The Idiot's Guide To Statistics Jr., Ph.D., Robert A. Donnelly Statistics, 2nd Edition Alpha (

User Manual: Pdf

Open the PDF directly: View PDF .
Page Count: 421

Download
Open PDF In Browser	View PDF

AbObWabWQa
ASQ]\R3RWbW]\
by Robert A. Donnelly, Jr., Ph.D.

A member of Penguin Group (USA) Inc.

To my wife, Debbie, who supported and encouraged me every step of the way.
I could not have done this without you, babe.

/:>6/0==9A
Published by the Penguin Group
Penguin Group (USA) Inc., 375 Hudson Street, New York, New York 10014, U.S.A.
Penguin Group (Canada), 10 Alcorn Avenue, Toronto, Ontario, Canada M4V 3B2 (a division of Pearson Penguin
Canada Inc.)
Penguin Books Ltd, 80 Strand, London WC2R 0RL, England
Penguin Ireland, 25 St Stephen’s Green, Dublin 2, Ireland (a division of Penguin Books Ltd)
Penguin Group (Australia), 250 Camberwell Road, Camberwell, Victoria 3124, Australia (a division of Pearson
Australia Group Pty Ltd)
Penguin Books India Pvt Ltd, 11 Community Centre, Panchsheel Park, New Delhi—110 017, India
Penguin Group (NZ), cnr Airborne and Rosedale Roads, Albany, Auckland 1310, New Zealand (a division of
Pearson New Zealand Ltd)
Penguin Books (South Africa) (Pty) Ltd, 24 Sturdee Avenue, Rosebank, Johannesburg 2196, South Africa
Penguin Books Ltd, Registered Offices: 80 Strand, London WC2R 0RL, England

1]^g`WUVb %Pg@]PS`b/2]\\SZZg8`
All rights reserved. No part of this book shall be reproduced, stored in a retrieval system, or transmitted by any
means, electronic, mechanical, photocopying, recording, or otherwise, without written permission from the publisher. No patent liability is assumed with respect to the use of the information contained herein. Although every
precaution has been taken in the preparation of this book, the publisher and author assume no responsibility for
errors or omissions. Neither is any liability assumed for damages resulting from the use of information contained
herein. For information, address Alpha Books, 800 East 96th Street, Indianapolis, IN 46240.
THE COMPLETE IDIOT’S GUIDE TO and Design are registered trademarks of Penguin Group (USA) Inc.
ISBN : 1-4295-1390-X
Library of Congress Catalog Card Number: 2006938600

Interpretation of the printing code: The rightmost number of the first series of numbers is the year of the book’s
printing; the rightmost number of the second series of numbers is the number of the book’s printing. For example,
a printing code of 07-1 shows that the first printing occurred in 2007.
Note: This publication contains the opinions and ideas of its author. It is intended to provide helpful and informative material on the subject matter covered. It is sold with the understanding that the author and publisher are not
engaged in rendering professional services in the book. If the reader requires personal assistance or advice, a competent professional should be consulted.
The author and publisher specifically disclaim any responsibility for any liability, loss, or risk, personal or otherwise, which is incurred as a consequence, directly or indirectly, of the use and application of any of the contents of
this book.
Publisher: Marie Butler-Knight
Editorial Director: Mike Sanders
Managing Editor: Billy Fields
Acquisitions Editor: Tom Stevens
Development Editor: Michael Thomas
Production Editor: Kayla Dugger
Copy Editor: Nancy Wagner

Cartoonist: Chris Eliopoulos
Cover Designer: Bill Thomas
Book Designer: Trina Wurst
Indexer: Angie Bess
Layout: Chad Dressler
Proofreader: Aaron Black

1]\bS\baObO5ZO\QS
>O`b(

BVS0OaWQa

2 Data, Data Everywhere and Not a Drop to Drink
All statistical analysis begins with the proper selection of the
source, type, and measurement scale of the data.

3 Displaying Descriptive Statistics
A vast array of methods display data and information effectively, such as frequency distributions, histograms, pie charts,
and bar charts.

4 Calculating Descriptive Statistics: Measures of
Central Tendency (Mean, Median, and Mode)
Using the mean, median, or mode is an effective way to summarize many pieces of data.
5 Calculating Descriptive Statistics: Measures
of Dispersion
The standard deviation, range, and quartiles reveal valuable
information about the variability of the data.

>O`b (

1 Let’s Get Started
Statistics plays a vital role in today’s society by providing the
foundation for sound decisions.

>`]POPWZWbgB]^WQa

6 Introduction to Probability
Basic probability theory, such as the intersection and union of
events, provides important groundwork for statistical concepts.

7 More Probability Stuff
Calculate the probability of winning your tennis match given
that you had a short warm-up period.

8 Counting Principles and Probability Distributions
Determine your odds at winning a state lottery drawing or
your chances of drawing a five-card flush in poker.

105

9 The Binomial Probability Distribution
121
Calculate the probability of correctly guessing the answer of 6
out of 12 multiple-choice questions when each question has five
choices.

BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
10 The Poisson Probability Distribution
Determine the probability that you will receive at least 3
spam e-mails tomorrow given that you average 2.5 such
e-mails per day.

131

11 The Normal Probability Distribution
145
Determine probabilities of events that follow this symmetrical,
bell-shaped distribution.

>O`b!(

7\TS`S\bWOZAbObWabWQa

12 Sampling
Discover how to choose between simple random, systematic,
cluster, and stratified sampling for statistical analysis.

165

13 Sampling Distributions
The central limit theorem tells us that sample means follow
the normal probability distribution as long as the sample size
is large enough.

177

14 Confidence Intervals
A confidence interval is a range of values used to estimate a
population parameter.

195

15 Introduction to Hypothesis Testing
213
A hypothesis test enables us to investigate an assumption about
a population parameter using a sample.

>O`b"(

16 Hypothesis Testing with One Sample
This procedure focuses on testing a statement concerning a
single population.

227

17 Hypothesis Testing with Two Samples
Use this test to see whether that new golf instructional video
will really lower your scores.

249

/RdO\QSR7\TS`S\bWOZAbObWabWQa
18 The Chi-Square Probability Distribution
This procedure enables us to test the independence of two categorical variables.

%
273

19 Analysis of Variance
289
Learn how to test the difference between more than two population means.
20 Correlation and Simple Regression
309
Determine the strength and direction of the linear relationship
between an independent and dependent variable.

1]\bS\baObO5ZO\QS
/^^S\RWfSa
A Solutions to “Your Turn”

333

B Statistical Tables

367

C Glossary

377

Index

387

1]\bS\ba
>O`b( BVS0OaWQa

:SbÂa5SbAbO`bSR

Where Is This Stuff Used? ............................................................4
Who Thought of This Stuff? ........................................................5
Early Pioneers ..............................................................................5
More Recent Famous People ..........................................................6
The Field of Statistics Today .........................................................6
Descriptive Statistics—the Minor League ......................................7
Inferential Statistics—the Major League .......................................8
Ethics and Statistics—It’s a Dangerous World Out There.........10
Your Turn......................................................................................12

2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

The Importance of Data ..............................................................16
The Sources of Data—Where Does All This Stuff Come
From?..........................................................................................17
Direct Observation—I’ll Be Watching You...................................19
Experiments—Who’s in Control? ................................................19
Surveys—Is That Your Final Answer? ........................................20
Types of Data................................................................................20
Types of Measurement Scales—a Weighty Topic .......................21
Nominal Level of Measurement ..................................................21
Ordinal Level of Measurement....................................................21
Interval Level of Measurement ...................................................22
Ratio Level of Measurement........................................................22
Computers to the Rescue.............................................................23
The Role of Computers in Statistics .............................................23
Installing the Data Analysis Add-In............................................24
Your Turn......................................................................................26

! 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

Frequency Distributions ..............................................................30
Constructing a Frequency Distribution ........................................31
(A Distant) Relative Frequency Distribution ...............................32
Cumulative Frequency Distribution ............................................33
Graphing a Frequency Distribution—the Histogram ...................34
Letting Excel Do Our Dirty Work ..............................................34

dWWW BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
Statistical Flower Power—the Stem and Leaf Display...............37
Charting Your Course ..................................................................39
What’s Your Favorite Pie Chart? ................................................39
Bar Charts .................................................................................41
Line Charts ................................................................................43
Your Turn......................................................................................44

" 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg
;SO\;SRWO\O\R;]RS

Measures of Central Tendency ....................................................48
Mean..........................................................................................48
Weighted Mean ..........................................................................50
Mean of Grouped Data from a Frequency Distribution................51
Median.......................................................................................54
Mode ..........................................................................................55
How Does One Choose?...............................................................56
Using Excel to Calculate Central Tendency ...............................56
Your Turn......................................................................................58

# 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

Range ............................................................................................62
Variance ........................................................................................63
Using the Raw Score Method (When Grilling)............................64
The Variance of a Population ......................................................65
Standard Deviation.......................................................................67
Calculating the Standard Deviation of Grouped Data ...............67
The Empirical Rule: Working the Standard Deviation..............69
Chebyshev’s Theorem ..................................................................71
Measures of Relative Position......................................................73
Quartiles ....................................................................................73
Interquartile Range ....................................................................74
Using Excel to Calculate Measures of Dispersion......................75
Your Turn......................................................................................76

>O`b ( >`]POPWZWbgB]^WQa

$ 7\b`]RcQbW]\b]>`]POPWZWbg

%'
&

What Is Probability? ....................................................................82
Classical Probability ....................................................................82

1]\bS\ba
Empirical Probability ..................................................................83
Subjective Probability..................................................................85
Basic Properties of Probability ....................................................86
The Intersection of Events ..........................................................87
The Union of Events: A Marriage Made in Heaven ..................88
Your Turn......................................................................................89

% ;]`S>`]POPWZWbgAbcTT

Conditional Probability................................................................94
Independent Versus Dependent Events ......................................96
Multiplication Rule of Probabilities ............................................97
Mutually Exclusive Events ...........................................................98
Addition Rule of Probabilities .....................................................99
Summarizing Our Findings .......................................................101
Bayes’ Theorem..........................................................................102
Your Turn....................................................................................103

& 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

Counting Principles ...................................................................106
The Fundamental Counting Principle .......................................106
Permutations ............................................................................107
Combinations............................................................................109
Using Excel to Calculate Permutations and Combinations..........111
Probability Distributions............................................................112
Random Variables ....................................................................112
Discrete Probability Distributions ..............................................113
Rules for Discrete Probability Distributions................................115
The Mean of a Discrete Probability Distribution........................115
The Variance and Standard Deviation of a Discrete
Probability Distribution ..........................................................116
Your Turn....................................................................................118

' BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\

Characteristics of a Binomial Experiment.................................122
The Binomial Probability Distribution .....................................123
Binomial Probability Tables.......................................................126
Using Excel to Calculate Binomial Probabilities ......................127
The Mean and Standard Deviation for the Binomial
Distribution ..............................................................................129
Your Turn....................................................................................129

BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\

BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

Characteristics of a Poisson Process..........................................132
The Poisson Probability Distribution .......................................133
Poisson Probability Tables .........................................................136
Using Excel to Calculate Poisson Probabilities ........................139
Using the Poisson Distribution as an Approximation to
the Binomial Distribution........................................................140
Your Turn....................................................................................142

BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

Characteristics of the Normal Probability Distribution...........146
Calculating Probabilities for the Normal Distribution ............148
Calculating the Standard Z-Score .............................................148
Using the Standard Normal Table.............................................150
The Empirical Rule Revisited ....................................................155
Calculating Normal Probabilities Using Excel ...........................156
Using the Normal Distribution as an Approximation to
the Binomial Distribution........................................................157
Your Turn....................................................................................161

>O`b!( 7\TS`S\bWOZAbObWabWQa

AO[^ZW\U

$!
$#

Why Sample?..............................................................................166
Random Sampling ......................................................................167
Simple Random Sampling.........................................................168
Systematic Sampling .................................................................170
Cluster Sampling......................................................................171
Stratified Sampling ..................................................................172
Sampling Errors .........................................................................173
Examples of Poor Sampling Techniques ...................................174
Your Turn....................................................................................176

! AO[^ZW\U2Wab`WPcbW]\a

What Is a Sampling Distribution?.............................................177
Sampling Distribution of the Mean...........................................178
The Central Limit Theorem .....................................................182
Standard Error of the Mean ......................................................185

1]\bS\ba
Why Does the Central Limit Theorem Work?........................186
Putting the Central Limit Theorem to Work ..........................188
Sampling Distribution of the Proportion..................................190
Calculating the Sample Proportion ............................................190
Calculating the Standard Error of the Proportion......................192
Your Turn....................................................................................193

" 1]\TWRS\QS7\bS`dOZa

Confidence Intervals for the Mean with Large Samples ..........196
Estimators ................................................................................196
Confidence Levels......................................................................197
Beware of the Interpretation of Confidence Interval! ..................199
The Effect of Changing Confidence Levels .................................200
The Effect of Changing Sample Size .........................................201
Determining Sample Size for the Mean ....................................202
Calculating a Confidence Interval When X Is Unknown ............202
Using Excel’s CONFIDENCE Function ...................................203
Confidence Intervals for the Mean with Small Samples...........204
When X Is Known ....................................................................204
When X Is Unknown ................................................................205
Confidence Intervals for the Proportion with Large
Samples .....................................................................................208
Calculating the Confidence Interval for the Proportion...............209
Determining Sample Size for the Proportion .............................210
Your Turn....................................................................................211

# 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U

Hypothesis Testing—the Basics.................................................214
The Null and Alternative Hypothesis ........................................215
Stating the Null and Alternative Hypothesis .............................216
Two-Tail Hypothesis Test ...........................................................217
One-Tail Hypothesis Test...........................................................218
Type I and Type II Errors..........................................................219
Example of a Two-Tail Hypothesis Test....................................220
Using the Scale of the Original Variable....................................221
Using the Standardized Normal Scale.......................................222
Example of a One-Tail Hypothesis Test....................................223
Your Turn....................................................................................225

fWW

BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\

$ 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

Hypothesis Testing for the Mean with Large Samples.............228
When Sigma Is Known.............................................................228
When Sigma Is Unknown.........................................................229
The Role of Alpha in Hypothesis Testing.................................231
Introducing the p-Value .............................................................233
The p-Value for a One-Tail Test................................................233
The p-Value for a Two-Tail Test................................................234
Hypothesis Testing for the Mean with Small Samples .............236
When Sigma Is Known.............................................................236
When Sigma Is Unknown.........................................................237
Using Excel’s TINV Function ...................................................241
Hypothesis Testing for the Proportion with Large Samples....242
One-Tail Hypothesis Test for the Proportion...............................243
Two-Tail Hypothesis Test for the Proportion ...............................245
Your Turn....................................................................................246

% 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

The Concept of Testing Two Populations ................................250
Sampling Distribution for the Difference in Means.................250
Testing for Differences Between Means with Large
Sample Sizes .............................................................................252
Testing a Difference Other Than Zero .....................................255
Testing for Differences Between Means with Small
Sample Sizes and Unknown Sigma .........................................256
Equal Population Standard Deviations......................................257
Unequal Population Standard Deviations..................................260
Letting Excel Do the Grunt Work............................................261
Testing for Differences Between Means with Dependent
Samples .....................................................................................263
Testing for Differences Between Proportions with
Independent Samples ...............................................................265
Your Turn....................................................................................269

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

& BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\

%
%!

Review of Data Measurement Scales.........................................274
The Chi-Square Goodness-of-Fit Test .....................................274
Stating the Null and Alternative Hypothesis .............................276

1]\bS\ba fWWW
Observed Versus Expected Frequencies .......................................276
Calculating the Chi-Square Statistic .........................................277
Determining the Critical Chi-Square Score...............................277
Using Excel’s CHIINV Function...............................................279
Characteristics of a Chi-Square Distribution............................279
A Goodness-of-Fit Test with the Binomial Distribution..........280
Chi-Square Test for Independence............................................282
Your Turn....................................................................................286

' /\OZgaWa]TDO`WO\QS

One-Way Analysis of Variance ..................................................290
Completely Randomized ANOVA ............................................291
Partitioning the Sum of Squares ...............................................292
Determining the Calculated F-Statistic .....................................295
Determining the Critical F-Statistic .........................................296
Using Excel to Perform One-Way ANOVA.............................298
Pairwise Comparisons ................................................................299
Completely Randomized Block ANOVA ..................................301
Partitioning the Sum of Squares ...............................................302
Determining the Calculated F-Statistic .....................................303
To Block or Not to Block, That Is the Question...........................304
Your Turn....................................................................................305

1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

Independent Versus Dependent Variables.................................310
Correlation .................................................................................311
Correlation Coefficient ..............................................................312
Testing the Significance of the Correlation Coefficient ................314
Using Excel to Calculate Correlation Coefficients .......................315
Simple Regression .....................................................................316
The Least Squares Method........................................................317
Confidence Interval for the Regression Line ...............................321
Testing the Slope of the Regression Line .....................................323
The Coefficient of Determination ..............................................324
Using Excel for Simple Regression .............................................325
A Simple Regression Example with Negative Correlation ..........326
Assumptions for Simple Regression ............................................330
Simple Versus Multiple Regression.............................................330
Your Turn....................................................................................331

/^^S\RWfSa

/ A]ZcbW]\ab]¿G]c`Bc`\À

!!!

0 AbObWabWQOZBOPZSa

!$%

1 5Z]aaO`g

!%%

7\RSf

!&%

4]`Se]`R
Statistics, statistics everywhere, but not a single word can we understand! Actually,
understanding statistics is a critically important skill that we all need to have in this
day and age. Every day, we are inundated with data about politics, sports, business,
the stock market, health issues, financial matters, and many other topics. Most of us
don’t pay much attention to most of the statistics we hear, but more importantly, most
of us don’t really understand how to make sense of the numbers, ratios, and percentages with which we are constantly barraged. In order to obtain the truth behind the
numbers, we must be able to ascertain what the data is really saying to us. We need to
determine whether the data is biased in a particular direction or whether the true, balanced picture is correctly represented in the numbers. That is the reason for reading
this book.
Statistics, as a field, is usually not the most popular topic or course in school. In fact,
many people will go to great lengths to avoid having to take a statistics course. Many
people think of it as a math course or something that is very quantitative, and that
scares them away. Others, who get past the math, do not have the patience to search
for what the numbers are actually saying. And still others don’t believe that statistics
can ever be used in a legitimate manner to point to the truth. But whether it is about
significant trends in the population, average salary and unemployment rates, or similarities and differences across stock prices, statistics are an extremely important input
to many decisions that we face daily. And understanding how to generate the statistics
and interpret them relating to your particular decision can make all the difference
between a good decision and a poor one.
For example, suppose that you are trying to sell your house and you need to set a selling price for it. The mean selling price of houses in your area is $250,000, so you set
your price at $265,000. Perhaps $250,000 is the price roughly in the middle of several
house prices that have ranged from $200,000 to $270,000, so you are in the ballpark. However, a mean of $250,000 could also occur with house prices of $175,000,
$150,000, $145,000, $100,000, and $780,000. One high price out of five causes the
mean to increase dramatically, so you have potentially priced yourself out of most of
the market. For this reason, it is important to understand what the term “mean” really
represents.
Another compelling reason to understand statistics is that we are living in a qualitydriven society. Everything nowadays is related to “improving quality,” a “quality job,”
or “quality improvement processes.” Companies are striving for higher quality in their
products and employees and are using such programs as “continuous improvement”
and “six-sigma” to achieve and measure this quality. Even the ordinary consumer has
heard these terms and needs to understand them in order to be an educated customer

or client. Here again, an understanding of statistics can help you make wise choices
related to purchasing behavior.
So as we move from the information age to the knowledge age, it is becoming increasingly important for us to at least understand, if not generate and use, statistics. In
this book, Bob Donnelly has done a wonderful job of presenting statistics so that you
can improve your ability to look at and comprehend the data you run across every
day. Bob’s many years of teaching statistics at all levels have provided the basis for his
phenomenal ability to explain difficult statistical concepts clearly. Even the most unsophisticated reader will soon understand the subtleties and power of telling the truth
with statistics!
Christine T. Kydd
2003 Delaware Professor of the Year
Associate Professor of Business Administration and Director of Undergraduate
Programs
University of Delaware

7\b`]RcQbW]\
Statistics. Why does this single word terrify so many of today’s students? The mere
mention of this word in the classroom causes a glassy-eyed, deer-in-the-headlights
reaction across a sea of faces. In one form or another, the topic of statistics has been
torturing innocent students for hundreds of years. You would think the word statistics
had been derived from the Latin words sta, meaning “Why” and tistica, meaning “Do I
have to take this %#!$@*% class?” But it really doesn’t have to be this way. The term
“stat” needn’t be a four-letter word in the minds of our students.
As you read this paragraph, you’re probably wondering what this book can do for
you. Well, it’s written by a person (that’s me) who (a) clearly remembers being in your
shoes as a student (even if it was in the last century), (b) sympathizes with your current
dilemma (I can feel your pain), and (c) has learned a thing or two over many years of
teaching (those many hours of tutorials were not for naught). The result of this experience has allowed me to discover ways to walk you through many of the concepts that
traditionally frustrate students. Armed with the tools that you will gain from the many
examples and numerous problems explained in detail, this task will not be as daunting
as it first appears.
Unfortunately, fancy terms such as inferential statistics, analysis of variance, and
hypothesis testing are enough to send many running for the hills. My goal has been
to show that these complicated terms are really used to describe ordinary, straightforward concepts. By applying many of the techniques to everyday (and sometimes
humorous) examples, I have attempted to show that not only is statistics a topic that
anyone can master, but it can also actually make sense and be helpful in numerous
situations.
To further help those in need, I have established a companion website for this book at
www.stat-guide.com. Here you will find additional problems with solutions and links
to other useful websites. If you have any feedback you would like to provide about this
book, please send me an e-mail via this website.
So hold on to your hats, we’re about to take a wild ride into the realm of numbers,
inequalities, and, oh yes, don’t forget all those Greek symbols! You will see equations
that look like the Chinese alphabet at first glance, but can, in fact, be simplified into
plain English. The step-by-step description of each problem will help you break down
the process into manageable pieces. As you work the example problems on your own,
you will gain confidence and success in your abilities to put numbers to work to provide usable information. And, guess what, that is sometimes how statisticians are born!

fdWWW BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
6]eBVWa0]]Y7a=`UO\WhSR
The book is organized into four parts:
In Part 1, “The Basics,” we start from the very beginning without any assumptions of prior knowledge. After a brief history lesson to warm you up, we dive into
the world of data and learn about the different types of data and the variety of measurement scales that we can use. We also cover how to display data graphically, both
manually and with the help of Microsoft Excel. We wrap up Part 1 with learning how
to calculate descriptive statistics of a sample, such as the mean and standard deviation.
In Part 2, “Probability Topics,” we introduce the scary world of probability theory.
Once again, I assume you have no prior knowledge of this topic (or if you did, I
assume you buried it in the deep recesses of your brain, hoping to never uncover it).
An important topic in this section is learning how to count the number of events,
which can really improve your poker skills. After easing you into the basics, we gently
slide into probability distributions, such as the normal and binomial. Once you master
these, we have set the stage for Part 3.
In Part 3, “Inferential Statistics,” we start off learning about sampling procedures
and the way samples behave statistically. When these concepts are understood, we
start acting like real statisticians by making estimates of populations using confidence
intervals. By this time, your own mother wouldn’t recognize you! We’ll top Part 3 off
with a procedure that’s near and dear to every statistician’s heart—hypothesis testing.
With this tool, you can do things like make bold comparisons between the male and
female population. I’ll leave that one to you.
In Part 4, “Advanced Inferential Statistics,” we build on earlier topics and explore
analysis of variance, a popular method to compare more than two populations to each
other. We will also learn about the chi-square tests, which enable us to determine
whether two variables are dependent. And last but not least, we’ll discover how simple
regression (which, by the way, is not so simple or else it wouldn’t be the last topic in
the book) describes the strength and direction of the relationship between two variables. When you’re done with these topics, your friends won’t believe the words they
hear coming from your mouth.

7\b`]RcQbW]\

fWf

3fb`Oa
Throughout this book, you will come across various sidebars that provide a helping
hand when things seem to get a little tough. Many are based on my experience as a
teacher with the concepts that I have found to cause students the most difficulty.

Random Thoughts
These are definitions of statistical jargon explained in a
nonthreatening manner, which
will help to clarify important
concepts. You’ll find that their
bark is often far worse than
their bite.

Bob’s Basics
These are tips and insights
that I have accumulated over
the years of helping students
master a particular topic. The
goal here is to have that light
bulb in the brain go off, resulting in the feeling of “I got it!”

In these sidebars I will give you
insights that I find interesting
(and hopefully you will, too!)
about the current topic. Statistics
is full of little-known facts that
can help relieve the intensity of
the topic at hand.

Wrong Number
These are warnings of
potential pitfalls lying in
wait for an unsuspecting student
to fall into. By taking note of
these, you’ll avoid the same
traps that have ensnarled many
of your predecessors.

/QY\]eZSRU[S\ba
There are many people whom I am indebted to for helping me with this project. I’d
like to thank Jessica Faust for her guidance and expertise to get me on track in the
beginning, Mike Sanders for going easy on me with his initial feedback, and Nancy
Lewis, for her valuable opinions during the writing process. I’d also like to thank Mike
Thomas and Nancy Wagner for their helpful suggestions with the second edition.
To my colleague and friend, Dr. Patricia Buhler, who introduced me to the publishing industry, convinced me to take on this project, and encouraged me throughout the
writing process. This all started with you, Pat.

BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
To my in-laws, Lindsay and Marge, who never failed to ask me what chapter I was
writing, which motivated me to stay on schedule. Your commitment to each other is a
true inspiration for all of us.
To my boss of 10 years at Goldey-Beacom College, Joyce Jones, who rearranged my
teaching schedule to accommodate my deadlines. Life at GBC will never be the same
after you retire, Joyce. I am really going to miss you. Thank you for your constant
support over the years. You have been a great boss and a true friend.
To my friend, Jerry Collarini, who provided many recommendations for changes that
appear in this second edition.
To my students who make teaching a pleasure. The lessons that I have leaned over the
years about teaching were invaluable to me as I wrote this book. Without all of you, I
would never have had the opportunity to be an author.
To my children, Christin, Brian, and John, and my stepchildren, Katie, Sam, and Jeff,
for your interest in this book and your willingness to let me use your antics as examples in many of the chapters.
And most importantly, to my wife, Debbie, who made this a team effort with all the
hours she spent contributing ideas, proofreading manuscripts, editing figures, and giving up family time to help me stay on schedule. Deb’s excitement over my opportunity
to write this book gave me the courage to accept this challenge. Deb was also the
inspiration for many of the examples used in the book, allowing me to share experiences from our wonderful life together. Thank you for your love and your patience
with me while writing this book.

B`ORS[O`Ya
All terms mentioned in this book that are known to be or are suspected of being
trademarks or service marks have been appropriately capitalized. Alpha Books and
Penguin Group (USA) Inc. cannot attest to the accuracy of this information. Use of a
term in this book should not be regarded as affecting the validity of any trademark or
service mark.

>O`b

BVS0OaWQa

The key to successfully mastering statistics is to have a solid foundation of
the basics. To get a firm grasp of the more advanced topics, you need to be
well grounded in the concepts presented in this part. After a quick history
lesson, these chapters focus on data, the starting point for any method in
statistics. You might be surprised with how much there really is to learn
about data and all of its properties. We will examine the different types of
data, how it is collected, how it is displayed, and how it is used to calculate
things called the mean and standard deviation.

1VO^bS`

:SbÂa5SbAbO`bSR
7\BVWa1VO^bS`
U The purpose of statistics—what’s in it for you?

U The history of statistics—where did this stuff come from?
U Brief overview of the field of statistics
U The ethical side of statistics

How many times have you asked yourself why you even need to learn
statistics? Well, you’re not alone. All too often students find themselves
drowning in a mathematical swamp of theories and concepts and never get
a chance to see the “big picture” before going under. My goal in this chapter is to provide you with that broader perspective and convince you that
statistics is a very useful tool in our current society. In other words, here
comes your life preserver. Grab on!
In today’s technologically advanced world, we are surrounded by a barrage
of data and information from sources trying to convince us to buy something or simply persuade us to agree with their point of view. When we
hear on TV that a politician is leading in the polls and in small print see +
or − 4 percent, do we know what that means? When a new product is recommended by 4 out of 5 doctors, do we question the validity of the claim?
(For instance, were the doctors paid for their endorsement?) Statistics can

>O`b( BVS0OaWQa
have a powerful influence on our feelings, our opinions, and our decisions that we
make in life. Getting a handle on this widely used tool is a good thing for all of us.

EVS`S7aBVWaAbcTTCaSRThe Funk and Wagnalls Dictionary defines statistics as “the science that deals with the
collection, tabulation, and systematic classification of quantitative data, especially as
a basis for inference and induction.” Now that’s a mouthful! In simpler terms, I view
statistics as a way to convert numbers into useful information so that good decisions
can be made.
These decisions can affect our lives in many ways. For instance, countless medical
studies have been performed to determine the effectiveness of new drugs. Statistics
form the basis of making an objective decision as to whether this new drug is actually an improvement over current treatments. The results of statistical studies and the
manner in which these results are presented often dictate government policies.

Wrong Number
Not interpreting statistical
information properly can
lead to disaster. Coca-Cola performed a major consumer study
in 1985 and, based on the
results, decided to reformulate
Coke, its flagship drink. After a
huge public outcry, Coca-Cola
had to backtrack and bring
the original formulation back to
market. What a mess!

Today’s corporations are making major business
decisions based on statistical analysis. In the 1980s,
Marriott conducted an extensive survey with potential customers on their attitudes about current hotel
offerings. After analyzing the data, the company
launched Courtyard by Marriott, which has been a
huge success.
The federal government heavily relies on the national
census that is conducted every 10 years to determine
funding levels for all the various parts of the country.
The statistical analysis performed on this census data
has far-reaching implications for many ongoing programs at the state and federal levels.

The entire sports industry is completely dependent
on the field of statistics. Can you even imagine baseball, football, or basketball without all the statistical analysis that surrounds them?
You would never know who the top players are, who is currently hot, and who is in a
slump. But then, without statistics, how could the players negotiate those outrageous
salaries? Hmmm, maybe I’m onto something here.
My point here is to make you aware of the fact that we are surrounded by statistics in
our society and that our world would be very different if this wasn’t the case. Statistics
is a useful, and sometimes even critical, tool in our everyday life.

1VO^bS`( :SbÂa5SbAbO`bSR

EV]BV]cUVb]TBVWaAbcTTThe field of statistics has been evolving for a very long time. Population surveys
appear to be the primary motivation for the historical development of statistics as we
know it today. In fact, according to the Bible, Moses conducted a census more than
3,000 years ago. The very word “statistics” comes from the Latin word status, which
means “state.” This etymological connection reflects the earliest focus of statistics on
measuring things such as the number of (taxable) subjects in a kingdom (or state) or
the number of subjects to send off to invade neighboring kingdoms.

3O`Zg>W]\SS`a
European mathematicians provided the basic foundation for the field of statistics.
In 1532, Sir William Petty provided the first accounts of the number of deaths in
London on a weekly basis. So began the insurance companies’ morbid fascination with
death statistics.
During the 1600s, Swiss mathematician James Bernoulli is credited with calculating the probability of a sequence of events, otherwise known as “independent trials.”
This term is an unfortunate choice of words, as many students over the generations
have struggled with this concept and felt like they were on “trial” themselves. You
might remember dealing with the problem of calculating (or trying desperately to
calculate) the probability of 7 “heads” in 10 coin tosses in a math class. You can thank
Mr. Bernoulli for providing you with a way to solve this type of problem. Chapter 9
explores Bernoulli trials in loving detail, and with a little practice you’ll get off with a
light sentence.
Later, during the 1700s, English mathematician Thomas Bayes developed probability
concepts that have also been very useful
to the field of statistics. Bayes used the
probability of known events of the past to
predict probabilities of the future. This concept of inference is widely used in statistical
techniques today. Chapter 7 covers one of
his particular contributions, appropriately
known as Bayes’ theorem.

The term inference refers to
a key concept in statistics in
which we draw a conclusion
from available evidence.

>O`b( BVS0OaWQa

;]`S@SQS\b4O[]ca>S]^ZS
But it wasn’t until the early twentieth century that statistics began to develop into the
field that we know it as today, when William Gossett developed the famous “t-test”
using the Student’s t-distribution while working at the Guinness brewery in Dublin,
Ireland. We will raise our glasses to Mr. Gossett as we investigate his efforts in
Chapter 14.
W. Edwards Deming has been credited with merging the science of statistics with the
field of quality control in manufacturing environments. Dr. Deming spent considerable time in Japan during the 1950s and 1960s promoting the concept of statistical
quality for businesses. This technique relies on control charts to monitor a process
and the use of statistics to determine whether the process is operating satisfactorily.
During the 1970s, the Japanese auto industry gained major market share in this country due mainly to superior quality. That’s the power of statistics!

Random Thoughts
Dr. Deming’s philosophy has been condensed to what is known as Deming’s 14
points. This list has proven to be invaluable for organizations seeking ways to use
statistics to make their processes more efficient. Through Dr. Deming’s efforts, statistics
has found a significant role in the business world. Check out his book The Deming
Management Method (Perigee, 1988) for more information.

