that we wish to reject the null hypothesis with a high probability if the
difference in the true mean yield at any two pressures is as great as 0.5. If a reasonable prior estimate of
the standard deviation of yield is 0.1, how many replicates should be run?
)2
n
2
)2
25
)
5
2
naD 2
n 3 0.5
2bV 2
2 3 0.1 2
X1
b 1
2
X2
12.5n
ab n 1
(3)(3)(1)
E
0.014
2 replications will be enough to detect the given difference.
5-28
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
5-19 The yield of a chemical process is being studied. The two factors of interest are temperature and
pressure. Three levels of each factor are selected; however, only 9 runs can be made in one day. The
experimenter runs a complete replicate of the design on each day. The data are shown in the following
table. Analyze the data assuming that the days are blocks.
Temperature
Low
Medium
High
250
86.3
88.5
89.1
Day 1
Pressure
260
84.0
87.3
90.2
Design Expert Output
Response: Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Block
13.01
1
Model
109.81
8
A
5.51
2
B
99.85
2
AB
4.45
4
Residual
4.25
8
Cor Total
127.07
17
270
85.8
89.0
91.3
Mean
Square
13.01
13.73
2.75
49.93
1.11
0.53
Day 2
Pressure
260
85.2
89.9
93.2
250
86.1
89.4
91.7
F
Value
25.84
5.18
93.98
2.10
270
87.3
90.3
93.7
Prob > F
< 0.0001
0.0360
< 0.0001
0.1733
significant
The Model F-value of 25.84 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Both main effects, temperature and pressure, are significant.
5-20 Consider the data in Problem 5-5. Analyze the data, assuming that replicates are blocks.
Design Expert Output
Response:
Warping
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Block
11.28
1
Model
968.22
15
A
698.34
3
B
156.09
3
AB
113.78
9
Residual
97.22
15
Cor Total
1076.72
31
Mean
Square
11.28
64.55
232.78
52.03
12.64
6.48
F
Value
Prob > F
9.96
35.92
8.03
1.95
< 0.0001
< 0.0001
0.0020
0.1214
significant
The Model F-value of 9.96 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Both temperature and copper content are significant. This agrees with the analysis in Problem 5-5.
5-21 Consider the data in Problem 5-6. Analyze the data, assuming that replicates are blocks.
5-29
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design-Expert Output
Response: Stength
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Block
1.04
1
Model
217.46
11
A
160.33
2
B
12.46
3
AB
44.67
6
Residual
44.46
11
Cor Total
262.96
23
Mean
Square
1.04
19.77
80.17
4.15
7.44
4.04
F
Value
Prob > F
4.89
19.84
1.03
1.84
0.0070
0.0002
0.4179
0.1799
significant
The Model F-value of 4.89 implies the model is significant. There is only
a 0.70% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A are significant model terms.
Only the operator factor (A) is significant. This agrees with the analysis in Problem 5-6.
5-22 An article in the Journal of Testing and Evaluation (Vol. 16, no.2, pp. 508-515) investigated the
effects of cyclic loading and environmental conditions on fatigue crack growth at a constant 22 MPa stress
for a particular material. The data from this experiment are shown below (the response is crack growth
rate).
Frequency
10
1
0.1
Air
Environment
H2O
Salt H2O
2.29
2.47
2.48
2.12
2.06
2.05
2.23
2.03
1.90
1.93
1.75
2.06
2.65
2.68
2.06
2.38
3.20
3.18
3.96
3.64
3.10
3.24
3.98
3.24
2.24
2.71
2.81
2.08
11.00
11.00
9.06
11.30
9.96
10.01
9.36
10.40
(a) Analyze the data from this experiment (use D = 0.05).
Design Expert Output
Response: Crack Growth
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
376.11
8
A
209.89
2
B
64.25
2
AB
101.97
4
Residual
5.42
27
Lack of Fit
0.000
0
Mean
Square
47.01
104.95
32.13
25.49
0.20
5-30
F
Value
234.02
522.40
159.92
126.89
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pure Error
Cor Total
5.42
381.53
27
35
0.20
The Model F-value of 234.02 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
Both frequency and environment, as well as their interaction are significant.
(b) Analyze the residuals.
The residual plots indicate that there may be some problem with inequality of variance. This is
particularly noticable on the plot of residuals versus predicted response and the plot of residuals versus
frequency.
Residuals vs. Predicted
Normal plot of residuals
0.71
99
2
95
Normal % probability
Res iduals
0.15
2
-0.41
-0.97
90
80
70
50
30
20
10
5
1
-1.53
1.91
4.08
6.25
8.42
10.59
-1.53
Predicted
-0.97
-0.41
0.71
Res idual
Residuals vs. Environment
Residuals vs. Frequency
0.71
0.71
2
2
0.15
0.15
2
Res iduals
Res iduals
0.15
-0.41
2
-0.41
-0.97
-0.97
-1.53
-1.53
1
2
3
1
Environm ent
2
3
Frequency
(c) Repeat the analyses from parts (a) and (b) using ln(y) as the response. Comment on the results.
5-31
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response: Crack Growth
Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
13.46
8
1.68
A
7.57
2
3.79
B
2.36
2
1.18
AB
3.53
4
0.88
Residual
0.25
27
9.367E-003
Lack of Fit
0.000
0
Pure Error
0.25
27
9.367E-003
Cor Total
13.71
35
Constant: 0.000
F
Value
179.57
404.09
125.85
94.17
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 179.57 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
Both frequency and environment, as well as their interaction are significant. The residual plots of the
based on the transformed data look better.
Normal plot of residuals
Residuals vs. Predicted
0.165324
99
95
Normal % probability
0.0827832
Res iduals
2
0.000242214
2
-0.0822988
90
80
70
50
30
20
10
5
1
-0.16484
0.65
1.07
1.50
1.93
2.36
-0.16484
Predicted
-0.0822988 0.000242214 0.0827832
Res idual
5-32
0.165324
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Frequency
Residuals vs. Environment
0.165324
0.165324
0.0827832
0.0827832
2
Res iduals
Res iduals
2
0.000242214
0.000242214
2
2
-0.0822988
-0.0822988
-0.16484
-0.16484
1
2
3
1
Environm ent
2
3
Frequency
5-23 An article in the IEEE Transactions on Electron Devices (Nov. 1986, pp. 1754) describes a study
on polysilicon doping. The experiment shown below is a variation of their study. The response variable
is base current.
Polysilicon
Doping (ions)
1 x 10 20
2 x 10 20
Anneal Temperature (°C)
900
950
1000
4.60
10.15
11.01
4.40
10.20
10.58
3.20
3.50
9.38
10.02
10.81
10.60
(a) Is there evidence (with D = 0.05) indicating that either polysilicon doping level or anneal temperature
affect base current?
Design Expert Output
Response:
Base Current
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
112.74
5
A
0.98
1
B
111.19
2
AB
0.58
2
Residual
0.39
6
Lack of Fit
0.000
0
Pure Error
0.39
6
Cor Total
113.13
11
Mean
Square
22.55
0.98
55.59
0.29
0.064
F
Value
350.91
15.26
865.16
4.48
Prob > F
< 0.0001
0.0079
< 0.0001
0.0645
significant
0.064
The Model F-value of 350.91 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Both factors, doping and anneal are significant. Their interaction is significant at the 10% level.
5-33
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(b) Prepare graphical displays to assist in interpretation of this experiment.
Interaction Graph
Doping
11.1051
Base Current
9.08882
7.0725
5.05618
A-
3.03986
A+
900
950
1000
Anneal
(c) Analyze the residuals and comment on model adequacy.
Normal plot of residuals
Residuals vs. Predicted
0.32
99
95
Res iduals
Normal % probability
0.16
8.88178E-016
-0.16
90
80
70
50
30
20
10
5
1
-0.32
3.35
5.21
7.07
8.93
10.80
-0.32
Predicted
-0.16
-8.88178E-016
Res idual
5-34
0.16
0.32
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals vs. Anneal
Residuals vs. Doping
0.16
0.16
8.88178E-016
8.88178E-016
-0.16
-0.16
-0.32
-0.32
Res iduals
Res iduals
0.32
0.32
1
2
1
2
Doping
3
Anneal
There is a funnel shape in the plot of residuals versus predicted, indicating some inequality of variance.
(d) Is the model y E 0 E1 x1 E 2 x2 E 22 x22 E12 x1 x2 H supported by this experiment (x1 = doping
level, x2 = temperature)? Estimate the parameters in this model and plot the response surface.
Design Expert Output
Response:
Base Current
ANOVA for Response Surface Reduced Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
112.73
4
28.18
A
0.98
1
0.98
B
93.16
1
93.16
B2
18.03
1
18.03
AB
0.56
1
0.56
Residual
0.40
7
0.057
Lack of Fit
0.014
1
0.014
Pure Error
0.39
6
0.064
Cor Total
113.13
11
F
Value
493.73
17.18
1632.09
315.81
9.84
0.22
Prob > F
< 0.0001
0.0043
< 0.0001
< 0.0001
0.0164
significant
0.6569 not significant
The Model F-value of 493.73 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, B2, AB are significant model terms.
Factor
Intercept
A-Doping
B-Anneal
B2
AB
Coefficient
Estimate
9.94
-0.29
3.41
-2.60
0.27
DF
1
1
1
1
1
Standard
Error
0.12
0.069
0.084
0.15
0.084
95% CI
Low
9.66
-0.45
3.21
-2.95
0.065
95% CI
High
10.22
-0.12
3.61
-2.25
0.46
VIF
1.00
1.00
1.00
1.00
All of the coefficients in the assumed model are significant. The quadratic effect is easily observable in
the response surface plot.
5-35
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1000.00
2
Base Current
2
11
950.00
10
2
2
9
8
925.00
7
Base Current
Anneal
975.00
12
11
10
9
8
7
6
5
4
3
6
5
900.00
2
1.00E+20
1.25E+20
1.50E+20
1000.00
4
1.75E+20
2
1.00E+20
975.00
1.25E+20
2.00E+20
950.00
1.50E+20
Doping
Doping
5-36
1.75E+20
2.00E+20
925.00
900.00
Anneal
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 6
k
The 2 Factorial Design
Solutions
6-1 An engineer is interested in the effects of cutting speed (A), tool geometry (B), and cutting angle on
the life (in hours) of a machine tool. Two levels of each factor are chosen, and three replicates of a 23
factorial design are run. The results follow:
A
B
C
Treatment
Combination
I
Replicate
II
+
-
-
(1)
a
22
32
31
43
25
29
-
+
-
b
35
34
50
+
+
-
ab
55
47
46
III
-
-
+
c
44
45
38
+
-
+
ac
40
37
36
-
+
+
bc
60
50
54
+
+
+
abc
39
41
47
(a) Estimate the factor effects. Which effects appear to be large?
From the normal probability plot of effects below, factors B, C, and the AC interaction appear to be
significant.
No rm a l p lo t
DE S IG N -E X P E RT P l o t
L i fe
A : C u tti n g S p e e d
B : T o ol G e o m e try
C: C u tti n g A n g l e
99
B
N orm al % probab ility
95
C
90
80
70
A
50
30
20
10
5
1
AC
-8 .8 3
-3 .7 9
1 .2 5
6 .2 9
1 1 .3 3
Effect
(b) Use the analysis of variance to confirm your conclusions for part (a).
The analysis of variance confirms the significance of factors B, C, and the AC interaction.
Design Expert Output
Response:
Life
in hours
6-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1612.67
7
230.38
A
0.67
1
0.67
B
770.67
1
770.67
C
280.17
1
280.17
AB
16.67
1
16.67
AC
468.17
1
468.17
BC
48.17
1
48.17
ABC
28.17
1
28.17
Pure Error
482.67
16
30.17
Cor Total
2095.33
23
F
Value
7.64
0.022
25.55
9.29
0.55
15.52
1.60
0.93
Prob > F
0.0004
0.8837
0.0001
0.0077
0.4681
0.0012
0.2245
0.3483
significant
The Model F-value of 7.64 implies the model is significant. There is only
a 0.04% chance that a "Model F-Value" this large could occur due to noise.
The reduced model ANOVA is shown below. Factor A was included to maintain hierarchy.
Design Expert Output
Response:
Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1519.67
4
379.92
A
0.67
1
0.67
B
770.67
1
770.67
C
280.17
1
280.17
AC
468.17
1
468.17
Residual
575.67
19
30.30
Lack of Fit
93.00
3
31.00
Pure Error
482.67
16
30.17
Cor Total
2095.33
23
F
Value
12.54
0.022
25.44
9.25
15.45
1.03
Prob > F
< 0.0001
0.8836
< 0.0001
0.0067
0.0009
0.4067
significant
not significant
The Model F-value of 12.54 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Effects B, C and AC are significant at 1%.
(c) Write down a regression model for predicting tool life (in hours) based on the results of this
experiment.
yijk
40.8333 0.1667 x A 5.6667 xB 3.4167 xC 4.4167 x A xC
Design Expert Output
Coefficient
Factor
Estimate
DF
Intercept
40.83
1
A-Cutting Speed
0.17
1
B-Tool Geometry
5.67
1
C-Cutting Angle
3.42
1
AC
-4.42
1
Final Equation in Terms of Coded Factors:
Life
+40.83
+0.17
+5.67
+3.42
-4.42
Standard
Error
1.12
1.12
1.12
1.12
1.12
95% CI
Low
38.48
-2.19
3.31
1.06
-6.77
=
*A
*B
*C
*A*C
Final Equation in Terms of Actual Factors:
6-2
95% CI
High
43.19
2.52
8.02
5.77
-2.06
VIF
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Life
+40.83333
+0.16667
+5.66667
+3.41667
-4.41667
=
* Cutting Speed
* Tool Geometry
* Cutting Angle
* Cutting Speed * Cutting Angle
The equation in part (c) and in the given in the computer output form a “hierarchial” model, that is, if an
interaction is included in the model, then all of the main effects referenced in the interaction are also
included in the model.
(d) Analyze the residuals. Are there any obvious problems?
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 1 .5
99
6 .7 9 16 7
90
80
70
R es idua ls
N orm al % probability
95
50
2 .0 8 33 3
30
20
10
-2 .6 25
5
1
-7 .3 33 3 3
-7 .3 33 3 3
-2 .6 25
2 .0 8 33 3
6 .7 9 16 7
1 1 .5
2 7 .1 7
R es idua l
3 3 .9 2
4 0 .6 7
4 7 .4 2
5 4 .1 7
Predicted
There is nothing unusual about the residual plots.
(e) Based on the analysis of main effects and interaction plots, what levels of A, B, and C would you
recommend using?
Since B has a positive effect, set B at the high level to increase life. The AC interaction plot reveals that
life would be maximized with C at the high level and A at the low level.
6-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
L i fe
60
One F a ctor P lot
DE S IG N-E X P E RT P l o t
C utting Ang le
L i fe
X = A : Cu tti n g S p ee d
Y = C: Cu tti n g A n gl e
60
X = B : T o o l G e o m e try
A c tu al Fa c tors
5 0 .5
A : Cu tti n g S p e ed = 0 .0 0
C: Cu tti n g A n g le = 0 .0 0
5 0 .5
Life
Life
C- -1 .0 0 0
C+ 1 .0 0 0
A ctu al Fa ctor
B : T o o l G e o m e try = 0 .0 0
41
41
3 1 .5
3 1 .5
22
22
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
-0 .5 0
C utting Speed
0 .0 0
0 .5 0
1 .0 0
Too l Geo m etry
6-2 Reconsider part (c) of Problem 6-1. Use the regression model to generate response surface and
contour plots of the tool life response. Interpret these plots. Do they provide insight regarding the
desirable operating conditions for this process?
The response surface plot and the contour plot in terms of factors A and C with B at the high level are
shown below. They show the curvature due to the AC interaction. These plots make it easy to see the
region of greatest tool life.
L ife
DE S IG N-E X P E RT P l o t
1 .0 0
L i fe
X = A : Cu tti n g S p ee d
Y = C: Cu tti n g A n gl e
DE S IG N-E X P E RT P l o t
L i fe
X = A : Cu tti n g S p ee d
Y = C: Cu tti n g A n gl e
45.8889
43.2778
A ctu al Fa ctor
B : T o o l G e o m e try = 0 .0 0 0 .5 0
A c tu al Fa c tor
B : T o o l G e o m e try = 0 .0 0
44.5833
40.6667
0 .0 0
40.6667
36.75
Life
C utting Ang le
48.5
32.8333
38.0556
-0 .5 0
1.00
35.4444
0.50
1.00
-1 .0 0
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
0.00
0.50
1 .0 0
C utting Sp eed
0.00
-0.50
-0.50
C utting Ang le
C utting Speed
-1.00
-1.00
6-3 Find the standard error of the factor effects and approximate 95 percent confidence limits for the
factor effects in Problem 6-1. Do the results of this analysis agree with the conclusions from the analysis
of variance?
6-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1
SE ( effect )
n2
Variable
A
B
AB
C
AC
BC
ABC
k 2
1
S2
3 2 3 2
Effect
0.333
11.333
-1.667
6.833
-8.833
-2.833
-2.167
30.17
CI
r4.395
r4.395
r4.395
r4.395
r4.395
r4.395
r4.395
2.24
*
*
*
The 95% confidence intervals for factors B, C and AC do not contain zero. This agrees with the analysis
of variance approach.
6-4 Plot the factor effects from Problem 6-1 on a graph relative to an appropriately scaled t distribution.
Does this graphical display adequately identify the important factors? Compare the conclusions from this
MSE
30.17
3.17
plot with the results from the analysis of variance. S
n
3
S c a le d t D i s tr ib u ti o n
AC
-1 0 . 0
C
0 .0
B
1 0 .0
F a c to r E f fe c ts
This method identifies the same factors as the analysis of variance.
6-5 A router is used to cut locating notches on a printed circuit board. The vibration level at the
surface of the board as it is cut is considered to be a major source of dimensional variation in the notches.
Two factors are thought to influence vibration: bit size (A) and cutting speed (B). Two bit sizes (1/16 and
1/8 inch) and two speeds (40 and 90 rpm) are selected, and four boards are cut at each set of conditions
shown below. The response variable is vibration measured as a resultant vector of three accelerometers (x,
y, and z) on each test circuit board.
A
B
Treatment
Combination
I
Replicate
II
III
IV
+
-
(1)
a
18.2
27.2
18.9
24.0
12.9
22.4
14.4
22.5
-
+
b
15.9
14.5
15.1
14.2
6-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+
+
ab
41.0
43.9
36.3
39.9
(a) Analyze the data from this experiment.
Design Expert Output
Response:
Vibration
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1638.11
3
546.04
A
1107.23
1
1107.23
B
227.26
1
227.26
AB
303.63
1
303.63
Residual
71.72
12
5.98
Lack of Fit
0.000
0
Pure Error
71.72
12
5.98
Cor Total
1709.83
15
F
Value
91.36
185.25
38.02
50.80
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 91.36 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
(b) Construct a normal probability plot of the residuals, and plot the residuals versus the predicted
vibration level. Interpret these plots.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
3 .6 2 5
99
1 .7 2 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
-0 .1 75
20
10
-2 .0 75
5
1
-3 .9 75
-3 .9 75
-2 .0 75
-0 .1 75
1 .7 2 5
3 .6 2 5
1 4 .9 2
R es idua l
2 1 .2 6
2 7 .6 0
3 3 .9 4
4 0 .2 7
Predicted
There is nothing unusual about the residual plots.
(c) Draw the AB interaction plot. Interpret this plot. What levels of bit size and speed would you
recommend for routine operation?
To reduce the vibration, use the smaller bit. Once the small bit is specified, either speed will work equally
well, because the slope of the curve relating vibration to speed for the small tip is approximately zero.
The process is robust to speed changes if the small bit is used.
6-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
V i b ra ti o n
C utting Speed
4 3 .9
X = A : B i t S i ze
Y = B : Cu tti n g S p ee d
3 6 .1 5
De si g n P o i n ts
Vibration
B - -1 .0 0 0
B + 1 .0 0 0
2 8 .4
2 0 .6 5
1 2 .9
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
Bit Size
6-6 Reconsider the experiment described in Problem 6-1. Suppose that the experimenter only
performed the eight trials from replicate I. In addition, he ran four center points and obtained the
following response values: 36, 40, 43, 45.
(a) Estimate the factor effects. Which effects are large?
No rm a l p lo t
DE S IG N-E X P E RT P l o t
L i fe
A : C u tti n g S p e e d
B : T o o l G e o m e try
C: C u tti n g A n g l e
99
N orm a l % proba bility
95
B
90
C
80
70
50
30
20
10
5
AC
1
-1 3 .7 5
-7 .1 3
-0 .5 0
6 .1 2
1 2 .7 5
Effect
Effects B, C, and AC appear to be large.
(b) Perform an analysis of variance, including a check for pure quadratic curvature. What are your
conclusions? SS PureQuadratic
Design Expert Output
Response:
Life
n F nC y F y C
n F nC
2
in hours
6-7
8 4 40.875 41.000
8 4
2
0.0417
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1048.88
7
149.84
A
3.13
1
3.13
B
325.13
1
325.13
C
190.12
1
190.12
AB
6.13
1
6.13
AC
378.12
1
378.12
BC
55.12
1
55.12
ABC
91.12
1
91.12
Curvature
0.042
1
0.042
Pure Error
46.00
3
15.33
Cor Total
1094.92
11
F
Value
Prob > F
9.77
0.0439
0.20
0.6823
21.20
0.0193
12.40
0.0389
0.40
0.5722
24.66
0.0157
3.60
0.1542
5.94
0.0927
2.717E-003 0.9617
significant
not significant
The Model F-value of 9.77 implies the model is significant. There is only
a 4.39% chance that a "Model F-Value" this large could occur due to noise.
The "Curvature F-value" of 0.00 implies the curvature (as measured by difference between the
average of the center points and the average of the factorial points) in the design space is not
significant relative to the noise. There is a 96.17% chance that a "Curvature F-value"
this large could occur due to noise.
Design Expert Output
Response:
Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
896.50
4
224.13
A
3.13
1
3.13
B
325.12
1
325.12
C
190.12
1
190.12
AC
378.12
1
378.12
Residual
198.42
7
28.35
Lack of Fit
152.42
4
38.10
Pure Error
46.00
3
15.33
Cor Total
1094.92
11
F
Value
7.91
0.11
11.47
6.71
13.34
Prob > F
0.0098
0.7496
0.0117
0.0360
0.0082
2.49
0.2402
significant
not significant
The Model F-value of 7.91 implies the model is significant. There is only
a 0.98% chance that a "Model F-Value" this large could occur due to noise.
Effects B, C and AC are significant at 5%. There is no effect of curvature.
(c) Write down an appropriate model for predicting tool life, based on the results of this experiment.
Does this model differ in any substantial way from the model in Problem 7-1, part (c)?
Design Expert Output
Final Equation in Terms of Coded Factors:
Life
+40.88
+0.62
+6.37
+4.87
-6.88
=
*A
*B
*C
*A*C
(d) Analyze the residuals.
6-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t o f re sid ua ls
Re sid ua ls vs. P re d icte d
3 .0 0
99
Studen tize d R es id uals
N orm al % probab ility
95
90
80
70
50
30
20
10
5
1 .5 0
0 .0 0
-1 .5 0
1
-3 .0 0
-2 .1 1
-1 .0 5
0 .0 0
1 .0 5
2 .1 1
2 2 .1 3
3 1 .1 9
Studen tize d R es id uals
4 0 .2 5
4 9 .3 1
5 8 .3 8
Pred icte d
(e) What conclusions would you draw about the appropriate operating conditions for this process?
To maximize life run with B at the high level, A at the low level and C at the high level
C ube Graph
Life
58.3 8
B+
34 .88
45 .88
B: Tool Ge om etry
49 .88
45.6 3
33 .13
C+
C : C utting Ang le
BA-
22 .13
37 .13
A: C uttin g Speed
CA+
6-7 An experiment was performed to improve the yield of a chemical process. Four factors were
selected, and two replicates of a completely randomized experiment were run. The results are shown in
the following table:
Treatment
Combination
Replicate
I
Replicate
II
Treatment
Combination
Replicate
I
Replicate
II
(1)
a
90
74
93
78
d
ad
98
72
95
76
b
81
85
bd
87
83
6-9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ab
83
80
abd
85
86
c
77
78
cd
99
90
ac
81
80
acd
79
75
bc
88
82
bcd
87
84
abc
73
70
abcd
80
80
(a) Estimate the factor effects.
Design Expert Output
Term
Model
Intercept
Error
A
Error
B
Error
C
Error
D
Error
AB
Error
AC
Error
AD
Error
BC
Error
BD
Error
CD
Error
ABC
Error
ABD
Error
ACD
Error
BCD
Error
ABCD
Effect
-9.0625
-1.3125
-2.6875
3.9375
4.0625
0.6875
-2.1875
-0.5625
-0.1875
1.6875
-5.1875
4.6875
-0.9375
-0.9375
2.4375
SumSqr
% Contribtn
657.031
13.7812
57.7813
124.031
132.031
3.78125
38.2813
2.53125
0.28125
22.7812
215.281
175.781
7.03125
7.03125
47.5313
40.3714
0.84679
3.55038
7.62111
8.11267
0.232339
2.3522
0.155533
0.0172814
1.3998
13.228
10.8009
0.432036
0.432036
2.92056
(b) Prepare an analysis of variance table, and determine which factors are important in explaining yield.
Design Expert Output
Response:
yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1504.97
15
100.33
A
657.03
1
657.03
B
13.78
1
13.78
C
57.78
1
57.78
D
124.03
1
124.03
AB
132.03
1
132.03
AC
3.78
1
3.78
AD
38.28
1
38.28
BC
2.53
1
2.53
BD
0.28
1
0.28
CD
22.78
1
22.78
ABC
215.28
1
215.28
ABD
175.78
1
175.78
ACD
7.03
1
7.03
BCD
7.03
1
7.03
ABCD
47.53
1
47.53
Residual
122.50
16
7.66
Lack of Fit
0.000
0
Pure Error
122.50
16
7.66
Cor Total
1627.47
31
F
Value
13.10
85.82
1.80
7.55
16.20
17.24
0.49
5.00
0.33
0.037
2.98
28.12
22.96
0.92
0.92
6.21
The Model F-value of 13.10 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, C, D, AB, AD, ABC, ABD, ABCD are significant model terms.
6-10
Prob > F
< 0.0001
< 0.0001
0.1984
0.0143
0.0010
0.0007
0.4923
0.0399
0.5733
0.8504
0.1038
< 0.0001
0.0002
0.3522
0.3522
0.0241
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
F0.01,1,16
8.53 , and F0.025,1,16
612
. therefore, factors A and D and interactions AB, ABC, and ABD are
significant at 1%. Factor C and interactions AD and ABCD are significant at 2.5%.
(b) Write down a regression model for predicting yield, assuming that all four factors were varied over
the range from -1 to +1 (in coded units).
Model with hierarchy maintained:
Design Expert Output
Final Equation in Terms of Coded Factors:
yield
+82.78
-4.53
-0.66
-1.34
+1.97
+2.03
+0.34
-1.09
-0.28
-0.094
+0.84
-2.59
+2.34
-0.47
-0.47
+1.22
=
*A
*B
*C
*D
*A*B
*A*C
*A*D
*B*C
*B*D
*C*D
*A*B*C
*A*B*D
*A*C*D
*B*C*D
*A*B*C*D
Model without hierarchy terms:
Design Expert Output
Final Equation in Terms of Coded Factors:
yield
+82.78
-4.53
-1.34
+1.97
+2.03
-1.09
-2.59
+2.34
+1.22
=
*A
*C
*D
*A*B
*A*D
*A*B*C
*A*B*D
*A*B*C*D
Confirmation runs might be run to see if the simpler model without hierarchy is satisfactory.
(d) Plot the residuals versus the predicted yield and on a normal probability scale. Does the residual
analysis appear satisfactory?
There appears to be one large residual both in the normal probability plot and in the plot of residuals
versus predicted.
6-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t o f re sid ua ls
Re sid ua ls vs. P re d icte d
6 .9 6 8 7 5
99
3 .9 6 8 7 5
90
80
R es iduals
N orm al % probab ility
95
70
50
0 .9 6 8 7 5
30
2
20
10
-2 .0 3 1 2 5
5
1
-5 .0 3 1 2 5
-5 .0 3 1 2 5
-2 .0 3 1 2 5
0 .9 6 8 7 5
3 .9 6 8 7 5
6 .9 6 8 7 5
7 1 .9 1
7 8 .3 0
8 4 .6 9
R es idual
9 1 .0 8
9 7 .4 7
Pred icte d
(e) Two three-factor interactions, ABC and ABD, apparently have large effects. Draw a cube plot in the
factors A, B, and C with the average yields shown at each corner. Repeat using the factors A, B, and
D. Do these two plots aid in data interpretation? Where would you recommend that the process be
run with respect to the four variables?
C ube Graph
C ube Graph
yield
yield
86.5 3
B: B
84 .03
86.0 0
B+
84 .22
85.4 1
77 .47
84 .56
B: B
B+
76 .34
C+
83 .50
77 .06
94.7 5
74 .75
C: C
BA-
93 .28
74 .97
A: A
D+
D: D
C-
B-
A+
A-
83 .94
77 .69
A: A
DA+
Run the process at A low B low, C low and D high.
6-8 A bacteriologist is interested in the effects of two different culture media and two different times on
the growth of a particular virus. She performs six replicates of a 22 design, making the runs in random
order. Analyze the bacterial growth data that follow and draw appropriate conclusions. Analyze the
residuals and comment on the model’s adequacy.
Culture Medium
Time
1
2
6-12
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
12 hr
18 hr
21
22
25
26
23
28
24
25
20
26
29
27
37
39
31
34
37
39
31
34
35
36
30
35
Design Expert Output
Response:
Virus growth
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
691.46
3
230.49
A
9.38
1
9.38
B
590.04
1
590.04
AB
92.04
1
92.04
Residual
102.17
20
5.11
Lack of Fit
0.000
0
Pure Error
102.17
20
5.11
Cor Total
793.63
23
F
Value
45.12
1.84
115.51
18.02
Prob > F
< 0.0001
0.1906
< 0.0001
0.0004
significant
The Model F-value of 45.12 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B, AB are significant model terms.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
4 .6 6 66 7
99
2 .6 6 66 7
90
80
70
R es idua ls
N orm al % probability
95
50
2
0 .6 6 66 6 7
30
20
2
10
-1 .3 33 3 3
5
1
-3 .3 33 3 3
-3 .3 33 3 3
-1 .3 33 3 3
0 .6 6 66 6 7
2 .6 6 66 7
4 .6 6 66 7
2 3 .3 3
R es idua l
2 6 .7 9
3 0 .2 5
3 3 .7 1
3 7 .1 7
Predicted
Growth rate is affected by factor B (Time) and the AB interaction (Culture medium and Time). There is
some very slight indication of inequality of variance shown by the small decreasing funnel shape in the
plot of residuals versus predicted.
6-13
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
V i ru s g ro wth
Tim e
39
2
X = A : Cu l tu re M e d i u m
Y = B: T im e
B - 1 2 .0 00
B + 1 8 .0 00
3 4 .2 5
Viru s grow th
De si g n P o i n ts
2 9 .5
2
2
2 4 .7 5
20
1
2
C ulture Med ium
6-9 An industrial engineer employed by a beverage bottler is interested in the effects of two different
typed of 32-ounce bottles on the time to deliver 12-bottle cases of the product. The two bottle types are
glass and plastic. Two workers are used to perform a task consisting of moving 40 cases of the product 50
feet on a standard type of hand truck and stacking the cases in a display. Four replicates of a 22 factorial
design are performed, and the times observed are listed in the following table. Analyze the data and draw
the appropriate conclusions. Analyze the residuals and comment on the model’s adequacy.
Worker
Bottle Type
1
1
2
2
Glass
5.12
4.98
4.89
5.00
6.65
5.49
6.24
5.55
Plastic
4.95
4.43
5.28
4.91
4.27
4.25
4.75
4.71
Design Expert Output
Response: Times
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
4.86
3
1.62
A
2.02
1
2.02
B
2.54
1
2.54
AB
0.30
1
0.30
Residual
1.49
12
0.12
Lack of Fit
0.000
0
Pure Error
1.49
12
0.12
Cor Total
6.35
15
Mean
Square
13.04
16.28
20.41
2.41
F
Value Prob > F
0.0004
0.0017
0.0007
0.1463
The Model F-value of 13.04 implies the model is significant. There is only
a 0.04% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
There is some indication of non-constant variance in this experiment.
6-14
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .6 6 75
99
0 .3 7 75
90
80
R es idua ls
N orm al % probability
95
70
50
30
0 .0 8 75
20
10
-0 .2 02 5
5
1
-0 .4 92 5
-0 .4 92 5
-0 .2 02 5
0 .0 8 75
0 .3 7 75
0 .6 6 75
4 .4 7
4 .8 5
R es idua l
5 .2 3
5 .6 1
5 .9 8
Predicted
Re sid ua ls vs. W orke r
0 .6 6 75
R es idua ls
0 .3 7 75
0 .0 8 75
-0 .2 02 5
-0 .4 92 5
1
2
Wo rker
6-10 In problem 6-9, the engineer was also interested in potential fatigue differences resulting from the
two types of bottles. As a measure of the amount of effort required, he measured the elevation of heart
rate (pulse) induced by the task. The results follow. Analyze the data and draw conclusions. Analyze the
residuals and comment on the model’s adequacy.
Worker
Bottle Type
1
1
2
2
Glass
39
58
45
35
20
16
13
11
Plastic
44
35
13
10
42
21
16
15
Design Expert Output
6-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Response:
Pulse
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2784.19
3
928.06
A
2626.56
1
2626.56
B
105.06
1
105.06
AB
52.56
1
52.56
Residual
694.75
12
57.90
Lack of Fit
0.000
0
Pure Error
694.75
12
57.90
Cor Total
3478.94
15
F
Value
16.03
45.37
1.81
0.91
Prob > F
0.0002
< 0.0001
0.2028
0.3595
significant
The Model F-value of 16.03 implies the model is significant. There is only
a 0.02% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A are significant model terms.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 3 .7 5
99
6 .6 8 75
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
-0 .3 75
-7 .4 37 5
5
1
-1 4 .5
-1 4 .5
-7 .4 37 5
-0 .3 75
6 .6 8 75
1 3 .7 5
1 3 .5 0
R es idua l
2 1 .1 9
2 8 .8 8
Predicted
Re sid ua ls vs. W orke r
1 3 .7 5
R es idua ls
6 .6 8 75
-0 .3 75
-7 .4 37 5
-1 4 .5
1
2
Wo rker
There is an indication that one worker exhibits greater variability than the other.
6-16
3 6 .5 6
4 4 .2 5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
6-11 Calculate approximate 95 percent confidence limits for the factor effects in Problem 6-10. Do the
results of this analysis agree with the analysis of variance performed in Problem 6-10?
1
SE ( effect )
n2
Variable
k 2
1
S2
4 2 2 2
Effect
57.90
3.80
C.I.
A
B
-25.625
-5.125
r3.80(1.96)= r7.448
r3.80(1.96)= r7.448
AB
-7.25
r3.80(1.96)= r7.448
The 95% confidence intervals for factors A does not contain zero. This agrees with the analysis of
variance approach.
6-12 An article in the AT&T Technical Journal (March/April 1986, Vol. 65, pp. 39-50) describes the
application of two-level factorial designs to integrated circuit manufacturing. A basic processing step is to
grow an epitaxial layer on polished silicon wafers. The wafers mounted on a susceptor are positioned
inside a bell jar, and chemical vapors are introduced. The susceptor is rotated and heat is applied until the
epitaxial layer is thick enough. An experiment was run using two factors: arsenic flow rate (A) and
deposition time (B). Four replicates were run, and the epitaxial layer thickness was measured (in mm).
The data are shown below:
Replicate
II
A
B
I
-
-
14.037
III
IV
16.165
13.972
13.907
+
-
13.880
13.860
14.032
13.914
-
+
14.821
14.757
14.843
14.878
+
+
14.888
14.921
14.415
14.932
Factor
Low (-)
Levels
High (+)
A
55%
59%
B
Short
Long
(10 min)
(15 min)
(a) Estimate the factor effects.
Design Expert Output
Term
Model
Intercept
Error
A
Error
B
Error
AB
Error
Lack Of Fit
Error
Pure Error
Effect
-0.31725
0.586
0.2815
SumSqr
0.40259
1.37358
0.316969
0
3.82848
% Contribtn
6.79865
23.1961
5.35274
0
64.6525
(b) Conduct an analysis of variance. Which factors are important?
Design Expert Output
Response:
Thickness
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.09
3
0.70
A
0.40
1
0.40
B
1.37
1
1.37
F
Value
2.19
1.26
4.31
6-17
Prob > F
0.1425
0.2833
0.0602
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
AB
Residual
Lack of Fit
Pure Error
Cor Total
0.32
3.83
0.000
3.83
5.92
1
12
0
12
15
0.32
0.32
0.99
0.3386
0.32
The "Model F-value" of 2.19 implies the model is not significant relative to the noise. There is a
14.25 % chance that a "Model F-value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case there are no significant model terms.
(c) Write down a regression equation that could be used to predict epitaxial layer thickness over the
region of arsenic flow rate and deposition time used in this experiment.
Design Expert Output
Final Equation in Terms of Coded Factors:
Thickness
+14.51
-0.16
+0.29
+0.14
=
*A
*B
*A*B
Final Equation in Terms of Actual Factors:
Thickness
+37.62656
-0.43119
-1.48735
+0.028150
=
* Flow Rate
* Dep Time
* Flow Rate * Dep Time
(d) Analyze the residuals. Are there any residuals that should cause concern? Observation #2 falls
outside the groupings in the normal probability plot and the plot of residual versus predicted.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .6 4 47 5
99
1 .0 8 02 5
90
80
70
R es idua ls
N orm al % probability
95
50
0 .5 1 57 5
30
20
10
-0 .0 48 7 5
5
1
-0 .6 13 2 5
-0 .6 13 2 5
-0 .0 48 7 5
0 .5 1 57 5
1 .0 8 02 5
1 .6 4 47 5
1 3 .9 2
1 4 .1 5
R es idua l
1 4 .3 7
1 4 .6 0
1 4 .8 2
Predicted
(e) Discuss how you might deal with the potential outlier found in part (d).
One approach would be to replace the observation with the average of the observations from that
experimental cell. Another approach would be to identify if there was a recording issue in the original
data. The first analysis below replaces the data point with the average of the other three. The second
analysis assumes that the reading was incorrectly recorded and should have been 14.165.
6-18
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Analysis with the run associated with standard order 2 replaced with the average of the remaining three
runs in the cell, 13.972:
Design Expert Output
Response:
Thickness
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.97
3
0.99
A
7.439E-003 1
7.439E-003
B
2.96
1
2.96
AB
2.176E-004 1
2.176E-004
Pure Error
0.22
12
0.018
Cor Total
3.19
15
F
Value
53.57
0.40
160.29
0.012
Prob > F
< 0.0001
0.5375
< 0.0001
0.9153
significant
The Model F-value of 53.57 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B are significant model terms.
Final Equation in Terms of Coded Factors:
Thickness
+14.38
-0.022
+0.43
+3.688E-003
=
*A
*B
*A*B
Final Equation in Terms of Actual Factors:
Thickness
+13.36650
-0.020000
+0.12999
+7.37500E-004
=
* Flow Rate
* Dep Time
* Flow Rate * Dep Time
No rm a l p lo t o f re sid ua ls
Re sid ua ls vs. P re d icte d
0 .1 4 3
99
0 .0 1 3 7 5
90
2
80
R es iduals
N orm al % probab ility
95
70
50
-0 .1 1 5 5
30
20
10
-0 .2 4 4 7 5
5
1
-0 .3 7 4
-0 .3 7 4
-0 .2 4 4 7 5
-0 .1 1 5 5
0 .0 1 3 7 5
0 .1 4 3
1 3 .9 2
R es idual
1 4 .1 5
1 4 .3 7
Pred icte d
A new outlier is present and should be investigated.
Analysis with the run associated with standard order 2 replaced with the value 14.165:
6-19
1 4 .6 0
1 4 .8 2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Thickness
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.82
3
0.94
A
0.018
1
0.018
B
2.80
1
2.80
AB
3.969E-003 1
3.969E-003
Pure Error
0.25
12
0.021
Cor Total
3.07
15
F
Value
45.18
0.87
134.47
0.19
Prob > F
< 0.0001
0.3693
< 0.0001
0.6699
significant
The Model F-value of 45.18 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B are significant model terms.
Final Equation in Terms of Coded Factors:
Thickness
+14.39
-0.034
+0.42
+0.016
=
*A
*B
*A*B
Final Equation in Terms of Actual Factors:
Thickness
+15.50156
-0.056188
-0.012350
+3.15000E-003
=
* Flow Rate
* Dep Time
* Flow Rate * Dep Time
No rm a l p lo t o f re sid ua ls
Re sid ua ls vs. P re d icte d
3 .0 0
99
Studen tize d R es id uals
N orm a l % proba bility
95
90
80
70
50
30
20
10
5
1 .5 0
0 .0 0
-1 .5 0
1
-3 .0 0
-3 .0 0
-1 .9 6
-0 .9 2
0 .1 2
1 .1 6
1 3 .9 2
Stude ntize d R es idua ls
1 4 .1 5
1 4 .3 7
1 4 .6 0
1 4 .8 2
Pred icte d
Another outlier is present and should be investigated.
6-13 Continuation of Problem 6-12. Use the regression model in part (c) of Problem 6-12 to generate a
response surface contour plot for epitaxial layer thickness. Suppose it is critically important to obtain
6-20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
layer thickness of 14.5 mm.
recommend?
What settings of arsenic flow rate and deposition time would you
Arsenic flow rate may be set at any of the experimental levels, while the deposition time should be set at
12.4 minutes.
DE S IG N-E X P E RT P l o t
1 5 .0 0
Thickne ss
4
T h i c kn e ss
T h i ckn e ss
X = A : Fl o w Ra te
Y = B : De p T i m e
De si g n P o i n ts
D ep Tim e
B - 1 0 .0 00
B + 1 5 .0 00
1 5 .4 95 3
Thicknes s
De si g n P o i n ts
1 2 .5 0
D ep Tim e
1 6 .1 65
X = A : Fl o w Ra te
Y = B : De p T i m e
14.6742
1 3 .7 5
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
4
1 4 .8 25 7
14.5237
14.3731
1 4 .1 56
1 1 .2 5
14.2226
14.072
4
1 3 .4 86 4
4
1 0 .0 0
5 5 .0 0
5 6 .0 0
5 7 .0 0
5 8 .0 0
5 9 .0 0
5 5 .0 0
5 6 .0 0
Flow R ate
5 7 .0 0
5 8 .0 0
5 9 .0 0
Flow R ate
6-14 Continuation of Problem 6-13. How would your answer to Problem 6-13 change if arsenic flow
rate was more difficult to control in the process than the deposition time?
Running the process at a high level of Deposition Time there is no change in thickness as flow rate
changes.
6-15 A nickel-titanium alloy is used to make components for jet turbine aircraft engines. Cracking is a
potentially serious problem in the final part, as it can lead to non-recoverable failure. A test is run at the
parts producer to determine the effects of four factors on cracks. The four factors are pouring temperature
(A), titanium content (B), heat treatment method (C), and the amount of grain refiner used (D). Two
replicated of a 24 design are run, and the length of crack (in Pm) induced in a sample coupon subjected to
a standard test is measured. The data are shown below:
A
B
C
D
Treatment
Combination
Replicate
I
Replicate
II
+
-
-
-
(1)
a
7.037
14.707
6.376
15.219
-
+
-
-
b
11.635
12.089
+
+
-
-
ab
17.273
17.815
-
-
+
-
c
10.403
10.151
+
-
+
-
ac
4.368
4.098
-
+
+
-
bc
9.360
9.253
+
+
+
-
abc
13.440
12.923
-
-
-
+
d
8.561
8.951
6-21
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+
-
-
+
ad
16.867
17.052
-
+
-
+
bd
13.876
13.658
+
+
-
+
abd
19.824
19.639
-
-
+
+
cd
11.846
12.337
+
-
+
+
acd
6.125
5.904
-
+
+
+
bcd
11.190
10.935
+
+
+
+
abcd
15.653
15.053
(a) Estimate the factor effects. Which factors appear to be large?
Design Expert Output
Term
Model
Intercept
Model
A
Model
B
Model
C
Model
D
Model
AB
Model
AC
Error
AD
Error
BC
Error
BD
Error
CD
Model
ABC
Error
ABD
Error
ACD
Error
BCD
Error
ABCD
Effect
3.01888
3.97588
-3.59625
1.95775
1.93412
-4.00775
0.0765
0.096
0.04725
-0.076875
3.1375
0.098
0.019125
0.035625
0.014125
SumSqr
% Contribtn
72.9089
126.461
103.464
30.6623
29.9267
128.496
0.046818
0.073728
0.0178605
0.0472781
78.7512
0.076832
0.00292613
0.0101531
0.00159613
12.7408
22.099
18.0804
5.35823
5.22969
22.4548
0.00818145
0.012884
0.00312112
0.00826185
13.7618
0.0134264
0.00051134
0.00177426
0.000278923
(b) Conduct an analysis of variance. Do any of the factors affect cracking? Use D=0.05.
Design Expert Output
Response:
Crack Lengthin mm x 10^-2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
570.95
15
38.06
468.99
A
72.91
1
72.91
898.34
B
126.46
1
126.46
1558.17
C
103.46
1
103.46
1274.82
D
30.66
1
30.66
377.80
AB
29.93
1
29.93
368.74
AC
128.50
1
128.50
1583.26
AD
0.047
1
0.047
0.58
BC
0.074
1
0.074
0.91
BD
0.018
1
0.018
0.22
CD
0.047
1
0.047
0.58
ABC
78.75
1
78.75
970.33
ABD
0.077
1
0.077
0.95
ACD
2.926E-003 1
2.926E-003
0.036
BCD
0.010
1
0.010
0.13
ABCD
1.596E-003 1
1.596E-003
0.020
Residual
1.30
16
0.081
Lack of Fit
0.000
0
Pure Error
1.30
16
0.081
Cor Total
572.25
31
The Model F-value of 468.99 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, D, AB, AC, ABC are significant model terms.
6-22
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.4586
0.3547
0.6453
0.4564
< 0.0001
0.3450
0.8518
0.7282
0.8902
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) Write down a regression model that can be used to predict crack length as a function of the significant
main effects and interactions you have identified in part (b).
Design Expert Output
Final Equation in Terms of Coded Factors:
Crack Length=
+11.99
+1.51
+1.99
-1.80
+0.98
+0.97
-2.00
+1.57
*A
*B
*C
*D
*A*B
*A*C
*A*B*C
(d) Analyze the residuals from this experiment.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .4 5 48 7 5
99
0 .2 3 26 8 8
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
0 .0 1 05
-0 .2 11 6 8 7
5
1
-0 .4 33 8 7 5
-0 .4 33 8 7 5
-0 .2 11 6 8 7
0 .0 1 05
0 .2 3 26 8 8
0 .4 5 48 7 5
4 .1 9
R es idua l
8 .0 6
1 1 .9 3
1 5 .8 0
1 9 .6 6
Predicted
There is nothing unusual about the residuals.
(e) Is there an indication that any of the factors affect the variability in cracking?
By calculating the range of the two readings in each cell, we can also evaluate the effects of the factors on
variation. The following is the normal probability plot of effects:
6-23
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t
DE S IG N -E X P E RT P l o t
Ra n g e
P o ur T e m p
T i tan i u m C o n ten t
H e at T re a t M e th o d
G ra i n Re fi n e r
99
CD
95
N orm al % probab ility
A:
B:
C:
D:
90
AB
80
70
50
30
20
10
5
1
-0 .1 0
-0 .0 2
0 .0 5
0 .1 3
0 .2 0
Effect
It appears that the AB and CD interactions could be significant. The following is the ANOVA for the
range data:
Design Expert Output
Response:
Range
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.29
2
0.14
AB
0.13
1
0.13
CD
0.16
1
0.16
Residual
0.16
13
0.013
Cor Total
0.45
15
F
Value
11.46
9.98
12.94
Prob > F
0.0014
0.0075
0.0032
significant
The Model F-value of 11.46 implies the model is significant. There is only
a 0.14% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case AB, CD are significant model terms.
Final Equation in Terms of Coded Factors:
Range =
+0.37
+0.089 * A * B
+0.10 * C * D
(f) What recommendations would you make regarding process operations?
Use interaction and/or main effect plots to assist in drawing conclusions. From the interaction plots,
choose A at the high level and B at the high level. In each of these plots, D can be at either level. From
the main effects plot of C, choose C at the high level. Based on the range analysis, with C at the high
level, D should be set at the low level.
From the analysis of the crack length data:
6-24
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Inte ra ctio n Graph
DE S IG N-E X P E RT P l o t
Cra ck L e n g th
1 9 .8 2 4
Inte ra ctio n Graph
DE S IG N-E X P E RT P l o t
B: Titan ium C o nten t
Cra c k L e n g th
X = A : Pour T em p
Y = C : He a t T re a t M e th o d
1 5 .8 9 2 5
B - -1 .0 0 0
B + 1 .0 0 0
A ctu a l Fa cto rs
C: H e a t T re a t M e th o d = 1
D: G ra i n Re fi n e r = 0 .0 0
1 1 .9 6 1
1 5 .8 9 2 5
C1 -1
C2 1
A c tu a l Fa cto rs
B : T i ta n i u m C o n te n t = 0 .0 0
D: G ra i n Re fi n e r = 0 .0 0
1 1 .9 6 1
8 .0 2 9 5
8 .0 2 9 5
4 .0 9 8
4 .0 9 8
C ra ck Leng th
X = A : Pour T em p
Y = B : T i ta n i u m Co n te n t
C ra ck Leng th
C : H ea t Trea t Metho d
1 9 .8 2 4
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
-0 .5 0
A: Pou r Tem p
Cra ck L e n g th
0 .5 0
1 .0 0
A: Pou r Tem p
One F a ctor P lot
DE S IG N -E X P E RT P l o t
0 .0 0
C ube Graph
DE S IG N -E X P E RT P l o t
C ra ck L eng th
Cra c k L e n g th
X = A: Pour T em p
Y = B : T i ta n i u m Co n te n t
Z = C: He a t T re a t M e th o d
1 9 .8 2 4
X = D : G ra i n Re fi n e r
14 .27
A c tu a l Fa cto r
D: G ra i n Re fi n e r = 0 .0 0
C ra ck L eng th
A ctu a l Fa cto rs
1 5 .8 9 2 5
A : P o ur T e m p = 0.0 0
B : T i tan i u m C o n ten t = 0 .0 0
C: H e at T re a t M e th o d = 1
10 .18
B+
12 .81
B: Titaniu m C o ntent
1 1 .9 6 1
8 .0 2 9 5
18 .64
11 .18
5.1 2
C : H eat Trea t Me tho
4 .0 9 8
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
B-
1 .0 0
A-
D : Grain R efiner
From the analysis of the ranges:
6-25
C+
7.7 3
15 .96
A: Pour Tem p
CA+
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Inte ra ctio n Graph
DE S IG N-E X P E RT P l o t
Ra n g e
0 .6 6 1
Inte ra ctio n Graph
DE S IG N-E X P E RT P l o t
B: Titan ium C o nten t
Ra n g e
X = C : He a t T re a t M e th o d
Y = D : G ra i n Re fi n e r
0 .5 2 2 5
B - -1 .0 0 0
B + 1 .0 0 0
A ctu a l Fa cto rs
C: H e a t T re a t M e th o d = 0 .0 0
D: G ra i n Re fi n e r = 0 .0 0
0 .3 8 4
0 .5 2 2 5
D- -1 .0 0 0
D+ 1 .0 0 0
A c tu a l Fa cto rs
A : P o u r T e m p = 0 .0 0
B : T i ta n i u m C o n te n t = 0 .0 0
0 .3 8 4
0 .2 4 5 5
0 .2 4 5 5
0 .1 0 7
0 .1 0 7
R an ge
X = A : Pour T em p
Y = B : T i ta n i u m Co n te n t
R an ge
D : Gra in R efine r
0 .6 6 1
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
A: Pou r Tem p
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
C : H ea t Trea t Metho d
6-16 Continuation of Problem 6-15. One of the variables in the experiment described in Problem 6-15,
heat treatment method (c), is a categorical variable. Assume that the remaining factors are continuous.
(a) Write two regression models for predicting crack length, one for each level of the heat treatment
method variable. What differences, if any, do you notice in these two equations?
Design Expert Output
Final Equation in Terms of Coded Factors
Heat Treat Method
Crack Length
+13.78619
+3.51331
+1.93994
+0.97888
-0.60169
-1
=
Heat Treat Method
Crack Length
+10.18994
-0.49444
+2.03594
+0.97888
+2.53581
1
=
* Pour Temp
* Titanium Content
* Grain Refiner
* Pour Temp * Titanium Content
* Pour Temp
* Titanium Content
* Grain Refiner
* Pour Temp * Titanium Content
(b) Generate appropriate response surface contour plots for the two regression models in part (a).
6-26
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
C ra ck L e ng th
DE S IG N -E X P E RT P l o t
1 .0 0
DE S IG N -E X P E RT P l o t
1 .0 0
Cra ck L e n g th
X = A: Pour T em p
Y = B : T i ta n i u m Co n te n t
18
A c tu a l Fa cto rs
C: H e at T re a t M e th o d = 10 .5 0
D: G ra i n Re fi n e r = 0 .0 0
14
0 .0 0
-0 .5 0
12
B: Titaniu m C o ntent
B: Titaniu m C o ntent
A ctu a l Fa cto rs
C: H e at T re a t M e th o d = -10 .5 0
D: G ra i n Re fi n e r = 0 .0 0
C ra ck L e ng th
Cra c k L e n g th
X = A: Pour T em p
Y = B : T i ta n i u m Co n te n t
16
12
0 .0 0
10
-0 .5 0
10
8
6
-1 .0 0
-1 .0 0
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
A: Pour Tem p
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
A: Pour Tem p
(c) What set of conditions would you recommend for the factors A, B and D if you use heat treatment
method C=+?
High level of A, low level of B, and low level of D.
(d) Repeat part (c) assuming that you wish to use heat treatment method C=-.
Low level of A, low level of B, and low level of D.
6-17 An experimenter has run a single replicate of a 24 design. The following effect estimates have been
calculated:
A = 76.95
B = -67.52
C = -7.84
D = -18.73
AB = -51.32
AC = 11.69
AD = 9.78
BC = 20.78
BD = 14.74
CD = 1.27
(a) Construct a normal probability plot of these effects.
The plot from Minitab follows.
6-27
ABC = -2.82
ABD = -6.50
ACD = 10.20
BCD = -7.98
ABCD = -6.25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot for Value
ML Estimates
99
ML Estimates
A
95
Mean
-1.57
StDev
31.2296
90
Goodness of Fit
Percent
80
AD*
70
60
50
40
30
20
AB
10
5
1.465
B
1
-50
0
50
Data
(b) Identify a tentative model, based on the plot of the effects in part (a).
Intercept 38.475 x A 33.76 x B 25.66 x A x B
ŷ
6-18 An article in Solid State Technology (“Orthogonal Design for Process Optimization and Its
Application in Plasma Etching,” May 1987, pp. 127-132) describes the application of factorial designs in
developing a nitride etch process on a single-wafer plasma etcher. The process uses C2F6 as the reactant
gas. Four factors are of interest: anode-cathode gap (A), pressure in the reactor chamber (B), C2F6 gas
flow (C), and power applied to the cathode (D). The response variable of interest is the etch rate for
silicon nitride. A single replicate of a 24 design in run, and the data are shown below:
Run
Actual
Run
Etch
Rate
Number
Order
A
B
C
D
(A/min)
Factor
Levels
Low (-)
High (+)
1
2
13
8
+
-
-
-
550
669
A (cm)
B (mTorr)
0.80
4.50
1.20
550
3
12
-
+
-
-
604
C (SCCM)
125
200
4
9
+
+
-
-
650
D (W)
275
325
5
4
-
-
+
-
633
6
15
+
-
+
-
642
7
16
-
+
+
-
601
8
3
+
+
+
-
635
9
1
-
-
-
+
1037
10
14
+
-
-
+
749
11
5
-
+
-
+
1052
12
10
+
+
-
+
868
13
11
-
-
+
+
1075
14
2
+
-
+
+
860
15
7
-
+
+
+
1063
6-28
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
16
6
+
+
+
+
729
(a) Estimate the factor effects. Construct a normal probability plot of the factor effects. Which effects
appear large?
No rm a l p lot
DE S IG N-E X P E RT P l o t
E tch R a te
G ap
P re ssure
G as Fl o w
P o we r
99
D
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
20
A
10
5
AD
1
-1 5 3 .6 3
-3 8 .69
7 6 .2 5
1 9 1 .19
3 0 6 .13
Effe ct
(b) Conduct an analysis of variance to confirm your findings for part (a).
Design Expert Output
Response:
Etch Rate in A/min
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
5.106E+005 3
1.702E+005 97.91
A
41310.56
1
41310.56
23.77
D
3.749E+005 1
3.749E+005 215.66
AD
94402.56
1
94402.56
54.31
Residual
20857.75
12
1738.15
Cor Total
5.314E+005 15
Prob > F
< 0.0001
0.0004
< 0.0001
< 0.0001
significant
The Model F-value of 97.91 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, D, AD are significant model terms.
(c) What is the regression model relating etch rate to the significant process variables?
Design Expert Output
Final Equation in Terms of Coded Factors:
Etch Rate
+776.06
-50.81
+153.06
-76.81
=
*A
*D
*A*D
Final Equation in Terms of Actual Factors:
Etch Rate =
-5415.37500
6-29
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+4354.68750
+21.48500
-15.36250
* Gap
* Power
* Gap * Power
(d) Analyze the residuals from this experiment. Comment on the model’s adequacy.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
6 6 .5
99
3 1 .7 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
-3
20
10
-3 7 .75
5
1
-7 2 .5
-7 2 .5
-3 7 .75
-3
3 1 .7 5
6 6 .5
5 9 7 .00
7 1 1 .94
R es idua l
8 2 6 .88
9 4 1 .81
1 0 5 6.7 5
Predicted
The residual versus predicted plot shows a slight football shape indicating very mild inequality of
variance.
(e) If not all the factors are important, project the 24 design into a 2k design with k<4 and conduct that
analysis of variance. The analysis of variance table is the same as in part (b).
Design Expert Output
Response:
Etch Rate in A/min
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
5.106E+005 3
1.702E+005 97.91
A
41310.56
1
41310.56
23.77
B
3.749E+005 1
3.749E+005 215.66
AB
94402.56
1
94402.56
54.31
Residual
20857.75
12
1738.15
Lack of Fit
0.000
0
Pure Error
20857.75
12
1738.15
Cor Total
5.314E+005 15
The Model F-value of 97.91 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
(f) Draw graphs to interpret any significant interactions.
6-30
Prob > F
< 0.0001
0.0004
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
E tch R a te
Po we r
1 0 8 8.8 7
X = A: Gap
Y = B : P o we r
9 5 4 .14 9
De si g n P o i n ts
Etch R ate
B - 2 7 5 .00 0
B + 3 2 5 .00 0
8 1 9 .43 3
6 8 4 .71 6
550
0 .8 0
0 .9 0
1 .0 0
1 .1 0
1 .2 0
Gap
(g) Plot the residuals versus the actual run order. What problems might be revealed by this plot?
Re sid ua ls vs. Run
6 6 .5
R es idua ls
3 1 .7 5
-3
-3 7 .75
-7 2 .5
1
4
7
10
13
16
R un N um ber
The plot of residuals versus run order can reveal trends in the process over time, inequality of variance
with time, and possibly indicate that there may be factors that were not included in the original
experiment.
6-19 Continuation of Problem 6-18. Consider the regression model obtained in part (c) of Problem 618.
(a) Construct contour plots of the etch rate using this model.
6-31
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
E tch Ra te
DE S IG N-E X P E RT P l o t
3 2 5 .00
E tch R a te
X = A: Gap
Y = D: P o we r
980.125
903.5
A ctu al Fa ctors
B : P re ssure = 5 0 0 .0 0 3 1 2 .50
C: G as Fl o w = 1 6 2 .5 0
Po we r
826.875
3 0 0 .00
750.25
2 8 7 .50
673.625
2 7 5 .00
0 .8 0
0 .9 0
1 .0 0
1 .1 0
1 .2 0
Gap
(b) Suppose that it was necessary to operate this process at an etch rate of 800 Å/min. What settings of
the process variables would you recommend?
Run at the low level of anode-cathode gap (0.80 cm) and at a cathode power level of about 286 watts. The
curve is flatter (more robust) on the low end of the anode-cathode variable.
6-20 Consider the single replicate of the 24 design in Example 6-2. Suppose we had arbitrarily decided
to analyze the data assuming that all three- and four-factor interactions were negligible. Conduct this
analysis and compare your results with those obtained in the example. Do you think that it is a good idea
to arbitrarily assume interactions to be negligible even if they are relatively high-order ones?
Design Expert Output
Response:
Etch Rate in A/min
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Prob > F
Model
5.212E+005 10
52123.41
25.58
0.0011
A
41310.56
1
41310.56
20.28
0.0064
B
10.56
1
10.56
5.184E-003 0.9454
C
217.56
1
217.56
0.11
0.7571
D
3.749E+005 1
3.749E+005 183.99
< 0.0001
AB
248.06
1
248.06
0.12
0.7414
AC
2475.06
1
2475.06
1.21
0.3206
AD
94402.56
1
94402.56
46.34
0.0010
BC
7700.06
1
7700.06
3.78
0.1095
BD
1.56
1
1.56
7.669E-004 0.9790
CD
18.06
1
18.06
8.866E-003 0.9286
Residual
10186.81
5
2037.36
Cor Total
5.314E+005 15
significant
The Model F-value of 25.58 implies the model is significant. There is only
a 0.11% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, D, AD are significant model terms.
This analysis of variance identifies the same effects as the normal probability plot of effects approach used
in Example 6-2. In general, it is not a good idea to arbitrarily pool interactions. Use the normal
6-32
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
probability plot of effect estimates as a guide in the choice of which effects to tentatively include in the
model.
6-21 An experiment was run in a semiconductor fabrication plant in an effort to increase yield. Five
factors, each at two levels, were studied. The factors (and levels) were A = aperture setting (small, large),
B = exposure time (20% below nominal, 20% above nominal), C = development time (30 s, 45 s), D =
mask dimension (small, large), and E = etch time (14.5 min, 15.5 min). The unreplicated 25 design
shown below was run.
(1) =
a=
7
9
d=
ad =
8
10
e=
ae =
8
12
de =
ade =
6
10
b=
34
bd =
32
be =
35
bde =
30
ab =
55
abd =
50
abe =
52
abde =
53
c=
16
cd =
18
ce =
15
cde =
15
ac =
20
acd =
21
ace =
22
acde =
20
bc =
40
bcd =
44
bce =
45
bcde =
41
abc =
60
abcd =
61
abce =
65
abcde =
63
(a) Construct a normal probability plot of the effect estimates. Which effects appear to be large?
No rm a l p lot
DE S IG N-E X P E RT P l o t
Yield
A p e rtu re
E xp o su re T i m e
De ve lo p T i m e
M a sk Di m e n si o n
E tch T i m e
B
99
95
N orm al % probability
A:
B:
C:
D:
E:
AB
90
C
A
80
70
50
30
20
10
5
1
-1 .1 9
7 .5 9
1 6 .3 8
2 5 .1 6
3 3 .9 4
Effe ct
(b) Conduct an analysis of variance to confirm your findings for part (a).
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
11585.13
4
2896.28
A
1116.28
1
1116.28
B
9214.03
1
9214.03
C
750.78
1
750.78
AB
504.03
1
504.03
Residual
78.84
27
2.92
Cor Total
11663.97
31
F
Value
991.83
382.27
3155.34
257.10
172.61
6-33
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The Model F-value of 991.83 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
(c) Write down the regression model relating yield to the significant process variables.
Design Expert Output
Final Equation in Terms of Actual Factors:
Aperture small
Yield
=
+0.40625
+0.65000 * Exposure Time
+0.64583 * Develop Time
Aperture
large
Yield
=
+12.21875
+1.04688 * Exposure Time
+0.64583 * Develop Time
(d) Plot the residuals on normal probability paper. Is the plot satisfactory?
No rm a l p lot o f re sid uals
99
N orm al % probability
95
90
80
70
50
30
20
10
5
1
-2 .7 81 2 5
-1 .3 90 6 3 -3 .5 52 7 1 E -0 1 5 1 .3 9 06 2
2 .7 8 12 5
R es idua l
There is nothing unusual about this plot.
(e) Plot the residuals versus the predicted yields and versus each of the five factors. Comment on the
plots.
6-34
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. A p e rture
Re sid ua ls vs. E xpo sure Tim e
2 .7 8 12 5
2
1 .3 9 06 2
3 .5 52 7 1 E -0 1 5
-1 .3 90 6 3
3
-2 .7 81 2 5
2
-1 3
-7
0
7
13
Ape rture
Expos ure Tim e
Re sid ua ls vs. D evelo p Tim e
Re sid ua ls vs. M a sk D im e nsio n
20
2 .7 8 12 5
2
1 .3 9 06 2
3
3 .5 52 7 1 E -0 1 5
2
2
R es idua ls
R es idua ls
2
-2 0
2 .7 8 12 5
3 .5 52 7 1 E -0 1 5
2
2
2
-1 .3 90 6 3
2
2
-1 .3 90 6 3
3
-2 .7 81 2 5
2
-2 .7 81 2 5
30
33
35
38
40
43
45
1
D evelop Tim e
Re sid ua ls vs. E tch Tim e
2
R es idua ls
2
3 .5 52 7 1 E -0 1 5
-1 .3 90 6 3
3
-2 .7 81 2 5
1 4 .5 0
1 4 .7 5
1 5 .0 0
2
Mas k D im en s ion
2 .7 8 12 5
1 .3 9 06 2
2
2
-2 .7 81 2 5
1
1 .3 9 06 2
2
3 .5 52 7 1 E -0 1 5
2
2
2
-1 .3 90 6 3
R es idua ls
3
R es idua ls
1 .3 9 06 2
2 .7 8 12 5
1 5 .2 5
1 5 .5 0
Etch Tim e
6-35
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The plot of residual versus exposure time shows some very slight inequality of variance. There is no
strong evidence of a potential problem.
(f) Interpret any significant interactions.
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
Yield
Ape rture
65
X = B : E xp osu re T i m e
Y = A : A p e rtu re
Yield
5 0 .2 5
A 1 sm a l l
A 2 l a rg e
A ctu al Fa ctors
C: De ve lo p T i m e = 3 7.5 0
D: M a sk Di m e n si o n = S m a3l5l .5
E : E tch T i m e = 1 5 .0 0
2 0 .7 5
6
-2 0 .00
-1 0 .00
0 .0 0
1 0 .0 0
2 0 .0 0
Expos ure Tim e
Factor A does not have as large an effect when B is at its low level as it does when B is at its high level.
(g) What are your recommendations regarding process operating conditions?
For the highest yield, run with B at the high level, A at the high level and C at the high level.
(h) Project the 25 design in this problem into a 2k design in the important factors. Sketch the design and
show the average and range of yields at each run. Does this sketch aid in interpreting the results of
this experiment?
DESIGN-EASE Analysis
Actual Yield
42.5000
R=5
B+
E
x
p
o
s
u
r
e
32.7500
R=5
62.2500
R=5
52.5000
R=5
16.0000
R=3
20.7500
R=2
C+
T
T
i
m
e
B-
7.2500
A- R=2
10.2500 CR=3
A+
Aperture
6-36
D
e
v
e
l
o
p
i
e
m
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
This cube plot aids in interpretation. The strong AB interaction and the large positive effect of C are
clearly evident.
6-22 Continuation of Problem 6-21. Suppose that the experimenter had run four runs at the center
points in addition to the 32 trials in the original experiment. The yields obtained at the center point runs
were 68, 74, 76, and 70.
(a) Reanalyze the experiment, including a test for pure quadratic curvature.
n F nC y F y C
n F nC
SS PureQuadratic
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
11461.09
4
2865.27
A
992.25
1
992.25
B
9214.03
1
9214.03
C
750.78
1
750.78
AB
504.03
1
504.03
Curvature
6114.34
1
6114.34
Residual
242.88
30
8.10
Cor Total
17818.31
35
2
32 4 30.53125 72
32 4
F
Value
353.92
122.56
1138.12
92.74
62.26
755.24
2
6114.337
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
significant
The Model F-value of 353.92 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
(b) Discuss what your next step would be.
Add axial points and fit a second-order model.
6-23 In a process development study on yield, four factors were studied, each at two levels: time (A),
concentration (B), pressure (C), and temperature (D). A single replicate of a 24 design was run, and the
resulting data are shown in the following table:
Run
Actual
Run
Number
Order
Yield
Factor
Levels
Low (-)
High (+)
A
B
C
D
(lbs)
1
5
-
-
-
-
12
3.0
9
+
-
-
-
18
A (h)
B (%)
2.5
2
14
18
3
8
-
+
-
-
13
C (psi)
60
80
4
13
+
+
-
-
16
D (ºC)
225
250
5
3
-
-
+
-
17
6
7
+
-
+
-
15
7
14
-
+
+
-
20
8
1
+
+
+
-
15
9
6
-
-
-
+
10
6-37
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
10
11
+
-
-
+
25
11
2
-
+
-
+
13
12
15
+
+
-
+
24
13
4
-
-
+
+
19
14
16
+
-
+
+
21
15
10
-
+
+
+
17
16
12
+
+
+
+
23
(a) Construct a normal probability plot of the effect estimates. Which factors appear to have large
effects?
No rm a l p lot
DE S IG N-E X P E RT P l o t
Yield
T im e
Co n cen tra ti o n
P re ssure
T e m p e ratu re
99
A
95
N orm al % probability
A:
B:
C:
D:
90
C
80
70
AD
D
50
30
20
10
5
AC
1
-4 .2 5
-2 .0 6
0 .1 3
2 .3 1
4 .5 0
Effe ct
A, C, D and the AC and AD interactions.
(b) Conduct an analysis of variance using the normal probability plot in part (a) for guidance in forming
an error term. What are your conclusions?
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
275.50
5
55.10
A
81.00
1
81.00
C
16.00
1
16.00
D
42.25
1
42.25
AC
72.25
1
72.25
AD
64.00
1
64.00
Residual
16.25
10
1.62
Cor Total
291.75
15
F
Value
33.91
49.85
9.85
26.00
44.46
39.38
Prob > F
< 0.0001
< 0.0001
0.0105
0.0005
< 0.0001
< 0.0001
significant
The Model F-value of 33.91 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, C, D, AC, AD are significant model terms.
(c) Write down a regression model relating yield to the important process variables.
6-38
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Final Equation in Terms of Coded Factors:
Yield
+17.38
+2.25
+1.00
+1.63
-2.13
+2.00
=
*A
*C
*D
*A*C
*A*D
Final Equation in Terms of Actual Factors:
Yield =
+209.12500
-83.50000 * Time
+2.43750 * Pressure
-1.63000 * Temperature
-0.85000 * Time * Pressure
+0.64000 * Time * Temperature
(d) Analyze the residuals from this experiment. Does your analysis indicate any potential problems?
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .3 7 5
99
0 .6 2 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
20
10
2
-0 .1 25
-0 .8 75
5
1
-1 .6 25
-1 .6 25
-0 .8 75
-0 .1 25
0 .6 2 5
1 .3 7 5
1 1 .6 3
R es idua l
1 4 .8 1
1 8 .0 0
Predicted
6-39
2 1 .1 9
2 4 .3 8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. Run
1 .3 7 5
R es idua ls
0 .6 2 5
-0 .1 25
-0 .8 75
-1 .6 25
1
4
7
10
13
16
R un N um ber
There is nothing unusual about the residual plots.
(e) Can this design be collapsed into a 23 design with two replicates? If so, sketch the design with the
average and range of yield shown at each point in the cube. Interpret the results.
DESIGN-EASE Analysis
Actual yield
22.0
R=2
18.0
R=2
D+
t
e
m
p
e
r
a
t
u
r
e
11.5
R=3
24.5
R=1
18.5
R=3
D-
12.5
A- R=1
15.0
R=0
C17.0
R=2 A+
r
p
e
C+
s
s
u
r
e
time
6-24 Continuation of Problem 6-23. Use the regression model in part (c) of Problem 6-23 to generate a
response surface contour plot of yield. Discuss the practical purpose of this response surface plot.
6-40
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The response surface contour plot shows the adjustments in the process variables that lead to an
increasing or decreasing response. It also displays the curvature of the response in the design region,
possibly indicating where robust operating conditions can be found. Two response surface contour plots
for this process are shown below. These were formed from the model written in terms of the original
design variables.
Yie ld
DE S IG N-E X P E RT P l o t
8 0 .0 0
Yie ld
DE S IG N-E X P E RT P l o t
2 5 0 .00
Yield
X = A: T im e
Y = C: P re ssu re
Yield
X = A: T im e
Y = D: T e m p e ratu re
A c tu al Fa c tors
B : Co n c en tra ti o n = 1 6 .0204 3 .75
C: P re ssure = 7 0 .00
Pres s ure
17.8333
7 0 .0 0
19.2917
Tem pera ture
A ctu al Fa ctors
B : Co n cen tra ti o n = 1 6 .0 07 5 .0 0
D: T e m p e ratu re = 2 3 7.5 0
17.8333
2 3 7 .50
16.375
19.2917
16.375
14.9167
6 5 .0 0
2 3 1 .25
13.4583
6 0 .0 0
2 2 5 .00
2 .5 0
2 .6 3
2 .7 5
2 .8 8
3 .0 0
2 .5 0
2 .6 3
Tim e
2 .7 5
2 .8 8
3 .0 0
Tim e
6-25 The scrumptious brownie experiment. The author is an engineer by training and a firm believer
in learning by doing. I have taught experimental design for many years to a wide variety of audiences and
have always assigned the planning, conduct, and analysis of an actual experiment to the class participants.
The participants seem to enjoy this practical experience and always learn a great deal from it. This
problem uses the results of an experiment performed by Gretchen Krueger at Arizona State University.
There are many different ways to bake brownies. The purpose of this experiment was to determine how
the pan material, the brand of brownie mix, and the stirring method affect the scrumptiousness of
brownies. The factor levels were
Factor
A = pan material
B = stirring method
C = brand of mix
Low (-)
Glass
Spoon
Expensive
High (+)
Aluminum
Mixer
Cheap
The response variable was scrumptiousness, a subjective measure derived from a questionnaire given to
the subjects who sampled each batch of brownies. (The questionnaire dealt with such issues as taste,
appearance, consistency, aroma, and so forth.) An eight-person test panel sampled each batch and filled
out the questionnaire. The design matrix and the response data are shown below:
Brownie
Batch
A
B
C
1
Test
2
Panel
3
Results
4
5
6
7
8
1
-
-
-
11
9
10
10
11
10
8
9
2
+
-
-
15
10
16
14
12
9
6
15
3
-
+
-
9
12
11
11
11
11
11
12
4
+
+
-
16
17
15
12
13
13
11
11
5
-
-
+
10
11
15
8
6
8
9
14
6-41
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
6
+
-
+
12
13
14
13
9
13
14
9
7
-
+
+
10
12
13
10
7
7
17
13
8
+
+
+
15
12
15
13
12
12
9
14
(a) Analyze the data from this experiment as if there were eight replicates of a 23 design. Comment on
the results.
Design Expert Output
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
93.25
7
13.32
A
72.25
1
72.25
B
18.06
1
18.06
C
0.063
1
0.063
AB
0.062
1
0.062
AC
1.56
1
1.56
BC
1.00
1
1.00
ABC
0.25
1
0.25
Residual
338.50
56
6.04
Lack of Fit
0.000
0
Pure Error
338.50
56
6.04
Cor Total
431.75
63
F
Value
2.20
11.95
2.99
0.010
0.010
0.26
0.17
0.041
Prob > F
0.0475
0.0010
0.0894
0.9194
0.9194
0.6132
0.6858
0.8396
significant
The Model F-value of 2.20 implies the model is significant. There is only
a 4.75% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A are significant model terms.
In this analysis, A, the pan material and B, the stirring method, appear to be significant. There are 56
degrees of freedom for the error, yet only eight batches of brownies were cooked, one for each recipe.
(b) Is the analysis in part (a) the correct approach? There are only eight batches; do we really have eight
replicates of a 23 factorial design?
The different rankings by the taste-test panel are not replicates, but repeat observations by differenttesters
on the same batch of brownies. It is not a good idea to use the analysis in part (a) because the estimate of
error may not reflect the batch-to-batch variation.
(c) Analyze the average and standard deviation of the scrumptiousness ratings. Comment on the results.
Is this analysis more appropriate than the one in part (a)? Why or why not?
6-42
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t
DE S IG N -E X P E RT P l o t
A ve ra g e
A : P a n M a te ri a l
B : S ti rri n g M e th o d
C: M i x B ra n d
A : P a n M a te ri a l
B : S ti rri n g M e th o d
C: M i x B ra n d
99
95
95
80
B
70
50
30
20
10
80
70
30
20
10
5
1
1
0 .2 9
0 .9 1
1 .5 2
2 .1 3
AC
-1 .5 7
Effect
Design Expert Output
Response:
Average
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
11.28
2
5.64
A
9.03
1
9.03
B
2.25
1
2.25
Residual
0.37
5
0.074
Cor Total
11.65
7
A
50
5
-0 .3 2
C
90
N orm al % probab ility
N orm al % probab ility
99
A
90
No rm a l p lo t
DE S IG N -E X P E RT P l o t
S td e v
-1 .0 1
-0 .4 5
0 .1 1
0 .6 8
Effect
F
Value
76.13
121.93
30.34
Prob > F
0.0002
0.0001
0.0027
significant
The Model F-value of 76.13 implies the model is significant. There is only
a 0.02% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Design Expert Output
Response:
Stdev
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
6.05
3
2.02
A
0.24
1
0.24
C
0.91
1
0.91
AC
4.90
1
4.90
Residual
0.82
4
0.21
Cor Total
6.87
7
F
Value
9.77
1.15
4.42
23.75
Prob > F
0.0259
0.3432
0.1034
0.0082
significant
The Model F-value of 9.77 implies the model is significant. There is only
a 2.59% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case AC are significant model terms.