BVS4WSZR]TAbObWabWQaB]ROg
The science of statistics has evolved into two basic categories known as descriptive statistics and inferential statistics. Because descriptive statistics is generally simpler, it can
be thought of as the “minor league” of the field; whereas inferential statistics, being
more challenging, can be considered the “major league” of the two.

The purpose of descriptive statistics is to summarize or display data so we can quickly
obtain an overview. Inferential statistics allows us to make claims or conclusions about
a population based on a sample of data from that population. A population represents
all possible outcomes or measurements of interest. A sample is a subset of a population.

1VO^bS`( :SbÂa5SbAbO`bSR

Today, computers and software play a dominant role in our use of statistics. Current
desktop computers have the capability of processing and analyzing huge amounts of
data and information. Specialized software such as SAS and SPSS allows you to conveniently perform all sorts of complicated statistical techniques without breaking a sweat.
In this book, I will show you how to perform many statistical techniques using
Microsoft Excel, a spreadsheet software package that’s readily available on most desktop computers (also included in the Microsoft Office software suite). Excel has many
easy-to-use statistics features that can save you time and energy. If this paragraph
causes your blood pressure to elevate (hey, wait a minute, nobody told me this was a
computer book!), have no fear. Feel free to just skip over these sections; subsequent
material in this book does not depend on this information. I promise it will not be on
the final exam.

2SaQ`W^bWdSAbObWabWQa¾bVS;W\]`:SOUcS
The main focus of descriptive statistics is to summarize and display data. Descriptive
statistics plays an important role today because of the vast amount of data readily
available at our fingertips. With a basic computer and an Internet connection, we can
access volumes of data in no time at all. Being able to accurately summarize all of this
data to get a look at the “big picture,” either graphically or numerically, is the job of
descriptive statistics.
There are many examples of descriptive statistics, but the most common is the average. As an example, let’s say I would like to get a perspective on the average attention
span of my Labrador retriever by using flash cards. I time each incident with a stopwatch and write it down on my clipboard. The following table lists our results, measured in seconds:
Observation

Seconds

1
2
3
4
5
6
7
8
9

4
8
5
10
2
4
7
12
7

>O`b( BVS0OaWQa
Using descriptive statistics, I can calculate the average attention span, as follows:
4 8 5 10 2 4 7 12 7
6.6 seconds
9
Descriptive statistics can also involve displaying the data graphically, as shown in
Figure 1.1. What a good dog!
14

Attention span graph.

Seconds

4WUc`S

10
8
6
4
2
0
1

Observation

We will delve into descriptive statistics in more detail in Chapters 3 and 4. But until
then, we’re ready to move up to the big leagues—inferential statistics.

7\TS`S\bWOZAbObWabWQa¾bVS;OX]`:SOUcS
As important as descriptive statistics is to us number crunchers, we really get excited
about inferential statistics. This category covers a large variety of techniques that
allow us to make actual claims about a population based on a sample of data. Suppose,
for instance, that I am interested in discovering in general who has the longer attention span, Labrador retrievers or, let’s say, teenage boys. (Based on personal observations, I suspect I know the answer to this, but I’ll keep it to myself.) Now, it’s not
possible to measure the attention span of every teenager and every dog, so the next
best thing is to take a sample of each and measure them.
At this point, I need to explain the difference between a population and a sample.
We use the term “population” in statistics to represent all possible measurements or
outcomes that are of interest to us in a particular study. The term “sample” refers to
a portion of the population that is representative of the population from which it was
selected.

1VO^bS`( :SbÂa5SbAbO`bSR

In this example, the population is all teenage boys and all Labrador retrievers. I need
to select a sample of teenagers and a sample of dogs that represent their respective
populations. Based on the results of my samples, I can infer the average attention span
of each population and determine which is longer.
Figure 1.2 shows the relationship between a population and a sample.

4WUc`S
Population

The relationship between a
population and a sample.

Sample

The following are other examples of inferential statistics:
U Based on a recent sample, I am 95 percent certain that the average age of my

customers is between 32 and 35 years old.
U The average salary for male employees in a particular job category across the

country was higher than the female employees’ salary, based on a random survey.
In each case, the findings were based on a sample from a larger population and were
used to make an inference on that entire population.
The basic difference between descriptive and inferential statistics is that descriptive
statistics reports only on the observations at hand and nothing more. Inferential statistics makes a statement about a population based solely on results from a sample taken
from that population.
I must tell you at this point that inferential statistics is the area of this field that
students find the most challenging. To be
able to make statements based on samples,
you need to use mathematical models that
involve probability theory. Now don’t panic.
Take a deep breath and count to 10 slowly.
That’s better. I realize that this is often the
stumbling block for many, so I have devoted
plenty of pages to that nasty “p” word.

Bob’s Basics
A good understanding of
probability concepts is an
essential stepping-stone for
properly digesting statistics.
Part 2 of this book covers probability.

>O`b( BVS0OaWQa

3bVWQaO\RAbObWabWQa¾7bÂaO2O\US`]caE]`ZR=cbBVS`S
People often use statistics when attempting to persuade you to their point of view.
Because they are motivated to convince you to purchase something from them or simply to support them, this motivation can lead to the misuse of statistics in several ways.
One of the most common misuses is choosing a sample that ensures results consistent
with the desired outcome, rather than choosing a sample representative of the population of interest. This is known as having a biased sample.
Suppose, for instance, that I’m an upstanding politician whose only concern is the best interest of my
constituents and I want to propose that Congress
A biased sample is a sample
establish a national golf holiday. During this honored
that does not represent the
day, all government and business offices would be
intended population and can
lead to distorted findings.
closed so that we could all run out to chase a little
Biased sampling can occur
white ball into a hole that’s way too small, with sticks
either intentionally or unintenpurposely designed by the evil golf companies to
tionally.
make this task impossible. Sounds like fun to me!
Somehow, I would need to demonstrate that the average level-headed American is in favor of this. Here is
where the genius part of my plan lies: rather than survey the general American public,
I pass out my survey form only at golf courses. But wait … it only gets better. I design
the survey to look like the following:
We would like to propose a national golf holiday, on which everybody gets the
day off from work and plays golf all day. (This means you would not need permission from your spouse.) Are you in favor of this proposal?
A. Yes, most definitely.
B. Sure, why not?
C. No, I would rather spend the entire day at work.
P.S. If you choose C, we will permanently revoke all your golfing privileges
everywhere in the country for the rest of your life. We are dead serious.
I can now honestly report back to Congress that the respondents of my survey were
overwhelmingly in favor of this new holiday. And from what we know about Congress,
they’d probably believe me.

1VO^bS`( :SbÂa5SbAbO`bSR

Another way to misuse statistics is to make differences seem greater than they actually
are by graphically presenting the data in a deceptive manner. Now that I have golf on
my brain, let me use my golf scores to demonstrate this point. Let’s say, hypothetically speaking of course, that my average golf score during the month of May was 98.
After taking some lessons in June, my average score in July dropped to 96. (For you
nongolfers, lower is better.) The graph in Figure 1.3 shows that this improvement was
nothing to write home about.

4WUc`S!

Average Golf Score

120

This graph shows the actual
difference between May and
July.

100
80
60
40
20
0
May

July
Month

However, to avoid feeling like I wasted my money on lessons, I can present the difference between May and July on a different scale, as in Figure 1.4.

4WUc`S"

Average Golf Score

This graph exaggerates the
difference between May and
July.

98
97
96
95
May

July
Month

>O`b( BVS0OaWQa
By changing the scale of the graph, it appears that I really made progress on my golf
game—when in reality, little progress was made. Oh well, back to the drawing board.
Many of the polls we see on the Internet represent another potential misuse of statistics. Many websites encourage visitors to vote on a question of the day. The results of
these informal polls are unreliable simply because those collecting the data have no
control over who responds or how many times they respond. As stated earlier, a valid
statistical study depends on selecting a sample representative from the population of
interest. This is not possible when any person surfing the Internet can participate in
the poll. Even though most of these polls state that the results are not scientific, it’s
still a natural human tendency to be influenced by the results we see.
The lesson here is that we are all consumers of statistics. We are constantly surrounded by information provided by someone who is trying to influence us or gain
our support. By having a basic understanding about the field of statistics, we increase
the likelihood that we can ward off those evil spirits in their attempts to distort the
truth. In Chapter 2, we’ll begin our journey to achieve this goal … oh, and to help
you pass your statistics course.

G]c`Bc`\
Identify each of the following statistics as either descriptive or inferential.
1. Seventy-three percent of Asian American households in the United States own a
computer.
2. Households with children under the age of 18 are more likely to have access to
the Internet (62 percent) than family households with no children (53 percent).
3. Hank Aaron hit 755 career home runs.
4. The average SAT score for incoming freshman at a local college was 950.
5. On a recent poll, 67 percent of Americans had a favorable opinion of the
President of the United States.
You can find additional sample problems on my website: www.stat-guide.com.

1VO^bS`( :SbÂa5SbAbO`bSR

BVS:SOabG]cO`b( BVS0OaWQa
Data also can be measured in many ways. The data measurement choice we make at
the start of the study will determine what kind of statistical techniques we can apply.

BVS7[^]`bO\QS]T2ObO
Data is simply defined as the value assigned to a specific observation or measurement.
If I’m collecting data on my wife’s snoring behavior, I can do so in different ways. I
can measure how many times Debbie snores over a 10-minute period. I can measure
the length of each snore in seconds. I could also measure how loud each snore is with
a descriptive phrase, like “That one sounded like a bear just waking up from hibernation” or “Wow! That one sounded like an Alaskan seal calling for its young.” (How a
sound like that can come from a person who can fit into a pair of size 2 jeans and still
be able to breathe I’ll never know.)
In each case, I’m recording data on the same event in a different form. In the first
case, I’m measuring a frequency or number of occurrences. In the second instance,
I’m measuring duration or length in time. And the final attempt measures the event by
describing volume using words rather than numbers. Each of these cases just shows a
different way to use data.
If you haven’t noticed yet, statistics people like to use all sorts of jargon, and here are a
couple more terms. Data that is used to describe something of interest about a population is called a parameter. However, if the data is describing a sample from that population, we refer to it as a statistic. For instance, let’s say that the population of interest is
my wife’s three-year-old preschool class and my measurement of interest is how many
times the little urchins use the bathroom in a day (according to Debbie, much more
than should be physically possible).
If we average the number of trips per child, this figure would be considered a parameter because the entire population was measured. However, if we want to make a statement about the average number of bathroom trips per day per three-year-old in the
country, then Debbie’s class could be our sample. We can consider the average that we
observe from her class a statistic if we assume it could be used to estimate all threeyear-olds in the country.
Data is the building blocks of all statistical studies. You can hire the most expensive,
well-known statisticians and provide them with the latest computer hardware and
software available, but if the data you provide them is inaccurate or not relevant to the
study, the final results will be worthless.

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

However, data all by its lonesome is not all that useful. By definition, data is just the
raw facts and figures that pertain to a measurement of interest. Information, on the
other hand, is derived from the facts for the purpose of making decisions. One of the
major reasons to use statistics is to transform data into information. For example, the
table that follows shows monthly sales data for a small retail store.

Data is the value assigned to an observation or a measurement and is the building
blocks to statistical analysis. The plural form is data and the singular form is datum,
referring to an individual observation or measurement.
Data that describes a characteristic about a population is known as a parameter. Data that describes a characteristic about a sample is known as a statistic.
Information is data that is transformed into useful facts that can be used for a specific
purpose, such as making a decision.

;]\bVZgAOZSa2ObO
Month

Sales ($)

January
February
March
April
May

15,178
14,293
13,492
12,287
11,321

Using statistical analysis, we can generate information that may be of interest, such
as “Wake up! You are doing something very wrong. At this rate, you will be out of
business by early next year.” Based on this valuable information, we can make some
important decisions about how to avoid this impending disaster.

BVSA]c`QSa]T2ObO¾EVS`S2]Sa/ZZBVWaAbcTT1][S
4`][We classify the sources of data into two broad categories: primary and secondary.
Secondary data is data that somebody else has collected and made available for others
to use. The U.S. government loves to collect and publish all sorts of interesting data,
just in case anyone should need it. The Department of Commerce handles census

>O`b( BVS0OaWQa
data, and the Department of Labor collects mountains of, you guessed it, labor statistics. The Department of the Interior provides all sorts of data about U.S. resources.
For instance, did you know there are 250 species of squirrels in this country? If you
don’t believe me, go to www.npwrc.usgs.gov/resource/distr/mammals/mammals/
_squirrel.htm and you can become the local “squirrel” expert.

Primary data is data that you
have collected for your own
use. Secondary data is data
collected by someone else that
you are “borrowing.”

The Canadian government has a great system for
providing statistical data to the public. Rather than
each department in the government being responsible
for collecting and disbursing data as in the United
States, Canada has a national statistical agency known
as Statistics Canada (www.statcan.ca/start.html). It’s
like one-stop shopping for the statistician. It’s a wonderful website that makes research of Canadian facts
a pleasure.

The main drawback of using secondary data is that you have no control over how the
data was collected. It’s a natural human tendency to believe anything that’s in print
(you believe me, don’t you?), and sometimes that requires a leap of faith. The advantage to secondary data is that it’s cheap (sometimes free) and it’s available now. That’s
called instant gratification.
Primary data, on the other hand, is data collected by the person who eventually uses
this data. It can be expensive to acquire, but the main advantage is that it’s your data
and you have nobody else to blame but yourself if you make a mess of it.
When collecting primary data, you want to ensure that the results will not be biased
by the manner in which it is collected. You can obtain primary data in many ways,
such as direct observation, surveys, and experiments.

Random Thoughts
The Internet has also become a rich source of data for statistics published by various industries. If you can muddle your way through the 63,278 sites that come back
from the typical Internet search engine, you might find something useful. I once found
a Japanese study on the effect of fluoride on toad embryos (www.fluoride-journal.
com/_1971.htm). Before this discovery, I was completely oblivious to the fact that
toads even had teeth, much less a cavity problem. I can’t wait to impress my friends at
the next neighborhood dinner party.

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

2W`SQb=PaS`dObW]\¾7ÂZZ0SEObQVW\UG]c
Most often, this method focuses on gathering data while the subjects of interest are
in their natural environment, oblivious to what is going on around them. Examples of
these studies would be observing wild animals stalking their prey in the forest or teenagers at the mall on Friday night (or is that the same example?). The advantage of this
method is that the subjects will unlikely be influenced by the data collection.
Focus groups are a direct observational technique where the subjects are aware that
data is being collected. Businesses use focus groups to gather information in a group
setting controlled by a moderator. The subjects are usually paid for their time and are
asked to comment on specific topics.

3f^S`W[S\ba¾EV]ÂaW\1]\b`]ZThis method is more direct than observation because the subjects will participate in
an experiment designed to determine the effectiveness of a treatment. An example of
a treatment could be the use of a new medical drug. Two groups would be established.
The first is the experimental group who receive the new drug, and the second is the
control group who think they are getting the new drug but are in fact getting no
medication. The reactions from each group are measured and compared to determine
whether the new drug was effective.
The claims that the experimental studies are attempting to verify need to be clear
and specific. I just recently read about an herb called ginkgo biloba. According to this
article, people who make money selling funny-sounding herbs claim ginkgo biloba will
keep your mind sharp as you age. Sounds like something everyone would want. Now
let’s see, where was I? As stated, this claim might prove difficult to verify. How do you
define “keeping your mind sharp”? And then, how do you measure sharpness of mind?
These are some of the challenges that statistical experiments face.
The benefit of experiments is that they allow the statistician to control factors that
could influence the results, such as gender, age, and education of the participants. The
concern about collecting data through experiments is that the response of the subjects
might be influenced by the fact that they are participating in a study. The design of
experiments for a statistical study is a very complex topic and goes beyond the scope
of this book.

>O`b( BVS0OaWQa

Ac`dSga¾7aBVObG]c`4W\OZ/\aeS`This technique of data collection involves directly asking the subject a series of questions. The questionnaire needs to be carefully designed to avoid any bias (see Chapter
1) or confusion for those participating. Concerns also exist about the influence the
survey will have on the participant’s responses. Some participants respond in a way
they feel the survey would like them to. This is very similar to the manner in which
hostages bond with their captors. The survey can be administered by e-mail, snailmail, or telephone. It’s the telephone survey that I’m most fond of, especially when I
get the call just as I’m sitting down to dinner, getting into the shower, or finally making some progress on the chapter I’m writing.

Bob’s Basics
Research has shown that the manner in which the questions are asked can affect
the responses a person provides on a questionnaire. A question posed in a positive tone will tend to invoke a more positive response and vice versa. A good
strategy is to test your questionnaire with a small group of people before releasing it to
the general public.

Whatever method you employ, your primary concern should always be that the sample is representative of the population in which you are interested.

Bg^Sa]T2ObO
Another way to classify data is by one of two types: quantitative or qualitative.
U Quantitative data uses numerical values to describe something of interest. An

example is Debbie’s age, which I have been forced by a legally bound document
to never, never, never reveal anywhere in this book, not even if it’s buried in an
appendix as an answer to an obscure question. (Hint: See page 167.)
U Qualitative data uses descriptive terms to measure or classify something of inter-

est. One example of qualitative data is the name of a respondent in a survey and
his or her level of education. The next section covers qualitative data in more
detail.

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

Bg^Sa]T;SOac`S[S\bAQOZSa¾OESWUVbgB]^WQ
Who would have thought of so many ways to look at data? The final way to classify
data is by the way it is measured. This distinction is critical because it affects which
statistical techniques we can use in our analysis of the data. Measurement classification
can be made in several levels.

<][W\OZ:SdSZ]T;SOac`S[S\b
A nominal level of measurement deals strictly with qualitative data. Observations are
simply assigned to predetermined categories. One example is gender of the respondent, with the categories being male and female. Another example is data indicating
the type of dog, if any, owned by families in a neighborhood. The categories for this
data are the various dog types (black Lab, terrier, stupid mangy mutt that keeps me
awake by barking all night at the moon). This data type does not allow us to perform
any mathematical operations, such as adding or multiplying. We also cannot rankorder this list in any way from highest to lowest (although I would put black Lab at
the top). This type is considered the lowest level of data and, as a result, is the most
restrictive when choosing a statistical technique to use for the analysis.
You can use numbers at the nominal level of measurement. Even in this case, the
rules of the nominal scale still remain. An example would be zip codes or telephone
numbers, which can’t be added or placed in a meaningful order of greater than or less
than. Even though the data appears to be numbers, it’s handled just like qualitative
data.

=`RW\OZ:SdSZ]T;SOac`S[S\b
On the food chain of data, ordinal is the next level up. It has all the properties of
nominal data with the added feature that we can rank-order the values from highest
to lowest. An example is if you were to have a lawnmower race. Let’s say the finishing
order was Scott, Tom, and Bob. We still can’t perform mathematical operations on
this data, but we can say that Scott’s lawnmower was faster than Bob’s. However, we
cannot say how much faster. Ordinal data does not allow us to make measurements
between the categories and to say, for instance, that Scott’s lawnmower is twice as
good as Bob’s (it’s not).
Ordinal data can be either qualitative or quantitative. An example of quantitative data
is rating movies with 1, 2, 3, or 4 stars. However, we still may not claim that a 4-star
movie is 4 times as good as a 1-star movie.

>O`b( BVS0OaWQa

7\bS`dOZ:SdSZ]T;SOac`S[S\b
Moving up the scale of data, we find ourselves at the interval level, which is strictly
quantitative data. Now we can get to work with the mathematical operations of
addition and subtraction when comparing values. For this data, we can measure the
difference between the different categories with actual numbers and also provide
meaningful information. Temperature measurement in degrees Fahrenheit is a common example here. For instance, 70 degrees is 5 degrees warmer than 65 degrees.
However, multiplication and division can’t be performed on this data. Why not?
Simply because we cannot argue that 100 degrees is twice as warm as 50 degrees.

@ObW]:SdSZ]T;SOac`S[S\b
The king of data types is the ratio level. This is as good as it gets as far as data is concerned. Now we can perform all four mathematical operations to compare values with
absolutely no feelings of guilt. Examples of this type of data are age, weight, height,
and salary. Ratio data has all the features of interval data with the added benefit of
a true 0 point. The term “true zero point” means that a 0 data value indicates the
absence of the object being measured. For instance, 0 salary indicates the absence of
any salary.

Wrong Number
Interval data does not
have a true 0 point. For
example, 0 degrees Fahrenheit
does not represent the absence
of temperature, even though it
may feel like it. To help explain
this, try baking a cake at twice
the recommended temperature
in half the recommended time.
Yuck!

With a true 0 point, we can use the rules of multiplication and division to compare data values. This
allows us to say that a person who is 6 feet in height
is twice as tall as a 3-foot person or that a 20-yearold person is half the age of a 40 year old.
The distinction between interval and ratio data is a
fine line. To help identify the proper scale, use the
“twice as much” rule. If the phrase “twice as much”
accurately describes the relationship between two
values that differ by a multiple of 2, then the data can
be considered ratio level.

There are endless examples of ratio data. Let’s look
at measuring typing speed in words per minute. I
happen to be a handicapped, two-finger, hunt-and-peck typist who has tried those
darned typing programs more than once and just can’t get it. I can type maybe 20
words a minute on a good day. My 15-year-old son, John, on the other hand, is one of
those show-offs who types while he’s not even looking and can still type 60 words a minute. Because we can correctly say that John types three times faster than me, typing
speed is an example of ratio data.

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

Figure 2.1 summarizes the different data scales and how they relate to one another. As
we explore different statistical techniques later in this book, we will revisit these different measurement scales. You will discover that specific techniques require certain
types of data.

4WUc`S

Types of Data

Summary of data measurement scales.
Qualitative

Nominal

Quantitative

Ordinal

Interval

Ratio

1][^cbS`ab]bVS@SaQcS
As mentioned in Chapter 1, we will explore the use of Excel in solving some of the
statistics problems in this book. If you have no interest in using Excel in this manner, just skip this section. I promise you won’t hurt my feelings. The purpose of this
last section is to talk about the use of computers with statistics in general and then to
make sure your computer is ready to follow us along.

BVS@]ZS]T1][^cbS`aW\AbObWabWQa
When I was a youthful engineering undergraduate student during the 1970s, the
words “personal computer” had no meaning. I performed calculations on a clever gadget fondly known as a “slide rule.” For those of you who weren’t even alive during this
time period, I’ve included a picture of one in Figure 2.2.

4WUc`S
Slide rule circa 1975.
(Courtesy of www.hpmuseum.org)

As you can see, this device looks like a ruler on steroids. It can perform all sorts of
mathematical calculations but is far from being user friendly. During my freshman
year in college, I purchased my first handheld calculator, a Texas Instrument model
that could only perform the basic math functions. It was the approximate size of a cash
register.

>O`b( BVS0OaWQa
At this point, the only serious statistical analysis was being performed on mainframe
computers by people with high levels of programming skill. These people were
somewhat “different” from the rest of us. Fortunately, we have advanced from the
Dark Ages and now have awesome, user-friendly computing power at our fingertips.
Powerful programs such as SAS, SPSS, Minitab, and Excel are readily accessible to
those of us who don’t know a lick of computer programming and allow us to perform
some of the most sophisticated statistical analysis known to humankind.
Parts of this book will demonstrate how to solve some of the statistical techniques
using Microsoft Excel. Choosing to skip these parts will not interfere with your grasp
of topics in subsequent chapters. This is simply optional material to expose you to
statistical analysis on the computer. I also assume you already have a basic working
knowledge of how to use Excel.

7\abOZZW\UbVS2ObO/\OZgaWa/RR7\
Our first task is to check whether Excel’s data analysis tools are available on your computer. Open the Excel program and left-click with your mouse on the Tools menu as
shown in Figure 2.3. From this point on in the book, I’ll use the term “click” to mean
click the left button on your mouse.

4WUc`S !
Excel’s Tools menu.

Notice in the figure that the highlighted item is Data Analysis, which may or may
not show up under your Tools menu. If Data Analysis does appear under your Tools
menu, skip the rest of this paragraph and the next two and proceed to the following
paragraph that begins with “Click on Data Analysis …”

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y
If Data Analysis does not appear under the
Tools menu, you need to add it to the menu. To
do so, click on Add-Ins under the same Tools
menu. If you don’t see Add-Ins under this menu
at first, expand the menu items by clicking on
the downward arrow at the bottom of the Tools
menu list. After clicking on Add-Ins, you should
see the dialog box in Figure 2.4.

Random Thoughts
If your Tools menu looks different from the one in Figure 2.3,
it might be because all of your
available menu items are not
currently visible. To make all the
menu items visible, click on the
Expand symbol at the bottom of
the list (the double-downwardpointing arrows).

4WUc`S "
Excel’s Add-Ins dialog box.

This dialog box provides a list of available add-ins for you to use. Click on the empty
box for Analysis ToolPak, which places a check mark in it, and then click OK. Now
click on the Tools menu again; Data Analysis will now appear in the list.

Random Thoughts
Don’t panic if you receive the following message: “Microsoft Excel can’t run this addin. This feature is not currently installed. Would you like to install it now?” If you want
to install the Analysis ToolPak, you might need to have your original Microsoft Office
CD close by. Click the Yes button and follow any further instructions. Then, the Data
Analysis option will become available on the Tools menu.

>O`b( BVS0OaWQa
Click on Data Analysis under the Tools menu to open the dialog box shown in
Figure 2.5.

4WUc`S #
Excel’s Data Analysis dialog
box.

Your Excel program is now ready to perform all sorts of statistical magic for you as we
explore various techniques throughout this book. At this point, you can click Cancel
and close out Excel. Each time you open Excel in the future, the Data Analysis option
will be available under the Tools menu.

G]c`Bc`\
Classify the following data as nominal, ordinal, interval, or ratio. Explain your choice.
1. Average monthly temperature in degrees Fahrenheit for the city of Wilmington
throughout the year
2. Average monthly rainfall in inches for the city of Wilmington throughout the
year
3. Education level of survey respondents
Level
High school
Bachelor’s degree
Master’s degree

Number of Respondents
168
784
212

4. Marital status of survey respondents
Status
Single
Married
Divorced

Number of Respondents
28
189
62

5. Age of the respondents in the survey
6. Gender of the respondents in the survey

1VO^bS` ( 2ObO2ObO3dS`geVS`SO\R<]bO2`]^b]2`W\Y

7. The year in which the respondent was born
8. The voting intentions of the respondents in the survey classified as Republican,
Democrat, or Undecided
9. The race of the respondents in the survey classified as White, African American,
Asian, or Other
10. Performance rating of employees classified as Above Expectations, Meets
Expectations, or Below Expectations
11. The uniform number of each member on a sports team
12. A list of the graduating high school seniors by class rank
13. Final exam scores for my statistics class on a scale of 0 to 100
14. The state in which the respondents in a survey reside

BVS:SOabG]cO`b( BVS0OaWQa

4`S_cS\Qg2Wab`WPcbW]\a
One of the most common ways to graphically describe data is through the use of frequency distributions. The best way to describe a frequency distribution is to start with
an example.
Ever since I was a young boy, I have been a devoted
fan of the Pittsburgh Pirates Major League Baseball
team. Why I still root for these guys, I’ll never know,
because they have not had a winning season since
1992. Anyway, below is a table of the batting averages
of individual Pirates at the end of the 2005 season. I
have not attached names with these averages in order
to protect their identities.

A frequency distribution is a
table that shows the number of
data observations that fall into
specific intervals.

Pittsburgh Pirates Final Batting Averages for the 2005 Season

.306
.255
.158

.257
.264
.113

.272
.221
.106

.291
.258
.119

.260
.341
.182

.273
.257
.143

.268
.222
.143

.251
.269
.192

.242
.241
.261

Source: www.espn.com

It is difficult to grasp what a tough year these guys had by just looking at this data in
this table format. Transforming this data into the frequency distribution shown here
makes this fact more obvious.
Batting Average

Number of Players

.000 to .049
.050 to .099
.100 to .149
.150 to .199
.200 to .249
.250 to .299
.300 to .349

0
0
5
3
4
13
2

As you can see, a frequency distribution is simply a table that organizes the number of
data values into intervals. In this example, the intervals are the batting average ranges
in the first column of the table. The number of data values is the number of players

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

who fall into each interval shown in the second column. Well, there’s always next season to look forward to.
The intervals in a frequency distribution are officially known as classes, and the number of observations in each class is known as class frequencies. Now let’s learn how to
arrange these classes.

1]\ab`cQbW\UO4`S_cS\Qg2Wab`WPcbW]\
You need to make some important decisions when constructing a frequency distribution. To illustrate these decisions, let’s use another example, something many of us can
relate to—cell phones! My son John and I are on one of those “family share plans,”
which means he gets all the peak minutes and I get to use my phone between the
hours of 3 A.M. and 6 A.M. every other Saturday. The following table represents the
number of calls each day during the month of May on John’s account.
Calls per Day

3
3
6
2
3
1

1
9
4
5
0
8

2
1
9
5
1
6

1
4
13
2
2
9

1
2
15
7
7
4

Source: A very confusing phone bill that requires a Ph.D. in metaphysical telecommunications to understand.

Using this data, I have constructed the following frequency distribution.

4`S_cS\Qg2Wab`WPcbW]\
Calls per Day

Number of Days

0–2
3–5
6–8
9–11
12–14
15–17

12
8
5
3
1
1

>O`b( BVS0OaWQa
When arranging these classes, I followed these rules:
1. From classes of equal size. I chose 3 data values to be in each class for this distribution. An example of a class is 0–2, which includes the number of days when
John made 0, 1, or 2 calls.
2. Make classes mutually exclusive, or in other words, prevent classes from overlapping. For instance, I wouldn’t want 2 classes to be 3–5 and 5–7 because 5 calls
would be in 2 different classes.
3. Try to have no fewer than 5 classes and no more than 15 classes. Too few or too
many classes tend to hide the true characteristics of the frequency distribution.
4. Avoid open-ended classes, if possible (for instance, a highest class of 15–over).

Classes are considered mutually exclusive when observations can only fall into one
class. For example, the gender
classes “male” and “female”
are mutually exclusive because
a person cannot belong to both
classes.

5. Include all data values from the original table
in a class. In other words, the classes should be
exhaustive.
Too few or too many classes will obscure patterns in
a frequency distribution. Consider the extreme case
where there are so many classes that no class has
more than one observation. The other extreme is
where there is only one class with all the observations
residing in that class. This would be a pretty useless
frequency distribution!

/2WabO\b@SZObWdS4`S_cS\Qg2Wab`WPcbW]\
Another way to display frequency data is by using the relative frequency distribution.
Rather than display the number of observations in each class, this method calculates
the percentage of observations in each class by dividing the frequency of each class by
the total number of observations. I can display John’s data as a relative frequency distribution, as I do in the following table.

Relative frequency distributions display the percentage of observations of each class
relative to the total number of observations.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

@SZObWdS4`S_cS\Qg2Wab`WPcbW]\
Calls per Day

0–2
3–5
6–8
9–11
12–14
15–17

Number of Days

Percentage
12/
30
8/
30
5/
30
3/
30
1/
30
1/
30

12
8
5
3
1
1
Total = 30

= 0.40
= 0.27
= 0.17
= 0.10
= 0.03
= 0.03
Total = 1.00

According to this distribution, John uses his phone 3 to 5 times 27 percent of the days
during a month.
The total percentage in a relative frequency distribution should be 100 percent or very
close (within 1 percent, because of rounding errors).

1c[cZObWdS4`S_cS\Qg2Wab`WPcbW]\
This “kissing cousin” of the relative frequency
distribution simply totals the percentages of
each class as you move down the column. (Get
it? Cousin, relative? Sorry, I couldn’t help
myself!) This provides you with the percentage
of observations that are less than or equal to
the class of interest. The resulting cumulative
frequency distribution is shown here.

Cumulative frequency distributions indicate the percentage of
observations that are less than
or equal to the current class.

1c[cZObWdS4`S_cS\Qg2Wab`WPcbW]\
Calls per Day

No. of Days

Percentage

Cumulative Percentage

0–2

3–5

6–8

9–11

12–14

15–17

12/
30
8/
30
5/
30
3/
30
1/
30
1/
30

Total = 30

= 0.40

0.40

= 0.27

0.67

= 0.17

0.84

= 0.10

0.94

= 0.03

0.97

= 0.03

1.00

Total = 1.00

>O`b( BVS0OaWQa
The value 0.67 in the fourth column is the result of adding 0.40 to 0.27. According to
this table, John used his phone 8 times or less on 84 percent of the days in the month.
If designed properly, frequency distribution is an excellent way to tease good information out of stubborn data. Now, let’s deal with how to display the distribution graphically.

5`O^VW\UO4`S_cS\Qg2Wab`WPcbW]\¾bVS6Wab]U`O[
A histogram is simply a bar graph showing the number of observations in each class as
the height of each bar. Figure 3.1 shows the histogram for John’s phone calls.

A histogram is a bar graph showing the number of observations in each class as the
height of each bar.

This graph gives us a good visual of John’s calling habits. At least the highest class on
the graph is the 0 to 2 calls per day. Things could be worse.
14

A histogram of John’s phone
calls.

Number of Observations

4WUc`S!

10
8
6
4
2
0
0-2

3-5

6-8

9-11

12-14

15-17

Calls per Day

:SbbW\U3fQSZ2]=c`2W`bgE]`Y
Excel will actually construct a frequency distribution for us and plot the histogram.
How nice!
1. The first thing we need to do is open Excel to a blank sheet and enter our data
in Column A starting in Cell A1 (use the data from the earlier table).

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa
2. Next enter the upper limits to each class
in Column B starting in Cell B1. For
example, in the class 0–2, the upper limit
would be 2. Figure 3.2 shows what the
spreadsheet should look like.

Random Thoughts
For some bizarre reason, Excel
refers to classes as bins. Go
figure.

4WUc`S!
Raw data for the frequency
distribution.

3. Go to the Tools menu at the top of the Excel window and select Data Analysis.
(Refer to the section “Installing the Data Analysis Add-in” from Chapter 2 if you
don’t see the Data Analysis command on the Tools menu.)
4. Select the Histogram option from the list of Analysis Tools (see Figure 3.3) and
click the OK button.

>O`b( BVS0OaWQa

4WUc`S!!
Data Analysis dialog box.

5. In the Histogram dialog box (as shown in Figure 3.4), click in the Input Range
list box, and then click in the worksheet to select cells A1 through A30 (the 30
original data values). Then, click in the Bin Range list box and click in the worksheet to select cells B1 through B6 (the upper limits for the 6 classes).
6. Click the New Worksheet Ply option button and the Chart Output check box
(see Figure 3.4).

4WUc`S!"
Histogram dialog box.

7. Click OK to generate the frequency distribution and histogram (see Figure 3.5).

4WUc`S!#
Frequency distribution and
histogram.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa
Notice that Excel has generated the frequency
distribution for us in columns A and B. Cool!
The problem here is that the histogram looks
like an elephant sat on it. Click on the chart to
select it, and then click on the bottom border
to drag the bottom of the chart down lower,
expanding the histogram to look like Figure 3.6.
Frequency distributions and histograms are
convenient ways to get an accurate picture of
what your data is trying to tell you. It sounds
like my data is telling me to “get more monthly
minutes on your cell phone plan.” Wonderful.

Random Thoughts
I prefer using Chart Wizard to
display the histogram because I
think the graph looks better than
when I use the Data Analysis
tool. The Chart Wizard allows
me more control over the final
appearance.

4WUc`S!$
Final histogram.

AbObWabWQOZ4Z]eS`>]eS`¾bVSAbS[O\R:SOT2Wa^ZOg
The stem and leaf display is another graphical technique you can use to display your
data. A statistician named John Tukey originated the idea during the 1970s. The major
benefit of this approach is that all the original data points are visible on the display.
To demonstrate this method, I will use my son Brian’s golf scores for his last 24
rounds, shown in the following table. Normally, Brian would only report his better
scores, but we statisticians must be unbiased and accurate.

>O`b( BVS0OaWQa
Brian’s Golf Scores

81
79
84

86
83
79

78
84
80

80
95
83

81
85
79

82
88
87

92
80
84

90
78
80

Figure 3.7 shows the stem and leaf display for these scores.

4WUc`S!%
Stem and leaf display.

7 88999
8 00001123344445678
9 025
Stem and Leaf Display

The “stem” in the display is the first column of numbers, which represents the first
digit of the golf scores. The “leaf” in the display is the second digit of the golf scores,
with 1 digit for each score. Because there were 5 scores in the 70s, there are 5 digits to
the right of 7.
If we choose to, we can break this display down further by adding more stems. Figure 3.8 shows this
approach.
The stem and leaf display
splits the data values into stems
(the first digit in the value) and
leaves (the remaining digits in
the value). By listing all of the
leaves to the right of each stem,
we can graphically describe
how the data is distributed.

Here, the stem labeled 7 (5) stores all the scores
between 75 and 79. The stem 8 (0) stores all the
scores between 80 and 84. After examining this display, I can see a pattern that’s not as obvious when
looking at Figure 3.7: Brian usually scores in the low
80s.
You can find an excellent source of information about
stem and leaf displays at the Statistics Canada website
at www.statcan.ca/english/edu/power/ch8/plots.htm.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa
7 (5)
8 (0)
8 (5)
9 (0)
9 (5)

88999
000011233444
5678
02
5

4WUc`S!&
A more detailed stem and leaf
display.

More Detailed Stem and Leaf Display

1VO`bW\UG]c`1]c`aS
Charts are yet another efficient way to summarize and display patterns in a set of data,
so let me demonstrate different types of charts that help us “tell it like it is.”

EVObÂaG]c`4Od]`WbS>WS1VO`bPie charts are commonly used to describe data from relative frequency distributions.
This type of chart is simply a circle divided into portions whose area is equal to the
relative frequency distribution. To illustrate the use of pie charts, let’s say some anonymous statistics professor submitted the following final grade distribution.

4W\OZAbObWabWQa5`ORSa
Grade

A
B
C
D

Number of Students

9
13
6
2
Total = 30

Relative Frequency
9/
30

= 0.30
= 0.43
6/ = 0.20
30
2/ = 0.07
30
Total = 1.00
13/
30

We can illustrate this relative frequency distribution by using the pie chart in Figure
3.9. This chart was done using Excel’s Chart Wizard.

>O`b( BVS0OaWQa

4WUc`S!'

D
7%

Pie chart illustrating a grade
distribution.

A
30%

C
20%

B
43%

Bob’s Basics
Pie charts are an excellent
way to colorfully present data
from a relative frequency
distribution. If you cannot use
colors, use patterns and textures
to display pie charts.

As you can see, the pie chart approach is much easier
on the eye when compared to looking at data from
a table. This person must be a pretty good statistics
teacher!
To construct a pie chart by hand, you first need to
calculate the center angle for each slice in the pie,
which is illustrated in Figure 3.10.

You determine the center angle of each slice by multiplying the relative frequency of the class by 360 (which is the number of degrees in a
circle). These results are shown in the following table.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

4WUc`S!
The center angle of a pie
chart slice.

Center
Angle

1S\bS`/\UZST]`>WS1VO`b1]\ab`cQbW]\
Grade

Relative Frequency

Central Angle

A
B
C
D

9/
30
13/
30
6/
30
2/
30

0.30 * 360 = 108 degrees
0.43 * 360 = 155 degrees
0.20 * 360 = 72 degrees
0.07 * 360 = 25 degrees

= 0.30
= 0.43
= 0.20
= 0.07
Total = 1.00

By using a device to measure angles, such as a protractor, you can now divide your pie
chart into slices of the appropriate size. This assumes, of course, you’ve mastered the
art of drawing circles.

0O`1VO`ba
Bar charts are a useful graphical tool when you are plotting individual data values next
to each other. To demonstrate this type of chart (see Figure 3.11), we’ll use the data
from the following table, which represents the monthly credit card balances for an
unnamed spouse of an unnamed person writing a statistics book. (Boy, I’m going to be
in big trouble when she sees this.)

>O`b( BVS0OaWQa

/\]\g[]ca1`SRWb1O`R0OZO\QSa
Month

Balance ($)

1
2
3
4
5
6

375
514
834
603
882
468

Source: An unnamed filing cabinet.

4WUc`S!
Credit Card Balance ($)

1000

Bar chart for somebody’s
credit card balances.

800
600
400
200
0
1

Month

Random Thoughts
By now you may have just said to yourself, “Hey, wait one minute! Haven’t I seen this
somewhere before?” By “this” I hope you’re referring to the type of chart rather than
my wife’s credit card statements. The histogram that we visited earlier in the chapter
is actually a special type of bar chart that plots frequencies rather than actual data
values.

I’m sure your inquisitive mind is now screaming with the question “How do I choose
between a pie chart and a bar chart?” If your objective is to compare the relative size
of each class to one another, use a pie chart. Bar charts are more useful when you want
to highlight the actual data values.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

:W\S1VO`ba
The last graphical tool discussed here is a line chart (sometimes called a line graph),
which is used to help identify patterns between two sets of data. To illustrate the use
of line charts, we’ll use a favorite topic of mine: teenagers and showers.
Our current resident teenagers seem to have a costly compulsion to take very long, very
hot showers, and sometimes more than once a day. As I lie awake at night listening to
the constant stream of hot water, all I can envision are dollar bills flowing down the
drain. So I have tabulated some data, which shows the number of showers the cleanest
kids on the block have taken in each of the recent months with the corresponding utility bill. Notice that at these rates we average more than one shower per day.
Month

Number of Showers

Utility Bill

1
2
3
4
5
6

72
91
98
82
76
85

$225
$287
$260
$243
$254
$275

To see whether there is any pattern between the number of showers and the utility
bill, we can plot the pairs of data for each month on a line chart, which is shown in
Figure 3.12.

Monthly Utility Bill ($)

300

4WUc`S!
A line chart for the number
of showers and the utility bill.

280

260

240

220
70

Number of Showers

100

>O`b( BVS0OaWQa
I have chosen to place the number of showers on the x-axis (horizontal) of the chart
and the utility bills on the y-axis (vertical) of the chart. Because the line connecting
the data points seems to have an overall upward trend, my suspicions hold true. It
seems the more showers our waterlogged darlings take, the higher the utility bill.
Line charts prove very useful when you are interested in exploring patterns between
two different types of data. They are also helpful when you have many data points and
want to show all of them on one graph.
Now that you have mastered the art of displaying descriptive statistics, you are ready
to move on to calculating them in the next chapter.

G]c`Bc`\
1. The following table represents the exam grades from 36 students from a certain
class that I might have taught. Construct a frequency distribution with 9 classes
ranging from 56 to 100.
Exam Scores
60

2. Construct a histogram using the solution from Problem 1.
3. Construct a relative and a cumulative frequency distribution from the data in
Problem 1.
4. Construct a pie chart from the solution to Problem 1.
5. Construct a stem and leaf diagram from the data in Problem 1 using one stem
for the scores in the 50s, 60s, 70s, 80s, and 90s.
6. Construct a stem and leaf diagram from the data in Problem 1 using two stems
for the scores in the 50s, 60s, 70s, 80s, and 90s.

1VO^bS`!( 2Wa^ZOgW\U2SaQ`W^bWdSAbObWabWQa

BVS:SOabG]cO`b( BVS0OaWQa
As mentioned in Chapter 1, descriptive statistics form the foundation for practically all
statistical analysis. If these are not calculated with loving care, our final analysis could
be misleading. And as everybody knows, statisticians never want to be misleading.
So this chapter focuses on how to calculate descriptive statistics manually and, if you
choose, how to verify these results with our good friend Excel.
This is the first chapter that uses mathematical formulas that have all those funnylooking Greek symbols that can make you break out into a cold sweat. But have no
fear. We will slay these demons one by one through careful explanation and, in the
end, victory will be ours. Onward!

;SOac`Sa]T1S\b`OZBS\RS\Qg
There exist two broad categories of descriptive statistics that are commonly used. The
first, measures of central tendency, describes the center point of our data set with a single
value. It’s a valuable tool to help us summarize many pieces of data with one number.
The second category, measures of dispersion, is the topic of Chapter 5. But let’s explore
the many ways to measure the central tendency of our data.

Measures of central tendency describe the center point of a data set with a single
value. Measures of dispersion describe how far individual data values have strayed
from the mean.

;SO\
The most common measure of central tendency is the mean or average, which we calculate by adding all the values in our data set and then dividing this result by the number of observations. The mathematical formula for the mean differs slightly depending
on whether you’re referring to the sample mean or the population mean. The formula
for the sample mean is as follows:
n

¤ xi
x

i 1

The mean or average is the most common measure of central tendency and is calculated by adding all the values in our data set and then dividing this result by the number of observations.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS

where:
x = the sample mean
x i = the values in the sample (x1 = the first data value, x2 = the second data value,
and so on)
n

¤ xi

= the sum of all the data values in the sample

i 1

n = the number of data values in the sample

Bob’s Basics
n

Don’t panic when you see the symbol

¤ x i , which means “the sum of x for
i

i 1

i = 1 to n.” If our data sample contains the values 5, 8, and 2, then n = 3,
x1 = 5, x2 = 8, and x3 = 2, resulting in the expression:
n

¤ xi x1
i 1

x 3 5 8 2 15

The formula for the population mean is as follows:
n

¤ xi
M

i 1

where:

M = the population mean (pronounced mu, as in “I hope you find this amusing”)
n

¤ xi

= the sum of all the data values in the population

i 1

N = the number of data values in the population
To demonstrate calculating measures of central tendency, let’s use the following example. As in many teenage households, video games are a common form of entertainment in our family room. Brian and John love to challenge me with a game and then
clean my clock before I can ask “Which team is mine?” I suspected John of sticking
me with the “bad” controller because it felt like a 10-second delay between pushing
a button and the game responding. (Turns out the delay was really between my brain
and my fingers.) Anyway, the following data set represents the number of hours each
week that video games are played in our household.

>O`b( BVS0OaWQa
3

Because this data represents a sample, we will calculate the sample mean:
n

¤ xi
s

i 1

3 7 4 9 5 4 6 17 4 7
6.6 hours
10

It looks like I need some serious practice time to catch up to these guys.

ESWUVbSR;SO\
When we calculated the mean number of hours in the previous example, we gave each
data value the same weight in the calculation as the other values. A weighted mean
refers to a mean that needs to go on a diet. Just kidding; I was checking to see whether
you were paying attention. A weighted mean allows you to assign more weight to certain values and less weight to others. For example, let’s say your statistics grade this
semester will be based on a combination of your final exam score, a homework score,
and a final project, each weighted according to the following table.
Type

Score

Weight (Percent)

Exam
Project
Homework

94
89
83

50
35
15

We can calculate your final grade using the following formula for a weighted average.
Note that here we are dividing by the sum of the weights rather than by the number
of data values.
n

¤ wi x i
x

i 1
n

¤ wi
i 1

Bob’s Basics
n

The symbol ¤ w x means “the sum of w times x.” Each pair of w and x is first
i i
i 1

multiplied together, and these results are then summed.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS

where:
wi = the weight for each data value xi
n

¤ wi

= the sum of the weights

i 1

The previous equation can be set up in the following table to demonstrate the procedure.

Type
i

Score
xi

Weight
wi

Weight × Score
(wi xi)

Exam 1
Project 2
Homework 3

94
89
83

0.50
0.35
0.15

47.0
31.2
12.4

¤ wi xi

¤ wi 1.0

90.6

i 1

We can obtain the same result by plugging the numbers directly in to the formula for
a weighted average:
x

0.50 94 0.35 89 0.15 83
0.50 0.35 0.15

47.0 31.2 121.4
90.6
1.0

Congratulations. You earned an A!
The weights in a weighted average do not need to add to 1 as in the previous example.
Let’s say I want a weighted average of my two most recent golf scores, 90 and 100,
and I want 90 to have twice the weight as 100 in my average. I would calculate this by:

2 90 1 100

93.3
2 1
By giving more weight to my lower score, the result is lower than the true average of
95. In this case, I think I’ll go with the weighted average.
x

;SO\]T5`]c^SR2ObOT`][O4`S_cS\Qg2Wab`WPcbW]\
Here’s some great news to get excited about! You can actually calculate the mean
of grouped data from a frequency distribution. Recall the data set from Chapter 3
regarding John’s cell phone calls per day shown in the following table.

>O`b( BVS0OaWQa
Calls per Day

3
3
6
2
3
1

1
9
4
5
0
8

2
1
9
5
1
6

1
4
13
2
2
9

1
2
15
7
7
4

The following table shows this data as a frequency distribution with the calls per day
as the class.

4`S_cS\Qg2Wab`WPcbW]\
Calls per Day

Number of Days

0–2
3–5
6–8
9–11
12–14
15–17

12
8
5
3
1
1

To calculate the mean of this distribution, we first need to determine the midpoint of
each class using the following method:
Lower Value Upper Value
Class Midpoint =
2
For instance, the class midpoint for the last class would be as follows:
15 17
16
Class Midpoint =
2
We can use the following table to assist in the calculations.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS
Class

Midpoint (x)

Frequency (f)

0–2
3–5
6–8
9–11
12–14
15–17

1
4
7
10
13
16

12
8
5
3
1
1

After we have determined the midpoint for each class, we can calculate the mean of
the frequency distribution using the following equation—which is basically a weighted
average formula:
n

¤ f i xi
x

i 1

¤ fi
i 1

where:

xi = the midpoint for each class for i = 1 to n
fi = the number of observations (frequency) of each class for i = 1 to n
n = the number of classes in the distribution
We determine the mean of this frequency distribution as follows:
x

12 1 8 4 5 7 3 10 1 13 1 16
12 8 5 3 1 1

4.6 calls

According to the mean of this frequency distribution, John averages 4.6 calls per day
on his cell phone.

Wrong Number
The mean of a frequency distribution where data is grouped into classes is only
an approximation to the mean of the original data set from which it was derived.
This is true because we make the assumption that the original data values are at the
midpoint of each class, which is not necessarily the case. The true mean of the 30
original data values in the cell phone example is only 4.5 calls per day rather than
4.6.

>O`b( BVS0OaWQa
If the classes in the frequency distribution are a single value rather than an interval,
calculate the mean by treating the distribution as a weighted mean. For example, let’s
say the following table represents the number of days that a hardware store experienced various daily demands for a particular hammer during the past 65 days of business.
Daily Demand (x)

Number of Days ( f )

0
1
2
3
4
5

10
15
12
18
6
4
Total = 65

For instance, there were 15 days in the past 65 days that the store experienced demand
for one hammer. What is the average daily demand during the past 65 days?
n

¤ f i xi
x

i 1

¤ fi
i 1

10 0

15 1 12 2 18 3 6 4 4 5
10 15 12 18 6 4

137
x
2.1 hammers per day
65
Now that we have become experts in every conceivable method for calculating a
mean, we are ready to move on to the other cool methods of measuring central tendency.

;SRWO\
Another way to measure central tendency is by finding the median. The median is the
value in the data set for which half the observations are higher and half the observations are lower. We find the median by arranging the data values in ascending order
and identifying the halfway point.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS

Using our example with the video games, we rearrange our data set in ascending
order:
3

Because we have an even number of data
points (10), the median is the average of the
two center points. In this case, that will be
the values 5 and 6, resulting in a median of
5.5 hours of video games per week. Notice
that there are four data values to the left (3,
4, 4, and 4) of these center points and four
data values to the right (7, 7, 9, and 17).

The median is a measure of
central tendency that represents
the value in the data set for
which half the observations are
higher and half the observations
are lower. When there is an
even number of data points, the
median will be the average of
the two center points.

To illustrate the median for a data set with
an odd number of values, let’s remove 17
from the video games data and repeat our
analysis.
3

In this instance, we only have one center point, which is the value 5. Therefore, the
median for this data set is 5 hours of video games per week. Again, there are four data
values to the left and right of this center point.

;]RS
The last measure of central tendency on my mind is the mode, which is simply the
observation in the data set that occurs the most frequently.
To illustrate the mode for a data set, let’s again use the original video game data.
3

The mode is 4 hours per week because this
value occurs three times in the data set.
That wraps up all the different ways to measure
central tendency of our data set. However, one
question is screaming to be answered, and that
is …

Random Thoughts
There can be more than one
mode of a data set if more than
one value occurs the most frequent number of times.

>O`b( BVS0OaWQa

6]e2]Sa=\S1V]]aSI bet you never thought you would have so many choices of measuring central tendency! It’s kind of like being in an ice cream store in front of 30 flavors. If you think
all the data in your data set is relevant, then the mean is your best choice. This measurement is affected by both the number and magnitude of your values. However, very
small or very large values can have a significant impact on the mean, especially if the
size of the sample is small. If this is a concern, perhaps you should consider using the
median. The median is not as sensitive to a very large or small value.
Consider the following data set from the original video game example:
3

The number 17 is rather large when compared to the rest of the data. The mean of
this sample was 6.6, whereas the median was 5.5. If you think 17 is not a typical value
that you would expect in this data set, the median would be your best choice for central tendency.
The poor lonely mode has limited applications. It is primarily used to describe data
at the nominal scale—that is, data that is grouped in descriptive categories such as
gender. If 60 percent of our survey respondents were male, then the mode of our data
would be male.

CaW\U3fQSZb]1OZQcZObS1S\b`OZBS\RS\Qg
Excel will kindly calculate the mean, median, and mode for you all at once with a few
mouse clicks. Let’s demonstrate this using the data set from the video game example.
1. To begin, open a blank Excel worksheet and enter the video game data (Figure
4.1).
2. Click on the Tools menu at the top of the spreadsheet (between Format and
Data) and select Data Analysis. (See the section “Installing the Data Analysis
Add-in” in Chapter 2 for more details on this step if you don’t see the Data
Analysis command.) After selecting Data Analysis, you should see the dialog box
shown in Figure 4.2.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS

4WUc`S"
Enter data from the video
game example.

4WUc`S"
Data Analysis dialog box.

3. Select Descriptive Statistics and click OK. The following dialog box will appear
(Figure 4.3).

4WUc`S"!
Descriptive Statistics dialog
box.

>O`b( BVS0OaWQa
4. For the Input Range, select cells A1 through A10, select the Output Range
option, and select cell C1. Then choose the Summary statistics check box and
click OK.
5. After you expand columns C and D slightly to see all the figures, your spreadsheet should look like Figure 4.4.

4WUc`S""
Measures of central tendency
for the video game example.

As you can see, the mean is 6.6 hours, the median is 5.5 hours, and the mode is 4.0
hours. Piece of cake!

G]c`Bc`\
1. Calculate the mean, median, and mode for the following data set: 20, 15, 24, 10,
8, 19, 24, 12, 21, 6.
2. Calculate the mean, median, and mode for the following data set: 84, 82, 90, 77,
75, 77, 82, 86, 82.
3. Calculate the mean, median, and mode for the following data set: 36, 27, 50, 42,
27, 36, 25, 40, 29, 15.
4. Calculate the mean, median, and mode for the following data set: 8, 11, 6, 2, 11,
6, 5, 6, 10.

1VO^bS`"( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T1S\b`OZBS\RS\Qg;SO\;SRWO\O\R;]RS

5. A company counted the number of their employees in each of the following age
classes. According to this distribution, what is the average age of the employees
in the company?
Age Range

Number of Employees

20–24

25–29

30–34

35–39

40–44

45–49

6. Calculate the weighted mean of the following values with the corresponding
weights.
Value

Weight

118

125

107

7. A company counted the number of employees at each level of years of service in
the following table. What is the average number of years of service in this company?
Years of Service

Number of Employees

>O`b( BVS0OaWQa

BVS:SOabG]cO`b( BVS0OaWQa
values varied between 1 and 12 hours. As you will see later, this distinction can be very
important.
To address this issue, we rely on the second major category of descriptive statistics,
measures of dispersion, which describes how far the individual data values have strayed
from the mean. So let’s look at the different ways we can measure dispersion.

@O\US
The range is the simplest measure of dispersion and is calculated by finding the difference between the highest value and the lowest value in the data set. To demonstrate
how to calculate the range, I’ll use the following example.
One of Debbie’s special qualities is that she is a dedicated grill–a-holic when it comes
to barbequing in the backyard. Recently, we needed to purchase a new grill since our
old one mysteriously caught fire one night when I was at school teaching. The cause
of the fire was labeled “suspicious” after Debbie saw this event as a wonderful opportunity to “upgrade.” My idea of the perfect grill is one you dump charcoal in, add
one can of lighter fluid to, toss in a match, and run for your life. The best part about
this kind of grill is that it has about four parts to assemble, which is something I can
easily put together in 3 or 4 weeks. Within minutes of arriving at the store, I felt a
stabbing pain in my chest upon finding my wife in an animated conversation with a
total stranger about a grill for the “serious barbeque
person” complete with three burners, electronic ignition, back-up propane tank, 300 horsepower, and
front disc brakes. This thing could barbeque a pig
Obtain the range of a sample
on a spit faster than I could say “oink.” I’d have betby subtracting the smallest
measurement from the largest
ter luck assembling a car from scratch. As protection
measurement.
from future acts of arson, I purchased a life insurance
policy for our new family member.
Anyway, the following data set represents the number of meals each month that
Debbie cranks up on the turbo-charged grill:
7

The range of this sample would be:
Range = 11 – 4 = 7 meals

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

The range is a “quick-and-dirty” way to get a feel for the spread of the data set.
However, the limitation is that it only relies on two data points to describe the variation in the sample. No other values between the highest and lowest points are part of
the range calculation.

DO`WO\QS
One of the most common measurements of dispersion in statistics is the variance,
which summarizes the squared deviation of each data value from the mean. The formula for the sample variance is:
n

¤ xi

i 1

n 1

where:
s2 = the variance of the sample

The variance is a measure of
dispersion that describes the
relative distance between the
data points in the set and the
mean of the data set. This measure is widely used in inferential
statistics.

x = the sample mean
n = the size of the sample (the number
of data values)

x i x = the deviation from the mean
for each value in the data set

The first step in calculating the variance is to determine the mean of the data set,
which in the grilling example is 7.8 meals per month. The rest of the calculations can
be facilitated by the following table.
xi

7
9
8
11
4

7.8
7.8
7.8
7.8
7.8

xi
-0.8
1.2
0.2
3.2
-3.8

0.64
1.44
0.04
10.24
14.44

>O`b( BVS0OaWQa
5

¤ xi

26.80

i 1

The final sample variance calculation becomes this:
26.8
6.7
5 1
For those of us who like to do things in one step, we can also do the entire variance
calculation in the following equation:
s2

7.8

7.8
5 1

7.8

6.7

CaW\UbVS@OeAQ]`S;SbV]REVS\5`WZZW\U
A more efficient way to calculate the variance of a data set is known as the raw score
method. Even though at first glance this equation may look more imposing, its bark
is much worse than its bite. Check it out and decide for yourself what works best for
you.
n

¤ x i2

¥ n ´
¦§ ¤ x i µ¶

i 1

n 1

where:
n

¤ x i2

= the sum of each data value after it has been squared

i 1

¥ n ´ = the square of the sum of all the data values
¦§ ¤ x i µ¶
i 1

Okay, don’t have heart failure just yet. Let me lay this out in the following table to
prove to you there are fewer calculations than with the previous method.
2

7
9
8
11
4

49
81
64
121
16

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\
n

¤ x i2 331

¤ x i 39
i 1

i 1

¥
´
2
¦§ ¤ x i µ¶ 39 1521
i 1

331

s
2

1521
5
4

331 304.2
6.7
4

As you can see, the results are the same
regardless of the method used. The benefits
of the raw score method become more obvious as the size of the sample (n) gets larger.

Bob’s Basics
If you are calculating the variance by hand, my advice is
to do your fingers and calculator battery a favor and use
the raw score method.

BVSDO`WO\QS]TO>]^cZObW]\
So far, we have discussed the variance in the context of samples. The good news is
the variance of a population is calculated in the same manner as the sample variance.
The bad news is I need to introduce another funny-looking Greek symbol: sigma. The
equation for the variance of a population is as follows:
n

¤ xi

i 1

where:

S 2 = the variance of the population
(pronounced “sigma squared”)
xi = the measurement of each item in
the population
R = the population mean
N = the size of the population

Wrong Number
Notice that the denominator for the population variance equation is N, whereas
the denominator for the sample
variance is n – 1.

>O`b( BVS0OaWQa
The raw score version of this equation is:
¥N ´
¦§ ¤ x i µ¶

i 1

¤ x i2

N
N
Even though this procedure is identical to the sample variance, let me demonstrate
with another example. Let’s say I am considering my statistics class as my population
and the following ages are the measurement of interest. (Can you guess which one is
me? My age adds a little spice to the variance.)
2

i 1

I’ll use the raw score method for this calculation with the population size (N) equal to
8. (I’d love to see a class this size.)
2

21
23
28
47
20
19
25
23

441
529
784
2209
400
361
625
529
n

¤ xi2 5878

¤ x i 206
i 1

i 1

¥ n ´
2
¦§ ¤ x i µ¶ 206 42436
i 1

5878

42436
8
8

5878 5304.5
71.7
8
Thanks to the old guy in the class, the population variance is 71.7.

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

AbO\RO`R2SdWObW]\
This method is pretty straightforward. The standard deviation is simply the square root
of the variance. Just as with the variance, there is a standard deviation for both the
sample and population, as shown in the following equations.
Sample standard deviation:
n

s s

¤ xi

i 1

n 1
Population standard deviation:
N

S S2

¤xi

A standard deviation is the
square root of a variance.

i 1

N
To calculate the standard deviation, you must first calculate the variance and then take
the square root of the result. Recall from the previous sections that the variance from
my sample of the number of meals Debbie grilled per month was 6.7. The standard
deviation of this sample is as follows:
s s 2 6.7 2.6 meals
Also recall the variance for the age of my class was 71.7. The standard deviation of the
age of this population is as follows:

S S 2 71.7 8.5 years
The standard deviation is actually a more useful measure than the variance because
the standard deviation is in the units of the original data set. In comparison, the units
of the variance for the grill example would be 6.7 “meals squared,” and the units of
the variance for the age example would be 71.7 “years squared.” I don’t know about
you, but I’m not too thrilled having my age reported as 2,209 squared years. I’ll take
the standard deviation over the variance any day.

1OZQcZObW\UbVSAbO\RO`R2SdWObW]\]T5`]c^SR2ObO
The following equation shows how to calculate the standard deviation of grouped data
in a frequency distribution.

>O`b( BVS0OaWQa

¤ ( xi
s

x )2 f i

i 1

n 1

where:
fi = the number of data values in each frequency class
m = the number of classes
m

¤ f i = the total number of values in the data set
i 1

The following table is a frequency distribution that represents the number of times
each child in Debbie’s 3-year-old preschool class needs a “potty break” in a day.
Number of Potty Breaks per Day ( xi )

Number of Children ( fi )

2
3
4
5
6

1
4
12
8
5

In this example, m = 5 and n = 30. From Chapter 4, we know the mean of this frequency distribution is this:
m

¤ f i xi
x

i 1
m

¤ fi
i 1

1 2 4 3 12 4 8 5 5 6
1 4 12 8 5

4.4 times per child per day

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

The following table summarizes the standard deviation calculations.
xi

2
3
4
5
6

1
4
12
8
5

4.4
4.4
4.4
4.4
4.4

¤ xi

-2.4
-1.4
-0.4
0.6
1.6

5.76
1.96
0.16
0.36
2.56

5.76
7.84
1.92
2.88
12.80

f i 31.20

i 1

¤ ( xi
s

x )2 f i

i 1

n 1

31.20
1.08 1.04 times per child per day
30 1

The potty break frequency distribution has a mean of 4.4 times per child per day and
a standard deviation of 1.04 times per child per day. The frequency of these potty
breaks must keep Debbie very busy.

BVS3[^W`WQOZ@cZS(E]`YW\UbVSAbO\RO`R2SdWObW]\
The values of many large data sets tend to cluster around the mean or median so
that the data distribution in the histogram resembles a bell-shape, symmetrical curve.
When this is the case, the empirical rule (sounds like a decree from the emperor) tells
us that approximately 68 percent of the data
values will be within one standard deviation
from the mean.
According to the empirical rule,
For example, suppose that the average exam
if a distribution follows a bellscore for my large statistics class is 88 points
shape—a symmetrical curve
and the standard deviation is 4.0 points and
centered around the mean—we
would expect approximately
that the distribution of grades is bell-shape
68, 95, and 99.7 percent of
around the mean, as shown in Figure 5.1.
the values to fall within one,
Because one standard deviation above the
two, and three standard deviamean would be 92 (88 + 4) and one standard
tions around the mean respecdeviation below the mean would be 84 (88
tively.
– 4), the empirical rule tells me that approximately 68 percent of the exam scores will fall
between 84 and 92 points.

>O`b( BVS0OaWQa

One standard deviation
around the mean.

Number of Students

4WUc`S#

68%

84
88
92
Exam Scores

100

104

The empirical rule also states that approximately 95 percent of the data values will fall
within two standard deviations from the mean. In our example, two standard deviations equal 8.0 points (2 * 4.0). Two standard deviations above the mean would be a
score of 96 (88 + 8), and two standard deviations below the mean would be 80 (88
– 8). According to Figure 5.2, approximately 95 percent of the exam scores will be
between 80 and 96 points.

Two standard deviations
around the mean.

Number of Students

4WUc`S#

95%

84
88
92
Exam Scores

100

104

Taking this one final step, the empirical rule states that, under these conditions,
approximately 99.7 percent of the data values will fall within three standard deviations
from the mean. According to Figure 5.3, virtually all the test scores should fall within
plus or minus 12 points (3 * 4.0) from the mean of 88. In this case, I would expect all
the exam scores to be between 76 and 100.

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\
4WUc`S#!
Number of Students

Three standard deviations
around the mean.

99.7%

84
88
92
Exam Scores

100

104

In general, we can use the following equation to express the range of values within k
standard deviations around the mean:

M p kS
We will revisit the empirical rule concept in subsequent chapters.