Variables A and B affect the mean rank of the brownies. Note that the AC interaction affects the standard
deviation of the ranks. This is an indication that both factors A and C have some effect on the variability
in the ranks. It may also indicate that there is some inconsistency in the taste test panel members. For the
analysis of both the average of the ranks and the standard deviation of the ranks, the mean square error is
6-43
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
now determined by pooling apparently unimportant effects. This is a more estimate of error than obtained
assuming that all observations were replicates.
6-26 An experiment was conducted on a chemical process that produces a polymer. The four factors
studied were temperature (A), catalyst concentration (B), time (C), and pressure (D). Two responses,
molecular weight and viscosity, were observed. The design matrix and response data are shown below:
Run
Actual
Run
Number
Order
Molecular
A
B
C
D
Weight
Viscosity
1
18
-
-
-
-
2400
1400
2
9
+
-
-
-
2410
1500
A (ºC)
B (%)
Factor
Levels
Low (-)
High (+)
100
120
4
8
3
13
-
+
-
-
2315
1520
C (min)
20
30
4
8
+
+
-
-
2510
1630
D (psi)
60
75
5
3
-
-
+
-
2615
1380
6
11
+
-
+
-
2625
1525
7
14
-
+
+
-
2400
1500
8
17
+
+
+
-
2750
1620
9
6
-
-
-
+
2400
1400
10
7
+
-
-
+
2390
1525
11
2
-
+
-
+
2300
1500
12
10
+
+
-
+
2520
1500
13
4
-
-
+
+
2625
1420
14
19
+
-
+
+
2630
1490
15
15
-
+
+
+
2500
1500
16
20
+
+
+
+
2710
1600
17
1
0
0
0
0
2515
1500
18
5
0
0
0
0
2500
1460
19
16
0
0
0
0
2400
1525
20
12
0
0
0
0
2475
1500
(a) Consider only the molecular weight response. Plot the effect estimates on a normal probability scale.
What effects appear important?
6-44
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
DE S IG N-E X P E RT P l o t
M o l ecu l a r Wt
T e m p e ratu re
Ca ta l yst Co n.
T im e
P re ssure
99
H alf N orm a l % prob ability
A:
B:
C:
D:
C
97
95
A
90
AB
85
80
70
60
40
20
0
0 .0 0
5 0 .3 1
1 0 0 .63
1 5 0 .94
2 0 1 .25
Effe ct
A, C and the AB interaction.
(b) Use an analysis of variance to confirm the results from part (a). Is there an indication of curvature?
A,C and the AB interaction are significant. While the main effect of B is not significant, it could be
included to preserve hierarchy in the model. There is no indication of quadratic curvature.
Design Expert Output
Response:
Molecular Wt
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
2.809E+005 3
93620.83
73.00
A
61256.25
1
61256.25
47.76
C
1.620E+005 1
1.620E+005 126.32
AB
57600.00
1
57600.00
44.91
Curvature
3645.00
1
3645.00
2.84
Residual
19237.50
15
1282.50
Lack of Fit
11412.50
12
951.04
0.36
Pure Error
7825.00
3
2608.33
Cor Total
3.037E+005 19
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.1125
not significant
0.9106
not significant
significant
The Model F-value of 73.00 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, C, AB are significant model terms.
(c) Write down a regression model to predict molecular weight as a function of the important variables.
Design Expert Output
Final Equation in Terms of Coded Factors:
Molecular Wt
+2506.25
+61.87
+100.63
+60.00
=
*A
*C
*A*B
(d) Analyze the residuals and comment on model adequacy.
6-45
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
4 2 .5
99
1 0 .6 25
90
80
2
R es idua ls
N orm al % probability
95
70
50
30
-2 1 .25
20
10
-5 3 .12 5
5
1
-8 5
-8 5
-5 3 .12 5
-2 1 .25
1 0 .6 25
4 2 .5
2 2 8 3.7 5
2 3 9 5.0 0
R es idua l
2 5 0 6.2 5
2 6 1 7.5 0
Predicted
There are two residuals that appear to be large and should be investigated.
(e) Repeat parts (a) - (d) using the viscosity response.
No rm a l p lot
DE S IG N-E X P E RT P l o t
V i sco si ty
T e m p e ratu re
Ca ta l yst Co n.
T im e
P re ssure
99
A
95
N orm al % probability
A:
B:
C:
D:
B
90
80
70
50
30
20
10
5
1
-2 5 .00
5 .3 1
3 5 .6 2
6 5 .9 4
9 6 .2 5
Effe ct
Design Expert Output
Response:
Viscosity
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
70362.50
2
35181.25
A
37056.25
1
37056.25
B
33306.25
1
33306.25
Curvature
61.25
1
61.25
Residual
15650.00
16
978.13
Lack of Fit
13481.25
13
1037.02
Pure Error
2168.75
3
722.92
Cor Total
86073.75
19
F
Value
35.97
37.88
34.05
0.063
1.43
6-46
Prob > F
< 0.0001
< 0.0001
< 0.0001
0.8056
not significant
0.4298
not significant
significant
2 7 2 8.7 5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The Model F-value of 35.97 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Final Equation in Terms of Coded Factors:
Viscosity =
+1500.62
+48.13
*A
+45.63
*B
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
3 5 .6 25
99
2
2
3
3 .1 2 5
90
2
80
R es idua ls
N orm al % probability
95
70
50
30
-2 9 .37 5
20
10
-6 1 .87 5
5
1
-9 4 .37 5
-9 4 .37 5
-6 1 .87 5
-2 9 .37 5
3 .1 2 5
3 5 .6 25
1 4 0 6.8 7
R es idua l
1 4 5 3.7 5
1 5 0 0.6 2
1 5 4 7.5 0
1 5 9 4.3 7
Predicted
There is one large residual that should be investigated.
6-27 Continuation of Problem 6-26. Use the regression models for molecular weight and viscosity to
answer the following questions.
(a) Construct a response surface contour plot for molecular weight. In what direction would you adjust
the process variables to increase molecular weight? Increase temperature, catalyst and time.
6-47
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
M o le cula r W t
DE S IG N -E X P E RT P l o t
8 .0 0 2400
M o l e cu l a r W t
X = A : T e m p e ra ture
Y = B : C a ta l yst Co n .
2600
2425
De si g n P o i n ts
B: C atalys t C on.
A ctu a l Fa cto rs
C: T i m e = 2 5 .0 0
D: P ressu re = 6 7 .5 0
7 .0 0
2575
2450
2550
2475
4
6 .0 0
2525
2500
5 .0 0
4 .0 0
1 0 0 .0 0
1 0 5 .0 0
1 1 0 .0 0
1 1 5 .0 0
1 2 0 .0 0
A: Tem p era ture
(a) Construct a response surface contour plot for viscosity. In what direction would you adjust the
process variables to decrease viscosity?
V isco sity
DE S IG N -E X P E RT P l o t
8 .0 0
V i sco si ty
X = A : T e m p e ra ture
Y = B : C a ta l yst Co n .
1575
1550
De si g n P o i n ts
B: C atalys t C on.
A ctu a l Fa cto rs
C: T i m e = 2 5 .0 0
D: P ressu re = 6 7 .5 0
7 .0 0
1525
4
1500
6 .0 0
1475
1450
5 .0 0
1425
4 .0 0
1 0 0 .0 0
1 0 5 .0 0
1 1 0 .0 0
1 1 5 .0 0
1 2 0 .0 0
A: Tem p era ture
Decrease temperature and catalyst.
(c) What operating conditions would you recommend if it was necessary to produce a product with a
molecular weight between 2400 and 2500, and the lowest possible viscosity?
6-48
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Overla y P lot
DE S IG N -E X P E RT P l o t
8 .0 0
O ve rl a y P l o t
X = A : T e m p e ra ture
Y = B : C a ta l yst Co n .
A ctu a l Fa cto rs
C: T i m e = 2 4 .5 0
D: P ressu re = 6 7 .5 0
B: C atalys t C on.
7 .0 0
6 .0 0
5 .0 0
Mole c ular Wt: 25 00
V is c os ity : 1450
4 .0 0
1 0 0 .0 0
1 0 5 .0 0
1 1 0 .0 0
1 1 5 .0 0
1 2 0 .0 0
A: Tem p era ture
Set the temperature between 100 and 105, the catalyst between 4 and 5%, and the time at 24.5 minutes.
The pressure was not significant and can be set at conditions that may improve other results of the process
such as cost.
6-28 Consider the single replicate of the 24 design in Example 6-2. Suppose that we ran five points at
the center (0,0,0,0) and observed the following responses: 73, 75, 71, 69, and 76. Test for curvature in
this experiment. Interpret the results.
Design Expert Output
Response:
Filtration Rate
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
5535.81
5
1107.16
A
1870.56
1
1870.56
C
390.06
1
390.06
D
855.56
1
855.56
AC
1314.06
1
1314.06
AD
1105.56
1
1105.56
Curvature
28.55
1
28.55
Residual
227.93
14
16.28
Lack of Fit
195.13
10
19.51
Pure Error
32.80
4
8.20
Cor Total
5792.29
20
F
Value
68.01
114.90
23.96
52.55
80.71
67.91
1.75
Prob > F
< 0.0001
< 0.0001
0.0002
< 0.0001
< 0.0001
< 0.0001
0.2066
not significant
2.38
0.2093
not significant
The Model F-value of 68.01 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, C, D, AC, AD are significant model terms.
The "Curvature F-value" of 1.75 implies the curvature (as measured by difference between the
average of the center points and the average of the factorial points) in the design space is not
significant relative to the noise. There is a 20.66% chance that a "Curvature F-value"
this large could occur due to noise.
There is no indication of curvature.
6-49
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
6-29 A missing value in a 2k factorial. It is not unusual to find that one of the observations in a 2k
design is missing due to faulty measuring equipment, a spoiled test, or some other reason. If the design is
replicated n times (n>1) some of the techniques discussed in Chapter 14 can be employed, including
estimating the missing observations. However, for an unreplicated factorial (n-1) some other method must
be used. One logical approach is to estimate the missing value with a number that makes the highestorder interaction contrast zero. Apply this technique to the experiment in Example 6-2 assuming that run
ab is missing. Compare the results with the results of Example 6-2.
Treatm ent
Com bination
(1)
a
b
ab
c
ac
bc
abc
d
ad
bd
abd
cd
acd
bcd
abcd
R esponse
45
71
48
m issing
68
60
80
65
43
100
45
104
75
86
70
96
R esponse *
ABCD
ABCD
45
1
-71
-1
-48
-1
m issing * 1
1
-68
-1
60
1
80
1
-65
-1
-43
-1
100
1
45
1
-104
-1
75
1
-86
-1
-70
-1
96
1
A
B
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
C
-1
-1
1
1
-1
-1
1
1
-1
-1
1
1
-1
-1
1
1
D
-1
-1
-1
-1
1
1
1
1
-1
-1
-1
-1
1
1
1
1
Contrast(ABCD )= m issing - 54 = 0
m issing = 54
Substitute the value 54 for the missing run at ab.
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Model
Term
Intercept
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD
Lenth's ME
Lenth's SME
Effect
SumSqr
% Contribtn
20.25
1.75
11.25
16
-1.25
-16.75
18
3.75
1
-2.5
3.25
5.5
-3
-4
0
11.5676
23.4839
1640.25
12.25
506.25
1024
6.25
1122.25
1296
56.25
4
25
42.25
121
36
64
0
27.5406
0.205684
8.50019
17.1935
0.104941
18.8431
21.7605
0.944465
0.067162
0.419762
0.709398
2.03165
0.604458
1.07459
0
6-50
-1
-1
-1
-1
-1
-1
-1
-1
1
1
1
1
1
1
1
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t
DE S IG N -E X P E RT P l o t
Fi l tra ti o n Ra te
T e m p e ra tu re
P ressu re
C o nce n tra ti o n
S ti rri n g R a te
99
A
95
90
N orm al % probab ility
A:
B:
C:
D:
AD
80
C
70
D
50
30
20
10
5
AC
1
-1 6 .7 5
-7 .5 0
1 .7 5
1 1 .0 0
2 0 .2 5
Effect
6-30 An engineer has performed an experiment to study the effect of four factors on the surface
roughness of a machined part. The factors (and their levels) are A = tool angle (12 degrees, 15 degrees),
B = cutting fluid viscosity (300, 400), C = feed rate (10 in/min, 15 in/min), and D = cutting fluid cooler
used (no, yes). The data from this experiment (with the factors coded to the usual -1, +1 levels) are shown
below.
Run
1
A
-
B
-
C
-
D
-
Surface Roughness
0.00340
2
+
-
-
-
0.00362
3
-
+
-
-
0.00301
4
+
+
-
-
0.00182
5
-
-
+
-
0.00280
6
+
-
+
-
0.00290
7
-
+
+
-
0.00252
8
+
+
+
-
0.00160
9
-
-
-
+
0.00336
10
+
-
-
+
0.00344
11
-
+
-
+
0.00308
12
+
+
-
+
0.00184
13
-
-
+
+
0.00269
14
+
-
+
+
0.00284
15
-
+
+
+
0.00253
16
+
+
+
+
0.00163
(a) Estimate the factor effects. Plot the effect estimates on a normal probability plot and select a tentative
model.
6-51
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
S u rfa ce Ro u g h n e ss
T ool Angle
V i sco si ty
Fe e d R a te
Cu tti n g Fl u i d
99
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
C
20
10
5
A
AB
B
1
-0 .0 00 9
-0 .0 00 6
-0 .0 00 4
-0 .0 00 1
0 .0 0 01
Effe ct
(b) Fit the model identified in part (a) and analyze the residuals. Is there any indication of model
inadequacy?
Design Expert Output
Response:
Surface Roughness
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
6.406E-006 4
1.601E-006
A
8.556E-007 1
8.556E-007
B
3.080E-006 1
3.080E-006
C
1.030E-006 1
1.030E-006
AB
1.440E-006 1
1.440E-006
Residual
1.532E-007 11
1.393E-008
Cor Total
6.559E-006 15
F
Value
114.97
61.43
221.11
73.96
103.38
The Model F-value of 114.97 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
6-52
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .0 0 01 6 6 2 5
99
8 .5 6 25 E -0 0 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
5 E -0 0 6
20
10
-7 .5 62 5 E -0 0 5
5
1
-0 .0 00 1 5 6 2 5
-0 .0 00 1 5 6 2 5-7 .5 62 5 E -0 0 5
5 E -0 0 6
8 .5 6 25 E -0 0 5 0 .0 0 01 6 6 2 5
0 .0 0 15
R es idua l
0 .0 0 20
0 .0 0 25
0 .0 0 30
0 .0 0 35
Predicted
Re sid ua ls vs. To ol A ng le
0 .0 0 01 6 6 2 5
R es idua ls
8 .5 6 25 E -0 0 5
5 E -0 0 6
-7 .5 62 5 E -0 0 5
-0 .0 00 1 5 6 2 5
12
13
14
15
Too l Angle
The plot of residuals versus predicted shows a slight “u-shaped” appearance in the residuals, and the plot of
residuals versus tool angle shows an outward-opening funnel.
(c) Repeat the analysis from parts (a) and (b) using 1/y as the response variable. Is there and indication
that the transformation has been useful?
The plots of the residuals are more representative of a model that does not violate the constant variance
assumption.
6-53
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
1 .0 /(S u rfa ce Ro u gh n e ss)
T ool Angle
V i sco si ty
Fe e d R a te
Cu tti n g Fl u i d
99
B
95
N orm al % probability
A:
B:
C:
D:
AB
A
90
C
80
70
50
30
20
10
5
1
-8 .3 0
3 1 .1 4
7 0 .5 9
1 1 0 .04
1 4 9 .49
Effe ct
Design Expert Output
Response:
Surface RoughnessTransform: Inverse
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.059E+005 4
51472.28
A
42610.92
1
42610.92
B
89386.27
1
89386.27
C
18762.29
1
18762.29
AB
55129.62
1
55129.62
Residual
388.94
11
35.36
Cor Total
2.063E+005 15
F
Value
1455.72
1205.11
2527.99
530.63
1559.16
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 1455.72 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
DE S IG N-E X P E RT P l o t
1 .0 /(S u rfa ce Ro u gh n e ss)
Re sid ua ls vs. P re d icted
DE S IG N-E X P E RT P l o t
1 .0 /(S u rfa c e Ro u gh n e ss)
4 .9 1 40 4
4 .9 1 40 4
Re sid ua ls vs. To ol A ng le
R es idua ls
8 .9 7 13
R es idua ls
8 .9 7 13
0 .8 5 67 9 1
0 .8 5 67 9 1
-3 .2 00 4 6
-3 .2 00 4 6
-7 .2 57 7 1
-7 .2 57 7 1
2 8 1 .73
3 6 5 .57
4 4 9 .41
5 3 3 .26
6 1 7 .10
12
Predicted
13
14
Too l Angle
6-54
15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(d) Fit a model in terms of the coded variables that can be used to predict the surface roughness. Convert
this prediction equation into a model in the natural variables.
Design Expert Output
Final Equation in Terms of Coded Factors:
1.0/(Surface Roughness)
+397.81
+51.61
*A
+74.74
*B
+34.24
*C
+58.70
*A*B
=
6-31 Resistivity on a silicon wafer is influenced by several factors.
experiment performed during a critical process step is shown below.
Run
A
B
C
D
Resistivity
1
2
+
-
-
-
1.92
11.28
3
-
+
-
-
1.09
4
+
+
-
-
5.75
5
-
-
+
-
2.13
6
+
-
+
-
9.53
7
-
+
+
-
1.03
8
+
+
+
-
5.35
9
-
-
-
+
1.60
10
+
-
-
+
11.73
11
-
+
-
+
1.16
12
+
+
-
+
4.68
13
-
-
+
+
2.16
14
+
-
+
+
9.11
15
-
+
+
+
1.07
16
+
+
+
+
5.30
The results of a 24 factorial
(a) Estimate the factor effects. Plot the effect estimates on a normal probability plot and select a tentative
model.
6-55
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
Re si sti vi ty
A
B
C
D
99
A
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
20
AB
10
5
B
1
-3 .0 0
-0 .6 7
1 .6 6
3 .9 9
6 .3 2
Effe ct
(b) Fit the model identified in part (a) and analyze the residuals. Is there any indication of model
inadequacy?
The normal probability plot of residuals is not satisfactory. The plots of residual versus predicted, residual
versus factor A, and the residual versus factor B are funnel shaped indicating non-constant variance.
Design Expert Output
Response:
Resistivity
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
214.22
3
71.41
A
159.83
1
159.83
B
36.09
1
36.09
AB
18.30
1
18.30
Residual
5.76
12
0.48
Cor Total
219.98
15
F
Value
148.81
333.09
75.21
38.13
The Model F-value of 148.81 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
6-56
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .3 1 75
99
0 .6 6 25
90
80
R es idua ls
N orm al % probability
95
70
50
30
0 .0 0 75
20
10
-0 .6 47 5
5
1
-1 .3 02 5
-0 .6 47 5
0 .0 0 75
0 .6 6 25
1 .3 1 75
1 .0 9
5 .7 5
8 .0 8
Predicted
Re sid ua ls vs. A
Re sid ua ls vs. B
1 .3 1 75
1 .3 1 75
0 .6 6 25
0 .6 6 25
0 .0 0 75
-0 .6 47 5
-1 .3 02 5
-1 .3 02 5
0
1
-1
A
1 0 .4 1
0 .0 0 75
-0 .6 47 5
-1
3 .4 2
R es idua l
R es idua ls
R es idua ls
-1 .3 02 5
0
1
B
(c) Repeat the analysis from parts (a) and (b) using ln(y) as the response variable. Is there any indication
that the transformation has been useful?
6-57
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
L n (Re si sti vi ty)
A
B
C
D
99
A
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
20
10
B
5
1
-0 .6 3
-0 .0 6
0 .5 0
1 .0 6
1 .6 3
Effe ct
Design Expert Output
Response:
Resistivity Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
12.15
2
6.08
A
10.57
1
10.57
B
1.58
1
1.58
Residual
0.14
13
0.011
Cor Total
12.30
15
Constant:
0.000
F
Value
553.44
962.95
143.94
Prob > F
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 553.44 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
The transformed data no longer indicates that the AB interaction is significant. A simpler model has
resulted from the log transformation.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .1 4 95 8 5
99
0 .0 5 79 8 3 3
90
80
R es idua ls
N orm al % probability
95
70
50
-0 .0 33 6 1 8
30
20
10
-0 .1 25 2 1 9
5
1
-0 .2 16 8 2 1
-0 .2 16 8 2 1
-0 .1 25 2 1 9
-0 .0 33 6 1 8
0 .0 5 79 8 3 3
0 .1 4 95 8 5
0 .0 6
R es idua l
0 .6 2
1 .1 9
Predicted
6-58
1 .7 5
2 .3 1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. A
Re sid ua ls vs. B
0 .0 5 79 8 3 3
0 .0 5 79 8 3 3
R es idua ls
0 .1 4 95 8 5
R es idua ls
0 .1 4 95 8 5
-0 .0 33 6 1 8
-0 .0 33 6 1 8
-0 .1 25 2 1 9
-0 .1 25 2 1 9
-0 .2 16 8 2 1
-0 .2 16 8 2 1
-1
0
1
-1
0
A
1
B
The residual plots are much improved.
(d) Fit a model in terms of the coded variables that can be used to predict the resistivity.
Design Expert Output
Final Equation in Terms of Coded Factors:
Ln(Resistivity)
+1.19
+0.81
*A
-0.31
*B
=
6.32 Continuation of Problem 6-31. Suppose that the experiment had also run four center points along
with the 16 runs in Problem 6-31. The resistivity measurements at the center points are: 8.15, 7.63, 8.95,
6.48. Analyze the experiment again incorporating the center points. What conclusions can you draw
now?
No rm a l p lot
DE S IG N-E X P E RT P l o t
Re si sti vi ty
A
B
C
D
99
A
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
20
AB
10
5
B
1
-3 .0 0
-0 .6 7
1 .6 6
Effe ct
6-59
3 .9 9
6 .3 2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Resistivity
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
214.22
3
71.41
A
159.83
1
159.83
B
36.09
1
36.09
AB
18.30
1
18.30
Curvature
31.19
1
31.19
Residual
8.97
15
0.60
Lack of Fit
5.76
12
0.48
Pure Error
3.22
3
1.07
Cor Total
254.38
19
F
Value
119.35
267.14
60.32
30.58
52.13
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
0.45
0.8632
not significant
significant
The Model F-value of 119.35 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .3 1 75
99
0 .6 5 75
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
-0 .0 02 5
-0 .6 62 5
5
1
-1 .3 22 5
-1 .3 22 5
-0 .6 62 5
-0 .0 02 5
0 .6 5 75
1 .3 1 75
1 .0 9
R es idua l
3 .4 2
5 .7 5
Predicted
Repeated analysis with the natural log transformation.
6-60
8 .0 8
1 0 .4 1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
L n (Re si sti vi ty)
A
B
C
D
99
A
95
N orm al % probability
A:
B:
C:
D:
90
80
70
50
30
20
10
5
B
1
-0 .6 3
-0 .0 6
0 .5 0
1 .0 6
1 .6 3
Effe ct
Design Expert Output
Response:
Resistivity Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
12.15
2
6.08
A
10.57
1
10.57
B
1.58
1
1.58
Curvature
2.38
1
2.38
Residual
0.20
16
0.012
Lack of Fit
0.14
13
0.011
Pure Error
0.056
3
0.019
Cor Total
14.73
19
Constant:
0.000
F
Value
490.37
853.20
127.54
191.98
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.59
0.7811
significant
significant
not significant
The Model F-value of 490.37 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
The "Curvature F-value" of 191.98 implies there is significant curvature (as measured by
difference between the average of the center points and the average of the factorial points) in
the design space. There is only a 0.01% chance that a "Curvature F-value" this large
could occur due to noise.
The curvature test indicates that the model has significant pure quadratic curvature.
6.33 Often the fitted regression model from a 2k factorial design is used to make predictions at points of
interest in the design space.
(a) Find the variance of the predicted response ŷ at the point x1 , x 2 ,…, x k in the design space. Hint:
Remember that the x’s are coded variables, and assume a 2k design with an equal number of replicates
V2
n at each design point so that the variance of a regression coefficient Ê is
and that the
n2 k
covariance between any pair of regression coefficients is zero.
Let’s assume that the model can be written as follows:
6-61
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
yˆ (x)=Eˆ0 Eˆ1 x1 Eˆ 2 x2 ... Eˆ p x p
where xc
[ x1 , x2 ,..., xk ] are the values of the original variables in the design at the point of interest
where a prediction is required , and the variables in the model x1 , x2 ,..., x p potentially include interaction
terms among the original k variables. Now the variance of the predicted response is
V [ yˆ (x)] V ( Eˆ0 Eˆ1 x1 Eˆ2 x2 ... Eˆ p x p )
V ( Eˆ0 ) V ( Eˆ1 x1 ) V ( Eˆ2 x2 ) ... V ( Eˆ p x p )
p
·
V2 §
1 ¦ xi2 ¸
k ¨
n2 © i 1 ¹
This result follows because the design is orthogonal and all model parameter estimates have the same
variance. Remember that some of the x’s involved in this equation are potentially interaction terms.
(b) Use the result of part (a) to find an equation for a 100(1-D)% confidence interval on the true mean
response at the point x1 , x 2 ,…, x k in the design space.
The confidence interval is
yˆ (x) tD / 2, df E V [ yˆ (x)] d y (x) d yˆ (x) tD / 2, df E V [ yˆ (x)]
where dfE is the number of degrees of freedom used to estimate V and the estimate of V has been
used in computing the variance of the predicted value of the response at the point of interest.
2
2
6.34 Hierarchical Models. Several times we have utilized the hierarchy principal in selecting a model;
that is, we have included non-significant terms in a model because they were factors involved in
significant higher-order terms. Hierarchy is certainly not an absolute principle that must be followed in
all cases. To illustrate, consider the model resulting from Problem 6-1, which required that a nonsignificant main effect be included to achieve hierarchy. Using the data from Problem 6-1:
(a) Fit both the hierarchical model and the non-hierarchical model.
Design Expert Output for Hierarchial Model
Response:
Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1519.67
4
379.92
A
0.67
1
0.67
B
770.67
1
770.67
C
280.17
1
280.17
AC
468.17
1
468.17
Residual
575.67
19
30.30
Lack of Fit
93.00
3
31.00
Pure Error
482.67
16
30.17
Cor Total
2095.33
23
F
Value
12.54
0.022
25.44
9.25
15.45
1.03
The Model F-value of 12.54 implies the model is significant. There is only
6-62
Prob > F
< 0.0001
0.8836
< 0.0001
0.0067
0.0009
0.4067
significant
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B, C, AC are significant model terms.
Std. Dev.
Mean
C.V.
PRESS
5.50
40.83
13.48
918.52
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.7253
0.6674
0.5616
10.747
The "Pred R-Squared" of 0.5616 is in reasonable agreement with the "Adj R-Squared" of
0.6674. A difference greater than 0.20 between the "Pred R-Squared" and the "Adj R-Squared"
indicates a possible problem with your model and/or data.
Design Expert Output for Non-Hierarchical Model
Response: Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source Squares DF
Square
Value
Prob > F
Model 1519.00 3
506.33
17.57
< 0.0001 significant
B
770.67
1
770.67
26.74
< 0.0001
C
280.17
1
280.17
9.72
0.0054
AC
468.17
1
468.17
16.25
0.0007
Residual
576.33
20
28.82
Lack of Fit
93.67
4
23.42
0.78
0.5566
not significant
Pure Error
482.67
16
30.17
Cor Total
2095.33 23
The Model F-value of 17.57 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B, C, AC are significant model terms.
The "Lack of Fit F-value" of 0.78 implies the Lack of Fit is not significant relative to the pure
error. There is a 55.66% chance that a "Lack of Fit F-value" this large could occur due
to noise. Non-significant lack of fit is good -- we want the model to fit.
Std. Dev.
Mean
C.V.
PRESS
5.37
40.83
13.15
829.92
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.7249
0.6837
0.6039
12.320
The "Pred R-Squared" of 0.6039 is in reasonable agreement with the "Adj R-Squared" of
0.6837. A difference greater than 0.20 between the "Pred R-Squared" and the "Adj R-Squared"
indicates a possible problem with your model and/or data.
(b) Calculate the PRESS statistic, the adjusted R2 and the mean square error for both models.
The PRESS and R2 are in the Design Expert Output above. The PRESS is smaller for the nonhierarchical model than the hierarchical model suggesting that the non-hierarchical model is a better
predictor.
(c) Find a 95 percent confidence interval on the estimate of the mean response at a cube corner
( x1 = x 2 = x3 = r1 ). Hint: Use the result of Problem 6-33.
Design Expert Output
Prediction
Life
27.45
Life
36.17
Life
38.67
Life
47.50
Life
43.00
Life
34.17
SE Mean
2.18
2.19
2.19
2.19
2.19
2.19
95% CI low
22.91
31.60
34.10
42.93
38.43
29.60
95% CI high
31.99
40.74
43.24
52.07
47.57
38.74
6-63
SE Pred
5.79
5.80
5.80
5.80
5.80
5.80
95% PI low
15.37
24.07
26.57
35.41
30.91
22.07
95% PI high
39.54
48.26
50.76
59.59
55.09
46.26
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Life
Life
54.33
45.50
2.19
2.19
49.76
40.93
58.90
50.07
5.80
5.80
42.24
33.41
66.43
57.59
(d) Based on the analyses you have conducted, which model would you prefer?
Notice that PRESS is smaller and the adjusted R2 is larger for the non-hierarchical model. This is an
indication that strict adherence to the hierarchy principle isn’t always necessary. Note also that the
confidence interval is shorter for the non-hierarchical model.
6-64
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 7
Blocking and Confounding in the 2k Factorial Design
Solutions
7-1 Consider the experiment described in Problem 6-1. Analyze this experiment assuming that each
replicate represents a block of a single production shift.
Source of
Variation
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
0.67
1
0.67
<1
Tool Geometry (B)
770.67
1
770.67
22.38*
Cutting Angle (C)
280.17
1
280.17
8.14*
AB
16.67
1
16.67
<1
AC
468.17
1
468.17
13.60*
BC
48.17
1
48.17
1.40
ABC
28.17
1
28.17
<1
0.58
2
0.29
Error
482.08
14
34.43
Total
2095.33
23
Cutting Speed (A)
Blocks
Design Expert Output
Response:
Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.58
2
0.29
Model
1519.67
4
379.92
A
0.67
1
0.67
B
770.67
1
770.67
C
280.17
1
280.17
AC
468.17
1
468.17
Residual
575.08
17
33.83
Cor Total
2095.33
23
F
Value
11.23
0.020
22.78
8.28
13.84
Prob > F
0.0001
0.8900
0.0002
0.0104
0.0017
significant
The Model F-value of 11.23 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B, C, AC are significant model terms.
These results agree with the results from Problem 6-1. Tool geometry, cutting angle and the cutting speed
x cutting angle factors are significant at the 5% level. The Design Expert program also includes A, speed,
in the model to preserve hierarchy.
7-2 Consider the experiment described in Problem 6-5. Analyze this experiment assuming that each
one of the four replicates represents a block.
Source of
Variation
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
Bit Size (A)
1107.23
1
1107.23
364.22*
7-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Cutting Speed (B)
227.26
1
227.26
74.76*
AB
303.63
1
303.63
99.88*
Blocks
44.36
3
14.79
Error
27.36
9
3.04
Total
1709.83
15
These results agree with those from Problem 6-5.
significant at the 1% level.
Design Expert Output
Response:
Vibration
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
44.36
3
14.79
Model
1638.11
3
546.04
A
1107.23
1
1107.23
B
227.26
1
227.26
AB
303.63
1
303.63
Residual
27.36
9
3.04
Cor Total
1709.83
15
Bit size, cutting speed and their interaction are
F
Value
Prob > F
179.61
364.21
74.75
99.88
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 179.61 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
7-3 Consider the alloy cracking experiment described in Problem 6-15. Suppose that only 16 runs
could be made on a single day, so each replicate was treated as a block. Analyze the experiment and draw
conclusions.
The analysis of variance for the full model is as follows:
Design Expert Output
Response:
Crack Lengthin mm x 10^-2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Block
0.016
1
0.016
Model
570.95
15
38.06
445.11
A
72.91
1
72.91
852.59
B
126.46
1
126.46
1478.83
C
103.46
1
103.46
1209.91
D
30.66
1
30.66
358.56
AB
29.93
1
29.93
349.96
AC
128.50
1
128.50
1502.63
AD
0.047
1
0.047
0.55
BC
0.074
1
0.074
0.86
BD
0.018
1
0.018
0.21
CD
0.047
1
0.047
0.55
ABC
78.75
1
78.75
920.92
ABD
0.077
1
0.077
0.90
ACD
2.926E-003 1
2.926E-003
0.034
BCD
0.010
1
0.010
0.12
ABCD
1.596E-003 1
1.596E-003
0.019
Residual
1.28
15
0.086
Cor Total
572.25
31
The Model F-value of 445.11 implies the model is significant. There is only
7-2
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.4708
0.3678
0.6542
0.4686
< 0.0001
0.3582
0.8557
0.7352
0.8931
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, D, AB, AC, ABC are significant model terms.
The analysis of variance for the reduced model based on the significant factors is shown below. The BC
interaction was included to preserve hierarchy.
Design Expert Output
Response:
Crack Lengthin mm x 10^-2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.016
1
0.016
Model
570.74
8
71.34
A
72.91
1
72.91
B
126.46
1
126.46
C
103.46
1
103.46
D
30.66
1
30.66
AB
29.93
1
29.93
AC
128.50
1
128.50
BC
0.074
1
0.074
ABC
78.75
1
78.75
Residual
1.49
22
0.068
Cor Total
572.25
31
F
Value
Prob > F
1056.10
1079.28
1872.01
1531.59
453.90
443.01
1902.15
1.09
1165.76
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.3075
< 0.0001
significant
The Model F-value of 1056.10 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, D, AB, AC, ABC are significant model terms.
Blocking does not change the results of Problem 6-15.
7-4 Consider the data from the first replicate of Problem 6-1. Suppose that these observations could not
all be run using the same bar stock. Set up a design to run these observations in two blocks of four
observations each with ABC confounded. Analyze the data.
Block 1
(1)
ab
ac
bc
Block 2
a
b
c
abc
From the normal probability plot of effects, B, C, and the AC interaction are significant.
Factor A was included in the analysis of variance to preserve hierarchy.
7-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t
DE S IG N -E X P E RT P l o t
L i fe
A : C u tti n g S p e e d
B : T o ol G e o m e try
C: C u tti n g A n g l e
99
95
B
N orm al % probab ility
90
80
C
70
50
30
20
10
5
AC
1
-1 3 .7 5
-7 .1 3
-0 .5 0
6 .1 3
1 2 .7 5
Effect
Design Expert Output
Response:
Life
in hours
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
91.13
1
91.13
Model
896.50
4
224.13
A
3.13
1
3.13
B
325.12
1
325.12
C
190.12
1
190.12
AC
378.13
1
378.13
Residual
61.25
2
30.62
Cor Total
1048.88
7
F
Value
Prob > F
7.32
0.10
10.62
6.21
12.35
0.1238
0.7797
0.0827
0.1303
0.0723
not significant
The "Model F-value" of 7.32 implies the model is not significant relative to the noise. There is a
12.38 % chance that a "Model F-value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case there are no significant model terms.
This design identifies the same significant factors as Problem 6-1.
7-5 Consider the data from the first replicate of Problem 6-7. Construct a design with two blocks of
eight observations each with ABCD confounded. Analyze the data.
Block 1
(1)
ab
ac
bc
ad
bd
cd
abcd
Block 2
a
b
c
d
abc
abd
acd
bcd
7-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The significant effects are identified in the normal probability plot of effects below:
No rm a l p lo t
DE S IG N -E X P E RT P l o t
yi e l d
A
B
C
D
99
D
95
N orm al % probab ility
A:
B:
C:
D:
90
ABD
AB
80
70
50
30
20
AD
10
5
ABC
A
1
-1 0 .0 0
-6 .2 5
-2 .5 0
1 .2 5
5 .0 0
Effect
AC, BC, and BD were included in the model to preserve hierarchy.
Design Expert Output
Response:
yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
42.25
1
42.25
Model
892.25
11
81.11
A
400.00
1
400.00
B
2.25
1
2.25
C
2.25
1
2.25
D
100.00
1
100.00
AB
81.00
1
81.00
AC
1.00
1
1.00
AD
56.25
1
56.25
BC
6.25
1
6.25
BD
9.00
1
9.00
ABC
144.00
1
144.00
ABD
90.25
1
90.25
Residual
25.25
3
8.42
Cor Total
959.75
15
F
Value
Prob > F
9.64
47.52
0.27
0.27
11.88
9.62
0.12
6.68
0.74
1.07
17.11
10.72
0.0438
0.0063
0.6408
0.6408
0.0410
0.0532
0.7531
0.0814
0.4522
0.3772
0.0256
0.0466
significant
The Model F-value of 9.64 implies the model is significant. There is only
a 4.38% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, D, ABC, ABD are significant model terms.
7-6 Repeat Problem 7-5 assuming that four blocks are required.
consequently CD) with blocks.
Block 1
(1)
ab
acd
Block 2
ac
bc
d
Block 3
c
abc
ad
7-5
Confound ABD and ABC (and
Block 4
a
b
cd
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
bcd
abd
bd
No rm a l p lo t
DE S IG N -E X P E RT P l o t
yi e l d
A
B
C
D
99
D
95
N orm al % probab ility
A:
B:
C:
D:
abcd
90
AB
ABC D
80
70
50
30
20
10
5
A
1
-1 0 .0 0
-6 .2 5
-2 .5 0
1 .2 5
5 .0 0
Effect
Design Expert Output
Response:
yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
243.25
3
81.08
Model
623.25
4
155.81
A
400.00
1
400.00
D
100.00
1
100.00
AB
81.00
1
81.00
ABCD
42.25
1
42.25
Residual
93.25
8
11.66
Cor Total
959.75
15
F
Value
Prob > F
13.37
34.32
8.58
6.95
3.62
0.0013
0.0004
0.0190
0.0299
0.0934
significant
The Model F-value of 13.37 implies the model is significant. There is only
a 0.13% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, D, AB are significant model terms.
7-7 Using the data from the 25 design in Problem 6-21, construct and analyze a design in two blocks
with ABCDE confounded with blocks.
Block 1
(1)
ab
ac
bc
ad
bd
cd
abcd
Block 1
ae
be
ce
abce
de
abde
acde
bcde
Block 2
a
b
c
abc
d
abd
acd
bcd
7-6
Block 2
e
abe
ace
bce
ade
bde
cde
abcde
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The normal probability plot of effects identifies factors A, B, C, and the AB interaction as being
significant. This is confirmed with the analysis of variance.
No rm a l p lo t
DE S IG N-E X P E RT P l o t
Yield
A p e rtu re
E xp o su re T i m e
D e ve l o p T i m e
M a sk Di m e n si o n
E tch T i m e
99
B
95
N orm a l % proba bility
A:
B:
C:
D:
E:
C
AB
90
80
A
70
50
30
20
10
5
1
-1 .1 9
7 .5 9
1 6 .3 8
2 5 .1 6
3 3 .9 4
Effect
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
0.28
1
0.28
Model
11585.13
4
2896.28
A
1116.28
1
1116.28
B
9214.03
1
9214.03
C
750.78
1
750.78
AB
504.03
1
504.03
Residual
78.56
26
3.02
Cor Total
11663.97
31
F
Value
Prob > F
958.51
369.43
3049.35
248.47
166.81
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 958.51 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
7-8 Repeat Problem 7-7 assuming that four blocks are necessary. Suggest a reasonable confounding
scheme.
Use ABC, CDE, confounded with ABDE. The four blocks follow.
Block 1
(1)
ab
acd
bcd
ace
bce
Block 2
a
b
cd
abcd
ce
abce
Block 3
ac
bc
d
abd
e
abe
7-7
Block 4
c
abc
ad
bd
ae
be
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
de
abde
ade
bde
acde
bcde
cde
abcde
The normal probability plot of effects identifies the same significant effects as in Problem 7-7.
No rm a l p lo t
DE S IG N -E X P E RT P l o t
Yield
A p ertu re
E xp o su re T i m e
D e ve l o p T i m e
M a sk Di m e n si o n
E tch T i m e
99
B
95
N orm al % probab ility
A:
B:
C:
D:
E:
90
C
AB
80
A
70
50
30
20
10
5
1
-1 .1 9
7 .5 9
1 6 .3 8
2 5 .1 6
3 3 .9 4
Effect
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
13.84
3
4.61
Model
11585.13
4
2896.28
A
1116.28
1
1116.28
B
9214.03
1
9214.03
C
750.78
1
750.78
AB
504.03
1
504.03
Residual
65.00
24
2.71
Cor Total
11663.97
31
F
Value
Prob > F
1069.40
412.17
3402.10
277.21
186.10
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 1069.40 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
7-9 Consider the data from the 25 design in Problem 6-21. Suppose that it was necessary to run this
design in four blocks with ACDE and BCD (and consequently ABE) confounded. Analyze the data from
this design.
Block 1
(1)
ae
cd
abc
acde
Block 2
a
e
acd
bc
cde
Block 3
b
abe
bcd
ac
abcde
7-8
Block 4
c
ace
d
ab
ade
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
bce
abd
bde
abce
bd
abde
ce
ab
de
be
abcd
bcde
Even with four blocks, the same effects are identified as significant per the normal probability plot and
analysis of variance below:
No rm a l p lo t
DE S IG N -E X P E RT P l o t
Yield
A p ertu re
E xp o su re T i m e
D e ve l o p T i m e
M a sk Di m e n si o n
E tch T i m e
99
B
95
N orm al % probab ility
A:
B:
C:
D:
E:
90
C
AB
80
70
A
50
30
20
10
5
1
-1 .1 9
7 .5 9
1 6 .3 7
2 5 .1 6
3 3 .9 4
Effect
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
2.59
3
0.86
Model
11585.13
4
2896.28
A
1116.28
1
1116.28
B
9214.03
1
9214.03
C
750.78
1
750.78
AB
504.03
1
504.03
Residual
76.25
24
3.18
Cor Total
11663.97
31
F
Value
Prob > F
911.62
351.35
2900.15
236.31
158.65
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 911.62 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, C, AB are significant model terms.
7-10 Design an experiment for confounding a 26 factorial in four blocks.
confounding scheme, different from the one shown in Table 7-8.
We choose ABCE and ABDF. Which also confounds with CDEF
Block 1
a
b
Block 2
c
abc
Block 3
ac
bc
7-9
Block 4
(1)
ab
Suggest an appropriate
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
cd
abcd
ace
bce
de
abde
cf
abcf
adf
bdf
ef
abef
acdef
bcdef
ad
bd
e
abe
acde
bcde
af
bf
cdf
abcdf
acef
bcef
def
abdef
d
abd
ae
be
cde
abcde
f
abf
acdf
bcdf
cef
abcef
adef
bdef
acd
bcd
ce
abce
ade
bde
acf
bcf
df
abdf
aef
bef
cdef
abcdef
7-11 Consider the 26 design in eight blocks of eight runs each with ABCD, ACE, and ABEF as the
independent effects chosen to be confounded with blocks. Generate the design. Find the other effects
confound with blocks.
Block 1
Block 2
Block 3
Block 4
Block 5
Block 6
Block 7
Block 8
b
acd
ce
abde
abcf
de
aef
bcdef
abc
d
ae
bcde
bf
acdf
cef
abdef
a
bcd
abce
de
cf
abdf
def
acdef
c
abd
be
acde
af
bcdf
abcef
def
ac
bd
abe
cde
f
abcdf
bcef
adef
(1)
abcd
bce
ade
acf
bdf
abef
cdef
bc
ad
e
abcde
abf
cdf
acef
bdef
ab
cd
ace
bde
bcf
adf
ef
abcdef
The factors that are confounded with blocks are ABCD, ABEF, ACE, BDE, CDEF, BCF, and ADF.
7-12 Consider the 22 design in two blocks with AB confounded. Prove algebraically that SSAB = SSBlocks.
If AB is confounded, the two blocks are:
Block 1
(1)
ab
(1) + ab
SS Blocks
SS Blocks
Block 2
a
b
a+b
> 1 ab@2 >a b@2 > 1 ab a b@2
2
4
1 2 ab2 2 1 ab a 2 b2 2ab
2
7-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1 2 ab 2 a 2 b 2 2 1 ab 2 1 a 2 1 b 2a ab 2b ab 2ab
4
2
2
2
2
1 ab a b 2 1 ab 2ab 2 1 a 2 1 b 2a ab 2b ab
4
1
2
> 1 ab a b@ SS AB
4
SS Blocks
SS Blocks
7-13 Consider the data in Example 7-2. Suppose that all the observations in block 2 are increased by 20.
Analyze the data that would result. Estimate the block effect. Can you explain its magnitude? Do blocks
now appear to be an important factor? Are any other effect estimates impacted by the change you made in
the data?
Block Effect
406 715
8
8
y Block1 y Block 2
309
8
38.625
This is the block effect estimated in Example 7-2 plus the additional 20 units that were added to each
observation in block 2. All other effects are the same.
Source of
Variation
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
A
C
1870.56
390.06
1
1
1870.56
390.06
89.93
18.75
D
855.56
1
855.56
41.13
AC
1314.06
1
1314.06
63.18
AD
1105.56
1
1105.56
53.15
Blocks
5967.56
1
5967.56
Error
187.56
9
20.8
Total
11690.93
15
Design Expert Output
Response:
Filtration in gal/hr
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
5967.56
1
5967.56
Model
5535.81
5
1107.16
A
1870.56
1
1870.56
C
390.06
1
390.06
D
855.56
1
855.56
AC
1314.06
1
1314.06
AD
1105.56
1
1105.56
Residual
187.56
9
20.84
Cor Total
11690.94
15
F
Value
Prob > F
53.13
89.76
18.72
41.05
63.05
53.05
< 0.0001
< 0.0001
0.0019
0.0001
< 0.0001
< 0.0001
The Model F-value of 53.13 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, C, D, AC, AD are significant model terms.
7-11
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
7-14 Suppose that the data in Problem 7-1 we had confounded ABC in replicate I, AB in replicate II, and
BC in replicate III. Construct the analysis of variance table.
Replicate I
(ABC Confounded)
1
2
(1)
a
ab
b
ac
c
bc
abc
Block->
Source of
Variation
Replicate II
(AB Confounded)
1
2
(1)
a
ab
b
abc
ac
c
bc
Replicate III
(BC Confounded)
1
2
(1)
b
bc
c
abc
ab
a
ac
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
A
B
0.67
770.67
1
1
0.67
770.67
<1
20.77
C
280.17
1
280.17
7.55
25.00
1
25.00
<1
468.17
1
468.17
12.62
22.56
1
22.56
<1
0.06
1
0.06
<1
119.83
3
15.87
0.58
2
Error
408.21
11
Total
2095.33
23
AB (reps 1 and III)
AC
BC (reps I and II)
ABC (reps II and III)
Blocks within replicates
Replicates
37.11
7-15 Repeat Problem 7-1 assuming that ABC was confounded with blocks in each replicate.
Block->
Source of
Variation
Replicate I, II, and III
(ABC Confounded)
1
2
(1)
a
ab
b
ac
c
bc
abc
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
A
B
0.67
770.67
1
1
0.67
770.67
<1
22.15
C
280.17
1
280.17
8.05
AB
16.67
1
16.67
<1
AC
468.17
1
468.17
13.46
BC
48.17
1
48.17
1.38
119.83
1
119.83
64.83
4
Blocks (or ABC)
Replicates/Lack of Fit
7-12
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
471.50
12
Total
2095.33
23
34.79
7-16 Suppose that in Problem 7-7 ABCD was confounded in replicate I and ABC was confounded in
replicate II. Perform the statistical analysis of variance.
Source of
Variation
A
B
Sum of
Squares
Degrees of
Freedom
Mean
Square
F0
657.03
13.78
1
1
657.03
13.78
84.89
1.78
C
57.78
1
57.78
7.46
D
124.03
1
124.03
16.02
AB
132.03
1
132.03
17.06
AC
3.78
1
3.78
<1
AD
38.28
1
38.28
4.95
BC
2.53
1
2.53
<1
BD
0.28
1
0.28
<1
CD
22.78
1
22.78
2.94
ABC
144.00
1
144.00
18.64
ABD
175.78
1
175.78
22.71
ACD
7.03
1
7.03
<1
BCD
7.03
1
7.03
<1
10.56
1
10.56
1.36
ABCD
Replicates
11.28
1
11.28
118.81
2
15.35
Error
100.65
13
7.74
Total
1627.47
31
Blocks
7-17 Construct a 23 design with ABC confounded in the first two replicates and BC confounded in the
third. Outline the analysis of variance and comment on the information obtained.
Block->
Replicate I
(ABC Confounded)
1
2
(1)
a
ab
b
ac
c
bc
abc
Replicate II
(ABC Confounded)
1
2
(1)
a
ab
b
ac
c
bc
abc
Source of
Variation
Degrees of
Freedom
A
1
B
1
C
1
7-13
Replicate III
(BC Confounded)
1
2
(1)
b
bc
c
abc
ab
a
ac
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
AB
1
AC
1
BC
1
ABC
1
Replicates
2
Blocks
3
Error
11
Total
23
This design provides “two-thirds” information on BC and “one-third” information on ABC.
7-14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 8
Two-Level Fractional Factorial Designs
Solutions
8-1 Suppose that in the chemical process development experiment in Problem 6-7, it was only possible
to run a one-half fraction of the 24 design. Construct the design and perform the statistical analysis, using
the data from replicate 1.
The required design is a 24-1 with I=ABCD.
A
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
-12
Model
B
-1
Model
C
4
Model
D
-1
Model
AB
6
Model
AC
-1
Model
AD
-5
Error
BC
Aliased
Error
BD
Aliased
Error
CD
Aliased
Error
ABC
Aliased
Error
ABD
Aliased
Error
ACD
Aliased
Error
BCD
Aliased
Error
ABCD
Aliased
Lenth's ME
Lenth's SME 54.0516
B
+
+
+
+
C
+
+
+
+
D=ABC
+
+
+
+
(1)
ad
bd
ab
cd
ac
bc
abcd
SumSqr
% Contribtn
288
2
32
2
72
2
50
64.2857
0.446429
7.14286
0.446429
16.0714
0.446429
11.1607
22.5856
8-1
90
72
87
83
99
81
88
80
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
A
90
85
AB
80
AD
70
60
D
B
40
20
0
0 .0 0
3 .0 0
6 .0 0
9 .0 0
1 2 .0 0
Effe ct
The largest effect is A. The next largest effects are the AB and AD interactions. A plausible tentative
model would be A, AB and AD, along with B and D to preserve hierarchy.
Design Expert Output
Response:
yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
414.00
5
82.80
A
288.00
1
288.00
B
2.00
1
2.00
D
2.00
1
2.00
AB
72.00
1
72.00
AD
50.00
1
50.00
Residual
34.00
2
17.00
Cor Total
448.00
7
F
Value
4.87
16.94
0.12
0.12
4.24
2.94
Prob > F
0.1791
0.0543
0.7643
0.7643
0.1758
0.2285
The "Model F-value" of 4.87 implies the model is not significant relative to the noise. There is a
17.91 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
4.12
85.00
4.85
544.00
Factor
Intercept
A-A
B-B
D-D
AB
AD
Coefficient
Estimate
85.00
-6.00
-0.50
-0.50
3.00
-2.50
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
1.46
1.46
1.46
1.46
1.46
1.46
0.9241
0.7344
-0.2143
6.441
95% CI
Low
78.73
-12.27
-6.77
-6.77
-3.27
-8.77
Final Equation in Terms of Coded Factors:
yield
+85.00
-6.00
-0.50
-0.50
+3.00
-2.50
=
*A
*B
*D
*A*B
*A*D
8-2
95% CI
High
91.27
0.27
5.77
5.77
9.27
3.77
VIF
1.00
1.00
1.00
1.00
1.00
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Final Equation in Terms of Actual Factors:
yield
+85.00000
-6.00000
-0.50000
-0.50000
+3.00000
-2.50000
=
*A
*B
*D
*A*B
*A*D
The Design-Expert output indicates that we really only need the main effect of factor A. The updated
analysis is shown below:
Design Expert Output
Response:
yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
288.00
1
288.00
A
288.00
1
288.00
Residual
160.00
6
26.67
Cor Total
448.00
7
F
Value
10.80
10.80
Prob > F
0.0167
0.0167
significant
The Model F-value of 10.80 implies the model is significant. There is only
a 1.67% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
5.16
85.00
6.08
284.44
Factor
Intercept
A-A
Coefficient
Estimate
85.00
-6.00
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Standard
Error
1.83
1.83
DF
1
1
0.6429
0.5833
0.3651
4.648
95% CI
Low
80.53
-10.47
95% CI
High
89.47
-1.53
VIF
1.00
Final Equation in Terms of Coded Factors:
yield
+85.00
-6.00
=
*A
Final Equation in Terms of Actual Factors:
yield
+85.00000
-6.00000
=
*A
8-2 Suppose that in Problem 6-15, only a one-half fraction of the 24 design could be run. Construct the
design and perform the analysis, using the data from replicate I.
The required design is a 24-1 with I=ABCD.
A
+
+
+
-
B
+
+
+
C
+
+
+
D=ABC
+
+
+
-
8-3
(1)
ad
bd
ab
cd
ac
bc
1.71
1.86
1.79
1.67
1.81
1.25
1.46
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
-0.285
Error
B
-0.215
Model
C
-0.415
Error
D
0.055
Error
AB
-0.08
Model
AC
-0.3
Error
AD
-0.16
Error
BC
Aliased
Error
BD
Aliased
Error
CD
Aliased
Error
ABC
Aliased
Error
ABD
Aliased
Error
ACD
Aliased
Error
BCD
Aliased
Error
ABCD
Aliased
Lenth's ME
1.21397
Lenth's SME 2.90528
+
+
abcd
SumSqr
0.85
% Contribtn
0.16245
0.09245
0.34445
0.00605
0.0128
0.18
0.0512
19.1253
10.8842
40.5522
0.712267
1.50695
21.1914
6.02778
C, A and AC + BD are the largest three effects. Now because the main effects of A and C are large, the
large effect estimate for the AC + BD alias chain probably indicates that the AC interaction is important.
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
C
90
85
80
AC
70
A
60
40
20
0
0 .0 0
0 .1 0
0 .2 1
0 .3 1
0 .4 2
Effe ct
Design Expert Output
Response:
Crack Lengthin mm x 10^-2
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.69
3
0.23
A
0.16
1
0.16
C
0.34
1
0.34
AC
0.18
1
0.18
Residual
0.16
4
0.041
Cor Total
0.85
7
F
Value
5.64
4.00
8.48
4.43
Prob > F
0.0641
0.1162
0.0436
0.1031
The Model F-value of 5.64 implies there is a 6.41% chance that a "Model F-Value"
this large could occur due to noise.
Std. Dev.
0.20
R-Squared
0.8087
8-4
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Mean
C.V.
PRESS
1.55
13.00
0.65
Factor
Intercept
A-Pour Temp
C-Heat Tr Mtd
AC
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
1.55
-0.14
-0.21
-0.15
DF
1
1
1
1
0.6652
0.2348
5.017
Standard
Error
0.071
0.071
0.071
0.071
95% CI
Low
1.35
-0.34
-0.41
-0.35
95% CI
High
1.75
0.055
-9.648E-003
0.048
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Crack Length
+1.55
-0.14
-0.21
-0.15
=
*A
*C
*A*C
Final Equation in Terms of Actual Factors:
Crack Length
+1.55000
-0.14250
-0.20750
-0.15000
=
* Pour Temp
* Heat Treat Method
* Pour Temp * Heat Treat Method
8-3 Consider the plasma etch experiment described in Problem 6-18. Suppose that only a one-half
fraction of the design could be run. Set up the design and analyze the data.
A
B
C
D=ABC
Etch
Rate
(A/min)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
550
650
642
601
749
1052
1075
729
Design Expert Output
Term
Effect
Model
Intercept
Error
A
4
Error
B
11.5
Model
C
290.5
Model
D
-127
Error
AB
-197.5
Error
AC
-25.5
Error
AD
-10
Error
BC
Aliased
Error
BD
Aliased
Model
CD
Aliased
Error
ABC
Aliased
Error
ABD
Aliased
Error
ACD
Aliased
Error
BCD
Aliased
Error
ABCD
Aliased
Lenth's ME
Lenth's SME
A (cm)
B (mTorr)
C (SCCM)
D (W)
SumSqr
% Contribtn
32
264.5
168780
32258
78012.5
1300.5
200
0.0113941
0.0941791
60.0967
11.4859
27.7775
0.463062
0.0712129
60.6987
145.264
8-5
Factor
Low (-)
Levels
High (+)
0.80
4.50
125
275
1.20
550
200
325
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
C
90
85
AB
80
D
70
60
40
20
0
0 .0 0
7 2 .6 3
1 4 5 .25
2 1 7 .88
2 9 0 .50
Effe ct
The large AB + CD alias chain is most likely the CD interaction.
Design Expert Output
Response:
Etch Rate in A/min
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
2.791E+005 3
93017.00
207.05
C
1.688E+005 1
1.688E+005 375.69
D
32258.00
1
32258.00
71.80
CD
78012.50
1
78012.50
173.65
Residual
1797.00
4
449.25
Cor Total
2.808E+005 7
Prob > F
< 0.0001
< 0.0001
0.0011
0.0002
significant
The Model F-value of 207.05 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
21.20
756.00
2.80
7188.00
Factor
Intercept
C-Gas Flow
D-Power
CD
Coefficient
Estimate
756.00
145.25
-63.50
-98.75
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
Standard
Error
7.49
7.49
7.49
7.49
0.9936
0.9888
0.9744
32.560
95% CI
Low
735.19
124.44
-84.31
-119.56
Final Equation in Terms of Coded Factors:
Etch Rate
+756.00
+145.25
-63.50
-98.75
=
*C
*D
*C*D
Final Equation in Terms of Actual Factors:
Etch Rate
-4246.41667
+35.47333
+14.57667
-0.10533
=
* Gas Flow
* Power
* Gas Flow * Power
8-6
95% CI
High
776.81
166.06
-42.69
-77.94
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
8-4 Problem 6-21describes a process improvement study in the manufacturing process of an integrated
circuit. Suppose that only eight runs could be made in this process. Set up an appropriate 25-2 design and
find the alias structure. Use the appropriate observations from Problem 6-21 as the observations in this
design and estimate the factor effects. What conclusions can you draw?
I = ABD = ACE = BCDE
A
B
C
D
E
BC
BE
(ABD)
(ABD)
(ABD)
(ABD)
(ABD)
(ABD)
(ABD)
=BD
=AD
=ABCD
=AB
=ABDE
=ACD
=ADE
A
B
C
D
E
BC
BE
A
+
+
+
+
B
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
11.25
Model
B
33.25
Model
C
10.75
Model
D
7.75
Error
E
2.25
Error
BC
-1.75
Error
BE
1.75
Lenth's ME
28.232
Lenth's SME 67.5646
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
=CE
=ABCE
=AE
=ACDE
=AC
=ABE
=ABC
C
+
+
+
+
D=AB
+
+
+
+
A
B
C
D
E
BC
BE
E=AC
+
+
+
+
SumSqr
% Contribtn
253.125
2211.13
231.125
120.125
10.125
6.125
6.125
8.91953
77.9148
8.1443
4.23292
0.356781
0.215831
0.215831
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
de
a
be
abd
cd
ace
bc
abcde
6
9
35
50
18
22
40
63
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
B
90
85
80
A
C
70
60
D
40
20
0
0 .0 0
8 .3 1
1 6 .6 3
Effe ct
8-7
2 4 .9 4
=ABCDE
=CDE
=BDE
=BCE
=BCD
=DE
=CD
3 3 .2 5
A=BD=CE=ABCDE
B=AD=ABCE=CDE
C=ABCD=AE=BDE
D=AB=ACDE=BCE
E=ABDE=AC=BCD
BC=ACD=ABE=DE
BE=ADE=ABC=CD
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The main A, B, C, and D are large. However, recall that you are really estimating A+BD+CE, B+AD,
C+DE and D+AD. There are other possible interpretations of the experiment because of the aliasing.
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2815.50
4
703.88
A
253.13
1
253.13
B
2211.12
1
2211.12
C
231.13
1
231.13
D
120.13
1
120.13
Residual
22.38
3
7.46
Cor Total
2837.88
7
F
Value
94.37
33.94
296.46
30.99
16.11
Prob > F
0.0017
0.0101
0.0004
0.0114
0.0278
significant
The Model F-value of 94.37 implies the model is significant. There is only
a 0.17% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.73
30.38
8.99
159.11
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
30.38
A-Aperture
5.63
B-Exposure Time 16.63
C-Develop Time
5.37
D-Mask Dimension 3.87
DF
1
1
1
1
1
Standard
Error
0.97
0.97
0.97
0.97
0.97
0.9921
0.9816
0.9439
25.590
95% CI
Low
27.30
2.55
13.55
2.30
0.80
Final Equation in Terms of Coded Factors:
Yield
+30.38
+5.63
+16.63
+5.37
+3.87
=
*A
*B
*C
*D
Final Equation in Terms of Actual Factors:
Aperture
Mask Dimension
Yield
-6.00000
+0.83125
+0.71667
small
Small
=
Aperture
Mask Dimension
Yield
+5.25000
+0.83125
+0.71667
large
Small
=
Aperture
Mask Dimension
Yield
+1.75000
+0.83125
+0.71667
small
Large
=
Aperture
Mask Dimension
Yield
+13.00000
large
Large
=
* Exposure Time
* Develop Time
* Exposure Time
* Develop Time
* Exposure Time
* Develop Time
8-8
95% CI
High
33.45
8.70
19.70
8.45
6.95
VIF
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+0.83125
+0.71667
* Exposure Time
* Develop Time
8-5 Continuation of Problem 8-4. Suppose you have made the eight runs in the 25-2 design in Problem
8-4. What additional runs would be required to identify the factor effects that are of interest? What are
the alias relationships in the combined design?
We could fold over the original design by changing the signs on the generators D = AB and E = AC to
produce the following new experiment.
A
B
C
D
E
BC
BE
(-ABD)
(-ABD)
(-ABD)
(-ABD)
(-ABD)
(-ABD)
(-ABD)
=-BD
=-AD
=-ABCD
=-AB
=-ABDE
=-ACD
=-ADE
A
+
+
+
+
B
+
+
+
+
C
+
+
+
+
A
B
C
D
E
BC
BE
(-ACE)
(-ACE)
(-ACE)
(-ACE)
(-ACE)
(-ACE)
(-ACE)
D=-AB
+
+
+
+
=-CE
=-ABCE
=-AE
=-ACDE
=-AC
=-ABE
=-ABC
E=-AC
+
+
+
+
A
B
C
D
E
BC
BE
(1)
ade
bd
abe
ce
acd
bcde
abc
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
(BCDE)
7
12
32
52
15
21
41
60
=ABCDE
=CDE
=BDE
=BCE
=BCD
=DE
=CD
A=-BD=-CE=ABCDE
B=-AD=-ABCE=CDE
C=-ABCD=-AE=BDE
D=-AB=-ACDE=BCE
E=-ABDE=-AC=BCD
BC=-ACD=-ABE=DE
BE=-ADE=-ABC=CD
Assuming all three factor and higher interactions to be negligible, all main effects can be separated from
their two-factor interaction aliases in the combined design.
8-6 R.D. Snee (“Experimenting with a Large Number of Variables,” in Experiments in Industry:
Design, Analysis and Interpretation of Results, by R.D. Snee, L.B. Hare, and J.B. Trout, Editors, ASQC,
1985) describes an experiment in which a 25-1 design with I=ABCDE was used to investigate the effects of
five factors on the color of a chemical product. The factors are A = solvent/reactant, B = catalyst/reactant,
C = temperature, D = reactant purity, and E = reactant pH. The results obtained were as follows:
e=
a=
b=
abe =
c=
ace =
bce =
abc =
-0.63
2.51
-2.68
1.66
2.06
1.22
-2.09
1.93
d=
ade =
bde =
abd =
cde =
acd =
bcd =
abcde =
6.79
5.47
3.45
5.68
5.22
4.38
4.30
4.05
(a) Prepare a normal probability plot of the effects. Which effects seem active?
Factors A, B, D, and the AB, AD interactions appear to be active.
8-9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lo t
DE S IG N-E X P E RT P l o t
Co l o r
S o l ve n t/Re a cta n t
C a ta l yst/R e a cta n t
T e m p e ra tu re
R e a cta n t P u ri ty
R e a cta n t p H
99
D
95
N orm a l % proba bility
A:
B:
C:
D:
E:
90
A
AB
80
70
50
30
20
10
B
5
AD
1
-1 .3 6
0 .0 9
1 .5 3
2 .9 8
4 .4 2
Effect
Design Expert Output
Term
Effect
Model
Intercept
Model
A
1.31
Model
B
-1.34
Error
C
-0.1475
Model
D
4.42
Error
E
-0.8275
Model
AB
1.275
Error
AC
-0.7875
Model
AD
-1.355
Error
AE
0.3025
Error
BC
0.1675
Error
BD
0.245
Error
BE
0.2875
Error
CD
-0.7125
Error
CE
-0.24
Error
DE
0.0875
Lenth's ME 1.95686
Lenth's SME 3.9727
SumSqr
% Contribtn
6.8644
7.1824
0.087025
78.1456
2.73902
6.5025
2.48062
7.3441
0.366025
0.112225
0.2401
0.330625
2.03063
0.2304
0.030625
Design Expert Output
Response:
Color
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
106.04
5
21.21
A
6.86
1
6.86
B
7.18
1
7.18
D
78.15
1
78.15
AB
6.50
1
6.50
AD
7.34
1
7.34
Residual
8.65
10
0.86
Cor Total
114.69
15
5.98537
6.26265
0.0758809
68.1386
2.38828
5.66981
2.16297
6.40364
0.319153
0.0978539
0.209354
0.288286
1.77059
0.200896
0.0267033
F
Value
24.53
7.94
8.31
90.37
7.52
8.49
Prob > F
< 0.0001
0.0182
0.0163
< 0.0001
0.0208
0.0155
The Model F-value of 24.53 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.93
2.71
34.35
22.14
Coefficient
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Standard
0.9246
0.8869
0.8070
14.734
95% CI
8-10
95% CI
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Estimate
Intercept
2.71
A-Solvent/Reactant 0.66
B-Catalyst/Reactant -0.67
D-Reactant Purity 2.21
AB
0.64
AD
-0.68
DF
1
1
1
1
1
1
Error
0.23
0.23
0.23
0.23
0.23
0.23
Low
2.19
0.14
-1.19
1.69
0.12
-1.20
High
3.23
1.17
-0.15
2.73
1.16
-0.16
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Color
+2.71
+0.66
-0.67
+2.21
+0.64
-0.68
=
*A
*B
*D
*A*B
*A*D
Final Equation in Terms of Actual Factors:
Color
+2.70750
+0.65500
-0.67000
+2.21000
+0.63750
-0.67750
=
* Solvent/Reactant
* Catalyst/Reactant
* Reactant Purity
* Solvent/Reactant * Catalyst/Reactant
* Solvent/Reactant * Reactant Purity
(b) Calculate the residuals. Construct a normal probability plot of the residuals and plot the residuals
versus the fitted values. Comment on the plots.
Design Expert Output
Diagnostics Case Statistics
Standard Actual
Predicted
Order
Value
Value
1
-0.63
0.47
2
2.51
1.86
3
-2.68
-2.14
4
1.66
1.80
5
2.06
0.47
6
1.22
1.86
7
-2.09
-2.14
8
1.93
1.80
9
6.79
6.25
10
5.47
4.93
11
3.45
3.63
12
5.68
4.86
13
5.22
6.25
14
4.38
4.93
15
4.30
3.63
16
4.05
4.86
Residual
-1.10
0.65
-0.54
-0.14
1.59
-0.64
0.053
0.13
0.54
0.54
-0.18
0.82
-1.03
-0.55
0.67
-0.81
Leverage
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
0.375
8-11
Student
Residual
-1.500
0.881
-0.731
-0.187
2.159
-0.874
0.071
0.180
0.738
0.738
-0.248
1.112
-1.398
-0.745
0.908
-1.105
Cook's
Distance
0.225
0.078
0.053
0.003
0.466
0.076
0.001
0.003
0.054
0.054
0.006
0.124
0.195
0.055
0.082
0.122
Outlier
t
-1.616
0.870
-0.713
-0.178
2.804
-0.863
0.068
0.171
0.720
0.720
-0.236
1.127
-1.478
-0.727
0.899
-1.119
Run
Order
2
6
14
11
8
15
10
3
4
5
16
12
9
1
13
7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .5 8 75
99
0 .9 1 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
0 .2 4 25
20
10
-0 .4 3
5
1
-1 .1 02 5
-1 .1 02 5
-0 .4 3
0 .2 4 25
0 .9 1 5
1 .5 8 75
-2 .1 4
R es idua l
-0 .0 5
2 .0 5
4 .1 5
6 .2 5
Predicted
The residual plots are satisfactory.
(c) If any factors are negligible, collapse the 25-1 design into a full factorial in the active factors.
Comment on the resulting design, and interpret the results.
The design becomes two replicates of a 23 in the factors A, B and D. When re-analyzing the data in three
factors, D becomes labeled as C.
Design Expert Output
Response:
Color
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
106.51
7
15.22
A
6.86
1
6.86
B
7.18
1
7.18
C
78.15
1
78.15
AB
6.50
1
6.50
AC
7.34
1
7.34
BC
0.24
1
0.24
ABC
0.23
1
0.23
Residual
8.18
8
1.02
Lack of Fit
0.000
0
Pure Error
8.18
8
1.02
Cor Total
114.69
15
F
Value
14.89
6.72
7.03
76.46
6.36
7.19
0.23
0.23
Prob > F
0.0005
0.0320
0.0292
< 0.0001
0.0357
0.0279
0.6409
0.6476
significant
The Model F-value of 14.89 implies the model is significant. There is only
a 0.05% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.01
2.71
37.34
32.71
Coefficient
Factor
Estimate
Intercept
2.71
A-Solvent/Reactant 0.66
B-Catalyst/Reactant -0.67
C-Reactant Purity 2.21
AB
0.64
AC
-0.68
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
0.25
0.25
0.25
0.25
0.25
0.25
0.9287
0.8663
0.7148
11.736
95% CI
Low
2.12
0.072
-1.25
1.63
0.055
-1.26
8-12
95% CI
High
3.29
1.24
-0.087
2.79
1.22
-0.095
VIF
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
BC
ABC
0.12
-0.12
1
1
0.25
0.25
-0.46
-0.70
0.71
0.46
1.00
1.00
Final Equation in Terms of Coded Factors:
Color
+2.71
+0.66
-0.67
+2.21
+0.64
-0.68
+0.12
-0.12
=
*A
*B
*C
*A*B
*A*C
*B*C
*A*B*C
Final Equation in Terms of Actual Factors:
Color
+2.70750
+0.65500
-0.67000
+2.21000
+0.63750
-0.67750
+0.12250
-0.12000
=
* Solvent/Reactant
* Catalyst/Reactant
* Reactant Purity
* Solvent/Reactant * Catalyst/Reactant
* Solvent/Reactant * Reactant Purity
* Catalyst/Reactant * Reactant Purity
* Solvent/Reactant * Catalyst/Reactant * Reactant Purity
8-7 An article by J.J. Pignatiello, Jr. And J.S. Ramberg in the Journal of Quality Technology, (Vol. 17,
1985, pp. 198-206) describes the use of a replicated fractional factorial to investigate the effects of five
factors on the free height of leaf springs used in an automotive application. The factors are A = furnace
temperature, B = heating time, C = transfer time, D = hold down time, and E = quench oil temperature.
The data are shown below:
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
7.78
8.15
7.50
7.59
7.54
7.69
7.56
7.56
7.50
7.88
7.50
7.63
7.32
7.56
7.18
7.81
Free Height
7.78
8.18
7.56
7.56
8.00
8.09
7.52
7.81
7.25
7.88
7.56
7.75
7.44
7.69
7.18
7.50
7.81
7.88
7.50
7.75
7.88
8.06
7.44
7.69
7.12
7.44
7.50
7.56
7.44
7.62
7.25
7.59
(a) Write out the alias structure for this design. What is the resolution of this design?
I=ABCD, Resolution IV
A
B
C
D
E
AB
AC
AD
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
8-13
BCD
ACD
ABD
ABC
ABCDE
CD
BD
BC
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
AE
BE
CE
DE
(ABCD)=
(ABCD)=
(ABCD)=
(ABCD)=
BCDE
ACDE
ABDE
ABCE
(b) Analyze the data. What factors influence the mean free height?
Design Expert Output
Term
Effect
Model
Intercept
Model
A
0.242083
Model
B
-0.16375
Model
C
-0.0495833
Model
D
0.09125
Model
E
-0.23875
Model
AB
-0.0295833
Model
AC
0.00125
Model
AD
-0.0229167
Model
AE
0.06375
Error
BC
Aliased
Error
BD
Aliased
Model
BE
0.152917
Error
CD
Aliased
Model
CE
-0.0329167
Model
DE
0.0395833
Error
Pure Error
Lenth's ME 0.088057
Lenth's SME 0.135984
SumSqr
% Contribtn
0.703252
0.321769
0.0295021
0.0999188
0.684019
0.0105021
1.875E-005
0.00630208
0.0487687
24.3274
11.1309
1.02056
3.45646
23.6621
0.363296
0.000648614
0.218006
1.68704
0.280602
9.70679
0.0130021
0.0188021
0.627067
0.449777
0.650415
21.6919
No rm a l p lot
A
99
BE
N orm al % probability
95
90
80
70
50
30
20
10
B
5
E
1
-0 .2 4
-0 .1 2
0 .0 0
0 .1 2
0 .2 4
Effe ct
Design Expert Output
Response:Free Height
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1.99
4
0.50
A
0.70
1
0.70
B
0.32
1
0.32
E
0.68
1
0.68
BE
0.28
1
0.28
Residual
0.90
43
0.021
Lack of Fit
0.27
11
0.025
F
Value
Prob > F
23.74
< 0.0001
33.56
< 0.0001
15.35
0.0003
32.64
< 0.0001
13.39
0.0007
1.27
8-14
significant
0.2844 not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Pure Error
Cor Total
0.63
2.89
32
47
0.020
The Model F-value of 23.74 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.14
7.63
1.90
1.12
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
7.63
A-Furnace Temp 0.12
B-Heating Time -0.082
E-Quench Temp -0.12
BE
0.076
DF
1
1
1
1
1
Standard
Error
0.021
0.021
0.021
0.021
0.021
0.6883
0.6593
0.6116
13.796
95% CI
Low
7.58
0.079
-0.12
-0.16
0.034
95% CI
High
7.67
0.16
-0.040
-0.077
0.12
VIF
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Free Height
+7.63
+0.12
-0.082
-0.12
+0.076
=
*A
*B
*E
*B*E
Final Equation in Terms of Actual Factors:
Free Height
+7.62562
+0.12104
-0.081875
-0.11937
+0.076458
=
* Furnace Temp
* Heating Time
* Quench Temp
* Heating Time * Quench Temp
(c) Calculate the range and standard deviation of the free height for each run. Is there any indication that
any of these factors affects variability in the free height?
Design Expert Output (Range)
Term
Model
Intercept
Model
A
Model
B
Model
C
Error
D
Model
E
Error
AB
Error
AC
Error
AD
Error
AE
Model
BC
Error
BD
Model
BE
Error
CD
Model
CE
Error
DE
Error
ABC
Error
ABD
Error
ABE
Error
ACD
Error
ACE
Error
ADE
Error
BCD
Model
BCE
Error
BDE
Error
CDE
Lenth's ME
Effect
0.11375
-0.12625
0.02625
0.06125
-0.01375
0.04375
-0.03375
0.03625
-0.00375
Aliased
Aliased
0.01625
Aliased
-0.13625
-0.02125
Aliased
Aliased
0.03125
Aliased
0.04875
0.13875
Aliased
Aliased
Aliased
Aliased
0.130136
SumSqr
% Contribtn
0.0517563
0.0637563
0.00275625
0.0150063
0.00075625
0.00765625
0.00455625
0.00525625
5.625E-005
16.2198
19.9804
0.863774
4.70277
0.236999
2.39937
1.42787
1.64724
0.017628
0.00105625
0.331016
0.0742562
0.00180625
23.271
0.566056
0.00390625
1.22417
0.00950625
0.0770062
2.97914
24.1328
8-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Lenth's SME
0.264194
Interaction ADE is aliased with BCE. Although the plot below identifies ADE, BCE was included in the
analysis.