1VSPgaVSdÂaBVS]`S[
Chebyshev’s theorem is a mathematical rule similar to the empirical rule except that
it applies to any distribution rather than just bell-shape, symmetrical distributions.
Chebyshev’s theorem states that for any number k greater than 1, at least
1´
¥
¦§ 1 2 µ¶ s 100 percent of the values will fall within k standard deviations from the
k
mean. Using this equation, we can state the following:
U At least 75 percent of the data values

will fall within two standard deviations
from the mean by setting k = 2 into
Chebyshev’s equation.
U At least 88.9 percent of the data values

will fall within three standard deviations from the mean by setting k = 3
into the equation.

Wrong Number
Chebyshev’s theorem can
be applied to any distribution of data but can only be
stated for values of k that are
greater than 1.

>O`b( BVS0OaWQa
U At least 93.7 percent of the data values will fall within four standard deviations

from the mean by setting k = 4 into the equation. This last example is shown as:
1´
¥
¦§ 1
µ s 100% 93.7%
42 ¶
Let’s check out Chebyshev’s theorem to see whether it really works. The following table shows the number of home runs hit by the top 40 players in Major League
Baseball during the 2002 season.
Number of Home Runs from Top 40 Players in 2002

57
38
33
29

52
38
32
29

49
37
31
29

46
37
31
28

43
35
31
28

42
34
30
28

41
34
30
28

39
33
29
27

Source: www.espn.com.

The following histogram shows that this distribution is neither bell-shape nor symmetrical, so we cannot apply the empirical rule (see Figure 5.4) but will need to use
Chebyshev’s theorem.
18

Home run histogram for
2002 season.

Number of Players

4WUc`S#"

14
12
10
8
6
4
2
0
23-27

28-32

33-37

38-42

43-47

Home Runs

48-52

53-57

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

The mean for this distribution is 34.7 home runs, and the standard deviation is 7.2
home runs. The following table summarizes various intervals around the mean with
the percentage of values within those intervals.
k

M kS

Chebyshev’s
Percentage

2
3
4

34.7
34.7
34.7

7.2
7.2
7.2

49.1
56.3
63.5

20.3
13.1
5.9

75.0%
88.9%
93.7%

Actual
Percentage

95.0%
97.5%
100.0%

This table supports Chebyshev’s theorem, which predicts that at least 75 percent of
the values will fall within two standard deviations from the mean. From the data set,
we can observe that 95 percent actually fall between 20.3 and 49.1 home runs (38 out
of 40). The same explanation holds true for three and four standard deviations around
the mean.

;SOac`Sa]T@SZObWdS>]aWbW]\
Another way of looking at dispersion of data is through measures of relative position, which describe the percentage of the data below a certain point. This technique
includes quartile and interquartile measurements.

?cO`bWZSa
Quartiles divide the data set into four equal
segments after it has been arranged in
ascending order. Approximately 25 percent
of the data points will fall below the first
quartile, Q1. Approximately 50 percent of the
data points will fall below the second quartile,
Q2. And, you guessed it, 75 percent should
fall below the third quartile, Q3. To demonstrate how to identify Q1, Q2, and Q3, let’s
use the following data set.
9

Quartiles measure the relative
position of the data values by
dividing the data set into four
equal segments.

>O`b( BVS0OaWQa
Step 1: Arrange your data in ascending order.
3

Step 2: Find the median of the data set. This is Q2.
3

Q2 = 9
Step 3: Find the median of the lower half of the data set (in parenthesis). This is Q1.
(3

Q1 = 5.5
Q2 = 9
Step 4: Find the median of the upper half of the data set (in parenthesis). This is Q3.
3

(10 12

14)

Q1 = 5.5
Q2 = 9
Q3 = 13

7\bS`_cO`bWZS@O\US
When you have established the quartiles, you can easily calculate the interquartile
range (IQR); the IQR measures the spread of the center half of our data set. It is simply the difference between the third and first quartiles, as follows:
IQR = Q3 – Q1
The interquartile range measures the spread of the center
half of the data set and identifies outliers, which are extreme
values that you should discard
before analysis.

The interquartile range is used to identify outliers,
which are the “black sheep” of our data set. These are
extreme values whose accuracy is questioned and can
cause unwanted distortions in statistical results. Any
values that are more than:
Q3 + 1.5IQR

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

or less than:
Q1 – 1.5IQR
should be discarded. Let’s use the following data set to determine if any nasty outliers
exist:
10

Since there are eight data values, Q1 will be the median of the first four values (the
midpoint between the second and third values).
42 45
43.5
2
Likewise, Q3 will be the median of the last four values (the midpoint between the sixth
and seventh values).
Q1

52 58
56
2

IRQ Q3 Q1 56 43.5 12.5
Any values greater than
Q3 1.5IRQ 56

1.5 12.5

74.75

or less than
Q1 1.5IRQ 43.5

1.5 12.5

24.75

would be considered an outlier. Therefore, the value 10 would be an outlier in this
data set.
Now that we have worked our fingers to the bones calculating all this stuff, let’s see
how Excel makes it look so easy.

CaW\U3fQSZb]1OZQcZObS;SOac`Sa]T2Wa^S`aW]\
Excel enables you to conveniently calculate the range, variance, and standard deviation
of a sample using the Data Analysis selection under the Tools menu. Use the exact
same steps to calculate these measures as those used to calculate measures of central
tendency shown in Chapter 4. Repeating those steps (see the section “Using Excel to
Calculate Central Tendency”) with the grilling example from this chapter will produce
Figure 5.5.

>O`b( BVS0OaWQa

4WUc`S##
Measures of dispersion for the
turbo grill example.

As you can see from Figure 5.5, the sample range equals 7 meals, the sample variance
equals 6.7, and the standard deviation equals 2.6 meals. Also note that this data set has
no mode since no value appears more than once.
This wraps up our discussion on the different ways to describe measures of dispersion.

Wrong Number
The values for variance and standard deviation reported by Excel are for a
sample. If your data set represents a population, you need to recalculate the
results using N in the denominator rather than n – 1.

G]c`Bc`\
1. Calculate the variance, standard deviation, and the range for the following
sample data set: 20, 15, 24, 10, 8, 19, 24.
2. Calculate the variance, standard deviation, and the range for the following
population data set: 84, 82, 90, 77, 75, 77, 82, 86, 82.
3. Calculate the variance, standard deviation, and the range for the following
sample data set: 36, 27, 50, 42, 27, 36, 25, 40.
4. Calculate the quartiles and the cutoffs for the outliers for the following data set:
8, 11, 6, 2, 11, 6, 5, 6, 10, 15.
5. A company counted the number of their employees in each of the age classes as
follows. According to this distribution, what is the standard deviation for the age

1VO^bS`#( 1OZQcZObW\U2SaQ`W^bWdSAbObWabWQa(;SOac`Sa]T2Wa^S`aW]\

of the employees in the company?
Age Range

Number of Employees

20–24

25–29

30–34

35–39

40–44

45–49

6. A company counted the number of employees at each level of years of service in
the table that follows. What is the standard deviation for the number of years of
service in this company?
Years of Service

Number of Employees

7. A data set that follows a bell-shape and symmetrical distribution has a mean
equal to 75 and a standard deviation equal to 10. What range of values centered
around the mean would represent 95 percent of the data points?
8. A data set that is not bell-shape and symmetrical has a mean equal to 50 and a
standard deviation equal to 6. What is the minimum percent of values that would
fall between 38 and 62?

>O`b( BVS0OaWQa

BVS:SOabG]cO`b

>`]POPWZWbgB]^WQa

The connection between descriptive and inferential statistics is based on
probability concepts. I know the topic of probability theory scares the living
daylights out of many students, but it is a very important topic in the world
of statistics. The topic of probability acts as a critical link between descriptive and inferential statistics. Without a firm grasp of probability concepts,
inferential statistics will seem like a foreign language. Because of this, Part
2 is designed to help you over this hurdle.

1VO^bS`

7\b`]RcQbW]\b]>`]POPWZWbg
7\BVWa1VO^bS`

U Distinguish between classical, empirical, and subjective probability
U Use frequency distributions to calculate probability
U Examine the basic properties of probability
U Demonstrate the intersection and union of simple events using a Venn

diagram
As we leave the happy world of descriptive statistics, you may feel like
you’re ready to take on the challenge of inferential statistics. But before
we enter that realm, we need to arm ourselves with probability theory.
Accurately predicting the probability that an event will occur has widespread applications. For instance, the gaming industry uses probability
theory to set odds for lotteries, card games, and sporting events.
The focus of this chapter is to start with the basics of probability, after
which we will gently proceed to more complex concepts in Chapters 7 and
8. We’ll discuss different types of probabilities and how to calculate the
probability of simple events. We’ll rely on data from frequency distributions to examine the likelihood of a combination of simple events. So pull
up a chair and let’s roll those dice!

>O`b ( >`]POPWZWbgB]^WQa

EVOb7a>`]POPWZWbgProbability concepts surround most of our daily lives. When I see that the weather
forecast shows an 80 percent chance of rain tomorrow and I want to play golf or that
my beloved Pittsburgh Pirates have only won 40 percent of their games this year
(which they also did last year and the year before that), there is a 65 percent chance I
will get moody.
In simple terms, probability is the likelihood of a particular event like rain or winning
a ballgame. But before we go any further, we need to tackle some new “stat jargon.”
The following terms are widely used when talking about probability:
U Experiment. The process of measuring or observing an activity for the purpose

of collecting data. An example is rolling a pair of dice.
U Outcome. A particular result of an experiment. An example is rolling a pair of

threes with the dice.
U Sample space. All the possible outcomes of the experiment. The sample space

for our experiment is the numbers {2, 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12}. Statistics
people like to put {} around the sample space values because they think it looks
cool.
U Event. One or more outcomes that are of interest for the experiment and which

is/are a subset of the sample space. An example is rolling a total of 2, 3, 4, or 5
with two dice.
To properly define probability, we need to consider which type of probability we are
referring to.

1ZOaaWQOZ>`]POPWZWbg
Classical probability refers to a situation when we know the number of possible outcomes of the event of interest and can calculate the probability of that event with the
following equation:
P[A] =

Number of possible outcomes in which Event A occurs
Total number of possible outcomes in the sample space

where:
P[A] = the probability that Event A will occur

1VO^bS`$( 7\b`]RcQbW]\b]>`]POPWZWbg

For example, if Event A = rolling a total of 2, 3, 4, or 5 with two dice, we need to
define the sample space for this experiment, which is shown in the following table.
{1,1}

{2,1}

{3,1}

{4,1)

{5,1}

{6,1}

{1,2}

{2,2}

{3,2}

{4,2}

{5,2}

{6,2}

{1,3}

{2,3}

{3,3}

{4,3}

{5,3}

{6,3}

{1,4}

{2,4}

{3,4}

{4,4}

{5,4}

{6,4}

{1,5}

{2,5}

{3,5}

{4,5}

{5,5}

{6,5}

{1,6}

{2,6}

{3,6}

{4,6}

{5,6}

{6,6}

There are 36 total outcomes for this experiment, each with the same chance of occurring. I have underlined the outcomes that
correspond to Event A. There is a total of
10 of them. Therefore:
10
= 0.28
P[A] =
36

Classical probability requires
that you know the number of
outcomes that pertain to a particular event of interest. You also
need to know the total number
of possible outcomes in the
sample space.

To use classical probability, you need to
understand the underlying process so you
can determine the number of outcomes associated with the event. You also need to be able to count the total number of possible
outcomes in the sample space. As you will see next, this may not always be possible.

3[^W`WQOZ>`]POPWZWbg
When we don’t know enough about the underlying process to determine the number
of outcomes associated with an event, we rely on empirical probability. This type of
probability observes the number of occurrences of an event through an experiment
and calculates the probability from a relative frequency distribution. Therefore:
P[A] =

Frequency in which Event A occurs
Total number of observations

>O`b ( >`]POPWZWbgB]^WQa
One example of empirical probability is to answer the
age-old question “What is the probability that John
will get out of bed in the morning for school after his
first wake-up call?” Because I cannot begin to understand the underlying process of why a teenager will
resist getting out of bed before 2 P.M., I need to rely
on empirical probability. The following table indicates the number of wake-up calls John required over
the past 20 school days.

Empirical probability requires
that you count the frequency
that an event occurs through an
experiment and calculate the
probability from the relative frequency distribution.

John’s Wake-Up Calls (Previous 20 School Days)

2
4

4
2

3
3

1
1

2
3

4
2

3
4

3
3

1
4

We can summarize this data with a relative frequency distribution.

@SZObWdS4`S_cS\Qg2Wab`WPcbW]\T]`8]V\ÂaEOYSC^1OZZa
Number of Wake-Up Calls

1
2
3
4

Number of Observations

3
4
8
5
Total = 20

Percentage
3/
20
4/
20
8/
20
5/
20

= 0.15
= 0.20
= 0.40
= 0.25

Based on these observations, if Event A = John getting out of bed on the first wake-up
call, then P[A] = 0.15.
Using the previous table, we can also examine the probability of other events. Let’s
say Event B = John requiring more than 2 wake-up calls to get out of bed; then P[B] =
0.40 + 0.25 = 0.65. That boy needs to go to bed earlier on school nights!

1VO^bS`$( 7\b`]RcQbW]\b]>`]POPWZWbg

Random Thoughts
The probability that you will win a typical state lottery, where you correctly choose
6 out of 50 numbers, is approximately 0.00000006, or 1 out of 16 million. This
is calculated using classical probability. Compare this to the probability that you will
be struck by lightning once during your lifetime, which is 0.0003 or 1 out of 3,000
(source: www.nws.noaa). This is an empirical probability determined by the number
of times people have been struck by lightning in the past. According to these statistics,
you are more than 5,000 times more likely to be struck by lightning than win the lottery! In spite of this, Debbie still makes me go buy a ticket when there’s a big jackpot,
even during a thunder storm.

If I choose to run another 20-day experiment of John’s waking behavior, I would
most likely see different results than those in the previous table. However, if I were to
observe 100 days of this data, the relative frequencies would approach the true or classical probabilities of the underlying process. This pattern is known as the law of large
numbers.
To demonstrate the law of large numbers,
let’s say I flip a coin three times and each
time the result is heads. For this experiment,
the empirical probability for the event heads
is 100 percent. However, if I were to flip the
coin 100 times, I would expect the empirical
probability of this experiment to be much
closer to the classical probability of 50 percent.

The law of large numbers states
that when an experiment is conducted a large number of times,
the empirical probabilities of
the process will converge to the
classical probabilities.

AcPXSQbWdS>`]POPWZWbg
We use subjective probability when classical and empirical probabilities are not available. Under these circumstances, we rely on experience and intuition to estimate the
probabilities.
Examples where we would apply subjective probability are “What is the probability
that my son Brian will ask to borrow my new car, which happens to have a 6-speed
manual transmission, for his junior prom?” (97 percent) or “What is the probability
that my new car will come back with all 6 gears in proper working order?” (18 percent). I based these probabilities on my personal observations after returning from a

>O`b ( >`]POPWZWbgB]^WQa
“practice run” where I heard noises from my poor transmission that chilled me to the
bone and to this day haunt me in my sleep. I need to use subjective probability in this
situation because my car would never survive several of these “experiments.”

0OaWQ>`]^S`bWSa]T>`]POPWZWbg
Our next step is to review the “rules and regulations” that govern probability theory.
The basic ones are as follows:
U If P[A] = 1, then Event A must occur with certainty. An example is Event A =

Debbie buying a pair of shoes this month.
U If P[A] = 0, then Event A will not occur with certainty. An example is Event A =

Bob will eventually finish the basement project that he started three years ago.
U The probability of Event A must be between 0 and 1.
U The sum of all the probabilities for the events in the sample space must be equal

to 1. For example, if the experiment is flipping a coin with Event A = heads and
Event B = tails, then A and B represent the entire sample space. We also know
that P[A] + P[B] = 0.5 + 0.5 = 1.
U The complement to Event A is defined as all the outcomes in the sample space

that are not part of Event A and is denoted as A’. Using this definition, we can
state the following: P[A] + P[A’] = 1 or P[A] = 1 – P[A’].
For example, if the experiment is rolling a single six-sided die, the sample space is
shown in Figure 6.1.

4WUc`S$
Sample space for a single die
experiment.

1VO^bS`$( 7\b`]RcQbW]\b]>`]POPWZWbg

If we say that Event A = rolling a 1, then Event A’ = rolling a 2, 3, 4, 5, or 6.
Therefore:
1
P[A] =
= 0.167
6
P[A’] = 1 – 0.167 = 0.833
Up to this point, all our examples would be considered cases of simple probability,
which is defined as the probability of a single event. Now we’ll expand this concept to
more than one event.

BVS7\bS`aSQbW]\]T3dS\ba
Sometimes we are interested in the probability of a combination of events rather than
just a simple event. To demonstrate this technique, I will use the following example.
Now that my children are older and living away from home, I cherish those moments
when the phone rings and I see one of their numbers appear on my caller ID.
Experience has taught me that I can categorize these calls as either “crisis,” involving
such things as a computer, a car, an ATM card, or a cell phone; or “noncrisis,” when
they call just to see if I’m alive and well enough to help with their next crisis.
The following table, called a contingency table, categorizes the last 50 phone calls by
child and type of call.

1]\bW\US\QgBOPZST]`>V]\S1OZZa
Child

Christin
Brian
John
Total

Crisis

Non-Crisis

Totals

14
10
4
28

6
4
12
22

20
14
16
50

Contingency tables show the actual or relative frequency of two types of data at the
same time. In this case, the data types are child and type of call.

>O`b ( >`]POPWZWbgB]^WQa
I’ll assume that this past pattern of calls will hold true in the near future. We’ll define
Events A and B as follows:
U Event A = the next phone call will come from Christin.
U Event B = the next phone call will involve a crisis.

We can use the contingency table to calculate the simple probability that the next
phone call will come from Christin as follows:
20
P[A] =
= 0.40
50
The probability that the next phone call will involve a crisis would be as follows:
28
= 0.56
P[B] =
50
A contingency table indicates
the number of observations
that are classified according to
two variables. The intersection
of Events A and B represents
the number of instances where
Events A and B occur at the
same time (that is, the same
phone call is both from Christin
and a crisis). The probability of
the intersection of two events is
known as a joint probability.

What about the probability that the next phone call
will come from Christin and will involve a crisis?
This event is known as the intersection of Events A
and B and is described by A B. The number of
phone calls from our contingency table that meet
both criteria is 14, so:
14
= 0.28
P[A and B] = P[A B ] =
50
This explains why I hold my breath as I pick up the
phone!
The probability of the intersection of two events is
known as a joint probability.

BVSC\W]\]T3dS\ba(/;O``WOUS;ORSW\6SOdS\
The union of Events A and B represents all the instances where either Event A or
Event B or both occur and is denoted as A B. Using our previous example, the following table shows the four combinations that include either a call from Christin or a
crisis phone call.

1VO^bS`$( 7\b`]RcQbW]\b]>`]POPWZWbg
Child

Type of Call

Christin
Christin
Brian
John

Crisis
Noncrisis
Crisis
Crisis

Number of Calls

14
6
10
4
Total = 34

Therefore, the probability that the next phone call is either from Christin or is a
crisis is as follows:
34
P[A and B] = P[A B ] =
= 0.68
50

The union of Events A and B represents the number of instances where either Event A
or B occur (that is, the number of calls that were either from Christin or were a crisis).

G]c`Bc`\
1. Define each of the following as classical, empirical, or subjective probability.
a. The probability that the baseball
player Derek Jeter will get a hit
during his next at bat.
b. The probability of drawing an Ace
from a deck of cards.

Bob’s Basics
The probability of the intersection of two events can
never be more than the
probability of the union of two
events. If your calculations don’t
agree with this, go back and
check for a mistake!

c. The probability that I will shoot
lower than a 90 during my next round of golf.
d. The probability of winning the next state lottery drawing.
e. The probability that the drive belt for my riding lawnmower will break this
summer (it did).
f. The probability that I will finish writing this book before my deadline.

>O`b ( >`]POPWZWbgB]^WQa
2. Identify whether each of the following are valid probabilities.
a. 65 percent
b. 1.9
c. 110 percent
d. –4.2
e. 0.75
f. 0
3. A survey of 125 families asked whether the household had Internet access. Each
family was classified by race. The contingency table is shown here.
Race

Caucasian
Asian American
African American
Total

Internet

No Internet

Total

15
23
14
52

22
18
33
73

37
41
47
125

A family from the survey is randomly selected. We define:
Event A: The selected family has an Internet connection in its home.
Event B: The selected family is Asian American.
a. Determine the probability that the selected family has an Internet connection.
b. Determine the probability that the selected family is Asian American.
c. Determine the probability that the selected family has an Internet connection
and is Asian American.
d. Determine the probability that the selected family has an Internet connection
or is Asian American.
4. Using the “crisis” and “noncrisis” phone call example, we define:
Event A: The next phone call will come from Brian.
Event B: The next phone call will be a “noncrisis.”

1VO^bS`$( 7\b`]RcQbW]\b]>`]POPWZWbg

a. Determine the probability that the next phone call will be from Brian and be a
“noncrisis.”
b. Determine the probability that the next phone call will be from Brian or be a
“noncrisis.”

BVS:SOabG]c`]POPWZWbgAbcTT
7\BVWa1VO^bS`
U Calculating conditional probabilities

U The distinction between independent and dependent events
U Using the multiplication rule of probability
U Defining mutually exclusive events
U Using the addition rule of probability
U Using the Bayes’ theorem to calculate conditional probabilities

Now that we have arrived at the second of three basic probability chapters, we’re ready for some new challenges. We need to take the probability
concepts that you’ve mastered from Chapter 6 and put them to work on
the next step up the ladder. Don’t worry if you’re afraid of heights like I
am—just keep looking up!
This chapter deals with the topic of manipulating the probability of different events in various ways. As new information about events becomes
available, we can revise the old information and make it more useful. This
revised information can sometimes lead to surprising results—as you’ll soon
see.

>O`b ( >`]POPWZWbgB]^WQa

1]\RWbW]\OZ>`]POPWZWbg
We define conditional probability as the probability of Event A knowing that Event B
has already occurred. To demonstrate this concept, consider this example.
Debbie is an avid tennis player, and we enjoy playing matches against each other.
We do, however, have one difference of opinion on the court. Debbie likes to have
a nice long warm-up session at the start, where we hit the ball back and forth and
back and forth and back and forth. All during this time, a little voice in my head is
saying, “Who’s winning?” and “What’s the score?” My ideal warm-up is to bend at
the waist to tie my sneakers and to adjust my shorts. Each tennis match becomes a
test of my manhood and the “warm-up” has nothing to do with “the thrill of victory
and the agony of defeat.” I can’t help it; it must be a guy thing that has been passed
down through thousands of years of conditioning. Debbie tells me that when we rush
through the warm-up, she doesn’t play as well. “Poppycock!” I say, and I’ll prove it.
The following table shows the outcomes of our last 20 matches, along with the type of
warm-up before we started keeping score.

1]\bW\US\QgBOPZST]`bVSBS\\Wa3fO[^ZS
Warm-Up Time

Less than 10 min (B)
10 min or more (B’)
Total

Debbie Wins (A)

Bob Wins (A’)

4
5
9

The events of interest are …
U Event A = Debbie wins the tennis match.
U Event B = the warm-up time is less than 10 minutes.
U Event A’ = Bob wins the tennis match.
U Event B’ = the warm-up time is 10 minutes or more.

9
2
11

Total

13
7
20

1VO^bS`%( ;]`S>`]POPWZWbgAbcTT

Without any additional information, the simple probability of each of these events is
as follows:
9
13
P[A]
0.45
P[B]
0.65
20
20
11
7
0.55
P[B’]
0.35
20
20
As if these probabilities don’t have enough names already, I have one more for you.
These are also known as prior probabilities because they are derived only from information that is currently available.
P[A’]

You might wonder, “What other information
is he talking about?” Well, suppose I know
that we had a warm-up period of less than
10 minutes. Knowing this piece of info, what
is the probability that Debbie will win the
match? This is the conditional probability
of Event A given that Event B has occurred.
Looking at the previous table, we can see
that Event B has occurred 13 times. Because
Debbie has won 4 of those matches (A), the
probability of A given B is calculated as follows:

Simple or prior probabilities
are always based on the total
number of observations. In
the previous example, it is 20
matches.

4
0.31
13
Debbie won’t be happy to see that probability.
P[A/B]

We can also calculate the probability that Debbie will win, given that the warm-up
is 10 minutes or longer (otherwise known as an eternity). According to the previous
table, these marathon warm-ups occurred 7 times, with Debbie winning 5 of these
matches. Therefore:
5
0.71
7
This one looks bad for Bob. I might have to hide this chapter from my live-in proofreader.
P[A/B’]

>O`b ( >`]POPWZWbgB]^WQa

Conditional probability is
defined as the probability of
Event A knowing that Event
B has already occurred.
Conditional probabilities are
also known as posterior probabilities.

Once again, I bring to you more “stat jargon.”
Conditional probabilities are also known as posterior
probabilities (I’ll resist using a butt joke here), which
are considered revisions of prior probabilities using
additional information. For example, the prior probability of Debbie winning is P[A] = 0.45. However,
with the additional information that the warm-up
was 10 minutes or longer, we revise the probability of
Debbie winning to P[A/B’] = 0.71.
Conditional probabilities are very useful for determining the probabilities of compound events as you
will see in the following sections.

7\RS^S\RS\bDS`aca2S^S\RS\b3dS\ba
Events A and B are said to be independent of each other if the occurrence of Event B
has no effect on the probability of Event A. Using conditional probability, Events A
and B are independent of one another if:
P[A/B] = P[A]
If Events A and B are not independent of one another, then they are said to be dependent events.
In the tennis example, Events A and B are dependent because the probability of
Debbie winning depends on whether the warm-up is more or less than 10 minutes.
We can also demonstrate this by observing that:
9
4
P[A]
0.45 and P[A/B] 0.31
20
13
These probabilities tell us that overall, Debbie wins 45 percent of the matches.
However, when there is a short warm-up, she only wins 31 percent of the time.
Because these probabilities are not equal, Events A and B are dependent.
An example of 2 independent events is the outcome of rolling two dice:
U Event A: Roll the number 4 on the first of two dice.
U Event B: Roll the number 6 on the second of two dice.

1VO^bS`%( ;]`S>`]POPWZWbgAbcTT

For these events, the simple probabilities are
as follows:
1
1
0.167 and P[B] 0.167
6
6
Even if we know that the first die rolled a 4,
the probability of the second die being a 6 is
not affected because dice, for the most part,
are pretty dim-witted and are not very aware
of what is going on around them. Knowing
this, we can say the following:
P[A]

Events A and B are said to be
independent of each other if
the occurrence of Event B has
no effect on the probability of
Event A. If Events A and B are
not independent of one another,
then they are said to be dependent events.

1
0.167
6
Therefore, Events A and B are independent of one another.
P[B/A] P[B]

;cZbW^ZWQObW]\@cZS]T>`]POPWZWbWSa
We use the multiplication rule of probabilities to calculate the joint probability of two
events. In other words, we are calculating the probability of these events occurring at
the same time. Chapter 6 referred to this as the intersection of two events. For two
independent events, the multiplication rule states the following:
P[A and B] = P[A] × P[B]
Recall from Chapter 6 that P[A and B] is also known as the joint probability of Events
A and B.
For example, we can use the multiplication rule to calculate the joint probability of
rolling “snake eyes” with a pair of dice. We define the events as follows:
U Event A: Roll a 1 on the first die.
U Event B: Roll a 1 on the second die.

Because these events are clearly independent, we can calculate the probability they will
occur simultaneously:
¥ 1´ ¥ 1´ 1
P[A and B] ¦ µ ¦ µ
§ 6 ¶ § 6 ¶ 36

>O`b ( >`]POPWZWbgB]^WQa
If the two events are dependent, things start to heat
up and the multiplication rule becomes:
For dependent events, the multiplication rule states that P[A
and B] = P[A/B] × P[B]. If the
events are independent, the
multiplication rule simplifies to
P[A and B] = P[A] × P[B].

P[A and B] = P[A/B] × P[B]
To demonstrate the multiplication rule with dependent events, let’s go back to the tennis court and
calculate P[A and B], the probability that Debbie will
win and that the warm-up is less than 10 minutes
(from my earlier results):
P[B] = 0.65 and P[A/B] = 0.31

Bob’s Basics
We can rearrange the multiplication rule algebraically
and use it to calculate the
conditional probability of Event
A, given that Event B has
occurred, with the following
equation:
P[A and B]
P[A/B]
P[B]

P[A and B] = (0.65) (0.31)
P[A and B] = 0.20
We can confirm this result by checking the original contingency table, where we see that out of 20
matches, Debbie won 4 times with a warm-up of less
than 10 minutes. Therefore:
4
0.20
20
Maybe Debbie has a valid complaint after all. I wonder whether she ever gets tired of being right!
P[A and B]

;cbcOZZg3fQZcaWdS3dS\ba
Two events are considered to be mutually exclusive if they cannot occur at the same
time during the experiment. For example, suppose my experiment is to roll a single
die and my events of interest are as follows:
U Event A: Roll a 1.
U Event B: Roll a 2.

Two events are considered to
be mutually exclusive if they
cannot occur at the same time
during the experiment.

Because there is no way for both of these events
to occur simultaneously, they are considered to be
mutually exclusive.
Events that can occur at the same time are, you
guessed it, not mutually exclusive. In our tennis

1VO^bS`%( ;]`S>`]POPWZWbgAbcTT
example, Events A and B are not mutually exclusive because (a) Debbie can win the
match and (b) the warm-up can be less than 10 minutes in the same experiment.

/RRWbW]\@cZS]T>`]POPWZWbWSa
We use the addition rule of probabilities to calculate the probability of the union of
events—that is, the probability that either Event A or Event B will occur. For two
events that are mutually exclusive, the addition rule states the following:
P[A or B] = P[A] + P[B]
As an example, for the single-die experiment
with mutually exclusive events:
U Event A: Roll a 1.
U Event B: Roll a 2.

For mutually exclusive events,
the addition rule states that
P[A or B] = P[A] + P[B]. If the
events are not mutually exclusive, the addition rule becomes
P[A or B] = P[A] + P[B] – P[A
and B].

The simple probabilities are as follows:
1
1
P[A] 0.167 and P[B] 0.167
6
6
The probability that either a 1 or a 2 will be rolled is as follows:
P[A or B] = P[A] + P[B]
P[A or B] = 0.167 + 0.167
P[A or B] = 0.334
For events that are not mutually exclusive, the addition rule states the following:
P[A or B] = P[A} + P[B] – P[A and B]
Going back to the tennis court, where …
U Event A = Debbie wins the tennis match.
U Event B = The warm-up time is less than 10 minutes.

Recall that:
P[A] = 0.45 and P[B] = 0.65
P[A and B] = 0.20

>O`b ( >`]POPWZWbgB]^WQa
Therefore, the probability that Debbie will either win the match or the warm-up will
be less than 10 minutes is as follows:
P[A or B] = P[A] + P[B] – P[A and B]
P[A or B] = 0.45 + 0.65 – 0.20
P[A or B] = 0.90
The logic behind subtracting P[A and B] in the addition rule is to avoid double counting. We can demonstrate this in the following table, which converts the frequency
distribution to a relative frequency distribution.

@SZObWdS4`S_cS\Qg2Wab`WPcbW]\T]`BS\\Wa;ObQVSa
Warm-Up Time

Debbie Wins

Less than 10
10 or more
Total

4/
20
5/
20
9/
20

= 0.20
= 0.25
= 0.45

Bob Wins
9/
20
2/
20
11/
20

Total

= 0.45
= 0.10
= 0.55

13/
20
7/
20
20/
20

= 0.65
= 0.35
= 1.00

The union of Events A and B can be displayed using Figure 7.1.

4WUc`S%

Warm-Up Time

Deb Wins

Bob Wins

Totals

The union of Events A and
B.

Less than 10 Min

0.20

0.45

0.65

10 Min or More

0.25

0.10

0.35

Totals

0.45

0.55

1.00

The Union of Events A and B

Bob’s Basics
When converting frequencies
to relative frequencies in a
contingency table, always
divide each number in the table
by the total number of observations. In the previous example,
that is 20 matches.

The probability of Debbie winning the match (Event
A) is represented by the box in the first column. The
probability of having a warm-up of less than 10 minutes (Event B) is represented by the box in the first
row. If we add P[A] + P[B], which would be the column plus the row in Figure 7.3, we are double counting P[A and B] = 0.20 and therefore need to subtract
this in the addition rule for events that are not mutually exclusive.

1VO^bS`%( ;]`S>`]POPWZWbgAbcTT

Ac[[O`WhW\U=c`4W\RW\Ua
Before moving on, let’s step back and take a look at what we’ve done so far. Figure 7.2
shows the simple, joint, and conditional probabilities in the relative frequency distribution for our tennis matches.

4WUc`S%

Summary of Probabilities for Tennis Example
Warm-Up Time

Less than 10 Min (B)
10 Min or More (B')

Totals

Deb Wins
(A)

Bob Wins
(A')

Totals

0.20

0.45

0.65

0.25

0.10

0.35

0.45

0.55

1.00

Joint Probabilities

Simple Probabilities

p[A and B] = 0.20
p[A' and B] = 0.45
p[A and B'] = 0.25
p[A' and B'] = 0.10

p[A ] = 0.45
p[A'] = 0.55
p[B] = 0.65
p[B'] = 0.35

Summary of probabilities for
the tennis example.