Ha lf No rm al p lo t
DE S IG N-E X P E RT P l o t
Ra n g e
Fu rn ace T e m p
He a ti n g T i m e
T ra n sfe r T i m e
Ho l d T i m e
Q ue n ch T e m p
99
H alf N orm a l % prob ability
A:
B:
C:
D:
E:
97
AD E
95
90
CE
85
80
B
A
70
60
40
C
20
E
0
0 .0 0
0 .0 3
0 .0 7
0 .1 0
0 .1 4
|Effect|
Design Expert Output (Range)
Response:
Range
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.28
8
0.035
A
0.052
1
0.052
B
0.064
1
0.064
C
2.756E-003
1
2.756E-003
E
7.562E-004
1
7.562E-004
BC
5.256E-003
1
5.256E-003
BE
1.056E-003
1
1.056E-003
CE
0.074
1
0.074
BCE
0.077
1
0.077
Residual
0.042
7
6.071E-003
Cor Total
0.32
15
F
Value
Prob > F
5.70
0.0167 significant
8.53
0.0223
10.50
0.0142
0.45
0.5220
0.12
0.7345
0.87
0.3831
0.17
0.6891
12.23
0.0100
12.69
0.0092
The Model F-value of 5.70 implies the model is significant. There is only
a 1.67% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.078
0.22
35.52
0.22
Coefficient
Factor
Estimate
Intercept
0.22
A-Furn Temp
0.057
B-Heat Time
-0.063
C-Transfer Time 0.013
E-Qnch Temp -6.875E-003
BC
0.018
BE
8.125E-003
CE
-0.068
BCE
0.069
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
1
1
1
Standard
Error
0.019
0.019
0.019
0.019
0.019
0.019
0.019
0.019
0.019
0.8668
0.7146
0.3043
7.166
95% CI
Low
0.17
0.011
-0.11
-0.033
-0.053
-0.028
-0.038
-0.11
0.023
Final Equation in Terms of Coded Factors:
8-16
95% CI
High
0.27
0.10
-0.017
0.059
0.039
0.064
0.054
-0.022
0.12
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Range
+0.22
+0.057
-0.063
+0.013
-6.875E-003
+0.018
+8.125E-003
-0.068
+0.069
=
*A
*B
*C
*E
*B*C
*B*E
*C*E
*B*C*E
Final Equation in Terms of Actual Factors:
Range
+0.21937
+0.056875
-0.063125
+0.013125
-6.87500E-003
+0.018125
+8.12500E-003
-0.068125
+0.069375
=
* Furnace Temp
* Heating Time
* Transfer Time
* Quench Temp
* Heating Time * Transfer Time
* Heating Time * Quench Temp
* Transfer Time * Quench Temp
* Heating Time * Transfer Time * Quench Temp
Design Expert Output (StDev)
Term
Effect
SumSqr
Model
Intercept
Model
A
0.0625896
0.0156698
Model
B
-0.0714887
0.0204425
Model
C
0.010567
0.000446646
Error
D
0.0353616
0.00500176
Model
E
-0.00684034
0.000187161
Error
AB
0.0153974
0.000948317
Error
AC
-0.0218505
0.00190978
Error
AD
0.0190608
0.00145326
Error
AE
-0.00329035
4.33057E-005
Model
BC
Aliased
Error
BD
Aliased
Model
BE
0.0087666
0.000307413
Error
CD
Aliased
Model
CE
-0.0714816
0.0204385
Error
DE
-0.00467792
8.75317E-005
Error
ABC
Aliased
Error
ABD
Aliased
Error
ABE
0.0155599
0.000968437
Error
ACD
Aliased
Error
ACE
0.0199742
0.00159587
Error
ADE
Aliased
Error
BCD
Aliased
Model
BCE
0.0764346
0.023369
Error
BDE
Aliased
Error
CDE
Aliased
Lenth's ME
0.0596836
Lenth's SME 0.121166
% Contribtn
16.873
22.0121
0.48094
5.3858
0.201532
1.02113
2.05641
1.56484
0.0466308
0.331017
22.0078
0.0942525
1.0428
1.7184
25.1633
Interaction ADE is aliased with BCE. Although the plot below identifies ADE, BCE was included in the
analysis.
8-17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
DE S IG N-E X P E RT P l o t
S tDe v
Fu rn ace T e m p
He a ti n g T i m e
T ra n sfe r T i m e
Ho l d T i m e
Q ue n ch T e m p
99
H alf N orm a l % prob ability
A:
B:
C:
D:
E:
97
AD E
95
90
B
85
80
CE
A
70
60
40
C
BE
20
0
0 .0 0
0 .0 2
0 .0 4
0 .0 6
0 .0 8
|Effect|
Design Expert Output (StDev)
Response:
StDev
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.082
8
0.010
A
0.016
1
0.016
B
0.020
1
0.020
C
4.466E-004 1
4.466E-004
E
1.872E-004 1
1.872E-004
BC
1.453E-003 1
1.453E-003
BE
3.074E-004 1
3.074E-004
CE
0.020
1
0.020
BCE
0.023
1
0.023
Residual
0.011
7
1.508E-003
Cor Total
0.093
15
F
Value
6.82
10.39
13.56
0.30
0.12
0.96
0.20
13.55
15.50
Prob > F
0.0101
0.0146
0.0078
0.6032
0.7350
0.3589
0.6653
0.0078
0.0056
significant
The Model F-value of 6.82 implies the model is significant. There is only
a 1.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.039
0.12
33.07
0.055
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
DF
Intercept
0.12
1
A-Furnace Temp
0.031
1
B-Heating Time
-0.036
1
C-Transfer Time
5.283E-003 1
E-Quench Temp -3.420E-003 1
BC
9.530E-003 1
BE
4.383E-003 1
CE
-0.036
1
BCE
0.038
1
0.8863
0.7565
0.4062
7.826
Standard
95% CI
95% CI
Error
Low
High
9.708E-003
0.094
0.14
9.708E-003
8.340E-003 0.054
9.708E-003
-0.059
-0.013
9.708E-003
-0.018
0.028
9.708E-003
-0.026
0.020
9.708E-003
-0.013
0.032
9.708E-003
-0.019
0.027
9.708E-003
-0.059
-0.013
9.708E-003
0.015
0.061
Final Equation in Terms of Coded Factors:
StDev
+0.12
+0.031
-0.036
+5.283E-003
=
*A
*B
*C
8-18
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-3.420E-003
+9.530E-003
+4.383E-003
-0.036
+0.038
*E
*B*C
*B*E
*C*E
*B*C*E
Final Equation in Terms of Actual Factors:
StDev
+0.11744
+0.031295
-0.035744
+5.28350E-003
-3.42017E-003
+9.53040E-003
+4.38330E-003
-0.035741
+0.038217
=
* Furnace Temp
* Heating Time
* Transfer Time
* Quench Temp
* Heating Time * Transfer Time
* Heating Time * Quench Temp
* Transfer Time * Quench Temp
* Heating Time * Transfer Time * Quench Temp
(d) Analyze the residuals from this experiment, and comment on your findings.
The residual plot follows. All plots are satisfactory.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .2 4 72 9 2
2
99
0 .1 0 18 7 5
90
2
2
2
80
R es idua ls
N orm al % probability
95
70
50
2
2
-0 .0 43 5 4 1 7
30
20
2
10
-0 .1 88 9 5 8
5
2
1
-0 .3 34 3 7 5
-0 .3 34 3 7 5
-0 .1 88 9 5 8
-0 .0 43 5 4 1 7
0 .1 0 18 7 5
0 .2 4 72 9 2
7 .3 8
R es idua l
7 .5 4
7 .7 0
Predicted
8-19
7 .8 6
8 .0 2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. F urnace Te m p
0 .2 4 72 9 2
0 .2 4 72 9 2
2
2
2
2
2
0 .1 0 18 7 5
2
2
2
R es idua ls
R es idua ls
0 .1 0 18 7 5
Re sid ua ls vs. He a ting Tim e
2
2
-0 .0 43 5 4 1 7
2
2
-0 .0 43 5 4 1 7
2
-0 .1 88 9 5 8
-0 .3 34 3 7 5
0
1
-1
H eating Tim e
Re sid ua ls vs. Tra nsfe r Tim e
Re sid ua ls vs. Ho ld Tim e
2
0 .2 4 72 9 2
2
2
2
2
2
-0 .0 43 5 4 1 7
-0 .1 88 9 5 8
2
-0 .3 34 3 7 5
-0 .1 88 9 5 8
2
-0 .3 34 3 7 5
-1
0
1
-1
Tra ns fer Tim e
2
2
2
2
2
2
-0 .0 43 5 4 1 7
2
-0 .1 88 9 5 8
2
-0 .3 34 3 7 5
-1
0
0
H old Tim e
Re sid ua ls vs. Q ue nch Te m p
0 .2 4 72 9 2
R es idua ls
2
2
2
1
2
0 .1 0 18 7 5
-0 .0 43 5 4 1 7
0 .1 0 18 7 5
0
Furnace Tem p
R es idua ls
R es idua ls
0 .1 0 18 7 5
2
-0 .3 34 3 7 5
-1
0 .2 4 72 9 2
2
-0 .1 88 9 5 8
2
1
Quench Tem p
8-20
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(e) Is this the best possible design for five factors in 16 runs? Specifically, can you find a fractional
design for five factors in 16 runs with a higher resolution than this one?
This was not the best design. A resolution V design is possible by setting the generator equal to the
highest order interaction, ABCDE.
8-8 An article in Industrial and Engineering Chemistry (“More on Planning Experiments to Increase
Research Efficiency,” 1970, pp. 60-65) uses a 25-2 design to investigate the effect of A = condensation, B =
amount of material 1, C = solvent volume, D = condensation time, and E = amount of material 2 on yield.
The results obtained are as follows:
e=
ab =
23.2
15.5
ad =
bc =
16.9
16.2
cd =
ace =
23.8
23.4
bde =
abcde =
16.8
18.1
(a) Verify that the design generators used were I = ACE and I = BDE.
A
+
+
+
+
B
+
+
+
+
C
+
+
+
+
D=BE
+
+
+
+
E=AC
+
+
+
+
e
ad
bde
ab
cd
ace
bc
abcde
(b) Write down the complete defining relation and the aliases for this design.
I=BDE=ACE=ABCD.
A
B
C
D
E
AB
AD
(BDE)
(BDE)
(BDE)
(BDE)
(BDE)
(BDE)
(BDE)
=ABDE
=DE
=BCDE
=BE
=BD
=ADE
=ABE
A
B
C
D
E
AB
AD
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
(ACE)
=CE
=ABCE
=AE
=ACDE
=AC
=BCE
=CDE
A
B
C
D
E
AB
AD
(ABCD)
(ABCD)
(ABCD)
(ABCD)
(ABCD)
(ABCD)
(ABCD)
=BCD
=ACD
=ABD
=ABC
=ABCDE
=CD
=BC
A=ABDE=CE=BCD
B=DE=ABCE=ACD
C=BCDE=AE=ABD
D=BE=ACDE=ABC
E=BD=AC=ABCDE
AB=ADE=BCE=CD
AD=ABE=CDE=BC
(c) Estimate the main effects.
Design Expert Output
Term
Model
Intercept
Model
A
Model
B
Model
C
Model
D
Model
E
Effect
-1.525
-5.175
2.275
-0.675
2.275
SumSqr
% Contribtn
4.65125
53.5613
10.3512
0.91125
10.3513
5.1831
59.6858
11.5349
1.01545
11.5349
(d) Prepare an analysis of variance table. Verify that the AB and AD interactions are available to use as
error.
The analysis of variance table is shown below. Part (b) shows that AB and AD are aliased with other
factors. If all two-factor and three factor interactions are negligible, then AB and AD could be pooled as
an estimate of error.
8-21
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
79.83
5
15.97
A
4.65
1
4.65
B
53.56
1
53.56
C
10.35
1
10.35
D
0.91
1
0.91
E
10.35
1
10.35
Residual
9.91
2
4.96
Cor Total
89.74
7
F
Value
3.22
0.94
10.81
2.09
0.18
2.09
Prob > F
0.2537
0.4349
0.0814
0.2853
0.7098
0.2853
not significant
The "Model F-value" of 3.22 implies the model is not significant relative to the noise. There is a
25.37 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.23
19.24
11.57
158.60
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
19.24
A-Condensation
-0.76
B-Material 1
-2.59
C-Solvent
1.14
D-Time
-0.34
E-Material 2
1.14
DF
1
1
1
1
1
1
Standard
Error
0.79
0.79
0.79
0.79
0.79
0.79
0.8895
0.6134
-0.7674
5.044
95% CI
Low
15.85
-4.15
-5.97
-2.25
-3.72
-2.25
95% CI
High
22.62
2.62
0.80
4.52
3.05
4.52
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Yield
+19.24
-0.76
-2.59
+1.14
-0.34
+1.14
=
*A
*B
*C
*D
*E
Final Equation in Terms of Actual Factors:
Yield
+19.23750
-0.76250
-2.58750
+1.13750
-0.33750
+1.13750
=
* Condensation
* Material 1
* Solvent
* Time
* Material 2
(e) Plot the residuals versus the fitted values. Also construct a normal probability plot of the residuals.
Comment on the results.
The residual plots are satisfactory.
8-22
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. P re d icted
No rm a l p lot o f re sid uals
1 .5 5
99
95
R es idua ls
N orm al % probability
0 .7 7 5
1 .7 76 3 6 E -0 1 5
-0 .7 75
90
80
70
50
30
20
10
5
1
-1 .5 5
1 3 .9 5
1 6 .3 8
1 8 .8 1
2 1 .2 4
2 3 .6 8
-1 .5 5
-0 .7 75
Predicted
-1 .7 76 3 6 E -0 1 5
0 .7 7 5
1 .5 5
R es idua l
8-9 Consider the leaf spring experiment in Problem 8-7. Suppose that factor E (quench oil
temperature) is very difficult to control during manufacturing. Where would you set factors A, B, C and D
to reduce variability in the free height as much as possible regardless of the quench oil temperature used?
Inte ra ctio n Graph
DE S IG N-E X P E RT P l o t
Fre e He i g h t
H ea ting Tim e
8 .1 8
X = E : Q u e n ch T e m p
Y = B : He a ti n g T i m e
Free H e igh t
7 .9 1 5
B - -1 .0 0 0
B + 1 .0 0 0
A ctu a l Fa cto rs
A : Fu rn a ce T e m p = 0 .0 0
C: T ra n sfe r T i m e = 0 .0 0 7 .6 5
D: H o l d T i m e = 0 .0 0
7 .3 8 5
7 .1 2
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
Que nch Tem p
Run the process with A at the high level, B at the low level, C at the low level and D at either level (the
low level of D may give a faster process).
8-10 Construct a 27-2 design by choosing two four-factor interactions as the independent generators.
Write down the complete alias structure for this design. Outline the analysis of variance table. What is
the resolution of this design?
I=CDEF=ABCG=ABDEFG, Resolution IV
A
B
C
D
E
8-23
F=CDE
G=ABC
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
(1)
ag
bg
ab
cfg
acf
bcf
abcfg
df
adfg
bdfg
abdf
cdg
acd
bcd
abcdg
ef
aefg
befg
abef
ceg
ace
bce
abceg
de
adeg
bdeg
abde
cdefg
acdef
bcdef
abcdefg
Alias Structure
A (CDEF)=
B (CDEF)=
C (CDEF)=
D (CDEF)=
E (CDEF)=
F (CDEF)=
G (CDEF)=
AB (CDEF)=
AC (CDEF)=
AD (CDEF)=
AE (CDEF)=
AF (CDEF)=
AG (CDEF)=
BD (CDEF)=
BE (CDEF)=
BF (CDEF)=
CD (CDEF)=
CE (CDEF)=
CF (CDEF)=
DG (CDEF)=
EG (CDEF)=
FG (CDEF)=
ACDEF
BCDEF
DEF
CEF
CDF
CDE
CDEFG
ABCDEF
ADEF
ACEF
ACDF
ACDE
ACDEFG
BCEF
BCDF
BCDE
EF
DF
DE
CEFG
CDFG
CDEG
A(ABCG)=
B(ABCG)=
C(ABCG)=
D(ABCG)=
E(ABCG)=
F(ABCG)=
G(ABCG)=
AB(ABCG)=
AC(ABCG)=
AD(ABCG)=
AE(ABCG)=
AF(ABCG)=
AG(ABCG)=
BD(ABCG)=
BE(ABCG)=
BF(ABCG)=
CD(ABCG)=
CE(ABCG)=
CF(ABCG)=
DG(ABCG)=
EG(ABCG)=
FG(ABCG)=
BCG
ACG
ABG
ABCDG
ABCEG
ABCFG
ABC
CG
BG
BCDG
BCEG
BCFG
BC
ACDG
ACEG
ACFG
ABDG
ABEG
ABFG
ABCD
ABCE
ABCF
A (ABDEFG)=
B (ABDEFG)=
C (ABDEFG)=
D (ABDEFG)=
E (ABDEFG)=
F (ABDEFG)=
G (ABDEFG)=
AB (ABDEFG)=
AC (ABDEFG)=
AD (ABDEFG)=
AE (ABDEFG)=
AF (ABDEFG)=
AG (ABDEFG)=
BD (ABDEFG)=
BE (ABDEFG)=
BF (ABDEFG)=
CD (ABDEFG)=
CE (ABDEFG)=
CF (ABDEFG)=
DG (ABDEFG)=
EG (ABDEFG)=
FG (ABDEFG)=
Analysis of Variance Table
Source
A
B
C
D
E
F
G
AB=CG
Degrees of Freedom
1
1
1
1
1
1
1
1
8-24
BDEFG
ADEFG
ABCDEFG
ABEFG
ABDFG
ABDEG
ABDEF
DEFG
BCDEFG
BEFG
BDFG
BDEG
BDEF
AEFG
ADFG
ADEG
ABCEFG
ABCDFG
ABCDEG
ABEF
ABDF
ABDE
A=ACDEF=BCG=BDEFG
B=BCDEF=ACG=ADEFG
C=DEF=ABG=ABCDEFG
D=CEF=ABCDG=ABEFG
E=CDF=ABCEG=ABDFG
F=CDE=ABCFG=ABDEG
G=CDEFG=ABC=ABDEF
AB=ABCDEF=CG=DEFG
AC=ADEF=BG=BCDEFG
AD=ACEF=BCDG=BEFG
AE=ACDF=BCEG=BDFG
AF=ACDE=BCFG=BDEG
AG=ACDEFG=BC=BDEF
BD=BCEF=ACDG=AEFG
BE=BCDF=ACEG=ADFG
BF=BCDE=ACFG=ADEG
CD=EF=ABDG=ABCEFG
CE=DF=ABEG=ABCDFG
CF=DE=ABFG=ABCDEG
DG=CEFG=ABCD=ABEF
EG=CDFG=ABCE=ABDF
FG=CDEG=ABCF=ABDE
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
AC=BG
AD
AE
AF
AG=BC
BD
BE
CD=EF
CE=DF
CF=DE
DG
EG
FG
Error
Total
1
1
1
1
1
1
1
1
1
1
1
1
1
9
31
8-11 Consider the 25 design in Problem 6-21. Suppose that only a one-half fraction could be run.
Furthermore, two days were required to take the 16 observations, and it was necessary to confound the 25-1
design in two blocks. Construct the design and analyze the data.
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
10.875
Model
B
33.625
Model
C
10.625
Error
D
-0.625
Error
E
0.375
Error
AB
Aliased
Error
AC
0.625
Error
AD
0.875
Error
AE
1.375
Error
BC
0.875
Error
BD
-0.375
Error
BE
0.125
Error
CD
0.625
Error
CE
0.625
Error
DE
-1.625
Lenth's ME 2.46263
Lenth's SME 5.0517
E=ABCD
+
+
+
+
+
+
+
+
Data
8
9
34
52
16
22
45
60
8
10
30
50
15
21
44
63
e
a
b
abe
c
ace
bce
abc
d
ade
bde
abd
cde
acd
bcd
abcde
SumSqr
% Contribtn
473.063
4522.56
451.562
1.5625
0.5625
8.6343
82.5455
8.24188
0.0285186
0.0102667
1.5625
3.0625
7.5625
3.0625
0.5625
0.0625
1.5625
1.5625
10.5625
Blocks = AB
+
+
+
+
+
+
+
+
Block
1
2
2
1
1
2
2
1
1
2
2
1
1
2
2
1
0.0285186
0.0558965
0.13803
0.0558965
0.0102667
0.00114075
0.0285186
0.0285186
0.192786
The AB interaction in the above table is aliased with the three-factor interaction BCD, and is also
confounded with blocks.
8-25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
B
97
95
A
90
C
85
80
70
60
40
20
0
0 .0 0
8 .4 1
1 6 .8 1
2 5 .2 2
3 3 .6 3
Effe ct
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Block
203.06
1
203.06
Model
5447.19
3
1815.73
A
473.06
1
473.06
B
4522.56
1
4522.56
C
451.56
1
451.56
Residual
31.69
11
2.88
Cor Total
5681.94
15
F
Value
Prob > F
630.31
164.22
1569.96
156.76
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
The Model F-value of 630.31 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.70
30.44
5.58
67.04
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
30.44
Block 1
3.56
Block 2
-3.56
A-Aperture
5.44
B-Exposure Time 16.81
C-Develop Time
5.31
DF
1
1
Standard
Error
0.42
1
1
1
0.42
0.42
0.42
0.9942
0.9926
0.9878
58.100
95% CI
Low
29.50
4.50
15.88
4.38
Final Equation in Terms of Coded Factors:
Yield
+30.44
+5.44
+16.81
+5.31
=
*A
*B
*C
Final Equation in Terms of Actual Factors:
Aperture
Yield
-1.56250
+0.84063
small
=
* Exposure Time
8-26
95% CI
High
31.37
VIF
6.37
17.75
6.25
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+0.70833
Aperture
Yield
+9.31250
+0.84063
+0.70833
* Develop Time
large
=
* Exposure Time
* Develop Time
8-12 Analyze the data in Problem 6-23 as if it came from a 2 4IV1 design with I = ABCD. Project the
design into a full factorial in the subset of the original four factors that appear to be significant.
Run
Number
1
2
3
4
5
6
7
8
A
+
+
+
+
B
+
+
+
+
C
+
+
+
+
D=ABC
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
3.75
Error
B
0.25
Model
C
2.75
Model
D
4.25
Error
AB
-0.75
Model
AC
-4.25
Model
AD
4.25
Lenth's ME
21.174
Lenth's SME 50.6734
Yield
(lbs)
12
25
13
16
19
15
20
23
(1)
ad
bd
ab
cd
ac
bc
abcd
A (h)
B (%)
C (psi)
D (ºC)
SumSqr
% Contribtn
28.125
0.125
15.125
36.125
1.125
36.125
36.125
18.3974
0.0817661
9.8937
23.6304
0.735895
23.6304
23.6304
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
AD
90
85
80
D
70
AC
60
A
C
40
20
0
0 .0 0
1 .0 6
2 .1 3
Effe ct
Design Expert Output
Response:
Yield
in lbs
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
8-27
3 .1 9
4 .2 5
Factor
Low (-)
2.5
14
60
225
Levels
High (+)
3.0
18
80
250
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Sum of
Squares
151.63
28.13
15.13
36.12
36.12
36.13
1.25
152.88
Source
Model
A
C
D
AC
AD
Residual
Cor Total
Mean
Square
30.32
28.13
15.13
36.12
36.12
36.13
0.62
DF
5
1
1
1
1
1
2
7
F
Value
48.52
45.00
24.20
57.80
57.80
57.80
Prob > F
0.0203
0.0215
0.0389
0.0169
0.0169
0.0169
significant
The Model F-value of 48.52 implies the model is significant. There is only
a 2.03% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.79
17.88
4.42
20.00
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
17.88
A-Time
1.87
C-Pressure
1.37
D-Temperature
2.13
AC
-2.13
AD
2.13
DF
1
1
1
1
1
1
0.9918
0.9714
0.8692
17.892
Standard
Error
0.28
0.28
0.28
0.28
0.28
0.28
95% CI
Low
16.67
0.67
0.17
0.92
-3.33
0.92
95% CI
High
19.08
3.08
2.58
3.33
-0.92
3.33
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Yield
+17.88
+1.87
+1.37
+2.13
-2.13
+2.13
=
*A
*C
*D
*A*C
*A*D
Final Equation in Terms of Actual Factors:
Yield
+227.75000
-94.50000
+2.47500
-1.70000
-0.85000
+0.68000
=
* Time
* Pressure
* Temperature
* Time * Pressure
* Time * Temperature
8-13 Repeat Problem 8-12 using I = -ABCD.
interpretation of the data?
Run
Number
1
2
3
4
5
6
7
8
A
+
+
+
+
Design Expert Output
Term
Model
Intercept
B
+
+
+
+
C
+
+
+
+
Effect
D=ABC
+
+
+
+
-
Does use of the alternate fraction change your
Yield
(lbs)
10
18
13
24
17
21
17
15
d
a
b
abd
c
acd
bcd
abc
SumSqr
% Contribtn
8-28
A (h)
B (%)
C (psi)
D (ºC)
Factor
Low (-)
2.5
14
60
225
Levels
High (+)
3.0
18
80
250
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Model
Error
Model
Model
Error
Model
Model
A
5.25
B
0.75
C
1.25
D
2.25
AB
-0.75
AC
-4.25
AD
3.75
Lenth's ME 12.7044
Lenth's SME 30.404
55.125
1.125
3.125
10.125
1.125
36.125
28.125
40.8712
0.834106
2.31696
7.50695
0.834106
26.7841
20.8526
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
A
90
85
AC
80
AD
70
60
D
C
40
20
0
0 .0 0
1 .3 1
2 .6 3
3 .9 4
5 .2 5
Effe ct
Design Expert Output
Response:
Yield
in lbs
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
132.63
5
26.52
A
55.13
1
55.13
C
3.13
1
3.13
D
10.13
1
10.13
AC
36.13
1
36.13
AD
28.13
1
28.13
Residual
2.25
2
1.12
Cor Total
134.88
7
F
Value
23.58
49.00
2.78
9.00
32.11
25.00
Prob > F
0.0412
0.0198
0.2375
0.0955
0.0298
0.0377
significant
The Model F-value of 23.58 implies the model is significant. There is only
a 4.12% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.06
16.88
6.29
36.00
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
16.88
A-Time
2.63
C-Pressure
0.63
D-Temperature
1.13
AC
-2.13
AD
1.88
DF
1
1
1
1
1
1
Standard
Error
0.37
0.37
0.37
0.37
0.37
0.37
0.9833
0.9416
0.7331
14.425
95% CI
Low
15.26
1.01
-0.99
-0.49
-3.74
0.26
Final Equation in Terms of Coded Factors:
Yield
=
8-29
95% CI
High
18.49
4.24
2.24
2.74
-0.51
3.49
VIF
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+16.88
+2.63
+0.63
+1.13
-2.13
+1.88
*A
*C
*D
*A*C
*A*D
Final Equation in Terms of Actual Factors:
Yield
+190.50000
-72.50000
+2.40000
-1.56000
-0.85000
+0.60000
=
* Time
* Pressure
* Temperature
* Time * Pressure
* Time * Temperature
8-14 Project the 2 4IV1 design in Example 8-1 into two replicates of a 22 design in the factors A and B.
Analyze the data and draw conclusions.
Design Expert Output
Response:
Filtration Rate
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
728.50
3
242.83
A
722.00
1
722.00
B
4.50
1
4.50
AB
2.00
1
2.00
Residual
2343.00
4
585.75
Lack of Fit
0.000
0
Pure Error
2343.00
4
585.75
Cor Total
3071.50
7
F
Value
Prob > F
0.41
0.7523
1.23
0.3291
7.682E-003 0.9344
3.414E-003 0.9562
not significant
The "Model F-value" of 0.41 implies the model is not significant relative to the noise. There is a
75.23 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
24.20
70.75
34.21
9372.00
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
70.75
A-Temperature
9.50
B-Pressure
0.75
AB
-0.50
DF
1
1
1
1
Standard
Error
8.56
8.56
8.56
8.56
0.2372
-0.3349
-2.0513
1.198
95% CI
Low
46.99
-14.26
-23.01
-24.26
Final Equation in Terms of Coded Factors:
Filtration Rate
+70.75
+9.50
+0.75
-0.50
=
*A
*B
*A*B
Final Equation in Terms of Actual Factors:
Filtration Rate
+70.75000
+9.50000
+0.75000
-0.50000
=
* Temperature
* Pressure
* Temperature * Pressure
8-30
95% CI
High
94.51
33.26
24.51
23.26
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
8-15 Construct a 2 6III3 design. Determine the effects that may be estimated if a second fraction of this
design is run with all signs reversed.
A
+
+
+
+
B
+
+
+
+
C
+
+
+
+
D=AB
+
+
+
+
Principal Fraction
lA=A+BD+CE
lB=B+AD+CF
lC=C+AE+BF
lD=D+AB+EF
lE=E+AC+DF
lF=F+BC+DE
lBE=BE+CD+AF
E=AC
+
+
+
+
F=BC
+
+
+
+
def
af
be
abd
cd
ace
bcf
abcdef
Second Fraction
l*A=A-BD-CE
l*B=B-AD-CF
l*C=C-AE-BF
l*D=D-AB-EF
l*E=E-AC-DF
l*F=F-BC-DE
l*BE=BE+CD+AF
By combining the two fractions we can estimate the following:
( li +l*I)/2
A
B
C
D
E
F
BE+CD+AF
( li -l*I)/2
BD+CE
AD+CF
AE+BF
AB+EF
AC+DF
BC+DE
8-16 Consider the 2 6III3 design in Problem 8-15. Determine the effects that may be estimated if a second
fraction of this design is run with the signs for factor A reversed.
Principal Fraction
lA=A+BD+CE
lB=B+AD+CF
lC=C+AE+BF
lD=D+AB+EF
lE=E+AC+DF
lF=F+BC+DE
lBE=BE+CD+AF
Second Fraction
l*A=-A+BD+CE
l*B=B-AD+CF
l*C=C-AE+BF
l*D=D-AB+EF
l*E=E-AC+DF
l*F=F+BC+DE
l*BE=BE+CD-AF
By combining the two fractions we can estimate the following:
( li -l*I)/2
A
AD
( li +l*I)/2
BD+CE
B+CF
8-31
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
AE
AB
AC
C+BF
D+EF
E+DF
F+BC+DE
AF
8-17 Fold over the 2 7III 4 design in Table 8-19 to produce a eight-factor design. Verify that the resulting
design is a 2 8IV 4 design. Is this a minimal design?
Original
Design
Second
Set of
Runs w/
all Signs
Switched
H
+
+
+
+
+
+
+
+
-
A
+
+
+
+
+
+
+
+
-
B
+
+
+
+
+
+
+
+
-
C
+
+
+
+
+
+
+
+
-
D=AB
+
+
+
+
+
+
+
+
-
E=AC
+
+
+
+
+
+
+
+
-
F=BC
+
+
+
+
+
+
+
+
-
G=ABC
+
+
+
+
+
+
+
+
-
After folding the original design over, we add a new factor H, and we have a design with generators
D=ABH, E=ACH, F=BCH, and G=ABC. This is a 28IV 4 design. It is a minimal design, since it contains
2k=2(8)=16 runs.
8-18 Fold over a 2 5III 2 design to produce a six-factor design. Verify that the resulting design is a 2 6IV2
design. Compare this 2 6IV2 design to the in Table 8-10.
Original
Design
Second
Set of
Runs w/
all Signs
Switched
F
+
+
+
+
+
+
+
+
-
A
+
+
+
+
+
+
+
+
-
B
+
+
+
+
+
+
+
+
-
C
+
+
+
+
+
+
+
+
-
D=AB
+
+
+
+
+
+
+
+
-
E=BC
+
+
+
+
+
+
+
+
-
If we relabel the factors from left to right as A, B, C, D, E, F, then this design becomes 2 6IV2 with
generators I=ABDF and I=BCEF. It is not a minimal design, since 2k=2(6)=12 runs, and the design
contains 16 runs.
8-32
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
8-19 An industrial engineer is conducting an experiment using a Monte Carlo simulation model of an
inventory system. The independent variables in her model are the order quantity (A), the reorder point
(B), the setup cost (C), the backorder cost (D), and the carrying cost rate (E). The response variable is
average annual cost. To conserve computer time, she decides to investigate these factors using a 2 5III 2
design with I = ABD and I = BCE. The results she obtains are de = 95, ae = 134, b = 158, abd = 190, cd
= 92, ac = 187, bce = 155, and abcde = 185.
(a) Verify that the treatment combinations given are correct. Estimate the effects, assuming three-factor
and higher interactions are negligible.
A
+
+
+
+
B
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
49
Model
B
45
Error
C
10.5
Error
D
-18
Error
E
-14.5
Error
AC
13.5
Error
AE
-14.5
Lenth's ME 81.8727
Lenth's SME 195.937
O rd e r Q u a n ti ty
R e -o rd er P o i n t
S e tu p Co st
B a cko rde r C o st
C a rryi n g Co st
D=AB
+
+
+
+
SumSqr
E=BC
+
+
+
+
4802
4050
220.5
648
420.5
364.5
420.5
43.9502
37.0675
2.01812
5.93081
3.84862
3.33608
3.84862
Ha lf No rm al p lo t
99
99
95
A
90
80
H alf N orm a l % prob ability
N orm a l % proba bility
de
ae
b
abd
cd
ac
bce
abcde
% Contribtn
No rm a l p lo t
DE S IG N-E X P E RT P l o t
A vg A n n u a l Co st
A:
B:
C:
D:
E:
C
+
+
+
+
B
70
50
30
20
10
5
97
95
A
90
85
80
B
70
60
40
20
1
0
-1 8 .0 0
-1 .2 5
1 5 .5 0
3 2 .2 5
4 9 .0 0
0 .0 0
Effect
1 2 .2 5
2 4 .5 0
3 6 .7 5
4 9 .0 0
Effe ct
(b) Suppose that a second fraction is added to the first, for example ade = 136, e = 93, ab = 187, bd =
153, acd = 139, c = 99, abce - 191, and bcde = 150. How was this second fraction obtained? Add
this data to the original fraction, and estimate the effects.
8-33
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
This second fraction is formed by reversing the signs of factor A.
A
+
+
+
+
-
B
+
+
+
+
Design Expert Output
Term
Effect
Model
Intercept
Model
A
44.25
Model
B
49.25
Error
C
6.5
Error
D
-8
Error
E
-8.25
Error
AB
-10
Error
AC
7.25
Error
AD
-4.25
Error
AE
-6
Error
BD
4.75
Error
CD
-8.5
Error
DE
6.25
Error
ACD
-6.25
Error
ADE
4
Lenth's ME 25.1188
Lenth's SME 51.5273
O rd e r Q u a n ti ty
R e -o rd er P o i n t
S e tu p Co st
B a cko rde r C o st
C a rryi n g Co st
D=AB
+
+
+
+
SumSqr
E=BC
+
+
+
+
% Contribtn
7832.25
9702.25
169
256
272.25
400
210.25
72.25
144
90.25
289
156.25
156.25
64
39.5289
48.9666
0.852932
1.29202
1.37403
2.01877
1.06112
0.364641
0.726759
0.455486
1.45856
0.788584
0.788584
0.323004
Ha lf No rm al p lo t
99
99
90
H alf N orm a l % prob ability
B
95
N orm a l % proba bility
ade
e
ab
bd
acd
c
abce
bcde
No rm a l p lo t
DE S IG N-E X P E RT P l o t
A vg A n n u a l Co st
A:
B:
C:
D:
E:
C
+
+
+
+
A
80
70
50
30
20
10
5
B
97
95
A
90
85
80
70
60
40
20
1
0
-9 .2 5
5 .6 3
2 0 .5 0
3 5 .3 8
5 0 .2 5
0 .0 0
1 2 .3 1
Effect
2 4 .6 3
3 6 .9 4
4 9 .2 5
Effe ct
(c) Suppose that the fraction abc = 189, ce = 96, bcd = 154, acde = 135, abe = 193, bde = 152, ad = 137,
and (1) = 98 was run. How was this fraction obtained? Add this data to the original fraction and
estimate the effects.
This second fraction is formed by reversing the signs of all factors.
A
+
B
+
C
+
D=AB
-
8-34
E=BC
-
abc
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+
+
+
-
+
+
+
-
Design Expert Output
Term
Effect
Model
Intercept
Model
A
43.75
Model
B
50.25
Error
C
4.5
Error
D
-8.75
Error
E
-7.5
Error
AB
-9.25
Error
AC
6
Error
AD
-5.25
Error
AE
-6.5
Error
BC
-7
Error
BD
5.25
Error
BE
6
Error
ABC
-8
Error
ABE
7.5
Lenth's ME 26.5964
Lenth's SME 54.5583
O rd e r Q u a n ti ty
R e -o rd er P o i n t
S e tu p Co st
B a cko rde r C o st
C a rryi n g Co st
+
+
+
+
-
SumSqr
+
+
+
+
% Contribtn
7656.25
10100.3
81
306.25
225
342.25
144
110.25
169
196
110.25
144
256
225
38.1563
50.3364
0.403678
1.52625
1.12133
1.70566
0.71765
0.549451
0.842242
0.976801
0.549451
0.71765
1.27582
1.12133
Ha lf No rm al p lo t
99
99
90
H alf N orm a l % prob ability
B
95
N orm a l % proba bility
bcd
acde
ce
abe
bde
ad
(1)
No rm a l p lo t
DE S IG N-E X P E RT P l o t
A vg A n n u a l Co st
A:
B:
C:
D:
E:
+
+
+
-
A
80
70
50
30
20
10
5
B
97
95
A
90
85
80
70
60
40
20
1
0
-9 .2 5
5 .6 3
2 0 .5 0
3 5 .3 8
5 0 .2 5
0 .0 0
1 2 .5 6
Effect
2 5 .1 3
3 7 .6 9
5 0 .2 5
Effe ct
8-20 Construct a 2 51 design. Show how the design may be run in two blocks of eight observations
each. Are any main effects or two-factor interactions confounded with blocks?
A
+
+
+
B
+
+
-
C
+
+
D
-
E=ABCD
+
+
+
8-35
e
a
b
abe
c
ace
Blocks = AB
+
+
+
-
Block
1
2
2
1
1
2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
bce
abc
d
ade
bde
abd
cde
acd
bcd
abcde
+
+
+
+
+
2
1
1
2
2
1
1
2
2
1
Blocks are confounded with AB and CDE.
8-21 Construct a 2 7 2 design. Show how the design may be run in four blocks of eight observations
each. Are any main effects or two-factor interactions confounded with blocks?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
A
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
F=CDE
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
G=ABC
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
(1)
ag
bg
ab
cfg
acf
bcf
abcfg
df
adfg
bdfg
abdf
cdg
acd
bcd
abcdg
ef
aefg
befg
abef
ceg
ace
bce
abceg
de
adeg
bdeg
abde
cdefg
acdef
bcdef
abcdefg
Block=ACE
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Block=BFG
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Block assignment
1
4
1
4
3
2
3
2
2
3
2
3
4
1
4
1
4
1
4
1
2
3
2
3
3
2
3
2
1
4
1
4
Blocks are confounded with ACE, BFG, and ABCEFG.
8-22 Irregular fractions of the 2k [John (1971)]. Consider a 24 design. We must estimate the four
main effects and the six two-factor interactions, but the full 24 factorial cannot be run. The largest
possible block contains 12 runs. These 12 runs can be obtained from the four one-quarter fractions
defined by I = r AB = r ACD = r BCD by omitting the principal fraction. Show how the remaining three
24-2 fractions can be combined to estimate the required effects, assuming that three-factor and higher
interactions are negligible. This design could be thought of as a three-quarter fraction.
The four 24-2 fractions are as follows:
8-36
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(1) I=+AB=+ACD=+BCD
Runs: c,d,ab,abcd
(2) I=+AB=-ACD=-BCD
Runs: (1), cd, abc, abd
(3) I=-AB=+ACD=-BCD
Runs: a, bc, bd, acd
(4) I=-AB=-ACD=+BCD
Runs: b, ac, ad, bcd
If we do not run the principal fraction (1), then we can combine the remaining 3 fractions to from 3 onehalf fractions of the 24 as follows:
Fraction 1: (2) + (3) implies I=-BCD. This fraction estimates: A, AB, AC, and AD
Fraction 2: (2) + (4) implies I=-ACD. This fraction estimates: B, BC, BD, and AB
Fraction 3: (3) + (4) implies I=-AB. This fraction estimates: C, D, and CD
In estimating these effects we assume that all three-factor and higher interactions are negligible. Note
that AB is estimated in two of the one-half fractions: 1 and 2. We would average these quantities and
obtain a single estimate of AB. John (1971, pp. 161-163) discusses this design and shows that the
estimates obtained above are also the least squares estimates. John also derives the variances and
covariances of these estimators.
8-23 Carbon anodes used in a smelting process are baked in a ring furnace. An experiment is run in the
furnace to determine which factors influence the weight of packing material that is stuck to the anodes
after baking. Six variables are of interest, each at two levels: A = pitch/fines ratio (0.45, 0.55); B =
packing material type (1, 2); C = packing material temperature (ambient, 325 C); D = flue location
(inside, outside); E = pit temperature (ambient, 195 C); and F = delay time before packing (zero, 24
hours). A 26-3 design is run, and three replicates are obtained at each of the design points. The weight of
packing material stuck to the anodes is measured in grams. The data in run order are as follows: abd =
(984, 826, 936); abcdef = (1275, 976, 1457); be = (1217, 1201, 890); af = (1474, 1164, 1541); def =
(1320, 1156, 913); cd = (765, 705, 821); ace = (1338, 1254, 1294); and bcf = (1325, 1299, 1253). We
wish to minimize the amount stuck packing material.
(a) Verify that the eight runs correspond to a 2 6III3 design. What is the alias structure?
A
+
+
+
+
B
+
+
+
+
C
+
+
+
+
D=AB
+
+
+
+
E=AC
+
+
+
+
F=BC
+
+
+
+
def
af
be
abd
cd
ace
bcf
abcdef
I=ABD=ACE=BCF=BCDE=ACDF=ABEF=DEF, Resolution III
A=BD=CE=CDF=BEF
B=AD=CF=CDE=AEF
C=AE=BF=BDE=ADF
8-37
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
D=AB=EF=BCE=ACF
E=AC=DF=BCD=ABF
F=BC=DE=ACD=ABE
CD=BE=AF=ABC=ADE=BDF=CEF
(b) Use the average weight as a response. What factors appear to be influential?
Design Expert Output
Term
Effect
Model
Intercept
Model
A
137.9
Error
B
-8.9
Error
C
0.221108
Model
D
-259.6
Model
E
99.7667
Model
F
243.567
Error
BC
-38.0306
Lenth's ME 563.322
Lenth's SME 1348.14
SumSqr
% Contribtn
37996.1
156.056
2094.02
136168
27246.7
107863
2629.69
12.0947
0.049675
0.666559
43.3443
8.67305
34.3345
0.837072
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
D
90
85
80
F
A
70
60
E
40
20
0
0 .0 0
6 4 .9 0
1 2 9 .80
1 9 4 .70
2 5 9 .60
Effe ct
Factors A, D, E and F (and their aliases) are apparently important.
(c) Use the range of the weights as a response. What factors appear to be influential?
Design Expert Output
Term
Effect
Model
Intercept
Error
A
44.5
Error
B
13.5
Model
C
-129
Error
D
75.5
Model
E
144
Model
F
163
Model
AF
145
Lenth's ME 728.384
Lenth's SME 1743.17
SumSqr
% Contribtn
3960.5
364.5
33282
11400.5
41472
53138
42050
2.13311
0.196319
17.9256
6.14028
22.3367
28.62
22.648
Factors C, E, F and the AF interaction (and their aliases) appear to be large.
8-38
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
H alf N orm a l % prob ability
99
97
95
F
90
85
AF
80
E
70
60
C
40
20
0
0 .0 0
4 0 .7 5
8 1 .5 0
1 2 2 .25
1 6 3 .00
Effe ct
(d) What recommendations would you make to the process engineers?
It is not known exactly what to do here, since A, D, E and F are large effects, and because the design is
resolution III, the main effects are aliased with two-factor interactions. Note, for example, that D is
aliased with EF and the main effect could really be a EF interaction. If the main effects are really
important, then setting all factors at the low level would minimize the amount of material stuck to the
anodes. It would be necessary to run additional experiments to confirm these findings.
8-24 A 16-run experiment was performed in a semiconductor manufacturing plant to study the effects of
six factors on the curvature or camber of the substrate devices produced. The six variables and their levels
are shown below:
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Lamination
Temperature
(c)
55
75
55
75
55
75
55
75
55
75
55
75
55
75
55
75
Lamination
Time
(s)
10
10
25
25
10
10
25
25
10
10
25
25
10
10
25
25
Lamination
Pressure
(tn)
5
5
5
5
10
10
10
10
5
5
5
5
10
10
10
10
Firing
Temperature
(c)
1580
1580
1580
1580
1580
1580
1580
1580
1620
1620
1620
1620
1620
1620
1620
1620
Firing
Cycle Time
(h)
17.5
29
29
17.5
29
17.5
17.5
29
17.5
29
29
17.5
29
17.5
17.5
29
Firing
Dew Point
(c)
20
26
20
26
26
20
26
20
26
20
26
20
20
26
20
26
Each run was replicated four times , and a camber measurement was taken on the substrate. The data are
shown below:
Run
1
2
Camber
1
0.0167
0.0062
for
2
0.0128
0.0066
Replicate
3
0.0149
0.0044
(in/in)
4
0.0185
0.0020
8-39
Total
(10-4 in/in)
629
192
Mean
(10-4 in/in)
157.25
48.00
Standard
Deviation
24.418
20.976
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
3
4
5
6
7
8
9
10
11
12
13
14
15
16
0.0041
0.0073
0.0047
0.0219
0.0121
0.0255
0.0032
0.0078
0.0043
0.0186
0.0110
0.0065
0.0155
0.0093
0.0043
0.0081
0.0047
0.0258
0.0090
0.0250
0.0023
0.0158
0.0027
0.0137
0.0086
0.0109
0.0158
0.0124
0.0042
0.0039
0.0040
0.0147
0.0092
0.0226
0.0077
0.0060
0.0028
0.0158
0.0101
0.0126
0.0145
0.0110
0.0050
0.0030
0.0089
0.0296
0.0086
0.0169
0.0069
0.0045
0.0028
0.0159
0.0158
0.0071
0.0145
0.0133
176
223
223
920
389
900
201
341
126
640
455
371
603
460
44.00
55.75
55.75
230.00
97.25
225.00
50.25
85.25
31.50
160.00
113.75
92.75
150.75
115.00
4.083
25.025
22.410
63.639
16.029
39.420
26.725
50.341
7.681
20.083
31.120
29.510
6.750
17.450
(a) What type of design did the experimenters use?
The 2 6IV2 , a 16-run design.
(b) What are the alias relationships in this design? The defining relation is I=ABCE=ACDF=BDEF
A (ABCE)=
B (ABCE)=
C (ABCE)=
D (ABCE)=
E (ABCE)=
F (ABCE)=
AB (ABCE)=
AC (ABCE)=
AD (ABCE)=
AE (ABCE)=
AF (ABCE)=
BD (ABCE)=
BF (ABCE)=
BCE
ACE
ABE
ABCDE
ABC
ABCEF
CE
BE
BCDE
BC
BCEF
ACDE
ACEF
A (ACDF)=
B (ACDF)=
C (ACDF)=
D (ACDF)=
E (ACDF)=
F (ACDF)=
AB (ACDF)=
AC (ACDF)=
AD (ACDF)=
AE (ACDF)=
AF (ACDF)=
BD (ACDF)=
BF (ACDF)=
CDF
ABCDF
ADF
ACF
ACDEF
ACD
BCDF
DF
CF
CDEF
CD
ABCF
ABCD
A (BDEF)=
B (BDEF)=
C (BDEF)=
D (BDEF)=
E (BDEF)=
F (BDEF)=
AB (BDEF)=
AC (BDEF)=
AD (BDEF)=
AE (BDEF)=
AF (BDEF)=
BD (BDEF)=
BF (BDEF)=
ABCDEF
DEF
BCDEF
BEF
BDF
BDE
ADEF
ABCDEF
ABEF
ABDF
ABDE
EF
DE
(c) Do any of the process variables affect average camber?
Yes, per the analysis below, variables A, C, D, and F affect average camber.
Design Expert Output
Term
Model
Intercept
Model
A
Error
B
Model
C
Error
D
Model
E
Model
F
Error
AB
Error
AC
Error
AD
Error
AE
Error
AF
Error
BC
Error
BD
Error
BE
Error
BF
Error
CD
Error
CE
Error
CF
Error
DE
Error
DF
Error
EF
Error
ABC
Error
ABD
Effect
SumSqr
% Contribtn
38.9063
5.78125
56.0313
-14.2188
-34.4687
-77.4688
19.1563
22.4063
-12.2188
18.1563
-19.7187
Aliased
23.0313
Aliased
7.40625
Aliased
Aliased
Aliased
Aliased
Aliased
Aliased
Aliased
0.53125
6054.79
133.691
12558
808.691
4752.38
24005.6
1467.85
2008.16
597.191
1318.6
1555.32
10.2962
0.227344
21.355
1.37519
8.08148
40.8219
2.49609
3.4149
1.01553
2.24229
2.64483
2121.75
3.60807
219.41
0.37311
1.12891
0.00191972
8-40
A=BCE=CDF=ABDEF
B=ACE=ABCDF=DEF
C=ABE=ADF=BCDEF
D=ABCDE=ACF=BEF
E=ABC=ABDEF=BDF
F=ABCEF=ACD=BDE
AB=CE=BCDF=ADEF
AC=BE=DF=ABCDEF
AD=BCDE=CF=ABEF
AE=BC=CDEF=ABDF
AF=BCEF=CD=ABDE
BD=ACDE=ABCF=EF
BF=ACEF=ABCD=DE
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
Error
ABE
ABF
Lenth's ME
Lenth's SME
Aliased
-17.3438
71.9361
146.041
1203.22
2.04609
No rm a l p lot
DE S IG N-E X P E RT P l o t
Ca m b e r A vg
Lam T emp
Lam T im e
L a m P re s
Fi re T e m p
Fi re T i m e
Fi re DP
99
C
95
N orm al % probability
A:
B:
C:
D:
E:
F:
90
A
80
70
50
30
20
10
E
5
F
1
-7 7 .47
-4 4 .09
-1 0 .72
2 2 .6 6
5 6 .0 3
Effe ct
Design Expert Output
Response:
Camber Avg in in/in
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
47370.80
4
11842.70
A
6054.79
1
6054.79
C
12558.00
1
12558.00
E
4752.38
1
4752.38
F
24005.63
1
24005.63
Residual
11435.01
11
1039.55
Cor Total
58805.81
15
F
Value
11.39
5.82
12.08
4.57
23.09
Prob > F
0.0007
0.0344
0.0052
0.0558
0.0005
significant
The Model F-value of 11.39 implies the model is significant. There is only
a 0.07% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
32.24
107.02
30.13
24193.08
Factor
Intercept
A-Lam Temp
C-Lam Pres
E-Fire Time
F-Fire DP
Coefficient
Estimate
107.02
19.45
28.02
-17.23
-38.73
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
Standard
Error
8.06
8.06
8.06
8.06
8.06
0.8055
0.7348
0.5886
11.478
95% CI
Low
89.27
1.71
10.27
-34.98
-56.48
Final Equation in Terms of Coded Factors:
Camber Avg
+107.02
+19.45
+28.02
-17.23
-38.73
=
*A
*C
*E
*F
Final Equation in Terms of Actual Factors:
8-41
95% CI
High
124.76
37.19
45.76
0.51
-20.99
VIF
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Camber Avg
+263.17380
+1.94531
+11.20625
-2.99728
-12.91146
=
* Lam Temp
* Lam Pres
* Fire Time
* Fire DP
(d) Do any of the process variables affect the variability in camber measurements?
Yes, A, B, F, and AF interaction affect the variability in camber measurements.
Effect
SumSqr
% Contribtn
15.9035
-16.5773
5.8745
-3.2925
-2.33725
-9.256
0.95525
2.524
-4.6265
-0.18025
-10.8745
Aliased
-4.85575
Aliased
8.21825
Aliased
Aliased
Aliased
Aliased
Aliased
Aliased
Aliased
-0.68125
Aliased
3.39825
17.8392
36.2162
1011.69
1099.22
138.039
43.3622
21.851
342.694
3.65001
25.4823
85.618
0.12996
473.019
27.6623
30.0558
3.77437
1.18564
0.597466
9.37021
0.0998014
0.696757
2.34103
0.00355347
12.9337
94.3132
2.57879
270.159
7.38689
1.85641
0.0507593
46.1924
1.26303
No rm a l p lot
DE S IG N-E X P E RT P l o t
Ca m b e r S tDe v
A:
B:
C:
D:
E:
F:
Lam T emp
Lam T im e
L a m P re s
Fi re T e m p
Fi re T i m e
Fi re DP
99
A
95
N orm al % probability
Design Expert Output
Term
Model
Intercept
Model
A
Model
B
Error
C
Error
D
Error
E
Model
F
Error
AB
Error
AC
Error
AD
Error
AE
Model
AF
Error
BC
Error
BD
Error
BE
Error
BF
Error
CD
Error
CE
Error
CF
Error
DE
Error
DF
Error
EF
Error
ABC
Error
ABD
Error
ABE
Error
ABF
Lenth's ME
Lenth's SME
90
80
70
50
30
20
F
AF
10
5
B
1
-1 6 .58
-8 .4 6
-0 .3 4
Effe ct
8-42
7 .7 8
1 5 .9 0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Response:
Camber StDev
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2926.62
4
731.65
A
1011.69
1
1011.69
B
1099.22
1
1099.22
F
342.69
1
342.69
AF
473.02
1
473.02
Residual
730.65
11
66.42
Cor Total
3657.27
15
F
Value
11.02
15.23
16.55
5.16
7.12
Prob > F
0.0008
0.0025
0.0019
0.0442
0.0218
significant
The Model F-value of 11.02 implies the model is significant. There is only
a 0.08% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
8.15
25.35
32.15
1545.84
Factor
Intercept
A-Lam Temp
B-Lam Time
F-Fire DP
AF
Coefficient
Estimate
25.35
7.95
-8.29
-4.63
-5.44
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
Standard
Error
2.04
2.04
2.04
2.04
2.04
0.8002
0.7276
0.5773
9.516
95% CI
Low
20.87
3.47
-12.77
-9.11
-9.92
95% CI
High
29.84
12.44
-3.80
-0.14
-0.95
VIF
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Camber StDev
+25.35
+7.95
-8.29
-4.63
-5.44
=
*A
*B
*F
*A*F
Final Equation in Terms of Actual Factors:
Camber StDev
-242.46746
+4.96373
-1.10515
+10.23804
-0.18124
=
* Lam Temp
* Lam Time
* Fire DP
* Lam Temp * Fire DP
(e) If it is important to reduce camber as much as possible, what recommendations would you make?
8-43
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One F a ctor P lot
One F a ctor P lot
1 8 0 .37 5
1 8 0 .37 5
C am b er Avg
230
C am b er Avg
230
1 3 0 .75
1 3 0 .75
8 1 .1 25
8 1 .1 25
3 1 .5
3 1 .5
5 5 .0 0
6 0 .0 0
6 5 .0 0
7 0 .0 0
7 5 .0 0
5 .0 0
8 .7 5
Lam Pre s
One F a ctor P lot
One F a ctor P lot
1 8 0 .37 5
1 8 0 .37 5
C am b er Avg
230
C am b er Avg
7 .5 0
Lam Tem p
230
1 3 0 .75
8 1 .1 25
3 1 .5
3 1 .5
2 0 .3 8
2 3 .2 5
2 6 .1 3
2 9 .0 0
2 0 .0 0
Fire Tim e
2 1 .5 0
2 3 .0 0
Fire D P
8-44
1 0 .0 0
1 3 0 .75
8 1 .1 25
1 7 .5 0
6 .2 5
2 4 .5 0
2 6 .0 0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
Ca m b e r S tDe v
6 3 .6 39
One F a ctor P lot
DE S IG N-E X P E RT P l o t
Fire D P
Ca m b e r S tDe v
X = A: Lam T emp
Y = F: Fire DP
6 3 .6 39
X = B: Lam T im e
A c tu al Fa c tors
4 8 .7 5
A : L a m T e m p = 6 5 .0 0
C: L a m P re s = 7.5 0
D: Fi re T e m p = 1 6 0 0 .0 0
E : Fi re T i m e = 2 3 .2 5
F: Fi re DP = 2 3 .0 0
3 3 .8 61
C am b er StD ev
C am b er StD ev
4 8 .7 5
F- 2 0.0 0 0
F+ 2 6.0 0 0
A ctu al Fa ctors
B : L a m T i m e = 1 7 .5 0
C: L a m P re s = 7.5 0
3 3 .8 61
D: Fi re T e m p = 1 6 0 0 .0 0
E : Fi re T i m e = 2 3 .2 5
1 8 .9 72
1 8 .9 72
4 .0 8 3
4 .0 8 3
5 5 .0 0
6 0 .0 0
6 5 .0 0
7 0 .0 0
7 5 .0 0
1 0 .0 0
Lam Tem p
1 3 .7 5
1 7 .5 0
2 1 .2 5
2 5 .0 0
Lam Tim e
Run A and C at the low level and E and F at the high level. B at the low level enables a lower variation
without affecting the average camber.
8-25 A spin coater is used to apply photoresist to a bare silicon wafer. This operation usually occurs
early in the semiconductor manufacturing process, and the average coating thickness and the variability in
the coating thickness has an important impact on downstream manufacturing steps. Six variables are
used in the experiment. The variables and their high and low levels are as follows:
Factor
Final Spin Speed
Acceleration Rate
Volume of Resist Applied
Time of Spin
Resist Batch Variation
Exhaust Pressure
Low Level
7350 rpm
5
3 cc
14 s
Batch 1
Cover Off
High Level
6650 rpm
20
5 cc
6s
Batch 2
Cover On
The experimenter decides to use a 26-1 design and to make three readings on resist thickness on each test
wafer. The data are shown in table 8-29.
Table 8-29
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
A
Volume
5
5
3
3
3
5
3
5
5
3
3
3
5
3
5
5
3
3
B
Batch
2
1
1
2
1
1
1
2
1
1
2
1
1
1
2
2
2
1
C
Time
14
6
6
14
14
6
6
14
14
14
14
6
6
6
14
6
14
14
D
Speed
7350
7350
6650
7350
7350
6650
7350
6650
6650
6650
6650
7350
6650
6650
7350
7350
7350
6650
E
Acc.
5
5
5
20
5
20
5
20
5
5
20
20
5
20
20
5
5
20
F
Cover
Off
Off
Off
Off
Off
Off
On
Off
Off
On
On
Off
On
On
On
On
On
Off
8-45
Left
4531
4446
4452
4316
4307
4470
4496
4542
4621
4653
4480
4221
4620
4455
4255
4490
4514
4494
Resist
Center
4531
4464
4490
4328
4295
4492
4502
4547
4643
4670
4486
4233
4641
4480
4288
4534
4551
4503
Thick
Right
4515
4428
4452
4308
4289
4495
4482
4538
4613
4645
4470
4217
4619
4466
4243
4523
4540
4496
ness
Avg.
4525.7
4446
4464.7
4317.3
4297
4485.7
4493.3
4542.3
4625.7
4656
4478.7
4223.7
4626.7
4467
4262
4515.7
4535
4497.7
Range
16
36
38
20
18
25
20
9
30
25
16
16
22
25
45
44
37
9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
19
20
21
22
23
24
25
26
27
28
29
30
31
32
5
3
5
3
5
3
5
3
3
5
5
5
5
3
2
2
1
2
1
2
1
2
1
2
1
2
2
2
6
6
14
6
14
6
14
6
14
6
6
6
14
14
7350
7350
6650
6650
7350
7350
7350
6650
7350
6650
7350
6650
6650
6650
20
5
20
5
20
20
5
20
20
5
20
20
5
5
Off
Off
On
On
Off
On
On
Off
On
Off
On
On
On
Off
4293
4534
4460
4650
4231
4225
4381
4533
4194
4666
4180
4465
4653
4683
4306
4545
4457
4688
4244
4228
4391
4521
4230
4695
4213
4496
4685
4712
4302
4512
4436
4656
4230
4208
4376
4511
4172
4672
4197
4463
4665
4677
4300.3
4530.3
4451
4664.7
4235
4220.3
4382.7
4521.7
4198.7
4677.7
4196.7
4474.7
4667.7
4690.7
13
33
24
38
14
20
15
22
58
29
33
33
32
35
(a) Verify that this is a 26-1 design. Discuss the alias relationships in this design.
I=ABCDEF. This is a resolution VI design where main effects are aliased with five-factor interactions
and two-factor interactions are aliased with four-factor interactions.
(b) What factors appear to affect average resist thickness?
Factors B, D, and E appear to affect the average resist thickness.
Design Expert Output
Term
Model
Intercept
Error
A
Model
B
Error
C
Model
D
Model
E
Error
F
Error
AB
Error
AC
Error
AD
Error
AE
Error
AF
Error
BC
Error
BD
Error
BE
Error
BF
Error
CD
Error
CE
Error
CF
Error
DE
Error
DF
Error
EF
Error
ABC
Error
ABD
Error
ABE
Error
ABF
Error
ACD
Error
ACE
Error
ACF
Error
ADE
Error
ADF
Error
AEF
Error
BCD
Error
BCE
Error
BCF
Error
BDE
Error
BDF
Error
BEF
Error
CDE
Error
CDF
Error
CEF
Effect
SumSqr
% Contribtn
9.925
73.575
3.375
-207.062
-182.925
-5.6625
-9
-7.3
-3.8625
-7.1
-26.9875
10.875
18.1125
-28.35
-30.2375
-24.9875
8.2
-6.7875
-38.5375
-3.2
-41.1625
0.375
Aliased
16.5
31.4125
15.5875
Aliased
Aliased
9.5375
Aliased
Aliased
29.0875
-1.625
Aliased
-1.8875
3.95
Aliased
Aliased
Aliased
3.1375
788.045
43306.2
91.125
342999
267692
256.511
648
426.32
119.351
403.28
5826.6
946.125
2624.5
6429.78
7314.45
4995
537.92
368.561
11881.1
81.92
13554.8
1.125
0.107795
5.92378
0.0124648
46.9182
36.6172
0.0350877
0.0886387
0.0583155
0.0163258
0.0551639
0.79701
0.129419
0.359001
0.879518
1.00053
0.683257
0.0735811
0.0504148
1.6252
0.0112057
1.85414
0.000153887
2178
7893.96
1943.76
0.297925
1.0798
0.265883
727.711
0.0995423
6768.66
21.125
0.925873
0.00288965
28.5013
124.82
0.00389863
0.0170739
78.7512
0.0107722
8-46
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
DEF
Aliased
Lenth's ME 28.6178
Lenth's SME 54.4118
No rm a l p lot
DE S IG N-E X P E RT P l o t
T h i ck A vg
Volum e
B a tch
T im e
Speed
A cc
Co ve r
99
B
95
N orm al % probability
A:
B:
C:
D:
E:
F:
90
80
70
50
30
20
10
5
1
E
D
-2 0 7 .0 6
-1 3 6 .9 0
-6 6 .74
3 .4 2
7 3 .5 7
Effe ct
Design Expert Output
Response:
Thick Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
F
Source
Squares
DF
Square
Value
Model
6.540E+005 3
2.180E+005 79.21
B
43306.24
1
43306.24
15.74
D
3.430E+005 1
3.430E+005 124.63
E
2.677E+005 1
2.677E+005 97.27
Residual
77059.83
28
2752.14
Cor Total
7.311E+005 31
Prob > F
< 0.0001
0.0005
< 0.0001
< 0.0001
significant
The Model F-value of 79.21 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
52.46
4458.51
1.18
1.006E+005
Factor
Intercept
B-Batch
D-Speed
E-Acc
Coefficient
Estimate
4458.51
36.79
-103.53
-91.46
DF
1
1
1
1
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Standard
Error
9.27
9.27
9.27
9.27
0.8946
0.8833
0.8623
24.993
95% CI
Low
4439.52
17.79
-122.53
-110.46
Final Equation in Terms of Coded Factors:
Thick Avg
+4458.51
+36.79
-103.53
-91.46
=
*B
*D
*E
Final Equation in Terms of Actual Factors:
Batch
Thick Avg
+6644.78750
Batch 1
=
8-47
95% CI
High
4477.51
55.78
-84.53
-72.47
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-0.29580
-12.19500
Batch
Thick Avg
+6718.36250
-0.29580
-12.19500
* Speed
* Acc
Batch 2
=
* Speed
* Acc
(c) Since the volume of resist applied has little effect on average thickness, does this have any important
practical implications for the process engineers?
Yes, less material could be used.
(d) Project this design into a smaller design involving only the significant factors. Graphically display
the results. Does this aid in interpretation?
C ube Graph
Thick Avg
422 6.7 3
D+
448 3.23
Sp eed
440 9.66
430 0.31
E+
450 7.37
443 3.7 9
Acc
D-
B-
469 0.29
461 6.72
Ba tch
B+
E-
The cube plot usually assists the experimenter in drawing conclusions.
(e) Use the range of resist thickness as a response variable. Is there any indication that any of these
factors affect the variability in resist thickness?
Design Expert Output
Term
Model
Intercept
Model
A
Model
B
Error
C
Error
D
Model
E
Model
F
Model
AB
Error
AC
Error
AD
Error
AE
Model
AF
Error
BC
Error
BD
Error
BE
Model
BF
Error
CD
Effect
SumSqr
% Contribtn
-0.625
2.125
-2.75
1.625
-5.375
7.75
0.625
-3.5
-0.125
1.875
1.75
0
0.125
-5.375
3.25
3.75
3.125
36.125
60.5
21.125
231.125
480.5
3.125
98
0.125
28.125
24.5
0
0.125
231.125
84.5
112.5
0.0777387
0.89866
1.50502
0.525514
5.74956
11.9531
0.0777387
2.43789
0.00310955
0.699649
0.609472
0
0.00310955
5.74956
2.10206
2.79859
8-48
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
Error
Error
Error
Model
Error
Error
Error
Model
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
CE
3.75
CF
4.875
DE
5.375
DF
5.5
EF
8
ABC
Aliased
ABD
Aliased
ABE
3.625
ABF
9
ACD
-6.5
ACE
Aliased
ACF
Aliased
ADE
-3.375
ADF
-0.5
AEF
1
BCD
Aliased
BCE
Aliased
BCF
Aliased
BDE
-2.625
BDF
-0.5
BEF
Aliased
CDE
Aliased
CDF
Aliased
CEF
2.125
DEF
2
Lenth's ME 9.15104
Lenth's SME 17.3991
112.5
190.125
231.125
242
512
2.79859
4.72962
5.74956
6.02009
12.7367
105.125
648
338
2.61513
16.1199
8.40822
91.125
2
8
2.26686
0.0497528
0.199011
55.125
2
1.37131
0.0497528
36.125
32
0.89866
0.796045
No rm a l p lot
DE S IG N-E X P E RT P l o t
T h i ck S tDe v
Volum e
B a tch
T im e
Speed
A cc
Co ve r
99
ABF
95
N orm al % probability
A:
B:
C:
D:
E:
F:
EF
F
90
80
70
B BF
AF
AB
50
30
20
A
10
5
E
1
-6 .5 0
-2 .6 2
1 .2 5
5 .1 3
9 .0 0
Effe ct
Design Expert Output
Response:
Thick StDev
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2023.00
9
224.78
A
3.13
1
3.13
B
36.13
1
36.13
E
231.12
1
231.12
F
480.50
1
480.50
AB
3.12
1
3.12
AF
24.50
1
24.50
BF
84.50
1
84.50
EF
512.00
1
512.00
ABF
648.00
1
648.00
F
Value
2.48
0.034
0.40
2.55
5.29
0.034
0.27
0.93
5.64
7.14
8-49
Prob > F
0.0400
0.8545
0.5346
0.1248
0.0313
0.8545
0.6086
0.3451
0.0267
0.0139
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residual
Cor Total
1996.88
4019.88
22
31
90.77
The Model F-value of 2.48 implies the model is significant. There is only
a 4.00% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
9.53
26.56
35.87
4224.79
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Factor
Intercept
A-Volume
B-Batch
E-Acc
F-Cover
AB
AF
BF
EF
ABF
Coefficient
Estimate
26.56
-0.31
1.06
-2.69
3.88
0.31
0.88
1.63
4.00
4.50
DF
1
1
1
1
1
1
1
1
1
1
Standard
Error
1.68
1.68
1.68
1.68
1.68
1.68
1.68
1.68
1.68
1.68
0.5032
0.3000
-0.0510
5.586
95% CI
Low
23.07
-3.81
-2.43
-6.18
0.38
-3.18
-2.62
-1.87
0.51
1.01
Final Equation in Terms of Coded Factors:
Thick StDev
+26.56
-0.31
+1.06
-2.69
+3.88
+0.31
+0.88
+1.63
+4.00
+4.50
=
*A
*B
*E
*F
*A*B
*A*F
*B*F
*E*F
*A*B*F
Final Equation in Terms of Actual Factors:
Batch
Cover
Thick StDev
+22.39583
+3.00000
-0.89167
Batch 1
Off
=
Batch
Cover
Thick StDev
+54.77083
-5.37500
-0.89167
Batch 2
Off
=
Batch
Cover
Thick StDev
+42.56250
-4.25000
+0.17500
Batch 1
On
=
Batch
Cover
Thick StDev
+9.43750
+5.37500
+0.17500
Batch 2
On
=
* Volume
* Acc
* Volume
* Acc
* Volume
* Acc
* Volume
* Acc
8-50
95% CI
High
30.06
3.18
4.56
0.81
7.37
3.81
4.37
5.12
7.49
7.99
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The model here for variability isn’t very strong. Notice the small value of R2, and in particular, the
adjusted R2. Often we find that obtaining a good model for a response that expresses variability isn’t as
easy as finding a satisfactory model for a response that essentially measures the mean.
(f) Where would you recommend that the process engineers run the process?
Considering only the average thickness results, the engineers could use factors B, D and E to put the
process mean at target. Then the engineer could consider the other factors on the range model to try to set
the factors to reduce the variation in thickness at that mean.
8-26 Harry and Judy Peterson-Nedry (two friends of the author) own a vineyard in Oregon. They grow
several varieties of grapes and manufacture wine. Harry and Judy have used factorial designs for process
and product development in the winemaking segment of their business. This problem describes the
experiment conducted for their 1985 Pinot Noir. Eight variables, shown below, were originally studied in
this experiment:
A
B
C
D
E
F
G
H
Variable
Pinot Noir Clone
Oak Type
Age of Barrel
Yeast/Skin Contact
Stems
Barrel Toast
Whole Cluster
Fermentation Temperature
Low Level
Pommard
Allier
Old
Champagne
None
Light
None
Low (75 F Max)
High Level
Wadenswil
Troncais
New
Montrachet
All
Medium
10%
High (92 F Max)
Harry and Judy decided to use a 28IV4 design with 16 runs. The wine was taste-tested by a panel of experts
on 8 March 1986. Each expert ranked the 16 samples of wine tasted, with rank 1 being the best. The
design and taste-test panel results are shown in Table 8-30.
Table 8-30
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
F
+
+
+
+
+
+
+
+
G
+
+
+
+
+
+
+
+
H
+
+
+
+
+
+
+
+
HPN
12
10
14
9
8
16
6
15
1
7
13
3
2
4
5
11
JPN
6
7
13
9
8
12
5
16
2
11
3
1
10
4
15
14
CAL
13
14
10
7
11
15
6
16
3
4
8
5
2
1
9
12
DCM
10
14
11
9
8
16
5
15
3
7
12
1
4
2
6
13
RGB
7
9
15
12
10
16
3
14
2
6
8
4
5
1
11
13
ybar
9.6
10.8
12.6
9.2
9.0
15.0
5.0
15.2
2.2
7.0
8.8
2.8
9.6
2.4
9.2
12.6
(a) What are the alias relationships in the design selected by Harry and Judy?
E = BCD, F = ACD, G = ABC, H = ABD
Defining Contrast : I = BCDE = ACDF = ABEF = ABCG = ADEG = BDFG = CEFG = ABDH
= ACEH = BCFH = DEFH = CDGH = BEGH = AFGH = ABCDEFGH
Aliases:
A = BCG = BDH = BEF = CDF = CEH = DEG = FGH
8-51
s
3.05
3.11
2.07
1.79
1.41
1.73
1.22
0.84
0.84
2.55
3.96
1.79
3.29
1.52
4.02
1.14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B = ACG = ADH = AEF = CDE = CFH = DFG = EGH
C = ABG = ADF = AEH = BDE = BFH = DGH = EFG
D = ABH = ACF = AEG = BCE = BFG = CGH = EFH
E = ABF = ACH = ADG = BCD = BGH = CFG = DFH
F = ABE = ACD = AGH = BCH = BDG = CEG = DEH
G = ABC = ADE = AFH = BDF = BEH = CDH = CEF
H = ABD = ACE = AFG = BCF = BEG = CDG = DEF
AB = CG = DH = EF
AC = BG = DF = EH
AD = BH = CF = EG
AE = BF = CH = DG
AF = BE = CD = GH
AG = BC = DE = FH
AH = BD = CE = FG
(b) Use the average ranks ( y ) as a response variable. Analyze the data and draw conclusions. You will
find it helpful to examine a normal probability plot of effect estimates.
Design Expert Output
Term
Effect
Model
Intercept
Error
A
1.125
Error
B
1.225
Error
C
1.875
Model
D
-3.975
Error
E
1.575
Model
F
-2.625
Model
G
3.775
Error
H
0.025
Error
AB
-0.075
Error
AC
1.975
Model
AD
-2.375
Error
AE
1.575
Error
AF
1.375
Error
AG
0.275
Error
AH
1.825
Lenth's ME
6.073
Lenth's SME 12.3291
SumSqr
% Contribtn
5.0625
6.0025
14.0625
63.2025
9.9225
27.5625
57.0025
0.0025
0.0225
15.6025
22.5625
9.9225
7.5625
0.3025
13.3225
2.00799
2.38083
5.57776
25.0687
3.93566
10.9324
22.6095
0.000991601
0.00892441
6.18858
8.9492
3.93566
2.99959
0.119984
5.28424
No rm a l p lot
DE S IG N-E X P E RT P l o t
T a ste A vg
99
G
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
90
80
70
50
30
20
AD
F
10
5
D
1
-3 .9 8
-2 .0 4
-0 .1 0
Effe ct
8-52
1 .8 4
3 .7 8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Taste Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
175.39
5
35.08
A
5.06
1
5.06
D
63.20
1
63.20
F
27.56
1
27.56
G
57.00
1
57.00
AD
22.56
1
22.56
Residual
76.72
10
7.67
Cor Total
252.12
15
F
Value
4.57
0.66
8.24
3.59
7.43
2.94
Prob > F
0.0198
0.4355
0.0167
0.0873
0.0214
0.1171
significant
The Model F-value of 4.57 implies the model is significant. There is only
a 1.98% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.77
8.81
31.43
196.42
Factor
Intercept
A-A
D-D
F-F
G-G
AD
Coefficient
Estimate
8.81
0.56
-1.99
-1.31
1.89
-1.19
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
0.69
0.69
0.69
0.69
0.69
0.69
0.6957
0.5435
0.2209
7.517
95% CI
Low
7.27
-0.98
-3.53
-2.86
0.34
-2.73
95% CI
High
10.36
2.11
-0.44
0.23
3.43
0.36
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Taste Avg
+8.81
+0.56
-1.99
-1.31
+1.89
-1.19
=
*A
*D
*F
*G
*A*D
Factors D, F, G and the AD interaction are important. Factor A is added to the model to preserve
hierarchy. Notice that the AD interaction is aliased with other two-factor interactions that could also be
important. So the interpretation of the two-factor interaction is somewhat uncertain. Normally, we would
add runs to the design to isolate the significant interactions, but that won’t work very well here because
each experiment requires a full growing season. In other words, it would require a very long time to add
runs to dealias the alias chain of interest.
8-53
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
T a ste A vg
D
1 5 .2
X = A: A
Y = D: D
De si g n P o i n ts
Tas te Avg
D1 D1
D2 D2
A ctu al Fa ctors
B: B = B1
C: C = C1
E: E = E1
F: F = F1
G: G = G1
H: H = H1
1 1 .9 5
8 .7
5 .4 5
2 .2
A1
A2
A
One F a ctor P lot
1 5 .2
1 5 .2
1 1 .9 5
1 1 .9 5
Tas te Avg
Tas te Avg
One F a ctor P lot
8 .7
8 .7
5 .4 5
5 .4 5
2 .2
2 .2
F1
F2
G1
F
G2
G
(c) Use the standard deviation of the ranks (or some appropriate transformation such as log s) as a
response variable. What conclusions can you draw about the effects of the eight variables on
variability in wine quality?
8-54
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
L n (T a ste S tDe v)
Wa rning! N o term s are s ele
99
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
90
80
70
50
30
20
10
5
1
-0 .3 9
-0 .2 0
-0 .0 1
0 .1 9
0 .3 8
Effe ct
There do not appear to be any significant factors.
(d) After looking at the results, Harry and Judy decide that one of the panel members (DCM) knows more
about beer than he does about wine, so they decide to delete his ranking. What affect would this have
on the results and on conclusions from parts (b) and (c)?
Design Expert Output
Term
Effect
Model
Intercept
Model
A
1.625
Error
B
2.0625
Error
C
1.5
Model
D
-4.5
Error
E
2.4375
Model
F
-2.375
Model
G
2.9375
Error
H
-0.6875
Error
AB
-0.5625
Error
AC
2.375
Model
AD
-1.5
Error
AE
0.6875
Error
AF
0.875
Error
AG
0.8125
Error
AH
2.3125
Lenth's ME 6.26579
Lenth's SME 12.7205
SumSqr
% Contribtn
10.5625
17.0156
9
81
23.7656
22.5625
34.5156
1.89063
1.26562
22.5625
9
1.89062
3.0625
2.64062
21.3906
4.02957
6.49142
3.43348
30.9013
9.06652
8.60753
13.1676
0.721268
0.482833
8.60753
3.43348
0.721268
1.16834
1.00739
8.16047
8-55
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
T a ste A vg
99
G
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
90
80
70
A
50
30
20
10
F
5
AD
D
1
-4 .5 0
-2 .6 4
-0 .7 8
1 .0 8
2 .9 4
Effe ct
Design Expert Output
Response:
Taste Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
157.64
5
31.53
A
10.56
1
10.56
D
81.00
1
81.00
F
22.56
1
22.56
G
34.52
1
34.52
AD
9.00
1
9.00
Residual
104.48
10
10.45
Cor Total
262.13
15
F
Value
3.02
1.01
7.75
2.16
3.30
0.86
Prob > F
0.0646
0.3384
0.0193
0.1724
0.0992
0.3752
not significant
The Model F-value of 3.02 implies there is a 6.46% chance that a "Model F-Value"
this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
3.23
8.50
38.03
267.48
Factor
Intercept
A-A
D-D
F-F
G-G
AD
Coefficient
Estimate
8.50
0.81
-2.25
-1.19
1.47
-0.75
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
0.81
0.81
0.81
0.81
0.81
0.81
0.6014
0.4021
-0.0204
5.778
95% CI
Low
6.70
-0.99
-4.05
-2.99
-0.33
-2.55
95% CI
High
10.30
2.61
-0.45
0.61
3.27
1.05
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Taste Avg
+8.50
+0.81
-2.25
-1.19
+1.47
-0.75
=
*A
*D
*F
*G
*A*D
The results are the same for average taste without DCM as they were with DCM.
8-56
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
T a ste S tDe v
Wa rning! N o term s are s ele
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
99
N orm al % probability
95
90
80
70
50
30
20
10
5
1
-0 .7 4
-0 .3 5
0 .0 3
0 .4 1
0 .7 9
Effe ct
The standard deviation response is much the same with or without DCM’s responses. Again, there are no
significant factors.
(e) Suppose that just before the start of the experiment, Harry and Judy discovered that the eight new
barrels they ordered from France for use in the experiment would not arrive in time, and all 16 runs
would have to be made with old barrels. If Harry and Judy just drop column C from their design,
what does this do to the alias relationships? Do they need to start over and construct a new design?
The resulting design is a 2 7IV 3 with defining relations: I = ABEF = ADEG = BDFG = ABDH = DEFH =
BEGH = AFGH.
(f) Harry and Judy know from experience that some treatment combinations are unlikely to produce good
results. For example, the run with all eight variables at the high level generally results in a poorly
rated wine. This was confirmed in the 8 March 1986 taste test. They want to set up a new design to
make the run with all eight factors at the high level. What design would you suggest?
By changing the sign of any of the design generators, a design that does not include the principal fraction
will be generated. This will give a design without an experimental run combination with all of the
variables at the high level.
8-27 In an article in Quality Engineering (“An Application of Fractional Factorial Experimental
Designs,” 1988, Vol. 1 pp. 19-23) M.B. Kilgo describes an experiment to determine the effect of CO2
pressure (A), CO2 temperature (B), peanut moisture (C), CO2 flow rate (D), and peanut particle size (E) on
the total yield of oil per batch of peanuts (y). The levels she used for these factors are as follows:
Coded
Level
-1
1
A
Pressure
(bar)
415
550
B
Temp
(C)
25
95
C
Moisture
(% by weight)
5
15
D
Flow
(liters/min)
40
60
She conducted the 16-run fractional factorial experiment shown below:
1
A
415
B
25
C
5
8-57
D
40
E
1.28
y
63
E
Particle Size
(mm)
1.28
4.05
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
550
415
550
415
550
415
550
415
550
415
550
415
550
415
550
25
95
95
25
25
95
95
25
25
95
95
25
25
95
95
5
5
5
15
15
15
15
5
5
5
5
15
15
15
15
40
40
40
40
40
40
40
60
60
60
60
60
60
60
60
4.05
4.05
1.28
4.05
1.28
1.28
4.05
4.05
1.28
1.28
4.05
1.28
4.05
4.05
1.28
21
36
99
24
66
71
54
23
74
80
33
63
21
44
96
(a) What type of design has been used? Identify the defining relation and the alias relationships.
A 2V51 , 16-run design, with I= -ABCDE.
A(-ABCDE)= -BCDE
B(-ABCDE)= -ACDE
C(-ABCDE)= -ABDE
D(-ABCDE)= -ABCE
E(-ABCDE)= -ABCD
AB(-ABCDE)= -CDE
AC(-ABCDE)= -BDE
AD(-ABCDE)= -BCE
AE(-ABCDE)= -BCD
BC(-ABCDE)= -ADE
BD(-ABCDE)= -ACE
BE(-ABCDE)= -ACD
CD(-ABCDE)= -ABE
CE(-ABCDE)= -ABD
DE(-ABCDE)= -ABC
A=
B=
C=
D=
E=
AB =
AC =
AD =
AE =
BC =
BD =
BE =
CD =
CE =
DE =
-BCDE
-ACDE
-ABDE
-ABCE
-ABCD
-CDE
-BDE
-BCE
-BCD
-ADE
-ACE
-ACD
-ABE
-ABD
-ABC
(b) Estimate the factor effects and use a normal probability plot to tentatively identify the important
factors.
Design Expert Output
Term
Effect
Model
Intercept
Error
A
7.5
Model
B
19.75
Error
C
1.25
Error
D
0
Model
E
44.5
Error
AB
5.25
Error
AC
1.25
Error
AD
-4
Error
AE
7
Error
BC
3
Error
BD
-1.75
Error
BE
0.25
Error
CD
2.25
Error
CE
-6.25
Error
DE
3.5
Lenth's ME 11.5676
Lenth's SME 23.4839
SumSqr
% Contribtn
225
1560.25
6.25
0
7921
110.25
6.25
64
196
36
12.25
0.25
20.25
156.25
49
2.17119
15.056
0.0603107
0
76.4354
1.06388
0.0603107
0.617582
1.89134
0.34739
0.118209
0.00241243
0.195407
1.50777
0.472836
8-58
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Ha lf No rm al p lo t
DE S IG N-E X P E RT P l o t
Yield
P re ssure
T e m p e ratu re
M o i stu re
Fl o w
P a rti cl e S i ze
99
H alf N orm a l % prob ability
A:
B:
C:
D:
E:
97
E
95
90
B
85
80
70
60
40
20
0
0 .0 0
1 1 .1 3
2 2 .2 5
3 3 .3 8
4 4 .5 0
|Effect|
(c) Perform an appropriate statistical analysis to test the hypothesis that the factors identified in part
above have a significant effect on the yield of peanut oil.
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
9481.25
2
4740.63
B
1560.25
1
1560.25
E
7921.00
1
7921.00
Residual
881.75
13
67.83
Cor Total
10363.00
15
F
Value
69.89
23.00
116.78
Prob > F
< 0.0001
0.0003
< 0.0001
significant
The Model F-value of 69.89 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
8.24
54.25
15.18
1335.67
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
54.25
B-Temperature
9.88
E-Particle Size
22.25
DF
1
1
1
Standard
Error
2.06
2.06
2.06
0.9149
0.9018
0.8711
18.017
95% CI
Low
49.80
5.43
17.80
95% CI
High
58.70
14.32
26.70
VIF
1.00
1.00
(d) Fit a model that could be used to predict peanut oil yield in terms of the factors that you have
identified as important.
Design Expert Output
Final Equation in Terms of Coded Factors:
Yield
+54.25
+9.88
+22.25
=
*B
*E
Final Equation in Terms of Actual Factors:
Yield
=
8-59
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-5.49175
+0.28214
+16.06498
* Temperature
* Particle Size
(e) Analyze the residuals from this experiment and comment on model adequacy.
The residual plots are satisfactory. There is a slight tendency for the variability of the residuals to
increase with the predicted value of y.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 2 .6 25
99
5 .6 2 5
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
-1 .3 75
2
2
-8 .3 75
5
1
-1 5 .37 5
-8 .3 75
-1 .3 75
5 .6 2 5
1 2 .6 25
2 2 .1 3
7 0 .3 1
Re sid ua ls vs. Te m pe ra ture
Re sid ua ls vs. P a rticle S ize
1 2 .6 25
5 .6 2 5
5 .6 2 5
2
2
-1 .3 75
-8 .3 75
-1 5 .37 5
-1 5 .37 5
37
48
60
72
83
95
2
1 .2 8
Tem pera ture
8 6 .3 8
2
-8 .3 75
25
5 4 .2 5
Predicted
1 2 .6 25
-1 .3 75
3 8 .1 9
R es idua l
R es idua ls
R es idua ls
-1 5 .37 5
1 .9 7
2 .6 7
3 .3 6
4 .0 5
Pa rticle Size
8-28 A 16-run fractional factorial experiment in 10 factors on sand-casting of engine manifolds was
conducted by engineers at the Essex Aluminum Plant of the Ford Motor Company and described in the
article “Evaporative Cast Process 3.0 Liter Intake Manifold Poor Sandfill Study,” by D. Becknell (Fourth
Symposium on Taguchi Methods, American Supplier Institute, Dearborn, MI, 1986, pp. 120-130). The
purpose was to determine which of 10 factors has an effect on the proportion of defective castings. The
design and the resulting proportion of nondefective castings p observed on each run are shown below.
8-60
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
This is a resolution III fraction with generators E=CD, F=BD, G=BC, H=AC, J=AB, and K=ABC.
Assume that the number of castings made at each run in the design is 1000.
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
F
+
+
+
+
+
+
+
+
G
+
+
+
+
+
+
+
+
H
+
+
+
+
+
+
+
+
J
+
+
+
+
+
+
+
+
K
+
+
+
+
+
+
+
+
p
0.958
1.000
0.977
0.775
0.958
0.958
0.813
0.906
0.679
0.781
1.000
0.896
0.958
0.818
0.841
0.955
arcsin
1.364
1.571
1.419
1.077
1.364
1.364
1.124
1.259
0.969
1.081
1.571
1.241
1.364
1.130
1.161
1.357
F&T’s
Modification
1.363
1.555
1.417
1.076
1.363
1.363
1.123
1.259
0.968
1.083
1.556
1.242
1.363
1.130
1.160
1.356
(a) Find the defining relation and the alias relationships in this design.
I=CDE=BDF=BCG=ACH=ABJ=ABCK=BCEF=BDEG=ADEH=ABCDEJ=ABDEK=CDFG=ABCDFH
= ADFJ=ACDFK=ABGH=ACGJ=AGK=BCHJ=BHK=CKJ
(b) Estimate the factor effects and use a normal probability plot to tentatively identify the important
factors.
Design Expert Output
Term
Model
Intercept
Error
A
Error
B
Error
C
Error
D
Error
E
Model
F
Error
G
Error
H
Error
J
Model
K
Error
AB
Error
AC
Error
AD
Error
AE
Error
AF
Error
BE
Error
DK
Lenth's ME
Lenth's SME
Effect
SumSqr
% Contribtn
-0.011875
0.006625
0.017625
-0.052125
0.036375
0.107375
-0.050875
0.028625
-0.012875
0.099625
Aliased
Aliased
0.004875
-0.034625
0.024875
-0.053125
0.015375
0.103145
0.209399
0.000564063
0.000175562
0.00124256
0.0108681
0.00529256
0.0461176
0.0103531
0.00327756
0.000663062
0.0397006
0.409171
0.127353
0.901355
7.88369
3.83923
33.4537
7.51011
2.37754
0.480986
28.7988
9.50625E-005
0.00479556
0.00247506
0.0112891
0.000945563
0.0689584
3.4787
1.79541
8.18909
0.685911
8-61
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
p
99
F
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
K: K
90
K
80
70
50
30
20
10
5
1
-0 .0 5
-0 .0 1
0 .0 3
0 .0 7
0 .1 1
Effe ct
(c)
Fit an appropriate model using the factors identified in part (b) above.
Design Expert Output
Response:
p
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.086
2
0.043
F
0.046
1
0.046
K
0.040
1
0.040
Residual
0.052
13
4.003E-003
Cor Total
0.14
15
F
Value
10.72
11.52
9.92
Prob > F
0.0018
0.0048
0.0077
significant
The Model F-value of 10.72 implies the model is significant. There is only
a 0.18% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
F-F
K-K
0.063
0.89
7.09
0.079
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
0.89
0.054
0.050
DF
1
1
1
Standard
Error
0.016
0.016
0.016
0.6225
0.5645
0.4282
7.556
95% CI
Low
0.86
0.020
0.016
95% CI
High
0.93
0.088
0.084
VIF
1.00
1.00
Final Equation in Terms of Coded Factors:
p
+0.89
+0.054
+0.050
=
*F
*K
Final Equation in Terms of Actual Factors:
p
+0.89206
+0.053688
+0.049812
=
*F
*K
(d) Plot the residuals from this model versus the predicted proportion of nondefective castings. Also
prepare a normal probability plot of the residuals. Comment on the adequacy of these plots.
8-62
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The residual versus predicted plot identifies an inequality of variances. This is likely caused by the
response variable being a proportion. A transformation could be used to correct this.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .0 8 88 1 2 5
99
2
0 .0 3 92 1 8 7
90
80
R es idua ls
N orm al % probability
95
70
50
2
-0 .0 10 3 7 5
30
20
10
-0 .0 59 9 6 8 8
5
1
-0 .1 09 5 6 3
-0 .1 09 5 6 3
-0 .0 59 9 6 8 8
-0 .0 10 3 7 5
0 .0 3 92 1 8 7
0 .0 8 88 1 2 5
0 .7 9
R es idua l
0 .8 4
0 .8 9
0 .9 4
1 .0 0
Predicted
(e) In part (d) you should have noticed an indication that the variance of the response is not constant
(considering that the response is a proportion, you should have expected this). The previous table
also shows a transformation on P, the arcsin square root, that is a widely used variance stabilizing
transformation for proportion data (refer to the discussion of variance stabilizing transformations is
Chapter 3). Repeat parts (a) through (d) above using the transformed response and comment on your
results. Specifically, are the residuals plots improved?
Design Expert Output
Term
Model
Intercept
Error
A
Error
B
Error
C
Error
D
Error
E
Model
F
Error
G
Error
H
Error
J
Model
K
Error
AD
Error
AF
Error
BE
Error
DH
Error
DK
Lenth's ME
Lenth's SME
Effect
SumSqr
% Contribtn
-0.032
0.00025
-0.02125
-0.0835
0.05875
0.19625
-0.0805
0.05625
-0.05325
0.1945
-0.032
0.05025
-0.104
-0.01125
0.0235
0.205325
0.41684
0.004096
2.5E-007
0.00180625
0.027889
0.0138062
0.154056
0.025921
0.0126562
0.0113422
0.151321
0.004096
0.0101003
0.043264
0.00050625
0.002209
0.884531
5.39875E-005
0.39006
6.02263
2.98146
33.2685
5.59764
2.73312
2.44936
32.6778
0.884531
2.18115
9.34286
0.109325
0.477034
As with the original analysis, factors F and K remain significant with a slight increase with the R2.
8-63
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
a rcsi n
99
F
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
K: K
90
K
80
70
50
30
20
10
5
1
-0 .1 0
-0 .0 3
0 .0 5
0 .1 2
0 .2 0
Effe ct
Design Expert Output
Response:
arcsin
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.31
2
0.15
F
0.15
1
0.15
K
0.15
1
0.15
Residual
0.16
13
0.012
Cor Total
0.46
15
F
Value
12.59
12.70
12.47
Prob > F
0.0009
0.0035
0.0037
significant
The Model F-value of 12.59 implies the model is significant. There is only
a 0.09% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
F-F
K-K
0.11
1.28
8.63
0.24
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
1.28
0.098
0.097
DF
1
1
1
Standard
Error
0.028
0.028
0.028
0.6595
0.6071
0.4842
8.193
95% CI
Low
1.22
0.039
0.038
95% CI
High
1.34
0.16
0.16
VIF
1.00
1.00
Final Equation in Terms of Coded Factors:
arcsin
+1.28
+0.098
+0.097
=
*F
*K
Final Equation in Terms of Actual Factors:
arcsin
+1.27600
+0.098125
+0.097250
=
*F
*K
The inequality of variance has improved; however, there remain hints of inequality in the residuals versus
predicted plot and the normal probability plot now appears to be irregular.
8-64
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .1 4 38 7 5
99
2
2
0 .0 5 93 7 5
90
80
R es idua ls
N orm al % probability
95
70
50
-0 .0 25 1 2 5
30
20
10
-0 .1 09 6 2 5
5
1
-0 .1 94 1 2 5
-0 .1 94 1 2 5
-0 .1 09 6 2 5
-0 .0 25 1 2 5
0 .0 5 93 7 5
0 .1 4 38 7 5
1 .0 8
R es idua l
1 .1 8
1 .2 8
1 .3 7
1 .4 7
Predicted
(f) There is a modification to the arcsin square root transformation, proposed by Freeman and Tukey
(“Transformations Related to the Angular and the Square Root,” Annals of Mathematical Statistics,
Vol. 21, 1950, pp. 607-611) that improves its performance in the tails. F&T’s modification is:
1ª
«arcsin
2 ¬«
np̂
arcsin
n 1
np̂ 1 º
»
n 1 ¼»
Rework parts (a) through (d) using this transformation and comment on the results. (For an interesting
discussion and analysis of this experiment, refer to “Analysis of Factorial Experiments with Defects or
Defectives as the Response,” by S. Bisgaard and H.T. Fuller, Quality Engineering, Vol. 7, 1994-5, pp.
429-443.)
Design Expert Output
Term
Model
Intercept
Error
A
Error
B
Error
C
Error
D
Error
E
Model
F
Error
G
Error
H
Error
J
Model
K
Error
AD
Error
AF
Error
BE
Error
DH
Error
DK
Lenth's ME
Lenth's SME
Effect
SumSqr
% Contribtn
-0.031125
0.000125
-0.017875
-0.082625
0.057875
0.192375
-0.080375
0.055875
-0.049625
0.190875
-0.027875
0.049625
-0.100625
-0.015375
0.023625
0.191348
0.388464
0.00387506
6.25E-008
0.00127806
0.0273076
0.0133981
0.148033
0.0258406
0.0124881
0.00985056
0.145733
0.00310806
0.00985056
0.0405016
0.000945563
0.00223256
0.871894
1.40626E-005
0.287566
6.14424
3.01458
33.3075
5.81416
2.80983
2.21639
32.7901
0.699318
2.21639
9.1129
0.212753
0.502329
As with the prior analysis, factors F and K remain significant.
8-65
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
F& T T ra n sfo rm
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
K: K
99
F
N orm al % probability
95
90
K
80
70
50
30
20
10
5
1
-0 .1 0
-0 .0 3
0 .0 5
0 .1 2
0 .1 9
Effe ct
Design Expert Output
Response: F&T Transform
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.29
2
0.15
F
0.15
1
0.15
K
0.15
1
0.15
Residual
0.15
13
0.012
Cor Total
0.44
15
F
Value
12.67
12.77
12.57
Prob > F
0.0009
0.0034
0.0036
significant
The Model F-value of 12.67 implies the model is significant. There is only
a 0.09% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
F-F
K-K
0.11
1.27
8.45
0.23
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
1.27
0.096
0.095
DF
1
1
1
Standard
Error
0.027
0.027
0.027
0.6610
0.6088
0.4864
8.221
95% CI
Low
1.22
0.038
0.037
95% CI
High
1.33
0.15
0.15
VIF
1.00
1.00
Final Equation in Terms of Coded Factors:
F&T Transform
+1.27
+0.096
+0.095
=
*F
*K
Final Equation in Terms of Actual Factors:
F&T Transform
+1.27356
+0.096188
+0.095437
=
*F
*K
The residual plots appears as they did with the arcsin square root transformation.
8-66
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .1 4 41 8 7
99
2
0 .0 6 06 8 7 5
90
80
R es idua ls
N orm al % probability
95
70
50
-0 .0 22 8 1 2 5
30
20
10
-0 .1 06 3 1 3
5
1
-0 .1 89 8 1 3
-0 .1 89 8 1 3
-0 .1 06 3 1 3
-0 .0 22 8 1 2 5 0 .0 6 06 8 7 5
0 .1 4 41 8 7
1 .0 8
1 .1 8
1 .2 7
R es idua l
1 .3 7
1 .4 7
Predicted
8-29 A 16-run fractional factorial experiment in 9 factors was conducted by Chrysler Motors
Engineering and described in the article “Sheet Molded Compound Process Improvement,” by P.I. Hsieh
and D.E. Goodwin (Fourth Symposium on Taguchi Methods, American Supplier Institute, Dearborn, MI,
1986, pp. 13-21). The purpose was to reduce the number of defects in the finish of sheet-molded grill
opening panels. The design, and the resulting number of defects, c, observed on each run, is shown
below. This is a resolution III fraction with generators E=BD, F=BCD, G=AC, H=ACD, and J=AB.
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
F
+
+
+
+
+
+
+
+
G
+
+
+
+
+
+
+
+
H
+
+
+
+
+
+
+
+
J
+
+
+
+
+
+
+
+
c
56
17
2
4
3
4
50
2
1
0
3
12
3
4
0
0
c
7.48
4.12
1.41
2.00
1.73
2.00
7.07
1.41
1.00
0.00
1.73
3.46
1.73
2.00
0.00
0.00
F&T’s
Modification
7.52
4.18
1.57
2.12
1.87
2.12
7.12
1.57
1.21
0.50
1.87
3.54
1.87
2.12
0.50
0.50
(a) Find the defining relation and the alias relationships in this design.
I=BDE=BCDF=CEF=ACG=ABCDEG=ABDEG=AEFG=ACDH=ABCEH=ABFH=ADEFH=DGH=
BEGH=BCRG=CDEFGH=ABJ=ADEJ=ACDFJ=ABCEFJ=BCGJ=CDEGJ=DEGJ=BEFGJ=BCDHJ=
CEHJ=FHJ=BDEFHJ=ABDGHJ=AEGHJ=ACEGJ=ABCDEFGHJ
(b) Estimate the factor effects and use a normal probability plot to tentatively identify the important
factors.
Design Expert Output
Term
Model
Intercept
Effect
SumSqr
% Contribtn
8-67
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Model
Model
Model
Model
Error
Model
Model
Error
Error
Model
Error
Model
Error
Model
Model
A
-9.375
B
-1.875
C
-3.625
D
-14.375
E
3.625
F
-16.625
G
-2.125
H
0.375
J
0.125
AD
11.625
AE
2.125
AF
9.875
AH
1.375
BC
11.375
BG
-12.625
Lenth's ME 13.9775
Lenth's SME 28.3764
351.562
14.0625
52.5625
826.562
52.5625
1105.56
18.0625
0.5625
0.0625
540.563
18.0625
390.063
7.5625
517.563
637.562
7.75573
0.310229
1.15957
18.2346
1.15957
24.3895
0.398472
0.0124092
0.0013788
11.9252
0.398472
8.60507
0.166834
11.4178
14.0651
No rm a l p lot
DE S IG N-E X P E RT P l o t
c
99
AD
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
90
BC
AF
80
70
50
30
20
10
D
5
BG
A
F
1
-1 6 .62
-9 .5 6
-2 .5 0
4 .5 6
1 1 .6 2
Effe ct
(c) Fit an appropriate model using the factors identified in part (b) above.
Design Expert Output
Response:
c
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
4454.13
10
445.41
A
351.56
1
351.56
B
14.06
1
14.06
C
52.56
1
52.56
D
826.56
1
826.56
F
1105.56
1
1105.56
G
18.06
1
18.06
AD
540.56
1
540.56
AF
390.06
1
390.06
BC
517.56
1
517.56
BG
637.56
1
637.56
Residual
78.81
5
15.76
Cor Total
4532.94
15
F
Value
28.26
22.30
0.89
3.33
52.44
70.14
1.15
34.29
24.75
32.84
40.45
The Model F-value of 28.26 implies the model is significant. There is only
a 0.09% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
3.97
R-Squared
0.9826
8-68
Prob > F
0.0009
0.0052
0.3883
0.1274
0.0008
0.0004
0.3333
0.0021
0.0042
0.0023
0.0014
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Mean
C.V.
PRESS
10.06
39.46
807.04
Factor
Intercept
A-A
B-B
C-C
D-D
F-F
G-G
AD
AF
BC
BG
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
10.06
-4.69
-0.94
-1.81
-7.19
-8.31
-1.06
5.81
4.94
5.69
-6.31
DF
1
1
1
1
1
1
1
1
1
1
1
Standard
Error
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.99
0.9478
0.8220
17.771
95% CI
Low
7.51
-7.24
-3.49
-4.36
-9.74
-10.86
-3.61
3.26
2.39
3.14
-8.86
95% CI
High VIF
12.61
-2.14
1.61
0.74
-4.64
-5.76
1.49
8.36
7.49
8.24
-3.76
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
c
+10.06
-4.69
-0.94
-1.81
-7.19
-8.31
-1.06
+5.81
+4.94
+5.69
-6.31
=
*A
*B
*C
*D
*F
*G
*A*D
*A*F
*B*C
*B*G
Final Equation in Terms of Actual Factors:
c
+10.06250
-4.68750
-0.93750
-1.81250
-7.18750
-8.31250
-1.06250
+5.81250
+4.93750
+5.68750
-6.31250
=
*A
*B
*C
*D
*F
*G
*A*D
*A*F
*B*C
*B*G
(d) Plot the residuals from this model versus the predicted number of defects. Also, prepare a normal
probability plot of the residuals. Comment on the adequacy of these plots.
8-69
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
3 .8 1 25
99
1 .9 0 62 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
0
20
10
-1 .9 06 2 5
5
1
-3 .8 12 5
-3 .8 12 5
-1 .9 06 2 5
0
1 .9 0 62 5
3 .8 1 25
-3 .8 1
R es idua l
1 0 .8 1
2 5 .4 4
4 0 .0 6
5 4 .6 9
Predicted
There is a significant problem with inequality of variance. This is likely caused by the response variable
being a count. A transformation may be appropriate.
(e) In part (d) you should have noticed an indication that the variance of the response is not constant
(considering that the response is a count, you should have expected this). The previous table also
shows a transformation on c, the square root, that is a widely used variance stabilizing transformation
for count data (refer to the discussion of variance stabilizing transformations in Chapter 3). Repeat
parts (a) through (d) using the transformed response and comment on your results. Specifically, are
the residual plots improved?
Design Expert Output
Term
Effect
Model
Intercept
Error
A
-0.895
Model
B
-0.3725
Error
C
-0.6575
Model
D
-2.1625
Error
E
0.4875
Model
F
-2.6075
Model
G
-0.385
Error
H
0.27
Error
J
0.06
Error
AD
1.145
Error
AE
0.555
Error
AF
0.86
Error
AH
0.0425
Error
BC
0.6275
Model
BG
-1.61
Lenth's ME 2.27978
Lenth's SME 4.62829
SumSqr
% Contribtn
3.2041
0.555025
1.72922
18.7056
0.950625
27.1962
0.5929
0.2916
0.0144
5.2441
1.2321
2.9584
0.007225
1.57502
10.3684
4.2936
0.743752
2.31722
25.0662
1.27387
36.4439
0.794506
0.390754
0.0192965
7.02727
1.65106
3.96436
0.00968175
2.11059
13.894
The analysis of the data with the square root transformation identifies only D, F, the BG interaction as being
significant. The original analysis identified factor A and several two factor interactions as being significant.
8-70
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
sq rt
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
99
N orm al % probability
95
90
80
70
50
30
20
10
BG
D
5
F
1
-2 .6 1
-1 .6 7
-0 .7 3
0 .2 1
1 .1 4
Effe ct
Design Expert Output
Response:
sqrt
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
57.42
5
11.48
B
0.56
1
0.56
D
18.71
1
18.71
F
27.20
1
27.20
G
0.59
1
0.59
BG
10.37
1
10.37
Residual
17.21
10
1.72
Cor Total
74.62
15
F
Value
6.67
0.32
10.87
15.81
0.34
6.03
Prob > F
0.0056
0.5826
0.0081
0.0026
0.5702
0.0340
significant
The Model F-value of 6.67 implies the model is significant. There is only
a 0.56% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.31
2.32
56.51
44.05
Factor
Intercept
B-B
D-D
F-F
G-G
BG
Coefficient
Estimate
2.32
-0.19
-1.08
-1.30
-0.19
-0.80
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
0.33
0.33
0.33
0.33
0.33
0.33
0.7694
0.6541
0.4097
8.422
95% CI
Low
1.59
-0.92
-1.81
-2.03
-0.92
-1.54
Final Equation in Terms of Coded Factors:
sqrt
+2.32
-0.19
-1.08
-1.30
-0.19
-0.80
=
*B
*D
*F
*G
*B*G
Final Equation in Terms of Actual Factors:
sqrt
=
8-71
95% CI
High
3.05
0.54
-0.35
-0.57
0.54
-0.074
VIF
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+2.32125
-0.18625
-1.08125
-1.30375
-0.19250
-0.80500
*B
*D
*F
*G
*B*G
The residual plots are acceptable; although, there appears to be a slight “u” shape to the residuals versus predicted
plot.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .9 7 5
99
0 .9 5 31 2 5
90
80
70
R es idua ls
N orm al % probability
95
50
-0 .0 68 7 5
30
20
10
-1 .0 90 6 2
5
1
-2 .1 12 5
-2 .1 12 5
-1 .0 90 6 2
-0 .0 68 7 5
0 .9 5 31 2 5
1 .9 7 5
-1 .2 5
R es idua l
0 .4 4
2 .1 3
3 .8 3
5 .5 2
Predicted
(f) There is a modification to the square root transformation proposed by Freeman and Tukey
(“Transformations Related to the Angular and the Square Root,” Annals of Mathematical Statistics,
Vol. 21, 1950, pp. 607-611) that improves its performance. F&T’s modification to the square root
transformation is:
1
2
> c
c 1
@
Rework parts (a) through (d) using this transformation and comment on the results. (For an interesting
discussion and analysis of this experiment, refer to “Analysis of Factorial Experiments with Defects or
Defectives as the Response,” by S. Bisgaard and H.T. Fuller, Quality Engineering, Vol. 7, 1994-5, pp.
429-443.)
Design Expert Output
Term
Model
Intercept
Error
A
Model
B
Error
C
Model
D
Error
E
Model
F
Model
G
Error
H
Error
J
Error
AD
Error
AE
Error
AF
Error
AH
Effect
SumSqr
% Contribtn
-0.86
-0.325
-0.605
-1.995
0.5025
-2.425
-0.4025
0.225
0.0275
1.1625
0.505
0.8825
0.0725
2.9584
0.4225
1.4641
15.9201
1.01002
23.5225
0.648025
0.2025
0.003025
5.40562
1.0201
3.11523
0.021025
4.38512
0.626255
2.17018
23.5977
1.49712
34.8664
0.960541
0.300158
0.00448383
8.01254
1.51205
4.61757
0.0311645
8-72
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
Model
BC
0.7525
BG
-1.54
Lenth's ME 2.14001
Lenth's SME 4.34453
2.26503
9.4864
3.35735
14.0613
As with the square root transformation, factors D, F, and the BG interaction remain significant.
No rm a l p lot
DE S IG N-E X P E RT P l o t
F& T
99
95
N orm al % probability
A: A
B: B
C: C
D: D
E: E
F: F
G: G
H: H
J: J
90
80
70
50
30
20
BG
10
D
5
F
1
-2 .4 2
-1 .5 3
-0 .6 3
0 .2 7
1 .1 6
Effe ct
Design Expert Output
Response:
F&T
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
50.00
5
10.00
B
0.42
1
0.42
D
15.92
1
15.92
F
23.52
1
23.52
G
0.65
1
0.65
BG
9.49
1
9.49
Residual
17.47
10
1.75
Cor Total
67.46
15
F
Value
5.73
0.24
9.12
13.47
0.37
5.43
Prob > F
0.0095
0.6334
0.0129
0.0043
0.5560
0.0420
significant
The Model F-value of 5.73 implies the model is significant. There is only
a 0.95% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.32
2.51
52.63
44.71
Factor
Intercept
B-B
D-D
F-F
G-G
BG
Coefficient
Estimate
2.51
-0.16
-1.00
-1.21
-0.20
-0.77
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
0.33
0.33
0.33
0.33
0.33
0.33
0.7411
0.6117
0.3373
7.862
95% CI
Low
1.78
-0.90
-1.73
-1.95
-0.94
-1.51
Final Equation in Terms of Coded Factors:
F&T
+2.51
-0.16
=
*B
8-73
95% CI
High VIF
3.25
0.57
-0.26
-0.48
0.53
-0.034
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-1.00
-1.21
-0.20
-0.77
*D
*F
*G
*B*G
Final Equation in Terms of Actual Factors:
F&T
+2.51125
-0.16250
-0.99750
-1.21250
-0.20125
-0.77000
=
*B
*D
*F
*G
*B*G
The following interaction plots appear as they did with the square root transformation; a slight “u” shape
is observed in the residuals versus predicted plot.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
2 .0 6 75
99
1 .0 4 62 5
90
80
R es idua ls
N orm al % probability
95
70
50
30
0 .0 2 5
20
10
-0 .9 96 2 5
5
1
-2 .0 17 5
-2 .0 17 5
-0 .9 96 2 5
0 .0 2 5
1 .0 4 62 5
2 .0 6 75
-0 .8 3
0 .7 6
R es idua l
2 .3 5
3 .9 4
5 .5 3
Predicted
8-30 An experiment is run in a semiconductor factory to investigate the effect of six factors on transistor
gain. The design selected is the 2 6IV 2 shown below.
Standard
Order
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Run
Order
2
8
5
9
3
14
11
10
15
13
1
6
12
4
7
16
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
8-74
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
F
+
+
+
+
+
+
+
+
Gain
1455
1511
1487
1596
1430
1481
1458
1549
1454
1517
1487
1596
1446
1473
1461
1563
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Use a normal plot of the effects to identify the significant factors.
Design Expert Output
Term
Effect
Model
Intercept
Model
A
76
Model
B
53.75
Model
C
-30.25
Error
D
3.75
Error
E
2
Error
F
1.75
Model
AB
26.75
Model
AC
-8.25
Error
AD
-0.75
Error
AE
-3.5
Error
AF
5.25
Error
BD
0.5
Error
BF
2.5
Error
ABD
3.5
Error
ABF
-2.5
Lenth's ME 9.63968
Lenth's SME 19.57
SumSqr
% Contribtn
23104
11556.2
3660.25
56.25
16
12.25
2862.25
272.25
2.25
49
110.25
1
25
49
25
55.2714
27.6459
8.75637
0.134566
0.0382766
0.0293055
6.84732
0.6513
0.00538265
0.117222
0.26375
0.00239229
0.0598072
0.117222
0.0598072
No rm a l p lot
DE S IG N-E X P E RT P l o t
Gain
A
B
C
D
E
F
99
A
95
N orm al % probability
A:
B:
C:
D:
E:
F:
90
B
AB
80
70
50
30
20
10
AC
5
C
1
-3 0 .25
-3 .6 9
2 2 .8 7
4 9 .4 4
7 6 .0 0
Effe ct
(b) Conduct appropriate statistical tests for the model identified in part (a).
Design Expert Output
Response:
Gain
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
41455.00
5
8291.00
A
23104.00
1
23104.00
B
11556.25
1
11556.25
C
3660.25
1
3660.25
AB
2862.25
1
2862.25
AC
272.25
1
272.25
Residual
346.00
10
34.60
Cor Total
41801.00
15
F
Value
239.62
667.75
334.00
105.79
82.72
7.87
The Model F-value of 239.62 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
8-75
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.0186
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Std. Dev.
Mean
C.V.
PRESS
5.88
1497.75
0.39
885.76
Factor
Intercept
A-A
B-B
C-C
AB
AC
Coefficient
Estimate
1497.75
38.00
26.87
-15.13
13.38
-4.12
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
Standard
Error
1.47
1.47
1.47
1.47
1.47
1.47
0.9917
0.9876
0.9788
44.419
95% CI
Low
1494.47
34.72
23.60
-18.40
10.10
-7.40
95% CI
High
1501.03
41.28
30.15
-11.85
16.65
-0.85
VIF
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Gain
+1497.75
+38.00
+26.87
-15.13
+13.38
-4.12
=
*A
*B
*C
*A*B
*A*C
Final Equation in Terms of Actual Factors:
Gain
+1497.75000
+38.00000
+26.87500
-15.12500
+13.37500
-4.12500
=
*A
*B
*C
*A*B
*A*C
(c) Analyze the residuals and comment on your findings.
The residual plots are acceptable. The normality and equality of variance assumptions are verified. There
does not appear to be any trends or interruptions in the residuals versus run order plot.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 0 .7 5
99
6 .1 2 5
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
2
1 .5
2
-3 .1 25
5
1
-7 .7 5
-7 .7 5
-3 .1 25
1 .5
6 .1 2 5
1 0 .7 5
1 4 3 5.2 5
R es idua l
1 4 7 5.2 5
1 5 1 5.2 5
Predicted
8-76
1 5 5 5.2 5
1 5 9 5.2 5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
6 .1 2 5
R es idua ls
6 .1 2 5
1 .5
2
1 .5
-3 .1 25
-3 .1 25
-7 .7 5
-7 .7 5
1
R es idua ls
Re sid ua ls vs. A
1 0 .7 5
4
7
10
13
16
A
Re sid ua ls vs. B
Re sid ua ls vs. C
1 0 .7 5
6 .1 2 5
6 .1 2 5
2
1 .5
2
2
1
2
1 .5
2
-3 .1 25
-7 .7 5
-7 .7 5
-1
0
1
-1
B
1
Re sid ua ls vs. E
1 0 .7 5
6 .1 2 5
6 .1 2 5
R es idua ls
1 0 .7 5
1 .5
-3 .1 25
0
C
Re sid ua ls vs. D
R es idua ls
0
R un N um ber
1 0 .7 5
-3 .1 25
2
-1
R es idua ls
R es idua ls
Re sid ua ls vs. Run
1 0 .7 5
2
2
1 .5
-3 .1 25
-7 .7 5
2
2
-7 .7 5
-1
0
1
-1
D
0
E
8-77
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. F
1 0 .7 5
R es idua ls
6 .1 2 5
1 .5
-3 .1 25
-7 .7 5
-1
0
1
F
(d) Can you find a set of operating conditions that produce gain of 1500 r 25 ?
Yes, see the graphs below.
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
Gain
1596
Gain
C
1596
X = A: A
Y = C: C
Gain
1 5 5 4.5
C- -1 .0 0 0
C+ 1 .0 0 0
A c tu al Fa c tors
B : B = 0 .0 0
D: D = 0 .0 0
E : E = 0 .0 0
F: F = 0 .0 0
1513
1 5 5 4.5
Gain
X = A: A
Y = B: B
B - -1 .0 0 0
B + 1 .0 0 0
A ctu al Fa ctors
C: C = 0 .0 0
D: D = 0 .0 0
E : E = 0 .0 0
F: F = 0 .0 0
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
B
1513
1 4 7 1.5
1 4 7 1.5
1430
1430
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
A
-0 .5 0
0 .0 0
A
8-78
0 .5 0
1 .0 0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Overla y P lot
DE S IG N-E X P E RT P l o t
1 .0 0
O ve rl a y P l o t
X = A: A
Y = B: B
A ctu al Fa ctors
C: C = -1 .0 0
D: D = 0 .0 0
E : E = 0 .0 0
F: F = 0 .0 0
B
0 .5 0
Gain: 1525
0 .0 0
Gain: 1475
-0 .5 0
-1 .0 0
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
A
8-31 Heat treating is often used to carbonize metal parts, such as gears. The thickness of the carbonized
layer is a critical output variable from this process, and it is usually measured by performing a carbon
analysis on the gear pitch (top of the gear tooth). Six factors were studied on a 2 6IV 2 design: A = furnace
temperature, B = cycle time, C = carbon concentration, D = duration of the carbonizing cycle, E = carbon
concentration of the diffuse cycle, and F = duration of the diffuse cycle. The experiment is shown below:
Standard
Order
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Run
Order
5
7
8
2
10
12
16
1
6
9
14
13
11
3
15
4
A
+
+
+
+
+
+
+
+
B
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
-
F
+
+
+
+
+
+
+
+
Pitch
74
190
133
127
115
101
54
144
121
188
135
170
126
175
126
193
(a) Estimate the factor effects and plot them on a normal probability plot. Select a tentative model.
Design Expert Output
Term
Model
Intercept
Model
A
Error
B
Model
C
Model
D
Model
E
Error
F
Error
AB
Error
AC
Error
AD
Error
AE
Effect
SumSqr
% Contribtn
50.5
-1
-13
37
34.5
4.5
-4
-2.5
4
1
10201
4
676
5476
4761
81
64
25
64
4
41.8777
0.016421
2.77515
22.4804
19.5451
0.332526
0.262737
0.102631
0.262737
0.016421
8-79
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
Model
Model
Error
Error
BD
4.5
CD
14.5
DE
-22
ABD
0.5
ABF
6
Lenth's ME 15.4235
Lenth's SME 31.3119
81
841
1936
1
144
0.332526
3.45252
7.94778
0.00410526
0.591157
Factors A, C, D, E and the two factor interactions CD and DE appear to be significant. The model can be
found in the Design Expert Output below.
No rm a l p lot
DE S IG N-E X P E RT P l o t
P i tch
A
B
C
D
E
F
99
A
95
N orm al % probability
A:
B:
C:
D:
E:
F:
90
80
70
CD
D
E
50
30
20
10
C
5
DE
1
-2 2 .00
-3 .8 8
1 4 .2 5
3 2 .3 8
5 0 .5 0
Effe ct
(b) Perform appropriate statistical tests on the model.
Design Expert Output
Response:
Pitch
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
23891.00
6
3981.83
A
10201.00
1
10201.00
C
676.00
1
676.00
D
5476.00
1
5476.00
E
4761.00
1
4761.00
CD
841.00
1
841.00
DE
1936.00
1
1936.00
Residual
468.00
9
52.00
Cor Total
24359.00
15
F
Value
76.57
196.17
13.00
105.31
91.56
16.17
37.23
Prob > F
< 0.0001
< 0.0001
0.0057
< 0.0001
< 0.0001
0.0030
0.0002
significant
The Model F-value of 76.57 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
Factor
Intercept
A-A
C-C
D-D
7.21
135.75
5.31
1479.11
Coefficient
Estimate
135.75
25.25
-6.50
18.50
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
Standard
Error
1.80
1.80
1.80
1.80
0.9808
0.9680
0.9393
28.618
95% CI
Low
131.67
21.17
-10.58
14.42
8-80
95% CI
High
139.83
29.33
-2.42
22.58
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
E-E
CD
DE
17.25
7.25
-11.00
1
1
1
1.80
1.80
1.80
13.17
3.17
-15.08
21.33
11.33
-6.92
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Pitch
+135.75
+25.25
-6.50
+18.50
+17.25
+7.25
-11.00
=
*A
*C
*D
*E
*C*D
*D*E
Final Equation in Terms of Actual Factors:
Pitch
+135.75000
+25.25000
-6.50000
+18.50000
+17.25000
+7.25000
-11.00000
=
*A
*C
*D
*E
*C*D
*D*E
(c) Analyze the residuals and comment on model adequacy.
The residual plots are acceptable. The normality and equality of variance assumptions are verified. There
does not appear to be any trends or interruptions in the residuals versus run order plot. The plots of the
residuals versus factors C and E identify reduced variation at the lower level of both variables while the
plot of residuals versus factor F identifies reduced variation at the upper level. Because C and E are
significant factors in the model, this might not affect the decision on the optimum solution for the process.
However, factor F is not included in the model and may be set at the upper level to reduce variation.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
8 .5
99
3 .1 2 5
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
10
-2 .2 5
-7 .6 25
5
1
-1 3
-1 3
-7 .6 25
-2 .2 5
3 .1 2 5
8 .5
5 0 .0 0
R es idua l
8 4 .1 3
1 1 8 .25
Predicted
8-81
1 5 2 .38
1 8 6 .50
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
3 .1 2 5
R es idua ls
3 .1 2 5
-2 .2 5
2
-2 .2 5
-7 .6 25
-7 .6 25
-1 3
-1 3
1
R es idua ls
Re sid ua ls vs. A
8 .5
4
7
10
13
16
-1
A
Re sid ua ls vs. B
Re sid ua ls vs. C
8 .5
8 .5
3 .1 2 5
3 .1 2 5
2
-2 .2 5
3
-7 .6 25
-1 3
-1 3
0
1
-1
B
1
Re sid ua ls vs. E
8 .5
3 .1 2 5
3 .1 2 5
R es idua ls
R es idua ls
Re sid ua ls vs. D
2
-2 .2 5
2
-2 .2 5
-7 .6 25
-7 .6 25
-1 3
-1 3
0
0
C
8 .5
-1
1
-2 .2 5
-7 .6 25
-1
0
R un N um ber
R es idua ls
R es idua ls
Re sid ua ls vs. Run
8 .5
1
-1
D
0
E
8-82
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. F
8 .5
R es idua ls
3 .1 2 5
3
-2 .2 5
-7 .6 25
-1 3
-1
0
1
F
(d) Interpret the results of this experiment. Assume that a layer thickness of between 140 and 160 is
desirable.
The graphs below identify a region that is acceptable between 140 and 160.
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
P i tch
193
P i tc h
E
193
X = D: D
Y = E: E
1 5 8 .25
Pitch
E - -1 .0 0 0
E + 1 .0 0 0
A c tu al Fa c tors
A : A = 0 .0 0
B : B = 0 .0 0
C: C = 0 .0 0
F: F = 0 .0 0
1 2 3 .5
1 5 8 .25
Pitch
X = C: C
Y = D: D
D- -1 .0 0 0
D+ 1 .0 0 0
A ctu al Fa ctors
A : A = 0 .0 0
B : B = 0 .0 0
E : E = 0 .0 0
F: F = 0 .0 0
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
D
1 2 3 .5
8 8 .7 5
8 8 .7 5
54
54
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
C
-0 .5 0
0 .0 0
D
8-83
0 .5 0
1 .0 0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DE S IG N-E X P E RT P l o t
1 .0 0
Overla y P lot
O ve rl a y P l o t
X = C: C
Y = D: D
Pitc h: 160
Co d e d Fa cto rs
A : A = 0 .5 0 0
B : B = 0 .0 0 0
E : E = 0 .0 0 0
F: F = 0 .0 0 0
D
0 .5 0
0 .0 0
Pitc h: 140
-0 .5 0
-1 .0 0
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
C
8-32 Five factors are studied in the irregular fractional factorial design of resolution V shown below.
Standard
Order
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Run
Order
1
10
5
4
15
19
16
7
8
3
13
11
12
20
9
22
21
6
23
18
24
17
2
14
A
+
+
+
+
+
+
+
+
+
+
+
+
-
B
+
+
+
+
+
+
+
+
+
+
+
+
C
+
+
+
+
+
+
+
+
+
+
+
+
D
+
+
+
+
+
+
+
+
+
+
+
+
E
+
+
+
+
+
+
+
+
+
+
+
+
Gain
16.33
18.43
27.07
16.95
14.58
19.12
18.96
23.56
29.15
15.74
20.73
21.52
15.58
21.03
26.78
13.39
18.63
19.01
17.96
20.49
29.31
17.62
16.03
21.42
(a) Analyze the data from this experiment. What factors influence the response y?
Design Expert Output
Term
Model
Intercept
Model
A
Model
B
Model
C
Model
D
Error
E
Model
AB
Model
AC
Error
AD
Effect
SumSqr
% Contribtn
2.9125
5.3275
-4.15917
2.1325
-0.4075
1.45428
-3.71585
-0.0282843
50.8959
170.294
103.792
27.2853
0.996338
12.6896
82.8451
0.0048
11.2736
37.7207
22.9903
6.04381
0.220693
2.8108
18.3505
0.00106322
8-84
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
Error
AE
BC
BD
BE
CD
CE
DE
ABC
ABD
ABE
ACD
ACE
ADE
BCD
BCE
BDE
CDE
Lenth's ME
Lenth's SME
0.113137
0.142887
0.133172
0.281664
-0.128458
0.0294628
0.291898
-0.130639
0.067361
Aliased
0.189835
Aliased
0.102062
0.155134
0.0898146
0.0408248
0.251073
0.455325
0.881839
0.0768
7.5E-005
0.102704
0.710704
0.0990083
0.00520833
0.511225
0.264033
0.027225
0.0170115
1.66128E-005
0.0227494
0.157424
0.0219307
0.00115367
0.113238
0.0584844
0.00603044
0.216225
0.0478947
0.0625
0.1444
0.0484
0.01
0.378225
0.013844
0.0319852
0.0107208
0.00221504
0.0837783
Factors A, B, C, D, and the AB and AC interactions appear to be significant.
No rm a l p lot
DE S IG N-E X P E RT P l o t
y
A
B
C
D
E
99
B
95
N orm al % probability
A:
B:
C:
D:
E:
A
90
D
AB
80
70
50
30
20
10
AC
5
C
1
-4 .1 6
-1 .7 9
0 .5 8
2 .9 6
5 .3 3
Effe ct
Design Expert Output
Response:
y
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
447.80
6
74.63
A
50.90
1
50.90
B
85.92
1
85.92
C
70.86
1
70.86
D
27.29
1
27.29
AB
12.69
1
12.69
AC
82.85
1
82.85
Residual
3.66
17
0.22
Cor Total
451.46
23
F
Value
346.86
236.54
399.32
329.32
126.81
58.98
385.02
The Model F-value of 346.86 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
0.46
19.97
2.32
R-Squared
Adj R-Squared
Pred R-Squared
0.9919
0.9890
0.9832
8-85
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
PRESS
7.60
Adeq Precision
Coefficient
Estimate
19.97
1.46
2.01
-1.82
1.07
0.77
-1.97
Factor
Intercept
A-A
B-B
C-C
D-D
AB
AC
60.974
Standard
Error
0.095
0.095
0.10
0.10
0.095
0.10
0.10
DF
1
1
1
1
1
1
1
95% CI
Low
19.77
1.26
1.79
-2.03
0.87
0.56
-2.18
95% CI
High VIF
20.17
1.66
2.22
-1.61
1.27
0.98
-1.76
1.00
1.13
1.13
1.00
1.12
1.12
Final Equation in Terms of Coded Factors:
y
+19.97
+1.46
+2.01
-1.82
+1.07
+0.77
-1.97
=
*A
*B
*C
*D
*A*B
*A*C
Final Equation in Terms of Actual Factors:
y
+19.97458
+1.45625
+2.00687
-1.82250
+1.06625
+0.77125
-1.97062
*A
*B
*C
*D
*A*B
*A*C
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
2 9 .3 1
y
X = A: A
Y = B: B
C
2 9 .3 1
X = A: A
Y = C: C
2 5 .3 3
C- -1 .0 0 0
C+ 1 .0 0 0
A c tu al Fa c tors
B : B = 0 .0 0
D: D = 0 .0 0
E : E = 0 .0 0
2 1 .3 5
y
B - -1 .0 0 0
B + 1 .0 0 0
A ctu al Fa ctors
C: C = 0 .0 0
D: D = 0 .0 0
E : E = 0 .0 0
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
B
2 5 .3 3
2 1 .3 5
y
y
=
1 7 .3 7
1 7 .3 7
1 3 .3 9
1 3 .3 9
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
-1 .0 0
A
-0 .5 0
0 .0 0
A
8-86
0 .5 0
1 .0 0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One F a ctor P lot
DE S IG N-E X P E RT P l o t
y
2 9 .3 1
X = D: D
A ctu al
A: A =
B: B =
C: C =
E: E =
Fa ctors
0 .0 0
0 .0 0
0 .0 0
0 .0 0
2 5 .3 3
y
2 1 .3 5
1 7 .3 7
1 3 .3 9
-1 .0 0
-0 .5 0
0 .0 0
0 .5 0
1 .0 0
D
(b) Analyze the residuals. Comment on model adequacy.
The residual plots are acceptable. The normality and equality of variance assumptions are verified. There
does not appear to be any trends or interruptions in the residuals versus run order plot.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .7 8 66 6 7
99
0 .3 7 07 2 9
90
80
70
R es idua ls
N orm al % probability
95
50
-0 .0 45 2 0 8 3
30
20
10
-0 .4 61 1 4 6
5
1
-0 .8 77 0 8 3
-0 .8 77 0 8 3
-0 .4 61 1 4 6
-0 .0 45 2 0 8 3
0 .3 7 07 2 9
0 .7 8 66 6 7
1 3 .7 9
R es idua l
1 7 .6 1
2 1 .4 3
Predicted
8-87
2 5 .2 5
2 9 .0 7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Re sid ua ls vs. Run
Re sid ua ls vs. A
0 .3 7 07 2 9
0 .3 7 07 2 9
R es idua ls
0 .7 8 66 6 7
R es idua ls
0 .7 8 66 6 7
-0 .0 45 2 0 8 3
-0 .0 45 2 0 8 3
-0 .4 61 1 4 6
-0 .4 61 1 4 6
-0 .8 77 0 8 3
-0 .8 77 0 8 3
1
4
7
10
13
16
19
22
-1
0
R un N um ber
A
Re sid ua ls vs. B
Re sid ua ls vs. C
0 .3 7 07 2 9
0 .3 7 07 2 9
R es idua ls
0 .7 8 66 6 7
R es idua ls
0 .7 8 66 6 7
1
-0 .0 45 2 0 8 3
-0 .0 45 2 0 8 3
-0 .4 61 1 4 6
-0 .4 61 1 4 6
-0 .8 77 0 8 3
-0 .8 77 0 8 3
-1
0
1
-1
B
0
1
C
Re sid ua ls vs. D
Re sid ua ls vs. E
0 .3 7 07 2 9
0 .3 7 07 2 9
R es idua ls
0 .7 8 66 6 7
R es idua ls
0 .7 8 66 6 7
-0 .0 45 2 0 8 3
-0 .0 45 2 0 8 3
-0 .4 61 1 4 6
-0 .4 61 1 4 6
-0 .8 77 0 8 3
-0 .8 77 0 8 3
-1
0
1
-1
D
0
E
8-88
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 9
Three-Level and Mixed-Level
Factorial and Fractional Factorial Design
Solutions
9-1 The effects of developer concentration (A) and developer time (B) on the density of photographic
plate film are being studied. Three strengths and three times are used, and four replicates of a 32 factorial
experiment are run. The data from this experiment follow. Analyze the data using the standard methods
for factorial experiments.
Developer Concentration
10%
12%
14%
10
0
5
4
7
7
8
Development Time (minutes)
14
2
1
3
4
4
2
6
6
8
5
7
7
10
10
10
7
8
7
Design Expert Output
Response:
Data
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
224.22
8
28.03
A
198.22
2
99.11
B
22.72
2
11.36
AB
3.28
4
0.82
Residual
71.00
27
2.63
Lack of Fit
0.000
0
Pure Error
71.00
27
2.63
Cor Total
295.22
35
F
Value
10.66
37.69
4.32
0.31
18
2
4
9
8
12
9
Prob > F
< 0.0001
< 0.0001
0.0236
0.8677
5
6
10
5
10
8
significant
The Model F-value of 10.66 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Concentration and time are significant. The interaction is not significant. By letting both A and B be
treated as numerical factors, the analysis can be performed as follows:
Design Expert Output
Response:
Data
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
221.01
5
44.20
A
192.67
1
192.67
B
22.04
1
22.04
A2
5.56
1
5.56
B2
0.68
1
0.68
AB
0.062
1
0.062
Residual
74.22
30
2.47
Lack of Fit
3.22
3
1.07
Pure Error
71.00
27
2.63
Cor Total
295.22
35
F
Value
17.87
77.88
8.91
2.25
0.28
0.025
0.41
9-1
Prob > F
< 0.0001
< 0.0001
0.0056
0.1444
0.6038
0.8748
0.7488
significant
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The Model F-value of 17.87 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
9-2
Compute the I and J components of the two-factor interaction in Problem 9-1.
11
22
32
A
AB Totals = 77, 78, 71; SS AB
B
10
28
35
17
32
39
77 2 782 712 2262
12
36
782 742 742 2262
12
36
I AB J AB 3.28
AB2 Totals = 78, 74, 74; SS AB 2
SS AB
2.39
I AB
0.89
J AB
9-3 An experiment was performed to study the effect of three different types of 32-ounce bottles (A) and
three different shelf types (B) -- smooth permanent shelves, end-aisle displays with grilled shelves, and
beverage coolers -- on the time it takes to stock ten 12-bottle cases on the shelves. Three workers (factor
C) were employed in this experiment, and two replicates of a 33 factorial design were run. The observed
time data are shown in the following table. Analyze the data and draw conclusions.
Worker
1
2
3
Bottle Type
Plastic
28-mm glass
38-mm glass
Plastic
28-mm glass
38-mm glass
Plastic
28-mm glass
38-mm glass
Permanent
3.45
4.07
4.20
4.80
4.52
4.96
4.08
4.30
4.17
Design Expert Output
Response:
Time
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
28.38
26
1.09
A
0.33
2
0.16
B
17.91
2
8.95
C
7.91
2
3.96
AB
0.11
4
0.027
AC
0.11
4
0.027
BC
1.59
4
0.40
ABC
0.43
8
0.053
Residual
2.26
27
0.084
Lack of Fit
0.000
0
Pure Error
2.26
27
0.084
Cor Total
30.64
53
Replicate I
EndAisle
4.14
4.38
4.26
5.22
5.15
5.17
3.94
4.53
4.86
Cooler
5.80
5.48
5.67
6.21
6.25
6.03
5.14
4.99
4.85
F
Value
13.06
1.95
107.10
47.33
0.33
0.32
4.76
0.64
The Model F-value of 13.06 implies the model is significant. There is only
9-2
Permanent
3.36
3.52
3.68
4.40
4.44
4.39
3.65
4.04
3.88
Prob > F
< 0.0001
0.1618
< 0.0001
< 0.0001
0.8583
0.8638
0.0049
0.7380
Replicate 2
EndAisle Cooler
4.19
5.23
4.26
4.85
4.37
5.58
4.70
5.88
4.65
6.20
4.75
6.38
4.08
4.49
4.08
4.59
4.48
4.90
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B, C, BC are significant model terms.
Factors B and C, shelf type and worker, and the BC interaction are significant. For the shortest time
regardless of worker chose the permanent shelves. This can easily be seen in the interaction plot below.
Intera ctio n Grap h
DE S IG N-E X P E RT P l o t
T im e
Sh elf Typ e
6 .5 2 16 2
X = C: W o rke r
Y = B : S h e l f T yp e
De si g n P o i n ts
5 .7 3 12 1
Tim e
B 1 P e rm a n en t
B 2 E n d A i sl e
B 3 Co o l e r
4 .9 4 08 1
A ctu al Fa ctor
A : B o ttl e T yp e = 2 8 m m g l a ss
4 .1 5 04
3 .3 6
1
2
3
Wo rker
9-4 A medical researcher is studying the effect of lidocaine on the enzyme level in the heart muscle of
beagle dogs. Three different commercial brands of lidocaine (A), three dosage levels (B), and three dogs
(C) are used in the experiment, and two replicates of a 33 factorial design are run. The observed enzyme
levels follow. Analyze the data from this experiment.
Lidocaine
Brand
1
2
3
Dosage
Strength
1
2
3
1
2
3
1
2
3
1
86
94
101
85
95
108
84
95
105
Design Expert Output
Response:
Enzyme Level
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
4490.33
26
172.71
A
31.00
2
15.50
B
4260.78
2
2130.39
C
28.00
2
14.00
AB
69.56
4
17.39
AC
3.33
4
0.83
BC
36.89
4
9.22
ABC
60.78
8
7.60
Replicate I
Dog
2
84
99
106
84
98
114
83
97
100
F
Value
16.99
1.52
209.55
1.38
1.71
0.082
0.91
0.75
9-3
3
85
98
98
86
97
109
81
93
106
1
84
95
105
80
93
110
83
92
102
Prob > F
< 0.0001
0.2359
< 0.0001
0.2695
0.1768
0.9872
0.4738
0.6502
Replicate 2
Dog
2
85
97
104
82
99
102
80
96
111
significant
3
86
90
103
84
95
100
79
93
108
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residual
Lack of Fit
Pure Error
Cor Total
274.50
0.000
274.50
4764.83
27
0
27
53
10.17
10.17
The Model F-value of 16.99 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B are significant model terms.
The dosage is significant.
9-5
Compute the I and J components of the two-factor interactions for Example 9-1.
B
134
-155
176
A
188
-348
127
44
-289
288
I totals = 74,75,16
J totals = -128,321,-28
I(AB) = 126.78
J(AB) = 6174.12
SSAB = 6300.90
C
-190
399
6
A
-58
230
-205
-211
394
-140
I totals = -100,342,-77
J totals = 25,141,-1
I(AC) = 6878.78
J(AC) = 635.12
SSAC = 7513.90
C
-93
-155
-104
B
-350
-133
-309
-16
533
74
I totals = -152,79,238
J totals =-253,287,131
I(BC) = 4273.00 J(BC) = 8581.34
SSBC = 12854.34
9-6 An experiment is run in a chemical process using a 32 factorial design. The design factors are
temperature and pressure, and the response variable is yield. The data that result from this experiment are
shown below.
Temperature, qC
80
90
100
Pressure, psig
120
64.97, 69.22
88.47, 84.23
96.57, 88.72
100
47.58, 48.77
51.86, 82.43
71.18, 92.77
140
80.92, 72.60
93.95, 88.54
76.58, 83.04
(a) Analyze the data from this experiment by conducting an analysis of variance. What conclusions can
you draw?
9-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
3187.13
8
A
1096.93
2
B
1503.56
2
AB
586.64
4
Pure Error
819.98
9
Cor Total
4007.10
17
Mean
Square
398.39
548.47
751.78
146.66
91.11
F
Value
4.37
6.02
8.25
1.61
Prob > F
0.0205
0.0219
0.0092
0.2536
significant
The Model F-value of 4.37 implies the model is significant. There is only
a 2.05% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B are significant model terms.
Temperature and pressure are significant. Their interaction is not. An alternate analysis is performed
below with the A and B treated as numeric factors:
Design Expert Output
Response:
Yield
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
3073.27
5
614.65
A
850.76
1
850.76
B
1297.92
1
1297.92
A2
246.18
1
246.18
B2
205.64
1
205.64
AB
472.78
1
472.78
Residual
933.83
12
77.82
Lack of Fit
113.86
3
37.95
Pure Error
819.98
9
91.11
Cor Total
4007.10
17
Mean
Square
7.90
10.93
16.68
3.16
2.64
6.08
0.42
F
Value
0.0017
0.0063
0.0015
0.1006
0.1300
0.0298
Prob > F
significant
0.7454 not significant
The Model F-value of 7.90 implies the model is significant. There is only
a 0.17% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case A, B, AB are significant model terms.
(b) Graphically analyze the residuals. Are there any concerns about underlying assumptions or model
adequacy?
9-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm al p lot o f re sid ua ls
Residuals vs. P redicte d
1 5 .2 8 5
99
7 .6 4 2 5
90
80
70
R e s id u a ls
N orm a l % p ro b a b ility
95
50
30
20
0
10
-7 .6 4 2 5
5
1
-1 5 .2 8 5
-1 5.2 85
-7 .6 4 25
0
7 .64 2 5
1 5.2 8 5
4 8 .1 8
5 9 .2 9
R es id u a l
Residuals vs. P ressure
9 2 .6 5
Residuals vs. Tem pera ture
1 5 .2 8 5
7 .6 4 2 5
7 .6 4 2 5
R e s id u a ls
R e s id u a ls
8 1 .5 3
Pre dicte d
1 5 .2 8 5
0
0
-7 .6 4 2 5
-7 .6 4 2 5
-1 5 .2 8 5
-1 5 .2 8 5
1
7 0 .4 1
2
3
1
Pre s s u re
2
3
Tem p e ratu re
The plot of residuals versus pressure shows a decreasing funnel shape indicating a non-constant variance.
(c) Verify that if we let the low, medium and high levels of both factors in this experiment take on the
levels -1, 0, and +1, then a least squares fit to a second order model for yield is
y
86. 81 10. 4 x1 8. 42 x 2 7.17 x12 7. 86 x22 7. 69 x1 x2
The coefficients can be found in the following table of computer output.
Design Expert Output
Final Equation in Terms of Coded Factors:
Yield =
+86.81
+8.42
+10.40
-7.84
-7.17
-7.69
*A
*B
* A2
* B2
*A*B
9-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(d) Confirm that the model in part (c) can be written in terms of the natural variables temperature (T) and
pressure (P) as
y
1335. 63 18. 56T 8. 59 P 0. 072T 2 0. 0196 P 2 0. 0384TP
The coefficients can be found in the following table of computer output.
Design Expert Output
Final Equation in Terms of Actual Factors:
Yield
-1335.62500
+8.58737
+18.55850
-0.019612
-0.071700
-0.038437
=
* Pressure
* Temperature
* Pressure2
* Temperature2
* Pressure * Temperature
(e) Construct a contour plot for yield as a function of pressure and temperature.
examination of this plot, where would you recommend running the process.
1 0 0 .0 0
Yie2 ld
2
Based on the
2
85
B: Tem perature
9 5 .0 0
90
9 0 .0 0
2
2
85
2
80
75
8 5 .0 0
70
65
55
60
50
2
2
2
8 0 .0 0
1 0 0 .0 0
1 1 0 .0 0
1 2 0 .0 0
1 3 0 .0 0
1 4 0 .0 0
A: Pres s ure
Run the process in the oval region indicated by the yield of 90.
9-7
(a) Confound a 33 design in three blocks using the ABC2 component of the three-factor interaction.
Compare your results with the design in Figure 9-7.
L = X1 + X2 + 2X3
Block 1
000
112
210
Block 2
100
212
010
9-7
Block 3
200
012
110
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
120
022
202
221
101
011
220
122
002
021
201
111
020
222
102
121
001
211
The new design is a 180q rotation around the Factor B axis.
(b) Confound a 33 design in three blocks using the AB2C component of the three-factor interaction.
Compare your results with the design in Figure 9-7.
L = X1 + 2X2 + X3
Block 1
000
022
011
212
220
201
110
102
121
Block 2
210
202
221
100
122
111
012
020
001
Block 3
112
120
101
010
002
021
200
222
211
The new design is a 180q rotation around the Factor C axis.
(c) Confound a 33 design three blocks using the ABC component of the three-factor interaction.
Compare your results with the design in Figure 9-7.
L = X1 + X2 + X3
Block 1
000
210
120
021
201
111
012
222
102
Block 2
112
022
202
100
010
220
121
001
211
Block 3
221
101
011
212
122
002
200
110
020
The new design is a 90q rotation around the Factor C axis along with switching layer 0 and layer 1 in the
C axis.
(d) After looking at the designs in parts (a), (b), and (c) and Figure 9-7, what conclusions can you draw?
All four designs are relatively the same. The only differences are rotations and swapping of layers.
9-8
Confound a 34 design in three blocks using the AB2CD component of the four-factor interaction.
9-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
L = X1 + 2X2 + X3 + X4
0000
0211
2102
1100
1222
0021
0110
2212
2001
0101
2221
2120
Block 1
2200
0122
1011
0220
2111
2022
0202
1121
0012
1210
1112
1002
1201
2010
1020
1021
0200
1000
1110
0022
1122
1202
0111
1211
0001
2002
2112
Block 2
0120
2121
2201
0212
2210
2020
1012
0010
2011
1101
0102
2100
1220
0221
2222
2012
1221
2021
2101
1010
2110
2220
1102
2202
1022
0020
0100
Block 3
1111
0112
0222
1200
0201
0011
2000
1001
0002
2121
1120
0121
2211
1212
0210
9-9 Consider the data from the first replicate of Problem 9-3. Assuming that all 27 observations could
not be run on the same day, set up a design for conducting the experiment over three days with AB2C
confounded with blocks. Analyze the data.
000
110
011
102
201
212
121
022
220
Totals->
Block 1
=
3.45
=
4.38
=
5.22
=
4.30
=
4.96
=
4.86
=
6.25
=
5.14
=
5.67
=
44.23
Design Expert Output
Response:
Time
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Block
0.23
2
Model
13.17
18
A
0.048
2
B
8.92
2
C
1.57
2
AB
1.31
4
AC
0.87
4
BC
0.45
4
Residual
1.03
6
Cor Total
14.43
26
100
210
111
202
001
012
221
122
020
Block 2
=
=
=
=
=
=
=
=
=
Mean
Square
0.11
0.73
0.024
4.46
0.78
0.33
0.22
0.11
0.17
The Model F-value of 4.27 implies the model is significant. There is only
a 4.04% chance that a "Model F-Value" this large could occur due to noise.
Values of "Prob > F" less than 0.0500 indicate model terms are significant.
In this case B are significant model terms.
9-9
4.07
4.26
4.14
4.17
4.80
3.94
4.99
6.03
5.80
43.21
200
010
211
002
101
112
021
222
120
F
Value
4.27
0.14
26.02
4.57
1.91
1.27
0.66
Block 3
=
4.20
=
4.14
=
5.17
=
4.08
=
4.52
=
4.53
=
6.21
=
4.85
=
5.48
43.18
Prob > F
0.0404
0.8723
0.0011
0.0622
0.2284
0.3774
0.6410
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
9-10 Outline the analysis of variance table for the 34 design in nine blocks. Is this a practical design?
Source
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC (AB2C,ABC2,AB2C2)
ABD (ABD,AB2D,ABD2)
ACD (ACD,ACD2,AC2D2)
BCD (BCD,BC2D,BCD2)
ABCD
Blocks (ABC,AB2C2,AC2D,BC2D2)
Total
DF
2
2
2
2
4
4
4
4
4
4
6
6
6
6
16
8
80
Any experiment with 81 runs is large. Instead of having three full levels of each factor, if two levels of
each factor could be used, then the overall design would have 16 runs plus some center points. This twolevel design could now probably be run in 2 or 4 blocks, with center points in each block. Additional
curvature effects could be determined by augmenting the experiment with the axial points of a central
composite design and additional enter points. The overall design would be less than 81 runs.
9-11 Consider the data in Problem 9-3. If ABC is confounded in replicate I and ABC2 is confounded in
replicate II, perform the analysis of variance.
L1 = X1 + X2 + X3
Block 1
000 = 3.45
111 = 5.15
222 = 4.85
120 = 5.48
102 = 4.30
210 = 4.26
201 = 4.96
012 = 3.94
021 = 6.21
Block 2
001 = 4.80
112 = 4.53
220 = 5.67
121 = 6.25
100 = 4.07
211 = 5.17
202 = 4.17
010 = 4.14
022 = 5.14
L2 = X1 + X2 + 2X2
002
110
221
122
101
212
200
011
020
Block 3
= 4.08
= 4.38
= 6.03
= 4.99
= 4.52
= 4.86
= 4.20
= 5.22
= 5.80
000
101
011
221
202
022
120
210
112
Block 1
= 3.36
= 4.44
= 4.70
= 6.38
= 3.88
= 4.49
= 4.85
= 4.37
= 4.08
100
201
111
021
002
122
220
010
212
Block 2
= 3.52
= 4.39
= 4.65
= 5.88
= 3.65
= 4.59
= 5.58
= 4.19
= 4.48
200
001
211
121
102
222
020
110
012
Block 3
= 3.68
= 4.40
= 4.75
= 6.20
= 4.04
= 4.90
= 5.23
= 4.26
= 4.08
The sums of squares for A, B, C, AB, AC, and BC are calculated as usual. The only sums of squares
presenting difficulties with calculations are the four components of the ABC interaction (ABC, ABC2,
AB2C, and AB2C2). ABC is computed using replicate I and ABC2 is computed using replicate II. AB2C
and AB2C2 are computed using data from both replicates.
We will show how to calculate AB2C and AB2C2 from both replicates. Form a two-way table of A x B at
each level of C. Find the I(AB) and J(AB) totals for each third of the A x B table.
C
B
0
0
6.81
A
1
7.59
2
7.88
9-10
I
26.70
J
27.55
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0
1
2
1
2
0
1
2
0
1
2
8.33
11.03
9.20
9.92
12.09
7.73
8.02
9.63
8.64
10.33
8.96
9.80
12.45
8.34
8.61
9.58
8.63
11.25
9.35
9.92
12.41
8.05
9.34
9.75
27.25
26.54
31.41
30.97
31.72
26.09
27.31
25.65
27.17
25.77
31.25
31.29
31.57
26.29
26.11
26.65
The I and J components for each third of the above table are used to form a new table of diagonal totals.
C
0
1
2
2.670
31.41
26.09
I(AB)
27.25
30.97
27.31
26.54
31.72
25.65
27.55
31.25
26.29
J(AB)
27.17
31.29
26.11
25.77
31.57
26.65
I Totals:
85.06,85.26,83.32
I Totals:
85.99,85.03,83.12
J Totals:
85.73,83.60,84.31
J Totals:
83.35,85.06,85.23
(85.06) 2 (85.26) 2 (83.32) 2 (253.64) 2
01265
.
18
54
(85.73) 2 (83.60) 2 (84.31) 2 (253.64) 2
and, AB2C = J[C x I(AB)]=
01307
.
18
54
Now, AB2C2 = I[C x I(AB)] =
If it were necessary, we could find ABC2 as ABC2= I[C x J(AB)] and ABC as J[C x J(AB)]. However,
these components must be computed using the data from the appropriate replicate.
The analysis of variance table:
Source
Replicates
Blocks within Replicates
A
B
C
AB
AC
BC
ABC (rep I)
ABC2 (rep II)
AB2C
AB2C2
Error
Total
SS
1.06696
0.2038
0.4104
17.7514
7.6631
0.1161
0.1093
1.6790
0.0452
0.1020
0.1307
0.1265
0.8998
30.3069
DF
1
4
2
2
2
4
4
4
2
2
2
2
22
53
MS
F0
0.2052
8.8757
3.8316
0.0290
0.0273
0.4198
0.0226
0.0510
0.0754
0.0633
0.0409
5.02
217.0
93.68
<1
<1
10.26
<1
1.25
1.60
1.55
9-12 Consider the data from replicate I in Problem 9-3. Suppose that only a one-third fraction of this
design with I=ABC is run. Construct the design, determine the alias structure, and analyze the design.
9-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The design is 000, 012, 021, 102, 201, 111, 120, 210, 222.
The alias structure is: A = BC = AB2C2
B = AC = AB2C
C = AB = ABC2
AB2 = AC2 = BC2
A
B
0
1
2
0
1
2
0
1
2
0
1
2
C
1
0
3.45
2
5.48
4.26
6.21
5.15
4.96
3.94
4.30
Source
A
B
C
AB2
Total
4.85
SS
2.25
0.30
2.81
0.30
5.66
DF
2
2
2
2
8
9-13 From examining Figure 9-9, what type of design would remain if after completing the first 9 runs,
one of the three factors could be dropped?
A full 32 factorial design results.
9-14 Construct a 34IV1 design with I=ABCD. Write out the alias structure for this design.
The 27 runs for this design are as follows:
0000
0012
0021
0102
0111
0120
0201
0210
0222
A = AB2C2D2 = BCD
AB = ABC2D2 = CD
BC = AB2C2D = AD
AD2 = AB2C2 = BCD2
B = AB2CD = ACD
AB2 = AC2D2 = BC2D2
BC2 = AB2D = AC2D
1002
1011
1020
1101
1110
1122
1200
1212
1221
2001
2010
2022
2100
2112
2121
2202
2211
2220
C = ABC2D = ABD
AC = AB2CD2 = BD
BD2 = AB2C = ACD2
9-12
D = ABCD2 = ABC
AC2 = AB2D2 = BC2D
CD2 = ABC2 = ABCD2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
9-15 Verify that the design in Problem 9-14 is a resolution IV design.
The design in Problem 9-14 is a Resolution IV design because no main effect is aliased with a component
of a two-factor interaction, but some two-factor interaction components are aliased with each other.
9-16 Construct a 35-2 design with I=ABC and I=CDE. Write out the alias structure for this design. What
is the resolution of this design?
The complete defining relation for this design is : I = ABC = CDE = ABC2DE = ABD2E2
This is a resolution III design. The defining contrasts are L1 = X1 + X2 + X3 and L2 = X3 + X4 + X5.
00000
00012
00022
01200
02100
10202
20101
11102
21200
11120
22111
21021
02111
01222
12012
02120
10210
12021
20111
22222
01210
12000
20120
11111
22201
21012
10222
To find the alias of any effect, multiply the effect by I and I2. For example, the alias of A is:
A = AB2C2 = ACDE = AB2CDE = AB2DE = BC = AC2D2E2 = BC2DE = BD2E2
9-17 Construct a 39-6 design, and verify that is a resolution III design.
Use the generators I = AC2D2, I = AB2C2E, I = BC2F2, I = AB2CG, I = ABCH2, and I = ABJ2
000000000
022110012
011220021
221111221
210221200
202001212
112222112
101002121
120112100
021201102
212012020
100120211
122200220
010011111
201122002
002121120
111010202
220202011
102211001
001212210
211100110
020022222
222020101
200210122
121021010
110101022
012102201
To find the alias of any effect, multiply the effect by I and I2. For example, the alias of C is:
C = C(BC2F2) = BF2, At least one main effect is aliased with a component of a two-factor interaction.
9-18 Construct a 4 x 23 design confounded in two blocks of 16 observations each. Outline the analysis of
variance for this design.
Design is a 4 x 23, with ABC at two levels, and Z at 4 levels. Represent Z with two pseudo-factors D and
E as follows:
9-13
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Z
Z1
Z2
Z3
Z4
PseudoD
0
1
0
1
Factors
E
0 = (1)
0=d
1=e
1 = de
The 4 x 23 is now a 25 in the factors A, B, C, D and E. Confound ABCDE with blocks. We have given
both the letter notation and the digital notation for the treatment combinations.
(1)
ab
ac
bc
abcd
abce
cd
ce
de
abde
bcde
be
ad
ae
acde
bd
Block 1
= 000
= 1100
= 1010
= 0110
= 1111
= 1112
= 0011
= 0012
= 0003
= 1103
= 0113
= 0102
= 1001
= 1002
= 1013
= 0101
a
b
c
abc
bcd
bce
acd
ace
ade
bde
abcde
abd
abe
d
e
cde
Source
A
B
C
Z (D+E+DE)
AB
AC
AZ (AD+AE+ADE)
BC
BZ (BD+BE+BDE)
CZ (CD+CE+CDE)
ABC
ABZ (ABD+ABE+ABDE)
ACZ (ACD+ACE+ACDE)
BCZ (BCD+BCE+BCDE)
ABCZ (ABCD+ABCE)
Blocks (or ABCDE)
Total
Block 2
= 1000
= 0100
= 0010
= 1110
= 0111
= 0112
= 1011
= 1012
= 1003
= 0103
= 1113
= 1101
= 1102
= 0001
= 0002
= 0013
DF
1
1
1
3
1
1
3
1
3
3
1
3
3
3
2
1
31
9-19 Outline the analysis of variance table for a 2232 factorial design. Discuss how this design may be
confounded in blocks.