Conditional Probabilities
P[A/B] =

P[A and B] 0.20
=
= 0.31
P[B]
0.65

P[A'/B] =

P[A' and B] 0.45
=
= 0.69
P[B]
0.65

P[A/B'] =

P[A and B'] 0.25
=
= 0.71
P[B']
0.35

P[A'/B']=

P[A' and B'] 0.10
=
= 0.29
P[B']
0.35

Note that:
U Event A’ = Bob wins the match.
U Event B’ = The warm-up is 10 minutes or more.

These conditional probabilities have revealed my secret to success on the court. The
probability of my winning after a short warm-up, P[A’/B], is 0.69; whereas the probability of my winning after a longer warm-up, P[A’/B’], is 0.29. I knew I should have
picked another example for this chapter.

>O`b ( >`]POPWZWbgB]^WQa

0OgSaÂBVS]`S[
Thomas Bayes (1701–1761) developed a mathematical rule that deals with calculating
P[B/A] from information about P[A/B]. Bayes’ theorem states the following:
P[B/A]

P[B]P[A/B]
(P[B]P[A/B])+(P[B’]P[A/B’])

where:
P[B’] = the probability of the complement of Event B
P[A/B’] = the probability of Event A, given that the complement to Event B has
occurred
Now that looks like a mouthful, but applying it in our tennis example will clear things
up. With Bayes’ theorem, we can calculate P[B/A], which is the probability that the
warm-up was less than 10 minutes, given that Debbie won the match. Using the values from the previous figure:
P[B/A]

0.65 0.31
0.65 0.31 0.35 0.71

P[B/A]

0.20
0.44
0.20 0.25

Random Thoughts
Not only was Thomas Bayes
a prominent mathematician,
but he was also a published
Presbyterian minister who used
mathematics to study religion.

Knowing that Debbie won the match, we can say
there is a 44 percent chance that the warm-up was
less than 10 minutes.
We can confirm this result by looking at the original
contingency table. Because Debbie won 9 matches
and from those, 4 had a warm-up of less than 10
minutes:
4
0.44
9
Ta da! Please hold your applause until the end of the
book.
P[B/A]

1VO^bS`%( ;]`S>`]POPWZWbgAbcTT

G]c`Bc`\
A political telephone survey of 260 people asked whether they were in favor or not in
favor of a proposed law. Each person was identified as Republican or Democrat. The
following contingency table shows the results.
Party

Republican
Democrat
Total

In Favor

Not in Favor

Total

98
79
177

54
29
83

152
108
260

A person from the survey is selected at random. We define:
U Event A: The person selected is in favor of the new law.
U Event B: The person selected is a Republican.

1. Determine the probability that the selected person is in favor of the new law.
2. Determine the probability that the selected person is a Republican.
3. Determine the probability that the selected person is not in favor of the new law.
4. Determine the probability that the selected person is a Democrat.
5. Determine the probability that the selected person is in favor of the new law
given that the person is a Republican.
6. Determine the probability that the selected person is not in favor of the new law
given that the person is a Republican.
7. Determine the probability that the selected person is in favor of the new law
given that the person is a Democrat.
8. Determine the probability that the selected person is in favor of the new law and
that the person is a Republican.
9. Determine the probability that the selected person is in favor of the new law and
that the person is a Democrat.
10. Determine the probability that the selected person is in favor of the new law or
that the person is a Republican.

>O`b ( >`]POPWZWbgB]^WQa
11. Determine the probability that the selected person is in favor of the new law or
that the person is a Democrat.
12. Using Bayes’ theorem, calculate the probability that the selected person was a
Republican, given that the person was in favor of the new law.

BVS:SOabG]c`W\QW^ZSaO\R
>`]POPWZWbg2Wab`WPcbW]\a
7\BVWa1VO^bS`
U Using the fundamental counting principle
U Distinguishing between permutations and combinations
U Defining a random variable and probability distribution

U Calculating the mean and variance of a discrete probability distribu-

tion
Well, we’ve finally arrived at our third and last chapter on general probability concepts. This chapter sets the stage for the last three chapters in Part
2, which will focus on specific types of probability distributions. Before you
know it, we’ll be knee deep with inferential statistics.
This chapter will also teach you how to count. This type of counting, however, goes far beyond what you’ve seen on Sesame Street. Counting events is
an important step in calculating probabilities and must be done with care.

>O`b ( >`]POPWZWbgB]^WQa

1]c\bW\U>`W\QW^ZSa
To use classical probability, which we introduced way back in Chapter 6, we need
to be able to count the number of events of interest along with the total number of
events that are possible in the sample space. For simple events, like rolling a single die,
the number of possible outcomes (six) is obvious. But for more complex events, like a
state lottery drawing, we need to rely on techniques known as counting principles to
arrive at the correct answer, so let’s look at these techniques.

BVS4c\RO[S\bOZ1]c\bW\U>`W\QW^ZS
After a tough round of golf on a hot afternoon, Brian, John, and I decide to revive our
spirits at the ice cream store on the way home. There I’m overwhelmed with deciding
between four flavors and three toppings to indulge in.
How many different combinations of ice cream and
toppings am I faced with? The fundamental counting
principle comes to my rescue by telling me that if one
According to the fundamental
event (my ice cream choice) can occur in m ways and
counting principle, if one event
a second event (my topping choice) can occur in n
can occur in m ways and a
ways, the total number of ways both events can occur
second event can occur in n
ways, the total number of ways
together is m n ways. In my case, I have m n comboth events can occur together
binations of flavors and toppings in which to blow my
is m • n ways. And we can
diet. (I’ll leave that topic for another chapter.)
extend this principle to more
Now I can extend this principle to more than two
than two events.
events. In addition to flavors and toppings, I have
another tempting choice between a small and large
serving. That leaves me with the mind-boggling decision of 4 3 2 24 combinations,
which are summarized in the table that follows my list of options.
Ice Cream Flavors

Toppings

Size

CH = Death by Chocolate
VA = Vanilla
ST = Strawberry
CF = Coffee

HF = Hot Fudge
BS = Butterscotch
SP = Sprinkles

LG = Large
SM = Small

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

List of Combinations (Flavor-Topping-Size)

CH-HF-LG
CH-HF-SM
CH-BS-LG
CH-BS-SM
CH-SP-LG
CH-SP-SM

VA-HF-LG
VA-HF-SM
VA-BS-LG
VA-BS-SM
VA-SP-LG
VA-SP-SM

ST-HF-LG
ST-HF-SM
ST-BS-LG
ST-BS-SM
ST-SP-LG
ST-SP-SM

CF-HF-LG
CF-HF-SM
CF-BS-LG
CF-BS-SM
CF-SP-LG
CF-SP-SM

Can you guess which choice a certain chocolate-loving author made?
Another demonstration of the fundamental counting principle is to calculate the number of unique combinations for a state’s automobile license plates. Suppose the state
plates have three letters followed by four numbers. The number zero and the letter O
are not eligible because their resemblance may cause confusion. Because we have 25
possible letters and 9 possible numbers, the total number of unique combinations is as
follows:
First
Letter

Second
Letter

Third
Letter

First
Number

Second
Number

Third
Number

Fourth
Number

25 × 25 × 25 × 9 × 9 × 9 × 9 = 102,515,625!
That’s 102,515,625 possible combinations!

>S`[cbObW]\a
Permutations are the number of different ways in which objects can be arranged in
order. In a permutation, each item appears only once. The number of permutations of
n distinct objects is n! (expressed as n factorial) and is defined as follows:
n ! n ( n 1) ( n 2) ( n 3) 4 3 2 1

>O`b ( >`]POPWZWbgB]^WQa
By definition, 0! = 1. For instance, 6 ! 6 5 4 3 2 1 720 . As an example, there are
six permutations for the numbers 1, 2, and 3, as shown here:
123 132 213 231 312 321
Because: 3! 3 2 1 6

Permutations are the number of
different ways in which objects
can be arranged in order. The
number of permutations of n
objects taken r at a time can
n!
.
be found by n Pr
(n r )!

Before the beginning of a professional basketball
game, the starting 5 players are announced one at a
time. How many different ways can we arrange the
order that the players are announced? The number
of permutations is:
5! 5 4 3 2 1 120 .
Suppose we want to select only some of the objects in
the group. The number of permutations of n objects
taken r at a time can be found as follows:
n!
n Pr
( n r )!

Bob’s Basics
It’s easier to calculate the number of permutations using this formula:
n!
n Pr
n (n 1) (n 2) (n r 1) .
(n r )!
This works because every value in the denominator (the bottom of the fraction) will
cancel out with many values in the numerator (the top of the fraction).

Using our basketball example again, if there are 12 players on the team, how many
different ways can any five players on the team be announced to start the game? In
this case, because n = 12 and r = 5, the number of permutations is as follows:
12

12!
12 11 10 9 8 7 6 5 4 3 2 1

(12 5)!
7 6 5 4 3 21

12!
12 11 10 9 8 95, 040
(12 5)!

I’m sure glad it’s not my job to decide who gets announced first.

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

Sometimes the order of events is not of consequence, and we’ll discuss those cases in
the next section.

1][PW\ObW]\a
Combinations are similar to permutations,
except that the order of the objects is not
important. The number of combinations of
n objects taken r at a time can be found as
follows:
n!
nC r
( n r )! r !
For example, in poker, five cards are selected
randomly from a deck of 52 cards. How
many five-card combinations exist?
52!
52 51 50 49 48

2, 598, 960
52C 5
(52 5)!5 !
5 4 3 21

Combinations are the number
of different ways in which
objects can be arranged without regard to order. The number
of combinations of n objects
taken r at a time can be found
n!
by nC r
.
(n r )!r !

How many five-card permutations exist?
52!
52 51 50 49 48 311, 875, 200
52P 5
(52 5)!

Bob’s Basics
It’s easier to calculate the number of combinations using the same logic as the permutation formula and this formula:
Cr

n!
n (n 1) (n 2) (n

r )!r !
r!

There are more five-card permutations because the following two poker hands would
be considered two different permutations but be counted as only one combination
because they are the same cards only in different order.

>O`b ( >`]POPWZWbgB]^WQa
Hand 1

Hand 2

Ace of Spades
Queen of Hearts
Ten of Spades
Ten of Diamonds
Three of Clubs

Ace of Spades
Ten of Spades
Queen of Hearts
Ten of Diamonds
Three of Clubs

Now that we know the total number of five-card combinations from a 52-card deck,
we can calculate the probability of a flush, which is any five cards that are all the same
suit (spades, clubs, hearts, or diamonds). For you poker veterans, I am including a
royal flush and a straight flush in this calculation. First, we need to count the number
of five-card flushes of one suit, let’s say diamonds. Because there are 13 diamonds in
the deck, the number of combinations of these 13 diamonds, taken five at a time, is as
follows:
13!
13 12 11 10 9

1, 287
13C 5
(13 5)! 5!
5 4 3 21
Because there are four suits in the deck, the total number of five-card flushes from
any suit is 1287 4 5, 148. Therefore, the probability of being dealt a flush, including
royal and straight, in a five-card hand is:
5, 148
P[Flush]
0.002
2, 598, 960
or roughly twice in 1,000 hands of poker. Ready to deal?

Bob’s Basics
An alternate notation for nC r
¥ n´
is ¦ µ , which you may
§r¶
come across in other textbooks.
Statisticians just love to have
different notations for the same
concept!

What about the probability of being dealt a hand
with two pairs of any suit? There are 13C 2 78 different two-pair combinations in the deck. Each pair
can have 4C 2 6 different combinations of the four
suits. There are 52 – 6 = 44 possible cards left for the
fifth card in the hand. The number of two-pair hands
would then be:
78 6 6 44 123, 552
Therefore, the probability of being dealt two pair is
123, 552
P[Two-pair]
0.0475
2, 598, 960

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

Combinations are also useful for calculating the probability of winning a state lottery
drawing. A typical lottery game requires you to pick six numbers out of a possible 49.
Because the order of the numbers does not matter, we use the combination rather
than the permutation formula. The number of six-number combinations from a pool
of 49 numbers is this:
49 !
49 48 47 46 45 44
49C 6
13, 983, 816

( 49 6 )! 6 !
6 5 4 3 21
Because there are nearly 14 million different six-number combinations, the probability
that your combination is the winner is as follows:
1
P[Winning a 6/49 Lottery]
0.00000007
13, 983, 816

Wrong Number
Probability does not have a memory. The same six numbers selected in last
week’s lottery drawing have the exact same probability of being chosen again in
this week’s lottery. That’s because the two drawings are independent events and have
absolutely no influence on each other. Therefore, choosing a lottery number because
it has not been selected recently does not increase your odds of winning. Sorry if I
ruined your favorite strategy!

With those chances of winning the lottery, you better not quit your day job just yet.

CaW\U3fQSZb]1OZQcZObS>S`[cbObW]\aO\R1][PW\ObW]\a
Here’s something that’s pretty cool—rather than deal with all those nasty factorial calculations, we can let Excel figure out the number of permutations or combinations for
us. The functions are:
=PERMUT(n, r)
=COMBIN(n, r)
For example, if we type =PERMUT(12,5) into Excel, the result will be 95,040. Be
sure to try this and give your poor calculator a rest!

>O`b ( >`]POPWZWbgB]^WQa
Random Thoughts
The odds of winning the state lottery drawing are so astronomically low, it’s hard to
really fathom them. Using the 6/49 lottery example, if I bought one ticket every day
of the year, I can expect to win once every 38,312 years. To give this some perspective, 38,000 years ago, people were living in caves during the Stone Age. I’m not
sure I want to wait that long, no matter how much money I win.

As we wrap up the topic of counting principles, many of you may be surprised at how
complicated it can be to count events. But this is an important concept in statistics
that we will revisit in Chapter 9.

>`]POPWZWbg2Wab`WPcbW]\a
Now let’s introduce you to probability distributions and prepare you for the last three
chapters of Part 2. However, first we need to discuss the topic of random variables,
which will lay the groundwork for specific probability distributions in Chapters 9, 10,
and 11.

@O\R][DO`WOPZSa
In Chapter 6, we talked about conducting experiments to acquire data. Examples of
experiments could be rolling dice or counting the number of times next month that
I can’t find something in the house and need to ask Debbie to help me. “She who
knows where all things are” has this mystical ability to make these items suddenly
appear before my very eyes after I have given up looking. Debbie then proceeds to
give me a pitiful look that says, “You would never survive a single day in this world
without me,” which sadly I’d have to agree with.

A random variable is an outcome that takes on a numerical
value as a result of an experiment. The value of the random
variable, which is not known
with certainty before the experiment, is often denoted by x.

The outcomes of these experiments are considered
random variables. By definition, these outcomes are
not known before the experiment. For example, I
can’t predict with certainty the number of times next
month I’ll need Debbie’s help finding something.
Once the outcome has occurred, I can determine the
value of the random variable. For instance, if I ask
Debbie to help me four times next month, the value
of that random variable is four.

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

All random variables are not created equal. The first type are known as continuous random variables, which are the result of a measurement on a continuous number scale.
For example, each morning when I take a deep breath and step on the bathroom scale
to weigh myself (taking a deep breath and holding it somehow makes me feel lighter),
I’m looking down in shock and disbelief at a continuous random variable. (Maybe I
should have chosen the small Death by Chocolate serving.) Examples of values for
continuous random variables of this sort could be 180, 181.5, 183.2, and so on. (I’ll
stop there.) Because this is a continuous variable, my morning weight could take on an
unlimited number of possible values, which is a very disconcerting thought.
The second type of random variable is discrete. Discrete random variables are the
result of counting outcomes rather than measuring them. Discrete random variables
can only take on a certain number of integer
values within an interval. An example of a
discrete random variable would be my golf
score for my next round because this value is
A random variable is continuarrived at by counting my total strokes over
ous if it can assume any numeri18 holes of play. Obviously, this value needs
cal value within an interval
to be an integer, such as 94, because there
as a result of measuring the
is no way to count a partial stroke (even
outcome of an experiment. A
random variable is discrete if it
though there are times my golf swing feels
is limited to assuming only spelike one).
cific integer values as a result
We will discuss continuous random variables
of counting the outcome of an
in more detail in Chapter 11. But here and
experiment.
in Chapters 9 and 10 we will focus solely on
discrete random variables.

2WaQ`SbS>`]POPWZWbg2Wab`WPcbW]\a
A listing of all the possible outcomes of an experiment for a discrete random variable
along with the relative frequency or probability of each outcome is called a discrete
probability distribution. To illustrate this concept, I’ll use this example.
My oldest daughter, Christin, was a very accomplished competitive swimmer between
the ages of 7 and 13, but her talent certainly didn’t come from my side of the family.
One day, I mustered the courage to ask Christin to teach me how to swim the butterfly stroke. My form was best described as “a beached whale having seizures.” The
lifeguards banned me from ever attempting this stroke again, claiming it too closely
resembled a person who was drowning. Somehow, in spite of this gene pool (Get it?
Swimming pool, gene pool?), Christin could not only swim, but she also could swim
fast.

>O`b ( >`]POPWZWbgB]^WQa
The following table is a relative frequency distribution showing the number of first-,
second-, third-, fourth-, and fifth-place finishes Christin earned during 50 races.
Place

Number of Races

1
2
3
4
5

27
12
7
3
1
Total = 50

Relative Frequency (Probability)
27/
50
12/
50
7/
50
3/
50
1/
50

= 0.54
= 0.24
= 0.14
= 0.06
= 0.02
Total = 1.00

If we define the random variable x = the place Christin finished in a race, the previous
table would be the discrete probability distribution for the variable x. From this table,
we can state the probability that Christin will finish first as follows:
P[x = 1] = 0.54
Or we can state the probability that Christin will finish either first or second as follows:
P[x = 1 or x = 2] = 0.54 + 0.24 = 0.78
Figure 8.1 shows the discrete probability distribution for x graphically.

0.6

The discrete probability distribution for Christin’s races.

0.5
Probability

4WUc`S&

0.4
0.3
0.2
0.1
0
1

3
Place

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

@cZSaT]`2WaQ`SbS>`]POPWZWbg2Wab`WPcbW]\a
Any discrete probability distribution needs to meet the following requirements:
U Each outcome in the distribution needs to be mutually exclusive—that is, the

value of the random variable cannot fall into more than one of the frequency
distribution classes. For example, it is not possible for Christin to take first and
second place in the same race.
U The probability of each outcome, P[x], must be between 0 and 1; that is,

0 b P[x] b 1 for all values of x. In the previous example, P[x 3] 0.14 , which
falls between 0 and 1.
U The sum of the probabilities for all the outcomes in the distribution needs to add
n

up to 1; that is,

¤ P[x ] 1. In the swimming example, the sum of the Relative
i

i 1

Frequency (Probability) column in the previous table adds up to 1.

BVS;SO\]TO2WaQ`SbS>`]POPWZWbg2Wab`WPcbW]\
The mean of a discrete probability distribution is simply a weighted average (discussed
in Chapter 4) calculated using the following formula:
n

M ¤ x i P[ x i ]
i 1

where:

M = the mean of the discrete probability distribution
x i = the value of the random variable for the ith outcome
P[x i ] = the probability that the ith outcome will occur
n = the number of outcomes in the distribution
The table that follows revisits Christin’s swimming probability distribution.
Place x i

1
2
3
4
5

Probability P[x i ]

0.54
0.24
0.14
0.06
0.02

>O`b ( >`]POPWZWbgB]^WQa
The mean of this discrete probability distribution is as follows:
n

M ¤ x i P[ x i ]
i 1

M 1 0.54

2 0.24 3 0.14 4 0.06 5 0.02

M 1.78
This mean is telling us that Christin’s average finish for a race is 1.78 place! How does
she do that? Obviously, this will never be the result of any one particular race. Rather,
it represents the average finish of many races. The mean of a discrete probability distribution does not have to equal one of the values of the random variable (1, 2, 3, 4, or
5 in this case).
Another term for describing the mean of a probability distribution is the expected value, E[x]. Therefore:
n

An expected value is the mean
of a probability distribution.

E[ x ] M ¤ x i P[ x i ]
i 1

Didn’t I say statisticians love all sorts of notation to
describe the same concept?

BVSDO`WO\QSO\RAbO\RO`R2SdWObW]\]TO2WaQ`SbS>`]POPWZWbg2Wab`WPcbW]\
Just when you thought it was safe to get back into the water, along comes another
variance! Well, if you’ve seen one variance calculation, you’ve seen them all. You can
calculate the variance for a discrete probability distribution as follows:
n

S 2 ¤ ( xi

M )2P[ x i ]

i 1

where:

S 2 = the variance of the discrete probability distribution
As before, the standard deviation of the distribution is as follows:

S S2
To demonstrate the use of these equations, we’ll rely on Christin’s swimming distribution. The calculations are summarized in the following table.

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a
xi

P[x i ]

1
2
3
4
5

0.54
0.24
0.14
0.06
0.02

1.78
1.78
1.78
1.78
1.78

( xi

-0.78
0.22
1.22
2.22
3.22

( xi

M )2

0.608
0.048
1.488
4.928
10.368

M )2 P[x i ]
0.328
0.012
0.208
0.296
0.208

S 2 ¤ ( xi

M )2P[x i ] 1.052

i 1

The standard deviation of this distribution is:

S S 2 1.052 1.026
A more efficient way to calculate the variance of a discrete probability distribution is:
´
¥ n
S 2 ¦ ¤ x i 2P[ x i ]µ M 2
¶
§ i 1
The following table summarizes these calculations using Christin’s swimming example.
xi

P[x i ]

xi 2

x i 2 P[ x i ]

1
2
3
4
5

0.54
0.24
0.14
0.06
0.02

1
4
9
16
25

0.54
0.96
1.26
0.96
0.50

¤x

2
i

P[ x i ] 4.22

i 1

´
¥ n
S 2 ¦ ¤ x i 2P[ x i ]µ
¶
§ i 1

S 2 4.22 (1.78)2 1.052
As you can see, the result is the same, but with less effort!

>O`b ( >`]POPWZWbgB]^WQa

G]c`Bc`\
1. A restaurant has a menu with three appetizers, eight entrées, four desserts, and
three drinks. How many different meals can you order?
2. A multiple-choice test has 10 questions, with each question having four choices.
What is the probability that a student, who randomly answers each question, will
answer each question correctly?
3. The NBA teams with the 13 worst records at the end of the season participate in
a lottery to determine the order in which they will draft new players for the next
season. How many different arrangements exist for the drafting order for these
13 teams?
4. In a race with eight swimmers, how many ways can swimmers finish first, second,
and third?
5. How many different ways can 10 new movies be ranked first and second by a
movie critic?
6. A combination lock has a total of 40 numbers and will unlock with the proper
three-number sequence. How many possible combinations exist?
7. I would like to select three paperback books from a list of 12 books to take on
vacation. How many different sets of three books can I choose?
8. A panel of 12 jurors needs to be selected from a group of 50 people. How many
different juries can be selected?
9. A survey of 450 families was conducted to find how many cats were owned by
each respondent. The following table summarizes the results.
Number of Cats

Number of Families

137

160

112

Develop a probability distribution for this data and calculate the mean, variance,
and standard deviation.
10. What is the probability of being dealt a full house (three-of-a-kind and a pair) in
five-card poker?

1VO^bS`&( 1]c\bW\U>`W\QW^ZSaO\R>`]POPWZWbg2Wab`WPcbW]\a

BVS:SOabG]c`]POPWZWbg
2Wab`WPcbW]\
7\BVWa1VO^bS`
U Describe the characteristics of a binomial experiment
U Calculate the probabilities for a binomial distribution
U Find probabilities using a binomial table
U Find binomial probabilities using Excel

U Calculate the mean and standard deviation of a binomial distribution

Our discussion of discrete probability distributions so far has been limited
to general distributions based on historical data that has been previously
collected. However, some theoretical probability distributions are based on
a mathematical formula rather than historical data. We will address the first
of these, the binomial probability distribution, in this chapter.
In many types of problems we are interested in the probability of an event
occurring several times. A classical example that has been torturing students
for many years is “What is the chance of getting 7 heads when tossing a
coin 10 times?” By the time you finish this chapter, answering this question
will be a piece of cake!

>O`b ( >`]POPWZWbgB]^WQa

1VO`OQbS`WabWQa]TO0W\][WOZ3f^S`W[S\b
If you remember, in Chapter 6 we defined experimenting as the process of measuring
or observing an activity for the purpose of collecting data. Let’s say our experiment
of interest involves a certain professional basketball player shooting free throws. Each
free throw would be considered a trial for the experiment. For this particular experiment, we have only two possible outcomes for each trial; either the free throw goes in
the basket (a success) or it doesn’t (a failure). Because we can have only two possible
outcomes for each trial, this is known as a binomial experiment.

A binomial experiment has the
following characteristics: (1) the
experiment consists of a fixed
number of trials denoted by n;
(2) each trial has only two possible outcomes, a success or
a failure; (3) the probability of
success and the probability of
failure are constant throughout
the experiment; (4) each trial is
independent of any other trial in
the experiment.

Random Thoughts
Binomial experiments are also
known as Bernoulli process,
named after Swiss mathematician James Bernoulli, who lived
during the 1600s. Repeating a
Bernoulli process several times
is referred to as Bernoulli trials,
a concept that has been haunting students for hundreds of
years!

Let’s say that our player of interest is Michael
Jordan, who historically has made 80 percent of his
free throws. So the probability of success, p, of any
given free throw is 0.80. Because there are only two
outcomes possible, the probability of failure for any
given free throw, q, is 0.20. For a binomial experiment, the values of p and q must be the same for
every trial in the experiment. Because only two outcomes are allowed in a binomial experiment, p 1 q
always holds true.
Finally, a binomial experiment requires that each trial
is independent of any other trial. In other words, the
probability of the second free throw being successful is not affected by whether the first free throw was
successful. Other examples of binomial experiments
include the following:
U Testing whether a part is defective after it has

been manufactured
U Observing the number of correct responses in a

multiple-choice exam
U Counting the number of American households

that have an Internet connection
Now that we have defined the ground rules for binomial experiments, we are ready to graduate to calculating binomial probabilities.

1VO^bS`'( BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\

BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\
The binomial probability distribution allows us to calculate the probability of a specific number of successes for a certain number of trials. Therefore, the random variable for this distribution would be the number of successes that were observed. To
demonstrate a binomial distribution, I will use the following example.
Debbie has trained our dog, Kaylee, to do an incredible trick. First thing every morning after she lets the dog out the back door, Kaylee runs like greased lightning around
the house, down our rather long driveway, grabs our newspaper, and races to the back
door, where she dutifully deposits it on the step. In return for this vital chore for our
household, she gets two cups of dry dog food in a plastic bowl. Amazing, you say. But
you’ve only heard half of it. Somehow, in the tiny recesses of Kaylee’s doggy brain,
she has worked out the remarkable deduction that “two breakfasts are better than
one” and at every opportunity goes on a neighborhood hunt for more newspapers to
deposit on our back step. Once she dragged an entire phone book back, thinking maybe
this would earn her a bonus. We have failed miserably trying to train Kaylee to return
these papers—apparently tiny doggy brains don’t work in reverse.
So my job on many afternoons is to careBob’s Basics
fully return the stolen merchandise, hoping
my neighbors fail to notice the dog slobber
Remember from Chapter 8
on their three-day-old paper. Anyway, let’s
n!
nC r , which
that
say on any particular day there is a 30 per(n r )!r !
represents the number of comcent probability that Kaylee will bring back
binations of n objects taken r at
one stolen paper and a 70 percent chance
a time.
that she won’t. We will assume that she will
not bring back more than one paper a day.
This scenario represents a binominal experiment, with each day being a Bernoulli trial with p 0.30 (the probability of a “success”) and q 0.70 (the probability of a “failure”). We can calculate the probability of
r successes in n trials using the binomial distribution, as follows:
P[r , n ]

n!
pr qn
( n r )! r !

With this equation, we can calculate the probability that Kaylee will bring back three
papers over the next five days.
P[3, 5]

5!
0.3 3 0.7
(5 3)! 3!

5 3

>O`b ( >`]POPWZWbgB]^WQa
¥ 120 ´
P[3, 5] ¦
0.027 0.49 0.1323
§ 2 6 ¶µ

Bob’s Basics
Remember from Chapter 8,
0! = 1. Also x0 = 1 for any
value of x.

There is a 13 percent chance that the neighborhood
paper bandit will strike 3 times during the next 5
days. We can also calculate the probability that she
will round up zero, one, two, four, or five papers over
the next five days.

For r = 0:
P[0, 5]

5!
0.3 0 0.7
(5 0 )! 0 !

5 0

¥ 120 ´
P[0, 5] ¦
1 0.1681 0.1681
§ 120 1µ¶
For r = 1:
5!
P[1, 5]
0.3 1 0.7 5 1
(5 1)!1!
¥ 120 ´
P[1, 5] ¦
0.3 0.2401 0.3601
§ 24 1µ¶
For r = 2:
P[2, 5]

5!
0.3 2 0.7
(5 2)! 2!

5 2

¥ 120 ´
P[2, 5] ¦
0.09 0.343 0.3087
§ 6 2µ¶
For r = 4:
P[4, 5]

5!
0.3 4 0.7
(5 4 )! 4 !

5 4

¥ 120 ´
P[4, 5] ¦
0.0081 0.7 0.0283
§ 1 24 µ¶
For r = 5:
P[5, 5]

5!
0.3 5 0.7
(5 5)! 5!

5 5

1VO^bS`'( BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\

¥ 120 ´
P[5, 5] ¦
0.0024 1 0.0024
§ 1 120 ¶µ

The following table summarizes all the previous probabilities.
r

P[r,5]

0
1
2
3
4
5

0.1681
0.3601
0.3087
0.1323
0.0283
0.0024
Total = 1.0

This table represents the binomial probability distribution for r successes in five trials
with the probability of success equal to 0.30. Notice that the sum of all the probabilities equals 1, which is a requirement for all probability distributions. Figure 9.1 shows
this probability distribution as a histogram.

4WUc`S'

0.4

Binomial probability
distribution.

Probability

0.3

0.2

0.1

0
0

Number of Successes

>O`b ( >`]POPWZWbgB]^WQa
From this figure, we can see that the most likely number of papers that Kaylee will
show up with over 5 days is 1.
Finally, we can calculate the probability of multiple events for this distribution. For
instance, the probability that Kaylee will steal at least three papers over the next five
days is this:
P[r r 3] P[3,5] P[4,5] P[5,5]
P[r r 3] 0.1323 0.0283 0.0024 0.163
Also, the probability that Kaylee will take no more than one paper over the next five
days is this:
P[r b 1] P[0,5] P[1,5]
P[r b 1] 0.1681 0.3601 0.5285
Our neighbors will be thrilled to see these figures!

0W\][WOZ>`]POPWZWbgBOPZSa
As the number of trials increases in a binomial experiment, calculating probabilities
using the previous formula will really drain the batteries in your calculator and possibly even your brain. An easier way to arrive at these probabilities is to use a binomial
probability table, which I have conveniently provided in Appendix B of this book.
Below is an excerpt from this appendix, with the probabilities from our previous
example underlined.
The probability table is organized by values of n, the total number of trials. The number of successes, r, are the rows of each section, whereas the probability of success, p,
are the columns. Notice that the sum of each block of probabilities for a particular
value of p adds to 1.0.

1VO^bS`'( BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\

DOZcSa]T]
n r 0.1

0.2

4 0 0.6561
1 0.2916

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.4096 0.2401 0.1296

0.0625

0.0256

0.0081

0.0016

0.0001

0.4096 0.4116 0.3456

0.2500

0.1536

0.0756

0.0256

0.0036

2 0.0486

0.1536 0.2646 0.3456

0.3750

0.3456

0.2646

0.1536

0.0486

3 0.0036

0.0256 0.0756 0.1536

0.2500

0.3456

0.4116

0.4096

0.2916

4 0.0001

0.0016 0.0081 0.0256

0.0625

0.1296

0.2401

0.4096

0.6561

5 0 0.5905

0.3277 0.1681 0.0778

0.0313

0.0102

0.0024

0.0003

0.0000

1 0.3280

0.4096 0.3601 0.2592

0.1563

0.0768

0.0284

0.0064

0.0005

2 0.0729

0.2048 0.3087 0.3456

0.3125

0.2304

0.1323

0.0512

0.0081

3 0.0081

0.0512 0.1323 0.2304

0.3125

0.3456

0.3087

0.2048

0.0729

4 0.0005

0.0064 0.0283 0.0768

0.1563

0.2592

0.3601

0.4096

0.3281

5 0.0000

0.0003 0.0024 0.0102

0.0313

0.0778

0.1681

0.3277

0.5905

One limitation of using binomial tables is that you are restricted to using only the
values of p that are shown in the table. For instance, the previous table would not be
useful for p 0.35. Other statistics books may contain binomial tables that are more
extensive than the one in Appendix B.

CaW\U3fQSZb]1OZQcZObS0W\][WOZ>`]POPWZWbWSa
A convenient way to calculate binomial probabilities is to rely on our friend Excel,
with its BINOMDIST function. This built-in function has the following characteristics:
BINOMDIST(r, n, p, cumulative)
where:
cumulative = FALSE if you want the probability of exactly r successes
cumulative = TRUE if you want the probability of r or fewer successes
For instance, Figure 9.2 shows the BINOMDIST function being used to calculate the
probability that Kaylee will bring back exactly two papers during the next five days.