9-14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Suppose we have n replicates of a 2232 factorial design. A and B are at 2 levels, and C and D are at 3
levels.
Source
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD
Error
Total
DF
1
1
2
2
1
2
2
2
2
4
2
2
4
4
4
36(n-1)
36n-1
Components for Confounding
A
B
C
D
AB
AC
AD
BD
CD,CD2
ABC
ABD
ACD,ACD2
BCD,BCD2
ABCD,ABCD2
Confounding in this series of designs is discussed extensively by Margolin (1967). The possibilities for a
single replicate of the 2232 design are:
2 blocks of 18 observations
3 blocks of 12 observations
4 blocks of 9 observations
6 blocks of 6 observations
9 blocks of 4 observations
For example, one component of the four-factor interaction, say ABCD2, could be selected to confound the
design in 3 blocks for 12 observations each, while to confound the design in 2 blocks of 18 observations 3
each we would select the AB interaction. Cochran and Cox (1957) and Anderson and McLean (1974)
discuss confounding in these designs.
9-20 Starting with a 16-run 24 design, show how two three-level factors can be incorporated in this
experiment. How many two-level factors can be included if we want some information on two-factor
interactions?
Use column A and B for one three-level factor and columns C and D for the other. Use the AC and BD
columns for the two, two-level factors. The design will be of resolution V.
9-21 Starting with a 16-run 24 design, show how one three-level factor and three two-level factors can be
accommodated and still allow the estimation of two-factor interactions.
Use columns A and B for the three-level factor, and columns C and D and ABCD for the three two-level
factors. This design will be of resolution V.
9-22 In Problem 9-26, you met Harry and Judy Peterson-Nedry, two friends of the author who have a
winery and vineyard in Newberg, Oregon. That problem described the application of two-level fractional
factorial designs to their 1985 Pinor Noir product. In 1987, they wanted to conduct another Pinot Noir
experiment. The variables for this experiment were
Variable
Clone of Pinot Noir
Berry Size
Levels
Wadenswil, Pommard
Small, Large
9-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Fermentation temperature
Whole Berry
Maceration Time
Yeast Type
Oak Type
80F, 85F, 90/80F, 90F
None, 10%
10 days, 21 days
Assmanhau, Champagne
Troncais, Allier
Harry and Judy decided to use a 16-run two-level fractional factorial design, treating the four levels of
fermentation temperature as two two-level variables. As in Problem 8-26, they used the rankings from a
taste-test panel as the response variable. The design and the resulting average ranks are shown below:
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Clone
+
+
+
+
+
+
+
+
Berry
Size
+
+
+
+
+
+
+
+
Ferm.
Temp.
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Whole
Berry
+
+
+
+
+
+
+
+
Macer.
Time
+
+
+
+
+
+
+
+
Yeast
Type
+
+
+
+
+
+
+
+
Oak
Type
+
+
+
+
+
+
+
+
Average
Rank
4
10
6
9
11
1
15
5
12
2
16
3
8
14
7
13
(a) Describe the aliasing in this design.
The design is a resolution IV design such that the main effects are aliased with three factor interactions.
Design Expert Output
Term
Intercept
A
B
C
D
E
F
G
H
AB
AC
AD
AE
AF
AG
AH
Aliases
ABCG ABDH ABEF ACDF ACEH ADEG AFGH BCDE BCFH BDFG BEGH CDGH CEFG DEFH
BCG BDH BEF CDF CEH DEG FGH ABCDE
ACG ADH AEF CDE CFH DFG EGH
ABG ADF AEH BDE BFH DGH EFG
ABH ACF AEG BCE BFG CGH EFH
ABF ACH ADG BCD BGH CFG DFH
ABE ACD AGH BCH BDG CEG DEH
ABC ADE AFH BDF BEH CDH CEF
ABD ACE AFG BCF BEG CDG DEF
CG DH EF ACDE ACFH ADFG AEGH BCDF BCEH BDEG BFGH
BG DF EH ABDE ABFH ADGH AEFG BCDH BCEF CDEG CFGH
BH CF EG ABCE ABFG ACGH AEFH BCDG BDEF CDEH DFGH
BF CH DG ABCD ABGH ACFG ADFH BCEG BDEH CDEF EFGH
BE CD GH ABCH ABDG ACEG ADEH BCFG BDFH CEFH DEFG
BC DE FH ABDF ABEH ACDH ACEF BDGH BEFG CDFG CEGH
BD CE FG ABCF ABEG ACDG ADEF BCGH BEFH CDFH DEGH
(b) Analyze the data and draw conclusions.
All of the main effects except Yeast and Oak are significant. The Macer Time is the most significant.
None of the interactions were significant.
9-16
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot
DE S IG N-E X P E RT P l o t
Ra n k
99
F
95
N orm al % probability
A : Cl o n e
B : B e rry S i ze
C: Fe rm T e m p 1
D: Fe rm T e m p 2
E : W h o l e B erry
F: M ace r T i m e
G : Y ea st
H: O ak
90
D
C
B
80
70
50
30
20
10
E
5
A
1
-2 .7 5
-0 .0 6
2 .6 3
5 .3 1
8 .0 0
Effe ct
Design Expert Output
Response:
Rank
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
328.75
6
54.79
A
30.25
1
30.25
B
9.00
1
9.00
C
9.00
1
9.00
D
12.25
1
12.25
E
12.25
1
12.25
F
256.00
1
256.00
Residual
11.25
9
1.25
Cor Total
340.00
15
F
Value
43.83
24.20
7.20
7.20
9.80
9.80
204.80
Prob > F
< 0.0001
0.0008
0.0251
0.0251
0.0121
0.0121
< 0.0001
significant
The Model F-value of 43.83 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.12
8.50
13.15
35.56
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Factor
Estimate
Intercept
8.50
A-Clone
-1.38
B-Berry Size
0.75
C-Ferm Temp 1
0.75
D-Ferm Temp 2
0.88
E-Whole Berry
-0.87
F-Macer Time
4.00
DF
1
1
1
1
1
1
1
Standard
Error
0.28
0.28
0.28
0.28
0.28
0.28
0.28
0.9669
0.9449
0.8954
19.270
95% CI
Low
7.87
-2.01
0.12
0.12
0.24
-1.51
3.37
Final Equation in Terms of Coded Factors:
Rank
+8.50
-1.38
+0.75
+0.75
+0.88
-0.87
+4.00
=
*A
*B
*C
*D
*E
*F
9-17
95% CI
High
9.13
-0.74
1.38
1.38
1.51
-0.24
4.63
VIF
1.00
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) What comparisons can you make between this experiment and the 1985 Pinot Noir experiment from
Problem 8-26?
The experiment from Problem 8-26 indicates that yeast, barrel, whole cluster and the clone x yeast
interactions were significant. This experiment indicates that maceration time, whole berry, clone and
fermentation temperature are significant.
9-23 An article by W.D. Baten in the 1956 volume of Industrial Quality Control described an
experiment to study the effect of three factors on the lengths of steel bars. Each bar was subjected to one
of two heat treatment processes, and was cut on one of four machines at one of three times during the day
(8 am, 11 am, or 3 pm). The coded length data are shown below
(a) Analyze the data from this experiment assuming that the four observations in each cell are replicates.
The Machine effect (C) is significant, the Heat Treat Process (B) is also significant, while the Time of Day
(A) is not significant. None of the interactions are significant.
Machine
Time of Heat Treat
Day
Process
1
8am
2
1
11 am
2
1
3 pm
2
1
6
1
4
0
6
1
3
1
5
9
6
3
Design Expert Output
Response:
Length
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
590.33
23
25.67
A
26.27
2
13.14
B
42.67
1
42.67
C
393.42
3
131.14
AB
23.77
2
11.89
AC
42.15
6
7.02
BC
13.08
3
4.36
ABC
48.98
6
8.16
Pure Error
447.50
72
6.22
Cor Total
1037.83
95
2
9
3
6
1
3
-1
1
-2
4
6
0
7
7
5
6
3
8
4
6
1
10
6
8
10
3
9
5
5
4
7
8
4
3
11
4
7
0
F
Value
4.13
2.11
6.86
21.10
1.91
1.13
0.70
1.31
The Model F-value of 4.13 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
2.49
3.96
R-Squared
Adj R-Squared
0.5688
0.4311
9-18
1
0
-1
0
3
1
2
-1
-1
6
0
4
Prob > F
< 0.0001
0.1283
0.0107
< 0.0001
0.1552
0.3537
0.5541
0.2623
4
2
4
0
1
2
0
0
1
2
1
-2
-4
6
7
4
5
7
11
9
6
10
4
4
7
significant
6
3
5
4
9
6
4
3
5
8
3
0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
C.V.
PRESS
62.98
795.56
Term
Intercept
A[1]
A[2]
B-Process
C[1]
C[2]
C[3]
A[1]B
A[2]B
A[1]C[1]
A[2]C[1]
A[1]C[2]
A[2]C[2]
A[1]C[3]
A[2]C[3]
BC[1]
BC[2]
BC[3]
A[1]BC[1]
A[2]BC[1]
A[1]BC[2]
A[2]BC[2]
A[1]BC[3]
A[2]BC[3]
Pred R-Squared
Adeq Precision
Coefficient
Estimate
3.96
0.010
-0.65
-0.67
-0.54
1.92
-3.08
0.010
0.60
0.32
-1.27
-0.39
-0.10
0.24
0.77
-0.25
-0.46
0.46
-0.094
-0.44
0.11
-1.10
-0.43
0.60
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Standard
Error
0.25
0.36
0.36
0.25
0.44
0.44
0.44
0.36
0.36
0.62
0.62
0.62
0.62
0.62
0.62
0.44
0.44
0.44
0.62
0.62
0.62
0.62
0.62
0.62
0.2334
7.020
95% CI
Low
3.45
-0.71
-1.36
-1.17
-1.42
1.04
-3.96
-0.71
-0.11
-0.92
-2.51
-1.63
-1.35
-1.00
-0.47
-1.13
-1.34
-0.42
-1.34
-1.68
-1.13
-2.35
-1.67
-0.64
95% CI
High
4.47
0.73
0.071
-0.16
0.34
2.80
-2.20
0.73
1.32
1.57
-0.028
0.86
1.14
1.48
2.01
0.63
0.42
1.34
1.15
0.80
1.36
0.14
0.82
1.85
VIF
1.00
Final Equation in Terms of Coded Factors:
Length
+3.96
+0.010
-0.65
-0.67
-0.54
+1.92
-3.08
+0.010
+0.60
+0.32
-1.27
-0.39
-0.10
+0.24
+0.77
-0.25
-0.46
+0.46
-0.094
-0.44
+0.11
-1.10
-0.43
+0.60
=
* A[1]
* A[2]
*B
* C[1]
* C[2]
* C[3]
* A[1]B
* A[2]B
* A[1]C[1]
* A[2]C[1]
* A[1]C[2]
* A[2]C[2]
* A[1]C[3]
* A[2]C[3]
* BC[1]
* BC[2]
* BC[3]
* A[1]BC[1]
* A[2]BC[1]
* A[1]BC[2]
* A[2]BC[2]
* A[1]BC[3]
* A[2]BC[3]
(b) Analyze the residuals from this experiment. Is there any indication that there is an outlier in one
cell? If you find an outlier, remove it and repeat the analysis from part (a). What are your
conclusions?
Standard Order 84, Time of Day at 3:00pm, Heat Treat #2, Machine #2, and length of 0, appears to be an
outlier.
9-19
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
4 .5
99
1 .8 1 25
90
3
80
R es idua ls
N orm al % probability
95
70
50
30
2
2
-0 .8 75
2
3
2
3
3
2
20
2
10
2
-3 .5 62 5
5
1
-6 .2 5
-6 .2 5
-3 .5 62 5
-0 .8 75
1 .8 1 25
4 .5
-0 .5 0
R es idua l
1 .6 9
3 .8 8
6 .0 6
8 .2 5
Predicted
The following analysis was performed with the outlier described above removed. As with the original
analysis, Machine is significant and Heat Treat Process is also significant, but now Time of Day, factor A,
is also significant with an F-value of 3.05 (the P-value is just above 0.05).
Design Expert Output
Response:
Length
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
626.58
23
27.24
A
34.03
2
17.02
B
33.06
1
33.06
C
411.89
3
137.30
AB
16.41
2
8.20
AC
50.19
6
8.37
BC
8.38
3
2.79
ABC
67.00
6
11.17
Pure Error
395.42
71
5.57
Cor Total
1022.00
94
F
Value
4.89
3.06
5.94
24.65
1.47
1.50
0.50
2.01
Prob > F
< 0.0001
0.0533
0.0173
< 0.0001
0.2361
0.1900
0.6824
0.0762
The Model F-value of 4.89 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.36
4.00
59.00
705.17
Term
Intercept
A[1]
A[2]
B-Process
C[1]
C[2]
C[3]
A[1]B
A[2]B
A[1]C[1]
A[2]C[1]
A[1]C[2]
A[2]C[2]
Coefficient
Estimate
4.05
-0.076
-0.73
-0.58
-0.63
2.18
-3.17
-0.076
0.52
0.41
-1.18
-0.65
-0.36
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
Standard
Error
0.24
0.34
0.34
0.24
0.42
0.43
0.42
0.34
0.34
0.59
0.59
0.60
0.60
0.6131
0.4878
0.3100
7.447
95% CI
Low
3.z56
-0.76
-1.41
-1.06
-1.46
1.33
-4.00
-0.76
-0.16
-0.77
-2.36
-1.83
-1.55
9-20
95% CI
High
VIF
4.53
0.61
-0.051
-0.096
1.00
0.21
3.03
-2.34
0.61
1.20
1.59
-6.278E-003
0.54
0.82
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A[1]C[3]
A[2]C[3]
BC[1]
BC[2]
BC[3]
A[1]BC[1]
A[2]BC[1]
A[1]BC[2]
A[2]BC[2]
A[1]BC[3]
A[2]BC[3]
0.33
0.86
-0.34
-0.20
0.37
-6.944E-003
-0.35
-0.15
-1.36
-0.34
0.69
1
1
1
1
1
1
1
1
1
1
1
0.59
0.59
0.42
0.43
0.42
0.59
0.59
0.60
0.60
0.59
0.59
-0.85
-0.32
-1.17
-1.05
-0.46
-1.18
-1.53
-1.33
-2.55
-1.52
-0.49
1.50
2.04
0.50
0.65
1.21
1.17
0.83
1.04
-0.18
0.84
1.87
Final Equation in Terms of Coded Factors:
Length
+4.05
-0.076
-0.73
-0.58
-0.63
+2.18
-3.17
-0.076
+0.52
+0.41
-1.18
-0.65
-0.36
+0.33
+0.86
-0.34
-0.20
+0.37
-6.944E-003
-0.35
-0.15
-1.36
-0.34
+0.69
=
* A[1]
* A[2]
*B
* C[1]
* C[2]
* C[3]
* A[1]B
* A[2]B
* A[1]C[1]
* A[2]C[1]
* A[1]C[2]
* A[2]C[2]
* A[1]C[3]
* A[2]C[3]
* BC[1]
* BC[2]
* BC[3]
* A[1]BC[1]
* A[2]BC[1]
* A[1]BC[2]
* A[2]BC[2]
* A[1]BC[3]
* A[2]BC[3]
The following residual plots are acceptable. Both the normality and constant variance assumptions are
satisfied
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
4 .5
99
2 .3 7 5
90
80
70
R es idua ls
N orm al % probability
95
50
30
20
3
0 .2 5
2
2
2
3
2
3
3
2
10
-1 .8 75
5
2
2
1
-4
-4
-1 .8 75
0 .2 5
2 .3 7 5
4 .5
-0 .5 0
R es idua l
1 .7 1
3 .9 2
Predicted
9-21
6 .1 2
8 .3 3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) Suppose that the observations in the cells are the lengths (coded) of bars processed together in heat
treating and then cut sequentially (that is, in order) on the three machines. Analyze the data to
determine the effects of the three factors on mean length.
The analysis with all effects and interactions included:
Design Expert Output
Response:
Length
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
147.58
23
6.42
A
6.57
2
3.28
B
10.67
1
10.67
C
98.35
3
32.78
AB
5.94
2
2.97
AC
10.54
6
1.76
BC
3.27
3
1.09
ABC
12.24
6
2.04
Pure Error
0.000
0
Cor Total
147.58
23
F
Value
Prob > F
The by removing the three factor interaction from the model and applying it to the error, the analysis
identifies factor C as being significant and B as being mildly significant.
Design Expert Output
Response:
Length
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
135.34
17
7.96
A
6.57
2
3.28
B
10.67
1
10.67
C
98.35
3
32.78
AB
5.94
2
2.97
AC
10.54
6
1.76
BC
3.27
3
1.09
Residual
12.24
6
2.04
Cor Total
147.58
23
F
Value
3.90
1.61
5.23
16.06
1.46
0.86
0.53
Prob > F
0.0502
0.2757
0.0623
0.0028
0.3052
0.5700
0.6756
not significant
When removing the remaining insignificant factors from the model, C, Machine, is the most significant
factor while B, Heat Treat Process, is moderately significant. Factor A, Time of Day, is not significant.
Design Expert Output
Response:
Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
109.02
4
27.26
B
10.67
1
10.67
C
98.35
3
32.78
Residual
38.56
19
2.03
Cor Total
147.58
23
F
Value
13.43
5.26
16.15
The Model F-value of 13.43 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.42
3.96
35.99
61.53
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.7387
0.6837
0.5831
9.740
9-22
Prob > F
< 0.0001
0.0335
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Coefficient
Estimate
3.96
-0.67
-0.54
1.92
-3.08
Term
Intercept
B-Process
C[1]
C[2]
C[3]
DF
1
1
1
1
1
Standard
Error
0.29
0.29
0.50
0.50
0.50
95% CI
Low
3.35
-1.28
-1.60
0.86
-4.14
95% CI
High
4.57
-0.058
0.51
2.97
-2.03
VIF
1.00
Final Equation in Terms of Coded Factors:
Avg
+3.96
-0.67
-0.54
+1.92
-3.08
=
*B
* C[1]
* C[2]
* C[3]
The following residual plots are acceptable. Both the normality and uniformity of variance assumptions
are verified.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
1 .9 1 66 7
99
0 .9 3 75
90
80
70
R es idua ls
N orm al % probability
95
50
-0 .0 41 6 6 6 7
30
20
10
-1 .0 20 8 3
5
1
-2
-2
-1 .0 20 8 3
-0 .0 41 6 6 6 7
0 .9 3 75
1 .9 1 66 7
0 .2 1
R es idua l
1 .7 9
3 .3 8
4 .9 6
6 .5 4
Predicted
(d) Calculate the log variance of the observations in each cell. Analyze the average length and the log
variance of the length for each of the 12 bars cut at each machine/heat treatment process combination.
What conclusions can you draw?
Factor B, Heat Treat Process, has an affect on the log variance of the observations while Factor A, Time of
Day, and Factor C, Machine, are not significant at the 5 percent level. However, A is significant at the 10
percent level, so Tome of Day has some effect on the variance.
Design Expert Output
Response:
Log(Var)
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2.79
11
0.25
A
0.58
2
0.29
B
0.50
1
0.50
C
0.59
3
0.20
AB
0.49
2
0.24
BC
0.64
3
0.21
Residual
1.22
12
0.10
F
Value
2.51
2.86
4.89
1.95
2.40
2.10
9-23
Prob > F
0.0648
0.0966
0.0471
0.1757
0.1324
0.1538
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Cor Total
4.01
23
The Model F-value of 2.51 implies there is a 6.48% chance that a "Model F-Value"
this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.32
0.65
49.02
4.86
Term
Intercept
A[1]
A[2]
B-Process
C[1]
C[2]
C[3]
A[1]B
A[2]B
BC[1]
BC[2]
BC[3]
Coefficient
Estimate
0.65
-0.054
-0.16
0.14
0.22
0.066
-0.19
-0.20
0.14
-0.18
-0.15
0.14
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
1
1
1
1
1
1
Standard
Error
0.065
0.092
0.092
0.065
0.11
0.11
0.11
0.092
0.092
0.11
0.11
0.11
0.6967
0.4186
-0.2133
5.676
95% CI
95% CI
Low
High
VIF
0.51
0.79
-0.25
0.15
-0.36
0.043
2.181E-003 0.29
1.00
-0.025
0.47
-0.18
0.31
-0.44
0.052
-0.40
3.237E-003
-0.065
0.34
-0.42
0.068
-0.39
0.098
-0.10
0.39
Final Equation in Terms of Coded Factors:
Log(Var)
+0.65
-0.054
-0.16
+0.14
+0.22
+0.066
-0.19
-0.20
+0.14
-0.18
-0.15
+0.14
=
* A[1]
* A[2]
*B
* C[1]
* C[2]
* C[3]
* A[1]B
* A[2]B
* BC[1]
* BC[2]
* BC[3]
The following residual plots are acceptable. Both the normality and uniformity of variance assumptions
are verified.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .3 4 04 6 1
99
0 .1 5 16 0 2
90
80
70
R es idua ls
N orm al % probability
95
50
-0 .0 37 2 5 5 6
30
20
10
-0 .2 26 1 1 4
5
1
-0 .4 14 9 7 2
-0 .4 14 9 7 2
-0 .2 26 1 1 4
-0 .0 37 2 5 5 6
0 .1 5 16 0 2
0 .3 4 04 6 1
-0 .1 2
R es idua l
0 .2 0
0 .5 2
Predicted
9-24
0 .8 4
1 .1 6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(e) Suppose the time at which a bar is cut really cannot be controlled during routine production. Analyze
the average length and the log variance of the length for each of the 12 bars cut at each machine/heat
treatment process combination. What conclusions can you draw?
The analysis of the average length is as follows:
Design Expert Output
Response:
Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
37.43
7
5.35
A
3.56
1
3.56
B
32.78
3
10.93
AB
1.09
3
0.36
Pure Error
0.000
0
Cor Total
37.43
7
F
Value
Prob > F
Because the Means Square of the AB interaction is much less than the main effects, it is removed from the
model and placed in the error. The average length is strongly affected by Factor B, Machine, and
moderately affected by Factor A, Heat Treat Process. The interaction effect was small and removed from
the model.
Design Expert Output
Response:
Avg
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
36.34
4
9.09
A
3.56
1
3.56
B
32.78
3
10.93
Residual
1.09
3
0.36
Cor Total
37.43
7
F
Value
25.00
9.78
30.07
Prob > F
0.0122
0.0522
0.0097
significant
The Model F-value of 25.00 implies the model is significant. There is only
a 1.22% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.60
3.96
15.23
7.75
Term
Intercept
A-Process
B[1]
B[2]
B[3]
Coefficient
Estimate
3.96
-0.67
-0.54
1.92
-3.08
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
Standard
Error
0.21
0.21
0.37
0.37
0.37
0.9709
0.9320
0.7929
13.289
95% CI
Low
3.28
-1.34
-1.72
0.74
-4.26
95% CI
High
4.64
0.012
0.63
3.09
-1.91
VIF
1.00
Final Equation in Terms of Coded Factors:
Avg
+3.96
-0.67
-0.54
+1.92
-3.08
=
*A
* B[1]
* B[2]
* B[3]
The following residual plots are acceptable. Both the normality and uniformity of variance assumptions
are verified.
9-25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .4 5 83 3 3
99
0 .2 2 91 6 7
90
80
R es idua ls
N orm al % probability
95
70
50
30
0
20
10
-0 .2 29 1 6 7
5
1
-0 .4 58 3 3 3
-0 .4 58 3 3 3
-0 .2 29 1 6 7
0
0 .2 2 91 6 7
0 .4 5 83 3 3
0 .2 1
1 .7 9
R es idua l
3 .3 8
4 .9 6
6 .5 4
Predicted
The Log(Var) is analyzed below:
Design Expert Output
Response:
Log(Var)
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.32
7
0.046
A
0.091
1
0.091
B
0.13
3
0.044
AB
0.098
3
0.033
Pure Error 0.000
0
Cor Total 0.32
7
F
Value
Prob > F
Because the AB interaction has the smallest Mean Square, it was removed from the model and placed in
the error. From the following analysis of variance, neither Heat Treat Process, Machine, nor the
interaction affect the log variance of the length.
Design Expert Output
Response: Log(Var)
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.22
4
0.056
A
0.091
1
0.091
B
0.13
3
0.044
Residual
0.098
3
0.033
Cor Total
0.32
7
F
Value
1.71
2.80
1.34
Prob > F
0.3441
0.1926
0.4071
not significant
The "Model F-value" of 1.71 implies the model is not significant relative to the noise. There is a
34.41 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.18
0.79
22.90
0.69
Coefficient
Term
Estimate DF
Intercept 0.79
1
A-Process 0.11
1
R-Squared 0.6949
Adj R-Squared
Pred R-Squared
Adeq Precision
Error
0.064
0.064
0.2882
-1.1693
3.991
Standard
Low
0.59
-0.096
95% CI
High
0.99
0.31
9-26
95% CI
VIF
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B[1]
B[2]
B[3]
0.15
0.030
-0.20
1
1
1
0.11
0.11
0.11
-0.20
-0.32
-0.55
0.51
0.38
0.15
Final Equation in Terms of Coded Factors:
Log(Var) =
+0.79
+0.11
*A
+0.15
* B[1]
+0.030
* B[2]
-0.20
* B[3]
The following residual plots are acceptable. Both the normality and uniformity of variance assumptions
are verified.
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0 .1 6 09 5 8
99
0 .0 8 04 7 9 1
90
80
70
R es idua ls
N orm al % probability
95
50
30
0
20
10
-0 .0 80 4 7 9 1
5
1
-0 .1 60 9 5 8
-0 .1 60 9 5 8
-0 .0 80 4 7 9 1
0
0 .0 8 04 7 9 1
0 .1 6 09 5 8
0 .4 8
R es idua l
0 .6 2
0 .7 6
Predicted
9-27
0 .9 1
1 .0 5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 10
Fitting Regression Models
Solutions
10-1 The tensile strength of a paper product is related to the amount of hardwood in the pulp. Ten
samples are produced in the pilot plant, and the data obtained are shown in the following table.
Strength
160
171
175
182
184
Percent Hardwood
10
15
15
20
20
Strength
181
188
193
195
200
Percent Hardwood
20
25
25
28
30
(a) Fit a linear regression model relating strength to percent hardwood.
Minitab Output
Regression Analysis: Strength versus Hardwood
The regression equation is
Strength = 144 + 1.88 Hardwood
Predictor
Constant
Hardwood
Coef
143.824
1.8786
S = 2.203
PRESS = 66.2665
SE Coef
2.522
0.1165
T
57.04
16.12
R-Sq = 97.0%
R-Sq(pred) = 94.91%
P
0.000
0.000
R-Sq(adj) = 96.6%
Regression Plot
Strength = 143.824 + 1.87864 Hardwood
S = 2.20320
R-Sq = 97.0 %
R-Sq(adj) = 96.6 %
200
Strength
190
180
170
160
10
20
30
Hardwood
(b) Test the model in part (a) for significance of regression.
Minitab Output
10-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Analysis of Variance
Source
Regression
Residual Error
Lack of Fit
Pure Error
Total
DF
1
8
4
4
9
SS
1262.1
38.8
13.7
25.2
1300.9
MS
1262.1
4.9
3.4
6.3
F
260.00
P
0.000
0.54
0.716
3 rows with no replicates
No evidence of lack of fit (P > 0.1)
(c) Find a 95 percent confidence interval on the parameter E1.
The 95 percent confidence interval is:
Eˆ1 tD
2, n p se
Eˆ1 d E 1 d Eˆ1 tD
2, n p se
Eˆ1
1.8786 2.3060 0.1165 d E 1 d 1.8786 2.3060 0.1165
1.6900 d E 1 d 2.1473
10-2 A plant distills liquid air to produce oxygen , nitrogen, and argon. The percentage of impurity in
the oxygen is thought to be linearly related to the amount of impurities in the air as measured by the
“pollution count” in part per million (ppm). A sample of plant operating data is shown below.
Purity(%)
Pollution count (ppm)
93.3
1.10
92.0
1.45
92.4
1.36
91.7
1.59
94.0
1.08
94.6
0.75
93.6
1.20
93.1
0.99
93.2
0.83
(a) Fit a linear regression model to the data.
Minitab Output
Regression Analysis: Purity versus Pollution
The regression equation is
Purity = 96.5 - 2.90 Pollution
Predictor
Constant
Pollutio
Coef
96.4546
-2.9010
S = 0.4277
PRESS = 3.43946
SE Coef
0.4282
0.3056
T
225.24
-9.49
R-Sq = 87.4%
R-Sq(pred) = 81.77%
P
0.000
0.000
R-Sq(adj) = 86.4%
10-2
92.9
1.22
92.2
1.47
91.3
1.81
90.1
2.03
91.6
1.75
91.9
1.68
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Regression Plot
Purity = 96.4546 - 2.90096 Pollution
S = 0.427745
R-Sq = 87.4 %
R-Sq(adj) = 86.4 %
95
Purity
94
93
92
91
90
1.0
1.5
2.0
Pollution
(b) Test for significance of regression.
Minitab Output
Analysis of Variance
Source
Regression
Residual Error
Total
DF
1
13
14
SS
16.491
2.379
18.869
MS
16.491
0.183
F
90.13
P
0.000
No replicates. Cannot do pure error test.
No evidence of lack of fit (P > 0.1)
(c) Find a 95 percent confidence interval on E1.
The 95 percent confidence interval is:
Eˆ1 tD
2, n p se
Eˆ1 d E 1 d Eˆ1 tD
2, n p se
Eˆ1
-2.9010 2.1604 0.3056 d E 1 d -2.9010 2.1604 0.3056
3.5612 d E1 d 2.2408
10-3 Plot the residuals from Problem 10-1 and comment on model adequacy.
10-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot of the Residuals
(response is Strength)
Normal Score
1
0
-1
-3
-2
-1
0
1
2
3
Residual
Residuals Versus the Fitted Values
(response is Strength)
3
2
Residual
1
0
-1
-2
-3
160
170
180
Fitted Value
10-4
190
200
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Order of the Data
(response is Strength)
3
2
Residual
1
0
-1
-2
-3
1
2
3
4
5
6
7
8
9
10
Observation Order
There is nothing unusual about the residual plots. The underlying assumptions have been met.
10-4 Plot the residuals from Problem 10-2 and comment on model adequacy.
Normal Probability Plot of the Residuals
(response is Purity)
2
Normal Score
1
0
-1
-2
-1.0
-0.5
0.0
Residual
10-5
0.5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Fitted Values
(response is Purity)
Residual
0.5
0.0
-0.5
-1.0
90.5
91.5
92.5
93.5
94.5
Fitted Value
Residuals Versus the Order of the Data
(response is Purity)
Residual
0.5
0.0
-0.5
-1.0
2
4
6
8
10
12
14
Observation Order
There is nothing unusual about the residual plots. The underlying assumptions have been met.
10-5 Using the results of Problem 10-1, test the regression model for lack of fit.
Minitab Output
Analysis of Variance
Source
Regression
Residual Error
DF
1
8
SS
1262.1
38.8
MS
1262.1
4.9
10-6
F
260.00
P
0.000
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Lack of Fit
Pure Error
Total
4
4
9
13.7
25.2
1300.9
3.4
6.3
0.54
0.716
3 rows with no replicates
No evidence of lack of fit (P > 0.1)
10-6 A study was performed on wear of a bearing y and its relationship to x1 = oil viscosity and x2 =
load. The following data were obtained.
y
193
230
172
91
113
125
x1
1.6
15.5
22.0
43.0
33.0
40.0
x2
851
816
1058
1201
1357
1115
(a) Fit a multiple linear regression model to the data.
Minitab Output
Regression Analysis: Wear versus Viscosity, Load
The regression equation is
Wear = 351 - 1.27 Viscosity - 0.154 Load
Predictor
Constant
Viscosit
Load
Coef
350.99
-1.272
-0.15390
S = 25.50
PRESS = 12696.7
SE Coef
74.75
1.169
0.08953
T
4.70
-1.09
-1.72
R-Sq = 86.2%
R-Sq(pred) = 10.03%
P
0.018
0.356
0.184
VIF
2.6
2.6
R-Sq(adj) = 77.0%
(b) Test for significance of regression.
Minitab Output
Analysis of Variance
Source
Regression
Residual Error
Total
DF
2
3
5
SS
12161.6
1950.4
14112.0
MS
6080.8
650.1
F
9.35
P
0.051
No replicates. Cannot do pure error test.
Source
Viscosit
Load
DF
1
1
Seq SS
10240.4
1921.2
* Not enough data for lack of fit test
(c) Compute t statistics for each model parameter. What conclusions can you draw?
Minitab Output
Regression Analysis: Wear versus Viscosity, Load
The regression equation is
Wear = 351 - 1.27 Viscosity - 0.154 Load
Predictor
Constant
Coef
350.99
SE Coef
74.75
T
4.70
10-7
P
0.018
VIF
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Viscosit
Load
-1.272
-0.15390
S = 25.50
PRESS = 12696.7
1.169
0.08953
-1.09
-1.72
R-Sq = 86.2%
R-Sq(pred) = 10.03%
0.356
0.184
2.6
2.6
R-Sq(adj) = 77.0%
The t-tests are shown in part (a). Notice that overall regression is significant (part(b)), but neither
variable has a large t-statistic. This could be an indicator that the regressors are nearly linearly dependent.
10-7 The brake horsepower developed by an automobile engine on a dynomometer is thought to be a
function of the engine speed in revolutions per minute (rpm), the road octane number of the fuel, and the
engine compression. An experiment is run in the laboratory and the data that follow are collected.
Brake Horsepower
225
212
229
222
219
278
246
237
233
224
223
230
rpm
2000
1800
2400
1900
1600
2500
3000
3200
2800
3400
1800
2500
Road Octane Number
90
94
88
91
86
96
94
90
88
86
90
89
Compression
100
95
110
96
100
110
98
100
105
97
100
104
(a) Fit a multiple linear regression model to the data.
Minitab Output
Regression Analysis: Horsepower versus rpm, Octane, Compression
The regression equation is
Horsepower = - 266 + 0.0107 rpm + 3.13 Octane + 1.87 Compression
Predictor
Constant
rpm
Octane
Compress
Coef
-266.03
0.010713
3.1348
1.8674
S = 8.812
PRESS = 2494.05
SE Coef
92.67
0.004483
0.8444
0.5345
T
-2.87
2.39
3.71
3.49
R-Sq = 80.7%
R-Sq(pred) = 22.33%
P
0.021
0.044
0.006
0.008
VIF
1.0
1.0
1.0
R-Sq(adj) = 73.4%
(b) Test for significance of regression. What conclusions can you draw?
Minitab Output
Analysis of Variance
Source
DF
SS
MS
Regression
3
2589.73
863.24
Residual Error
8
621.27
77.66
Total
11
3211.00
r No replicates. Cannot do pure error test.
Source
rpm
Octane
Compress
DF
1
1
1
F
11.12
P
0.003
Seq SS
509.35
1132.56
947.83
Lack of fit test
Possible interactions with variable Octane (P-Value = 0.028)
Possible lack of fit at outer X-values
(P-Value = 0.000)
Overall lack of fit test is significant at P = 0.000
10-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(c) Based on t tests, do you need all three regressor variables in the model?
Yes, all of the regressor variables are important.
10-8 Analyze the residuals from the regression model in Problem 10-7. Comment on model adequacy.
Normal Probability Plot of the Residuals
(response is Horsepow)
2
Normal Score
1
0
-1
-2
-10
0
10
Residual
Residuals Versus the Fitted Values
(response is Horsepow)
Residual
10
0
-10
210
220
230
240
Fitted Value
10-9
250
260
270
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Order of the Data
(response is Horsepow)
Residual
10
0
-10
2
4
6
8
10
12
Observation Order
The normal probability plot is satisfactory, as is the plot of residuals versus run order (assuming that
observation order is run order). The plot of residuals versus predicted response exhibits a slight “bow”
shape. This could be an indication of lack of fit. It might be useful to consider adding some ineraction
terms to the model.
10-9 The yield of a chemical process is related to the concentration of the reactant and the operating
temperature. An experiment has been conducted with the following results.
Yield
81
89
83
91
79
87
84
90
Concentration
1.00
1.00
2.00
2.00
1.00
1.00
2.00
2.00
Temperature
150
180
150
180
150
180
150
180
(a) Suppose we wish to fit a main effects model to this data. Set up the X’X matrix using the data
exactly as it appears in the table.
ª1
«1
«
«1
1
1
1
1
1
1
1 º«
ª 1
«1.00 1.00 2.00 2.00 1.00 1.00 2.00 2.00» «1
«
» «1
«¬ 150 180 150 180 150 180 150 180 »¼ «
«1
«1
«
«¬1
1.00 150º
1.00 180»»
2.00 150»
»
2.00 180»
1.00 150»
»
1.00 180»
2.00 150»
»
2.00 180»¼
(b) Is the matrix you obtained in part (a) diagonal? Discuss your response.
10-10
12
1320 º
ª8
«12
20
1980 »»
«
«¬1320 1980 219600»¼
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The X’X is not diagonal, even though an orthogonal design has been used. The reason is that we have
worked with the natural factor levels, not the orthogonally coded variables.
(c) Suppose we write our model in terms of the “usual” coded variables
x1
Conc 1.5
, x2
0.5
Temp 165
15
Set up the X’X matrix for the model in terms of these coded variables. Is this matrix diagonal? Discuss
your response.
ª1
«
«1
«1
ª 1 1 1 1 1 1 1 1º «
« 1 1 1 1 1 1 1 1» «1
«
» «1
«¬ 1 1 1 1 1 1 1 1»¼ «
«1
«1
«
¬«1
1
1
1
1
1
1
1
1
1º
»
1»
1»
»
1»
1»
»
1»
1»
»
1¼»
ª8 0 0º
« 0 8 0»
«
»
«¬ 0 0 8»¼
The X’X matrix is diagonal because we have used the orthogonally coded variables.
(d) Define a new set of coded variables
x1
Conc 1.0
, x2
1.0
Temp 150
30
Set up the X’X matrix for the model in terms of this set of coded variables. Is this matrix diagonal?
Discuss your response.
ª1 1 1 1 1 1 1 1º
«0 0 1 1 0 0 1 1»
«
»
«¬0 1 0 1 0 1 0 1»¼
ª1
«1
«
«1
«
«1
«1
«
«1
«1
«
«¬1
0 0º
0 1 »»
1 0»
»
1 1»
0 0»
»
0 1»
1 0»
»
1 1 »¼
ª8 4 4º
« 4 4 2»
«
»
«¬4 2 4»¼
The X’X is not diagonal, even though an orthogonal design has been used. The reason is that we have not
used orthogonally coded variables.
(e) Summarize what you have learned from this problem about coding the variables.
If the design is orthogonal, use the orthogonal coding. This not only makes the analysis somewhat easier,
but it also results in model coefficients that are easier to interpret because they are both dimensionless and
uncorrelated.
10-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
10-10 Consider the 24 factorial experiment in Example 6-2. Suppose that the last observation in missing.
Reanalyze the data and draw conclusions. How do these conclusions compare with those from the
original example?
The regression analysis with the one data point missing indicates that the same effects are important.
Minitab Output
Regression Analysis: Rate versus A, B, C, D, AB, AC, AD, BC, BD, CD
The regression equation is
Rate = 69.8 + 10.5 A + 1.25 B + 4.63 C + 7.00 D - 0.25 AB - 9.38 AC + 8.00 AD
+ 0.87 BC - 0.50 BD - 0.87 CD
Predictor
Constant
A
B
C
D
AB
AC
AD
BC
BD
CD
Coef
69.750
10.500
1.250
4.625
7.000
-0.250
-9.375
8.000
0.875
-0.500
-0.875
S = 5.477
PRESS = 1750.00
SE Coef
1.500
1.500
1.500
1.500
1.500
1.500
1.500
1.500
1.500
1.500
1.500
T
46.50
7.00
0.83
3.08
4.67
-0.17
-6.25
5.33
0.58
-0.33
-0.58
R-Sq = 97.6%
R-Sq(pred) = 65.09%
P
0.000
0.002
0.452
0.037
0.010
0.876
0.003
0.006
0.591
0.756
0.591
VIF
1.1
1.1
1.1
1.1
1.1
1.1
1.1
1.1
1.1
1.1
R-Sq(adj) = 91.6%
Analysis of Variance
Source
Regression
Residual Error
Total
DF
10
4
14
SS
4893.33
120.00
5013.33
MS
489.33
30.00
No replicates. Cannot do pure error test.
Source
A
B
C
D
AB
AC
AD
BC
BD
CD
DF
1
1
1
1
1
1
1
1
1
1
Seq SS
1414.40
4.01
262.86
758.88
0.06
1500.63
924.50
16.07
1.72
10.21
10-12
F
16.31
P
0.008
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot of the Residuals
(response is Rate)
2
Normal Score
1
0
-1
-2
-5
0
5
Residual
Residuals Versus the Fitted Values
(response is Rate)
Residual
5
0
-5
40
50
60
70
80
90
100
Fitted Value
Residuals Versus the Order of the Data
(response is Rate)
Residual
5
0
-5
2
4
6
8
10
Observation Order
10-13
12
14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The residual plots are acceptable; therefore, the underlying assumptions are valid.
10-11 Consider the 24 factorial experiment in Example 6-2. Suppose that the last two observations are
missing. Reanalyze the data and draw conclusions. How do these conclusions compare with those from
the original example?
The regression analysis with the one data point missing indicates that the same effects are important.
Minitab Output
Regression Analysis: Rate versus A, B, C, D, AB, AC, AD, BC, BD, CD
The regression equation is
Rate = 71.4 + 10.1 A + 2.87 B + 6.25 C + 8.62 D - 0.66 AB - 9.78 AC + 7.59 AD
+ 2.50 BC + 1.12 BD + 0.75 CD
Predictor
Constant
A
B
C
D
AB
AC
AD
BC
BD
CD
Coef
71.375
10.094
2.875
6.250
8.625
-0.656
-9.781
7.594
2.500
1.125
0.750
S = 4.732
PRESS = 1493.06
SE Coef
1.673
1.323
1.673
1.673
1.673
1.323
1.323
1.323
1.673
1.673
1.673
T
42.66
7.63
1.72
3.74
5.15
-0.50
-7.39
5.74
1.49
0.67
0.45
R-Sq = 98.7%
R-Sq(pred) = 70.20%
P
0.000
0.005
0.184
0.033
0.014
0.654
0.005
0.010
0.232
0.549
0.684
VIF
1.1
1.7
1.7
1.7
1.1
1.1
1.1
1.7
1.7
1.7
R-Sq(adj) = 94.2%
Analysis of Variance
Source
Regression
Residual Error
Total
DF
10
3
13
SS
4943.17
67.19
5010.36
MS
494.32
22.40
No replicates. Cannot do pure error test.
Source
A
B
C
D
AB
AC
AD
BC
BD
CD
DF
1
1
1
1
1
1
1
1
1
1
Seq SS
1543.50
1.52
177.63
726.01
1.17
1702.53
738.11
42.19
6.00
4.50
10-14
F
22.07
P
0.014
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot of the Residuals
(response is Rate)
2
Normal Score
1
0
-1
-2
-3
-2
-1
0
1
2
3
Residual
Residuals Versus the Fitted Values
(response is Rate)
3
2
Residual
1
0
-1
-2
-3
40
50
60
70
80
90
100
Fitted Value
Residuals Versus the Order of the Data
(response is Rate)
3
2
Residual
1
0
-1
-2
-3
2
4
6
8
Observation Order
10-15
10
12
14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The residual plots are acceptable; therefore, the underlying assumptions are valid.
10-12 Given the following data, fit the second-order polynomial regression model
y
E 0 E 1 x1 E 2 x 2 E11 x12 E 22 x 22 E 12 x1 x 2 H
y
26
24
175
160
163
55
62
100
26
30
70
71
x1
1.0
1.0
1.5
1.5
1.5
0.5
1.5
0.5
1.0
0.5
1.0
0.5
x2
1.0
1.0
4.0
4.0
4.0
2.0
2.0
3.0
1.5
1.5
2.5
2.5
After you have fit the model, test for significance of regression.
Minitab Output
Regression Analysis: y versus x1, x2, x1^2, x2^2, x1x2
The regression equation is
y = 24.4 - 38.0 x1 + 0.7 x2 + 35.0 x1^2 + 11.1 x2^2 - 9.99 x1x2
Predictor
Constant
x1
x2
x1^2
x2^2
x1x2
Coef
24.41
-38.03
0.72
34.98
11.066
-9.986
SE Coef
26.59
40.45
11.69
21.56
3.158
8.742
T
0.92
-0.94
0.06
1.62
3.50
-1.14
S = 6.042
R-Sq = 99.4%
PRESS = 1327.71
R-Sq(pred) = 96.24%
r Analysis of Variance
Source
Regression
Residual Error
Lack of Fit
Pure Error
Total
DF
5
6
3
3
11
SS
35092.6
219.1
91.1
128.0
35311.7
P
0.394
0.383
0.953
0.156
0.013
0.297
DF
1
1
1
1
1
89.6
52.1
103.9
104.7
105.1
R-Sq(adj) = 98.9%
MS
7018.5
36.5
30.4
42.7
7 rows with no replicates
Source
x1
x2
x1^2
x2^2
x1x2
VIF
Seq SS
11552.0
22950.3
21.9
520.8
47.6
10-16
F
192.23
P
0.000
0.71
0.607
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Normal Probability Plot of the Residuals
(response is y)
2
Normal Score
1
0
-1
-2
-5
0
5
10
Residual
Residuals Versus the Fitted Values
(response is y)
10
Residual
5
0
-5
20
70
120
170
Fitted Value
Residuals Versus the Order of the Data
(response is y)
10
Residual
5
0
-5
2
4
6
8
Observation Order
10-17
10
12
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
10-13
(a) Consider the quadratic regression model from Problem 10-12. Compute t statistics for each model
parameter and comment on the conclusions that follow from the quantities.
Minitab Output
Predictor
Constant
x1
x2
x1^2
x2^2
x1x2
Coef
24.41
-38.03
0.72
34.98
11.066
-9.986
SE Coef
26.59
40.45
11.69
21.56
3.158
8.742
T
0.92
-0.94
0.06
1.62
3.50
-1.14
P
0.394
0.383
0.953
0.156
0.013
0.297
VIF
89.6
52.1
103.9
104.7
105.1
x 22 is the only model parameter that is statistically significant with a t-value of 3.50. A logical model
might also include x2 to preserve model hierarchy.
(b) Use the extra sum of squares method to evaluate the value of the quadratic terms, x12 , x 22 and x1 x 2 to
the model.
The extra sum of squares due to E2 is
SS R E 2 E 0, E1
SS R E 0 , E1 , E 2 SS R E 0 , E1
SS R E1 , E 2 E 0 SS R E1 E 0
SS R E1 , E 2 E 0 sum of squares of regression for the model in Problem 10-12 = 35092.6
SS R E1 E 0 =34502.3
SS R E 2 E 0, E1
F0
Since F0.05 ,3,6
35092.6 34502.3
SS R E 2 E 0, E1 3
MS E
590.3 3
36.511
590.3
5.3892
4.76 , then the addition of the quadratic terms to the model is significant. The P-values
2
indicate that it’s probably the term x2 that is responsible for this.
10-14 Relationship between analysis of variance and regression. Any analysis of variance model can be
expressed in terms of the general linear model y = XE
E + H , where the X matrix consists of zeros and
ones. Show that the single-factor model y ij P W i H ij , i=1,2,3, j=1,2,3,4 can be written in general
linear model form. Then
ˆ
(a) Write the normal equations ( XcX)E
the model in Chapter 3.
ˆ
The normal equations are ( XcX)E
Xcy and compare them with the normal equations found for
Xcy
10-18
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ª12
«4
«
«4
«
¬4
4
4
0
0
4
0
4
0
4
0
0
4
º ª Pˆ º
» «Wˆ »
»« 1 »
» «Wˆ2 »
»« »
¼ ¬Wˆ3 ¼
ª y.. º
« »
« y1. »
« y 2. »
« »
¬ y 3. ¼
which are in agreement with the results of Chapter 3.
(b) Find the rank of XcX . Can ( XcX)
1
be obtained?
XcX is a 4 x 4 matrix of rank 3, because the last three columns add to the first column. Thus (X’X)-1
does not exist.
(c) Suppose the first normal equation is deleted and the restriction
¦i 1 nWˆi
3
0 is added. Can the
resulting system of equations be solved? If so, find the solution. Find the regression sum of squares
Eˆ cXcy , and compare it to the treatment sum of squares in the single-factor model.
Imposing
¦i 1 nWˆi
3
0 yields the normal equations
ª0
«4
«
«4
«
¬4
4
4
0
0
4
0
4
0
4º ªPˆ º
« »
0»» «Wˆ1 »
0» «Wˆ 2 »
»« »
4¼ ¬Wˆ3 ¼
ª y .. º
« »
« y1. »
« y 2. »
« »
¬ y 3. ¼
The solution to this set of equations is
Pˆ
Wˆi
y ..
y ..
12
y i. y ..
This solution was found be solving the last three equations for Wˆi , yielding Wˆi
substituting in the first equation to find Pˆ y..
y i. Pˆ , and then
The regression sum of squares is
SS R E
a
Eˆ c X’y = y .. y .. ¦ y i. y ..
i 1
2
a
y ..2
y2 y2
¦ i. ..
an i 1 n
an
a
¦
i 1
y i2.
n
with a degrees of freedom. This is the same result found in Chapter 3. For more discussion of the
relationship between analysis of variance and regression, see Montgomery and Peck (1992).
10-15 Suppose that we are fitting a straight line and we desire to make the variance of as small as
possible. Restricting ourselves to an even number of experimental points, where should we place these
points so as to minimize V Ê 1 ? (Note: Use the design called for in this exercise with great caution
because, even though it minimized V Ê 1 , it has some undesirable properties; for example, see Myers and
10-19
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Montgomery (1995). Only if you are very sure the true functional relationship is linear should you
consider using this design.
V2
, we may minimize V Ê 1 by making Sxx as large as possible. Sxx is maximized by
S xx
spreading out the xj’s as much as possible. The experimenter usually has a “region of interest” for x. If n is
even, n/2 of the observations should be run at each end of the “region of interest”. If n is odd, then run
one of the observations in the center of the region and the remaining (n-1)/2 at either end.
Since V Eˆ1
10-16 Weighted least squares. Suppose that we are fitting the straight line y
variance of the y’s now depends on the level of x; that is,
V2
, i 1,2,..., n
wi
V2
V y xi
E 0 E 1 x H , but the
where the wi are known constants, often called weights.
Show that if we choose estimates of the
n
regression coefficients to minimize the weighted sum of squared errors given by
¦ wi
i 1
the resulting least squares normal equations are
n
n
i 1
i 1
n
Eˆ 0 ¦ wi Eˆ1 ¦ wi x i
n
n
i 1
i 1
¦ wi y i
i 1
n
Eˆ 0 ¦ wi x i Eˆ1 ¦ wi x i2
¦ wi xi y i
i 1
The least squares normal equations are found:
n
¦
L
y i E 0 E 1 x1
2
wi
i 1
n
wL
wE 0
2¦ y i Eˆ 0 Eˆ1 x1 wi
wL
wE 1
2¦ y i Eˆ 0 Eˆ1 x1 x1 wi
0
i 1
n
0
i 1
which simplify to
n
n
i 1
n
i 1
Eˆ 0 ¦ wi Eˆ1 ¦ x1 wi
n
¦ wi y i
i 1
n
Eˆ 0 ¦ x1 wi Eˆ1 ¦ x12 wi
i 1
i 1
10-17 Consider the 2 4IV1 design discussed in Example 10-5.
10-20
n
wi x1 y i
¦
i
1
y i E 0 E 1 xi
2
,
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Suppose you elect to augment the design with the single run selected in that example. Find the
variances and covariances of the regression coefficients in the model (ignoring blocks):
y
X' X
ª 1
« 1
« 1
«
« 1
« 1
« 1
«
¬ 1
( X' X) 1
1
1
1
E 0 E 1 x1 E 2 x 2 E 3 x 3 E 4 x 4 E 12 x1 x 2 E 34 x 3 x 4 H
1
1
1 1
1 1
1 1
1 1
1
1
1 1 1
1
1 1 1
1
1
1
1
1 1
1 1 1
1 1
1
1 1 1
1 1
1
1
1
1
1
1
1 1 1 1
ª1
«
«1
1º
«1
1» «
»
1 «1
»
1» «1
«
1» 1
«
»
1 «
»1
1¼ «
«1
«
¬1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
0
0
0
ª 0.125
«
0.125
0
0
« 0
« 0
0
0.125
0
«
0
0
0
0.125
«
« 0
0
0
0
«
0
.
0625
0
.
0625
0
.
0625
0.0625
«
« 0.0625 0.0625 0.0625 0.0625
¬
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1º
1»»
1»
»
1»
1»
»
1»
1»
»
1»
»
1¼
ª 9 1 1 1 1 1 1º
«
»
1 1 1 1 1»
« 1 9
« 1 1 9
1 1 1 1»
«
»
« 1 1 1 9 1 1 1»
« 1 1 1 1 9
1 1»
«
»
« 1 1 1 1 1 9 7»
« 1 1 1 1 1 7 9»
¬
¼
0.0625
0
0.0625º
»
0
0.0625
0.0625»
0
0.0625 0.0625»
»
0
0.0625 0.0625»
0.125 0.0625
0.0625»
»
0.0625
0.4375 0.375 »
0.0625 0.375
0.4375»¼
(b) Are there any other runs in the alternate fraction that
Any other run from the alternate fraction will dealias AB from CD.
(c) Suppose you augment the design with four runs suggested in Example 10-5. Find the variance and
the covariances of the regression coefficients (ignoring blocks) for the model in part (a).
Choose 4 runs that are one of the quarter fractions not used in the principal half fraction.
X' X
ª 1
«
« 1
« 1
«
« 1
« 1
«
« 1
« 1
¬
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1 1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1 1
10-21
1
1
1
1
1
1
1
ª1
«1
«
«1
1º «
1
1»» «
«1
1» «
» «1
1» «
1
1» «
» «1
1» «
1
1»¼ «
«1
«
«1
«1
¬
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1º
1»»
1»
»
1»
1»
»
1»
1»
»
1»
»
1»
1»
»
1»
1»¼
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0
0 0 0 0º
ª12 0
« 0 12
0
0 4 4 0»»
«
« 0 0 12 4 0 0 0»
«
»
« 0 0 4 12 0 0 0»
« 0 4
0
0 12 4 0»
«
»
0
0 4 12 0»
« 0 4
« 0 0
0
0 0 0 12»¼
¬
X' X
X' X
1
0
ª0.0833
«0
0.1071
«
«0
0
«
0
0
«
«0
0.0179
«
0.0536
«0
«0
0.0357
¬
0
0
0.0938
0.0313
0
0
0
0
0
0
0
º
0.0179 0.0536
0
0.0357»»
»
0.0313
0
0
0
»
0.0938
0
0
0
»
0
0.1071 0.0536
0.0357»
»
0.0536
0
0.2142 0.1429»
0
0.0357 0.1429
0.1785»¼
(d) Considering parts (a) and (c), which augmentation strategy would you prefer and why?
If you only have the resources to run one more run, then choose the one-run augmentation. But if
resources are not scarce, then augment the design in multiples of two runs, to keep the design orthogonal.
Using four runs results in smaller variances of the regression coefficients and a simpler covariance
structure.
10-18 Consider the 2 7III 4 . Suppose after running the experiment, the largest observed effects are A +
BD, B + AD, and D + AB. You wish to augment the original design with a group of four runs to dealias
these effects.
(a) Which four runs would you make?
Take the first four runs of the original experiment and change the sign on A.
Design Expert Output
Std
1
2
3
4
5
6
7
8
9
10
11
12
Run
1
2
3
4
5
6
7
8
9
10
11
12
Block
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 2
Block 2
Block 2
Block 2
Factor 1
A:x1
-1.00
1.00
-1.00
1.00
-1.00
1.00
-1.00
1.00
1.00
1.00
-1.00
-1.00
Factor 2
B:x2
-1.00
-1.00
1.00
1.00
-1.00
-1.00
1.00
1.00
1.00
-1.00
-1.00
-1.00
Factor 3
C:x3
-1.00
-1.00
-1.00
-1.00
1.00
1.00
1.00
1.00
1.00
-1.00
1.00
-1.00
Factor 4
D:x4
1.00
-1.00
-1.00
1.00
1.00
-1.00
-1.00
1.00
-1.00
1.00
1.00
-1.00
Factor 5
E:x5
1.00
-1.00
1.00
-1.00
-1.00
1.00
-1.00
1.00
-1.00
-1.00
-1.00
-1.00
Factor 6
F:x6
1.00
1.00
-1.00
-1.00
-1.00
-1.00
1.00
1.00
-1.00
-1.00
-1.00
-1.00
x1x2
1
x1x4
-1
x2x4
-1
Main effects and interactions of interest are:
x1
-1
x2
-1
x4
1
10-22
Factor 7
G:x7
-1.00
1.00
1.00
-1.00
1.00
-1.00
-1.00
1.00
-1.00
-1.00
-1.00
-1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1
-1
1
-1
1
-1
1
1
-1
1
-1
-1
1
1
-1
-1
1
1
-1
-1
1
1
-1
-1
1
1
-1
-1
1
1
-1
-1
1
-1
-1
1
1
-1
-1
1
-1
1
1
-1
-1
1
1
-1
-1
1
1
1
1
-1
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
(b) Find the variances and covariances of the regression coefficients in the model
y
E 0 E 1 x1 E 2 x 2 E 4 x 4 E12 x1 x 2 E 14 x1 x 4 E 24 x 2 x 4 H
X' X
X' X
1
0
0
0
0
0
0º
ª12
«
0
12
0
0
0
0
4»»
«
« 0
0 12
0
0 4
0»
«
»
0
0 12 4
0
0»
« 0
« 0
0
0 4 12
0
0»
«
»
0 4
0
0 12
0»
« 0
« 0 4
0
0
0
0 12»¼
¬
0
0
ª0.0833
«0
0.1071 0.0178
«
«0
0.0179
0.1071
«
0
0
«0
«0
0
0
«
0.0536
0.0714
«0
«0
0.0714 0.0536
¬
0
0
0
0.0938
0.0313
0
0
0
0
0
º
0
0.0536
0.0714»»
0
0.0714 0.0536»
»
0.0313
0
0
»
»
0.0938
0
0
»
0
0.2143 0.1607»
0.1607
0
0.2143»¼
(c) Is it possible to dealias these effects with fewer than four additional runs?
It is possible to dealias these effects in only two runs. By utilizing Design Expert’s design augmentation
function, the runs 9 and 10 (Block 2) were generated as follows:
Design Expert Output
Std
1
2
3
4
5
6
7
8
9
10
Run
1
2
3
4
5
6
7
8
9
10
Block
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 1
Block 2
Block 2
Factor 1
A:x1
-1.00
1.00
-1.00
1.00
-1.00
1.00
-1.00
1.00
-1.00
1.00
Factor 2
B:x2
-1.00
-1.00
1.00
1.00
-1.00
-1.00
1.00
1.00
1.00
-1.00
Factor 3
C:x3
-1.00
-1.00
-1.00
-1.00
1.00
1.00
1.00
1.00
-1.00
-1.00
Factor 4
D:x4
1.00
-1.00
-1.00
1.00
1.00
-1.00
-1.00
1.00
1.00
-1.00
10-23
Factor 5
E:x5
1.00
-1.00
1.00
-1.00
-1.00
1.00
-1.00
1.00
-1.00
-1.00
Factor 6
F:x6
1.00
1.00
-1.00
-1.00
-1.00
-1.00
1.00
1.00
-1.00
-1.00
Factor 7
G:x7
-1.00
1.00
1.00
-1.00
1.00
-1.00
-1.00
1.00
-1.00
-1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 11
Response Surface Methods and
Other Approaches to Process Optimization
Solutions
11-1 A chemical plant produces oxygen by liquefying air and separating it into its component gases by
fractional distillation. The purity of the oxygen is a function of the main condenser temperature and the
pressure ratio between the upper and lower columns. Current operating conditions are temperature
( [1 ) -220°C and pressure ratio ( [ 2 ) 1.2. Using the following data find the path of steepest ascent.
Temperature (x1)
-225
-225
-215
-215
-220
-220
-220
-220
Design Expert Output
Response:
Purity
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
3.14
2
A
2.89
1
B
0.25
1
Curvature
0.080
1
Residual
0.24
4
Lack of Fit
0.040
1
Pure Error
0.20
3
Cor Total
3.46
7
Pressure Ratio (x2)
1.1
1.3
1.1
1.3
1.2
1.2
1.2
1.2
Mean
Square
1.57
2.89
0.25
0.080
0.060
0.040
0.067
Purity
82.8
83.5
84.7
85.0
84.1
84.5
83.9
84.3
F
Value
26.17
48.17
4.17
1.33
Prob > F
0.0050
0.0023
0.1108
0.3125
not significant
0.60
0.4950
not significant
significant
The Model F-value of 26.17 implies the model is significant. There is only
a 0.50% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.24
84.10
0.29
1.00
Factor
Intercept
A-Temperature
B-Pressure Ratio
Center Point
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
84.00
0.85
0.25
0.20
DF
1
1
1
1
0.9290
0.8935
0.7123
12.702
Standard
Error
0.12
0.12
0.12
0.17
Final Equation in Terms of Coded Factors:
Purity =
+84.00
+0.85 * A
+0.25 * B
Final Equation in Terms of Actual Factors:
Purity =
+118.40000
11-1
95% CI
Low
83.66
0.51
-0.090
-0.28
95% CI
High
84.34
1.19
0.59
0.68
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+0.17000
+2.50000
* Temperature
* Pressure Ratio
From the computer output use the model ŷ 84 0.85 x1 0.25 x2 as the equation for steepest ascent.
Suppose we use a one degree change in temperature as the basic step size. Thus, the path of steepest
ascent passes through the point (x1=0, x2=0) and has a slope 0.25/0.85. In the coded variables, one degree
of temperature is equivalent to a step of 'x1 1/5=0.2. Thus, 'x 2 (0.25/0.85)0.2=0.059. The path of
steepest ascent is:
Origin
'
Origin + '
Origin +5 '
Origin +7 '
Coded
x1
Variables
x2
Natural
[1
Variables
[2
0
0.2
0.2
1.0
1.40
0
0.059
0.059
0.295
0.413
-220
1
-219
-215
-213
1.2
0.0059
1.2059
1.2295
1.2413
11-2 An industrial engineer has developed a computer simulation model of a two-item inventory system.
The decision variables are the order quantity and the reorder point for each item. The response to be
minimized is the total inventory cost. The simulation model is used to produce the data shown in the
following table. Identify the experimental design. Find the path of steepest descent.
Item 1
Order
Reorder
Quantity (x1)
Point (x2)
100
25
140
45
140
25
140
25
100
45
100
45
100
25
140
45
120
35
120
35
120
35
Item 2
Order
Reorder
Quantity (x3)
Point (x4)
250
40
250
40
300
40
250
80
300
40
250
80
300
80
300
80
275
60
275
60
275
60
Total
Cost
625
670
663
654
648
634
692
686
680
674
681
The design is a 24-1 fractional factorial with generator I=ABCD, and three center points.
Design Expert Output
Response:
Total Cost
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
3880.00
6
646.67
A
684.50
1
684.50
C
1404.50
1
1404.50
D
450.00
1
450.00
AC
392.00
1
392.00
AD
264.50
1
264.50
CD
684.50
1
684.50
Curvature
815.52
1
815.52
Residual
30.67
3
10.22
Lack of Fit
2.00
1
2.00
Pure Error
28.67
2
14.33
Cor Total
4726.18
10
F
Value
63.26
66.96
137.40
44.02
38.35
25.88
66.96
79.78
Prob > F
0.0030
0.0038
0.0013
0.0070
0.0085
0.0147
0.0038
0.0030
significant
0.14
0.7446
not significant
The Model F-value of 63.26 implies the model is significant. There is only
11-2
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a 0.30% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
3.20
664.27
0.48
192.50
Factor
Intercept
A-Item 1 QTY
C-Item 2 QTY
D-Item 2 Reorder
AC
AD
CD
Center Point
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
659.00
9.25
13.25
7.50
-7.00
-5.75
9.25
19.33
DF
1
1
1
1
1
1
1
1
0.9922
0.9765
0.9593
24.573
Standard
Error
1.13
1.13
1.13
1.13
1.13
1.13
1.13
2.16
95% CI
Low
655.40
5.65
9.65
3.90
-10.60
-9.35
5.65
12.44
95% CI
High
662.60
12.85
16.85
11.10
-3.40
-2.15
12.85
26.22
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Total Cost =
+659.00
+9.25 * A
+13.25 * C
+7.50 * D
-7.00 * A * C
-5.75 * A * D
+9.25 * C * D
Final Equation in Terms of Actual Factors:
Total Cost =
+175.00000
+5.17500 * Item 1 QTY
+1.10000 * Item 2 QTY
-2.98750 * Item 2 Reorder
-0.014000 * Item 1 QTY * Item 2 QTY
-0.014375 * Item 1 QTY * Item 2 Reorder
+0.018500 * Item 2 QTY * Item 2 Reorder
+0.019 * Item 2 QTY * Item 2 Reorder
The equation used to compute the path of steepest ascent is ŷ 659 9.25 x1 13.25 x3 7.50 x4 . Notice
that even though the model contains interaction, it is relatively common practice to ignore the interactions
in computing the path of steepest ascent. This means that the path constructed is only an approximation
to the path that would have been obtained if the interactions were considered, but it’s usually close enough
to give satisfactory results.
It is helpful to give a general method for finding the path of steepest ascent. Suppose we have a first-order
model in k variables, say
ŷ
Eˆ 0
k
¦ Eˆ x
i i
i 1
The path of steepest ascent passes through the origin, x=0, and through the point on a hypersphere of
radius, R where ŷ is a maximum. Thus, the x’s must satisfy the constraint
k
¦x
2
i
R2
i 1
To find the set of x’s that maximize ŷ subject to this constraint, we maximize
11-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Eˆ 0
L
k
¦
i 1
ª k
º
Eˆ i xi O « xi2 R 2 »
«¬ i 1
»¼
¦
where O is a LaGrange multiplier. From wL / wxi
wL / wO
0 , we find
Eˆ i
2O
xi
It is customary to specify a basic step size in one of the variables, say ' xj, and then calculate 2 O as
2 O = Eˆ j / 'x j . Then this value of 2 O can be used to generate the remaining coordinates of a point on
the path of steepest ascent.
We demonstrate using the data from this problem. Suppose that we use -10 units in [1 as the basic step
size. Note that a decrease in [1 is called for, because we are looking for a path of steepest decent. Now
-10 units in [1 is equal to -10/20 = -0.5 units change in x1.
Thus, 2 O = Eˆ 1 / 'x1 = 9.25/(-0.5) = -18.50
Consequently,
'x 3
'x 4
Eˆ 3
2O
Eˆ 4
2O
13.25
18.50
0.716
7.50
18.50
0.705
are the remaining coordinates of points along the path of steepest decent, in terms of the coded variables.
The path of steepest decent is shown below:
Origin
'
Origin + '
Origin +2 '
Coded
x1
Variables
x2
x3
x4
Natural
[1
Variables
[2
0
-0.50
-0.50
-1.00
0
0
0
0
0
-0.716
-0.716
-1.432
0
-0.405
-0.405
-0.810
120
-10
110
100
35
0
35
35
[3
275
-17.91
257.09
239.18
[4
60
-8.11
51.89
43.78
11-3 Verify that the following design is a simplex. Fit the first-order model and find the path of steepest
ascent.
Position
1
x1
0
2
3
- 2
0
4
2
x2
2
0
- 2
0
11-4
x3
-1
y
18.5
1
19.8
-1
17.4
1
22.5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1
4
2
x2
x3
3
x1
The graphical representation of the design identifies a tetrahedron; therefore, the design is a simplex.
Design Expert Output
Response:
y
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
14.49
3
A
3.64
1
B
0.61
1
C
10.24
1
Pure Error
0.000
0
Cor Total
14.49
3
Mean
Square
4.83
3.64
0.61
10.24
F
Value
Prob > F
Std. Dev.
R-Squared 1.0000
Mean 19.55
Adj R-Squared
C.V.
Pred R-Squared N/A
PRESS N/A
Adeq Precision 0.000
Case(s) with leverage of 1.0000: Pred R-Squared and PRESS statistic not defined
Factor
Intercept
A-x1
B-x2
C-x3
Coefficient
Estimate
19.55
1.35
0.55
1.60
DF
1
1
1
1
Standard
Error
95% CI
Low
95% CI
High
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
y =
+19.55
+1.35 * A
+0.55 * B
+1.60 * C
Final Equation in Terms of Actual Factors:
y =
+19.55000
+0.95459 * x1
+0.38891 * x2
+1.60000 * x3
11-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The first order model is ŷ 19.55 1.35x1 0.55 x2 1.60 x3 .
To find the path of steepest ascent, let the basic step size be 'x3
the previous problem, we obtain
'x 3
1 . Then using the results obtained in
Eˆ 3
1.60
or 1.0 =
2O
2O
which yields 2O 1.60 . Then the coordinates of points on the path of steepest ascent are defined by
'x1
'x 2
Eˆ 1
2O
Eˆ 2
2O
0.96
1.60
0.60
0.24
1.60
0.24
Therefore, in the coded variables we have:
Origin
'
Origin + '
Origin +2 '
11-4
For the first-order model ŷ
variables are coded as 1 d xi d 1 .
Let the basic step size be 'x3
1 . 'x 3
Coded
x1
0
0.60
0.60
1.20
Variables
x2
0
0.24
0.24
0.48
x3
0
1.00
1.00
2.00
60 1.5 x1 0.8x 2 2.0 x3 find the path of steepest ascent. The
Eˆ 3
2.0
or 1.0 =
. Then 2O
2O
2O
Eˆ 1 1.50
'x1
0.75
2O 2.0
Eˆ 2 0.8
0.40
'x 2
2O
2.0
2.0
Therefore, in the coded variables we have
Origin
'
Origin + '
Origin +2 '
Coded
x1
0
0.75
0.75
1.50
Variables
x2
0
-0.40
-0.40
-0.80
x3
0
1.00
1.00
2.00
11-5
The region of experimentation for three factors are time ( 40 d T1 d 80 min), temperature
( 200 d T2 d 300 °C), and pressure ( 20 d P d 50 psig). A first-order model in coded variables has been fit
to yield data from a 23 design. The model is
11-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
30 5 x1 2.5 x 2 3.5 x3
ŷ
Is the point T1 = 85, T2 = 325, P=60 on the path of steepest ascent?
The coded variables are found with the following:
x1
T1 60
20
x2
'T1
'x1
Origin
'
Origin + '
Origin +5 '
Coded
x1
0
0.25
0.25
1.25
5
P 35
T2 250
x3 1
50
15
5
'x1
0.25
20
Eˆ 1
20
or 0.25 =
2O
2O
2O
Eˆ 2 2.5
'x 2
0.125
2O 20
Eˆ 3 3.5
'x 3
0.175
2O 20
Variables
x2
0
0.125
0.125
0.625
x3
0
0.175
0.175
0.875
20
Natural
T1
60
5
65
85
Variables
T2
250
6.25
256.25
281.25
P
35
2.625
37.625
48.125
The point T1=85, T2=325, and P=60 is not on the path of steepest ascent.
11-6 The region of experimentation for two factors are temperature ( 100 d T d 300q F) and catalyst feed
rate ( 10 d C d 30 lb/h). A first order model in the usual r 1 coded variables has been fit to a molecular
weight response, yielding the following model.
ŷ
2000 125x1 40 x2
(a) Find the path of steepest ascent.
x1
T 200
100
'T
'x1
Origin
'
100
C 20
10
100
1
100
x2
'x1
Eˆ 1
125
or 1
2O 125
2O
2O
Eˆ 2
40
'x2
0.32
2O 125
Coded
x1
0
1
Variables
x2
0
0.32
11-7
Natural
T
200
100
Variables
C
20
3.2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Origin + '
Origin +5 '
1
5
0.32
1.60
300
700
23.2
36.0
(a) It is desired to move to a region where molecular weights are above 2500. Based on the information
you have from the experiment, in this region, about how may steps along the path of steepest ascent
might be required to move to the region of interest?
'ŷ
'x1 Eˆ 1 'x 2 Eˆ 2
1 125 0.32 40
2500 2000
137.8
# Steps
137.8
3.63 o 4
11-7 The path of steepest ascent is usually computed assuming that the model is truly first-order.; that is,
there is no interaction. However, even if there is interaction, steepest ascent ignoring the interaction still
usually produces good results. To illustrate, suppose that we have fit the model
ŷ
20 5 x1 8 x2 3 x1 x2
using coded variables (-1 d x1 d +1)
(a) Draw the path of steepest ascent that you would obtain if the interaction were ignored.
Path of Steepest Ascent for
Main Effects Model
0
-1
X2
-2
-3
-4
-5
0
1
2
3
4
5
X1
(b) Draw the path of steepest ascent that you would obtain with the interaction included in the model.
Compare this with the path found in part (a).
11-8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Path of Steepest Ascent for
Full Model
0
-1
X2
-2
-3
-4
-5
-2
-1
0
1
2
3
X1
11-8 The data shown in the following table were collected in an experiment to optimize crystal growth
as a function of three variables x1, x2, and x3. Large values of y (yield in grams) are desirable. Fit a
second order model and analyze the fitted surface. Under what set of conditions is maximum growth
achieved?
x1
x2
x3
y
-1
-1
-1
-1
1
1
1
1
-1.682
1.682
0
0
0
0
0
0
0
0
0
0
-1
-1
1
1
-1
-1
1
1
0
0
-1.682
1.682
0
0
0
0
0
0
0
0
-1
1
-1
1
-1
1
-1
1
0
0
0
0
-1.682
1.682
0
0
0
0
0
0
66
70
78
60
80
70
100
75
100
80
68
63
65
82
113
100
118
88
100
85
Design Expert Output
Response:
Yield
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
3662.00
9
406.89
A
22.08
1
22.08
B
25.31
1
25.31
C
30.50
1
30.50
F
Value
2.19
0.12
0.14
0.16
11-9
Prob > F
0.1194
0.7377
0.7200
0.6941
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A2
B2
204.55
1
204.55
1.10
0.3191
2226.45
1
2226.45
11.96
0.0061
C2
AB
AC
BC
Residual
Lack of Fit
Pure Error
Cor Total
1328.46
66.12
55.13
171.13
1860.95
1001.61
859.33
5522.95
1
1
1
1
10
5
5
19
1328.46
66.12
55.13
171.13
186.09
200.32
171.87
7.14
0.36
0.30
0.92
0.0234
0.5644
0.5982
0.3602
1.17
0.4353
not significant
The "Model F-value" of 2.19 implies the model is not significant relative to the noise. There is a
11.94 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
13.64
83.05
16.43
8855.23
Factor
Intercept
A-x1
B-x2
C-x3
A2
B2
C2
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
100.67
1.27
1.36
-1.49
DF
1
1
1
1
0.6631
0.3598
-0.6034
3.882
Standard
Error
5.56
3.69
3.69
3.69
95% CI
Low
88.27
-6.95
-6.86
-9.72
95% CI
High
113.06
9.50
9.59
6.73
VIF
1.00
1.00
1.00
-3.77
1
3.59
-11.77
4.24
1.02
-12.43
1
3.59
-20.44
-4.42
1.02
-9.60
2.87
-2.63
-4.63
1
1
1
1
3.59
4.82
4.82
4.82
-17.61
-7.87
-13.37
-15.37
-1.59
13.62
8.12
6.12
1.02
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Yield =
+100.67
+1.27 * A
+1.36 * B
-1.49 * C
-3.77 * A2
-12.43
-9.60
+2.87
-2.63
-4.63
* B2
* C2
*A*B
*A*C
*B*C
Final Equation in Terms of Actual Factors:
Yield =
+100.66609
+1.27146 * x1
+1.36130 * x2
-1.49445 * x3
-12.42955
* x12
* x22
-9.60113
+2.87500
-2.62500
-4.62500
* x32
* x1 * x2
* x1 * x3
* x2 * x3
-3.76749
There are so many nonsignificant terms in this model that we should consider eliminating some of them.
A reasonable reduced model is shown below.
Design Expert Output
11-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Response:
Yield
ANOVA for Response Surface Reduced Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
3143.00
4
785.75
B
25.31
1
25.31
C
30.50
1
30.50
B2
2115.31
1
2115.31
C2
1239.17
1
1239.17
Residual
2379.95
15
158.66
Lack of Fit
1520.62
10
152.06
Pure Error
859.33
5
171.87
Cor Total
5522.95
19
F
Value
4.95
0.16
0.19
13.33
7.81
Prob > F
0.0095
0.6952
0.6673
0.0024
0.0136
0.88
0.5953
significant
not significant
The Model F-value of 4.95 implies the model is significant. There is only
a 0.95% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
12.60
83.05
15.17
4735.52
Factor
Intercept
B-x2
C-x3
B2
C2
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
97.58
1.36
-1.49
-12.06
-9.23
DF
1
1
1
1
1
0.5691
0.4542
0.1426
5.778
Standard
Error
4.36
3.41
3.41
3.30
3.30
95% CI
Low
88.29
-5.90
-8.76
-19.09
-16.26
Final Equation in Terms of Coded Factors:
Yield =
+97.58
+1.36 * B
-1.49 * C
-12.06 * B2
-9.23
* C2
Final Equation in Terms of Actual Factors:
Yield =
+97.58260
+1.36130 * x2
-1.49445 * x3
-12.05546
-9.22703
* x22
* x32
The contour plot identifies a maximum near the center of the design space.
11-11
95% CI
High
106.88
8.63
5.77
-5.02
-2.19
VIF
1.00
1.00
1.01
1.01
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Yie ld
DE S IG N-E X P E RT P l o t
1.00
80
Yield
X = B : x2
Y = C: x3
80
85
De si g n P o i n ts
0.50
P red ic t ion9 7. 6 82
9 5% Lo w 6 9. 2 73
9 5% H ig h1 26 . 09 0
S E m e an 4 .3 5 58 4
S E pre d 1 3. 3 28 1
X
0 .0 6
Y
-0. 0 8 6
A ctu al Fa ctor
A : x1 = 0 .0 0
C : x3
85
0.00
-0.50
95
85
80
90
-1.00
-1.00
-0.50
0.00
0.50
1.00
B: x2
11-9 The following data were collected by a chemical engineer. The response y is filtration time, x1 is
temperature, and x2 is pressure. Fit a second-order model.
x1
x2
y
-1
-1
1
1
-1.414
1.414
0
0
0
0
0
0
0
-1
1
-1
1
0
0
-1.414
1.414
0
0
0
0
0
54
45
32
47
50
53
47
51
41
39
44
42
40
Design Expert Output
Response:
y
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
264.22
4
66.06
A
13.11
1
13.11
B
25.72
1
25.72
A2
81.39
1
81.39
AB
Residual
Lack of Fit
Pure Error
Cor Total
144.00
205.78
190.98
14.80
470.00
1
8
4
4
12
144.00
25.72
47.74
3.70
F
Value
2.57
0.51
1.00
Prob > F
0.1194
0.4955
0.3467
3.16
5.60
0.1132
0.0455
12.90
0.0148
The "Model F-value" of 2.57 implies the model is not significant relative to the noise. There is a
11.94 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
5.07
R-Squared
0.5622
11-12
not significant
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Mean
C.V.
PRESS
45.00
11.27
716.73
Factor
Intercept
A-Temperature
B-Pressure
A2
AB
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
42.91
1.28
-1.79
DF
1
1
1
3.39
6.00
1
1
0.3433
-0.5249
4.955
Standard
Error
1.83
1.79
1.79
1.91
2.54
95% CI
Low
38.69
-2.85
-5.93
95% CI
High
47.14
5.42
2.34
-1.01
0.15
7.79
11.85
VIF
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Time =
+42.91
+1.28 * A
-1.79 * B
+3.39 * A2
+6.00
*A*B
Final Equation in Terms of Actual Factors:
Time =
+42.91304
+1.28033 * Temperature
-1.79289 * Pressure
+3.39130 * Temperature2
+6.00000
* Temperature * Pressure
The lack of fit test in the above analysis is significant. Also, the residual plot below identifies an outlier
which happens to be standard order number 8.
No rm a l p lot o f re sid uals
99
N orm al % probability
95
90
80
70
50
30
20
10
5
1
-5.23112
-1.26772
2.69568
6.65909
10.6225
R es idua l
We chose to remove this run and re-analyze the data.
Design Expert Output
Response:
y
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
407.34
4
101.84
A
13.11
1
13.11
11-13
F
Value
30.13
3.88
Prob > F
0.0002
0.0895
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B
A2
AB
Residual
Lack of Fit
Pure Error
Cor Total
132.63
155.27
144.00
23.66
8.86
14.80
431.00
1
1
1
7
3
4
11
132.63
155.27
144.00
3.38
2.95
3.70
39.25
45.95
42.61
0.0004
0.0003
0.0003
0.80
0.5560
not significant
The Model F-value of 30.13 implies the model is significant. There is only
a 0.02% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.84
44.50
4.13
80.66
Factor
Intercept
A-Temperature
B-Pressure
A2
AB
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
40.68
1.28
-4.82
DF
1
1
1
4.88
6.00
1
1
0.9451
0.9138
0.8129
18.243
Standard
Error
0.73
0.65
0.77
95% CI
Low
38.95
-0.26
-6.64
0.72
0.92
95% CI
High
42.40
2.82
-3.00
3.18
3.83
6.59
8.17
Final Equation in Terms of Coded Factors:
Time =
+40.68
+1.28 * A
-4.82 * B
+4.88 * A2
+6.00
*A*B
Final Equation in Terms of Actual Factors:
Time =
+40.67673
+1.28033 * Temperature
-4.82374 * Pressure
+4.88218 * Temperature2
+6.00000
* Temperature * Pressure
The lack of fit test is satisfactory as well as the following normal plot of residuals:
No rm a l p lot o f re sid uals
99
N orm al % probability
95
90
80
70
50
30
20
10
5
1
-1.67673
-0.42673
0.82327
R es idua l
11-14
2.07327
3.32327
VIF
1.00
1.02
1.02
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) What operating conditions would you recommend if the objective is to minimize the filtration time?
Tim e
1.00
34
B: Pres s ure
P red ic t ion
3 3. 1 95
9 5% Lo w
2 7. 8 85
9 5% H ig h
3 8. 5 06
S E m e an
1 .2 9 00 7
S E pre d
2 .2 4 58 1
X
-0. 6 8
0.50
Y
1 .0 0
36
38
40
46
5
0.00
42
-0.50
46
48
50
44
52
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: Tem pera ture
(b) What operating conditions would you recommend if the objective is to operate the process at a mean
filtration time very close to 46?
Tim e
1.00
34
36
B: Pres s ure
0.50
38
40
46
5
0.00
42
-0.50
46
48
50
44
52
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: Tem pera ture
There are two regions that enable a filtration time of 46. Either will suffice; however, higher temperatures
and pressures typically have higher operating costs. We chose the operating conditions at the lower
pressure and temperature as shown.
11-10 The hexagon design that follows is used in an experiment that has the objective of fitting a secondorder model.
x1
1
x2
0
11-15
y
68
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0.5
0.75
0.75
0
- 0.75
- 0.75
0
0
0
0
0
-0.5
-1
-0.5
0.5
0
0
0
0
0
74
65
60
63
70
58
60
57
55
69
(a) Fit the second-order model.
Design Expert Output
Response:
y
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
245.26
5
49.05
A
85.33
1
85.33
B
9.00
1
9.00
A2
25.20
1
25.20
B2
AB
Residual
Lack of Fit
Pure Error
Cor Total
129.83
1.00
129.47
10.67
118.80
374.73
1
1
5
1
4
10
129.83
1.00
25.89
10.67
29.70
F
Value
1.89
3.30
0.35
Prob > F
0.2500
0.1292
0.5811
0.97
0.3692
5.01
0.039
0.0753
0.8519
0.36
0.5813
not significant
not significant
The "Model F-value" of 1.89 implies the model is not significant relative to the noise. There is a
25.00 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
5.09
63.55
8.01
569.63
Factor
Intercept
A-x1
B-x2
A2
B2
AB
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
59.80
5.33
1.73
DF
1
1
1
0.6545
0.3090
-0.5201
3.725
Standard
Error
2.28
2.94
2.94
95% CI
Low
53.95
-2.22
-5.82
95% CI
High
65.65
12.89
9.28
VIF
1.00
1.00
4.20
1
4.26
-6.74
15.14
1.00
9.53
1.15
1
1
4.26
5.88
-1.41
-13.95
20.48
16.26
1.00
1.00
Final Equation in Terms of Coded Factors:
y =
+59.80
+5.33 * A
+1.73 * B
+4.20 * A2
+9.53
+1.15
* B2
*A*B
(a) Perform the canonical analysis. What type of surface has been found?
The full quadratic model is used in the following analysis because the reduced model is singular.
11-16
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Solution
Variable
Critical Value
X1
-0.627658
X2
-0.052829
Predicted Value at Solution
58.080492
Variable
X1
X2
Eigenvalues and Eigenvectors
9.5957
4.1382
0.10640
0.99432
0.99432
-0.10640
Since both eigenvalues are positive, the response is a minimum at the stationary point.
(c) What operating conditions on x1 and x2 lead to the stationary point?
The stationary point is (x1,x2) = (-0.62766, -0.05283)
(d) Where would you run this process if the objective is to obtain a response that is as close to 65 as
possible?
y
0.87
75
70
0.43
B: x2
65
5
0.00
60
-0.43
70
-0.87
-1.00
-0.50
0.00
0.50
1.00
A: x1
Any value of x1 and x2 that give a point on the contour with value of 65 would be satisfactory.
11-11 An experimenter has run a Box-Behnken design and has obtained the results below, where the
response variable is the viscosity of a polymer.
Level
Temp.
Agitation
Rate
Pressure
High
Middle
Low
200
175
150
10.0
7.5
5.0
25
20
15
11-17
x1
+1
0
-1
x2
+1
0
-1
x3
+1
0
-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Run
x1
x2
x3
y1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
-1
1
-1
1
-1
1
-1
1
0
0
0
0
0
0
0
-1
-1
1
1
0
0
0
0
-1
1
-1
1
0
0
0
0
0
0
0
-1
-1
1
1
-1
-1
1
1
0
0
0
535
580
596
563
645
458
350
600
595
648
532
656
653
599
620
(a) Fit the second-order model.
Design Expert Output
Response:
Viscosity
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
89652.58
9
9961.40
A
703.12
1
703.12
B
6105.12
1
6105.12
C
5408.00
1
5408.00
A2
20769.23
1
20769.23
B2
C2
AB
AC
BC
Residual
Lack of Fit
Pure Error
Cor Total
F
Value
9.54
0.67
5.85
5.18
Prob > F
0.0115
0.4491
0.0602
0.0719
19.90
0.0066
1404.00
1
1404.00
1.35
0.2985
4719.00
1521.00
47742.25
1260.25
5218.75
3736.75
1482.00
94871.33
1
1
1
1
5
3
2
14
4719.00
1521.00
47742.25
1260.25
1043.75
1245.58
741.00
4.52
1.46
45.74
1.21
0.0868
0.2814
0.0011
0.3219
1.68
0.3941
significant
not significant
The Model F-value of 9.54 implies the model is significant. There is only
a 1.15% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
32.31
575.33
5.62
63122.50
Factor
Intercept
A-Temperatue
B-Agitation Rate
C-Pressure
A2
B2
C2
AB
AC
BC
Coefficient
Estimate
624.00
9.37
27.62
-26.00
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
0.9450
0.8460
0.3347
10.425
Standard
Error
18.65
11.42
11.42
11.42
95% CI
Low
576.05
-19.99
-1.74
-55.36
95% CI
High
671.95
38.74
56.99
3.36
VIF
1.00
1.00
1.00
-75.00
1
16.81
-118.22
-31.78
19.50
1
16.81
-23.72
62.72
1.01
-35.75
-19.50
109.25
17.75
1
1
1
1
16.81
16.15
16.15
16.15
-78.97
-61.02
67.73
-23.77
7.47
22.02
150.77
59.27
1.01
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
11-18
1.01
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Viscosity =
+624.00
+9.37 * A
+27.62 * B
-26.00 * C
-75.00 * A2
+19.50
-35.75
-19.50
+109.25
+17.75
* B2
* C2
*A*B
*A*C
*B*C
Final Equation in Terms of Actual Factors:
Viscosity =
-629.50000
+27.23500 * Temperatue
-9.55000 * Agitation Rate
-111.60000 * Pressure
-0.12000 * Temperatue2
+3.12000
-1.43000
-0.31200
+0.87400
+1.42000
* Agitation Rate2
* Pressure2
* Temperatue * Agitation Rate
* Temperatue * Pressure
* Agitation Rate * Pressure
(b) Perform the canonical analysis. What type of surface has been found?
Solution
Variable
Critical Value
X1
2.1849596
X2
-0.871371
X3
2.7586015
Predicted Value at Solution
586.34437
Variable
X1
X2
X3
Eigevalues and Eigevectors
20.9229
2.5208
-0.02739
0.58118
0.99129
-0.08907
0.12883
0.80888
-114.694
0.81331
0.09703
-0.57368
The system is a saddle point.
(c) What operating conditions on x1, x2, and x3 lead to the stationary point?
The stationary point is (x1, x2, x3) = (2.18496, -0.87167, 2.75860). This is outside the design region. It
would be necessary to either examine contour plots or use numerical optimization methods to find desired
operating conditions.
(d) What operating conditions would you recommend if it is important to obtain a viscosity that is as
close to 600 as possible?
11-19
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
V iscosity
DE S IG N-E X P E RT P l o t
25.00
V i sco si ty
X = A : T e m p e ra tu e
Y = C: P re ssu re
4 00
4 50
5 00
De si g n P o i n ts
22.50
5 50
A ctu a l Fa cto r
B : A g i ta ti o n R a te = 7 .5 0
C : Pre s s u re
6 00
3
20.00
6 00
17.50
5 50
5 00
15.00
150.00
162.50
175.00
187.50
200.00
A: Tem pe ra tue
Any point on either of the contours showing a viscosity of 600 is satisfactory.
11-12 Consider the three-variable central composite design shown below. Analyze the data and draw
conclusions, assuming that we wish to maximize conversion (y1) with activity (y2) between 55 and 60.
Run
Time
(min)
Temperature
(qC)
Catalyst
(%)
Conversion (%)
y1
Activity
y2
1
-1.000
-1.000
-1.000
74.00
53.20
2
1.000
-1.000
-1.000
51.00
62.90
3
-1.000
1.000
-1.000
88.00
53.40
4
1.000
1.000
-1.000
70.00
62.60
5
-1.000
-1.000
1.000
71.00
57.30
6
1.000
-1.000
1.000
90.00
67.90
7
-1.000
1.000
1.000
66.00
59.80
8
1.000
1.000
1.000
97.00
67.80
9
0.000
0.000
0.000
81.00
59.20
10
0.000
0.000
0.000
75.00
60.40
11
0.000
0.000
0.000
76.00
59.10
12
0.000
0.000
0.000
83.00
60.60
13
-1.682
0.000
0.000
76.00
59.10
14
1.682
0.000
0.000
79.00
65.90
15
0.000
-1.682
0.000
85.00
60.00
16
0.000
1.682
0.000
97.00
60.70
17
0.000
0.000
-1.682
55.00
57.40
18
0.000
0.000
1.682
81.00
63.20
19
0.000
0.000
0.000
80.00
60.80
20
0.000
0.000
0.000
91.00
58.90
Quadratic models are developed for the Conversion and Activity response variables as follows:
11-20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Conversion
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2555.73
9
283.97
A
14.44
1
14.44
B
222.96
1
222.96
C
525.64
1
525.64
A2
48.47
1
48.47
B2
124.48
1
124.48
C2
388.59
1
388.59
AB
36.13
1
36.13
AC
1035.13
1
1035.13
BC
120.12
1
120.12
Residual
222.47
10
22.25
Lack of Fit
56.47
5
11.29
Pure Error
166.00
5
33.20
Cor Total
287.28
19
F
Value
12.76
0.65
10.02
23.63
2.18
5.60
17.47
1.62
46.53
5.40
Prob > F
0.0002
0.4391
0.0101
0.0007
0.1707
0.0396
0.0019
0.2314
< 0.0001
0.0425
0.34
0.8692
significant
not significant
The Model F-value of 12.76 implies the model is significant. There is only
a 0.02% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
4.72
78.30
6.02
676.22
Factor
Intercept
A-Time
B-Temperature
C-Catalyst
A2
B2
C2
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
81.09
1.03
4.04
6.20
-1.83
2.94
-5.19
2.13
11.38
-3.87
DF
1
1
1
1
1
1
1
1
1
1
0.9199
0.8479
0.7566
14.239
Standard
Error
1.92
1.28
1.28
1.28
1.24
1.24
1.24
1.67
1.67
1.67
Final Equation in Terms of Coded Factors:
Conversion
+81.09
+1.03
+4.04
+6.20
-1.83
+2.94
-5.19
+2.13
+11.38
-3.87
=
*A
*B
*C
* A2
* B2
* C2
*A*B
*A*C
*B*C
Final Equation in Terms of Actual Factors:
Conversion
+81.09128
+1.02845
+4.04057
+6.20396
-1.83398
+2.93899
-5.19274
+2.12500
+11.37500
=
* Time
* Temperature
* Catalyst
* Time2
* Temperature2
* Catalyst2
* Time * Temperature
* Time * Catalyst
11-21
95% CI
Low
76.81
-1.82
1.20
3.36
-4.60
0.17
-7.96
-1.59
7.66
-7.59
95% CI
High
85.38
3.87
6.88
9.05
0.93
5.71
-2.42
5.84
15.09
-0.16
VIF
1.00
1.00
1.00
1.02
1.02
1.02
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-3.87500 * Temperature * Catalyst
Design Expert Output
Response:
Activity
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
256.20
9
28.47
A
175.35
1
175.35
B
0.89
1
0.89
C
67.91
1
67.91
10.05
1
10.05
A2
0.081
1
0.081
B2
0.047
1
0.047
C2
AB
1.20
1
1.20
AC
0.011
1
0.011
BC
0.78
1
0.78
Residual
31.08
10
3.11
Lack of Fit
27.43
5
5.49
Pure Error
3.65
5
0.73
Cor Total
287.28
19
F
Value
Prob > F
9.16
0.0009
56.42
< 0.0001
0.28
0.6052
21.85
0.0009
3.23
0.1024
0.026
0.8753
0.015
0.9046
0.39
0.5480
3.620E-003 0.9532
0.25
0.6270
7.51
0.0226
significant
significant
The Model F-value of 9.16 implies the model is significant. There is only
a 0.09% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.76
60.51
2.91
214.43
Factor
Intercept
A-Time
B-Temperature
C-Catalyst
A2
B2
C2
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
59.85
3.58
0.25
2.23
0.83
0.075
0.057
-0.39
-0.038
0.31
DF
1
1
1
1
1
1
1
1
1
1
0.8918
0.7945
0.2536
10.911
Standard
Error
0.72
0.48
0.48
0.48
0.46
0.46
0.46
0.62
0.62
0.62
Final Equation in Terms of Coded Factors:
Conversion =
+59.85
+3.58 * A
+0.25 * B
+2.23 * C
+0.83 * A2
+0.075
+0.057
-0.39
-0.038
+0.31
* B2
* C2
*A*B
*A*C
*B*C
Final Equation in Terms of Actual Factors:
Conversion =
+59.84984
+3.58327 * Time
+0.25462 * Temperature
+2.22997 * Catalyst
+0.83491 * Time2
+0.074772
* Temperature2
11-22
95% CI
Low
58.25
2.52
-0.81
1.17
-0.20
-0.96
-0.98
-1.78
-1.43
-1.08
95% CI
High
61.45
4.65
1.32
3.29
1.87
1.11
1.09
1.00
1.35
1.70
VIF
1.00
1.00
1.00
1.02
1.02
1.02
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+0.057094
-0.38750
-0.037500
+0.31250
* Catalyst2
* Time * Temperature
* Time * Catalyst
* Temperature * Catalyst
Because many of the terms are insignificant, the reduced quadratic model is fit as follows:
Design Expert Output
Response:
Activity
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
253.20
3
84.40
A
175.35
1
175.35
C
67.91
1
67.91
A2
9.94
1
9.94
Residual
34.07
16
2.13
Lack of Fit
30.42
11
2.77
Pure Error
3.65
5
0.73
Cor Total
287.28
19
F
Value
39.63
82.34
31.89
4.67
Prob > F
< 0.0001
< 0.0001
< 0.0001
0.0463
3.78
0.0766
significant
not significant
The Model F-value of 39.63 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.46
60.51
2.41
106.24
Factor
Intercept
A-Time
C-Catalyst
A2
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
59.95
3.58
2.23
0.82
DF
1
1
1
1
0.8814
0.8591
0.6302
20.447
Standard
Error
0.42
0.39
0.39
0.38
Final Equation in Terms of Coded Factors:
Activity
+59.95
+3.58
+2.23
+0.82
=
*A
*C
* A2
Final Equation in Terms of Actual Factors:
Activity
+59.94802
+3.58327
+2.22997
+0.82300
=
* Time
* Catalyst
* Time2
11-23
95% CI
Low
59.06
2.75
1.39
0.015
95% CI
High
60.83
4.42
3.07
1.63
VIF
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
C o nve rsio n
DE S IG N-E X P E RT P l o t
1.00
Co n ve rsi o n
X = A : T im e
Y = C : Ca ta l yst
92
76
90
88
78
1.00
A c ti v i ty
X = A : T im e
Y = C : Ca ta l y st
86
De si g n P o i n ts
0.50
0.50
C : C ata lys t
C : C ata lys t
De si g n P o i n ts
64
A c tu a l Fa c to r
B : T e m p e ra tu re = -1 .0 0
82
0.00
66
84
A ctu a l Fa cto r
B : T e m p e ra tu re = -1 .0 0
A ctivity
DE S IG N-E X P E RT P l o t
74
80
78
60
62
0.00
0.50
0.00
58
76
74
-0.50
-0.50
72
70
68
66
6462
56
60 58
5 65 4
-1.00
-1.00
-1.00
-0.50
0.00
0.50
1.00
-1.00
A: Tim e
-0.50
1.00
A: Tim e
Overla y P lot
DE S IG N-E X P E RT P l o t
1.00
O ve rl a y P l o t
X = A : T im e
Y = C : Ca ta l yst
De si g n P o i n ts
0.50
C : C ata lys t
A ctu a l Fa cto r
B : T e m p e ra tu re = -1 .0 0
C o n v ers io n: 8 2
A c t iv it y : 6 0
0.00
-0.50
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: Tim e
The contour plots visually describe the models while the overlay plots identifies the acceptable region for
the process.
11-24
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
11-13 A manufacturer of cutting tools has developed two empirical equations for tool life in hours (y1)
and for tool cost in dollars (y2). Both models are linear functions of steel hardness (x1) and
manufacturing time (x2). The two equations are
ŷ1 10 5 x1 2 x 2
ŷ 2 23 3x1 4 x 2
and both equations are valid over the range -1.5dx1d1.5. Unit tool cost must be below $27.50 and life
must exceed 12 hours for the product to be competitive. Is there a feasible set of operating conditions for
this process? Where would you recommend that the process be run?
The contour plots below graphically describe the two models. The overlay plot identifies the feasible
operating region for the process.
L ife
1.50
C o st
1.50
20
32
30
18
0.75
28
2 7. 5
0.75
26
16
0.00
8
10
12
B: Tim e
B: Tim e
24
6
14
0.00
22
4
20
-0.75
-0.75
2
18
16
14
-1.50
-1.50
-1.50
-0.75
0.00
0.75
1.50
-1.50
A: H ardn es s
C o s t : 2 7. 5
B: Tim e
0.75
L if e : 12
0.00
-0.75
-1.50
-1.50
-0.75
0.00
0.00
A: H ardn es s
Overla y P lot
1.50
-0.75
0.75
1.50
A: H ardn es s
10 5 x1 2 x 2 t 12
23 3x1 4 x 2 d 27.50
11-26
0.75
1.50
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
11-14 A central composite design is run in a chemical vapor deposition process, resulting in the
experimental data shown below. Four experimental units were processed simultaneously on each run of
the design, and the responses are the mean and variance of thickness, computed across the four units.
x1
x2
y
s2
-1
-1
1
1
1.414
-1.414
0
0
0
0
0
0
-1
1
-1
1
0
0
1.414
-1.414
0
0
0
0
360.6
445.2
412.1
601.7
518.0
411.4
497.6
397.6
530.6
495.4
510.2
487.3
6.689
14.230
7.088
8.586
13.130
6.644
7.649
11.740
7.836
9.306
7.956
9.127
(a) Fit a model to the mean response. Analyze the residuals.
Design Expert Output
Response:
Mean Thick
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
47644.26
5
9528.85
A
22573.36
1
22573.36
B
15261.91
1
15261.91
A2
2795.58
1
2795.58
B2
AB
Residual
Lack of Fit
Pure Error
Cor Total
5550.74
2756.25
3546.83
2462.04
1084.79
51191.09
1
1
6
3
3
11
F
Value
16.12
38.19
25.82
Prob > F
0.0020
0.0008
0.0023
4.73
0.0726
9.39
4.66
0.0221
0.0741
2.27
0.2592
5550.74
2756.25
591.14
820.68
361.60
significant
not significant
The Model F-value of 16.12 implies the model is significant. There is only
a 0.20% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
24.31
472.31
5.15
19436.37
Factor
Intercept
A-x1
B-x2
A2
B2
AB
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
505.88
53.12
43.68
DF
1
1
1
0.9307
0.8730
0.6203
11.261
Standard
Error
12.16
8.60
8.60
95% CI
Low
476.13
32.09
22.64
95% CI
High
535.62
74.15
64.71
VIF
1.00
1.00
-20.90
1
9.61
-44.42
2.62
1.04
-29.45
26.25
1
1
9.61
12.16
-52.97
-3.50
-5.93
56.00
1.04
1.00
Final Equation in Terms of Coded Factors:
Mean Thick =
+505.88
+53.12 * A
+43.68 * B
11-27
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
-20.90
-29.45
+26.25
* A2
* B2
*A*B
Final Equation in Terms of Actual Factors:
Mean Thick =
+505.87500
+53.11940 * x1
+43.67767 * x2
-20.90000
-29.45000
+26.25000
* x12
* x22
* x1 * x2
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
24.725
99
12.4493
90
80
R es idua ls
N orm al % probability
95
70
50
30
0.173533
20
10
-12.1022
5
1
-24.3779
-24.3779
-12.1022
0.173533
12.4493
24.725
384.98
433.38
R es idua l
481.78
530.17
578.57
Predicted
A modest deviation from normality can be observed in the Normal Plot of Residuals; however, not enough
to be concerned.
(b) Fit a model to the variance response. Analyze the residuals.
Design Expert Output
Response:
Var Thick
ANOVA for Response Surface 2FI Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
65.80
3
A
41.46
1
B
15.21
1
AB
9.13
1
Residual
4.89
8
Lack of Fit
3.13
5
Pure Error
1.77
3
Cor Total
70.69
11
Mean
Square
21.93
41.46
15.21
9.13
0.61
0.63
0.59
F
Value
35.86
67.79
24.87
14.93
Prob > F
< 0.0001
< 0.0001
0.0011
0.0048
1.06
0.5137
The Model F-value of 35.86 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.78
9.17
8.53
7.64
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.9308
0.9048
0.8920
18.572
11-28
significant
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Intercept
A-x1
B-x2
AB
Coefficient
Estimate
9.17
2.28
-1.38
-1.51
DF
1
1
1
1
Standard
Error
0.23
0.28
0.28
0.39
95% CI
Low
8.64
1.64
-2.02
-2.41
95% CI
High
9.69
2.91
-0.74
-0.61
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Var Thick =
+9.17
+2.28 * A
-1.38 * B
-1.51 * A * B
Final Equation in Terms of Actual Factors:
Var Thick =
+9.16508
+2.27645 * x1
-1.37882 * x2
-1.51075 * x1 * x2
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0.745532
99
0.226878
90
80
R es idua ls
N orm al % probability
95
70
50
30
-0.291776
20
10
-0.810429
5
1
-1.32908
-1.32908
-0.810429
-0.291776
0.226878
0.745532
5.95
8.04
R es idua l
10.14
12.23
14.33
Predicted
The residual plots are not acceptable. A transformation should be considered. If not successful at
correcting the residual plots, further investigation into the two apparently unusual points should be made.
(c) Fit a model to the ln(s2). Is this model superior to the one you found in part (b)?
Design Expert Output
Response:
Var Thick
Transform: Natural log
ANOVA for Response Surface 2FI Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
0.67
3
0.22
A
0.46
1
0.46
B
0.14
1
0.14
AB
0.079
1
0.079
Residual
0.049
8
6.081E-003
Lack of Fit
0.024
5
4.887E-003
Pure Error
0.024
3
8.071E-003
Cor Total
0.72
11
11-29
Constant:
0
F
Value
36.94
74.99
22.80
13.04
Prob > F
< 0.0001
< 0.0001
0.0014
0.0069
0.61
0.7093
significant
not significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The Model F-value of 36.94 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.078
2.18
3.57
0.087
Factor
Intercept
A-x1
B-x2
AB
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
2.18
0.24
-0.13
-0.14
DF
1
1
1
1
0.9327
0.9074
0.8797
18.854
Standard
Error
0.023
0.028
0.028
0.039
95% CI
Low
2.13
0.18
-0.20
-0.23
95% CI
High
2.24
0.30
-0.068
-0.051
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Ln(Var Thick) =
+2.18
+0.24 * A
-0.13 * B
-0.14 * A * B
Final Equation in Terms of Actual Factors:
Ln(Var Thick) =
+2.18376
+0.23874 * x1
-0.13165 * x2
-0.14079 * x1 * x2
No rm a l p lot o f re sid uals
Re sid ua ls vs. P re d icted
0.0930684
99
0.0385439
90
80
R es idua ls
N orm al % probability
95
70
50
-0.0159805
30
20
10
-0.070505
5
1
-0.125029
-0.125029
-0.070505
-0.0159805
0.0385439
0.0930684
1.85
R es idua l
2.06
2.27
2.48
2.69
Predicted
The residual plots are much improved following the natural log transformation; however, the two runs
still appear to be somewhat unusual and should be investigated further. They will be retained in the
analysis.
(d) Suppose you want the mean thickness to be in the interval 450±25. Find a set of operating conditions
that achieve the objective and simultaneously minimize the variance.
11-30
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
M e a n Thick
1.00
L n(V a r Thick)
1.00
5 75
5 50
0.50
0.50
2 .1
0.00
B: x2
B: x2
5 25
4
5 00
2 .2
2
4
0.00
2 .3
4 75
2 .4
4 50
-0.50
-0.50
2 .5
4 25
2 .6
4 00
-1.00
-1.00
-1.00
-0.50
0.00
0.50
1.00
-1.00
-0.50
0.00
A: x1
0.50
1.00
A: x1
Overla y P lot
1.00
B: x2
0.50
L n(V ar Thic k ): 2 . 00 0
4
0.00
Me an Th ic k : 47 5
-0.50
Me an Th ic k : 42 5
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: x1
The contour plots describe the two models while the overlay plot identifies the acceptable region for the
process.
(e) Discuss the variance minimization aspects of part (d). Have you minimized total process variance?
The within run variance has been minimized; however, the run-to-run variation has not been minimized
in the analysis. This may not be the most robust operating conditions for the process.
11-15
Verify that an orthogonal first-order design is also first-order rotatable.
To show that a first order orthogonal design is also first order rotatable, consider
V ( ŷ ) V ( Eˆ 0
k
¦
Eˆ i xi ) V ( Eˆ 0 )
i 1
k
¦ x V ( Eˆ )
i 1
11-31
2
i
i
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
since all covariances between Ei and E j are zero, due to design orthogonality. Furthermore, we have:
V Eˆ 0
V ( Eˆ 1 ) V ( Eˆ 2 ) ... V ( Eˆ k )
V2 V2
n
n
V ( ŷ )
2
V ( ŷ )
V
n
§ V
¨1
¨
n
©
2
k
V2
, so
n
¦x
2
i
i 1
k
¦x
i 1
2
i
·
¸
¸
¹
which is a function of distance from the design center (i.e. x=0), and not direction. Thus the design must
be rotatable. Note that n is, in general, the number of points in the exterior portion of the design. If there
V2
are nc centerpoints, then V ( Eˆ 0 )
.
( n nc )
11-16 Show that augmenting a 2k design with nc center points does not affect the estimates of the Ei (i=1,
2, . . . , k), but that the estimate of the intercept E0 is the average of all 2k + nc observations.
In general, the X matrix for the 2k design with nc center points and the y vector would be:
X
y
E1
E2
Ek º
ª E0
«
1
1
1
1 »»
«
« 1
1
1
1 »
«
»
»
«
« 1
1
1
1 »
«
»
« »
« 1
0
0
0 »
«
»
0
0
0 »
« 1
«
»
0
0
0 »
« 1
«
»
«
»
0
0
0 »¼
«¬ 1
ª y1 º
«
»
« y2 »
« »
«
»
« y 2k »
« » Å 2k+nc
«
» observations
« n01 »
«n » .
« 02 »
« »
«
»
«¬ n 0c »¼
Å The lower half of the matrix represents the
center points (nc rows)
ª2 nc
«
«
«
«
¬«
k
X' X
Å The upper half of the matrix is the usual r 1
notation of the 2k design
0
2
11-32
k
0º
»
0»
»
»
2k ¼»
X' y
ª g0 º
« »
« g1 »
«g 2 »
« »
« »
«g k »
¬ ¼
Å Grand total of
all 2k+nc
observations
Å usual contrasts
from 2k
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
gi
g0
, which is the average of all 2 k n c observations, while Eˆ i
, which does
2 nc
2k
not depend on the number of center points, since in computing the contrasts gi, all observations at the
center are multiplied by zero.
Therefore, Eˆ 0
k
11-17 The rotatable central composite design. It can be shown that a second-order design is rotatable if
¦
n
x a xb
u 1 iu ju
¦
0 if a or b (or both) are odd and if
composite design these conditions lead to D
1/ 4
nf
n
x4
u 1 iu
¦
3
n
x2 x 2 .
u 1 iu ju
Show that for the central
for rotatability, where nf is the number of points in
the factorial portion.
The balance between +1 and -1 in the factorial columns and the orthogonality among certain column in
the X matrix for the central composite design will result in all odd moments being zero. To solve for D
use the following relations:
n
¦
x iu4
n
¦x
n f 2D 4 ,
u 1
2 2
iu x ju
nf
u 1
then
¦
¦
n
x4
u 1 iu
3
n f 2D 4
2D 4
D
11-18
3( n f )
2n f
4
D
n
x2 x2
u 1 iu ju
nf
nf
4
Verify that the central composite design shown below blocks orthogonally.
x1
Block 1
x2
x3
x1
Block 2
x2
x3
x1
Block 3
x2
x3
0
0
1
1
-1
-1
0
0
1
-1
-1
1
0
0
1
-1
1
-1
0
0
1
1
-1
-1
0
0
1
-1
1
-1
0
0
-1
1
1
-1
-1.633
1.633
0
0
0
0
0
0
0
0
-1.633
1.633
0
0
0
0
0
0
0
0
-1.633
1.633
0
0
Note that each block is an orthogonal first order design, since the cross products of elements in different
columns add to zero for each block. To verify the second condition, choose a column, say column x2.
Now
k
¦x
2
2u
13.334 , and n=20
u 1
11-33
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
For blocks 1 and 2,
¦x
2
2m
4 , nm=6
m
So
¦x
m
n
2
2m
nm
¦
6
x 22u
u 1
4
13.334
6
20
0.3 = 0.3
and condition 2 is satisfied by blocks 1 and 2. For block 3, we have
¦x
2
2m
5.334 , nm = 8, so
m
¦x
m
n
2
2m
¦x
2
2u
nm
n
u 1
5.334
13.334
8
20
0.4 = 0.4
And condition 2 is satisfied by block 3. Similar results hold for the other columns.
11-19 Blocking in the central composite design. Consider a central composite design for k = 4 variables
in two blocks. Can a rotatable design always be found that blocks orthogonally?
To run a central composite design in two blocks, assign the nf factorial points and the n01 center points to
block 1 and the 2k axial points plus n02 center points to block 2. Both blocks will be orthogonal first order
designs, so the first condition for orthogonal blocking is satisfied.
The second condition implies that
¦x
2
im
block1
2
im
block 2
m
¦x
m
11-34
n f nc1
2k nc 2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
However,
¦x
2
im
n f in block 1 and
m
¦x
2
im
2D 2 in block 2, so
m
n f nc1
nf
2D
2
2k nc 2
Which gives:
1
D
Since D
4
ª n f 2k nc 2 º 2
«
»
¬« 2 n f nc1 »¼
n f if the design is to be rotatable, then the design must satisfy
nf
ª n f 2k n c 2 º
«
»
¬« 2 n f nc1 »¼
2
It is not possible to find rotatable central composite designs which block orthogonally for all k. For
example, if k=3, the above condition cannot be satisfied. For k=2, there must be an equal number of
center points in each block, i.e. nc1 = nc2. For k=4, we must have nc1 = 4 and nc2 = 2.
11-20 How could a hexagon design be run in two orthogonal blocks?
The hexagonal design can be blocked as shown below. There are nc1 = nc2 = nc center points with nc even.
1
2
n
6
5
3
4
Put the points 1,3,and 5 in block 1 and 2,4,and 6 in block 2. Note that each block is a simplex.
11-21 Yield during the first four cycles of a chemical process is shown in the following table. The
variables are percent concentration (x1) at levels 30, 31, and 32 and temperature (x2) at 140, 142, and
144qF. Analyze by EVOP methods.
Cycle
(1)
(2)
Conditions
(3)
11-35
(4)
(5)
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
1
2
3
4
60.7
59.1
56.6
60.5
59.8
62.8
59.1
59.8
60.2
62.5
59.0
64.5
64.2
64.6
62.3
61.0
57.5
58.3
61.1
60.1
Cycle: n=1 Phase 1
Calculation of Averages
Operating Conditions
(i)
Previous Cycle Sum
(ii)
Previous Cycle Average
(iii)
New Observation
(iv)
Differences
(v)
New Sums
(vi)
New Averages
Calculation of Standard Deviation
(1)
(2)
(3)
(4)
60.7
59.8
60.2
64.2
57.5
60.7
60.7
59.8
59.8
60.2
60.2
64.2
64.2
57.5
57.5
Calculation of Effects
CIM
(5)
Previous Sum S=
Previous Average =
New S=Range x fk,n
Range=
New Sum S=
New average S = New Sum S/(n-1)=
Calculation of Error Limits
A
1
y 3 y 4 y 2 y5
2
3.55
B
1
y 3 y 4 y 2 y5
2
-3.55
AB
1
y3 y 4 y 2 y5
2
-0.85
1
y 3 y 4 y 2 y 5 4 y1
2
-0.22
§ 2 ·
¸S
¸
© n¹
§ 2 ·
¸S
For New Effects: ¨¨
¸
© n¹
§ 1.78 ·
¸S
For CIM: ¨¨
¸
© n ¹
For New Average: ¨¨
Cycle: n=2 Phase 1
Calculation of Averages
Operating Conditions
(i)
Previous Cycle Sum
(ii)
Previous Cycle Average
(iii)
New Observation
(iv)
Differences
(v)
New Sums
(vi)
New Averages
Calculation of Standard Deviation
(1)
(2)
(3)
(4)
(5)
60.7
60.7
59.1
1.6
119.8
59.90
59.8
59.8
62.8
-3.0
122.6
61.30
60.2
60.2
62.5
-2.3
122.7
61.35
64.2
64.2
64.6
-0.4
128.8
64.40
57.5
57.5
58.3
-0.8
115.8
57.90
Calculation of Effects
CIM
Previous Sum S=
Previous Average =
New S=Range x fk,n=1.38
Range=4.6
New Sum S=1.38
New average S = New Sum S/(n-1)=1.38
Calculation of Error Limits
A
1
y 3 y 4 y 2 y5
2
3.28
B
1
y 3 y 4 y 2 y5
2
-3.23
AB
1
y3 y 4 y 2 y5
2
0.18
1
y 3 y 4 y 2 y 5 4 y1
2
1.07
§ 2 ·
¸S
¸
© n¹
§ 2 ·
¸S
For New Effects: ¨¨
¸
© n¹
§ 1.78 ·
¸S
For CIM: ¨¨
¸
© n ¹
For New Average: ¨¨
1.95
1.95
1.74
Cycle: n=3 Phase 1
Calculation of Averages
Operating Conditions
(i)
Previous Cycle Sum
Calculation of Standard Deviation
(1)
(2)
(3)
(4)
(5)
119.8
122.6
122.7
128.8
115.8
11-36
Previous Sum S=1.38
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(ii)
(iii)
(iv)
(v)
(vi)
Previous Cycle Average
New Observation
Differences
New Sums
New Averages
59.90
56.6
3.30
176.4
58.80
61.30
59.1
2.20
181.7
60.57
61.35
59.0
2.35
181.7
60.57
64.40
62.3
2.10
191.1
63.70
Calculation of Effects
CIM
57.90
61.1
-3.20
176.9
58.97
Previous Average =1.38
New S=Range x fk,n=2.28
Range=6.5
New Sum S=3.66
New average S = New Sum S/(n-1)=1.38
Calculation of Error Limits
A
1
y 3 y 4 y 2 y5
2
2.37
B
1
y 3 y 4 y 2 y5
2
-2.37
AB
1
y3 y 4 y 2 y5
2
-0.77
1
y 3 y 4 y 2 y 5 4 y1
2
1.72
§ 2 ·
¸S
¸
© n¹
§ 2 ·
¸S
For New Effects: ¨¨
¸
© n¹
§ 1.78 ·
¸S
For CIM: ¨¨
¸
© n ¹
For New Average: ¨¨
2.11
2.11
1.74
Cycle: n=4 Phase 1
Calculation of Averages
Operating Conditions
(i)
Previous Cycle Sum
(ii)
Previous Cycle Average
(iii)
New Observation
(iv)
Differences
(v)
New Sums
(vi)
New Averages
Calculation of Standard Deviation
(1)
(2)
(3)
(4)
(5)
176.4
58.80
60.5
-1.70
236.9
59.23
181.7
60.57
59.8
0.77
241.5
60.38
181.7
60.57
64.5
-3.93
245.2
61.55
191.1
63.70
61.0
2.70
252.1
63.03
176.9
58.97
60.1
-1.13
237.0
59.25
Calculation of Effects
CIM
Previous Sum S=3.66
Previous Average =1.83
New S=Range x fk,n=2.45
Range=6.63
New Sum S=6.11
New average S = New Sum S/(n-1)=2.04
Calculation of Error Limits
A
1
y 3 y 4 y 2 y5
2
2.48
B
1
y 3 y 4 y 2 y5
2
-1.31
AB
1
y3 y 4 y 2 y5
2
-0.18
1
y 3 y 4 y 2 y 5 4 y1
2
1.46
§ 2 ·
¸S
¸
© n¹
§ 2 ·
¸S
For New Effects: ¨¨
¸
© n¹
§ 1.78 ·
¸S
For CIM: ¨¨
¸
© n ¹
For New Average: ¨¨
2.04
2.04
1.82
From studying cycles 3 and 4, it is apparent that A (and possibly B) has a significant effect. A new phase
should be started following cycle 3 or 4.
11-22 Suppose that we approximate a response surface with a model of order d1, such as y=X1E1+HH, when
the true surface is described by a model of order d2>d1; that is E(y)= X1E1+ X2E2.
E1+AE
E2, where A=(X’1X1)(a) Show that the regression coefficients are biased, that is, that E( E 1 )=E
1 ’
X 1X2. A is usually called the alias matrix.
11-37
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
> @
ˆ
Eβ
1
E ª X1' X1
«¬
1
X1' y º
»¼
X1' X1
1
X1' E >y @
X1' X1
1
X1' X1β1 X 2 β 2
1
X1' X1 X1' X1β1 X1' X1
β1 Aβ 2
where A
X '1 X1
1
1
X1' X 2 β 2
X '1 X 2
(a) If d1=1 and d2=2, and a full 2k is used to fit the model, use the result in part (a) to determine the alias
structure.
In this situation, we have assumed the true surface to be first order, when it is really second order. If a full
factorial is used for k=2, then
E 0 E1 E 2
ª1 1 1º
X1 = «1 1 1 »
«
»
«1 1 1»
«
»
1¼
¬1 1
E11
ª1
X2 = «1
«
«1
«
¬1
ª Eˆ 0 º
« »
Then, E E 1 = E « Eˆ 1 »
« Eˆ »
¬ 2¼
> @
E 22
1
1
1
1
E 12
1º
1»» and
1»
»
1¼
ª 1 1 0º
A = « 0 0 0»
«
»
«¬ 0 0 0»¼
ª E 0 º ª1 1 0º ª E 11 º
« » «0 0 0» «
»
« E1 » «
» « E 22 »
«¬ E 2 »¼ «¬0 0 0»¼ «¬ E 12 »¼
ª E 0 E 11 E 22 º
«
»
E1
«
»
«¬
»¼
E2
The pure quadratic terms bias the intercept.
(b) If d1=1, d2=2 and k=3, find the alias structure assuming that a 23-1 design is used to fit the model.
E 0 E1 E 2 E 3
ª1 1 1 1 º
X1 = «1 1 1 1»
«
»
«1 1 1 1»
«
»
1
1¼
¬1 1
> @
Then, E E 1
ª Eˆ 0 º
«ˆ »
E
= E« 1»
« Eˆ »
« 2»
«¬ Eˆ 3 »¼
E11
ª1
X2 = «1
«
«1
«
¬1
E 22
1
1
1
1
ª E 0 º ª1
« E » «0
« 1»«
« E 2 » «0
« » «
¬ E 3 ¼ ¬0
E 33 E12 E13
1 1 1
1 1 1
1 1 1
1 1 1
1
0
0
0
0
0
0
0
0
0
0
1
0
0
1
0
E 23
1º
1»
»
1»
»
1¼
ª1
«0
and A = «
«0
«
¬0
ª E 11 º
0º «« E 22 »»
1»» « E 33 »
«
»
0» « E 12 »
»
0¼ « E 13 »
«
»
¬« E 23 ¼»
1
0
0
0
0
0
0
0
0
0
0
1
0
0
1
0
0º
1»
»
0»
»
0¼
ª E 0 E 11 E 22 E 22 º
«
»
E1 E 23
«
»
«
»
E 2 E 13
«
»
E 3 E12
¬
¼
(d) If d1=1, d2=2, k=3, and the simplex design in Problem 11-3 is used to fit the model, determine the
alias structure and compare the results with part (c).
11-38
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
E0
E1 E 2
E11 E 22 E 33 E12 E13 E 23
ª0 2 1 0 0 2 º
»
X2 = « 2 0 1 0 2 0 » and A =
«
«0 2 1 0 0 2 »
»
«
¬2 0 1 0 2 0 ¼
E3
ª1 1 1 1 º
X1 = «1 1 1 1»
«
»
«1 1 1 1»
«
»
1
1¼
¬1 1
> @
Then, E E 1
ª Eˆ 0 º
«ˆ »
E
= E « 1»
« Eˆ »
« 2»
«¬ Eˆ 3 »¼
ª E 0 º ª1 1
«E » «
« 1 » «0 0
« E 2 » «0 0
« » «
¬ E 3 ¼ ¬1 1
1
0
0
0
0
0
0
0
0
1
0
0
ª E 11 º
«
»
0 º « E 22 »
»
0 » « E 33 »
«
»
1» « E 12 »
»«
0 ¼ E 13 »
«
»
¬« E 23 ¼»
ª1 1
«0 0
«
«0 0
«
¬1 1
1
0
0
0
0
0
0
0
0 0º
1 0»
»
0 1»
0 0 »¼
ª E 0 E 11 E 22 E 22 º
«
»
E 1 E 13
«
»
«
»
E 2 E 23
«
»
¬ E 3 E 11 E 22 ¼
Notice that the alias structure is different from that found in the previous part for the 23-1 design. In
general, the A matrix will depend on which simplex design is used.
11-23 In an article (“Let’s All Beware the Latin Square,” Quality Engineering, Vol. 1, 1989, pp. 453465) J.S. Hunter illustrates some of the problems associated with 3k-p fractional factorial designs. Factor A
is the amount of ethanol added to a standard fuel and factor B represents the air/fuel ratio. The response
variable is carbon monoxide (CO) emission in g/m2. The design is shown below.
Design
A
0
1
2
0
1
2
0
1
2
B
0
0
0
1
1
1
2
2
2
Observations
x1
-1
0
1
-1
0
1
-1
0
1
x2
-1
-1
-1
0
0
0
1
1
1
y
66
78
90
72
80
75
68
66
60
y
62
81
94
67
81
78
66
69
58
Notice that we have used the notation system of 0, 1, and 2 to represent the low, medium, and high levels
for the factors. We have also used a “geometric notation” of -1, 0, and 1. Each run in the design is
replicated twice.
(a) Verify that the second-order model
ŷ
78.5 4.5 x1 7.0 x2 4.5 x12 4.0 x22 9.0 x1x2
is a reasonable model for this experiment. Sketch the CO concentration contours in the x1, x2 space.
In the computer output that follows, the “coded factors” model is in the -1, 0, +1 scale.
Design Expert Output
Response:
CO Emis
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1624.00
5
324.80
A
243.00
1
243.00
11-39
F
Value
50.95
38.12
Prob > F
< 0.0001
< 0.0001
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B
A2
B2
AB
Residual
Lack of Fit
Pure Error
Cor Total
588.00
1
588.00
92.24
< 0.0001
81.00
1
81.00
12.71
0.0039
64.00
648.00
76.50
30.00
46.50
1700.50
1
1
12
3
9
17
64.00
648.00
6.37
10.00
5.17
10.04
101.65
0.0081
< 0.0001
1.94
0.1944
not significant
The Model F-value of 50.95 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.52
72.83
3.47
169.71
Factor
Intercept
A-Ethanol
B-Air/Fuel Ratio
A2
B2
AB
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
78.50
4.50
-7.00
0.9550
0.9363
0.9002
21.952
Standard
Error
1.33
0.73
0.73
DF
1
1
1
95% CI
Low
75.60
2.91
-8.59
95% CI
High
81.40
6.09
-5.41
VIF
1.00
1.00
-4.50
1
1.26
-7.25
-1.75
1.00
-4.00
-9.00
1
1
1.26
0.89
-6.75
-10.94
-1.25
-7.06
1.00
1.00
Final Equation in Terms of Coded Factors:
CO Emis =
+78.50
+4.50 * A
-7.00 * B
-4.50 * A2
-4.00
-9.00
* B2
*A*B
CO E
2 m is
2
1.00
2
65
70
B: Air/Fuel R atio
0.50
75
2
2
2
0.00
80
-0.50
85
26 5
2
2
-1.00
-1
-0.5
0
0.5
1
A: Eth anol
(b) Now suppose that instead of only two factors, we had used four factors in a 34-2 fractional factorial
design and obtained exactly the same data in part (a). The design would be as follows:
A
B
C
Design
D
x1
x2
11-40
x3
x4
Observations
y
y
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0
1
2
0
1
2
0
1
2
0
0
0
1
1
1
2
2
2
0
1
2
2
0
1
1
2
0
0
1
2
1
2
0
2
0
1
-1
0
+1
-1
0
+1
-1
0
+1
-1
-1
-1
0
0
0
+1
+1
+1
-1
0
+1
+1
-1
0
0
+1
-1
-1
0
+1
0
+1
-1
+1
-1
0
66
78
90
72
80
75
68
66
60
62
81
94
67
81
78
66
69
58
Confirm that this design is an L9 orthogonal array.
This is the same as the design in Table 11-22.
(c) Calculate the marginal averages of the CO response at each level of the four factors A, B, C, and D.
Construct plots of these marginal averages and interpret the results. Do factors C and D appear to
have strong effects? Do these factors really have any effect on CO emission? Why is their apparent
effect strong?
One F a ctor P lot
94
94
85
85
C O Em is
C O Em is
One F a ctor P lot
76
76
67
67
58
58
0
1
2
0
A: A
1
B: B
11-41
2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
One F a ctor P lot
94
85
85
C O Em is
C O Em is
One F a ctor P lot
94
76
76
67
67
58
58
0
1
2
0
1
C: C
2
D: D
Both Factors C and D appear to have an effect on CO emission. This is probably because both C and D
are aliased with components of interaction involving A and B, and there is a strong AB interaction.
(a) The design in part (b) allows the model
E0
y
4
¦
Ei xi
i 1
4
¦E x
2
ii i
H
i 1
to be fitted. Suppose that the true model is
y
E0
4
4
¦E x ¦E
i i
i 1
2
ii x i
i 1
¦¦ E
ij x i x j
H
i j
Show that if Ej represents the least squares estimates of the coefficients in the fitted model, then
E Eˆ 0
E Eˆ
1
E Eˆ 2
E Eˆ 3
E Eˆ
4
E0 E13 E14 E34
E1 E 23 E 24 / 2
E 2 E13 E14 E34 / 2
E3 E12 E 24 / 2
E 4 E12 E 23 / 2
E Eˆ11
E Eˆ 22
E Eˆ
E11 E 23 E 24 / 2
33
E33 E 24 E12 / 2 E14
E Eˆ 44
E 44 E12 E 23 / 2 E13
E 22 E13 E14 E34 / 2
11-42
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
E 0 E1
ª1 1
«1 0
«
«1 1
«
1 1
Let X1 = «
«1 0
«
«1 1
«1 1
«
«1 0
«1 1
¬
Then, A
ª Eˆ 0 º
« ˆ »
« E1 »
« Eˆ »
« 2»
« Eˆ 3 »
E « Eˆ 4 »
«
»
« Eˆ 11 »
«ˆ »
« E 22 »
« Eˆ 33 »
«ˆ »
«¬ E 44 »¼
E2
1
1
1
0
0
0
1
1
1
E3
1
0
1
1
1
0
1
1
0
E 4 E11
1 1
0 0
1 1
0 1
1 0
1 1
1 1
1 0
0 1
E 22
1
1
1
0
0
0
1
1
1
E 33 E 44
1 1º
0 0»»
1 1»
»
1 0»
1 1»
»
0 1»
1 1»
»
1 1»
0 0»¼
E12 E13
ª1 1
«0 0
«
« 1 1
«
0 1
and X2 = «
«0 0
«
«0 0
« 1 0
«
«0 0
«1 1
¬
E14 E 23 E 24
1 1 1
0 0 0
1 1 1
0 0 0
0 0 0
1 0 0
1 0 1
0 1 1
0 1 0
E 34
1º
0 »»
1»
»
0»
1»
»
0»
0»
»
1»
0 »¼
0
0
1
1
1 º
ª 0
« 0
0
0
1 2 1 2
0 »»
«
« 0
1 2 1 2
0
0
1 2»
«
»
0
0
0
0 »
1 2
« 1 2
1
X1' X1 X1' X 2 = A = « 1 2
0
0
1 2
0
0 »
«
»
0
0
1 2 1 2
0 »
« 0
« 0
12
12
0
0
12»
«
»
0
1
0
1 2
0 »
«12
«
»
1
0
12
0
0 ¼
¬ 1 2
E 0 E 13 E 14 E 34
1
1
1 º
0
0
ª E0 º ª 0
ª
º
«
» «
«
»
»
1 2 1 2
E 1 1 2 E 23 1 2 E 24
0
0
0 »
« E1 » « 0
»
ª E12 º «
« E2 » « 0
« E 2 1 2 E 13 1 2 E 14 1 2 E 34 »
1 2 1 2
1 2» «
0
0
»
«
» «
» E
«
»
1 2
E 3 1 2 E 12 1 2 E 24
0
0
0
0 » « 13 » «
»
« E 3 » « 1 2
« E14 »
« E » « 1 2
»
E 4 1 2 E 12 1 2 E 23
1 2
0
0
0
0 »«
» «
4
«
» «
» « E 23 » «
»
1 2 1 2
E 11 1 2 E 23 1 2 E 24
0
0
0 »«
« E 11 » « 0
«
»
»
E
«E » « 0
» « 24 » « E 1 2 E 1 2 E 1 2 E »
1
2
1
2
0
0
1
2
13
14
34 »
« 22 » «
» « E » « 22
« E 33 » « 1 2
1 2
0
1
0
0 » ¬ 34 ¼ « E 33 1 2E 12 E 14 1 2 E 24 »
«
»
«
» «
»
1
0
12
0
0 ¼
¬ E 44 ¼ ¬ 1 2
¬ E 44 1 2 E 12 E 13 1 2 E 23 ¼
11-24 Suppose that you need to design an experiment to fit a quadratic model over the region
1 d x i d 1 , i=1,2 subject to the constraint x1 x 2 d 1 . If the constraint is violated, the process will
not work properly. You can afford to make no more than n=12 runs. Set up the following designs:
(a) An “inscribed” CCD with center points at x1
x2
0
x1
x2
-0.5
-0.5
0.5
-0.5
-0.5
0.5
0.5
0.5
-0.707
0
0.707
0
11-43
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0 -0.707
(a)* An “inscribed” CCD with center points at x1
the constrained region
0
0.707
0
0
0
0
0
0
0
0
0.25 so that a larger design could be fit within
x2
x1
x2
-1
-1
0.5
-1
-1
0.5
0.5
0.5
-1.664
-0.25
1.164
-0.25
-0.25 -1.664
-0.25
1.164
-0.25
-0.25
-0.25
-0.25
-0.25
-0.25
-0.25
-0.25
(a) An “inscribed” 32 factorial with center points at x1
x 2 0.25
x1
x2
-1
-1
-0.25
-1
0.5
-1
-1 -0.25
-0.25 -0.25
0.5 -0.25
-1
0.5
-0.25
0.5
0.5
0.5
-0.25 -0.25
-0.25 -0.25
-0.25 -0.25
(a) A D-optimal design.
x1 x2
-1
-1
1
-1
-1
1
1
0
0
1
0
0
-1
0
0
-1
11-44
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
0.5
0.5
-1
-1
1
-1
-1
1
(a) A modified D-optimal design that is identical to the one in part (c), but with all replicate runs at the
design center.
x1 x2
1
0
0
0
0
1
-1
-1
1
-1
-1
1
-1
0
0
-1
0.5
0.5
0
0
0
0
0
0
(a) Evaluate the ( X cX) 1 criteria for each design.
X cX
1
(a)
(a)*
(b)
(c)
(d)
0.5
0.00005248
0.007217
0.0001016
0.0002294
(a) Evaluate the D-efficiency for each design relative to the D-optimal design in part (c).
D-efficiency
(a)
(a)*
(b)
(c)
(d)
24.25%
111.64%
49.14%
100.00%
87.31%
(a) Which design would you prefer? Why?
The offset CCD, (a)*, is the preferred design based on the D-efficiency. Not only is it better than the Doptimal design, (c), but it maintains the desirable design features of the CCD.
11-45
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
11-25 Consider a 23 design for fitting a first-order model.
(a) Evaluate the D-criterion ( X cX) 1 for this design.
( X cX) 1 = 2.441E-4
(b) Evaluate the A-criterion tr ( X cX) 1 for this design.
tr ( X cX) 1 = 0.5
(c) Find the maximum scaled prediction variance for this design. Is this design G-optimal?
vx
NVar ŷ x
V2
Nx c 1 X cX
1
x1
4 . Yes, this is a G-optimal design.
11-26 Repeat Problem 11-25 using a first order model with the two-factor interaction.
( X cX) 1 = 4.768E-7
tr ( X cX) 1 = 0.875
vx
NVar ŷ x
V2
Nx c 1 X cX
1
x1
7 . Yes, this is a G-optimal design.
11-27 A chemical engineer wishes to fit a calibration curve for a new procedure used to measure the
concentration of a particular ingredient in a product manufactured in his facility. Twelve samples can be
prepared, having known concentration. The engineer’s interest is in building a model for the measured
concentrations. He suspects that a linear calibration curve will be adequate to model the measured
concentration as a function of the known concentrations; that is, where x is the actual concentration. Four
experimental designs are under consideration. Design 1 consists of 6 runs at known concentration 1 and 6
runs at known concentration 10. Design 2 consists of 4 runs at concentrations 1, 5.5, and 10. Design 3
consists of 3 runs at concentrations 1, 4, 7, and 10. Finally, design 4 consists of 3 runs at concentrations 1
and 10 and 6 runs at concentration 5.5.
(a) Plot the scaled variance of prediction for all four designs on the same graph over the concentration
range. Which design would be preferable, in your opinion?
11-46
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
S ca le d V a ria n c e o f P re dic tion
3.5
3
Design 4
Design 3
2.5
Design 2
2
Design 1
1.5
1
0.5
0
1
3
5
7
9
Because it has the lowest scaled variance of prediction at all points in the design space with the exception
of 5.5, Design 1 is preferred.
(b) For each design calculate the determinant of ( X cX) 1 . Which design would be preferred according
to the “D” criterion?
Design
( X cX) 1
1
2
3
4
0.000343
0.000514
0.000617
0.000686
Design 1 would be preferred.
(c) Calculate the D-efficiency of each design relative to the “best” design that you found in part b.
Design
1
2
3
4
D-efficiency
100.00%
81.65%
74.55%
70.71%
(a) For each design, calculate the average variance of prediction over the set of points given by x = 1, 1.5,
2, 2.5, . . ., 10. Which design would you prefer according to the V-criterion?
Average Variance of Prediction
Design
Actual
Coded
1
1.3704
0.1142
2
1.5556
0.1296
3
1.6664
0.1389
4
1.7407
0.1451
11-47
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design 1 is still preferred based on the V-criterion.
(e) Calculate the V-efficiency of each design relative to the best design you found in part (d).
Design
1
2
3
4
V-efficiency
100.00%
88.10%
82.24%
78.72%
(f) What is the G-efficiency of each design?
Design G-efficiency
1
100.00%
2
80.00%
3
71.40%
4
66.70%
11-28 Rework Problem 11-27 assuming that the model the engineer wishes to fit is a quadratic.
Obviously, only designs 2, 3, and 4 can now be considered.
S ca le d V a ria n c e o f P re dic tion
4.5
4
3.5
3
2.5
2
2
1.5
Design 4
1
Design 3
0.5
Design 2
0
1
3
5
7
9
Based on the plot, the preferred design would depend on the region of interest. Design 4 would be
preferred if the center of the region was of interest; otherwise, Design 2 would be preferred.
Design
2
3
4
( X cX) 1
4.704E-07
6.351E-07
5.575E-07
Design 2 is preferred based on ( X cX) 1 .
11-48
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design
2
3
4
D-efficiency
100.00%
90.46%
94.49%
Average Variance of Prediction
Design
Actual
Coded
2
2.441
0.2034
3
2.393
0.1994
4
2.242
0.1869
Design 4 is preferred.
Design
2
3
4
V-efficiency
91.89%
93.74%
100.00%
Design G-efficiency
2
100.00%
3
79.00%
4
75.00%
11-29 An experimenter wishes to run a three-component mixture experiment. The constraints are the
components proportions are as follows:
0.2 d x1 d 0.4
0.1 d x 2 d 0.3
0.4 d x 3 d 0.7
(a) Set up an experiment to fit a quadratic mixture model. Use n=14 runs, with 4 replicates. Use the Dcriteria.
Std
1
2
3
4
5
6
7
8
9
10
11
12
13
14
x1
0.2
0.3
0.3
0.2
0.4
0.4
0.2
0.275
0.35
0.3
0.2
0.3
0.2
0.4
(a) Draw the experimental design region.
11-49
x2
0.3
0.3
0.15
0.1
0.2
0.1
0.2
0.25
0.175
0.1
0.3
0.3
0.1
0.1
x3
0.5
0.4
0.55
0.7
0.4
0.5
0.6
0.475
0.475
0.6
0.5
0.4
0.7
0.5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A: x1
0.50
2
0.40
0.10
2
2
2
0.40
B: x2
0.20
0.70
C: x3
(c) Set up an experiment to fit a quadratic mixture model with n=12 runs, assuming that three of these
runs are replicated. Use the D-criterion.
Std
1
2
3
4
5
6
7
8
9
10
11
12
x1
0.3
0.2
0.3
0.2
0.4
0.4
0.2
0.275
0.35
0.2
0.4
0.4
x2
0.15
0.3
0.3
0.1
0.2
0.1
0.2
0.25
0.175
0.1
0.1
0.2
11-50
x3
0.55
0.5
0.4
0.7
0.4
0.5
0.6
0.475
0.475
0.7
0.5
0.4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A: x1
0.50
2
2
0.40
0.10
2
0.40
B: x2
0.20
0.70
C: x3
(d) Comment on the two designs you have found.
The design points are the same for both designs except that the edge center on the x1-x3 edge is not
included in the second design. None of the replicates for either design are in the center of the
experimental region. The experimental runs are fairly uniformly spaced in the design region.
11-30 Myers and Montgomery (1995) describe a gasoline blending experiment involving three mixture
components. There are no constraints on the mixture proportions, and the following 10 run design is
used.
Design Point
1
2
3
4
5
6
7
8
9
10
x2
0
1
0
½
0
½
1/3
1/6
2/3
1/6
x1
1
0
0
½
½
0
1/3
2/3
1/6
1/6
x3
0
0
1
0
½
½
1/3
1/6
1/6
2/3
y(mpg)
24.5, 25.1
24.8, 23.9
22.7, 23.6
25.1
24.3
23.5
24.8, 24.1
24.2
23.9
23.7
(a) What type of design did the experimenters use?
A simplex centroid design was used.
(b) Fit a quadratic mixture model to the data. Is this model adequate?
Design Expert Output
Response:
y
ANOVA for Mixture Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
11-51
F
Value
Prob > F
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Model
Linear Mixture
AB
AC
BC
Residual
Lack of Fit
Pure Error
Cor Total
4.22
3.92
0.15
0.081
0.077
1.73
0.50
1.24
5.95
5
2
1
1
1
8
4
4
13
0.84
1.96
0.15
0.081
0.077
0.22
0.12
0.31
3.90
9.06
0.69
0.38
0.36
0.0435
0.0088
0.4289
0.5569
0.5664
significant
0.40
0.8003
not significant
The Model F-value of 3.90 implies the model is significant. There is only
a 4.35% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.47
24.16
1.93
5.27
Component
A-x1
B-x2
C-x3
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
24.74
24.31
23.18
1.51
1.11
-1.09
DF
1
1
1
1
1
1
0.7091
0.5274
0.1144
5.674
Standard
Error
0.32
0.32
0.32
1.82
1.82
1.82
95% CI
Low
24.00
23.57
22.43
-2.68
-3.08
-5.28
95% CI
High
25.49
25.05
23.92
5.70
5.30
3.10
Final Equation in Terms of Pseudo Components:
y =
+24.74 * A
+24.31 * B
+23.18 * C
+1.51 * A * B
+1.11 * A * C
-1.09 * B * C
Final Equation in Terms of Real Components:
y =
+24.74432 * x1
+24.31098 * x2
+23.17765 * x3
+1.51364 * x1 * x2
+1.11364 * x1 * x3
-1.08636 * x2 * x3
The quadratic terms appear to be insignificant. The analysis below is for the linear mixture model:
Design Expert Output
Response:
y
ANOVA for Mixture Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
3.92
2
Linear Mixture
3.92
2
Residual
2.03
11
Lack of Fit
0.79
7
Pure Error
1.24
4
Cor Total
5.95
13
Mean
Square
1.96
1.96
0.18
0.11
0.31
F
Value
10.64
10.64
Prob > F
0.0027
0.0027
significant
0.37
0.8825
not significant
The Model F-value of 10.64 implies the model is significant. There is only
a 0.27% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.43
24.16
1.78
3.62
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.6591
0.5972
0.3926
8.751
11-52
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Component
A-x1
B-x2
C-x3
Coefficient
Estimate
24.93
24.35
23.19
DF
1
1
1
Standard
Error
0.25
0.25
0.25
95% CI
Low
24.38
23.80
22.64
95% CI
High
25.48
24.90
23.74
Component
A-x1
B-x2
C-x3
Adjusted
Effect
1.16
0.29
-1.45
DF
1
1
1
Adjusted
Std Error
0.33
0.33
0.33
Approx t for H0
Effect=0
3.49
0.87
-4.36
Prob > |t|
0.0051
0.4021
0.0011
Final Equation in Terms of Pseudo Components:
y =
+24.93 * A
+24.35 * B
+23.19 * C
Final Equation in Terms of Real Components:
y =
+24.93048 * x1
+24.35048 * x2
+23.19048 * x3
(c) Plot the response surface contours. What blend would you recommend to maximize the MPG?
A: x1
2
1.00
2 4. 8
2 4. 6
0.00
0.00
2 4. 4
2 4. 2
24
2 3. 8
2 3. 6
2 3. 4
2
1.00
B: x2
2
0.00
1.00
C : x3
y
To maximize the miles per gallon, the recommended blend is x1 = 1, x2 = 0, and x3 = 0.
11-31 Consider the bottle filling experiment in Example 6-1. Suppose that the percent carbonation (A) is
a noise variable (in coded units V z2 1 ).
(a) Fit the response model to these data. Is there a robust design problem?
From the analysis below, the AB interaction appears to have some importance. Because of this, there is
opportunity for improvement in the robustness of the process.
Design Expert Output
11-53
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Response:
Fill Height
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Source
Squares
DF
Model
73.00
7
A
36.00
1
B
20.25
1
C
12.25
1
AB
2.25
1
AC
0.25
1
BC
1.00
1
ABC
1.00
1
Pure Error
5.00
8
Cor Total
78.00
15
Mean
Square
10.43
36.00
20.25
12.25
2.25
0.25
1.00
1.00
0.63
F
Value
16.69
57.60
32.40
19.60
3.60
0.40
1.60
1.60
Prob > F
0.0003
< 0.0001
0.0005
0.0022
0.0943
0.5447
0.2415
0.2415
significant
The Model F-value of 16.69 implies the model is significant. There is only
a 0.03% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.79
1.00
79.06
20.00
Factor
Intercept
A-Carbination
B-Pressure
C-Speed
AB
AC
BC
ABC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
1.00
1.50
1.13
0.88
0.38
0.13
0.25
0.25
0.9359
0.8798
0.7436
13.416
Standard
Error
0.20
0.20
0.20
0.20
0.20
0.20
0.20
0.20
DF
1
1
1
1
1
1
1
1
95% CI
Low
0.54
1.04
0.67
0.42
-0.081
-0.33
-0.21
-0.21
Final Equation in Terms of Coded Factors:
Fill Height =
+1.00
+1.50 * A
+1.13 * B
+0.88 * C
+0.38 * A * B
+0.13 * A * C
+0.25 * B * C
+0.25 * A * B * C
Final Equation in Terms of Actual Factors:
Fill Height =
-225.50000
+21.00000 * Carbination
+7.80000 * Pressure
+1.08000 * Speed
-0.75000 * Carbination * Pressure
-0.10500 * Carbination * Speed
-0.040000 * Pressure * Speed
+4.00000E-003 * Carbination * Pressure * Speed
(b) Find the mean model and either the variance model or the POE.
The mean model in coded terms is:
E z >y x , z1
@
1.00 1.13B 0.88C 0.25BC
Contour plots of the mean model and POE are shown below:
11-54
95% CI
High
1.46
1.96
1.58
1.33
0.83
0.58
0.71
0.71
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
F ill He ig ht
DE S IG N-E X P E RT P l o t
1.000
1.000
Fi l l He i gh t
X = B : P re ssu re
Y = C: S p e e d
P O E (Fi l l H ei g h t)
X = B : P re ssu re
Y = C: S p e e d
2 .5
0.500
2
0 .5
0.500
2
1 .4
1 .5
0.000
2 .2
Co d e d Fa c to r
A : Ca rb i n a tio n = 0 .0 00
C : Spe ed
C : Spe ed
Co d e d Fa cto r
A : Ca rb i n a tio n = 0 .0 00
P OE (F ill He ig ht)
DE S IG N-E X P E RT P l o t
3
1
1 .8
0.000
1 .6
0
-0.500
-0.500
-0. 5
-1.000
-1.000
-1.000
-0.500
0.000
0.500
1.000
-1.000
B: Pres s ure
-0.500
0.000
0.500
1.000
B: Pres s ure
(c) Find a set of conditions that result in mean fill deviation as close to zero as possible with minimum
transmitted variability from carbonation.
The overlay plot below identifies a region that meets these requirements. The Pressure should be set at its
low level and the Speed should be set between approximately 0.0 and 0.5 in coded terms.
DE S IG N-E X P E RT P l o t
1.00
Overla y P lot
O ve rl a y P l o t
X = B : P re ssu re
Y = C: S p e e d
C : Spe ed
Co d e d Fa cto r
A : Ca rb i n a tio n = 0 .0 00
0.50
0.00
F ill H eig ht : 0 .2 5
-0.50
F ill H eig ht : -0. 2 5
-1.00
-1.00
-0.50
0.00
0.50
1.00
B: Pres s ure
11-32 Consider the experiment in Problem 11-12. Suppose that temperature is a noise variable ( V z2 1
in coded units). Fit response models for both responses. Is there a robust design problem with respect to
both responses? Find a set of conditions that maximize conversion with activity between 55 and 60, and
that minimize the variability transmitted from temperature.
The following is the analysis of variance for the Conversion response. Because of a significant BC
interaction, there is some opportunity for improvement in the robustness of the process with regards to
Conversion.
Design Expert Output
11-55
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Response:
Conversion
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
2555.73
9
283.97
A
14.44
1
14.44
B
222.96
1
222.96
C
525.64
1
525.64
A2
48.47
1
48.47
B2
124.48
1
124.48
C2
388.59
1
388.59
AB
36.13
1
36.13
AC
1035.13
1
1035.13
BC
120.12
1
120.12
Residual
222.47
10
22.25
Lack of Fit
56.47
5
11.29
Pure Error
166.00
5
33.20
Cor Total
287.28
19
F
Value
12.76
0.65
10.02
23.63
2.18
5.60
17.47
1.62
46.53
5.40
Prob > F
0.0002
0.4391
0.0101
0.0007
0.1707
0.0396
0.0019
0.2314
< 0.0001
0.0425
0.34
0.8692
significant
not significant
The Model F-value of 12.76 implies the model is significant. There is only
a 0.02% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
4.72
78.30
6.02
676.22
Factor
Intercept
A-Time
B-Temperature
C-Catalyst
A2
B2
C2
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
81.09
1.03
4.04
6.20
-1.83
2.94
-5.19
2.13
11.38
-3.87
DF
1
1
1
1
1
1
1
1
1
1
0.9199
0.8479
0.7566
14.239
Standard
Error
1.92
1.28
1.28
1.28
1.24
1.24
1.24
1.67
1.67
1.67
Final Equation in Terms of Coded Factors:
Conversion
+81.09
+1.03
+4.04
+6.20
-1.83
+2.94
-5.19
+2.13
+11.38
-3.87
=
*A
*B
*C
* A2
* B2
* C2
*A*B
*A*C
*B*C
Final Equation in Terms of Actual Factors:
Conversion
+81.09128
+1.02845
+4.04057
+6.20396
-1.83398
+2.93899
-5.19274
+2.12500
+11.37500
-3.87500
=
* Time
* Temperature
* Catalyst
* Time2
* Temperature2
* Catalyst2
* Time * Temperature
* Time * Catalyst
* Temperature * Catalyst
11-56
95% CI
Low
76.81
-1.82
1.20
3.36
-4.60
0.17
-7.96
-1.59
7.66
-7.59
95% CI
High
85.38
3.87
6.88
9.05
0.93
5.71
-2.42
5.84
15.09
-0.16
VIF
1.00
1.00
1.00
1.02
1.02
1.02
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The following is the analysis of variance for the Activity response. Because there is not a significant
interaction term involving temperature, there is no opportunity for improvement in the robustness of the
process with regards to Activity.
Design Expert Output
Response:
Activity
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
256.20
9
28.47
A
175.35
1
175.35
B
0.89
1
0.89
C
67.91
1
67.91
10.05
1
10.05
A2
0.081
1
0.081
B2
0.047
1
0.047
C2
AB
1.20
1
1.20
AC
0.011
1
0.011
BC
0.78
1
0.78
Residual
31.08
10
3.11
Lack of Fit
27.43
5
5.49
Pure Error
3.65
5
0.73
Cor Total
287.28
19
F
Value
Prob > F
9.16
0.0009
56.42
< 0.0001
0.28
0.6052
21.85
0.0009
3.23
0.1024
0.026
0.8753
0.015
0.9046
0.39
0.5480
3.620E-003 0.9532
0.25
0.6270
7.51
0.0226
significant
significant
The Model F-value of 9.16 implies the model is significant. There is only
a 0.09% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.76
60.51
2.91
214.43
Factor
Intercept
A-Time
B-Temperature
C-Catalyst
A2
B2
C2
AB
AC
BC
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
59.85
3.58
0.25
2.23
0.83
0.075
0.057
-0.39
-0.038
0.31
DF
1
1
1
1
1
1
1
1
1
1
0.8918
0.7945
0.2536
10.911
Standard
Error
0.72
0.48
0.48
0.48
0.46
0.46
0.46
0.62
0.62
0.62
Final Equation in Terms of Coded Factors:
Conversion =
+59.85
+3.58 * A
+0.25 * B
+2.23 * C
+0.83 * A2
+0.075 * B2
+0.057 * C2
-0.39 * A * B
-0.038 * A * C
+0.31 * B * C
Final Equation in Terms of Actual Factors:
Conversion =
+59.84984
+3.58327 * Time
+0.25462 * Temperature
+2.22997 * Catalyst
+0.83491 * Time2
11-57
95% CI
Low
58.25
2.52
-0.81
1.17
-0.20
-0.96
-0.98
-1.78
-1.43
-1.08
95% CI
High
61.45
4.65
1.32
3.29
1.87
1.11
1.09
1.00
1.35
1.70
VIF
1.00
1.00
1.00
1.02
1.02
1.02
1.00
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
+0.074772
+0.057094
-0.38750
-0.037500
+0.31250
* Temperature2
* Catalyst2
* Time * Temperature
* Time * Catalyst
* Temperature * Catalyst
Because many of the terms are insignificant, the reduced quadratic model is fit as follows:
Design Expert Output
Response:
Activity
ANOVA for Response Surface Quadratic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
253.20
3
84.40
A
175.35
1
175.35
C
67.91
1
67.91
A2
9.94
1
9.94
Residual
34.07
16
2.13
Lack of Fit
30.42
11
2.77
Pure Error
3.65
5
0.73
Cor Total
287.28
19
F
Value
39.63
82.34
31.89
4.67
Prob > F
< 0.0001
< 0.0001
< 0.0001
0.0463
3.78
0.0766
significant
not significant
The Model F-value of 39.63 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
1.46
60.51
2.41
106.24
Factor
Intercept
A-Time
C-Catalyst
A2
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
59.95
3.58
2.23
0.82
DF
1
1
1
1
0.8814
0.8591
0.6302
20.447
Standard
Error
0.42
0.39
0.39
0.38
95% CI
Low
59.06
2.75
1.39
0.015
95% CI
High
60.83
4.42
3.07
1.63
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Activity
+59.95
+3.58
+2.23
+0.82
=
*A
*C
* A2
Final Equation in Terms of Actual Factors:
Activity
+59.94802
+3.58327
+2.22997
+0.82300
=
* Time
* Catalyst
* Time2
Contour plots of the mean models for the responses along with POE for Conversion are shown below:
11-58
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
C o nve rsio n
DE S IG N-E X P E RT P l o t
1.00
Co n ve rsi o n
X = A : T im e
Y = C : Ca ta l yst
92
76
90
88
78
1.00
A c ti v i ty
X = A : T im e
Y = C : Ca ta l y st
86
De si g n P o i n ts
0.50
De si g n P o i n ts
0.50
82
80
0.00
64
A c tu a l Fa c to r
B : T e m p e ra tu re = -1 .0 0
C : C ata lys t
C : C ata lys t
66
84
A ctu a l Fa cto r
B : T e m p e ra tu re = -1 .0 0
A ctivity
DE S IG N-E X P E RT P l o t
74
78
60
62
0.00
0.50
0.00
58
76
74
-0.50
-0.50
72
70
68
66
6462
56
60 58
5 65 4
-1.00
-1.00
-1.00
-0.50
0.00
0.50
1.00
-1.00
A: Tim e
-0.50
A: Tim e
P OE (C o nve rsio n)
DE S IG N-E X P E RT P l o t
1.00
8.5
P O E (C o n ve rsi o n )
X = A : T im e
Y = C : Ca ta l yst
8
7 .5
7
6.5
De si g n P o i n ts
6
0.50
C : C ata lys t
A ctu a l Fa cto r
B : T e m p e ra tu re = -1 .0 0
5.5
0.00
5
-0.50
5
5.5
6
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: Tim e
The overlay plot shown below identifies a region near the center of the design space that meets the
constraints for the process.