>O`b ( >`]POPWZWbgB]^WQa

4WUc`S'
BINOMDIST function in
Excel for exactly r successes.

Cell A1 contains the Excel formula =BINOMDIST(2,5,0.3,FALSE) with the result
being 0.3087.
Excel will also calculate the probability that Kaylee will bring back no more than two
papers over the next five days, as shown in Figure 9.3.

4WUc`S'!
BINOMDIST function in
Excel for no more than r
successes.

Cell A1 contains the Excel formula =BINOMDIST(2,5,0.3,TRUE) with the result
being 0.8369, which is the same as this:
P[r b 2] P[0,5] P[1,5] P[2,5]
P[r b 2] 0.1681 0.3601 0.3087 0.8369
In other words, there is more than an 83 percent chance Kaylee will show up at our
back door with 0, 1, or 2 papers that don’t belong to us during the next 5 days. That
dog sure does keep me busy!
One benefit of using Excel to determine binomial probabilities is that you are
not limited to the values of p shown in the binomial table in Appendix B. Excel’s
BINOMDIST function allows you to use any value between 0 and 1 for p.

1VO^bS`'( BVS0W\][WOZ>`]POPWZWbg2Wab`WPcbW]\

BVS;SO\O\RAbO\RO`R2SdWObW]\T]`bVS0W\][WOZ
2Wab`WPcbW]\
You can calculate the mean for a binomial probability distribution by using the following equation:

M np
where:
n = the number of trials
p = the probability of a success
For Kaylee’s example, the mean of the distribution is as follows:

M np (5)( 0.3) 1.5 papers
In other words, Kaylee brings back, on average, 1.5 papers every 5 days.
You can calculate the standard deviation for a binomial probability distribution using
the following equation:

S npq
where:
q = the probability of a failure
For our example, the standard deviation of the distribution is as follows:

S npq (5)( 0.3)( 0.7) 1.02 papers
Well, that about covers the binomial probability distribution discussion. Don’t be too
sad, though; you’ll see this again in future chapters.

G]c`Bc`\
1. What is the probability of seeing exactly 7 heads after tossing a coin 10 times?
2. Goldey-Beacom College accepts 75 percent of applications that are submitted
for entrance. What is the probability that they will accept exactly three of the
next six applications?

>O`b ( >`]POPWZWbgB]^WQa
3. Michael Jordan makes 80 percent of his free throws. What is the probability that
he will make at least six of his next eight free-throw attempts?
4. A student randomly guesses at a 12-question, multiple-choice test where each
question has 5 choices. What is the probability that the student will correctly
answer exactly six questions?
5. Historical records show that 5 percent of people who visit a particular website
purchase something. What is the probability that no more than two people out
of the next seven will purchase something?
6. During the 2005 Major League Baseball season, Derrek Lee had a 0.335 batting
average. Construct a binomial probability distribution for the number of successes (hits) for four official at bats during this season.
7. Sixty percent of a particular college population are female students. What is the
probability that a class of 10 students has exactly 4 female students?

BVS:SOabG]c]Waa]\>`]POPWZWbg
2Wab`WPcbW]\
7\BVWa1VO^bS`
U Describe the characteristics of a Poisson process
U Calculate probabilities using the Poisson equation
U Use the Poisson probability tables
U Use Excel to calculate Poisson probabilities

U Use the Poisson equation to approximate the binomial equation

Now that we have mastered the binomial probability distribution, we are
ready to move on to the next discrete theoretical distribution, the Poisson.
This probability distribution is named after Simeon Poisson, a French
mathematician who developed the distribution during the early 1800s.
The Poisson distribution is useful for calculating the probability that a certain number of events will occur over a specific period of time. We could
use this distribution to determine the likelihood that 10 customers will walk
into a store during the next hour or that 2 car accidents will occur at a busy
intersection this month. So let’s grab some crêpes and croissants and learn
about some French math.

>O`b ( >`]POPWZWbgB]^WQa

1VO`OQbS`WabWQa]TO>]Waa]\>`]QSaa
In Chapter 9, we defined a binomial experiment, otherwise known as a Bernoulli
process, as counting the number of successes over a specific number of trials. The
result of each trial is either a success or a failure. A Poisson process counts the number
of occurrences of an event over a period of time, area, distance, or any other type of
measurement.
Rather than being limited to only two outcomes, the Poisson process can have any
number of outcomes over the unit of measurement. For instance, the number of customers who walk into our local convenience store during the next hour could be zero,
one, two, three, or so on. The random variable for the Poisson distribution would be
the actual number of occurrences—in this case, the number of customers arriving during the next hour.
The mean for a Poisson distribution is the average number of occurrences that would
be expected over the unit of measurement. For a Poisson process, the mean has to
be the same for each interval of measurement. For instance, if the average number of
customers walking into the store each hour is 11, this average needs to apply to every
one-hour increment.
The last characteristic of a Poisson process is that the number of occurrences during
one interval is independent of the number of occurrences in other intervals. In other
words, if six customers walk into the store during the first hour of business, this would
have no effect on the number of customers arriving during the second hour.

A Poisson process has the following characteristics: (1) the experiment consists of
counting the number of occurrences of an event over a period of time, area, distance,
or any other type of measurement; (2) the mean of the Poisson distribution has to be
the same for each interval of measurement; (3) the number of occurrences during one
interval is independent of the number of occurrences in any other interval.

Examples of random variables that may follow a Poisson probability distribution
include the following:
U The number of cars that arrive at a tollbooth over a specific period of time
U The number of typographical errors found in a manuscript

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

U The number of students who are absent in my Monday-morning statistics class
U The number of professional football players who are placed on the injured list

each week
Now that you understand the basics of a Poisson process, let’s move into probability
calculations.

BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\
If a random variable follows a pattern consistent with a Poisson probability distribution, we can calculate the probability of a certain number of occurrences over a given
interval. To make this calculation, we need to know the average number of occurrences for the event over this interval. To demonstrate the use of the Poisson probability distribution, I’ll use this example.
The following story is true, but the names have not been changed because nobody in
this story is innocent. Not that any of the previous stories have been false, but this one
is “really” true. Each year, Brian, John, and I make a golf pilgrimage to Myrtle Beach,
South Carolina. On our last night one particular year, we were browsing through a
golf store. Brian somehow convinced me to purchase a used, fancy, brand-name golf
club that he swore he absolutely had to have in order to reach his full potential as a
golfer. Even used, this club cost more than any I had purchased new, but teenagers
have this special talent that allows them to disregard any rational adult logic when
their minds are made up.
Early the following morning, we packed our bags, checked out of the hotel, and drove
to our final round of golf, which I had cleverly planned to be along our route back
home. On the first tee, Brian pulled out his new, used prize possession and proceeded
to hit a “duck hook,” which is a golfer’s term for a ball that goes very short and very
left, often into a bunch of trees never to be seen again. I smiled nervously at Brian and
tried to convince myself that he’d be fine on the next hole. After hitting duck hooks
on holes two, three, and four, I found myself physically restraining Brian from throwing his new, used prized possession into the lake.
After our round was over, I drove back to Myrtle Beach to return the club, adding an
hour to what would have been a 10-hour car ride. (I just hope Brian remembers times
like these when I’m a frail old man drooling away in a retirement home.) At the golf
store, the woman cheerfully said she would take the club back, but I needed to show
her … the receipt. Now I vaguely remembered putting the receipt someplace “special”
just in case I would need it, but after packing, checking out, and playing golf, I would

>O`b ( >`]POPWZWbgB]^WQa
have had a better chance of discovering a cure for cancer than remembering where I
had put that piece of paper.
Not being one to give up easily, I marched back to the car and started unpacking
everything. After a short while, during which time I had spread out my dirty underwear and socks all over their parking lot, the same woman walked out to tell me the
store would gladly refund my money without the receipt if I would just pack up my
things and put them back in the car.
I discovered a very powerful technique here that I am going to pass along to you. Just
consider this a bonus for using my book. Whenever I can’t find a receipt when I need
to return something, I simply take along some dirty clothes in a suitcase and reenact
my Myrtle Beach scenario right in front of the store. It works like a charm.
Anyway, let’s assume that the number of tee shots that Brian normally hits that actually land in the fairway during a round of golf is five. The fairway is the area of short
grass where the people who have designed this nerve-wracking game intended your
tee shot to land. We will also assume that the actual number of fairways that Brian
“hits” during one round follows the Poisson distribution.

Wrong Number
How do I know that the actual number of fairways that Brian “hits” during one
round follows the Poisson distribution? At this point, I really don’t know for sure.
What I would need to do to verify this claim is record the number of fairways hit over
several rounds and then perform a “Goodness of Fit” test to decide whether the data
fits the pattern of a Poisson distribution. I promise you that we will perform this test in
Chapter 18, so please be patient.

We can now use the Poisson probability distribution to calculate the probability that
Brian will hit x number of fairways during his next round, as follows:
M xe M
P[x ]
x!
where:
x = the number of occurrences of interest over the interval

M = the mean number of occurrences over the interval
e = the mathematical constant 2.71828

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

P[x] the probability of exactly x occurrences over the interval
We can now calculate the probability that
Brian will hit exactly seven fairways during
his next round. With M 5, the equation
becomes this:

5 2.71838
7

P[7]

Bob’s Basics
Some statistics books use
the symbol Q, pronounced
lambda, to denote the mean
of a Poisson probability distribution. However, regardless of
the notation, it’s still the same
equation.

78125 0.006738
P[7]
7 6 5 4 3 21

0.1044

In other words, Brian has slightly more
than a 10 percent chance of hitting exactly
seven fairways.

Bob’s Basics
Remember from Chapter 8,
0! = 1. Also x0 = 1 for any
value of x.

We can also calculate the cumulative probability that Brian will hit no more than two
fairways using the following equations:
P[x b 2] P[x 0] P[x 1] P[x 2]

5 2.71838
0

P[x 0]

5 2.71838

5 2.71838
2

P[x 2]

0!
1

P[x 1]

1 0.006738

0.0067

5 0.006738
1

0.0337

25 0.006738
21

0.0842

P[x b 2] 0.0067 0.0337 0.0842 0.1246
There is a 12.46 percent chance that Brian will hit no more than two fairways during
his next round.
In the previous example, the mean of the Poisson distribution happened to be an
integer (5). However, this doesn’t have to always be the case. Suppose the number
of absent students for my Monday-morning statistics follows a Poisson distribution,
with the average being 2.4 students. The probability that there will be three students
absent next Monday is as follows.

>O`b ( >`]POPWZWbgB]^WQa

2.4 2.71838
3

P[x 3]

2.4

13.824 0.090718
3 21

0.2090

Looks like I need to start taking roll on Mondays!
There’s one more cool feature of the Poisson distribution: the variance of the distribution is the same as the mean. In other words:

S2 M
This means that there are no nasty variance calculations like the ones we dealt with in
previous chapters for this distribution.

>]Waa]\>`]POPWZWbgBOPZSa
Just like the binomial distribution, the Poisson probability distribution has a table
that allows you to look up the probabilities for certain mean values. You can find the
Poisson distribution table in Appendix B of this book. The following is an excerpt
from this appendix with the probabilities from our Myrtle Beach example underlined.

DOZcSa]T M
x

3.2

3.4

3.6

3.8

4.0

4.2

4.4

4.6

4.8

5.0

0.0408 0.0334 0.0273 0.0224 0.0183 0.0150 0.0123 0.0101 0.0082 0.0067

0.1304 0.1135 0.0984 0.0850 0.0733 0.0630 0.0540 0.0462 0.0395 0.0337

0.2087 0.1929 0.1771 0.1615 0.1465 0.1323 0.1188 0.1063 0.0948 0.0842

0.2226 0.2186 0.2125 0.2046 0.1954 0.1852 0.1743 0.1631 0.1517 0.1404

0.1781 0.1858 0.1912 0.1944 0.1954 0.1944 0.1917 0.1875 0.1820 0.1755

0.1140 0.1264 0.1377 0.1477 0.1563 0.1633 0.1687 0.1725 0.1747 0.1755

0.0608 0.0716 0.0826 0.0936 0.1042 0.1143 0.1237 0.1323 0.1398 0.1462

0.0278 0.0348 0.0425 0.0508 0.0595 0.0686 0.0778 0.0869 0.0959 0.1044

0.0111 0.0148 0.0191 0.0241 0.0298 0.0360 0.0428 0.0500 0.0575 0.0653

0.0040 0.0056 0.0076 0.0102 0.0132 0.0168 0.0209 0.0255 0.0307 0.0363

10 0.0013 0.0019 0.0028 0.0039 0.0053 0.0071 0.0092 0.0118 0.0147 0.0181

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

11 0.0004 0.0006 0.0009 0.0013 0.0019 0.0027 0.0037 0.0049 0.0064 0.0082
12 0.0001 0.0002 0.0003 0.0004 0.0006 0.0009 0.0013 0.0019 0.0026 0.0034
13 0.0000 0.0000 0.0001 0.0001 0.0002 0.0003 0.0005 0.0007 0.0009 0.0013
14 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0001 0.0002 0.0003 0.0005
15 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0002

The probability table is organized by values of R, the average number of occurrences.
Notice that the sum of each block of probabilities for a particular value of R adds to 1.
As with the binomial tables, one limitation of using the Poisson tables is that you are
restricted to using only the values of R that are shown in the table. For instance, the
previous table would not be useful for R = 0.45. However, other statistics books might
contain Poisson tables that are more extensive than the one in Appendix B.
The Poisson distribution for R = 5 is shown graphically in the following histogram.
The probabilities in Figure 10.1 are taken from the last column in the previous table.

4WUc`S

0.2

Poisson probability
distribution.

Probability

0.16
0.12
0.08
0.04
0
0

Number of Occurrences

Note that the most likely number of occurrences for this distribution is four and five.
Here’s another example. Let’s assume that the number of car accidents each month
at a busy intersection that I pass on my way to work follows the Poisson distribution
with a mean of 1.8 accidents per month. What is the probability that three or more
accidents will occur next month? You can express this as:
P[x r 3] P[x 3] P[ x 4 ] P[x 5] P[ x d ]

>O`b ( >`]POPWZWbgB]^WQa
Technically, with a Poisson distribution, there is no upper limit to the number of
occurrences during the interval. You’ll notice from the Poisson tables that the probability of a large number of occurrences is practically zero. Because we cannot add all
the probabilities of an infinite number of occurrences (if you can, you’re a much better
statistician than I am!), we need to take 1 minus the complement of P[x r 3] or:
P[x r 3] 1 P[ x 3]
because:
P[x 0 ] P[x 1] P[x 2] P[x 3] P[ x d ] 1.0
Therefore, to find the probability of three or more accidents, we’ll use the following:
P[x r 3] 1

P[x 0]

P[x 1] P[x 2]

Using the probabilities underlined in the following Poisson table (I seem to have misplaced my calculator), we have this:

DOZcSa]TR
x

1.1

1.2

0.3329 0.3012 0.2725 0.2466 0.2231 0.2019 0.1827 0.1653 0.1496 0.1353

0.3662 0.3614 0.3543 0.3452 0.3347 0.3230 0.3106 0.2975 0.2842 0.2707

0.2014 0.2169 0.2303 0.2417 0.2510 0.2584 0.2640 0.2678 0.2700 0.2707

0.0738 0.0867 0.0998 0.1128 0.1255 0.1378 0.1496 0.1607 0.1710 0.1804

0.0203 0.0260 0.0324 0.0395 0.0471 0.0551 0.0636 0.0723 0.0812 0.0902

0.0045 0.0062 0.0084 0.0111 0.0141 0.0176 0.0216 0.0260 0.0309 0.0361

0.0008 0.0012 0.0018 0.0026 0.0035 0.0047 0.0061 0.0078 0.0098 0.0120

0.0001 0.0002 0.0003 0.0005 0.0008 0.0011 0.0015 0.0020 0.0027 0.0034

0.0000 0.0000 0.0001 0.0001 0.0001 0.0002 0.0003 0.0005 0.0006 0.0009

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0001 0.0001 0.0001 0.0002

P[x r 3] 1

1.3

0.1653

1.4

1.5

1.6

1.7

1.8

1.9

2.0

0.2975 0.2678

P[x r 3] 1 0.7306 0.2694
There is almost a 27 percent chance that three or more accidents will occur in this
intersection next month. Looks like I better find a safer way to work!

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

CaW\U3fQSZb]1OZQcZObS>]Waa]\>`]POPWZWbWSa
You can also conveniently calculate Poisson probabilities using Excel. The built-in
POISSON function has the following characteristics:
POISSON(x, R, cumulative)
where:
cumulative = FALSE if you want the probability of exactly x occurrences
cumulative = TRUE if you want the probability of x or fewer occurrences
For instance, Figure 10.2 shows the POISSON function being used to calculate the
probability that there will be exactly two accidents in the intersection next month.

4WUc`S
POISSON function in Excel
for exactly x occurrences.

Cell A1 contains the Excel formula =POISSON(2,1.8,FALSE) with the result being
0.2678. This probability is underlined in the previous table.
Excel will also calculate the cumulative probability that there will be no more than
two accidents in the intersection, as shown in Figure 10.3.

4WUc`S!
POISSON function in Excel
for no more than x occurrences.

Cell A1 contains the Excel formula =POISSON(2,1.8,TRUE) with the result being
0.7306, a probability that we saw in the last calculation and which is also the sum of
the underlined probabilities in the previous table.

>O`b ( >`]POPWZWbgB]^WQa
One benefit of using Excel to determine Poisson probabilities is that you are not limited to the values of R shown in the Poisson table in Appendix B. Excel’s POISSON
function allows you to use any value for R.

CaW\UbVS>]Waa]\2Wab`WPcbW]\OaO\/^^`]fW[ObW]\b]
bVS0W\][WOZ2Wab`WPcbW]\
I don’t know about you, but when I have two ways to do something, I like to choose
the one that’s less work. If you don’t agree with me, feel free to skip this material. If
you do, read on!
We can use the Poisson distribution to calculate binomial probabilities under the following conditions:
U When the number of trials, n, is greater than or equal to 20 and …
U When the probability of a success, p, is less than or equal to 0.05 …

The Poisson formula would look like this:
P[x ]

np x e np
x!

where:
n = the number of trials
p = the probability of a success

Bob’s Basics
If you need to calculate binomial probabilities with the
number of trials, n, greater
than or equal to 20 and the
probability of a success, p, less
than or equal to 0.05, you can
use the equation for the Poisson
distribution to approximate the
binomial probabilities.

You might be asking yourself at this moment just why
you would want to do this. The answer is because
the Poisson formula has fewer computations than the
binomial formula and, under the stated conditions,
the distributions are very close to one another.

Just in case you are from Missouri (the “Show Me”
state), I’ll demonstrate this point with an example.
Suppose there are 20 traffic lights in my town and
each has a 3 percent chance of not working properly
(a success) on any given day. What is the probability
that exactly 1 of the 20 lights will not work today?
This is a binomial experiment with n = 20, r = 1, and p = 0.03. From Chapter 9, we
know that the binomial probability is this:

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\
P[r , n ]

n!
pr qn
( n r )! r !

P[1, 20 ]

20 !
0.03 1 0.97
( 20 1)!1!

20 1

P[1, 20 ] 20 0.03 0.560613 0.3364
The Poisson approximation is as follows:
P[x ]

np x e np

x!
Because np ( 20 )( 0.03) 0.6 :
P[1]

0.6 1 e 0.6
1!

P[1] 0.6 0.548812 0.3293
Even if you’re from Missouri, I think you would have to agree that the Poisson calculation is easier and the two results are very close. But if you need further proof …
Figures 10.4 and 10.5 show the histogram for each distribution for this example.

4WUc`S"
The binomial probability
distribution with n = 20,
p = 0.03.

>O`b ( >`]POPWZWbgB]^WQa

4WUc`S#
The Poisson probability distribution with the mean = 0.6.

Even to a skeptic, these two distributions look very much alike. So my advice to you is
to use the Poisson equation if you’re faced with calculating binomial probabilities with
n r 20 and p b 0.05.
This concludes our discussion of discrete probability distributions. I hope you’ve had
as much fun with these as I’ve had!

G]c`Bc`\
1. The number of rainy days per month at a particular town follows a Poisson distribution with a mean value of six days. What is the probability that it will rain
four days next month?
2. The number of customers arriving at a particular store follows a Poisson distribution with a mean value of 7.5 customers per hour. What is the probability that
five customers will arrive during the next hour?
3. The number of pieces of mail that I receive daily follows a Poisson distribution
with a mean value of 4.2 per day. What is the probability that I will receive more
than two pieces of mail tomorrow?
4. The number of employees who call in sick on Monday follows a Poisson distribution with a mean value of 3.6. What is the probability that no more than three
employees will call in sick next Monday?

1VO^bS`( BVS>]Waa]\>`]POPWZWbg2Wab`WPcbW]\

5. The number of spam e-mails that I receive each day follows a Poisson distribution with a mean value of 2.5. What is the probability that I will receive exactly
one spam e-mail tomorrow?
6. Historical records show that 5 percent of people who visit a particular website
purchase something. What is the probability that exactly 2 people out of the next
25 will purchase something? Use the Poisson distribution to estimate this binomial probability.
7. The number of times that Debbie proves me wrong each month follows a
Poisson distribution with a mean of 2.5 times. What is the probability that “she
who is never wrong” will fail to prove me wrong next month?

BVS:SOabG]c`]POPWZWbg
2Wab`WPcbW]\
7\BVWa1VO^bS`

U Examining the properties of a normal probability distribution
U Using the standard normal table to calculate probabilities of a normal

random variable
U Using Excel to calculate normal probabilities
U Using the normal distribution as an approximation to the binomial

distribution
Now let’s take on a new challenge, continuous random variables and a
continuous probability distribution known as the normal distribution.
Remember that in Chapter 8 we defined a continuous random variable as
one that can assume any numerical value within an interval as a result of
measuring the outcome of an experiment. Some examples of continuous
random variables are weight, distance, speed, or time.
The normal distribution is a statistician’s workhorse. This distribution is
the foundation for many types of inferential statistics that we rely on today.
We will continue to refer to this distribution through many of the remaining chapters in this book.

>O`b ( >`]POPWZWbgB]^WQa

1VO`OQbS`WabWQa]TbVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\
A continuous random variable that follows the normal probability distribution has
several distinctive features. Let’s say the monthly rainfall in inches for a particular city
follows the normal distribution with an average of 3.5 inches and a standard deviation of 0.8 inches. The probability distribution for such a random variable is shown in
Figure 11.1.

4WUc`S

Normal Probability Distribution

Normal distribution with
a mean = 3.5, standard
deviation = 0.8.

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

Mean = 3.5, Standard Deviation = 0.8

From this figure, we can make the following observations about the normal distribution:
U The mean, median, and mode are the same value—in this case, 3.5 inches.
U The distribution is bell-shaped and symmetrical around the mean.
U The total area under the curve is equal to 1.
U The left and right sides of the normal probability distribution extend indefinitely,

never quite touching the horizontal axis.
The standard deviation plays an important role in the shape of the curve. Looking
at the previous figure, we can see that nearly all the monthly rainfall measurements
would fall between 1.0 and 6.0 inches. Now look at Figure 11.2, which shows the normal distribution with the same mean of 3.5 inches, but with a standard deviation of
only 0.5 inches.

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

4WUc`S

Normal Probability Distribution

Normal distribution with
a mean = 3.5, standard
deviation = 0.5.

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

Mean = 3.5, Standard Deviation = 0.5

Bob’s Basics

Here you see a curve that’s much tighter
around the mean. Almost all the rainfall
measurements will be between 2.0 and 5.0
inches per month.
Figure 11.3 shows the impact of changing
the mean of the distribution to 5.0 inches,
leaving the standard deviation at 0.8 inches.

A smaller standard deviation
results in a “skinnier” curve
that’s tighter and taller around
the mean. A larger X (standard
deviation) makes for a “fatter”
curve that’s more spread out
and not as tall.

4WUc`S!

Normal Probability Distribution

Normal distribution with
a mean = 5.0, standard
deviation = 0.8.

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

Mean = 5.0, Standard Deviation = 0.8

In each of the previous figures, the characteristics of the normal probability distribution hold true. In each case, the values of R—the mean—and X—the standard
deviation—completely describe the shape of the distribution.

>O`b ( >`]POPWZWbgB]^WQa
The probability function for the normal distribution has a particularly mean personality (that pun was surely intended) and is shown as follows:
f [x]

1/ 2 ª̈ x

M / S ·¹

S 2P
I promise you this will be the last you’ll see of this beast. Fortunately, we have other
methods for calculating probabilities for this distribution that are more civilized and
which we will discuss in the next section.

1OZQcZObW\U>`]POPWZWbWSaT]`bVS<]`[OZ2Wab`WPcbW]\
There are a couple of approaches to calculate probabilities for a normal random variable. The following example demonstrates how this is done.
One morning a few days ago, Debbie called me on my cell phone while I was out
running errands and spoke the two words that I had feared hearing for the past year.
“They’re back,” she said. “Okay,” I replied somberly, and then hung up the phone and
headed straight toward the hardware store. My manhood was once again being challenged, and I’d be darned if I was going to take this lying down. This was war, and I
was coming home fully prepared for battle! I am referring to, of course, my annual
struggle with the most vile, the most dastardly, the most hungry creature that God has
ever placed on this planet … the Japanese beetle.
By the time I returned home from the hardware store, half of our beautiful plum tree
looked like Swiss cheese. I quickly counterattacked with a vengeance, spraying the
most potent chemicals money could buy. In the end, after the toxic spray cleared, I
stood alone, master of my domain.
Alright, let’s say that the amount of toxic spray I use each year follows a normal distribution with a mean of 60 ounces and a standard deviation of 5 ounces. This means
that each year I do battle with these demons, the most likely amount of spray I’ll use
is 60 ounces, but it will vary year to year. The probability of other amounts above and
below 60 ounces will drop off according to the bell-shaped curve. Armed with this
information, we are now ready to determine probabilities of various usages each year.

1OZQcZObW\UbVSAbO\RO`RHAQ]`S
Because the total area under a normal distribution curve equals 1 and the curve is
symmetrical, we can say the probability that I will use 60 ounces or more of spray
is 50 percent, as is the probability that I will use 60 ounces or less. This is shown in
Figure 11.4.

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\
4WUc`S"

= 60
=5

Normal distribution with
a mean = 60, standard
deviation = 5.0.
50%

50%

Ounces of Toxic Spray

How would you calculate the probability that I would use 64.3 ounces of spray or less
next year? I’m glad you asked. For this task, we need to define the standard normal
distribution, which is a normal distribution with R" and X", and is shown in Figure
11.5.

4WUc`S#

= 0
= 1

Standard normal distribution
with a mean = 0, standard
deviation = 1.0.
50%

50%

0
Number of Standard Deviations

This standard normal distribution is the
basis for all normal probability calculations,
and I’ll use it throughout this chapter.
My next step is to determine how many
standard deviations the value 64.3 is from
the mean of 60 and show this value on the
standard normal distribution curve. We do
this using the following formula:
z

The standard normal distribution is a normal distribution with
a mean equal to 0 and a standard deviation equal to 1.0.

>O`b ( >`]POPWZWbgB]^WQa
where:
x " the normally distributed random variable of interest

R" the mean of the normal distribution
X" the standard deviation of the normal distribution
z " the number of standard deviations between x and R, otherwise known as the
standard z-score.
For the current example, the standard z-score is as follows:
64.3 60
0.86
5
Now I know that 64.3 is 0.86 standard deviations away from 60 in my distribution.
z64.3

CaW\UbVSAbO\RO`R<]`[OZBOPZS
Now that I have my standard z-score, I can use the following table to determine the
probability that I will use 64.3 ounces of toxic spray or less next year. This table is an
excerpt from Appendix B and shows the area of the standard normal curve up to and
including certain values of z. Because z = 0.86 in this example, we go to the 0.8 row
and the 0.06 column to find a value of 0.8051, which is underlined.

ASQ]\RRWUWb]TH
z

0.00

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.0 0.5000 0.5040 0.5080 0.5120 0.5160 0.5199 0.5239 0.5279 0.5319 0.5359
0.1 0.5398 0.5438 0.5478 0.5517 0.5557 0.5596 0.5636 0.5675 0.5714 0.5753
0.2 0.5793 0.5832 0.5871 0.5910 0.5948 0.5987 0.6026 0.6064 0.6103 0.6141
0.3 0.6179 0.6217 0.6255 0.6293 0.6331 0.6368 0.6406 0.6443 0.6480 0.6517
0.4 0.6554 0.6591 0.6628 0.6664 0.6700 0.6736 0.6772 0.6808 0.6844 0.6879
0.5 0.6915 0.6950 0.6985 0.7019 0.7054 0.7088 0.7123 0.7157 0.7190 0.7224
0.6 0.7257 0.7291 0.7324 0.7357 0.7389 0.7422 0.7454 0.7486 0.7517 0.7549
0.7 0.7580 0.7611 0.7642 0.7673 0.7704 0.7734 0.7764 0.7794 0.7823 0.7852
0.8 0.7881 0.7910 0.7939 0.7967 0.7995 0.8023 0.8051 0.8078 0.8106 0.8133
0.9 0.8159 0.8186 0.8212 0.8238 0.8264 0.8289 0.8315 0.8340 0.8365 0.8389
1.0 0.8413 0.8438 0.8461 0.8485 0.8508 0.8531 0.8554 0.8577 0.8599 0.8621

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

This area is shown graphically in Figure 11.6.

4WUc`S$

=1
=0

The shaded area represents
the probability that z will be
less than or equal to 0.86.
0.8051

0.86

Number of Standard Deviations

The probability that the standard z-score will be less than or equal to 0.86 is 80.51
percent. Because:
P[z b 0.86 ] P[x b 64.3] 0.8051
There is an 80.51 percent chance I will use 64.3 ounces of spray or less next year
against those evil Japanese beetles. This can be seen in Figure 11.7.

4WUc`S%

= 60
=5

The shaded area represents
the probability that x will
be less than or equal to
64.3 ounces.

0.8051

64.3

Ounces of Toxic Spray

>O`b ( >`]POPWZWbgB]^WQa
Wrong Number
With continuous random variables, we cannot determine the probability of using
exactly 64.3 ounces of spray because this would be an infinitely small probability. This is because I can use an infinite amount of quantities in any given year. One
year, I could use 61.757 ounces and another year, 53.472 ounces. That’s why with
continuous random variables we can only calculate the probabilities of certain intervals, like less than 64.3 ounces or between 50.5 and 58.1 ounces. Compare this
to discrete random variables from previous chapters. Because there were only a finite
number of values for these variables, we could calculate the probability of exactly x
occurrences or r successes.

What about the probability that I will use more than 62.5 ounces of spray next year?
Because the standard normal table only has probabilities that are less than or equal to
the z-scores, we need to look at the complement to this event.
P[ x 62.5] 1 P[ x b 62.5]
The z-score now becomes this:
62.5 60
0.50
5
According to our normal table:
z62.5

P[ z b 0.50 ] 0.6915
But we want:
P[ z 0.50 ] 1 0.6915 0.3085
This probability is shown graphically in Figure 11.8.

4WUc`S&
The shaded area represents
the probability that z will
be more than 0.50.

=0
=1

0.3085
0 0.5
Number of Standard Deviations

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

Because:
P[z 0.50 ] P[x 62.5] 0.3085
There is a 30.85 percent chance that I will use more than 62.5 ounces of toxic spray.
Beetles beware!
What about the probability that I will use more than 54 ounces of spray? Again, I
need the complement, which would be this:
P[ x 54 ] 1 P[x b 54 ]
The z-score becomes this:
54 60
1.20
5
The negative score indicates that we are to the left of the distribution mean. Notice
that the standard normal table only shows positive z values. But this is no problem
because the distribution is symmetric. Figure 11.9 shows that the shaded area to the
left of –1.2 standard deviations from the mean is the same as the shaded area to the
right of +1.2 standard deviations from the mean.
z54

4WUc`S'
The shaded areas are equal.

0.1151
-1.2

0.1151
1.2

Number of Standard Deviations

We can determine the area to the right of +1.2 standard deviations as follows:
P[ z 1.2] 1 P[ z b 1.2] 1 0.8849 0.1151
Therefore, the area to the left of –1.2 standard deviations from the mean is also
0.1151. We now can calculate the area to the right of –1.2 standard deviations from
the mean.
P[ z 1.2] 1 P[ z b 1.2] 1 0.1151 0.8849

>O`b ( >`]POPWZWbgB]^WQa
Because:
P[ x 54 ] P[z 1.2] 0.8849
There is an 88.49 percent chance I will use more than 54 ounces of spray. This probability is shown graphically in Figure 11.10.

4WUc`S

= 60
=5

The shaded area is the probability that x will be more
than 54 ounces.