11-59
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Overla y P lot
DE S IG N-E X P E RT P l o t
1.00
O ve rl a y P l o t
X = A : T im e
Y = B : T e m p e ra tu re
De si g n P o i n ts
B: Tem p era ture
0.50
A ctu a l Fa cto r
C: C a ta l yst = 0 .0 0
P O E (C on v e rs io n ): 8
C o n v ers io n: 8 2
6
A c t iv it y : 6 0
0.00
-0.50
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: Tim e
11-33 An experiment has been run in a process that applies a coating material to a wafer. Each run in the
experiment produced a wafer, and the coating thickness was measured several times at different locations
on the wafer. Then the mean y1, and standard deviation y2 of the thickness measurement was obtained.
The data [adapted from Box and Draper (1987)] are shown in the table below.
Run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
Speed
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
-1.000
0.000
1.000
Pressure
-1.000
-1.000
-1.000
0.000
0.000
0.000
1.000
1.000
1.000
-1.000
-1.000
-1.000
0.000
0.000
0.000
1.000
1.000
1.000
-1.000
-1.000
-1.000
0.000
0.000
0.000
1.000
1.000
1.000
Distance
-1.000
-1.000
-1.000
-1.000
-1.000
-1.000
-1.000
-1.000
-1.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
1.000
Mean (y1)
24.0
120.3
213.7
86.0
136.6
340.7
112.3
256.3
271.7
81.0
101.7
357.0
171.3
372.0
501.7
264.0
427.0
730.7
220.7
239.7
422.0
199.0
485.3
673.7
176.7
501.0
1010.0
Std Dev (y2)
12.5
8.4
42.8
3.5
80.4
16.2
27.6
4.6
23.6
0.0
17.7
32.9
15.0
0.0
92.5
63.5
88.6
21.1
133.8
23.5
18.5
29.4
44.7
158.2
55.5
138.9
142.4
(a) What type of design did the experimenters use? Is this a good choice of design for fitting a quadratic
model?
The design is a 33. A better choice would be a 23 central composite design. The CCD gives more
information over the design region with fewer points.
11-60
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(b) Build models of both responses.
The model for the mean is developed as follows:
Design Expert Output
Response:
Mean
ANOVA for Response Surface Reduced Cubic Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
1.289E+006
7
1.841E+005
A
5.640E+005
1
5.640E+005
B
2.155E+005
1
2.155E+005
C
3.111E+005
1
3.111E+005
AB
52324.81
1
52324.81
AC
68327.52
1
68327.52
BC
22794.08
1
22794.08
ABC
54830.16
1
54830.16
Residual
57874.57
19
3046.03
Cor Total
1.347E+006
26
F
Value
60.45
185.16
70.75
102.14
17.18
22.43
7.48
18.00
Prob > F
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.0006
0.0001
0.0131
0.0004
significant
The Model F-value of 60.45 implies the model is significant. There is only
a 0.01% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
55.19
314.67
17.54
1.271E+005
Factor
Intercept
A-Speed
B-Pressure
C-Distance
AB
AC
BC
ABC
Coefficient
Estimate
314.67
177.01
109.42
131.47
66.03
75.46
43.58
82.79
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
DF
1
1
1
1
1
1
1
1
0.9570
0.9412
0.9056
33.333
Standard
Error
10.62
13.01
13.01
13.01
15.93
15.93
15.93
19.51
95% CI
Low
292.44
149.78
82.19
104.24
32.69
42.11
10.24
41.95
95% CI
High
336.90
204.24
136.65
158.70
99.38
108.80
76.93
123.63
VIF
1.00
1.00
1.00
1.00
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Mean =
+314.67
+177.01 * A
+109.42 * B
+131.47 * C
+66.03 * A * B
+75.46 * A * C
+43.58 * B * C
+82.79 * A * B * C
Final Equation in Terms of Actual Factors:
Mean =
+314.67037
+177.01111 * Speed
+109.42222 * Pressure
+131.47222 * Distance
+66.03333 * Speed * Pressure
+75.45833 * Speed * Distance
+43.58333 * Pressure * Distance
+82.78750 * Speed * Pressure * Distance
The model for the Std. Dev. response is as follows. A square root transformation was applied to correct
problems with the normality assumption.
11-61
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Design Expert Output
Response:
Std. Dev.
Transform: Square root
ANOVA for Response Surface Linear Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
116.75
3
38.92
A
16.52
1
16.52
B
26.32
1
26.32
C
73.92
1
73.92
Residual
206.17
23
8.96
Cor Total
322.92
26
Constant:
0
F
Value
4.34
1.84
2.94
8.25
Prob > F
0.0145
0.1878
0.1001
0.0086
significant
The Model F-value of 4.34 implies the model is significant. There is only
a 1.45% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
2.99
6.00
49.88
279.05
Factor
Intercept
A-Speed
B-Pressure
C-Distance
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
6.00
0.96
1.21
2.03
DF
1
1
1
1
0.3616
0.2783
0.1359
7.278
Standard
Error
0.58
0.71
0.71
0.71
95% CI
Low
4.81
-0.50
-0.25
0.57
95% CI
High
7.19
2.42
2.67
3.49
VIF
1.00
1.00
1.00
Final Equation in Terms of Coded Factors:
Sqrt(Std. Dev.) =
+6.00
+0.96 * A
+1.21 * B
+2.03 * C
Final Equation in Terms of Actual Factors:
Sqrt(Std. Dev.) =
+6.00273
+0.95796 * Speed
+1.20916 * Pressure
+2.02643 * Distance
Because Factor A is insignificant, it is removed from the model. The reduced linear model analysis is
shown below:
Design Expert Output
Response:
Std. Dev.
Transform: Square root
ANOVA for Response Surface Reduced Linear Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
100.23
2
50.12
B
26.32
1
26.32
C
73.92
1
73.92
Residual
222.68
24
9.28
Cor Total
322.92
26
Constant:
0
F
Value
5.40
2.84
7.97
Prob > F
0.0116
0.1051
0.0094
The Model F-value of 5.40 implies the model is significant. There is only
a 1.16% chance that a "Model F-Value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
3.05
6.00
50.74
275.24
Coefficient
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
0.3104
0.2529
0.1476
6.373
Standard
11-62
95% CI
95% CI
significant
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Estimate
6.00
1.21
2.03
Intercept
B-Pressure
C-Distance
DF
1
1
1
Error
0.59
0.72
0.72
Low
4.79
-0.27
0.54
High
7.21
2.69
3.51
VIF
1.00
1.00
Final Equation in Terms of Coded Factors:
Sqrt(Std. Dev.) =
+6.00
+1.21 * B
+2.03 * C
Final Equation in Terms of Actual Factors:
Sqrt(Std. Dev.) =
+6.00273
+1.20916 * Pressure
+2.02643 * Distance
The following contour plots graphically represent the two models:
Mean
DE S IG N-E X P E RT P l o t
1.00
8 50
8 00
0.50
50
A c tu al Fa c tor
A : S p e e d = 1 .0 0
45
C : Dis tance
5 50
5 00
4 50
4 00
65
55
0.50
6 50
0.00
80
75
70
De si g n P o i n ts
6 00
C : Dis tance
1.00
S q rt(S td . De v.)
X = B : P re ssu re
Y = C: Di sta n c e
60
7 50
7 00
De si g n P o i n ts
A ctu al Fa ctor
A : S p e e d = 1 .0 0
S td. D e v.
DE S IG N-E X P E RT P l o t
9 50
9 00
M ean
X = B : P re ssu re
Y = C: Di sta n ce
40
35
0.00
30
25
-0.50
-0.50
3 50
20
15
3 00
10
2 50
-1.00
-1.00
-1.00
-0.50
0.00
0.50
1.00
-1.00
B: Pres s ure
-0.50
0.00
0.50
1.00
B: Pres s ure
(c) Find a set of optimum conditions that result in the mean as large as possible with the standard
deviation less than 60.
The overlay plot identifies a region that meets the criteria of the mean as large as possible with the
standard deviation less than 60. The optimum conditions in coded terms are approximately Speed = 1.0,
Pressure = 0.75 and Distance = 0.25.
11-63
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Overla y P lot
DE S IG N-E X P E RT P l o t
1.00
O ve rl a y P l o t
X = B : P re ssu re
Y = C: Di sta n ce
S t d. D e v . : 60
De si g n P o i n ts
Me an : 70 0
0.50
C : Dis tance
A ctu al Fa ctor
A : S p e e d = 1 .0 0
0.00
-0.50
-1.00
-1.00
-0.50
0.00
0.50
1.00
B: Pres s ure
11-34 A variation of Example 6-2. In example 6-2 we found that one of the process variables
(B=pressure) was not important. Dropping this variable produced two replicates of a 23 design. The data
are shown below.
C
D
A(+)
A(-)
y
+
+
+
+
45, 48
68, 80
43, 45
75, 70
71, 65
60, 65
100, 104
86, 96
57.75
68.25
73.00
81.75
s2
121.19
72.25
1124.67
134.92
Assume that C and D are controllable factors and that A is a noise factor.
(a) Fit a model to the mean response.
The following is the analysis of variance with all terms in the model:
Design Expert Output
Response:
Mean
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
300.05
3
100.02
A
92.64
1
92.64
B
206.64
1
206.64
AB
0.77
1
0.77
Pure Error
0.000
0
Cor Total
300.05
3
F
Value
Prob > F
Based on the above analysis, the AB interaction is removed from the model and used as error.
Design Expert Output
Response:
Mean
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
11-64
F
Value
Prob > F
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Model
A
B
Residual
Cor Total
299.28
92.64
206.64
0.77
300.05
2
1
1
1
3
149.64
92.64
206.64
0.77
195.45
121.00
269.90
0.0505
0.0577
0.0387
not significant
The Model F-value of 195.45 implies there is a 5.05% chance that a "Model F-Value"
this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.87
70.19
1.25
12.25
Factor
Intercept
A-Concentration
B-Stir Rate
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
70.19
4.81
7.19
0.9974
0.9923
0.9592
31.672
Standard
Error
0.44
0.44
0.44
DF
1
1
1
95% CI
Low
64.63
-0.75
1.63
95% CI
High
75.75
10.37
12.75
Final Equation in Terms of Coded Factors:
Mean =
+70.19
+4.81 * A
+7.19 * B
Final Equation in Terms of Actual Factors:
Mean =
+70.18750
+4.81250 * Concentration
+7.18750 * Stir Rate
The following is a contour plot of the mean model:
Mean
1.00
80
75
B: Stir R ate
0.50
70
0.00
65
-0.50
60
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: C oncentration
(b) Fit a model to the ln(s2) response.
The following is the analysis of variance with all terms in the model:
Design Expert Output
Response:
Variance
Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
11-65
Constant:
0
VIF
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Sum of
Squares
4.42
1.74
2.03
0.64
0.000
4.42
Source
Model
A
B
AB
Pure Error
Cor Total
DF
3
1
1
1
0
3
Mean
Square
1.47
1.74
2.03
0.64
F
Value
Prob > F
Based on the above analysis, the AB interaction is removed from the model and applied to the residual
error.
Design Expert Output
Response:
Variance
Transform: Natural log
ANOVA for Selected Factorial Model
Analysis of variance table [Partial sum of squares]
Sum of
Mean
Source
Squares
DF
Square
Model
3.77
2
1.89
A
1.74
1
1.74
B
2.03
1
2.03
Residual
0.64
1
0.64
Cor Total
4.42
3
Constant:
0
F
Value
2.94
2.71
3.17
Prob > F
0.3815
0.3477
0.3260
not significant
The "Model F-value" of 2.94 implies the model is not significant relative to the noise. There is a
38.15 % chance that a "Model F-value" this large could occur due to noise.
Std. Dev.
Mean
C.V.
PRESS
0.80
5.25
15.26
10.28
Factor
Intercept
A-Concentration
B-Stir Rate
R-Squared
Adj R-Squared
Pred R-Squared
Adeq Precision
Coefficient
Estimate
5.25
-0.66
0.71
DF
1
1
1
0.8545
0.5634
-1.3284
3.954
Standard
Error
0.40
0.40
0.40
95% CI
Low
0.16
-5.75
-4.38
95% CI
High
10.34
4.43
5.81
Final Equation in Terms of Coded Factors:
Ln(Variance) =
+5.25
-0.66 * A
+0.71 * B
Final Equation in Terms of Actual Factors:
Ln(Variance) =
+5.25185
-0.65945 * Concentration
+0.71311 * Stir Rate
The following is a contour plot of the variance model in the untransformed form:
11-66
VIF
1.00
1.00
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
V a ria nce
1.00
6 50 6 00
5 50 5 00
4 50
4 00
3 50
0.50
B: Stir R ate
3 00
2 50
2 00
0.00
1 50
1 00
-0.50
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: C oncentration
(c) Find operating conditions that result in the mean filtration rate response exceeding 75 with minimum
variance.
The overlay plot shown below identifies the region required by the process:
Overla y P lot
1.00
Me an : 75
B: Stir R ate
0.50
0.00
V aria nc e: 1 30
-0.50
-1.00
-1.00
-0.50
0.00
0.50
1.00
A: C oncentration
(d) Compare your results with those from Example 11-6 which used the transmission of error approach.
How similar are the two answers.
The results are very similar. Both require the Concentration to be held at the high level while the stirring
rate is held near the middle.
11-67
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 12
Experiments with Random Factors
Solutions
12-1
A textile mill has a large number of looms. Each loom is supposed to provide the same output of
cloth per minute. To investigate this assumption, five looms are chosen at random and their output is noted
at different times. The following data are obtained:
Loom
1
2
3
4
5
14.0
13.9
14.1
13.6
13.8
Output (lb/min)
14.2
14.0
13.9
14.0
14.1
14.0
14.0
13.9
13.9
13.8
14.1
13.8
14.2
13.8
13.6
14.1
14.0
13.9
13.7
14.0
(a) Explain why this is a random effects experiment. Are the looms equal in output? Use D = 0.05.
The looms used in the experiment are a random sample of all the looms in the manufacturing area. The
following is the analysis of variance for the data:
Minitab Output
ANOVA: Output versus Loom
Factor
Loom
Type Levels Values
random
5
1
2
3
4
5
Analysis of Variance for Output
Source
Loom
Error
Total
Source
1 Loom
2 Error
DF
4
20
24
SS
0.34160
0.29600
0.63760
MS
0.08540
0.01480
F
5.77
P
0.003
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
0.01412
2
(2) + 5(1)
0.01480
(2)
(b) Estimate the variability between looms.
Vˆ W2
MS Model MS E
n
0.0854 0.0148
5
0.01412
(c) Estimate the experimental error variance.
Vˆ 2
MS E
0.0148
(d) Find a 95 percent confidence interval for V W2 V W2 V 2 .
L
º
1 ª MS Model
1
1»
«
n ¬« MS E FD 2 ,a 1,n a »¼
12-1
0.1288
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
U
º
1 ª MS Model
1
1»
«
n «¬ MS E F1D 2 ,a 1,n a »¼
3.851
2
VW
L
U
d
d
L 1 V W2 V 2 U 1
0.144 d
V W2
V W2 V 2
d 0.794
(e) Analyze the residuals from this experiment. Do you think that the analysis of variance assumptions are
satisfied?
There is nothing unusual about the residual plots; therefore, the analysis of variance assumptions are
satisfied.
Normal Probability Plot of the Residuals
(response is Output)
2
Normal Score
1
0
-1
-2
-0.2
-0.1
0.0
0.1
0.2
Residual
Residuals Versus the Fitted Values
(response is Output)
0.2
Residual
0.1
0.0
-0.1
-0.2
13.8
13.9
14.0
Fitted Value
12-2
14.1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus Loom
(response is Output)
0.2
Residual
0.1
0.0
-0.1
-0.2
1
2
3
4
5
Loom
12-2 A manufacturer suspects that the batches of raw material furnished by her supplier differ
significantly in calcium content. There are a large number of batches currently in the warehouse. Five of
these are randomly selected for study. A chemist makes five determinations on each batch and obtains the
following data:
Batch 1
23.46
23.48
23.56
23.39
23.40
Batch 2
23.59
23.46
23.42
23.49
23.50
Batch 3
23.51
23.64
23.46
23.52
23.49
Batch 4
23.28
23.40
23.37
23.46
23.39
Batch 5
23.29
23.46
23.37
23.32
23.38
(a) Is there significant variation in calcium content from batch to batch? Use D = 0.05.
Yes, as shown in the Minitab Output below, there is a difference.
Minitab Output
ANOVA: Calcium versus Batch
Factor
Batch
Type Levels Values
random
5
1
2
3
4
5
Analysis of Variance for Calcium
Source
Batch
Error
Total
Source
1 Batch
2 Error
DF
4
20
24
SS
0.096976
0.087600
0.184576
MS
0.024244
0.004380
F
5.54
P
0.004
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
0.00397
2
(2) + 5(1)
0.00438
(2)
(b) Estimate the components of variance.
V W2
MS Model MS E
n
.024244 .004380
5
12-3
0.00397
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
V 2
MS E
0.004380
(c) Find a 95 percent confidence interval for V W2 V W2 V 2 .
L
U
º
1 ª MS Model
1
1» 0.1154
«
n «¬ MS E FD 2 ,a 1,n a »¼
º
1 ª MS Model
1
1» 9.276
«
n «¬ MS E F1D 2 ,a 1,n a »¼
V2
L
U
d 2 W 2 d
L 1 V W V
U 1
0.1035 d
V W2
V W2 V 2
d 0.9027
(d) Analyze the residuals from this experiment. Are the basic analysis of variance assumptions satisfied?
There are five residuals that stand out in the normal probability plot. From the Residual vs. Batch plot, we
see that one point per batch appears to stand out. A natural log transformation was applied to the data but
did not change the results of the residual analysis. Further investigation should probably be performed to
determine if these points are outliers.
Normal Probability Plot of the Residuals
(response is Calcium)
2
Normal Score
1
0
-1
-2
-0.1
0.0
Residual
12-4
0.1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Fitted Values
(response is Calcium)
0.1
Residual
0.0
-0.1
23.35
23.37
23.39
23.41
23.43
23.45
23.47
23.49
23.51
23.53
Fitted Value
Residuals Versus Batch
(response is Calcium)
0.1
Residual
0.0
-0.1
1
2
3
4
5
Batch
12-3 Several ovens in a metal working shop are used to heat metal specimens. All the ovens are supposed
to operate at the same temperature, although it is suspected that this may not be true. Three ovens are
selected at random and their temperatures on successive heats are noted. The data collected are as follows:
Oven
1
2
3
Temperature
491.50 498.30 498.10
488.50 484.65 479.90
490.10 484.80 488.25
493.50
477.35
473.00
493.60
471.85
478.65
(a) Is there significant variation in temperature between ovens? Use D = 0.05.
The analysis of variance shown below identifies significant variation in temperature between the ovens.
Minitab Output
General Linear Model: Temperature versus Oven
Factor
Type Levels Values
12-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Oven
random
3 1 2 3
Analysis of Variance for Temperat, using Adjusted SS for Tests
Source
Oven
Error
Total
DF
2
12
14
Seq SS
594.53
413.81
1008.34
Adj SS
594.53
413.81
Adj MS
297.27
34.48
F
8.62
P
0.005
Expected Mean Squares, using Adjusted SS
Source
1 Oven
2 Error
Expected Mean Square for Each Term
(2) + 4.9333(1)
(2)
Error Terms for Tests, using Adjusted SS
Source
1 Oven
Error DF
12.00
Error MS
34.48
Synthesis of Error MS
(2)
Variance Components, using Adjusted SS
Source
Oven
Error
Estimated Value
53.27
34.48
(b) Estimate the components of variance.
n0
1 ª
¦ ni2 º» 1 ª15 25 16 36 º 4.93
«¦ ni
»
a 1 «
15
¼
¦ ni »¼ 2 «¬
¬
MS Model MS E 297.27 34.48
V W2
53.30
n
4.93
V 2 MS E 34.48
(c) Analyze the residuals from this experiment. Draw conclusions about model adequacy.
There is a funnel shaped appearance in the plot of residuals versus predicted value indicating a possible
non-constant variance. There is also some indication of non-constant variance in the plot of residuals
versus oven. The inequality of variance problem is not severe.
Normal Probability Plot of the Residuals
(response is Temperat)
2
Normal Score
1
0
-1
-2
-10
0
Residual
12-6
10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Fitted Values
(response is Temperat)
Residual
10
0
-10
480
485
490
495
Fitted Value
Residuals Versus Oven
(response is Temperat)
Residual
10
0
-10
1
2
3
Oven
12-4 An article in the Journal of the Electrochemical Society (Vol. 139, No. 2, 1992, pp. 524-532)
describes an experiment to investigate the low-pressure vapor deposition of polysilicon. The experiment
was carried out in a large-capacity reactor at Sematech in Austin, Texas. The reactor has several wafer
positions, and four of these positions are selected at random. The response variable is film thickness
uniformity. Three replicates of the experiments were run, and the data are as follows:
Wafer Position
1
2
3
4
2.76
1.43
2.34
0.94
Uniformity
5.67
1.70
1.97
1.36
4.49
2.19
1.47
1.65
(a) Is there a difference in the wafer positions? Use D = 0.05.
Yes, there is a difference.
12-7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Minitab Output
ANOVA: Uniformity versus Wafer Position
Factor
Wafer Po
Type Levels Values
fixed
4
1
2
3
4
Analysis of Variance for Uniformi
Source
Wafer Po
Error
Total
DF
3
8
11
SS
16.2198
5.2175
21.4373
MS
5.4066
0.6522
F
8.29
P
0.008
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Wafer Po
2
(2) + 3Q[1]
2 Error
0.6522
(2)
(b) Estimate the variability due to wafer positions.
VW2
VW2
MSTreatment MS E
n
54066
0.6522
.
.
15844
3
(c) Estimate the random error component.
V 2
0.6522
(d) Analyze the residuals from this experiment and comment on model adequacy.
Variability in film thickness seems to depend on wafer position. These observations also show up as
outliers on the normal probability plot. Wafer position number 1 appears to have greater variation in
uniformity than the other positions.
Normal Probability Plot of the Residuals
(response is Uniformi)
2
Normal Score
1
0
-1
-2
-1
0
Residual
12-8
1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus the Fitted Values
(response is Uniformi)
Residual
1
0
-1
1
2
3
4
Fitted Value
Residuals Versus Wafer Po
(response is Uniformi)
Residual
1
0
-1
1
2
3
4
Wafer Po
12-5 Consider the vapor deposition experiment described in Problem 12-4.
(a) Estimate the total variability in the uniformity response.
Vˆ W2 Vˆ 2
1.5848 0.6522 2.2370
(b) How much of the total variability in the uniformity response is due to the difference between positions
in the reactor?
Vˆ W2
Vˆ 2 Vˆ W2
1.5848
2.2370
0.70845
(c) To what level could the variability in the uniformity response be reduced, if the position-to-position
variability in the reactor could be eliminated? Do you believe this is a significant reduction?
12-9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
The variability would be reduced from 2.2370 to Vˆ 2
0.6522 which is a reduction of approximately:
2.2370 0.6522
2.2370
71%
12-6 An article in the Journal of Quality Technology (Vol. 13, No. 2, 1981, pp. 111-114) describes and
experiment that investigates the effects of four bleaching chemicals on pulp brightness. These four
chemicals were selected at random from a large population of potential bleaching agents. The data are as
follows:
Chemical
1
2
3
4
Pulp Brightness
74.466
92.746
79.306
81.914
78.017
91.596
78.358
77.544
77.199
80.522
79.417
78.001
76.208
80.346
80.802
77.364
82.876
73.385
80.626
77.386
(a) Is there a difference in the chemical types? Use D = 0.05.
The computer output shows that the null hypothesis cannot be rejected. Therefore, there is no evidence that
there is a difference in chemical types.
Minitab Output
ANOVA: Brightness versus Chemical
Factor
Type Levels Values
Chemical random
4
1
2
3
4
Analysis of Variance for Brightne
Source
Chemical
Error
Total
DF
3
16
19
SS
53.98
383.99
437.97
MS
17.99
24.00
F
0.75
P
0.538
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Chemical
-1.201
2
(2) + 5(1)
2 Error
23.999
(2)
(b) Estimate the variability due to chemical types.
MSTreatment MSE
n
17
994
23.999
.
VW2
1201
.
5
which agrees with the Minitab output.
Because the variance component cannot be negative, this likely means that the variability due to chemical
types is zero.
VW2
(c) Estimate the variability due to random error.
V 2
23.999
(d) Analyze the residuals from this experiment and comment on model adequacy.
12-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Two data points appear to be outliers in the normal probability plot of effects. These outliers belong to
chemical types 1 and 3 and should be investigated. There seems to be much less variability in brightness
with chemical type 4.
Normal Probability Plot of the Residuals
(response is Brightne)
2
Normal Score
1
0
-1
-2
-5
0
5
10
Residual
Residuals Versus the Fitted Values
(response is Brightne)
Residual
10
5
0
-5
78
79
80
Fitted Value
12-11
81
82
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Residuals Versus Chemical
(response is Brightne)
Residual
10
5
0
-5
1
2
3
4
Chemical
12-7
Consider the one-way balanced, random effects method. Develop a procedure for finding a 100(1D) percent confidence interval for V 2 / (V W2 V 2 ) .
ª
º
V2
We know that P « L d W2 d U »
V
«¬
»¼
2
2
ª
º
V
V
P « L 1 d W2 2 d U 1»
V
V
«¬
»¼
2
2
ª
º
V V
P« L 1 d W 2
d U 1»
V
«¬
»¼
ª L
U º
V2
t 2
t
P«
»
2
1 U »¼
«¬ 1 L V W V
1 D
1D
1D
1D
12-8 Refer to Problem 12-1.
(a) What is the probability of accepting H0 if V W2 is four times the error variance V 2 ?
O
X1
1
X2
a 1 4
nV W2
1
5 4V 2
21 4.6
V2
V2
N a 25 5 20 E | 0.035 , from the OC curve.
(b) If the difference between looms is large enough to increase the standard deviation of an observation by
20 percent, we wish to detect this with a probability of at least 0.80. What sample size should be used?
X1
a 1 4
O
>
X2
N a
@
1 n 1 0.01P 2 1
25 5
>
20
D
0.05
1 n 1 0.01 20
12-12
2
@
1
P ( accept ) d 0.2
1 0.44n
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Trial and Error yields:
X2
20
45
65
n
5
10
14
P(accept)
O
1.79
2.32
2.67
0.6
0.3
0.2
Choose n t 14, therefore N t 70
12-9 An experiment was performed to investigate the capability of a measurement system. Ten parts
were randomly selected, and two randomly selected operators measured each part three times. The tests
were made in random order, and the data below resulted.
Part
Number
1
2
3
4
5
6
7
8
9
10
Operator 1
Measurements
-------------------------1
2
3
50
49
50
52
52
51
53
50
50
49
51
50
48
49
48
52
50
50
51
51
51
52
50
49
50
51
50
47
46
49
Operator 2
Measurements
------------------------1
2
3
50
48
51
51
51
51
54
52
51
48
50
51
48
49
48
52
50
50
51
50
50
53
48
50
51
48
49
46
47
48
(a) Analyze the data from this experiment.
Minitab Output
ANOVA: Measurement versus Part, Operator
Factor
Part
Type Levels Values
random
10
1
8
Operator random
2
1
2
9
2
3
10
4
5
6
7
Analysis of Variance for Measurem
Source
Part
Operator
Part*Operator
Error
Total
DF
9
1
9
40
59
SS
99.017
0.417
5.417
60.000
164.850
MS
11.002
0.417
0.602
1.500
F
18.28
0.69
0.40
P
0.000
0.427
0.927
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Part
1.73333
3
(4) + 3(3) + 6(1)
Operator
-0.00617
3
(4) + 3(3) + 30(2)
Part*Operator -0.29938
4
(4) + 3(3)
Error
1.50000
(4)
(b) Find point estimates of the variance components using the analysis of variance method.
V 2
MS E
12-13
V 2
15
.
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2
VWE
V W2
MS AB MS E
0.6018519 15000000
.
2
2
V WE
=0
0 , assume V WE
n
3
MS B MS AB
11.001852 0.6018519
V E2
Vˆ E2
1.7333
an
23
MS A MS AB
bn
0.416667 0.6018519
0 , assume V W2 =0
10 3
Vˆ W2
All estimates agree with the Minitab output.
12-10 Reconsider the data in Problem 5-6. Suppose that both factors, machines and operators, are chosen
at random.
(a) Analyze the data from this experiment.
Operator
1
1
109
110
2
110
115
Machine
3
108
109
4
110
108
2
110
112
110
111
111
109
114
112
3
116
114
112
115
114
119
120
117
The following Minitab output contains the analysis of variance and the variance component estimates:
Minitab Output
ANOVA: Strength versus Operator, Machine
Factor
Type Levels Values
Operator random
3
1
Machine random
4
1
2
2
3
3
4
Analysis of Variance for Strength
Source
Operator
Machine
Operator*Machine
Error
Total
DF
2
3
6
12
23
SS
160.333
12.458
44.667
45.500
262.958
MS
80.167
4.153
7.444
3.792
F
10.77
0.56
1.96
P
0.010
0.662
0.151
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Operator
9.0903
3
(4) + 2(3) + 8(1)
Machine
-0.5486
3
(4) + 2(3) + 6(2)
Operator*Machine
1.8264
4
(4) + 2(3)
Error
3.7917
(4)
(b) Find point estimates of the variance components using the analysis of variance method.
2
VWE
V 2 MS E
MS AB MS E
2
V WE
n
V 2 3.79167
7.44444 3.79167
2
12-14
.
182639
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
V E2
MS B MS AB
an
V W2
4.15278 7.44444
0 , assume V E2
3(2)
V E2
MS A MS AB
bn
V W2
8016667
7.44444
.
4(2)
9.09028
These results agree with the Minitab variance component analysis.
12-11 Reconsider the data in Problem 5-13. Suppose that both factors are random.
(a) Analyze the data from this experiment.
Row Factor
1
2
3
Column
2
39
20
37
1
36
18
30
Factor
3
36
22
33
4
32
20
34
Minitab Output
General Linear Model: Response versus Row, Column
Factor
Row
Column
Type Levels Values
random
3 1 2 3
random
4 1 2 3 4
Analysis of Variance for Response, using Adjusted SS for Tests
Source
Row
Column
Row*Column
Error
Total
DF
2
3
6
0
11
Seq SS
580.500
28.917
28.833
0.000
638.250
Adj SS
580.500
28.917
28.833
0.000
Adj MS
290.250
9.639
4.806
0.000
F
60.40
2.01
**
** Denominator of F-test is zero.
Expected Mean Squares, using Adjusted SS
Source
1 Row
2 Column
3 Row*Column
4 Error
Expected Mean Square for Each Term
(4) +
(3) + 4.0000(1)
(4) +
(3) + 3.0000(2)
(4) +
(3)
(4)
Error Terms for Tests, using Adjusted SS
Source
1 Row
2 Column
3 Row*Column
Error DF
*
*
*
Error MS
4.806
4.806
*
Synthesis of Error MS
(3)
(3)
(4)
Variance Components, using Adjusted SS
Source
Row
Column
Row*Column
Error
Estimated Value
71.3611
1.6111
4.8056
0.0000
(b) Estimate the variance components.
12-15
P
**
**
0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Because the experiment is unreplicated and the interaction term was included in the model, there is no
estimate of MSE, and therefore, no estimate of V 2 .
2
VWE
V E2
V W2
MS AB MS E
4.8056 0
2
Vˆ WE
4.8056
n
1
MS B MS AB
9.6389 4.8056
Vˆ E2
1.6111
an
31
MS A MS AB
bn
Vˆ W2
290.2500 4.8056
41
71.3611
These estimates agree with the Minitab output.
12-12 Suppose that in Problem 5-11 the furnace positions were randomly selected, resulting in a mixed
model experiment. Reanalyze the data from this experiment under this new assumption. Estimate the
appropriate model components.
Position
1
2
800
570
565
583
Temperature (°C)
825
1063
1080
1043
850
565
510
590
528
547
521
988
1026
1004
526
538
532
The following analysis assumes a restricted model:
Minitab Output
ANOVA: Density versus Position, Temperature
Factor
Type Levels Values
Position random
2
1
Temperat fixed
3
800
2
825
850
Analysis of Variance for Density
Source
Position
Temperat
Position*Temperat
Error
Total
DF
1
2
2
12
17
SS
7160
945342
818
5371
958691
MS
F
7160
16.00
472671 1155.52
409
0.91
448
P
0.002
0.001
0.427
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Position
745.83
4
(4) + 9(1)
Temperat
3
(4) + 3(3) + 6Q[2]
Position*Temperat
-12.83
4
(4) + 3(3)
Error
447.56
(4)
2
VWE
V 2
MS AB MS E
n
Vˆ 2
MS E
2
V̂ WE
447.56
409 448
2
0 assume V WE
3
12-16
0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Vˆ W2
MS A MS E
bn
Vˆ W2
7160 448
33
745.83
These results agree with the Minitab output.
12-13 Reanalyze the measurement systems experiment in Problem 12-9, assuming that operators are a
fixed factor. Estimate the appropriate model components.
The following analysis assumes a restricted model:
Minitab Output
ANOVA: Measurement versus Part, Operator
Factor
Part
Type Levels Values
random
10
1
8
Operator fixed
2
1
2
9
2
3
10
4
5
F
7.33
0.69
0.40
P
0.000
0.427
0.927
6
7
Analysis of Variance for Measurem
Source
Part
Operator
Part*Operator
Error
Total
DF
9
1
9
40
59
SS
99.017
0.417
5.417
60.000
164.850
MS
11.002
0.417
0.602
1.500
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Part
1.5836
4
(4) + 6(1)
Operator
3
(4) + 3(3) + 30Q[2]
Part*Operator -0.2994
4
(4) + 3(3)
Error
1.5000
(4)
V 2
2
Vˆ WE
Vˆ 2 1.5000
MS AB MS E
0.60185 1.5000
2
2
Vˆ WE
0 assume V WE
3
n
MS A MS E
11.00185 1.50000
Vˆ W2
Vˆ W2
1.58364
23
bn
MS E
0
These results agree with the Minitab output.
12-14 In problem 5-6, suppose that there are only four machines of interest, but the operators were selected
at random.
(a) What type of model is appropriate?
A mixed model is appropriate.
(b) Perform the analysis and estimate the model components.
The following analysis assumes a restricted model:
Minitab Output
ANOVA: Strength versus Operator, Machine
Factor
Type Levels Values
12-17
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Operator random
Machine
fixed
3
4
1
1
2
2
3
3
4
Analysis of Variance for Strength
Source
Operator
Machine
Operator*Machine
Error
Total
DF
2
3
6
12
23
SS
160.333
12.458
44.667
45.500
262.958
MS
80.167
4.153
7.444
3.792
F
21.14
0.56
1.96
P
0.000
0.662
0.151
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Operator
9.547
4
(4) + 8(1)
Machine
3
(4) + 2(3) + 6Q[2]
Operator*Machine
1.826
4
(4) + 2(3)
Error
3.792
(4)
V 2 MS E
Vˆ 2 3.792
MS AB MS E
7.444 3.792
2
Vˆ WE
1.826
n
2
MS A MS E
80.167 3.792
Vˆ W2
9.547
42
bn
2
VWE
Vˆ W2
These results agree with the Minitab output.
12-15 By application of the expectation operator, develop the expected mean squares for the two-factor
factorial, mixed model. Use the restricted model assumptions. Check your results with the expected mean
squares given in Table 12-11 to see that they agree.
The sums of squares may be written as
a
SS A
bn
¦
yi .. y...
2
b
,
SS B
an
i 1
a
SS AB
n
. j.
y ij . y i .. y . j . y ...
a
2
,
SS E
i 1 j 1
Using the model y ijk
..
2
b
n
¦¦ ¦ y
ijk
y...
2
i 1 j 1 k 1
P W i E j WE
ij
H ijk , we may find that
y i ..
P W i WE
y. j.
P E j H . j.
y ij .
P W i E j WE
y ...
P E . H ...
i.
H i ..
ij
H ij .
Using the assumptions for the restricted form of the mixed model, W .
WE
y...
j 1
b
¦¦
¦y
0 , WE
0 . Substituting these expressions into the sums of squares yields
12-18
.j
0 , which imply that
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
a
SS A
bn
¦ W WE
i 1
b
SS B
¦E
an
j
H i .. H ...
i.
2
2
H . j . H ...
j 1
a
b
¦¦
SS AB
WE )ij WE
n
i.
H ij . H i .. H . j . H ...
2
i 1 j 1
a
SS E
b
n
¦¦¦ H
2
H ij .
ijk
i 1 j 1 k 1
Using the assumption that E H ijk
0 , V (H ijk )
0 , and E H ijk H i' j' k'
0 , we may divide each sum of
squares by its degrees of freedom and take the expectation to produce
E MS A
ª bn º
V2 «
»E
¬ a 1 ¼ i
E MSB
ª an º
V2 «
»
¬ b 1 ¼ j
a
¦W
E MSE
¦E
i.
2
j
1
ª
n
V2 «
¬ a 1 b 1
V
2
WE
1
b
E MS AB
i
º
»E
¼ i
a
b
¦¦
WE
ij
WE
2
i.
1 j 1
2
Note that E MSB and E MS E are the results given in Table 8-3. We need to simplify E MS A and
E MS AB . Consider E MS A
since WE
ij
bn ª
«
a 1 «¬ i
a
¦EW
2
a
¦ E WE
E MS A
V2
E MS A
ª
ª a 1
« a
«
bn
2
« W i a ¬ a
V2
a 1 « i 1
b
«
¬
E MS A
2
V 2 nV WE
i
crossproducts
i 1
1
¦
bn
a 1
2
i.
a
¦W
i
º
º
»
»
¼ V2 »
WE
»
»
¼
2
i 1
§ a 1 2 ·
V WE ¸ . Consider E MS AB
is NID¨ 0,
a
©
¹
E MS AB
V2
E MS AB
V2
E MS AB
2
V
n
a 1 b 1
n
a 1 b 1
a
b
¦¦ E WE
ij
WE
2
i.
i 1 j 1
a
b
§ b 1 ·§ a 1 · 2
¸¨
¸V WE
b ¹© a ¹
1
¦¦ ¨©
i 1 j
2
nV WE
12-19
º
0»
»¼
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Thus E MS A and E MS AB agree with table 12-8.
12-16 Consider the three-factor factorial design in Example 12-6. Propose appropriate test statistics for all
main effects and interactions. Repeat for the case where A and B are fixed and C is random.
If all three factors are random there are no exact tests on main effects. We could use the following:
MS A MS ABC
MS AB MS AC
MSB MS ABC
MS AB MSBC
MSC MS ABC
MS AC MSBC
A: F
B:F
C:F
If A and B are fixed and C is random, the expected mean squares are (assuming the restricted for m of the
model):
Factor
F
a
i
F
b
j
R
c
k
R
n
l
E(MS)
Wi
0
b
c
n
V 2 bnV WJ2 bcn
Ej
a
0
c
n
2
acn
V 2 anV EJ
Jk
a
b
1
n
V 2 abnV J2
0
0
c
n
2
cn
V 2 nV WEJ
WE
WJ
ij
ik
2
0
b
1
n
V
jk
a
0
1
n
2
V 2 anV EJ
WEJ
ijk
0
0
1
n
2
V 2 nV WEJ
1
1
1
1
V2
ijk l
E 2j
¦ b 1
WE
2
ji
¦¦ a 1 b 1
bnV WJ2
EJ
H
W i2
¦ a 1
These are exact tests for all effects.
12-17 Consider the experiment in Example 12-7. Analyze the data for the case where A, B, and C are
random.
Minitab Output
ANOVA: Drop versus Temp, Operator, Gauge
Factor
Type Levels Values
Temp
random
3
60
Operator random
4
1
Gauge
random
3
1
75
2
2
90
3
3
4
Analysis of Variance for Drop
Source
Temp
DF
2
SS
1023.36
MS
511.68
12-20
F
2.30
P
0.171 x
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Operator
Gauge
Temp*Operator
Temp*Gauge
Operator*Gauge
Temp*Operator*Gauge
Error
Total
3
2
6
4
6
12
36
71
423.82
7.19
1211.97
137.89
209.47
166.11
770.50
3950.32
141.27
3.60
202.00
34.47
34.91
13.84
21.40
0.63
0.06
14.59
2.49
2.52
0.65
0.616 x
0.938 x
0.000
0.099
0.081
0.788
x Not an exact F-test.
Source
1
2
3
4
5
6
7
8
Variance Error Expected Mean Square for
component term (using restricted model)
Temp
12.044
*
(8) + 2(7) + 8(5) + 6(4)
Operator
-4.544
*
(8) + 2(7) + 6(6) + 6(4)
Gauge
-2.164
*
(8) + 2(7) + 6(6) + 8(5)
Temp*Operator
31.359
7
(8) + 2(7) + 6(4)
Temp*Gauge
2.579
7
(8) + 2(7) + 8(5)
Operator*Gauge
3.512
7
(8) + 2(7) + 6(6)
Temp*Operator*Gauge
-3.780
8
(8) + 2(7)
Error
21.403
(8)
Each Term
+ 24(1)
+ 18(2)
+ 24(3)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 Temp
2 Operator
3 Gauge
Error DF
6.97
7.09
5.98
Error MS
222.63
223.06
55.54
Synthesis of
(4) + (5) (4) + (6) (5) + (6) -
Error MS
(7)
(7)
(7)
Since all three factors are random there are no exact tests on main effects. Minitab uses an approximate F
test for the these factors.
12-18 Derive the expected mean squares shown in Table 12-14.
Factor
F
a
i
R
b
j
R
c
k
R
n
l
E(MS)
Wi
0
b
c
n
2
2
bnV WJ2 cnV WE
bcn
V 2 nV WEJ
Ej
a
1
c
n
2
V 2 anV EJ
acnV E2
Jk
a
b
1
n
2
V 2 anV EJ
abnV J2
WE
ij
0
1
c
n
2
2
V 2 nV WEJ
cnV WE
WJ
ik
0
b
1
n
2
V 2 nV WEJ
bnV WJ2
EJ
jk
a
1
1
n
2
V 2 anV EJ
WEJ
ijk
0
1
1
n
2
V 2 nV WEJ
1
1
1
1
V2
H ijkl
W i2
¦ a 1
12-19 Consider a four-factor factorial experiment where factor A is at a levels, factor B is at b levels, factor
C is at c levels, factor D is at d levels, and there are n replicates. Write down the sums of squares, the
degrees of freedom, and the expected mean squares for the following cases. Do exact tests exist for all
effects? If not, propose test statistics for those effects that cannot be directly tested. Assume the restricted
model on all cases. You may use a computer package such as Minitab.
The four factor model is:
12-21
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
y ijklh
P W i E j J k G l WE
WEJ
ijk
WEG
ijl
ij
WJ
EJG
jkl
ik
WG
WJG
ikl
il
EJ
WEJG
EG
jk
ijkl
jl
JG
kl
H ijklh
To simplify the expected mean square derivations, let capital Latin letters represent the factor effects or
¦W
2
i
F
b
j
b
0
b
b
0
b
b
0
0
b
0
0
0
b
0
1
F
c
k
c
c
0
c
c
0
c
0
c
0
0
c
0
0
0
1
bcdn
variance components. For example, A
a1
, or B
acdnV E2 .
(a) A, B, C, and D are fixed factors.
F
a
i
0
a
a
a
0
0
0
a
a
a
0
0
a
0
0
1
Factor
Wi
Ej
Jk
Gl
(WE ) ij
(WJ ) ik
(WG ) il
( EJ ) jk
( EG ) jl
(JG ) kl
(WEJ ) ijk
(WEG ) ijl
( EJG ) jkl
(WJG ) ikl
(WEJG ) ijkl
H (ijkl ) h
F
d
l
d
d
d
0
d
d
0
d
0
0
d
0
0
0
0
1
R
n
h
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
1
E(MS)
V2 A
V2 B
V2 C
V2 D
V 2 AB
V 2 AC
V 2 AD
V 2 BC
V 2 BD
V 2 CD
V 2 ABC
V 2 ABD
V 2 BCD
V 2 ACD
V 2 ABCD
V2
There are exact tests for all effects. The results can also be generated in Minitab as follows:
Minitab Output
ANOVA: y versus A, B, C, D
Factor
Type Levels Values
A
fixed
2
H
B
fixed
2
H
C
fixed
2
H
D
fixed
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
DF
1
1
1
1
1
1
1
1
1
1
1
1
SS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
MS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
F
0.49
0.01
0.09
0.01
0.25
0.25
0.25
0.25
0.25
0.25
0.25
2.27
P
0.492
0.921
0.767
0.921
0.622
0.622
0.622
0.622
0.622
0.622
0.622
0.151
12-22
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
16
31
3.13
3.13
3.13
198.00
264.88
3.13
3.13
3.13
12.38
0.25
0.25
0.25
0.622
0.622
0.622
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 A
16
(16) + 16Q[1]
2 B
16
(16) + 16Q[2]
3 C
16
(16) + 16Q[3]
4 D
16
(16) + 16Q[4]
5 A*B
16
(16) + 8Q[5]
6 A*C
16
(16) + 8Q[6]
7 A*D
16
(16) + 8Q[7]
8 B*C
16
(16) + 8Q[8]
9 B*D
16
(16) + 8Q[9]
10 C*D
16
(16) + 8Q[10]
11 A*B*C
16
(16) + 4Q[11]
12 A*B*D
16
(16) + 4Q[12]
13 A*C*D
16
(16) + 4Q[13]
14 B*C*D
16
(16) + 4Q[14]
15 A*B*C*D
16
(16) + 2Q[15]
16 Error
12.38
(16)
(b) A, B, C, and D are random factors.
Factor
Wi
Ej
Jk
Gl
(WE ) ij
(WJ ) ik
(WG ) il
( EJ ) jk
( EG ) jl
(JG ) kl
(WEJ ) ijk
(WEG ) ijl
( EJG ) jkl
(WJG ) ikl
(WEJG ) ijkl
H (ijkl ) h
R
a
i
1
a
a
a
1
1
1
a
a
a
1
1
a
1
1
1
R
b
j
b
1
b
b
1
b
b
1
1
b
1
1
1
b
1
1
R
c
k
c
c
1
c
c
1
c
1
c
1
1
c
1
1
1
1
R
d
l
d
d
d
1
d
d
1
d
1
1
d
1
1
1
1
1
R
n
h
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
1
E(MS)
V 2 ABCD ACD ABD ABC AD AC AB A
V 2 ABCD BCD ABD ABC BD BC AB B
V 2 ABCD ACD BCD ABC AB BC CD C
V 2 ABCD ACD BCD ABD BD AD CD D
V 2 ABCD ABC ABD AB
V 2 ABCD ABC ACD AC
V 2 ABCD ABD ACD AD
V 2 ABCD ABC BCD BC
V 2 ABCD ABD BCD BD
V 2 ABCD ACD BCD CD
V 2 ABCD ABC
V 2 ABCD ABD
V 2 ABCD BCD
V 2 ABCD ACD
V 2 ABCD
V2
No exact tests exist on main effects or two-factor interactions. For main effects use statistics such as:
A: F
MS A MS ABC MS ABD MS ACD
MS AB MS AC MS AD MS ABCD
For testing two-factor interactions use statistics such as: AB: F
The results can also be generated in Minitab as follows:
12-23
MS AB MS ABCD
MS ABC MS ABD
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
random
2
H
random
2
H
random
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
31
SS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
MS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
F
**
**
0.36
**
0.11
1.00
0.11
1.00
0.11
1.00
1.00
9.00
1.00
1.00
0.25
P
0.843 x
0.796
0.667
0.796
0.667
0.796
0.667
0.500
0.205
0.500
0.500
0.622
x
x
x
x
x
x
x Not an exact F-test.
** Denominator of F-test is zero.
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 A
1.7500
*
(16) + 2(15) + 4(13) + 4(12) + 4(11)
+ 8(5) + 16(1)
2 B
1.3750
*
(16) + 2(15) + 4(14) + 4(12) + 4(11)
+ 8(5) + 16(2)
3 C
-0.1250
*
(16) + 2(15) + 4(14) + 4(13) + 4(11)
+ 8(6) + 16(3)
4 D
1.3750
*
(16) + 2(15) + 4(14) + 4(13) + 4(12)
+ 8(7) + 16(4)
5 A*B
-3.1250
*
(16) + 2(15) + 4(12) + 4(11) + 8(5)
6 A*C
0.0000
*
(16) + 2(15) + 4(13) + 4(11) + 8(6)
7 A*D
-3.1250
*
(16) + 2(15) + 4(13) + 4(12) + 8(7)
8 B*C
0.0000
*
(16) + 2(15) + 4(14) + 4(11) + 8(8)
9 B*D
-3.1250
*
(16) + 2(15) + 4(14) + 4(12) + 8(9)
10 C*D
0.0000
*
(16) + 2(15) + 4(14) + 4(13) + 8(10)
11 A*B*C
0.0000 15
(16) + 2(15) + 4(11)
12 A*B*D
6.2500 15
(16) + 2(15) + 4(12)
13 A*C*D
0.0000 15
(16) + 2(15) + 4(13)
14 B*C*D
0.0000 15
(16) + 2(15) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
+ 8(7) + 8(6)
+ 8(9) + 8(8)
+ 8(10) + 8(8)
+ 8(10) + 8(9)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 A
2 B
3 C
4 D
5 A*B
6 A*C
7 A*D
Error DF
0.56
0.56
0.14
0.56
0.98
0.33
0.98
Error MS
*
*
3.13
*
28.13
3.13
28.13
Synthesis of
(5) + (6) +
(5) + (8) +
(6) + (8) +
(7) + (9) +
(11) + (12)
(11) + (13)
(12) + (13)
Error MS
(7) - (11) - (12) - (13) + (15)
(9) - (11) - (12) - (14) + (15)
(10) - (11) - (13) - (14) + (15)
(10) - (12) - (13) - (14) + (15)
- (15)
- (15)
- (15)
12-24
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
8 B*C
9 B*D
10 C*D
0.33
0.98
0.33
3.13
28.13
3.13
(11) + (14) - (15)
(12) + (14) - (15)
(13) + (14) - (15)
(c) A is fixed and B, C, and D are random.
Factor
Wi
Ej
Jk
Gl
(WE ) ij
(WJ ) ik
(WG ) il
( EJ ) jk
( EG ) jl
(JG ) kl
(WEJ ) ijk
(WEG ) ijl
( EJG ) jkl
(WJG ) ikl
(WEJG ) ijkl
H (ijkl ) h
F
a
i
0
a
a
a
0
0
0
a
a
a
0
0
a
0
0
1
R
b
j
b
1
b
b
1
b
b
1
1
b
1
1
1
b
1
1
R
c
k
c
c
1
c
c
1
c
1
c
1
1
c
1
1
1
1
R
d
l
d
d
d
1
d
d
1
d
1
1
d
1
1
1
1
1
R
n
h
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
1
E(MS)
V 2 ABCD ACD ABD ABC AD AC AB A
V 2 BCD ABD BC B
V 2 BCD BC CD C
V 2 BCD BD CD D
V 2 ABCD ABC ABD AB
V 2 ABCD ABC ACD AC
V 2 ABCD ABD ACD AD
V 2 BCD BC
V 2 BCD BD
V 2 BCD CD
V 2 ABCD ABC
V 2 ABCD ABD
V 2 BCD
V 2 ABCD ACD
V 2 ABCD
V2
No exact tests exist on main effects or two-factor interactions involving the fixed factor A. To test the fixed
factor A use
A: F
MS A MS ABC MS ABD MS ACD
MS AB MS AC MS AD MS ABCD
MS D MS ABCD
MS ABC MS ABD
Random main effects could be tested by, for example: D: F
MS AB MS ABCD
MS ABC MS ABD
For testing two-factor interactions involving A use: AB: F
The results can also be generated in Minitab as follows:
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
random
2
H
random
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
DF
1
1
SS
6.13
0.13
MS
6.13
0.13
F
**
0.04
P
0.907 x
12-25
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
1
1
1
1
1
1
1
1
1
1
16
31
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
0.36
0.04
0.11
1.00
0.11
1.00
1.00
1.00
1.00
9.00
1.00
0.25
0.25
0.761
0.907
0.796
0.667
0.796
0.500
0.500
0.500
0.500
0.205
0.500
0.622
0.622
x
x
x
x
x
x Not an exact F-test.
** Denominator of F-test is zero.
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 A
*
(16) + 2(15) + 4(13) + 4(12) + 4(11) + 8(7) + 8(6)
+ 8(5) + 16Q[1]
2 B
-0.1875
*
(16) + 4(14) + 8(9) + 8(8) + 16(2)
3 C
-0.1250
*
(16) + 4(14) + 8(10) + 8(8) + 16(3)
4 D
-0.1875
*
(16) + 4(14) + 8(10) + 8(9) + 16(4)
5 A*B
-3.1250
*
(16) + 2(15) + 4(12) + 4(11) + 8(5)
6 A*C
0.0000
*
(16) + 2(15) + 4(13) + 4(11) + 8(6)
7 A*D
-3.1250
*
(16) + 2(15) + 4(13) + 4(12) + 8(7)
8 B*C
0.0000 14
(16) + 4(14) + 8(8)
9 B*D
0.0000 14
(16) + 4(14) + 8(9)
10 C*D
0.0000 14
(16) + 4(14) + 8(10)
11 A*B*C
0.0000 15
(16) + 2(15) + 4(11)
12 A*B*D
6.2500 15
(16) + 2(15) + 4(12)
13 A*C*D
0.0000 15
(16) + 2(15) + 4(13)
14 B*C*D
-2.3125 16
(16) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 A
2 B
3 C
4 D
5 A*B
6 A*C
7 A*D
Error DF
0.56
0.33
0.33
0.33
0.98
0.33
0.98
Error MS
*
3.13
3.13
3.13
28.13
3.13
28.13
Synthesis of Error MS
(5) + (6) + (7) - (11) - (12) - (13) + (15)
(8) + (9) - (14)
(8) + (10) - (14)
(9) + (10) - (14)
(11) + (12) - (15)
(11) + (13) - (15)
(12) + (13) - (15)
(d) A and B are fixed and C and D are random.
Factor
Wi
Ej
Jk
Gl
(WE ) ij
(WJ ) ik
(WG ) il
F
a
i
0
a
a
a
0
0
0
F
b
j
b
0
b
b
0
b
b
R
c
k
c
c
1
c
c
1
c
R
d
l
d
d
d
1
d
d
1
R
n
h
n
n
n
n
n
n
n
12-26
E(MS)
V 2 ACD AD AC A
V 2 BCD BC BD B
V 2 CD C
V 2 CD D
V 2 ABCD ABC ABD AB
V 2 ACD AC
V 2 ACD AD
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
( EJ ) jk
( EG ) jl
(JG ) kl
(WEJ ) ijk
(WEG ) ijl
( EJG ) jkl
(WJG ) ikl
(WEJG ) ijkl
H (ijkl ) h
a
a
a
0
0
a
0
0
1
0
0
b
0
0
0
b
0
1
1
c
1
1
c
1
1
1
1
d
1
1
d
1
1
1
1
1
n
n
n
n
n
n
n
n
1
V 2 BCD BC
V 2 BCD BD
V 2 CD
V 2 ABCD ABC
V 2 ABCD ABD
V 2 BCD
V 2 ACD
V 2 ABCD
V2
There are no exact tests on the fixed factors A and B, or their two-factor interaction AB. The appropriate
test statistics are:
A: F
B: F
AB: F
MS A MS ACD
MS AC MS AD
MS B MS BCD
MS BC MS BD
MS AB MS ABCD
MS ABC MS ABD
The results can also be generated in Minitab as follows:
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
fixed
2
H
random
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
31
SS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
MS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
F
1.96
0.04
0.36
0.04
0.11
1.00
1.00
1.00
1.00
0.25
1.00
9.00
0.25
0.25
0.25
P
0.604 x
0.907 x
0.656
0.874
0.796 x
0.500
0.500
0.500
0.500
0.622
0.500
0.205
0.622
0.622
0.622
x Not an exact F-test.
Source
1 A
2 B
3 C
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
*
(16) + 4(13) + 8(7) + 8(6) + 16Q[1]
*
(16) + 4(14) + 8(9) + 8(8) + 16Q[2]
-0.1250 10
(16) + 8(10) + 16(3)
12-27
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4
5
6
7
8
9
10
11
12
13
14
15
16
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
-0.1875
10
*
13
13
14
14
16
15
15
16
16
16
0.0000
0.0000
0.0000
0.0000
-1.1563
0.0000
6.2500
-2.3125
-2.3125
-4.6250
12.3750
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
(16)
+
+
+
+
+
+
+
+
+
+
+
+
8(10)
2(15)
4(13)
4(13)
4(14)
4(14)
8(10)
2(15)
2(15)
4(13)
4(14)
2(15)
+
+
+
+
+
+
16(4)
4(12) + 4(11) + 8Q[5]
8(6)
8(7)
8(8)
8(9)
+ 4(11)
+ 4(12)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 A
2 B
5 A*B
Error DF
0.33
0.33
0.98
Error MS
3.13
3.13
28.13
Synthesis of
(6) + (7) (8) + (9) (11) + (12)
Error MS
(13)
(14)
- (15)
(e) A, B and C are fixed and D is random.
Factor
Wi
Ej
Jk
Gl
(WE ) ij
(WJ ) ik
(WG ) il
( EJ ) jk
( EG ) jl
(JG ) kl
(WEJ ) ijk
(WEG ) ijl
( EJG ) jkl
(WJG ) ikl
(WEJG ) ijkl
H (ijkl ) h
F
a
i
0
a
a
a
0
0
0
a
a
a
0
0
a
0
0
1
F
b
j
b
0
b
b
0
b
b
0
0
b
0
0
0
b
0
1
F
c
k
c
c
0
c
c
0
c
0
c
0
0
c
0
0
0
1
R
d
l
d
d
d
1
d
d
1
d
1
1
d
1
1
1
1
1
R
n
h
n
n
n
n
n
n
n
n
n
n
n
n
n
n
n
1
E(MS)
V 2 AD A
V 2 BD B
V 2 CD C
V2 D
V 2 ABD AB
V 2 ACD AC
V 2 AD
V 2 BCD BC
V 2 BD
V 2 CD
V 2 ABCD ABC
V 2 ABD
V 2 BCD
V 2 ACD
V 2 ABCD
V2
There are exact tests for all effects. The results can also be generated in Minitab as follows:
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
fixed
2
H
fixed
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
DF
SS
MS
F
12-28
P
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
31
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
1.96
0.04
0.36
0.01
0.11
1.00
0.25
1.00
0.25
0.25
1.00
2.27
0.25
0.25
0.25
0.395
0.874
0.656
0.921
0.795
0.500
0.622
0.500
0.622
0.622
0.500
0.151
0.622
0.622
0.622
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 A
7
(16) + 8(7) + 16Q[1]
2 B
9
(16) + 8(9) + 16Q[2]
3 C
10
(16) + 8(10) + 16Q[3]
4 D
-0.7656 16
(16) + 16(4)
5 A*B
12
(16) + 4(12) + 8Q[5]
6 A*C
13
(16) + 4(13) + 8Q[6]
7 A*D
-1.1563 16
(16) + 8(7)
8 B*C
14
(16) + 4(14) + 8Q[8]
9 B*D
-1.1563 16
(16) + 8(9)
10 C*D
-1.1563 16
(16) + 8(10)
11 A*B*C
15
(16) + 2(15) + 4Q[11]
12 A*B*D
3.9375 16
(16) + 4(12)
13 A*C*D
-2.3125 16
(16) + 4(13)
14 B*C*D
-2.3125 16
(16) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
12-20 Reconsider cases (c), (d) and (e) of Problem 12-19. Obtain the expected mean squares assuming the
unrestricted model. You may use a computer package such as Minitab. Compare your results with those
for the restricted model.
A is fixed and B, C, and D are random.
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
random
2
H
random
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
DF
1
1
1
1
1
1
1
1
1
1
1
1
SS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
MS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
F
**
**
0.36
**
0.11
1.00
0.11
1.00
0.11
1.00
1.00
9.00
P
0.843 x
0.796
0.667
0.796
0.667
0.796
0.667
0.500
0.205
12-29
x
x
x
x
x
x
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
16
31
3.13
3.13
3.13
198.00
264.88
3.13
3.13
3.13
12.38
1.00
1.00
0.25
0.500
0.500
0.622
x Not an exact F-test.
** Denominator of F-test is zero.
Source
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 A
*
(16) + 2(15) + 4(13) + 4(12) + 4(11)
+ 8(5) + Q[1]
2 B
1.3750
*
(16) + 2(15) + 4(14) + 4(12) + 4(11)
+ 8(5) + 16(2)
3 C
-0.1250
*
(16) + 2(15) + 4(14) + 4(13) + 4(11)
+ 8(6) + 16(3)
4 D
1.3750
*
(16) + 2(15) + 4(14) + 4(13) + 4(12)
+ 8(7) + 16(4)
5 A*B
-3.1250
*
(16) + 2(15) + 4(12) + 4(11) + 8(5)
6 A*C
0.0000
*
(16) + 2(15) + 4(13) + 4(11) + 8(6)
7 A*D
-3.1250
*
(16) + 2(15) + 4(13) + 4(12) + 8(7)
8 B*C
0.0000
*
(16) + 2(15) + 4(14) + 4(11) + 8(8)
9 B*D
-3.1250
*
(16) + 2(15) + 4(14) + 4(12) + 8(9)
10 C*D
0.0000
*
(16) + 2(15) + 4(14) + 4(13) + 8(10)
11 A*B*C
0.0000 15
(16) + 2(15) + 4(11)
12 A*B*D
6.2500 15
(16) + 2(15) + 4(12)
13 A*C*D
0.0000 15
(16) + 2(15) + 4(13)
14 B*C*D
0.0000 15
(16) + 2(15) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
+ 8(7) + 8(6)
+ 8(9) + 8(8)
+ 8(10) + 8(8)
+ 8(10) + 8(9)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 A
2 B
3 C
4 D
5 A*B
6 A*C
7 A*D
8 B*C
9 B*D
10 C*D
Error DF
0.56
0.56
0.14
0.56
0.98
0.33
0.98
0.33
0.98
0.33
Error MS
*
*
3.13
*
28.13
3.13
28.13
3.13
28.13
3.13
Synthesis of
(5) + (6) +
(5) + (8) +
(6) + (8) +
(7) + (9) +
(11) + (12)
(11) + (13)
(12) + (13)
(11) + (14)
(12) + (14)
(13) + (14)
Error MS
(7) - (11) - (12) - (13) + (15)
(9) - (11) - (12) - (14) + (15)
(10) - (11) - (13) - (14) + (15)
(10) - (12) - (13) - (14) + (15)
- (15)
- (15)
- (15)
- (15)
- (15)
- (15)
A and B are fixed and C and D are random.
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
fixed
2
H
random
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
B
C
D
A*B
A*C
A*D
DF
1
1
1
1
1
1
1
SS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
MS
6.13
0.13
1.13
0.13
3.13
3.13
3.13
F
1.96
0.04
0.36
**
0.11
1.00
0.11
P
0.604 x
0.907 x
0.843 x
0.796 x
0.667 x
0.796 x
12-30
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
1
1
1
1
1
16
31
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
1.00
0.11
1.00
1.00
9.00
1.00
1.00
0.25
0.667 x
0.796 x
0.667 x
0.500
0.205
0.500
0.500
0.622
x Not an exact F-test.
** Denominator of F-test is zero.
Source
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 A
*
(16) + 2(15) + 4(13) + 4(12) + 4(11)
+ Q[1,5]
2 B
*
(16) + 2(15) + 4(14) + 4(12) + 4(11)
+ Q[2,5]
3 C
-0.1250
*
(16) + 2(15) + 4(14) + 4(13) + 4(11)
+ 8(6) + 16(3)
4 D
1.3750
*
(16) + 2(15) + 4(14) + 4(13) + 4(12)
+ 8(7) + 16(4)
5 A*B
*
(16) + 2(15) + 4(12) + 4(11) + Q[5]
6 A*C
0.0000
*
(16) + 2(15) + 4(13) + 4(11) + 8(6)
7 A*D
-3.1250
*
(16) + 2(15) + 4(13) + 4(12) + 8(7)
8 B*C
0.0000
*
(16) + 2(15) + 4(14) + 4(11) + 8(8)
9 B*D
-3.1250
*
(16) + 2(15) + 4(14) + 4(12) + 8(9)
10 C*D
0.0000
*
(16) + 2(15) + 4(14) + 4(13) + 8(10)
11 A*B*C
0.0000 15
(16) + 2(15) + 4(11)
12 A*B*D
6.2500 15
(16) + 2(15) + 4(12)
13 A*C*D
0.0000 15
(16) + 2(15) + 4(13)
14 B*C*D
0.0000 15
(16) + 2(15) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
+ 8(7) + 8(6)
+ 8(9) + 8(8)
+ 8(10) + 8(8)
+ 8(10) + 8(9)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 A
2 B
3 C
4 D
5 A*B
6 A*C
7 A*D
8 B*C
9 B*D
10 C*D
Error DF
0.33
0.33
0.14
0.56
0.98
0.33
0.98
0.33
0.98
0.33
Error MS
3.13
3.13
3.13
*
28.13
3.13
28.13
3.13
28.13
3.13
Synthesis of
(6) + (7) (8) + (9) (6) + (8) +
(7) + (9) +
(11) + (12)
(11) + (13)
(12) + (13)
(11) + (14)
(12) + (14)
(13) + (14)
Error MS
(13)
(14)
(10) - (11) - (13) - (14) + (15)
(10) - (12) - (13) - (14) + (15)
- (15)
- (15)
- (15)
- (15)
- (15)
- (15)
(e) A, B and C are fixed and D is random.
Minitab Output
ANOVA: y versus A, B, C, D
Factor
A
B
C
D
Type Levels Values
fixed
2
H
fixed
2
H
fixed
2
H
random
2
H
L
L
L
L
Analysis of Variance for y
Source
A
DF
1
SS
6.13
MS
6.13
F
1.96
P
0.395
12-31
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B
C
D
A*B
A*C
A*D
B*C
B*D
C*D
A*B*C
A*B*D
A*C*D
B*C*D
A*B*C*D
Error
Total
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
31
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
198.00
264.88
0.13
1.13
0.13
3.13
3.13
3.13
3.13
3.13
3.13
3.13
28.13
3.13
3.13
3.13
12.38
0.04
0.36
**
0.11
1.00
0.11
1.00
0.11
1.00
1.00
9.00
1.00
1.00
0.25
0.874
0.656
0.795
0.500
0.796 x
0.500
0.796 x
0.667 x
0.500
0.205
0.500
0.500
0.622
x Not an exact F-test.
** Denominator of F-test is zero.
Source
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 A
7
(16) + 2(15) + 4(13) + 4(12) + 8(7) + Q[1,5,6,11]
2 B
9
(16) + 2(15) + 4(14) + 4(12) + 8(9) + Q[2,5,8,11]
3 C
10
(16) + 2(15) + 4(14) + 4(13) + 8(10) + Q[3,6,8,11]
4 D
1.3750
*
(16) + 2(15) + 4(14) + 4(13) + 4(12) + 8(10) + 8(9)
+ 8(7) + 16(4)
5 A*B
12
(16) + 2(15) + 4(12) + Q[5,11]
6 A*C
13
(16) + 2(15) + 4(13) + Q[6,11]
7 A*D
-3.1250
*
(16) + 2(15) + 4(13) + 4(12) + 8(7)
8 B*C
14
(16) + 2(15) + 4(14) + Q[8,11]
9 B*D
-3.1250
*
(16) + 2(15) + 4(14) + 4(12) + 8(9)
10 C*D
0.0000
*
(16) + 2(15) + 4(14) + 4(13) + 8(10)
11 A*B*C
15
(16) + 2(15) + Q[11]
12 A*B*D
6.2500 15
(16) + 2(15) + 4(12)
13 A*C*D
0.0000 15
(16) + 2(15) + 4(13)
14 B*C*D
0.0000 15
(16) + 2(15) + 4(14)
15 A*B*C*D
-4.6250 16
(16) + 2(15)
16 Error
12.3750
(16)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
4 D
7 A*D
9 B*D
10 C*D
Error DF
0.56
0.98
0.98
0.33
Error MS
*
28.13
28.13
3.13
Synthesis of
(7) + (9) +
(12) + (13)
(12) + (14)
(13) + (14)
Error MS
(10) - (12) - (13) - (14) + (15)
- (15)
- (15)
- (15)
12-21 In Problem 5-17, assume that the three operators were selected at random. Analyze the data under
these conditions and draw conclusions. Estimate the variance components.
Minitab Output
ANOVA: Score versus Cycle Time, Operator, Temperature
Factor
Type Levels Values
Cycle Ti fixed
3
40
Operator random
3
1
Temperat fixed
2
300
50
2
350
60
3
Analysis of Variance for Score
Source
Cycle Ti
Operator
Temperat
DF
2
2
1
SS
436.000
261.333
50.074
MS
218.000
130.667
50.074
12-32
F
2.45
39.86
8.89
P
0.202
0.000
0.096
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Cycle Ti*Operator
Cycle Ti*Temperat
Operator*Temperat
Cycle Ti*Operator*Temperat
Error
Total
4
2
2
4
36
53
355.667
78.815
11.259
46.185
118.000
1357.333
88.917
39.407
5.630
11.546
3.278
27.13
3.41
1.72
3.52
0.000
0.137
0.194
0.016
Source
1
2
3
4
5
6
7
8
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Cycle Ti
4
(8) + 6(4) + 18Q[1]
Operator
7.0772
8
(8) + 18(2)
Temperat
6
(8) + 9(6) + 27Q[3]
Cycle Ti*Operator
14.2731
8
(8) + 6(4)
Cycle Ti*Temperat
7
(8) + 3(7) + 9Q[5]
Operator*Temperat
0.2613
8
(8) + 9(6)
Cycle Ti*Operator*Temperat
2.7562
8
(8) + 3(7)
Error
3.2778
(8)
The following calculations agree with the Minitab results:
2
V WEJ
2
VEJ
V 2 MS E
Vˆ 2 3.27778
MS ABC MS E
11.546296 3.277778
2
Vˆ WEJ
2.7562
n
3
MS BC MS E
88.91667 3.277778
2
Vˆ EJ
14.27315
an
23
V WJ2
MS AC MS E
bn
V J2
MS C MS E
abn
5.629630 3.277778
33
Vˆ WJ2
Vˆ J2
130.66667 3.277778
23 3
0.26132
7.07716
12-22 Consider the three-factor model
yijk
P W i E j J k WE
ij
EJ
jk
H ijk
Assuming that all the factors are random, develop the analysis of variance table, including the expected
mean squares. Propose appropriate test statistics for all effects.
Source
DF
E(MS)
A
a-1
2
V 2 cV WE
bcV W2
B
b-1
2
2
V 2 cV WE
aV EJ
acV E2
C
c-1
2
V 2 aV EJ
abV J2
AB
(a-1)(b-1)
2
V 2 cV WE
BC
(b-1)(c-1)
2
V 2 aV EJ
Error (AC + ABC)
Total
b(a-1)(c-1)
abc-1
V2
There are exact tests for all effects except B. To test B, use the statistic F
12-33
MS B MS E
MS AB MS BC
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
12-23 The three-factor model for a single replicate is
yijk
P W i E j J k ( WE) ij (EJ ) jk ( WJ ) ik ( WEJ ) ijk H ijk
If all the factors are random, can any effects be tested? If the three-factor interaction and the ( WE) ij
interaction do not exist, can all the remaining effects be tested.
The expected mean squares are found by referring to Table 12-9, deleting the line for the error term H( ijk ) l
and setting n=1. The three-factor interaction now cannot be tested; however, exact tests exist for the twofactor interactions and approximate F tests can be conducted for the main effects. For example, to test the
main effect of A, use
F
MS A MS ABC
MS AB MS AC
If (WEJ ) ijk and (WE ) ij can be eliminated, the model becomes
y ijk
P W i E j J k WE
ij
EJ
jk
WJ
ik
WEJ
ijk
H ijk
For this model, the analysis of variance is
Source
DF
E(MS)
A
a-1
V 2 bV WJ2 bcV W2
B
b-1
2
V 2 aV EJ
acV E2
C
c-1
2
V 2 aV EJ
bV WJ2 abV J2
AC
(a-1)(c-1)
V 2 bV WJ2
BC
(b-1)(c-1)
2
V 2 aV EJ
Error (AB + ABC)
Total
c(a-1)(b-1)
abc-1
V2
There are exact tests for all effect except C. To test the main effect of C, use the statistic:
F
MS C MS E
MS BC MS AC
12-24 In Problem 5-6, assume that both machines and operators were chosen randomly. Determine the
power of the test for detecting a machine effect such that V 2E V 2 , where V E2 is the variance component
for the machine factor. Are two replicates sufficient?
O
If V E2
V 2 , then an estimate of V 2
V E2
1
anV E2
2
V 2 nV WE
3.79 , and an estimate of V 2
of variance table. Then
12-34
2
nV WE
7.45 , from the analysis
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
O
1
3 2 3.79
7.45
2.22
1.49
and the other OC curve parameters are X1 3 and X 2 6 . This results in E | 0.75 approximately, with
D 0.05 , or E | 0.9 with D 0.01 . Two replicates does not seem sufficient.
>
12-25 In the two-factor mixed model analysis of variance, show that Cov WE ij , WE
i' j
@
2
1 a WEV
for
izi'.
º
ª a
WE ij »
0 (constant) we have V «
«¬ i 1
»¼
a
Since
¦ WE
¦
ij
i 1
a
¦ V WE
i 1
ij
>
§a·
2¨¨ ¸¸Cov WE ij , WE
© 2¹
i' j
@
0 , which implies that
0
>
a!
ª a 1º 2
a«
»V WE 2! a 2 ! 2 Cov WE ij , WE
a
¬
¼
>
2
a a 1 Cov WE ij ,W E
a 1 V WE
>
Cov W E ij , WE
i' j
@
i' j
@
i' j
@
0
0
§1· 2
¨ ¸V WE
©a¹
12-26 Show that the method of analysis of variance always produces unbiased point estimates of the
variance component in any random or mixed model.
Let g be the vector of mean squares from the analysis of variance, chosen so that E(g) does not contain any
fixed effects. Let V 2 be the vector of variance components such that E(g) AV 2 , where A is a matrix of
constants. Now in the analysis of variance method of variance component estimation, we equate observed
and expected mean squares, i.e.
g = As 2 ˆs 2
A -1 g
Since A -1 always exists then,
E s 2 = E A -1 g
A -1 E g = A -1 As 2
s2
Thus V 2 is an unbiased estimator of V 2 . This and other properties of the analysis of variance method are
discussed by Searle (1971a).
12-27 Invoking the usual normality assumptions, find an expression for the probability that a negative
estimate of a variance component will be obtained by the analysis of variance method. Using this result,
write a statement giving the probability that V 2W 0 in a one-factor analysis of variance. Comment on the
usefulness of this probability statement.
12-35
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
MS1 MS2
, where MSi for i=1,2 are two mean squares and c is a constant. The
c
probability that V̂ W2 0 (negative) is
Suppose V 2
^
`
P Vˆ 2 0
MS1
½
1¾
P^MS1 MS 2 0` P ®
¯ MS 2
¿
MS1
° E MS
E MS1
°
1
P®
MS
E MS 2
2
°
°¯ E MS 2
½
°
°
¾
°
°¿
E MS1 ½
P ®Fu ,v
¾
E MS 2 ¿
¯
where u is the number of degrees of freedom for MS1 and v is the number of degrees of freedom for MS 2 .
For the one-way model, this equation reduces to
^
`
P V 2 0
where k
°
½°
V2
P ® Fa 1, N a 2
2¾
V nV W °¿
°̄
1 ½
P ® Fa 1, N a
¾
1 nk ¿
¯
V W2
. Using arbitrary values for some of the parameters in this equation will give an
V2
experimenter some idea of the probability of obtaining a negative estimate of V̂ W2 0 .
12-28 Analyze the data in Problem 12-9, assuming that the operators are fixed, using both the unrestricted
and restricted forms of the mixed models. Compare the results obtained from the two models.
The restricted model is as follows:
Minitab Output
ANOVA: Measurement versus Part, Operator
Factor
Part
Type Levels Values
random
10
1
8
Operator fixed
2
1
2
9
2
3
10
4
5
F
7.33
0.69
0.40
P
0.000
0.427
0.927
6
7
Analysis of Variance for Measurem
Source
Part
Operator
Part*Operator
Error
Total
DF
9
1
9
40
59
SS
99.017
0.417
5.417
60.000
164.850
MS
11.002
0.417
0.602
1.500
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Part
1.5836
4
(4) + 6(1)
Operator
3
(4) + 3(3) + 30Q[2]
Part*Operator -0.2994
4
(4) + 3(3)
Error
1.5000
(4)
The second approach is the unrestricted mixed model.