0.8849
0.1151

Ounces of Toxic Spray

Bob’s Basics
A shortcut to the previous example would be to recognize the following:
P[z 1.20] P[z b 1.20]
P[z 1.20] 0.8849
In general, you can use the following two relationships for any value a when dealing
with negative z-scores:
P[z a] P[z b a]
P[z b a] 1 P[z b a]

Finally, let’s look at the probability that I will use between 54 and 62.5 ounces of spray
next year. This probability is shown graphically in Figure 11.11.

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

4WUc`S

= 60
=5

The shaded area is the probability that x will be between
54 and 62.5 ounces.
0.5764

0.1151
0.3085
54

60 62.5

Ounces of Toxic Spray

We know from previous examples that the area to the left of 54 ounces is 0.1151 and
that the area to the right of 62.5 ounces is 0.3085. Because the total area under the
curve is 1:
P[54 b x b 62.5] 1 0.1151 0.3085 0.5764
There is a 57.64 percent chance that I will use between 54 and 62.5 ounces of spray
next year. I can’t wait.

BVS3[^W`WQOZ@cZS@SdWaWbSR
Remember way back in Chapter 5 we discussed the empirical rule, which stated that if
a distribution follows a bell-shaped, symmetrical curve centered around the mean, we
would expect approximately 68, 95, and 99.7 percent of the values to fall within 1.0,
2.0, and 3.0 standard deviations around the mean respectively. I’m glad to inform you
that we now have the ability to demonstrate these results.
The shaded area in Figure 11.12 shows the percentage of observations that we would
expect to fall within 1.0 standard deviation of the mean.

4WUc`S
The shaded area is the probability that x will be between
–1.0 and +1.0 standard
deviation from the mean.

0.6826
0.1587

0.1587

-1

Number of Standard Deviations

>O`b ( >`]POPWZWbgB]^WQa
Where did 68 percent come from? We can look in the normal table to get the probability that an observation will be less than one standard deviation from the mean.
P[ z b 1.0 ] 0.8413
Therefore, the area to the right of +1.0 standard deviations is this:
P[ z 1.0 ] 1 0.8413 0.1587
By symmetry, the area to the left of –1.0 standard deviations is also 0.1587. That
leaves the area between –1.0 and +1.0 as this:
P[ 1.0 b z b 1.0 ] 1 0.1587 0.1587 0.6826
The same logic is used to demonstrate the probabilities of 2.0 and 3.0 standard deviations from the mean. I’ll leave those for you to try.

1OZQcZObW\U<]`[OZ>`]POPWZWbWSaCaW\U3fQSZ
Once again we can rely on Excel to do some of the grunt work for us. The first builtin function is NORMDIST, which has the following characteristics:
NORMDIST(x, mean, standard dev, cumulative)
where:
cumulative = FALSE if you want the probability mass function (we don’t)
cumulative = TRUE if you want the cumulative probability (we do)

Bob’s Basics
Don’t be alarmed if the values that are returned using
the NORMDIST function
in Excel are slightly different
than those found in Table 3
in Appendix B. This is due to
rounding differences that are
small enough to be ignored.

For instance, Figure 11.13 shows the NORMDIST
function being used to calculate the probability that I
will use less than 64.3 ounces of spray on those nasty
beetles next year.
Cell A1 contains the Excel formula =NORMDIST
(64.3,60,5,TRUE) with the result being 0.8051. This
probability is underlined in the previous table.
Excel also has a cool function called NORMSINV,
which has the following characteristics:
NORMSINV(probability)

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

You provide this function a probability between 0 and 1, and it returns the corresponding z-score. Figure 11.14 shows the NORMSINV function returning a z-score
for a probability of 0.8413, which is 1.0 standard deviation from the mean.

4WUc`S!
NORMDIST function in
Excel for less than 64.3
ounces.

4WUc`S"
NORMSINV function in
Excel for 1.0 standard
deviation.

Cell A1 contains the Excel formula =NORMSINV(0.8413) with the result being
0.9998 (close enough to 1.0). If you look back to Figure 11.12, notice that the area to
the left of 1.0 standard deviation from the mean totals to 0.8413. You can also find this
value in the standard normal table next to z = 1.0.

CaW\UbVS<]`[OZ2Wab`WPcbW]\OaO\/^^`]fW[ObW]\b]
bVS0W\][WOZ2Wab`WPcbW]\
Remember how nasty our friend the binomial distribution can get sometimes? Well,
the normal distribution may be able help us out during these difficult times under the
right conditions. Recall from Chapter 9 that the binomial equation will calculate the
probability of r successes in n trials with p = the probability of a success for each trial
and q = the probability of a failure. If np r 5 and nq r 5, we can use the normal distribution to approximate the binomial.

>O`b ( >`]POPWZWbgB]^WQa
As an example, suppose my statistics class is composed of 60 percent females. If I
select 15 students at random, what is the probability that this group will include 8,
9, 10, or 11 female students? For this example, n = 15; p = 0.6; q = 0.4; and r = 8, 9,
10, and 11. We can use the normal approximation because np = (15)(0.6) = 9 and
nq = (15)(0.4) = 6. (Sorry, guys. I didn’t mean to infer picking you would be classified
a failure!)

Bob’s Basics
Even if you are not interested in learning how the normal distribution can be
used to approximate the binomial, I strongly encourage you to work through the
example in this section. It will be good practice for determining probabilities for a
normal distribution. And we all know that practice makes perfect!

Also recall from Chapter 9 that the mean and standard deviation of this binomial distribution is this:

M np 15 0.6 9
S npq

15 0.6 0.4

1.897

The probability that the group of 15 students will include 8, 9, 10, or 11 female
students can be calculated using the following equations:
P[8, 15]

15!
0.6
(15 8)! 8 !

P[9, 15]

15!
0.6
(15 9 )! 9 !

0.4 15 8 6435 0.0168 0.0016

0.1730

0.4 15 9 5005 0.0101 0.0041

0.2073

P[10, 15]

15!
0.6
(15 10 )!10 !

P[11, 15]

15!
0.6
(15 11)!11!

0.4 15 10 3003 0.0060 0.0102

0.4 15 11 1365 0.0036 0.0256

0.1838
0.1258

P[ r 8, 9, 10, or 11] 0.1730 0.2073 0.1838 0..1258 0.6899
Now let’s solve this problem using the normal distribution and compare the results.
Figure 11.15 shows the normal distribution with R" and X".

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

4WUc`S#

=9
= 1.897

The normal approximation to
the binomial distribution.

7.5

11.5

Number of Female Students

Notice that the shaded interval goes from 7.5 to 11.5 rather than 8 to 11. Don’t
worry; I didn’t make a mistake. I subtracted 0.5 from 8 and added 0.5 to 11 to compensate for the fact that the normal distribution is continuous and the binomial is
discrete. Adding and subtracting 0.5 is known as the continuity correction factor. For
larger values of n, like 100 or more, you can ignore this correction factor.
Now we need to calculate the z-scores.
z11.5
z7.5

S
M

11.5 9
1.32
1.897

7.5 9
0.79
1.897

According to the normal table:
P[ z b 1.32] 0.9066
This area is shown in the shaded region of Figure 11.16.

4WUc`S$

=0
=1

The probability that
z f +1.32 standard
deviations from the mean.

0.7852

+1.32

Number of Standard Deviations

>O`b ( >`]POPWZWbgB]^WQa
We also know because of symmetry with the normal curve that:
P[ z b 0.79 ] 1 P[ z b 0.79 ]
According to the normal table:
P[ z b 0.79 ] 0.7852
Therefore:
P[ z b 0.79 ] 1 0.7852 0.2148
This probability is shown in the shaded area in Figure 11.17.

4WUc`S%
The probability that
z f –0.79 standard
deviations from the mean.

=0
=1

0.2148
-0.79

Number of Standard Deviations

The probability of interest for this example is the area between z-scores of –0.79 and
+1.32. We can use the following calculations to find this area:
P[ 0.79 b z b 1.32] P[ z b 1.32] P[ z b 0.79 ]
P[ 0.79 b z b 1.32] 0.9066 0.2148 0.6918
This probability is shown in the shaded area in Figure 11.18.

4WUc`S&
The probability that
–0.79 f z f +1.32 standard
deviations from the mean.

=0
=1

0.6918

-0.79

+1.32

Number of Standard Deviations

1VO^bS`( BVS<]`[OZ>`]POPWZWbg2Wab`WPcbW]\

Using the normal distribution, we have determined the probability that my group of
15 students will contain 8, 9, 10, or 11 females is 0.6916. As you can see, this probability is very close to the result we obtained using the binomial equations, which was
0.6899.
Well, this ends our chapter on the normal probability distribution. And I feel much
better prepared for next year’s return visit of my archenemy, the Japanese beetle. Wish
me luck.

G]c`Bc`\
1. The speed of cars passing through a checkpoint follows a normal distribution
with R = 62.6 miles per hour and X = 3.7 miles per hour. What is the probability
that the next car passing will …
a. Be exceeding 65.5 miles per hour?
b. Be exceeding 58.1 miles per hour?
c. Be between 61 and 70 miles per hour?
2. The selling price of various homes in a community follows the normal distribution with R = $176,000 and X = $22,300. What is the probability that the next
house will sell for …
a. Less than $190,000?
b. Less than $158,000?
c. Between $150,000 and $168,000?
3. The age of customers for a particular retail store follows a normal distribution
with R = 37.5 years and X = 7.6 years. What is the probability that the next customer who enters the store will be …
a. More than 31 years old?
b. Less than 42 years old?
c. Between 40 and 45 years old?
4. A coin is flipped 14 times. Use the normal approximation to the binomial distribution to calculate the probability of a total of 4, 5, or 6 heads. Compare this to
the binomial probability.

>O`b ( >`]POPWZWbgB]^WQa
5. A certain statistics author’s golf scores follow the normal distribution with a
mean of 92 and a standard deviation of 4. What is the probability that, during his
next round of golf, his score will be …
a. More than 97?
b. More than 90?
6. The number of text messages that Debbie’s son Jeff sends and receives a month
follows the normal distribution with a mean of 4,580 (I am not making this up!)
and a standard deviation of 550. What is the probability that next month he will
send and receive …
a. Between 4,000 and 5,000 text messages?
b. Less than 4,200 text messages?

BVS:SOabG]cO`b

7\TS`S\bWOZAbObWabWQa

Now we can take all those wonderful concepts that we have stuffed into
our overloaded brains from Parts 1 and 2 and put them to work using
statistically sounding words, such as confidence interval and hypothesis test.
Inferential statistics enables us to make statements about a general population using the results of a random sample from that population. For
instance, using inferential statistics, the winner of a political election can be
accurately predicted very early in the polling process based on the results of
a relatively small random sample that is properly chosen. Pretty cool stuff!

1VO^bS`

AO[^ZW\U
7\BVWa1VO^bS`

U The reason for measuring a sample rather than the population
U The various methods for collecting a random sample
U Defining sampling errors
U Consequences for poor sampling techniques

This first chapter dealing with the long-awaited topic of inferential statistics focuses on the subject of sampling. Way back in Chapter 1, we defined
a population as representing all possible outcomes or measurements of
interest, and a sample as a subset of a population. Here we’ll talk about
why we use samples in statistics and what can go wrong if they are not used
properly.
Virtually all statistical results are based on the measurements of a sample
drawn from a population. Major decisions are often made based on information from samples. For instance, the Nielson ratings gather information
from a small sample of homes and are used to infer the television-viewing
patterns of the entire country. The future of your favorite TV show rests
in the hands of these select few! So choosing the proper sample is a critical
step to ensure accurate statistical conclusions.

>O`b!( 7\TS`S\bWOZAbObWabWQa

EVgAO[^ZSMost statistical studies are based on a sample of the population at large. The relationship between a population and sample is shown in Figure 12.1 (and also described in
Chapter 1).

4WUc`S
The relationship between a
population and sample.

Population

Sample

Random Thoughts
Nielsen Media Research
surveys 5,000 households
nationwide to infer the television
habits of millions of people.
Because the results of these surveys are the basis for decisions
such as show cancellations and
advertising revenue, you better
believe they select this sample
very carefully.

Bob’s Basics
Often it is just not feasible to
measure an entire population. Even when it is feasible,
measuring an entire population can be a waste of time
and money and provides little
added benefit beyond measuring a sample.

Why not just measure the whole population rather
than rely on only a sample? That’s a very good question! Depending on the study, measuring an entire
population could be very expensive or just plain
impossible. If I want to measure the life span of a
certain breed of pesky mosquitoes (extremely short
if I had any say in the matter), I could not possibly
observe every single mosquito in the population.
Rather, I would need to rely on a sample of the mosquito population, measure their life span, and make
a statement about the life span of the entire population. That’s the whole concept of inferential statistics in one paragraph! Unfortunately, doing what I
just wrote is a whole lot harder than just writing it.
Doing it is what the rest of this book is all about!
Even if we could feasibly measure the entire population, to do so would often be a wasteful decision.
If a sample is collected properly and the analysis
performed correctly, we can make a very accurate
assessment of the entire population. There is very
little added benefit to continue beyond the sample
and measure everything in sight. Measuring the
population often is a waste of both time and money,
resources that seem to be very scarce these days.

1VO^bS` ( AO[^ZW\U

One example where such a decision was recently made occurred at Goldey-Beacom
College, where I presently teach. I am also the Chair of the Academic Honor Code
Committee and was involved in a project whose goal was to gather information
regarding the attitude of our student body on the topic of academic integrity. It would
have been possible to ask every student at our college to respond to the survey, but
it was really unnecessary with the availability of inferential statistics. We eventually
made the intelligent decision and sampled only a portion of the students to infer the
attitudes of the population.

@O\R][AO[^ZW\U
The term random sampling refers to a sampling procedure where every member in the
population has a chance of being selected. The objective of the sampling procedure is
to ensure that the final sample to be measured is representative of the population from
which it was taken. If this is not the case, then we have a biased sample, which can lead
to misleading results. If you recall, we discussed an example of a biased sample back in
Chapter 1 with the golf course survey. The selection of a proper sample is critical to
the accuracy of the statistical analysis.
Since you can select a random sample in several ways, I’ll use the following example to
demonstrate these techniques.

Random sampling refers to
a sampling procedure where
Most of the time, I consider Debbie to be
every member in the populaa person of sound mind and judgment (she
tion has a chance of being
married me, after all). Lately, however, I
selected. A biased sample is a
have had some concerns about her behavsample that does not represent
ior dealing with the fact that she is reachthe intended population and
can lead to distorted findings.
ing a major milestone in life before I am.
Although I am not permitted to mention
exactly what this milestone is (under penalty
of her not proofreading any more chapters and other certain activities), I will say it
involves dividing the number 100 by 2. (You do the math.)

Anyway, recently we were walking through the local mall when Debbie suddenly ran to
a sales counter where they were selling fake ponytails for your hair. I had never heard
of such a thing in my life and never would have conceived of the idea in a million
years. Debbie, on the other hand, thought it was absolutely brilliant. Within seconds,
a total stranger appeared from nowhere and before I could say, “That’s my wife,” had
rearranged Debbie’s hair and, in his final crowning moment, expertly arranged a fake
hairpiece that somewhat resembled a small, furry animal on the top of her head.

>O`b!( 7\TS`S\bWOZAbObWabWQa
Debbie, beaming with her “new look,” turned to me to ask what I thought. Because
this also happened to be our wedding anniversary, I weakly said it looked great as I
handed this total stranger my credit card. (I might be a little slow in these matters, but
I’m not stupid.) Debbie spent the rest of the evening prancing through the mall with
her new cute furry animal hanging on for dear life. I have to admit, once I got used to
the idea, it did look pretty cute.
Now let’s say we wanted to conduct a survey to collect opinions of Debbie’s new look.
In fact, you, the reader, can render your opinion of Debbie after observing Figure 12.2
by sending me an e-mail from the book’s website at www.stat-guide.com.

4WUc`S
Debbie’s new look; what do
you think?

If I consider the current shoppers at the mall that night as my population, I need to
decide how to select the random sample from whose opinion I will ask. As we will see
in the following sections, there are four different ways to gather a random sample:
simple random, systematic, cluster, and stratified.

AW[^ZS@O\R][AO[^ZW\U
A simple random sample is a sample in which every member of the population has an
equal chance of being chosen. Unfortunately, this is easier said than done. In our mall
example, I can randomly approach people to ask their opinion. However, I might have

1VO^bS` ( AO[^ZW\U
some biases in my selection. For instance,
if I observe that a certain menacing-looking
person has a tattoo that says, “Death to All
Statisticians,” I might choose not to ask him
what he thinks of Debbie’s new ponytail. But
in doing so, I might be biasing my sample.

A simple random sample is a
sample in which every member
of the population has an equal
chance of being chosen.

Assuming I can rid myself of any biased
selection, Figure 12.3 would describe a
simple random sample at the mall.
X
Store 3

4WUc`S !
Store 4

Simple random sample.
X

X
X

Store 2

Store 5

X
X

X
Store 1

Store 6
X
Store 7

Each “X” represents a shopper, and each “X” that’s circled represents a shopper in my
sample.
There would be other options for choosing a simple random sample for the Academic
Integrity survey mentioned earlier in the chapter. I could randomly choose students
using a random number table, which is aptly named. (After all, it is simply a table of
numbers that are completely random.) An excerpt of such a table is shown here:
57245
42726
82768
97742
48332
26700
66156
64062
24713
90417

39666
58321
32694
58918
38634
40484
16407
10061
95591
18344

18545
59267
62828
33317
20510
28341
57395
01923
26970
22436

50534
72742
19097
34192
09198
25428
86230
29260
37647
77006

57654
53968
09877
06286
56256
08806
47495
32771
26282
87841

25519
63679
32093
39824
04431
98858
13908
71002
89759
94322

35477
54095
23518
74264
22753
04816
97015
58132
69034
45526

71309
56563
08654
01941
20944
16317
58225
58646
55281
38145

12212
09820
64815
95810
95319
94928
82255
69089
64853
86554

98911
86291
19894
26247
29515
05512
01956
63694
50837
42733

>O`b!( 7\TS`S\bWOZAbObWabWQa
Suppose we had 1,000 students in the population from which we were drawing a
sample size of 100. (We’ll discuss sample size in Chapter 14.) I would list these students with assigned numbers from 0 to 999. The random number table would tell me
to select student 572, followed by student 427, and so forth until I had selected 100
students. Using this technique, my sample of 100 students would be chosen with complete randomness.
Bob’s Basics

Each time a change is made
in the spreadsheet, Excel
automatically recalculates
all the functions and formulas,
resulting in the generation of a
new random number for each
RAND function being used.

Random numbers can also be generated with Excel
using the RAND function. Figure 12.4 demonstrates
how this is done.
Cell A1 contains the formula =RAND(), which
provides a random number between 0 and 1. This
random number would result in student 357 being
chosen for the sample.

4WUc`S "
Excel’s random number
generator.

AgabS[ObWQAO[^ZW\U
One way to avoid a personal bias when selecting people at random is to use systematic sampling. This technique results in selecting every kth member of the population
to be in your sample. The value of k will depend on
the size of the sample and the size of the population.
Using my Academic Integrity survey, with a populaIn systematic sampling, every
tion of 1,000 students and a sample of 100,
th
k member of the population is
k = 10. From a listing of the entire population, I
chosen for the sample, with the
would choose every tenth student to be included in
value of k being approximately
the sample. In general, if N = the size of the populaN
.
tion and n = the size of the sample, the value of k
n
N
.
would be approximately
n

1VO^bS` ( AO[^ZW\U

We could also apply this sampling technique to the mall survey. Figure 12.5 shows
every third customer walking into the mall being asked his or her opinion of Debbie’s
ponytail, even if the customer does have a tattoo.
Again, each “X” represents a shopper, and each “X” that’s circled represents a shopper
in my sample.

4WUc`S #
Store 4

Store 3

Systematic sampling.

Store 5

Store 2
X
X

Store 6

X
Store 1

X
Store 7

X
X
X

The benefit of systematic sampling is that it’s easier to conduct than a simple random
sample, often resulting in less time and money. The downside is the danger of selecting a biased sample if there is a pattern in the population that is consistent with the
value of k. For instance, let’s say I’m conducting a survey on campus asking students
how many hours they are studying during the week, and I select every fourth week to
collect my data. Because we are on an 8-week semester schedule at Goldey-Beacom,
every fourth week could end up being mid-terms and finals week, which would result
in a higher number of study hours than normal (or at least I would hope so!).

1ZcabS`AO[^ZW\U
If we can divide the population into groups, or clusters, then we can select a simple
random sample from these clusters to form the final sample. Using the Academic
Integrity survey, the clusters could be defined as classes. We would randomly choose
different classes to participate in the survey. In each class chosen, every student would
be selected to be part of the sample.

>O`b!( 7\TS`S\bWOZAbObWabWQa

A cluster sample is a simple
random sample of groups, or
clusters, of the population. Each
member of the chosen clusters
would be part of the final
sample.

4WUc`S $

We could also conduct the mall survey using cluster
sampling. Clusters could be defined as stores in the
mall population. We could randomly choose different stores and ask each customer in these stores his
or her opinion about Debbie’s ponytail. Figure 12.6
shows cluster sampling graphically.

Cluster sampling.

Store 3

X
X

Store 4
X

X
X

Store 2

Store 5
X

Store 6

X
Store 1
X

X
Store 7

According to this figure, stores one, three, and four have been chosen to participate in
the survey.
For cluster sampling to be effective, it is assumed that each cluster selected for the
sample is representative of the population at large. In effect, each cluster is a miniaturized version of the overall population. If used properly, cluster sampling can be a
very cost-effective way of collecting a random sample from the population. In the mall
example, I would only have to visit three stores to conduct my survey, saving me valuable time on my wedding anniversary.

Ab`ObWTWSRAO[^ZW\U
In stratified sampling, we divide the population into mutually exclusive groups, or
strata, and randomly sample from each of these groups. Using our mall example, we
could define our strata as male and female shoppers. Using stratified sampling, I can
be sure that my final sample contains an equal number of male and female shoppers.
This can be shown graphically in Figure 12.7.

1VO^bS` ( AO[^ZW\U
There are many different ways to establish strata from the population. Using the
Academic Integrity survey, we could define
our strata as undergraduate and graduate
students. If 20 percent of our college population are graduate students, I could use
stratified sampling to ensure that 20 percent
of my final sample are also graduate students. Other examples of criteria that we
can use to divide the population into strata
are age, income, or occupation.
F

Store 3

A stratified sample is obtained
by dividing the population into
mutually exclusive groups, or
strata, and randomly sampling
from each of these groups.

Store 4

4WUc`S %
Stratified sampling.

F
F
F

M
M
Store 5

Store 2
M
M

M
Store 6

Store 1
F

M
F
Store 7

Stratified sampling is helpful when it is important that the final sample has certain
characteristics of the overall population. If we chose to use a simple random sample at
the mall, the final sample may not have the desired proportion of males and females.
This would lead to a biased sample if males feel differently about Debbie’s new look
than females.

AO[^ZW\U3``]`a
Up to this point, we have stressed the benefits of drawing a sample from a population
rather than measuring every member of the population. However, in statistics, as in
life, there’s no such thing as a free lunch. By relying on a sample, we expose ourselves
to errors that can lead to inaccurate conclusions about the population.

>O`b!( 7\TS`S\bWOZAbObWabWQa
The type of error that a statistician is most concerned about is called sampling error,
which occurs when the sample measurement is different from the population measurement. Because the population is rarely measured in its entirety, the sampling error
cannot be directly calculated. However, with inferential statistics, we’ll be able to
assign probabilities to certain amounts of sampling error later in Chapter 15.

Sampling error occurs when
the sample measurement is
different from the population
measurement. It is the result of
selecting a sample that is not
a perfect match to the entire
population.

Sampling errors occur because we might have the
unfortunate luck of selecting a sample that is not a
perfect match to the entire population. If the majority
of mall shoppers really did like Debbie’s new look but
we just happened to choose a bunch of morons for
our sample who did not fully appreciate a good thing
when they saw it, Debbie might never wear her new
ponytail again.

Sampling errors are expected and usually are a small
price to pay to avoid measuring an entire population.
One way to reduce the sampling error of a statistical
study is to increase the size of the sample. In general, the larger the sample size, the
smaller the sampling error. If you increase the sample size until it reaches the size of
the population, then the sampling error will be reduced to zero. But in doing so, you
forfeit the benefits of sampling.

3fO[^ZSa]T>]]`AO[^ZW\UBSQV\W_cSa
The technique of sampling has been widely used, both properly and improperly, in
the area of politics. One of the most famous mishaps with sampling occurred during
the 1936 presidential race when the Literary Digest predicted Alf Landon to win the
election over Franklin D. Roosevelt. Even if history is not your best subject, you can
realize somebody had egg on his face after this election day. Literary Digest drew their
sample from phonebooks and automobile registrations. The problem was that people
with phones and cars in 1936 tended to be wealthier Republicans and were not representative of the entire voting population.
Another sampling blunder occurred in the 1948 presidential race when the Gallup poll
predicted Thomas Dewey to be the winner over Harry Truman. The picture in Figure
12.8 shows a victorious Truman holding up the morning copy of the Chicago Tribune
with the headline “Dewey Defeats Truman.”

1VO^bS` ( AO[^ZW\U

4WUc`S &
Dewey Defeats Truman.

The failure of the Gallup poll stemmed from the fact that there were a large number
of undecided voters in the sample. It was wrongly assumed that these voters were representative of the decided voters who happened to favor Dewey. Truman easily won
the election with 303 electoral votes compared to Dewey’s 189.

Wrong Number
Have you ever participated in an online survey on a sports or news website that
allowed you to view the results? These surveys can be fun and interesting, but
you need to take the results with a grain of salt. That’s because the respondents are
self-selected, which means the sample is not randomly chosen. The results of these
surveys are most likely biased because the respondents would not be representative
of the population at large. For example, people without Internet access would not
be part of the sample and might respond differently than people with access to the
Internet.

As you can see, choosing the proper sample is a critical step when using inferential
statistics. Even a large sample size cannot hide the errors of choosing a sample that
is not representative of the population at large. History has shown that large sample
sizes are not needed to ensure accuracy. For example, the Gallup poll predicted that
Richard Nixon would receive 43 percent of the votes for the 1968 presidential election and in fact he won 42.9 percent. This Gallup poll was based on a sample size of
only 2,000; whereas the disastrous 1936 Literary Digest poll sampled 2,000,000 people
(source: www.personal.psu.edu/faculty/g/e/gec7/Sampling.html).

>O`b!( 7\TS`S\bWOZAbObWabWQa

G]c`Bc`\
1. You are to gather a systematic sample from a local phone book with 75,000
names. If every kth name in the phone book is to be selected, what value of k
would you choose to gather a sample size of 500?
2. Consider a population that is defined as every employee in a particular company.
How could you use cluster sampling to gather a sample to participate in a survey
involving employee satisfaction?
3. Consider a population that is defined as every employee in a particular company.
How could you use stratified sampling to gather a sample to participate in a survey involving employee satisfaction?

BVS:SOabG]cO`b!( 7\TS`S\bWOZAbObWabWQa
sample size of 10 (n = 10) qualified individuals and record how many miles they drove
yesterday. I then choose another 10 drivers and record the same information. I do this
three more times, with the results in the following table.
Sample Number

Average Number of Miles (Sample Mean)

1
2
3
4
5

40.4
76.0
58.9
43.6
62.6

As you can see, each sample has its own mean value, and each value is different. We
can continue this experiment by selecting many more samples and observe the pattern
of sample means. This pattern of sample means represents the sampling distribution
for the number of miles the average person drives in one day.

AO[^ZW\U2Wab`WPcbW]\]TbVS;SO\
The distribution from the previous example represents the sampling distribution of the
mean because the mean of each sample was the measurement of interest. This particular distribution has some interesting properties that I will discuss with the following
example.
On a recent beach vacation, the resort we had chosen advertised a ping-pong tournament, which caught the eye of my 15-year-old son, John, who has enough skill to
embarrass his poor old father. (This is the thanks I get for teaching him how to play
the game when he had to stand on top of a cooler to see over the table.) As fate would
have it, we were paired against each other and, to my relief, I found myself losing
10–8. With John needing one more point to win, I fed him two serves that he usually
crams down my throat, but he somehow missed them both, and the score was tied at
10–10.

The sampling distribution of
the mean refers to the pattern
of sample means that will occur
as samples are drawn from the
population at large.

If we had been playing back home in our basement,
I’d have been dancing for joy and feeding him trash
talk. But standing at that resort surrounded by spectators, all I could think about was a ruined vacation
over a silly ping-pong game. On John’s next serve, I
attacked the ball with a motion that somewhat resembled a person having an epileptic seizure and hit the
ball into the net. I gave my best “I can’t believe I just

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a

did that” expression and quickly sat down, breathing a hidden sigh of relief. But that
was a small price to pay since John’s pride was saved, as well as the rest of my vacation
week. The things we do for our children!
Anyway, using ping-pong balls to describe the way sample means behave, assume that
I have 100 ping-pong balls in a container in which 20 balls are marked with the number 1, 20 are marked with 2, 20 with 3, 20 with 4, and 20 with 5.
We can look at the probability distribution of this population in the following table.
Ball Number

Frequency

Relative Frequency

Probability

1
2
3
4
5

20
20
20
20
20

20/100
20/100
20/100
20/100
20/100

0.20
0.20
0.20
0.20
0.20

This is known as a discrete uniform probability distribution because each event has the
same probability, as you can see in Figure 13.1.

4WUc`S!

0.25

Discrete uniform probability
distribution.

Probability

0.2
0.15
0.1
0.05
0
1

3
Ball Number

>O`b!( 7\TS`S\bWOZAbObWabWQa

A discrete uniform probability
distribution is a distribution that
assigns the same probability to
each discrete event (and is discrete if it is countable).

We can calculate the mean and variance of a discrete
uniform distribution as follows:
1
M a b
2
S2

1
b a 1 2
12

where:
a = minimum value of the distribution
b = maximum value of the distribution
For the ping-pong ball population:
1
M 1 5 3.0
2
1
25
5 1 1 2 12 2.08
12
Keep these results in mind. We’ll be referring to them later in the chapter.
S2

Now for my demonstration. With the balls evenly mixed, I select one ball, record the
number, place it back in the container, and then select a second ball, doing the same.
This is my first sample with a size of 2 (n = 2). After doing this 25 times, I calculate
the means of each sample and show the results in the following table.

AO[^ZW\U2Wab`WPcbW]\]TbVS;SO\[+
Sample

First Ball

Second Ball

Sample Mean x

1
2
3
4
5
6
7
8
9
10
11
12

1
1
2
1
4
1
1
3
2
1
3
4

3
1
1
1
2
3
2
1
5
3
3
2

2.0
1.0
1.5
1.0
3.0
2.0
1.5
2.0
3.5
2.0
3.0
3.0

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a
13
14
15
16
17
18
19
20
21
22
23
24
25

5
3
1
4
2
2
1
2
1
5
3
5
2

2
1
4
4
2
2
1
5
2
5
2
5
1

3.5
2.0
2.5
4.0
2.0
2.0
1.0
3.5
1.5
5.0
2.5
5.0
1.5

Wrong Number

I have a slight confession to make here. I
really didn’t buy 100 ping-pong balls and
mark each one. The numbers from the previous table came from Excel’s random number function that we discussed in Chapter
12.

Students often confuse
sample size, n, and
number of samples. In the
previous example, the sample
size equals 2 (n = 2), and the
number of samples equals 25.
In other words, we have 25
samples, each of size 2.

We can convert this table into a relative frequency distribution, which is shown in the
following table.
Sample Mean x

Frequency

Relative Frequency

Probability

1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0

3
4
7
2
3
3
1
0
2

3/25
4/25
7/25
2/25
3/25
3/25
1/25
0/25
2/25

0.12
0.16
0.28
0.08
0.12
0.12
0.04
0.00
0.08

>O`b!( 7\TS`S\bWOZAbObWabWQa
The previous table represents the sampling distribution of the mean for our pingpong experiment with n = 2. We can show this distribution graphically in Figure 13.2.
I’m sure by now your highly inquisitive mind is screaming, “What happens to the
sampling distribution if we increase the sample size?” That’s an excellent question that
I will address in the next section.

4WUc`S!

0.3

Sampling distribution of the
mean for n = 2.

0.25

Probability

0.2
0.15
0.1
0.05
0
1.0

1.5

2.0

2.5 3.0 3.5
Sample Means

4.0

4.5

5.0

Bob’s Basics
The central limit theorem, in my humble opinion, is the most powerful concept for
inferential statistics. It forms the foundation for many statistical models that are used
today. It’s a good idea to cozy up to this theorem.

BVS1S\b`OZ:W[WbBVS]`S[
As I mentioned earlier, sample means behave in a very special way. According to the
central limit theorem, as the sample size, n, gets larger, the sample means tend to follow
a normal probability distribution. This holds true regardless of the distribution of the
population from which the sample was drawn. Amazing, you say.

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a
As you look at Figure 13.2, you’re probably
scratching your head and thinking, “That
distribution doesn’t look like a normal curve,
which I know is bell-shaped and symmetrical.” You’re absolutely right because a sample size of two is generally not big enough
for the central limit theorem to kick in.

According to the central limit
theorem, as the sample size, n,
gets larger, the sample means
tend to follow a normal probability distribution and tend to
cluster around the true population mean. This holds true
regardless of the distribution of
the population from which the
sample was drawn.