Minitab Output
ANOVA: Measurement versus Part, Operator
Factor
Part
Type Levels Values
random
10
1
8
Operator fixed
2
1
2
9
2
3
10
4
12-36
5
6
7
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Analysis of Variance for Measurem
Source
Part
Operator
Part*Operator
Error
Total
DF
9
1
9
40
59
SS
99.017
0.417
5.417
60.000
164.850
MS
11.002
0.417
0.602
1.500
F
18.28
0.69
0.40
P
0.000
0.427
0.927
Source
1
2
3
4
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
Part
1.7333
3
(4) + 3(3) + 6(1)
Operator
3
(4) + 3(3) + Q[2]
Part*Operator -0.2994
4
(4) + 3(3)
Error
1.5000
(4)
Source
Sum of
Squares
DF
Mean
Square
F-test
E(MS)
F
a
A
0.416667
a-1=1
0.416667
B
99.016667
b-1=9
11.00185
5.416667
(a-1)(b-1)=9
60.000000
164.85000
40
nabc-1=59
AB
Error
Total
2
bn
V 2 nV WE
¦W
2
i
i 1
MS A
MS AB
F
a 1
2
V 2 nV WE
anV E2
F
0.60185
2
V 2 nV WE
F
1.50000
V2
0.692
MS B
MS AB
MS AB
MS E
18.28
0.401
In the unrestricted model, the F-test for B is different. The F-test for B in the unrestricted model should
generally be more conservative, since MSAB will generally be larger than MSE. However, this is not the
case with this particular experiment.
12-29 Consider the two-factor mixed model. Show that the standard error of the fixed factor mean (e.g. A)
is >MS AB / bn@1 2 .
The standard error is often used in Duncan’s Multiple Range test. Duncan’s Multiple Range Test requires
the variance of the difference in two means, say
V y i .. y m..
where rows are fixed and columns are random. Now, assuming all model parameters to be independent, we
have the following:
y i .. y m ..
W i W m
1
b
b
¦
WE
j 1
ij
1
b
b
¦
WE
mj
j 1
1
bn
b
n
¦¦
H ijk
j 1 k 1
1
bn
b
n
¦¦ H
mjk
j 1 k 1
and
2
V yi .. ym ..
2
2
2
§1·
§1·
§ 1 ·
§ 1 ·
2
2
¨ ¸ bV WE
¨ ¸ bnV 2 ¨ ¸ bnV 2
¨ ¸ bV WE
b
b
bn
© ¹
© ¹
© ¹
© bn ¹
12-37
2
2 V 2 nV WE
bn
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2
, we would use
Since MS AB estimates V 2 nV WE
2 MS AB
bn
as the standard error to test the difference. However, the table of ranges for Duncan’s Multiple Range test
already include the constant 2.
12-30 Consider the variance components in the random model from Problem 12-9.
(a) Find an exact 95 percent confidence interval on V2.
f E MS E
FD2 2, f E
dV2 d
f E MS E
F12D
2, f E
40 1.5
40 1.5
dV 2 d
59.34
24.43
2
1011
.
d V d 2.456
(b) Find approximate 95 percent confidence intervals on the other variance components using the
Satterthwaite method.
2
V WE
and V W2 are negative, and the Satterthwaithe method does not apply. The confidence interval on V E2
is
VE2
r
MS B MS AB
an
Vˆ E2
MSB MS AB 2
2
MSB2
MS AB
b 1
a 1 b 1
rVO2
FD2 2,r
11.001852 0.6018519
1.7333
23
11.001852 0.6018519 2
1.0018522 0.60185192
9
1 9
2
rV E
d V E2 d 2
F1D 2,r
8.01826
8.01826 1.7333
8.01826 1.7333
d V E2 d
17.55752
2.18950
2
0.79157 d V E d 6.34759
12-31 Use the experiment described in Problem 5-6 and assume that both factor are random. Find an exact
95 percent confidence interval on V2. Construct approximate 95 percent confidence interval on the other
variance components using the Satterthwaite method.
V 2 MS E
V 2 3.79167
f E MS E
f E MS E
dV2 d 2
FD2 2, f E
F1D 2, f E
12 3.79167
12 3.79167
dV 2 d
23.34
4.40
12-38
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
19494
.
d V 2 d 10.3409
Satterthwaite Method:
2
VWE
MS AB MS E
n
MS AB MS E
r
2
V WE
2
7.44444 3.79167
2
7.44444 3.79167
2
MS AB
MS E2
a 1 b 1
df E
.
182639
2
2.2940
7.44444 2 3.79167 2
2 3
12
rV E2
FD2 2 ,r
d V E2 d
rV E2
F12D
2 ,r
2.2940 1.82639
2.2940 1.82639
d V E2 d
7.95918
0.09998
2
0.52640 d V E d 4190577
.
V E2 0 , this variance component does not have a confidence interval using Satterthwaite’s Method.
V W2
MS A MS AB
bn
Vˆ W2
MS A MS AB 2
2
MS A2
MS AB
a 1
a 1 b 1
rV W2
r
FD2 2 ,r
80.16667 7.44444
42
9.09028
80.16667 7.44444 2
80.166672 7.444442
2
2 3
2
rV W
d V W2 d 2
F1D 2 ,r
1.64108
(164108
.
)(9.09028)
(164108
.
)(9.09028)
d V W2 d
6.53295
0.03205
2.28348 d V W2 d 465.45637
12-32 Consider the three-factor experiment in Problem 5-17 and assume that operators were selected at
random. Find an approximate 95 percent confidence interval on the operator variance component.
MS C MS E
abn
V J2
r
Vˆ J2
MSC MS E 2
MSC2 MS E2
c 1
df E
130.66667 3.277778
23 3
130.66667 3.27778 2
130.666672 3.277782
2
36
rV J2
FD2 2 ,r
d V J2 d
1.90085
rV J2
F12D
2 ,r
1.90085 7.07716
1.90085 7.07716
d V J2 d
9.15467
0.04504
2
.
d V J d 4298.66532
146948
12-39
7.07716
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
12-33 Rework Problem 12-30 using the modified large-sample approach described in Section 12-7.2.
Compare the two sets of confidence intervals obtained and discuss.
V O2
G1 1
1
F0.05 ,9 ,f
1
H1
F.95 ,9i ,f
1
V E2
1
1.88
1
1
F .295 ,9
MS B MS AB
an
11.001852 0.6018519
1.7333
23
Vˆ O2
0.46809
1 .
1
1 1.7027
0.370
9
2
FD , f i , f j 1 G12 FD , f i , f j H12
Gij
3.18 1 2 0.46809 2 3.18 1.70272
3.18
FD , f i , f j
VL
2
G12 c12 MS B2 H 12 c 22 MS AB
G11 c1 c 2 MS B MS AB
2
VL
VL
0.36366
2
§1·
§1·
§ 1 ·§ 1 ·
0.46809 2 ¨ ¸ 11.00185 2 1.7027 2 ¨ ¸ 0.60185 2 0.36366 ¨ ¸¨ ¸ 11.00185 0.60185
6
6
© ¹
© ¹
© 6 ¹© 6 ¹
0.83275
V E2 V L
L
.
0.83275
17333
0.82075
12-34 Rework Problem 12-32 using the modified large-sample method described in Section 12-7.2.
Compare this confidence interval with he one obtained previously and discuss.
V J2
G1 1
H1
1
F0.05 ,3,f
1
F.95 ,36 ,f
1
1
1
2.60
1
F .295 ,36
MS C MS E
abn
Vˆ J2
130.66667 3.277778
23 3
7.07716
0.61538
1 .
1
1 0.54493
0.64728
36
Gij
VL
FD , f i , f j 1 2 G12 FD , fi , f j H12
FD , fi , f j
VL
L
0.74542
2
G11 c1 c 2 MS B MS AB
G12 c12 MS B2 H 12 c 22 MS AB
2
VL
2.88 1 2 0.61538 2 2.88 0.544932
2.88
§1·
0.61538 2 ¨ ¸ 130.66667
© 18 ¹
20.95112
V J2 V L
2
2
§1·
§ 1 ·§ 1 ·
0.54493 2 ¨ ¸ 3.27778 2 0.74542 ¨ ¸¨ ¸ 130.66667 3.27778
© 18 ¹
© 18 ¹© 18 ¹
7.07716 20.95112
2.49992
12-40
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 13
Nested and Split-Plot Designs
Solutions
In this chapter we have not shown residual plots and other diagnostics to conserve space. A complete
analysis would, of course, include these model adequacy checking procedures.
13-1 A rocket propellant manufacturer is studying the burning rate of propellant from three production
processes. Four batches of propellant are randomly selected from the output of each process and three
determinations of burning rate are made on each batch. The results follow. Analyze the data and draw
conclusions.
Batch
Process 2
1 2 3 4
19 23 18 35
17 24 21 27
14 21 17 25
Process 1
1 2 3 4
25 19 15 15
30 28 17 16
26 20 14 13
Process 3
1 2 3 4
14 35 38 25
15 21 54 29
20 24 50 33
Minitab Output
ANOVA: Burn Rate versus Process, Batch
Factor
Type Levels Values
Process
fixed
3
1
Batch(Process) random
4
1
2
2
3
3
4
Analysis of Variance for Burn Rat
Source
Process
Batch(Process)
Error
Total
DF
2
9
24
35
SS
676.06
2077.58
454.00
3207.64
MS
338.03
230.84
18.92
F
1.46
12.20
P
0.281
0.000
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Process
2
(3) + 3(2) + 12Q[1]
2 Batch(Process)
70.64
3
(3) + 3(2)
3 Error
18.92
(3)
There is no significant effect on mean burning rate among the different processes; however, different
batches from the same process have significantly different burning rates.
13-2 The surface finish of metal parts made on four machines is being studied. An experiment is
conducted in which each machine is run by three different operators and two specimens from each
operator are collected and tested. Because of the location of the machines, different operators are used on
each machine, and the operators are chosen at random. The data are shown in the following table.
Analyze the data and draw conclusions.
Operator
Machine 1
1 2 3
79 94 46
62 74 57
Machine 2
1 2 3
92 85 76
99 79 68
Minitab Output
ANOVA: Finish versus Machine, Operator
13-1
Machine 3
1 2 3
88 53 46
75 56 57
Machine 4
1 2 3
36 40 62
53 56 47
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Type Levels Values
Machine
fixed
4
1
Operator(Machine) random
3
1
2
2
3
3
4
Analysis of Variance for Finish
Source
Machine
Operator(Machine)
Error
Total
DF
3
8
12
23
SS
3617.67
2817.67
1014.00
7449.33
MS
1205.89
352.21
84.50
F
3.42
4.17
P
0.073
0.013
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Machine
2
(3) + 2(2) + 6Q[1]
2 Operator(Machine)
133.85
3
(3) + 2(2)
3 Error
84.50
(3)
There is a slight effect on surface finish due to the different processes; however, the different operators
running the same machine have significantly different surface finish.
13-3 A manufacturing engineer is studying the dimensional variability of a particular component that is
produced on three machines. Each machine has two spindles, and four components are randomly selected
from each spindle. These results follow. Analyze the data, assuming that machines and spindles are fixed
factors.
Spindle
Machine 2
1
2
14
12
15
10
13
11
14
13
Machine 1
1
2
12
8
9
9
11
10
12
8
Machine 3
1
2
14
16
10
15
12
15
11
14
Minitab Output
ANOVA: Variability versus Machine, Spindle
Factor
Machine
Spindle(Machine)
Type Levels Values
fixed
3
1
fixed
2
1
2
2
3
Analysis of Variance for Variabil
Source
Machine
Spindle(Machine)
Error
Total
DF
2
3
18
23
SS
55.750
43.750
26.500
126.000
MS
27.875
14.583
1.472
F
18.93
9.91
P
0.000
0.000
There is a significant effects on dimensional variability due to the machine and spindle factors.
13-4 To simplify production scheduling, an industrial engineer is studying the possibility of assigning
one time standard to a particular class of jobs, believing that differences between jobs is negligible. To see
if this simplification is possible, six jobs are randomly selected. Each job is given to a different group of
three operators. Each operator completes the job twice at different times during the week, and the
following results were obtained. What are your conclusions about the use of a common time standard for
all jobs in this class? What value would you use for the standard?
Job
1
Operator 1
158.3 159.4
Operator 2
159.2 159.6
13-2
Operator 3
158.9 157.8
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
2
3
4
5
6
154.6
162.5
160.0
156.3
163.7
154.9
162.6
158.7
158.1
161.0
157.7
161.0
157.5
158.3
162.3
156.8
158.9
158.9
156.9
160.3
154.8
160.5
161.1
157.7
162.6
156.3
159.5
158.5
156.9
161.8
Minitab Output
ANOVA: Time versus Job, Operator
Factor
Type Levels Values
Job
random
6
1
Operator(Job) random
3
1
2
2
3
3
4
5
6
Analysis of Variance for Time
Source
Job
Operator(Job)
Error
Total
DF
5
12
18
35
SS
148.111
12.743
27.575
188.430
MS
29.622
1.062
1.532
F
27.89
0.69
P
0.000
0.738
Source
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Job
4.7601
2
(3) + 2(2) + 6(1)
2 Operator(Job) -0.2350
3
(3) + 2(2)
3 Error
1.5319
(3)
The jobs differ significantly; the use of a common time standard would likely not be a good idea.
13-5 Consider the three-stage nested design shown in Figure 13-5 to investigate alloy hardness. Using
the data that follow, analyze the design, assuming that alloy chemistry and heats are fixed factors and
ingots are random.
Alloy Chemistry
Heats
Ingots
1
1
2
3
1
2
2
1 2
40 27
63 30
1 2
95 69
67 47
1 2
65 78
54 45
1 2
22 23
10 39
1 2
83 75
62 64
Minitab Output
ANOVA: Hardness versus Alloy, Heat, Ingot
Factor
Type Levels Values
Alloy
fixed
2
1
Heat(Alloy)
fixed
3
1
Ingot(Alloy Heat) random
2
1
2
2
2
3
Analysis of Variance for Hardness
Source
Alloy
Heat(Alloy)
Ingot(Alloy Heat)
Error
Total
DF
1
4
6
12
23
SS
315.4
6453.8
2226.3
2141.5
11137.0
MS
315.4
1613.5
371.0
178.5
Source
F
0.85
4.35
2.08
P
0.392
0.055
0.132
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
1 Alloy
3
(4) + 2(3) + 12Q[1]
2 Heat(Alloy)
3
(4) + 2(3) + 4Q[2]
3 Ingot(Alloy Heat)
96.29
4
(4) + 2(3)
4 Error
178.46
(4)
13-3
3
1 2
61 35
77 42
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Alloy hardness differs significantly due to the different heats within each alloy.
13-6 Reanalyze the experiment in Problem 13-5 using the unrestricted form of the mixed model.
Comment on any differences you observe between the restricted and unrestricted model results. You may
use a computer software package.
Minitab Output
ANOVA: Hardness versus Alloy, Heat, Ingot
Factor
Type Levels Values
Alloy
fixed
2
1
Heat(Alloy)
fixed
3
1
Ingot(Alloy Heat) random
2
1
2
2
2
3
Analysis of Variance for Hardness
Source
Alloy
Heat(Alloy)
Ingot(Alloy Heat)
Error
Total
DF
1
4
6
12
23
SS
315.4
6453.8
2226.3
2141.5
11137.0
MS
315.4
1613.5
371.0
178.5
F
0.85
4.35
2.08
P
0.392
0.055
0.132
Source
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
1 Alloy
3
(4) + 2(3) + Q[1,2]
2 Heat(Alloy)
3
(4) + 2(3) + Q[2]
3 Ingot(Alloy Heat)
96.29
4
(4) + 2(3)
4 Error
178.46
(4)
13-7 Derive the expected means squares for a balanced three-stage nested design, assuming that A is
fixed and that B and C are random. Obtain formulas for estimating the variance components.
Wi
F
a
i
0
R
b
j
b
R
c
k
c
R
n
l
n
E j (i )
1
1
c
n
V 2 nV J2 cnV E2
J k ( ij)
1
1
1
n
V 2 nV J2
H (ijk ) l
1
1
1
1
V2
V 2
MS E
Factor
Vˆ J2
MS C
B
E(MS)
V 2 nV J2 cnV E2
MS E
Vˆ E2
n
MS B
The expected mean squares can be generated in Minitab as follows:
Minitab Output
ANOVA: y versus A, B, C
Factor
A
B(A)
C(A B)
Type Levels Values
fixed
2
-1
random
2
-1
random
2
-1
1
1
1
Analysis of Variance for y
Source
DF
SS
MS
F
13-4
P
bcn
a 1
A
¦W
2
i
MS C
cn
B
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A
B(A)
C(A B)
Error
Total
1
2
4
8
15
Source
1
2
3
4
0.250
8.500
49.000
46.000
103.750
0.250
4.250
12.250
5.750
0.06
0.35
2.13
0.831
0.726
0.168
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
2
(4) + 2(3) + 4(2) + 8Q[1]
-2.000
3
(4) + 2(3) + 4(2)
3.250
4
(4) + 2(3)
5.750
(4)
A
B(A)
C(A B)
Error
13-8 Repeat Problem 13-7 assuming the unrestricted form of the mixed model. You may use a computer
software package. Comment on any differences you observe between the restricted and unrestricted model
analysis and conclusions.
Minitab Output
ANOVA: y versus A, B, C
Factor
A
B(A)
C(A B)
Type Levels Values
fixed
2
-1
random
2
-1
random
2
-1
1
1
1
Analysis of Variance for y
Source
A
B(A)
C(A B)
Error
Total
DF
1
2
4
8
15
Source
1
2
3
4
A
B(A)
C(A B)
Error
SS
0.250
8.500
49.000
46.000
103.750
MS
0.250
4.250
12.250
5.750
F
0.06
0.35
2.13
P
0.831
0.726
0.168
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
2
(4) + 2(3) + 4(2) + Q[1]
-2.000
3
(4) + 2(3) + 4(2)
3.250
4
(4) + 2(3)
5.750
(4)
In this case there is no difference in results between the restricted and unrestricted models.
13-9 Derive the expected means squares for a balanced three-stage nested design if all three factors are
random. Obtain formulas for estimating the variance components. Assume the restricted form of the
mixed model.
Wi
R
a
i
1
R
b
j
b
R
c
k
c
R
n
l
n
E j (i )
1
1
c
n
V 2 nV J2 cnV E2
J k ( ij)
1
1
1
n
V 2 nV J2
H (ijk ) l
1
1
1
1
V2
Factor
V 2
MS E
V J2
MS C ( B ) MS E
n
V E2
E(MS)
V 2 nV J2 cnV E2 bcnV W2
MS B ( A ) MS C ( B )
cn
The expected mean squares can be generated in Minitab as follows:
13-5
V J2
MS A MS B ( A )
bcn
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Minitab Output
ANOVA: y versus A, B, C
Factor
A
B(A)
C(A B)
Type Levels Values
random
2
-1
random
2
-1
random
2
-1
1
1
1
Analysis of Variance for y
Source
A
B(A)
C(A B)
Error
Total
Source
1
2
3
4
A
B(A)
C(A B)
Error
DF
1
2
4
8
15
SS
0.250
8.500
49.000
46.000
103.750
MS
0.250
4.250
12.250
5.750
F
0.06
0.35
2.13
P
0.831
0.726
0.168
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
-0.5000
2
(4) + 2(3) + 4(2) + 8(1)
-2.0000
3
(4) + 2(3) + 4(2)
3.2500
4
(4) + 2(3)
5.7500
(4)
13-10 Verify the expected mean squares given in Table 13-1.
Wi
F
a
i
0
F
b
j
b
R
n
l
n
Ej i
1
0
n
H
1
1
1
Wi
R
a
i
1
R
b
j
b
R
n
l
n
Ej i
1
1
n
V 2 nV E2
H
1
1
1
V2
Wi
F
a
i
0
R
b
j
b
R
n
l
n
Ej i
1
1
n
V 2 nV E2
H
1
1
1
V2
Factor
ijk l
Factor
ijk l
Factor
ijk l
E(MS)
bn
W i2
a 1
n
V2
a b 1
V2
¦
¦¦ E
2
j i
V2
E(MS)
V 2 nV E2 bnV W2
E(MS)
V 2 nV E2
bn
a 1
¦W
2
i
13-11 Unbalanced designs. Consider an unbalanced two-stage nested design with bj levels of B under the
ith level of A and nij replicates in the ijth cell.
13-6
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(a) Write down the least squares normal equations for this situation. Solve the normal equations.
The least squares normal equations are:
a
P
n.. Pˆ
¦
i 1
Wi
bi
a
ni .Wˆ i
¦¦ n
ˆ
ij E j i
y ...
i 1 j 1
ni . Pˆ ni .Wˆ i
bi
¦n
ˆ
ij E j i
y i .. , for i
1,2,..., a
j 1
Ej i
nij Pˆ nijWˆ i nij Eˆ j i
yij . , for i
1,2,..., a and j
1,2,..., bi
There are 1+a+b equations in 1+a+b unknowns. However, there are a+1linear dependencies in these
equations, and consequently, a+1 side conditions are needed to solve them. Any convenient set of a+1
linearly independent equations can be used. The easiest set is P 0 , Wi 0 , for i=1,2,…,a. Using these
conditions we get
P
0 , Ej ( i )
0 , Wi
yij.
as the solution to the normal equations. See Searle (1971) for a full discussion.
(b) Construct the analysis of variance table for the unbalanced two-stage nested design.
The analysis of variance table is
Source
SS
a
y i2..
¦n
A
i 1
a
bi
i.
DF
y ...2
a-1
n..
y ij2 .
a
ij
i 1
y i2..
¦¦ n ¦ n
B
i 1 j 1
a
Error
bi
nij
¦¦¦
2
y ijk
i 1 j 1 k 1
a
bi
bi
nij
y ij2.
¦¦ n
i 1 j 1
¦¦¦
Total
a
b.-a
i.
2
y ijk
i 1 j 1 k 1
n..-b
ij
y ...2
n ..
n..-1
(c) Analyze the following data, using the results in part (b).
Factor A
Factor B
1
1
6
4
8
2
-3
1
1
5
7
9
6
2
2
2
4
3
3
1
0
-3
Note that a=2, b1=2, b2=3, b.=b1+b2=5, n11=3, n12=2, n21=4, n22=3 and n23=3
Source
SS
13-7
DF
MS
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
A
B
Error
Total
0.13
153.78
35.42
189.33
1
3
10
14
0.13
51.26
3.54
The analysis can also be performed in Minitab as follows. The adjusted sum of squares is utilized by
Minitab’s general linear model routine.
Minitab Output
General Linear Model: y versus A, B
Factor
A
B(A)
Type Levels Values
fixed
2 1 2
fixed
5 1 2 1 2 3
Analysis of Variance for y, using Adjusted SS for Tests
Source
A
B(A)
Error
Total
DF
1
3
10
14
Seq SS
0.133
153.783
35.417
189.333
Adj SS
0.898
153.783
35.417
Adj MS
0.898
51.261
3.542
F
0.25
14.47
P
0.625
0.001
13-12 Variance components in the unbalanced two-stage nested design. Consider the model
y ijk
P W i E j i H k ij
i 1,2 ,...,a
°
® j 1,2 ,...,b
°k 1,2,..., n
ij
¯
where A and B are random factors. Show that
V 2 c1V E2 c2V W2
E MS A
E MSB
V 2 c0V E2
A
V2
E MSE
where
§ bi nij2 ·
¨
¸
¨
¸
n
i 1 © j 1 i. ¹
ba
a § bi
nij2 ·¸ a
¨
¨
n ¸ i 1
i 1 © j 1 i. ¹
a 1
a
N
c0
¦¦
¦¦
c1
c2
nij2
¦¦ N
a
N
bi
¦n
2
i.
i 1
N
a 1
13-8
j 1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
See “Variance Component Estimation in the 2-way Nested Classification,” by S.R. Searle, Annals of
Mathematical Statistics, Vol. 32, pp. 1161-1166, 1961. A good discussion of variance component
estimation from unbalanced data is in Searle (1971a).
13-13 A process engineer is testing the yield of a product manufactured on three machines. Each
machine can be operated at two power settings. Furthermore, a machine has three stations on which the
product is formed. An experiment is conducted in which each machine is tested at both power settings,
and three observations on yield are taken from each station. The runs are made in random order, and the
results follow. Analyze this experiment, assuming all three factors are fixed.
Station
1
Power Setting 1 34.1
30.3
31.6
Power Setting 2 24.3
26.3
27.1
The linear model is y ijkl
Machine 1
2
3
33.7 36.2
34.9 36.8
35.0 37.1
28.1 25.7
29.3 26.1
28.6 24.9
P W i E j WE
1
32.1
33.5
34.0
24.1
25.0
26.3
ij
J k
Machine 2
2
3
33.1 32.8
34.7 35.1
33.9 34.3
24.1 26.0
25.1 27.1
27.9 23.9
j
WJ
ik ( j )
H
1
32.9
33.0
33.1
24.2
26.1
25.3
Machine 3
2
3
33.8 33.6
33.4 32.8
32.8 31.7
23.2 24.7
27.4 22.0
28.0 24.8
ijk l
Minitab Output
ANOVA: Yield versus Machine, Power, Station
Factor
Machine
Power
Station(Machine)
Type Levels Values
fixed
3
1
fixed
2
1
fixed
3
1
2
2
2
3
3
Analysis of Variance for Yield
Source
Machine
Power
Station(Machine)
Machine*Power
Power*Station(Machine)
Error
Total
DF
2
1
6
2
6
36
53
SS
21.143
853.631
32.583
0.616
28.941
58.893
995.808
MS
10.572
853.631
5.431
0.308
4.824
1.636
F
6.46
521.80
3.32
0.19
2.95
P
0.004
0.000
0.011
0.829
0.019
Source
1
2
3
4
5
6
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Machine
6
(6) + 18Q[1]
Power
6
(6) + 27Q[2]
Station(Machine)
6
(6) + 6Q[3]
Machine*Power
6
(6) + 9Q[4]
Power*Station(Machine)
6
(6) + 3Q[5]
Error
1.636
(6)
13-14 Suppose that in Problem 13-13 a large number of power settings could have been used and that the
two selected for the experiment were chosen randomly. Obtain the expected mean squares for this
situation and modify the previous analysis appropriately.
Wi
R
2
i
1
F
3
j
3
F
3
k
3
R
3
l
3
Ej
2
0
3
3
Factor
E(MS)
V 2 27V W2
2
V 2 9V WE
9
13-9
¦E
2
j
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
(WE ) ij
1
0
3
3
2
V 2 9V WE
J k( j)
2
1
0
3
V 2 3V WJ2
(WJ ) ik ( j )
1
1
0
3
V 2 3V WJ2
H (ijk ) l
1
1
1
1
V2
¦¦J
2
k( j )
The analysis of variance and the expected mean squares can be completed in Minitab as follows:
Minitab Output
ANOVA: Yield versus Machine, Power, Station
Factor
Type Levels Values
Machine
fixed
3
1
Power
random
2
1
Station(Machine) fixed
3
1
2
2
2
3
3
Analysis of Variance for Yield
Source
Machine
Power
Station(Machine)
Machine*Power
Power*Station(Machine)
Error
Total
DF
2
1
6
2
6
36
53
SS
21.143
853.631
32.583
0.616
28.941
58.893
995.808
MS
10.572
853.631
5.431
0.308
4.824
1.636
F
34.33
521.80
1.13
0.19
2.95
P
0.028
0.000
0.445
0.829
0.019
Source
1
2
3
4
5
6
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Machine
4
(6) + 9(4) + 18Q[1]
Power
31.5554
6
(6) + 27(2)
Station(Machine)
5
(6) + 3(5) + 6Q[3]
Machine*Power
-0.1476
6
(6) + 9(4)
Power*Station(Machine)
1.0625
6
(6) + 3(5)
Error
1.6359
(6)
13-15 Reanalyze the experiment in Problem 13-14 assuming the unrestricted form of the mixed model.
You may use a computer software program to do this. Comment on any differences between the restricted
and unrestricted model analysis and conclusions.
ANOVA: Yield versus Machine, Power, Station
Factor
Type Levels Values
Machine
fixed
3
1
Power
random
2
1
Station(Machine) fixed
3
1
2
2
2
3
3
Analysis of Variance for Yield
Source
Machine
Power
Station(Machine)
Machine*Power
Power*Station(Machine)
Error
Total
DF
2
1
6
2
6
36
53
SS
21.143
853.631
32.583
0.616
28.941
58.893
995.808
MS
F
10.572
34.33
853.631 2771.86
5.431
1.13
0.308
0.06
4.824
2.95
1.636
Source
1
2
3
4
5
6
P
0.028
0.000
0.445
0.939
0.019
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
Machine
4
(6) + 3(5) + 9(4) + Q[1,3]
Power
31.6046
4
(6) + 3(5) + 9(4) + 27(2)
Station(Machine)
5
(6) + 3(5) + Q[3]
Machine*Power
-0.5017
5
(6) + 3(5) + 9(4)
Power*Station(Machine)
1.0625
6
(6) + 3(5)
Error
1.6359
(6)
13-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
There are differences between several of the expected mean squares. However, the conclusions that could
be drawn do not differ in any meaningful way from the restricted model analysis.
13-16 A structural engineer is studying the strength of aluminum alloy purchased from three vendors.
Each vendor submits the alloy in standard-sized bars of 1.0, 1.5, or 2.0 inches. The processing of
different sizes of bar stock from a common ingot involves different forging techniques, and so this factor
may be important. Furthermore, the bar stock if forged from ingots made in different heats. Each vendor
submits two tests specimens of each size bar stock from the three heats. The resulting strength data
follow. Analyze the data, assuming that vendors and bar size are fixed and heats are random.
Heat
Bar Size: 1 inch
1 1/2 inch
2 inch
1
1.230
1.259
1.316
1.300
1.287
1.292
Vendor 1
2
3
1.346 1.235
1.400 1.206
1.329 1.250
1.362 1.239
1.346 1.273
1.382 1.215
y ijkl
1
1.301
1.263
1.274
1.268
1.247
1.215
P W i E j WE
Vendor 2
2
3
1.346 1.315
1.392 1.320
1.384 1.346
1.375 1.357
1.362 1.336
1.328 1.342
ij
J k
j
1
1.247
1.296
1.273
1.264
1.301
1.262
( WJ )ik
j
H
F
0.26
41.32
1.37
0.65
2.27
P
0.776
0.000
0.290
0.640
0.037
Vendor 3
2
3
1.275 1.324
1.268 1.315
1.260 1.392
1.265 1.364
1.280 1.319
1.271 1.323
ijk l
Minitab Output
ANOVA: Strength versus Vendor, Bar Size, Heat
Factor
Type Levels Values
Vendor
fixed
3
1
Heat(Vendor) random
3
1
Bar Size
fixed
3
1.0
2
2
1.5
3
3
2.0
Analysis of Variance for Strength
Source
Vendor
Heat(Vendor)
Bar Size
Vendor*Bar Size
Bar Size*Heat(Vendor)
Error
Total
DF
2
6
2
4
12
27
53
SS
0.0088486
0.1002093
0.0025263
0.0023754
0.0110303
0.0109135
0.1359034
MS
0.0044243
0.0167016
0.0012631
0.0005939
0.0009192
0.0004042
Source
1
2
3
4
5
6
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Vendor
2
(6) + 6(2) + 18Q[1]
Heat(Vendor)
0.00272
6
(6) + 6(2)
Bar Size
5
(6) + 2(5) + 18Q[3]
Vendor*Bar Size
5
(6) + 2(5) + 6Q[4]
Bar Size*Heat(Vendor) 0.00026
6
(6) + 2(5)
Error
0.00040
(6)
13-17 Reanalyze the experiment in Problem 13-16 assuming the unrestricted form of the mixed model.
You may use a computer software program to do this. Comment on any differences between the restricted
and unrestricted model analysis and conclusions.
Minitab Output
ANOVA: Strength versus Vendor, Bar Size, Heat
Factor
Type Levels Values
Vendor
fixed
3
1
Heat(Vendor) random
3
1
2
2
3
3
13-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Bar Size
fixed
3
1.0
1.5
2.0
Analysis of Variance for Strength
Source
Vendor
Heat(Vendor)
Bar Size
Vendor*Bar Size
Bar Size*Heat(Vendor)
Error
Total
DF
2
6
2
4
12
27
53
SS
0.0088486
0.1002093
0.0025263
0.0023754
0.0110303
0.0109135
0.1359034
MS
0.0044243
0.0167016
0.0012631
0.0005939
0.0009192
0.0004042
F
0.26
18.17
1.37
0.65
2.27
P
0.776
0.000
0.290
0.640
0.037
Source
1
2
3
4
5
6
Variance Error Expected Mean Square for Each Term
component term (using unrestricted model)
Vendor
2
(6) + 2(5) + 6(2) + Q[1,4]
Heat(Vendor)
0.00263
5
(6) + 2(5) + 6(2)
Bar Size
5
(6) + 2(5) + Q[3,4]
Vendor*Bar Size
5
(6) + 2(5) + Q[4]
Bar Size*Heat(Vendor) 0.00026
6
(6) + 2(5)
Error
0.00040
(6)
There are some differences in the expected mean squares. However, the conclusions do not differ from
those of the restricted model analysis.
13-18 Suppose that in Problem 13-16 the bar stock may be purchased in many sizes and that the three
sizes are actually used in experiment were selected randomly. Obtain the expected mean squares for this
situation and modify the previous analysis appropriately. Use the restricted form of the mixed model.
Minitab Output
ANOVA: Strength versus Vendor, Bar Size, Heat
Factor
Type Levels Values
Vendor
fixed
3
1
Heat(Vendor) random
3
1
Bar Size
random
3
1.0
2
2
1.5
3
3
2.0
Analysis of Variance for Strength
Source
Vendor
Heat(Vendor)
Bar Size
Vendor*Bar Size
Bar Size*Heat(Vendor)
Error
Total
DF
2
6
2
4
12
27
53
SS
0.0088486
0.1002093
0.0025263
0.0023754
0.0110303
0.0109135
0.1359034
MS
0.0044243
0.0167016
0.0012631
0.0005939
0.0009192
0.0004042
F
0.27
18.17
1.37
0.65
2.27
P
0.772 x
0.000
0.290
0.640
0.037
x Not an exact F-test.
Source
1
2
3
4
5
6
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Vendor
*
(6) + 2(5) + 6(4) + 6(2) + 18Q[1]
Heat(Vendor)
0.00263
5
(6) + 2(5) + 6(2)
Bar Size
0.00002
5
(6) + 2(5) + 18(3)
Vendor*Bar Size
-0.00005
5
(6) + 2(5) + 6(4)
Bar Size*Heat(Vendor) 0.00026
6
(6) + 2(5)
Error
0.00040
(6)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
1 Vendor
Error DF Error MS
5.75 0.0163762
Synthesis of Error MS
(2) + (4) - (5)
Notice that a Satterthwaite type test is used for vendor.
13-12
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
13-19 Steel in normalized by heating above the critical temperature, soaking, and then air cooling. This
process increases the strength of the steel, refines the grain, and homogenizes the structure. An
experiment is performed to determine the effect of temperature and heat treatment time on the strength of
normalized steel. Two temperatures and three times are selected. The experiment is performed by
heating the oven to a randomly selected temperature and inserting three specimens. After 10 minutes one
specimen is removed, after 20 minutes the second specimen is removed, and after 30 minutes the final
specimen is removed. Then the temperature is changed to the other level and the process is repeated.
Four shifts are required to collect the data, which are shown below. Analyze the data and draw
conclusions, assume both factors are fixed.
Shift
1
Temperature (F)
1500
1600
63
89
54
91
61
62
50
80
52
72
59
69
48
73
74
81
71
69
54
88
48
92
59
64
Time(minutes)
10
20
30
10
20
30
10
20
30
10
20
30
2
3
4
This is a split-plot design. Shifts correspond to blocks, temperature is the whole plot treatment, and time
is the subtreatments (in the subplot or split-plot part of the design). The expected mean squares and
analysis of variance are shown below.
W i (blocks)
R
4
i
1
F
2
j
2
F
3
k
3
R
1
l
1
E j (temp)
4
0
3
1
2
V 2 3V WE
12 / 3
WE
ij
1
0
3
1
2
V 2 2V WE
J k (time)
4
2
0
1
V 2 2V WJ2 8 / 2
WJ
1
2
0
1
V 2 2V WJ2
Factor
ik
E(MS)
V 2 6V W2
jk
4
0
0
1
2
V 2 V WEJ
12 / 3
WEJ
ijk
1
0
0
1
2
V 2 V WEJ
1
1
1
1
V 2 (not estimable)
¦J
EJ
H
ijk l
¦E
2
j
2
k
¦¦ EJ
2
jk
The following Minitab Output has been modified to display the results of the split-plot analysis. Minitab
will calculate the sums of squares correctly, but the expected mean squares and the statistical tests are not,
in general, correct. Notice that the Error term in the analysis of variance is actually the three factor
interaction.
Minitab Output
ANOVA: Strength versus Shift, Temperature, Time
Factor
Type Levels Values
Shift
random
4
1
2
Temperat fixed
2 1500 1600
Time
fixed
3
10
20
3
4
30
Analysis of Variance for Strength
13-13
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Source
Shift
Temperat
Shift*Temperat
Time
Shift*Time
Temperat*Time
Error
Total
DF
3
1
3
2
6
2
6
23
SS
145.46
2340.38
240.46
159.25
478.42
795.25
244.42
4403.63
Standard
F
P
1.19 0.390
29.20 0.012
1.97 0.220
1.00 0.422
1.96 0.217
9.76 0.013
MS
48.49
2340.38
80.15
79.63
79.74
397.63
40.74
Split Plot
F
P
29.21
0.012
1.00
0.422
9.76
0.013
Source
1
2
3
4
5
6
7
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Shift
1.292
7
(7) + 6(1)
Temperat
3
(7) + 3(3) + 12Q[2]
Shift*Temperat
13.139
7
(7) + 3(3)
Time
5
(7) + 2(5) + 8Q[4]
Shift*Time
19.500
7
(7) + 2(5)
Temperat*Time
7
(7) + 4Q[6]
Error
40.736
(7)
13-20 An experiment is designed to study pigment dispersion in paint. Four different mixes of a
particular pigment are studied. The procedure consists of preparing a particular mix and then applying
that mix to a panel by three application methods (brushing, spraying, and rolling). The response
measured is the percentage reflectance of the pigment. Three days are required to run the experiment, and
the data obtained follow. Analyze the data and draw conclusions, assuming that mixes and application
methods are fixed.
Mix
Day
1
App Method
1
2
3
1
2
3
1
2
3
2
3
1
64.5
68.3
70.3
65.2
69.2
71.2
66.2
69.0
70.8
2
66.3
69.5
73.1
65.0
70.3
72.8
66.5
69.0
74.2
3
74.1
73.8
78.0
73.8
74.5
79.1
72.3
75.4
80.1
4
66.5
70.0
72.3
64.8
68.3
71.5
67.7
68.6
72.4
This is a split plot design. Days correspond to blocks, mix is the whole plot treatment, and method is the
subtreatment (in the subplot or split plot part of the design). The expected mean squares are:
W i (blocks)
R
3
i
1
F
4
j
4
F
3
k
3
R
1
l
1
E j (temp)
3
0
3
1
2
V 2 3V WE
9/3
WE
ij
1
0
3
1
2
V 2 3V WE
J k (time)
3
4
0
1
V 2 4V WJ2 12 / 2
WJ
1
4
0
1
V 2 4V WJ2
Factor
ik
E(MS)
V 2 12V W2
EJ
jk
3
0
0
1
2
V 2 V WEJ
3/ 6
WEJ
ijk
1
0
0
1
2
V 2 V WEJ
1
1
1
1
V 2 (not estimable)
H
ijk l
¦E
2
j
¦J
2
k
¦¦ EJ
2
jk
The following Minitab Output has been modified to display the results of the split-plot analysis. Minitab
will calculate the sums of squares correctly, but the expected mean squares and the statistical tests are not,
13-14
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
in general, correct. Notice that the Error term in the analysis of variance is actually the three factor
interaction.
Minitab Output
ANOVA: Reflectance versus Day, Mix, Method
Factor
Day
Mix
Method
Type Levels Values
random
3
1
fixed
4
1
fixed
3
1
2
2
2
3
3
3
Analysis of Variance for Reflecta
Source
Day
Mix
Day*Mix
Method
Day*Method
Mix*Method
Error
Total
DF
2
3
6
2
4
6
12
35
SS
2.042
307.479
4.529
222.095
1.963
10.036
8.786
556.930
MS
1.021
102.493
0.755
111.047
0.491
1.673
0.732
4
Standard
Split
F
P
F
1.39 0.285
135.77 0. 000 135.75
1.03 0.451
226.24 0.000 226.16
0.67 0.625
2.28 0.105
2.28
Plot
P
0.000
0.000
0.105
Source
1
2
3
4
5
6
7
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Day
0.02406
7
(7) + 12(1)
Mix
3
(7) + 3(3) + 9Q[2]
Day*Mix
0.00759
7
(7) + 3(3)
Method
5
(7) + 4(5) + 12Q[4]
Day*Method -0.06032
7
(7) + 4(5)
Mix*Method
7
(7) + 3Q[6]
Error
0.73213
(7)
13-21 Repeat Problem 13-20, assuming that the mixes are random and the application methods are fixed.
The expected mean squares are:
W i (blocks)
R
3
i
1
R
4
j
4
F
3
k
3
R
1
l
1
E j (temp)
3
1
3
1
2
V 2 3V WE
19V E2
WE
ij
1
1
3
1
2
V 2 3V WE
J k (time)
3
4
0
1
2
V 2 V WEJ
4V WJ2 12 / 2
WJ
1
4
0
1
2
V 2 V WEJ
4V WJ2
Factor
ik
E(MS)
2
V 2 3V WE
12V W2
EJ
jk
3
1
0
1
2
2
V 2 V WEJ
3V EJ
WEJ
ijk
1
1
0
1
2
V 2 V WEJ
1
1
1
1
V 2 (not estimable)
H
ijk l
¦J
2
k
The F-tests are the same as those in Problem 13-20. The following Minitab Output has been edited to
display the results of the split-plot analysis. Minitab will calculate the sums of squares correctly, but the
expected mean squares and the statistical tests are not, in general, correct. Again, the Error term in the
analysis of variance is actually the three factor interaction.
Minitab Output
ANOVA: Reflectance versus Day, Mix, Method
13-15
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Factor
Day
Mix
Method
Type Levels Values
random
3
1
random
4
1
fixed
3
1
2
2
2
3
3
3
Analysis of Variance for Reflecta
Source
Day
Mix
Day*Mix
Method
Day*Method
Mix*Method
Error
Total
DF
2
3
6
2
4
6
12
35
SS
2.042
307.479
4.529
222.095
1.963
10.036
8.786
556.930
MS
1.021
102.493
0.755
111.047
0.491
1.673
0.732
4
Standard
F
P
1.35 0.328
135.77 0.000
1.03 0.451
77.58 0.001 x
0.67 0.625
2.28 0.105
Split Plot
F
P
135.75
0.000
226.16
0.000
2.28
0.105
x Not an exact F-test.
Source
1
2
3
4
5
6
7
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Day
0.0222
3
(7) + 3(3) + 12(1)
Mix
11.3042
3
(7) + 3(3) + 9(2)
Day*Mix
0.0076
7
(7) + 3(3)
Method
*
(7) + 3(6) + 4(5) + 12Q[4]
Day*Method -0.0603
7
(7) + 4(5)
Mix*Method
0.3135
7
(7) + 3(6)
Error
0.7321
(7)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
4 Method
Error DF
3.59
Error MS
1.431
Synthesis of Error MS
(5) + (6) - (7)
13-22 Consider the split-split-plot design described in example 13-3. Suppose that this experiment is
conducted as described and that the data shown below are obtained. Analyze and draw conclusions.
Blocks
1
2
3
4
Dose Strengths
Wall Thickness
1
2
3
4
1
2
3
4
1
2
3
4
1
2
3
4
1
1
2
3
95
104
101
108
95
106
103
109
96
105
106
113
90
100
102
114
71
82
85
85
78
84
86
84
70
81
88
90
68
84
85
88
108
115
117
116
110
109
116
110
107
106
112
117
109
112
115
118
1
Technician
2
2
3
1
3
2
3
96
99
95
97
100
101
99
112
94
100
104
121
98
102
100
118
70
84
83
85
72
79
80
86
66
84
87
90
68
81
85
85
108
100
105
109
104
102
108
109
100
101
109
117
106
103
110
116
95
102
105
107
92
100
101
108
90
97
100
110
98
102
105
110
70
81
84
87
69
76
80
86
73
75
82
91
72
78
80
95
100
106
113
115
101
104
109
113
98
100
104
112
101
105
110
120
Using the computer output, the F-ratios were calculated by hand using the expected mean squares found in
Table 13-18. The following Minitab Output has been edited to display the results of the split-plot
analysis. Minitab will calculate the sums of squares correctly, but the expected mean squares and the
statistical tests are not, in general, correct. Notice that the Error term in the analysis of variance is
actually the four factor interaction.
Minitab Output
13-16
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
ANOVA: Time versus Day, Tech, Dose, Thick
Factor
Day
Tech
Dose
Thick
Type Levels Values
random
4
1
fixed
3
1
fixed
3
1
fixed
4
1
2
2
2
2
3
3
3
3
4
4
Analysis of Variance for Time
Source
Day
Tech
Day*Tech
Dose
Day*Dose
Tech*Dose
Day*Tech*Dose
Thick
Day*Thick
Tech*Thick
Day*Tech*Thick
Dose*Thick
Day*Dose*Thick
Tech*Dose*Thick
Error
Total
DF
3
2
6
2
6
4
12
3
9
6
18
6
18
12
36
143
SS
48.41
248.35
161.15
20570.06
112.11
125.94
113.89
3806.91
313.12
126.49
167.57
402.28
70.44
205.89
172.06
26644.66
MS
16.14
124.17
26.86
10285.03
18.69
31.49
9.49
1268.97
34.79
21.08
9.31
67.05
3.91
17.16
4.78
Standard
F
P
3.38 0.029
4.62 0.061
5.62 0.000
550.44 0.000
3.91 0.004
3.32 0.048
1.99 0.056
36.47 0.000
7.28 0.000
2.26 0.084
1.95 0.044
17.13 0.000
0.82 0.668
3.59 0.001
Split Plot
F
P
4.62
0.061
550.30
0.000
3.32
0.048
36.48
0.000
2.26
0.084
17.15
0.000
3.59
0.001
Source
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Day
0.3155 15
(15) + 36(1)
Tech
3
(15) + 12(3) + 48Q[2]
Day*Tech
1.8400 15
(15) + 12(3)
Dose
5
(15) + 12(5) + 48Q[4]
Day*Dose
1.1588 15
(15) + 12(5)
Tech*Dose
7
(15) + 4(7) + 16Q[6]
Day*Tech*Dose
1.1779 15
(15) + 4(7)
Thick
9
(15) + 9(9) + 36Q[8]
Day*Thick
3.3346 15
(15) + 9(9)
Tech*Thick
11
(15) + 3(11) + 12Q[10]
Day*Tech*Thick
1.5100 15
(15) + 3(11)
Dose*Thick
13
(15) + 3(13) + 12Q[12]
Day*Dose*Thick
-0.2886 15
(15) + 3(13)
Tech*Dose*Thick
15
(15) + 4Q[14]
Error
4.7793
(15)
13-23 Rework Problem 13-22, assuming that the dosage strengths are chosen at random.
restricted form of the mixed model.
Use the
The following Minitab Output has been edited to display the results of the split-plot analysis. Minitab will
calculate the sums of squares correctly, but the expected mean squares and the statistical tests are not, in
general, correct. Again, the Error term in the analysis of variance is actually the four factor interaction.
Minitab Output
ANOVA: Time versus Day, Tech, Dose, Thick
Factor
Day
Tech
Dose
Thick
Type Levels Values
random
4
1
fixed
3
1
random
3
1
fixed
4
1
2
2
2
2
3
3
3
3
4
4
Analysis of Variance for Time
Source
Day
Tech
Day*Tech
Dose
DF
3
2
6
2
SS
48.41
248.35
161.15
20570.06
MS
16.14
124.17
26.86
10285.03
Standard
F
P
0.86 0.509
2.54 0.155
2.83 0.059
550.44 0.000
13-17
Split Plot
F
P
4.62
0.061
550.30
0.000
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Day*Dose
Tech*Dose
Day*Tech*Dose
Thick
Day*Thick
Tech*Thick
Day*Tech*Thick
Dose*Thick
Day*Dose*Thick
Tech*Dose*Thick
Error
Total
6
4
12
3
9
6
18
6
18
12
36
143
112.11
125.94
113.89
3806.91
313.12
126.49
167.57
402.28
70.44
205.89
172.06
26644.66
18.69
31.49
9.49
1268.97
34.79
21.08
9.31
67.05
3.91
17.16
4.78
3.91
3.32
1.99
12.96
8.89
0.97
1.95
17.13
0.82
3.59
0.004
0.048
0.056
0.001 x
0.000
0.475 x
0.044
0.000
0.668
0.001
3.32
0.048
36.48
0.000
2.26
0.084
17.15
0.000
3.59
0.001
x Not an exact F-test.
Source
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Variance Error Expected Mean Square for Each Term
component term (using restricted model)
Day
-0.071
5
(15) + 12(5) + 36(1)
Tech
*
(15) + 4(7) + 16(6) + 12(3) + 48Q[2]
Day*Tech
1.447
7
(15) + 4(7) + 12(3)
Dose
213.882
5
(15) + 12(5) + 48(4)
Day*Dose
1.159 15
(15) + 12(5)
Tech*Dose
1.375
7
(15) + 4(7) + 16(6)
Day*Tech*Dose
1.178 15
(15) + 4(7)
Thick
*
(15) + 3(13) + 12(12) + 9(9) + 36Q[8]
Day*Thick
3.431 13
(15) + 3(13) + 9(9)
Tech*Thick
*
(15) + 4(14) + 3(11) + 12Q[10]
Day*Tech*Thick
1.510 15
(15) + 3(11)
Dose*Thick
5.261 13
(15) + 3(13) + 12(12)
Day*Dose*Thick
-0.289 15
(15) + 3(13)
Tech*Dose*Thick
3.095 15
(15) + 4(14)
Error
4.779
(15)
* Synthesized Test.
Error Terms for Synthesized Tests
Source
2 Tech
8 Thick
10 Tech*Thick
Error DF
6.35
10.84
15.69
Error MS
48.85
97.92
21.69
Synthesis of Error MS
(3) + (6) - (7)
(9) + (12) - (13)
(11) + (14) - (15)
The expected mean squares can also be shown as follows:
Wi
R
4
i
1
F
3
j
3
R
3
k
3
F
4
h
4
R
1
l
1
Ej
4
0
3
4
1
2
2
2
V 2 4V WEJ
16V EJ
12V WE
( 48 / 2 )
1
0
3
4
1
2
2
V 2 4V WEJ
12V WE
4
3
1
4
1
2
2
V 2 3V WJG
12V JG
12V WJ2 48V J2
1
3
1
4
1
V 2 12V WJ2
Factor
WE
ij
Jk
WJ
ik
E(MS)
V 2 12V WJ2 36V W2
¦E
EJ
jk
4
0
1
4
1
2
2
V 2 4V WEJ
16V EJ
WEJ
ijk
1
0
1
4
1
2
V 2 4V WEJ
4
3
3
0
1
2
2
2
V 2 3V WJG
12V JG
9V WG
36 / 3
1
3
3
0
1
2
2
V 2 3V WJG
9V WG
Gh
WG
ih
¦G
EG
jh
4
0
3
0
1
2
2
2
V 2 V WEJG
4V EJG
3V WEG
12 / 6
WEG
ijh
1
0
3
0
1
2
2
V 2 V WEJG
3V WEG
4
3
1
0
1
2
2
V 2 3V WJG
12V JG
JG
kh
13-18
2
j
2
h
¦¦ EG
2
jh
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
WJG
ikh
1
3
1
0
1
2
V 2 3V WJG
EJG
jkh
4
0
1
0
1
2
2
V 2 V WEJG
4V EJG
WEJG
ijkh
1
0
1
0
1
2
V 2 V WEJG
1
1
1
1
1
V2
H
ijk l
There are no exact tests on technicians E j , dosage strengths J k , wall thickness Gh , or the technician x
wall thickness interaction EG
jh
. The approximate F-tests are as follows:
H0: E j =0
F
MS B MS ABC
MS AB MS BC
2
MS B MS ABC
p
2
MS B2 MS ABC
2
12
2.291
124.174 9.491 2
2.315
124.174 2 9.4912
2
12
2
MS AB MS BC
q
124.174 9.491
26.859 31.486
2
2
MS AB
MS BC
6
4
26.859 31.486
2
9.248
26.859 2 31.486 2
6
4
Do not reject H0: E j =0
H0: J k =0
F
MS C MS ACD
MS CD MS AD
MS C MS ACD
p
10285.028 3.914
101.039
67.046 34.791
2
2
MS C2 MS ACD
2
18
MS CD MS AD
q
10285.028 3.914
2
10285.028 2 3.914 2
2
18
2
2
2
MS CD
MS AD
6
9
67.046 34.791 2
2.002
11.736
67.046 2 34.7912
6
9
Reject H0: J k =0
H0: Gh =0
F
p
MS D MS ACD
MS CD MS AD
MS D MS ACD
2
MS D2 MS ACD
3
18
2
1268.970 3.914
12.499
67.046 34.791
1268.970 3.914
2
1268.970 2 3.914 2
3
18
13-19
3.019
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
MS CD MS AD
q
2
67.046 34.791 2
2
2
MS CD
MS AD
6
9
67.046 2 34.7912
6
9
11.736
Reject H0: Gh =0
H0: EG
jh
=0
F
F<1, Do not reject H0: EG
jh
MS BD MS ABCD
MS BCD MS ABD
21.081 4.779
17.157 9.309
0.977
=0
13-24 Suppose that in Problem 13-22 four technicians had been used. Assuming that all the factors are
fixed, how many blocks should be run to obtain an adequate number of degrees of freedom on the test for
differences among technicians?
The number of degrees of freedom for the test is (a-1)(4-1)=3(a-1), where a is the number of blocks used.
Number of Blocks (a)
2
3
4
5
DF for test
3
6
9
12
At least three blocks should be run, but four would give a better test.
13-25 Consider the experiment described in Example 13-3. Demonstrate how the order in which the
treatments combinations are run would be determined if this experiment were run as (a) a split-split-plot,
(b) a split-plot, (c) a factorial design in a randomized block, and (d) a completely randomized factorial
design.
(a) Randomization for the split-split plot design is described in Example 13-3.
(b) In the split-plot, within a block, the technicians would be the main treatment and within a blocktechnician plot, the 12 combinations of dosage strength and wall thickness would be run in random
order. The design would be a two-factor factorial in a split-plot.
(c) To run the design in a randomized block, the 36 combinations of technician, dosage strength, and
wall thickness would be ran in random order within each block. The design would be a three factor
factorial in a randomized block.
(d) The blocks would be considered as replicates, and all 144 observations would be 4 replicates of a
three factor factorial.
13-20
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Chapter 14
Other Design and Analysis Topics
Solutions
14-1 Reconsider the experiment in Problem 5-22. Use the Box-Cox procedure to determine if a
transformation on the response is appropriate (or useful) in the analysis of the data from this experiment.
DE S IG N-E X P E RT P l o t
Cra ck G ro w th
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
5.62
Lam bda
Cu rre n t = 1
B e st = 0 .1 1
L o w C .I. = -0 .4 4
Hi g h C .I. = 0 .5 6
Ln (R es idu alSS)
4.49
Re co m m e n d tra n sfo rm :
Log
(L a m b d a = 0 )
3.36
2.23
1.10
-3
-2
-1
0
1
2
3
La m b da
With the value of lambda near zero, and since the confidence interval does not include one, a natural log
transformation would be appropriate.
14-2 In example 6-3 we selected a log transformation for the drill advance rate response. Use the BoxCox procedure to demonstrate that this is an appropriate data transformation.
14-1
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
DE S IG N-E X P E RT P l o t
A d va n ce Ra te
6.85
Lam bda
Cu rre n t = 1
B e st = -0 .2 3
L o w C .I. = -0 .7 9
Hi g h C .I. = 0 .3 2
5.40
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
Log
(L a m b d a = 0 )
3.95
2.50
1.05
-3
-2
-1
0
1
2
3
La m b da
Because the value of lambda is very close to zero, and the confidence interval does not include one, the
natural log was the correct transformation chosen for this analysis.
14-3 Reconsider the smelting process experiment in Problem 8-23, where a 26-3 fractional factorial
design was used to study the weight of packing material stuck to carbon anodes after baking. Each of the
eight runs in the design was replicated three times and both the average weight and the range of the
weights at each test combination were treated as response variables. Is there any indication that that a
transformation is required for either response?
DE S IG N-E X P E RT P l o t
We ig h t
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
11.05
Lam bda
Cu rre n t = 1
B e st = 1 .3 3
L o w C .I. = -0 .7 1
Hi g h C .I. = 4 .2 9
10.26
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
13.06
Lam bda
Cu rre n t = 1
B e st = 0 .5 8
L o w C .I. = -1 .7 4
Hi g h C .I. = 2 .9 2
12.12
Re c o m m e n d tra n sfo rm :
No n e
(L a m b d a = 1 )
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
No n e
(L a m b d a = 1 )
Ln (R es idu alSS)
DE S IG N-E X P E RT P l o t
Ra n g e
9.47
11.17
8.68
10.23
7.89
9.29
-3
-2
-1
0
1
2
3
-3
La m b da
-2
-1
0
1
2
3
La m b da
There is no indication that a transformation is required for either response.
14-4 In Problem 8-24 a replicated fractional factorial design was used to study substrate camber in
semiconductor manufacturing. Both the mean and standard deviation of the camber measurements were
used as response variables. Is there any indication that a transformation is required for either response?
14-2
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DE S IG N-E X P E RT P l o t
Ca m b e r A vg
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
Lam bda
Cu rre n t = 1
B e st = -0 .0 3
L o w C .I. = -0 .7 9
Hi g h C .I. = 0 .7 4
13.00
Lam bda
Cu rre n t = 1
B e st = 0 .5 7
L o w C .I. = -0 .0 3
Hi g h C .I. = 1 .1 6
11.35
Re co m m e n d tra n sfo rm :
Log
(L a m b d a = 0 )
11.35
Re c o m m e n d tra n sfo rm :
No n e
(L a m b d a = 1 )
Ln (R es idu alSS)
Ln (R es idu alSS)
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
DE S IG N-E X P E RT P l o t
Ca m b e r S tDe v
12.22
10.49
9.70
9.62
8.05
8.76
6.40
-3
-2
-1
0
1
2
3
-3
La m b da
-2
-1
0
1
2
3
La m b da
The Box-Cox plot for the Camber Average suggests a natural log transformation should be applied. This
decision is based on the confidence interval for lambda not including one and the point estimate of lambda
being very close to zero. With a lambda of approximately 0.5, a square root transformation could be
considered for the Camber Standard Deviation; however, the confidence interval indicates that no
transformation is needed.
14-5 Reconsider the photoresist experiment in Problem 8-25. Use the variance of the resist thickness at
each test combination as the response variable. Is there any indication that a transformation is required?
DE S IG N-E X P E RT P l o t
T h i ck S tDe v
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
9.93
Lam bda
Cu rre n t = 1
B e st = -0 .0 4
L o w C .I. = -0 .7 7
Hi g h C .I. = 0 .7 6
9.28
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
Log
(L a m b d a = 0 )
8.62
7.97
7.31
-3
-2
-1
0
1
2
3
La m b da
With the point estimate of lambda near zero, and the confidence interval for lambda not inclusive of one,
a natural log transformation would be appropriate.
14-3
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
14-6 In the grill defects experiment described in Problem 8-29 a variation of the square root
transformation was employed in the analysis of the data. Use the Box-Cox method to determine if this is
the appropriate transformation.
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
DE S IG N-E X P E RT P l o t
c
12.75
Lam bda
Cu rre n t = 1
B e st = -0 .0 6
L o w C .I. = -0 .6 9
Hi g h C .I. = 0 .7 4
10.35
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
Log
(L a m b d a = 0 )
k = 0 .5 6
(u se d to m a ke
re sp o n se va l u e s
p o si ti ve )
7.95
5.55
3.15
-3
-2
-1
0
1
2
3
La m b da
Because the confidence interval for the minimum lambda does not include one, the decision to use a
transformation is correct. Because the lambda point estimate is close to zero, the natural log
transformation would be appropriate. This is a stronger transformation than the square root.
14-7 In the central composite design of Problem 11-14, two responses were obtained, the mean and
variance of an oxide thickness. Use the Box-Cox method to investigate the potential usefulness of
transformation for both of these responses. Is the log transformation suggested in part (c) of that problem
appropriate?
DE S IG N-E X P E RT P l o t
M e a n T h i ck
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
8.73
Lam bda
Cu rre n t = 1
B e st = -0 .2
L o w C .I. = -3 .5 8
Hi g h C .I. = 3 .1 8
8.57
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
2.63
Lam bda
Cu rre n t = 1
B e st = -0 .4 7
L o w C .I. = -2 .8 5
Hi g h C .I. = 1 .5 1
2.30
Re c o m m e n d tra n sfo rm :
No n e
(L a m b d a = 1 )
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
No n e
(L a m b d a = 1 )
Ln (R es idu alSS)
DE S IG N-E X P E RT P l o t
V a r T h ick
8.41
1.97
8.25
1.65
8.09
1.32
-3
-2
-1
0
1
2
3
-3
La m b da
-2
-1
0
1
2
3
La m b da
The Box-Cox plot for the Mean Thickness model suggests that a natural log transformation could be
applied; however, the confidence interval for lambda includes one. Therefore, a transformation would
14-4
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
have a minimal effect. The natural log transformation applied to the Variance of Thickness model
appears to be acceptable; however, again the confidence interval for lambda includes one.
14-8 In the 33 factorial design of Problem 11-33 one of the responses is a standard deviation. Use the
Box-Cox method to investigate the usefulness of transformations for this response. Would your answer
change if we used the variance of the response?
DE S IG N-E X P E RT P l o t
S td . D e v.
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
22.32
Lam bda
Cu rre n t = 1
B e st = 0 .2 9
L o w C .I. = 0 .0 1
Hi g h C .I. = 0 .6 1
Ln (R es idu alSS)
19.22
Re co m m e n d tra n sfo rm :
S q u a re R o o t
(L a m b d a = 0 .5 )
k = 1 .5 8 2
(u se d to m a ke
re sp o n se va l u e s
p o si ti ve )
16.13
13.03
9.94
-3
-2
-1
0
1
2
3
La m b da
Because the confidence interval for lambda does not include one, a transformation should be applied. The
natural log transformation should not be considered due to zero not being included in the confidence
interval. The square root transformation appears to be acceptable. However, notice that the value of zero
is very close to the lower confidence limit, and the minimizing value of lambda is between 0 and 0.5. It is
likely that either the natural log or the square root transformation would work reasonably well.
14-9 Problem 11-34 suggests using the ln(s2) as the response (refer to part b). Does the Box-Cox method
indicate that a transformation is appropriate?
14-5
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
DE S IG N-E X P E RT P l o t
V a ri a n ce
B o x-C o x P lo t fo r P o we r Tra nsfo rm s
17.56
Lam bda
Cu rre n t = 1
B e st = -1 .1 7
L o w C .I. = -1 .5 3
Hi g h C .I. = -0 .7 2
14.13
Ln (R es idu alSS)
Re co m m e n d tra n sfo rm :
In ve rse
(L a m b d a = -1 )
10.70
7.28
3.85
-3
-2
-1
0
1
2
3
La m b da
Because the confidence interval for lambda does not include one, a transformation should be applied. The
confidence interval does not include zero; therefore, the natural log transformation is inappropriate. With
the point estimate of lambda at –1.17, the reciprocal transformation is appropriate.
14-10 A soft drink distributor is studying the effectiveness of delivery methods. Three different types of
hand trucks have been developed, and an experiment is performed in the company’s methods engineering
laboratory. The variable of interest is the delivery time in minutes (y); however, delivery time is also
strongly related to the case volume delivered (x). Each hand truck is used four times and the data that
follow are obtained. Analyze the data and draw the appropriate conclusions. Use D=0.05.
1
y
27
44
33
41
1
x
24
40
35
40
Hand
2
y
25
35
46
26
Truck
2
x
26
32
42
25
Type
3
y
40
22
53
18
3
x
38
26
50
20
From the analysis performed in Minitab, hand truck does not have a statistically significant effect on
delivery time. Volume, as expected, does have a significant effect.
Minitab Output
General Linear Model: Time versus Truck
Factor
Truck
Type Levels Values
fixed
3 1 2 3
Analysis of Variance for Time, using Adjusted SS for Tests
Source
Volume
Truck
Error
Total
DF
1
2
8
11
Seq SS
1232.07
11.65
41.95
1285.67
Term
Constant
Volume
Coef
-4.747
1.17326
SE Coef
2.638
0.07699
Adj SS
1217.55
11.65
41.95
T
-1.80
15.24
Adj MS
1217.55
5.82
5.24
P
0.110
0.000
14-6
F
232.20
1.11
P
0.000
0.375
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
14-11 Compute the adjusted treatment means and the standard errors of the adjusted treatment means for
the data in Problem 14-10.
adj yi .
adj y1.
adj y 2.
adj y3.
yi . Eˆ xi . x..
145
§ 139 398 ·
1.173 ¨
¸ 34.39
4
12 ¹
© 4
132
§ 125 398 ·
1.173 ¨
¸ 35.25
4
12 ¹
© 4
133
§ 134 398 ·
1.173 ¨
¸ 32.86
4
12 ¹
© 4
S adj . yi .
ª
° 1
x x ..
« MS E ® i .
E xx
«¬
°̄ n
1
2
½°º 2
¾»
°¿»¼
1
2
S adj . y1.
ª
° 1 34.75 33.17
«5.24®
884.50
°̄ 4
«¬
2
S adj . y2.
ª
° 1 31.25 33.17
«5.24®
884.50
°̄ 4
«¬
2
S adj . y3.
ª
° 1 33.50 33.17
«5.24®
884.50
°̄ 4
«¬
½°º 2
¾»
°¿»¼
1.151
1
½°º 2
¾»
°¿»¼
1.154
1
½°º 2
¾»
°¿»¼
1.145
The solutions can also be obtained with Minitab as follows:
Minitab Output
Least Squares Means for Time
Truck
1
2
3
Mean
34.39
35.25
32.86
SE Mean
1.151
1.154
1.145
14-12 The sums of squares and products for a single-factor analysis of covariance follow. Complete the
analysis and draw appropriate conclusions. Use D = 0.05.
Source of
Variation
Treatment
Error
Total
Source
Treatment
Error
Total
df
3
12
15
Degrees of
Freedom
3
12
15
Sums of
x
1500
6000
7500
Squares &
xy
1000
1200
2200
Sums of
x
1500
6000
7500
Squares and
xy
1000
1200
2200
Products
y
650
550
1200
14-7
y
310
559.67
Products
x
650
550
1200
Adjusted
df
11
14
MS
28.18
F0
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Adjusted
Treat.
244.67
3
81.56
2.89
Treatments differ only at 10%.
14-13 Find the standard errors of the adjusted treatment means in Example 14-4.
From Example 14-4 y1.
40.38 , adj y2.
4142
. , adj y3.
37.78
1
2
S adj . y1.
ª
° 1 25.20 24.13
«2.54®
195.60
°̄ 5
«¬
2
S adj . y2.
ª
° 1 26.00 24.13
«2.54®
195.60
°̄ 5
«¬
2
S adj . y3.
ª
° 1 21.20 24.13
«2.54®
195.60
°̄ 5
«¬
½°º 2
¾»
°¿»¼
0.7231
1
½°º 2
¾»
°¿»¼
0.7439
1
°½º 2
¾»
°¿»¼
0.7871
14-14 Four different formulations of an industrial glue are being tested. The tensile strength of the glue
when it is applied to join parts is also related to the application thickness. Five observations on strength
(y) in pounds and thickness (x) in 0.01 inches are obtained for each formulation. The data are shown in
the following table. Analyze these data and draw appropriate conclusions.
1
y
46.5
45.9
49.8
46.1
44.3
1
x
13
14
12
12
14
2
y
48.7
49.0
50.1
48.5
45.2
Glue
2
x
12
10
11
12
14
Formulation
3
y
46.3
47.1
48.9
48.2
50.3
3
x
15
14
11
11
10
4
y
44.7
43.0
51.0
48.1
48.6
4
x
16
15
10
12
11
From the analysis performed in Minitab, glue formulation does not have a statistically significant effect on
strength. As expected, glue thickness does affect strength.
Minitab Output
General Linear Model: Strength versus Glue
Factor
Glue
Type Levels Values
fixed
4 1 2 3 4
Analysis of Variance for Strength, using Adjusted SS for Tests
Source
Thick
Glue
Error
Total
DF
1
3
15
19
Term
Constant
Thick
Coef
60.089
-1.0099
Seq SS
68.852
1.771
20.962
91.585
SE Coef
1.944
0.1547
Adj SS
59.566
1.771
20.962
T
30.91
-6.53
Adj MS
59.566
0.590
1.397
P
0.000
0.000
Unusual Observations for Strength
14-8
F
42.62
0.42
P
0.000
0.740
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Obs
3
Strength
49.8000
Fit
47.5299
SE Fit
0.5508
Residual
2.2701
St Resid
2.17R
R denotes an observation with a large standardized residual.
Expected Mean Squares, using Adjusted SS
Source
1 Thick
2 Glue
3 Error
Expected Mean Square for Each Term
(3) + Q[1]
(3) + Q[2]
(3)
Error Terms for Tests, using Adjusted SS
Source
1 Thick
2 Glue
Error DF
15.00
15.00
Error MS
1.397
1.397
Synthesis of Error MS
(3)
(3)
Variance Components, using Adjusted SS
Source
Error
Estimated Value
1.397
14-15 Compute the adjusted treatment means and their standard errors using the data in Problem 14-14.
adj y1.
adj y 2.
adj y3.
adj y 4.
adj yi . yi . Eˆ xi . x..
46.52 1.0099 13.00 12.45
47.08
47.64
47.91
47.43
48.30 1.0099 11.80 12.45
48.16 1.0099 12.20 12.45
47.08 1.0099 12.80 12.45
S adj . yi .
ª
° 1
x x ..
« MS E ® i .
E xx
«¬
°̄ n
1
2
½°º 2
¾»
°¿»¼
1
S adj . y1.
ª
° 1 13.00 12.45
«1.40®
58.40
°̄ 5
«¬
S adj . y2.
ª
° 1 11.80 12.45 2 °½º 2
«1.40®
¾»
58.40
°¿»¼
°̄ 5
«¬
2
S adj . y3.
ª
° 1 12.20 12.45
«1.40®
58.40
°̄ 5
«¬
2
S adj . y4.
ª
° 1 12.80 12.45
«1.40®
58.40
°̄ 5
«¬
2
½°º 2
¾»
°¿»¼
0.5360
1
0.5386
1
½°º 2
¾»
°¿»¼
0.5306
1
½°º 2
¾»
°¿»¼
0.5319
The adjusted treatment means can also be generated in Minitab as follows:
Minitab Output
Least Squares Means for Strength
Glue
1
2
3
Mean
47.08
47.64
47.91
SE Mean
0.5355
0.5382
0.5301
14-9
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
4
47.43
0.5314
14-16 An engineer is studying the effect of cutting speed on the rate of metal removal in a machining
operation. However, the rate of metal removal is also related to the hardness of the test specimen. Five
observations are taken at each cutting speed. The amount of metal removed (y) and the hardness of the
specimen (x) are shown in the following table. Analyze the data using and analysis of covariance. Use
D=0.05.
1000
y
68
90
98
77
88
1000
x
120
140
150
125
136
Cutting
1200
y
112
94
65
74
85
Speed
1200
x
165
140
120
125
133
(rpm)
1400
y
118
82
73
92
80
1400
x
175
132
124
141
130
As shown in the analysis performed in Minitab, there is no difference in the rate of removal between the
three cutting speeds. As expected, the hardness does have an impact on rate of removal.
Minitab Output
General Linear Model: Removal versus Speed
Factor
Speed
Type Levels Values
fixed
3 1000 1200 1400
Analysis of Variance for Removal, using Adjusted SS for Tests
Source
Hardness
Speed
Error
Total
DF
1
2
11
14
Seq SS
3075.7
2.4
95.5
3173.6
Adj SS
3019.3
2.4
95.5
Adj MS
3019.3
1.2
8.7
Term
Constant
Hardness
Speed
1000
1200
Coef
-41.656
0.93426
SE Coef
6.907
0.05008
T
-6.03
18.65
P
0.000
0.000
0.478
0.036
1.085
1.076
0.44
0.03
0.668
0.974
F
347.96
0.14
P
0.000
0.872
Unusual Observations for Removal
Obs
8
Removal
65.000
Fit
70.491
SE Fit
1.558
Residual
-5.491
St Resid
-2.20R
R denotes an observation with a large standardized residual.
Expected Mean Squares, using Adjusted SS
Source
1 Hardness
2 Speed
3 Error
Expected Mean Square for Each Term
(3) + Q[1]
(3) + Q[2]
(3)
Error Terms for Tests, using Adjusted SS
Source
1 Hardness
2 Speed
Error DF
11.00
11.00
Error MS
8.7
8.7
Synthesis of Error MS
(3)
(3)
Variance Components, using Adjusted SS
Source
Estimated Value
14-10
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
Error
8.677
Means for Covariates
Covariate
Hardness
Mean
137.1
StDev
15.94
Least Squares Means for Removal
Speed
1000
1200
1400
Mean
86.88
86.44
85.89
SE Mean
1.325
1.318
1.328
14-17 Show that in a single factor analysis of covariance with a single covariate a 100(1-D) percent
confidence interval on the ith adjusted treatment mean is
ª
§ 1 x x ..
y i . Eˆ x i . x .. r tD 2 ,a n 1 1 « MS E ¨ i .
¨n
E xx
«¬
©
1
2
·º 2
¸»
¸»
¹¼
Using this formula, calculate a 95 percent confidence interval on the adjusted mean of machine 1 in
Example 14-4.
The 100(1-D) percent interval on the ith adjusted treatment mean would be
y i . Eˆ x i . x .. r tD
2 ,a n 1 1 S adjyi .
since yi . Eˆ x i . x .. is an estimator of the ith adjusted treatment mean. The standard error of the
adjusted treatment mean is found as follows:
>
V y i . Eˆ xi . x ..
V adj . y i .
@
V y i . xi . x.. 2 V Eˆ
Since the ^ y i. ` and E are independent. From regression analysis, we have V Eˆ
V adj .y i .
x x.. 2 V 2
V2
i.
n
E xx
ª1 x x
V 2 « i . ..
E xx
«¬ n
2
V2
. Therefore,
E xx
º
»
»¼
Replacing V 2 by its estimator MSE, yields
V̂ adj .y i .
ª 1 x x ..
MS E « i .
E xx
¬« n
S adj . yi .
°
ª 1 xi . x ..
®MS E «
E xx
°̄
¬« n
Substitution of this result into y i. Eˆ xi . x.. r tD
2
º
» or
¼»
1
2
º ½° 2
»¾
¼» °¿
2,a ( n 1) 1 S adjyi .
will produce the desired
confidence interval. A 95% confidence interval on the mean of machine 1 would be found as follows:
14-11
Solutions from Montgomery, D. C. (2001) Design and Analysis of Experiments, Wiley, NY
y i . Eˆ x i . x .. 40.38
S adj . y i . 0.7231
>40.38 r t 0.025 ,11 0.7231 @
adj . y i .
>40.38 r 2.20
>40.38 r 1.59@
0.7231 @
Therefore, 38.79 d P1 d 41.96 , where P1 denotes the true adjusted mean of treatment one.
14-18 Show that in a single-factor analysis of covariance with a single covariate, the standard error of the
difference between any two adjusted treatment means is
1
S Adjyi . Adjy j .
adj . yi . adj .y j .
ª
§ 2 x x .. 2 ·º 2
¸»
« MS E ¨ i .
¸»
¨n
E
«¬
xx
¹¼
©
ˆ
ˆ
yi . E xi . x.. y j . E x j . x..
>
adj. y i. adj. y j .
@
yi . y j . Eˆ xi. x j .
The variance of this statistic is
>
V y i . y j . Eˆ xi . x j .
@
V yi . V y j . xi . x j . 2 V Eˆ
xi . x j . 2 V 2
V2 V2
n
n
E xx
ª 2 xi . x j .
V 2«
E xx
«¬ n
2
º
»
»¼
Replacing V 2 by its estimator MSE, , and taking the square root yields the standard error
S Adjyi . Adjy j .
ª
§ 2 x x ..
« MS E ¨ i .
¨n
E xx
«¬
©
1
2
·º 2
¸»
¸»
¹¼
14-19 Discuss how the operating characteristic curves for the analysis of variance can be used in the
analysis of covariance.
To use the operating characteristic curves, fixed effects case, we would use as the parameter )2,
)2
a
¦W
2
i
nV 2
The test has a-1 degrees of freedom in the numerator and a(n-1)-1 degrees of freedom in the denominator.
14-12
Source Exif Data:
File Type : PDF File Type Extension : pdf MIME Type : application/pdf PDF Version : 1.3 Linearized : Yes Page Count : 488 Page Mode : UseOutlines XMP Toolkit : XMP toolkit 2.9.1-13, framework 1.6 About : uuid:69a57763-7a32-4004-8d84-42e66f2e2804 Producer : Acrobat Distiller 4.0 for Windows Modify Date : 2005:04:12 11:32:21+02:00 Create Date : 2000:10:24 14:36:28Z Creator Tool : ADOBEPS4.DRV Version 4.24 Metadata Date : 2005:04:12 11:32:21+02:00 Document ID : uuid:f01c2cf4-7883-4ebe-a8a4-ccc28e72d29c Format : application/pdf Title : Microsoft Word - DOX Solutions 02.doc Creator : jperea Author : jpereaEXIF Metadata provided by EXIF.tools