Let’s satisfy your curiosity and repeat my
experiment by gathering 25 samples each
consisting of 5 ping-pong balls (n = 5). I calculate the average of each sample and plot
them in Figure 13.3.

Notice the impact that increasing the sample size has on the shape of the sample distribution. It’s starting to appear somewhat bell-shaped with a little more symmetry.
Let’s look at sample sizes of 10 and 20 in Figures 13.4 and 13.5.

4WUc`S!!

0.45

Sampling distribution of the
mean for n = 5.

0.4

Probability

0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1.0

1.5

2.0

2.5 3.0 3.5
Sample Means

4.0

4.5

5.0

>O`b!( 7\TS`S\bWOZAbObWabWQa

4WUc`S!"

0.6

Sampling distribution of the
mean for n = 10.

0.5

Probability

0.4
0.3
0.2
0.1
0
1.0

4WUc`S!#

1.5

2.0

2.5 3.0 3.5
Sample Means

4.0

4.5

5.0

2.5 3.0 3.5
Sample Means

4.0

4.5

5.0

0.4

Sampling distribution of the
mean for n = 20.

0.35

Probability

0.3
0.25
0.2
0.15
0.1
0.05
0
1.0

1.5

2.0

Note that as the sample size increases, the sampling distribution tends to resemble a
normal probability distribution. I don’t know about you, but I find this pretty impressive considering the fact that the population that these samples were drawn from was
not even close to being a normal distribution. If you recall, the ping-pong ball population followed a uniform distribution as shown in Figure 13.1.
Also, notice that as the sample size increases, the sample means tend to cluster around
the true population mean, which if you recall we calculated as 3.0. This is another
important feature of the central limit theorem.
And believe it or not, the central limit theorem has even one more important feature.

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a

AbO\RO`R3``]`]TbVS;SO\
Notice in the last four figures that as the sample size increased, the distribution of
sample means tended to converge closer together. In other words, as the sample size
increased, the standard deviation of the sample means became smaller. According to
the central limit theorem (here we go again!), the standard deviation of the sample
means can be calculated as follows:
Sx

S
n

where:
S x = the standard deviation of the sample means
S = the standard deviation of the population
n = sample size
The standard deviation of the sample means
is formally known as the standard error of the
mean.
Recall that earlier in the chapter, in the section “Sampling Distribution of the Mean,”
we determined that the variance of the pingpong ball population was 2.08. Therefore:
S S 2 2.08 1.44

The standard error of the mean
is the standard deviation of
sample means. According to
the central limit theorem, the
standard error of the mean can
S
be determined by S x
.
n

We can now calculate the standard error of
the mean for n = 2 in our example:
Sx

S 1.44

1.02
n
2

Bob’s Basics
Students often confuse X and S x . The symbol X, the standard deviation of the
population, measures the variation within the population and was discussed in
Chapter 5. The symbol S x , the standard error, measures the variation of the sample means and will decrease as the sample size increases.

>O`b!( 7\TS`S\bWOZAbObWabWQa
The following table summarizes how the standard error varies with sample size in our
ping-pong ball example.

AbO\RO`R3``]`DO`WSaeWbVAO[^ZSAWhS
Sample Size

Standard Error

2
5
10
20

1.02
0.64
0.46
0.32

EVg2]SabVS1S\b`OZ:W[WbBVS]`S[E]`YLet me explain why the central limit theorem behaves the way it does. If this concept
does not interest you, feel free to skip this section. I promise you won’t hurt my feelings.
Going back to our original experiment with a sample size of two, the following table
shows all the two-ball combinations that are possible along with the sample mean.
This represents the theoretical sampling distribution of the mean because it represents all
the possible combinations of samples along with their respective probabilities.
Sample

First Ball

Second Ball

Sample Mean x

1
2
3
4
5
6
7
8
9

1
1
1
1
1
2
2
2
2

1
2
3
4
5
1
2
3
4

1.0
1.5
2.0
2.5
3.0
1.5
2.0
2.5
3.0

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25

2
3
3
3
3
3
4
4
4
4
4
5
5
5
5
5

5
1
2
3
4
5
1
2
3
4
5
1
2
3
4
5

3.5
2.0
2.5
3.0
3.5
4.0
2.5
3.0
3.5
4.0
4.5
3.0
3.5
4.0
4.5
5.0

We can convert this table into a relative frequency distribution, which is shown in the
following table.
Sample Mean x

Frequency

Relative Frequency

Probability

1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
5.0

1
2
3
4
5
4
3
2
1

1/25
2/25
3/25
4/25
5/25
4/25
3/25
2/25
1/25

0.04
0.08
0.12
0.16
0.20
0.16
0.12
0.08
0.04

This distribution is shown graphically in Figure 13.6.

>O`b!( 7\TS`S\bWOZAbObWabWQa

4WUc`S!$

0.25

Theoretical sampling
distribution of the mean.

Probability

0.2
0.15
0.1
0.05
0
1.0

The theoretical sampling distribution of the mean displays
all the possible sample means
along with their classical probabilities. See Chapter 6 for a
review of classical probability.

1.5

2.0

2.5 3.0 3.5
Sample Means

4.0

4.5

5.0

You can see by this figure that the most common
sample average is 3.0, whereas sample averages of
1.0 and 5.0 occur the least number of times. This is
because there are more possible combinations of twoball samples that average to 3.0 (5 to be exact) than
two-ball samples that average to 1.0 or 5.0 (1 to be
exact). In other words, we have five times the likelihood of drawing a two-ball sample that averages 3.0
when compared to sample averages of 1.0 or 5.0.

As we increase our sample size to 5, 10, and 20, the probability of drawing a sample
with an average of 1.0 or 5.0 decreases while the probability of drawing a sample with
an average of 3.0 increases. This explains why as sample size grows, more sample averages center around 3.0 and fewer around 1.0 and 5.0.

>cbbW\UbVS1S\b`OZ:W[WbBVS]`S[b]E]`Y
I can just sense your need right now to do something really neat with this wonderful
new tool. Look no further. If we know the sample means follow the normal probability distribution and we also know the mean and standard deviation of that distribution,
we can predict the likelihood that the sample means will be greater or less than certain
values.

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a

For example, let’s take our ping-pong ball experiment with n = 20. From the central
limit theorem, we know the sample means follow a normal distribution with:
M 3.0
S 1.44

0.32
n
20
What is the probability that our next sample of 20 ping-pong balls will have a sample
average of 3.5 or less? The sample mean distribution is shown in Figure 13.7, with the
shaded region indicating the probability of interest.
Sx

Sampling Distribution of the Mean
n = 20

4WUc`S!%
Probability the next sample
mean will be less than or
equal to 3.5.

= 3.0
x = 0.32

3.0
Sample Means

3.5

As we did in Chapter 11, we need to calculate the z-score. The equation looks slightly
different because we are working with sample means, but in reality, it is identical to
what we saw in Chapter 11.
z

x M
Sx

3.5 3.0
1.56
0.32
Using the standard z-table in Appendix B:
z3.5

P[ x b 3.5] P[ z b 1.56 ] 0.9406
This probability is shown in Figure 13.8.

>O`b!( 7\TS`S\bWOZAbObWabWQa

4WUc`S!&

Sampling Distribution of the Mean
n = 20

Probability the next sample
mean will be less than or
equal to 1.56 standard deviations from the population
mean.

= 3.0
x = 0.32

0.9406

0
+1.56
Number of Standard Deviations

According to the shaded region, the probability that our next sample of 20 ping-pong
balls will have a sample mean of 3.5 or less is approximately 94 percent.
The power of the central limit theorem lies in the fact that you need little information
about the distribution of the population to apply it. The sample means will behave
very nicely as long as the sample size is large enough. It’s a very versatile theorem that
has countless applications in the real world. I knew you’d be impressed!

AO[^ZW\U2Wab`WPcbW]\]TbVS>`]^]`bW]\
The sample mean is not the only statistical measurement that is performed. What if
I want to measure the percentage of teenagers in this country who would agree with
the following statement: “My parents are an excellent resource when I’m looking for
advice on an important matter in my life.” Because each respondent has only two
choices (agree or disagree), this experiment follows the binomial probability distribution, which I discussed in Chapter 9.

1OZQcZObW\UbVSAO[^ZS>`]^]`bW]\
My measurement of interest is the proportion of teenagers in my sample of size n,
who will agree with the statement “My parents are an excellent resource when I’m
looking for advice on an important matter in my life.” The sample proportion, ps, is
calculated by:
ps

Number of Successes in the Sample
n

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a
Because I don’t know the population proportion, p, who would agree with the statement, I need to collect data from samples
and approximate the population proportion.
With proportion data, I want the sample
size to be large enough so I can use the normal probability distribution to approximate
the binomial distribution. As you recall from
Chapter 11, if np r 5 and nq r 5, we can use
the normal distribution to approximate the
binomial (q = 1 – p, the probability of a failure). I’m hopeful that p will be at least 5 percent
(at least a few teenagers might listen to their
parents), so if I choose n = 150, then:

Wrong Number
It’s important to remember
that a proportion, either p
or p s, cannot be less than 0 or
greater than 1. A common mistake that students make is when
told that the proportion equals
10 percent, they set p = 10
rather than p = 0.10.

np = (150)(0.05) = 7.5
nq = (150)(0.95) = 142.5
Suppose I choose 10 samples, each of size 150, and record the number of agreements
(successes) in each sample in the table that follows.
Sample

Number of Successes ps

Sample Proportion

1
2
3
4
5
6
7
8
9
10

26
18
21
30
24
21
16
28
35
27

26/150 = 0.173
18/150 = 0.120
21/150 = 0.140
30/150 = 0.200
24/150 = 0.160
21/150 = 0.140
16/150 = 0.107
28/150 = 0.187
35/150 = 0.233
27/150 = 0.180

Next I average the sample proportions to approximate the population proportion, p:
p z ps

0.173 0.12 0.14 ... 0.233 0.18
0.164
10

>O`b!( 7\TS`S\bWOZAbObWabWQa

1OZQcZObW\UbVSAbO\RO`R3``]`]TbVS>`]^]`bW]\
I now need to calculate the standard deviation of this sampling distribution, which is
known as the standard error of the proportion, or S p , with the following equation:
Sp

p 1 p
n

0.164 1 0.164
0.000914 0.030
150

The standard error of the proportion is the standard deviation of the sample proportions
and can be calculated by
Sp

p1 p .
n

4WUc`S!'

Now I’m ready to answer the age-old question,
“What is the probability that from my next sample
of 150 teenagers, 20 percent or less will agree with
the statement: ‘My parents are an excellent resource
when I’m looking for advice on an important matter
in my life’?” The shaded area in Figure 13.9 represents this probability, which displays the sampling
distribution of the proportion for this example.

Sampling Distribution of the Proportion

Sampling distribution of the
proportion.
p = 0.164
␴p = 0.030

0.164

0.20

Sample Proportions

Because our sample size allows us to use the normal approximation to the binomial
distribution, we now calculate the z-score for the proportion using the following
equation:

1VO^bS`!( AO[^ZW\U2Wab`WPcbW]\a
z

ps p
Sp

0.20 0.164
1.20
0.030
Using the standard z-table in Appendix B:
z0.20

P[ ps b 0.20 ] P[ z b 1.20 ] 0.8849
This probability is also shown graphically in the shaded region in Figure 13.10.
Sampling Distribution of the Proportion

4WUc`S!
Probability that next sample
proportion will be less than or
equal to 1.2 standard deviations from the population
proportion.

p = 0.164
p = 0.030
0.8849

+1.20

Number of Standard Deviations

According to our results, there is an 88.49 percent chance that 20 percent or fewer
teenagers will agree with our statement from the next sample of size 150. Oh well,
maybe when they get older, they’ll discover the real wisdom of their parents.

G]c`Bc`\
1. Calculate the standard error of the mean when …
a. S 10, n 15
b. S 4.7, n 12
c. S 7, n 20
2. A population has a mean value of 16.0 and a standard deviation of 7.5. Calculate
the following with a sample size of 9.
a. P ¨ª x b 17 ·¹

>O`b!( 7\TS`S\bWOZAbObWabWQa
b. P ¨ª x 18 ·¹
c. P ¨ª14.5 b x b 16.5·¹
3. Calculate the standard error of the proportion when …
a. p = 0.25, n = 200
b. p = 0.42, n = 100
c. p = 0.06, n = 175
4. A population proportion has been estimated at 0.32. Calculate the following with
a sample size of 160.
a. P ª̈ ps b 0.30 ·¹
b. P ª̈ ps 0.36 ·¹
c. P ª̈0.29 b ps b 0.37 ·¹
5. A hypothetical statistics author is obsessed with making 10-foot putts. Each day
that he practices, he putts 60 times and counts the number he makes. Over the
last 20 practice sessions, he has averaged 24 made putts. What is the probability
that he will make at least 30 putts during his next session?

BVS:SOabG]cO`b!( 7\TS`S\bWOZAbObWabWQa

1]\TWRS\QS7\bS`dOZaT]`bVS;SO\eWbV:O`USAO[^ZSa
So let’s learn how to construct a confidence interval for a population mean using a
large sample size. By a large sample size, we are generally referring to n v 30. The
first step in developing a confidence interval for a population involves the following
discussion on estimators.

3abW[Ob]`a
The simplest estimate of a population is the point estimate, the most common being
the sample mean. A point estimate is a single value that best describes the population
of interest. Let me explain this concept by using the following example.

A point estimate is a single
value that best describes the
population of interest, the
sample mean being the most
common. An interval estimate
provides a range of values that
best describes the population.

I think my wife has been kidnapped and secretly
replaced by a Debbie look-alike who also happens to
be completely addicted to the QVC home shopping
channel. No one or nothing in our household has
escaped the products Debbie has found on her new
favorite TV show. She has purchased stuff for the car,
the kitchen floor, the dog, her skin, her hair, and my
back (an inversion table that she wants me to hang
upside down on!).

Suddenly “Diamonique Week” has become a
major holiday in our household. I’m not really sure
what Diamonique actually is, but I suspect it is “available for a limited time only.”
Whenever I turn on any TV in the house, the channel always seems to be set to a very
convincing home shopping channel-type person pleading with me to “Call now! Only
three left!”
Anyway, let’s say I want to estimate the average dollar value of an order for the home
shopping channel population. If my sample average was $78.25, I could use that as my
point estimate for the entire population of home shopping customers.
The advantage of a point estimate is that it is easy to calculate and easy to understand.
The disadvantage, however, is that I have no clue as to how accurate this estimate
really is.
To deal with this uncertainty, we can use an interval estimate, which provides a range of
values that best describes the population. To develop an interval estimate, we need to
learn about confidence levels.

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

1]\TWRS\QS:SdSZa
A confidence level is the probability that the interval estimate will include the population
parameter. A parameter is defined as a numerical description of a population characteristic, such as the mean.
Remember from Chapter 13 that sample
means will follow the normal probability distribution for large sample sizes. Let’s say we
want to construct an interval estimate with a
90 percent confidence level. This confidence
level corresponds to a z-score from the standard normal table equal to 1.64 as shown in
Figure 14.1.

A confidence level is the probability that the interval estimate
will include the population
parameter, such as the mean.
A parameter is data that
describes a characteristic about
a population.

4WUc`S"

90% Confidence Interval

90 percent confidence interval.

90%

-1.64

+1.64

0.95

Bob’s Basics
Notice that in Figure 14.1, 5 percent of the area under the curve lies to the right
of +1.64 and 95 percent of the area under the curve lies to the left. That’s why
you see 0.9495 (close enough to 0.95) corresponding to a z-score of 1.64 in
Table 3 of Appendix B. Remember, however, that z = 1.64 corresponds to a 90 percent confidence interval, the shaded region in the figure.

>O`b!( 7\TS`S\bWOZAbObWabWQa
In general, we can construct a confidence interval around our sample mean using the
following equations:
x

zc S x (upper limit of confidence interval)

zc S x (lower limit of confidence interval)

where:
x = the sample mean
zc = the critical z-score, which is the number of standard deviations based on
the confidence level
S x = the standard error of the mean (remember our friend from Chapter 13?)
The term zc S x is referred to as the margin of error, or E, a phrase often referred to in
polls and surveys.

A confidence interval is a
range of values used to estimate a population parameter
and is associated with a specific confidence level. The margin of error, E, determines the
width of the confidence interval
and is calculated using z c S x .

Going back to our home shopping example, let’s say
from a sample of 32 customers the average order
is $78.25 and the population standard deviation is
$37.50. (This represents the variation among orders
within the population.) We can calculate our 90 percent confidence interval as follows:
x $78.25
n = 32
X = $37.50
z c = 1.64

$37.50
S

$6.63
Sx
n
32
Upper limit = x 1.64 S x $78.25 1.64 $6.63 $89.12
Lower limit = x 1.64 S x $78.25 1.64 $6.63 $67.38
According to these results, our 90 percent confidence interval for this random sample
of home shoppers is between $67.38 and $89.12 or ($67.38, $89.12). This interval is
shown in Figure 14.2.

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa
Interval Estimate for the Average Order
Size of a Home Shopping Customer

$67.38

$78.25

$89.12

4WUc`S"
Interval estimate for the
average dollar value of a
home shopping order.

0SeO`S]TbVS7\bS`^`SbObW]\]T1]\TWRS\QS7\bS`dOZ
As described earlier, a confidence interval is a range of values used to estimate a
population parameter and is associated with a specific confidence level. A confidence
interval needs to be described in the context of several samples. If we select 10 samples
from our home shopping population and construct 90 percent confidence intervals
around each of the sample means, then theoretically 9 of the 10 intervals will contain
the true population mean, which remains unknown. Figure 14.3 shows this concept.

4WUc`S"!
Interpreting the definition of
a confidence interval.

As you can see, Samples 1 through 9 have confidence intervals that include the true
population mean, whereas Sample 10 does not.

Wrong Number
It is easy to misinterpret the definition of a confidence interval. For example, it is
not correct to state that “there is a 90 percent probability that the true population
mean is within the interval ($67.38, $89.12).” Rather, a correct statement would be
that “there is a 90 percent probability that any given confidence interval from a random sample will contain the true population mean.”

>O`b!( 7\TS`S\bWOZAbObWabWQa
Because there is a 90 percent probability that any given confidence interval will contain the true population mean in the previous example, we have a 10 percent chance
that it won’t. This 10 percent value is known as the level of significance, F, which is represented by the total white area in both tails of Figure 14.4.

4WUc`S""

Level of Significance

The level of significance.

1–

The level of significance (F)
is the probability of making a
Type I error.

The probability for the confidence interval is a complement to the significance level. For example, the
significance level for a 95 percent confidence interval
is 5 percent, the significance level for a 99 percent
confidence interval is 1 percent, and so on. In general, a (1 – F) confidence interval has a significance
level equal to F.
We will revisit the level of significance in more detail
in later chapters.

BVS3TTSQb]T1VO\UW\U1]\TWRS\QS:SdSZa
So far, we have only referred to a 90 percent confidence interval. However, we can
choose other confidence levels to suit our needs. The following table shows our home
shopping example with confidence levels of 90, 95, and 99 percent.

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

1]\TWRS\QS7\bS`dOZaeWbVDO`W]ca1]\TWRS\QS:SdSZa
Confidence
Level

90
95
99

Sample
Mean

Lower
Limit

Upper
Limit

1.64
1.96
2.57

$6.63
$6.63
$6.63

$78.25
$78.25
$78.25

$67.38
$65.26
$61.21

$89.12
$91.24
$95.29

From the previous table, you can see that
there’s a price to pay for increasing the
confidence level—our interval estimate of
the true population mean becomes wider
and less precise. We have proven that once
again, there is no free lunch with statistics.
If we want more certainty that our confidence interval will contain the true population mean, that confidence interval will
become wider.

Bob’s Basics
I recommend that you confirm
the z-scores in this table for
yourself by checking with
Table 3 in Appendix B. Practice
makes perfect! Review Chapter
11 if you need to.

BVS3TTSQb]T1VO\UW\UAO[^ZSAWhS
There is one way, however, to reduce the width of our confidence interval while
maintaining the same confidence level. We can do this by increasing the sample size.
There is still no free lunch though because increasing the sample size has a cost associated with it. Let’s say we increase our sample size to include 64 home shoppers. This
change will affect our standard error as follows:
$37.50
S

$4.69
n
64
Our new 90 percent confidence interval for our original sample will be:
Sx

x $78.25
n = 64
S x $4.69
Upper limit = x 1.64 S x $78.25 1.64 $4.69 $85.94
Lower limit = x 1.64 S x $78.25 1.64 $4.69 $70.56

>O`b!( 7\TS`S\bWOZAbObWabWQa
Increasing our sample size from 32 to 64 has reduced the 90 percent confidence interval from ($67.38, $89.12) to ($70.56, $85.94), which is a more precise interval.

2SbS`[W\W\UAO[^ZSAWhST]`bVS;SO\
We can also calculate a minimum sample size that would be needed to provide a specific margin of error. What sample size would we need for a 95 percent confidence
interval that has a margin of error of $8.00 (E = $8.00) in our home shopping example?
E zS x
zS
n
zS
n
E

¥ zS ´
n¦ µ
§ E¶

¥ 1.96 $37.50 ´
n¦
µ¶ 84.4 z 85
$8.00
§
Therefore, to obtain a 95 percent confidence interval that ranges from
$78.25 – $8.00 = $70.25 to $78.25 + $8.00 = $86.25 would require a sample size
of 85 home shopping-addicted people.

1OZQcZObW\UO1]\TWRS\QS7\bS`dOZEVS\X7aC\Y\]e\
Here’s a simple section for you. (It’s about time!) So far, all of our examples have
assumed that we knew X, the population standard deviation. What happens if X is
unknown? Don’t panic, because as long as n r 30, we can substitute s, the sample standard deviation, for X, the population standard deviation, and follow the same procedure as before. To demonstrate this technique, consider the following table that shows
the order size in dollars of 30 home shoppers.

6][SAV]^^W\UAO[^ZS[+!
75
29
99

109
70
140

32
89
112

54
100
87

121
48
122

80
40
75

96
137
54

47
75
92

67
39
89

115
88
153

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

Using Excel, we can confirm that:
x $84.47 and s $32.98
A 99 percent confidence interval around this sample mean would be:
x $84.47
n = 30
s = $32.98
zc = 2.57
Sˆ x

s
n

$32.98
$6.02
30

We use Ŝ x to indicate that we have approximated the standard error of the mean by
using s instead of X. We statisticians just love to put little hats on top of letters.
Upper limit = x

2.57Sˆ x $84.47 2.57 $6.02 $99.94

Lower limit = x 2.57Sˆ x $84.47 2.57 $6.02 $69.00
See! That wasn’t too bad.

CaW\U3fQSZÂa1=<4723<134c\QbW]\
Excel has a pretty cool built-in function that calculates confidence intervals for us.
The CONFIDENCE function has the following characteristics:
CONFIDENCE(alpha, standard_dev, size)
where:
alpha = the significance level of the confidence interval
standard_dev = the standard deviation of the population
size = sample size
For instance, Figure 14.5 shows the CONFIDENCE function being used to calculate
the confidence interval for our original home shopping example.

>O`b!( 7\TS`S\bWOZAbObWabWQa

4WUc`S"#
CONFIDENCE function in
Excel for the home shopping
sample.

Cell A1 contains the Excel formula =CONFIDENCE(0.1,37.5,32) with the result
being 10.90394. This value represents the margin of error, or the amount to add and
subtract from the sample mean, as follows:
$78.25 + $10.90 = $89.15
$78.25 – $10.90 = $67.35
This confidence interval is slightly different from the one calculated earlier in the
chapter due to the rounding of numbers. This sure beats using tables and square root
functions on the calculator.

1]\TWRS\QS7\bS`dOZaT]`bVS;SO\eWbVA[OZZAO[^ZSa
So far, this entire chapter has dealt with the case where n r 30. I’m sure you are now
wondering about how to construct a confidence interval when our sample size is less
than 30. Well, as with many things in life, it depends.
With a small sample size, we lose the use of our faithful friend, the central limit
theorem, and we need to assume that the population is normally (or approximately)
distributed for all cases. The first case that we’ll examine is when we know X, the
population standard deviation.

EVS\X7a9\]e\
When X is known, the procedure reverts back to the large sample size case. We can do
this because we are now assuming the population is normally distributed. Let’s construct a 95 percent confidence interval from the following home shopping sample of
size 10.

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

6][SAV]^^W\UAO[^ZS[+
75

109

121

115

We know the following information:
x $79.60
n = 10
X= $37.50 (given from the original example)
zc = 1.96
$37.50
S

$11.86
Sx
n
10
Upper Limit = x 1.96S x $79.60 1.96 $11.86 $102.85
Lower Limit = x 1.96S x $79.60 1.96 $11.86 $56.35
Notice that the small sample size has resulted in a wide confidence interval. Again, we
are assuming here that the population from which the sample was drawn is normally
distributed, which is the first time we have made such an assumption in this chapter so
far.

EVS\X7aC\Y\]e\
More often, we don’t know the value of X. Here, we make a similar adjustment that
we made earlier and substitute s, the sample standard deviation, for X, the population
standard deviation. However, because of the small sample size, this substitution forces
us to use a new probability distribution known
Random Thoughts
as the Student’s t-distribution (named in honor
of you, the student).
The Student’s t-distribution was
The t-distribution is a continuous probability
developed by William Gosset
distribution with the following properties:
(1876–1937) while working for the Guinness Brewing
U It is bell-shaped and symmetrical around
Company in Ireland. He
the mean.
published his findings using
the pseudonym Student. Now
U The shape of the curve depends on the
there’s a rare statistical event—
degrees of freedom (d.f.) which, when deala bashful Irishman!
ing with the sample mean, would be equal
to n – 1.

>O`b!( 7\TS`S\bWOZAbObWabWQa
U The area under the curve is equal to 1.0.

The degrees of freedom are the
number of values that are free
to be varied given information,
such as the sample mean, is
known.

4WUc`S"$

U The t-distribution is flatter than the normal dis-

tribution. As the number of degrees of freedom
increase, the shape of the t-distribution becomes
similar to the normal distribution as seen in
Figure 14.6. With more than 30 degrees of freedom (a sample size of 30 or more), the two distributions are practically identical.
The t-Distribution Compared to the Normal Curve

The Student’s t-distribution
compared to the normal distribution.

d.f. = 15
Standard Normal Curve

d.f. = 2

Students often struggle with the concept of degrees of freedom, which represent the
number of remaining free choices you have after something has been decided, such
as the sample mean. For example, if I know that my sample of size 3 has a mean of
10, I can only vary two values (n – 1). After I set those two values, I have no control
over the third value because my sample average must be 10. For this sample, I have 2
degrees of freedom.
We can now set up our confidence intervals for the mean using a small sample:
x t c Ŝ x (upper limit of confidence interval)
x t c Ŝ x (lower limit of confidence interval)
where:
t c = critical t-value (can be found in Table 4 in Appendix B)
s
, the estimated standard error of the mean
Ŝ x
n

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

To demonstrate this procedure, let’s assume the population of home shopping orders
follows a normal distribution and the following sample of size 10 was collected.

6][SAV]^^W\UAO[^ZST`][O<]`[OZ2Wab`WPcbW]\[+
29

100

137

With X unknown, we will construct a 95 percent confidence interval around the sample mean.
To determine the value of t c for this example, I need to calculate the number
of degrees of freedom. Because n = 10, I have n – 1 = 9 d.f. This corresponds to
t c 2.262, which is underlined in the following table taken from Table 4 in Appendix
B.

3fQS`^bT`][bVSAbcRS\bÂab2Wab`WPcbW]\
Selected right-tail areas with confidence levels underneath
Alpha
0.2000 0.1500 0.1000 0.0500
Conf lev 0.6000 0.7000 0.8000 0.9000
df

0.0250
0.9500

0.0100
0.9800

0.0050
0.9900

0.0010
0.9980

0.0005
0.9990

1.376

1.963

3.078

6.314

12.706

31.821

63.657

318.31

636.62

1.061

1.386

1.886

2.920

4.303

6.965

9.925

22.327

31.599

0.978

1.250

1.638

2.353

3.182

4.541

5.841

10.215

12.924

0.941

1.190

1.533

2.132

2.776

3.747

4.604

7.173

8.610

0.920

1.156

1.476

2.015

2.571

3.365

4.032

5.893

6.869

0.906

1.134

1.440

1.943

2.447

3.143

3.707

5.208

5.959

0.896

1.119

1.415

1.895

2.365

2.998

3.499

4.785

5.408

0.889

1.108

1.397

1.860

2.306

2.896

3.355

4.501

5.041

0.883

1.100

1.383

1.833

2.262

2.821

3.250

4.297

4.781

0.879

1.093

1.372

1.812

2.228

2.764

3.169

4.144

4.587

We next need to calculate the sample mean and sample standard deviation, which,
according to Excel, are as follows:
x $71.50 and s = $33.50

>O`b!( 7\TS`S\bWOZAbObWabWQa
We can now approximate the standard error of the mean:
$33.50

$10.59
n
10
and can construct our 95 percent confidence interval:
Sˆ x

Upper limit = x t c Sˆ x $71.50 2.262 $10.59 $95.45
Lower limit = x t c Sˆ x $71.50 2.262 $10.59 $47.55
Now that wasn’t too bad!

Bob’s Basics
We can use the t-distribution when all of the following conditions have been met:
U The population follows the normal (or approximately normal) distribution.
U The sample size is less than 30.
U The population standard deviation, X, is unknown and must be approximated

by s, the sample standard deviation.

That ends our discussion on confidence intervals around the mean. Next on the menu
are proportions!

1]\TWRS\QS7\bS`dOZaT]`bVS>`]^]`bW]\eWbV:O`US
AO[^ZSa
We can also estimate the proportion of a population by constructing a confidence
interval from a sample. As you might recall from Chapter 13, proportion data follow
the binomial distribution that can be approximated by the normal distribution under
the following conditions:
np r 5 and nq r 5
where:
p = the probability of a success in the population
q = the probability of a failure in the population (q = 1 – p)

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

Suppose I want to estimate the proportion of home shopping customers who are
female based on the results of a sample. In Chapter 13, we learned that we can calculate the proportion of a sample using:
Number of Successes in the Sample
n

1OZQcZObW\UbVS1]\TWRS\QS7\bS`dOZT]`bVS>`]^]`bW]\
The confidence interval around the sample proportion can be calculated by:
ps

zc S p (upper limit of confidence interval)

zc S p (lower limit of confidence interval)

where S p is the standard error of the proportion (which is the standard deviation of
the sample proportions) using:
p 1 p
n
There’s extra credit for anyone who can see a problem arising here. Our challenge is
that we are trying to estimate p, the population proportion, but we need a value for p
to set up the confidence interval. Our solution—estimate the standard error by using
the sample proportion as an approximation for the population proportion as follows:
Sp

ps 1 ps
n
We now can construct a confidence interval around the sample proportion by:
Ŝ p

zc Ŝ p (upper limit of confidence interval)

zc Ŝ p (lower limit of confidence interval)

Let’s put these equations to work. In my efforts to estimate the proportion of female
home shopping customers, I sample 175 random customers, of which 110 are female.
I can now calculate ps , the sample proportion:
Number of Successes in the Sample 110
0.629

5
n
175
The estimated standard error of the proportion would be:
ps

Sˆ p

ps 1 ps

0.629 0.371
175

0.0365

>O`b!( 7\TS`S\bWOZAbObWabWQa
We are now ready to construct a 90 percent confidence interval around our sample
proportion (zc = 1.64):
Upper limit = ps

1.64 S p 0.629 1.64 0.0365 0.689

Lower limit = ps 1.64 S p 0.629 1.64 0.0365 0.569
Our 90 percent confidence interval for the proportion of female home shopping customers is (0.569, 0.689). Debbie must be in there somewhere!

2SbS`[W\W\UAO[^ZSAWhST]`bVS>`]^]`bW]\
Almost done. Just as we did for the mean, we can determine a required sample size
that would be needed to provide a specific margin of error. What sample size would
we need for a 99 percent confidence interval that has a margin of error of 6 percent (E
= 0.06) in our home shopping example? The formula to calculate n, the sample size is:
2

¥z ´
n pq ¦ c µ
§ E¶
Notice that we need a value for p and q. If we don’t have a preliminary estimate of the
values, set p = q = 0.50. Because half the population is female, that sounds like a good
strategy to me.
2
¥ 2.57 ´
n 0.50 0.50 ¦
459
µ
§ 0.06 ¶
Therefore, to obtain a 99 percent confidence interval that provides a margin of error
no more than 6 percent would require a sample size of 459 home shoppers.

Random Thoughts
The reason we use p = q = 0.50 if we don’t have an estimate of the population proportion is that these values provide the largest sample size when compared to other
combinations of p and q. It’s like being penalized for not having specific information about your population. This way you are sure your sample size is large enough,
regardless of the population proportion.

1VO^bS`"( 1]\TWRS\QS7\bS`dOZa

G]c`Bc`\
1. Construct a 97 percent confidence interval around a sample mean of 31.3 taken
from a population that is not normally distributed with a standard deviation of
7.6 using a sample of size 40.
2. What sample size would be necessary to ensure a margin of error of 5 for a 98
percent confidence interval taken from a population that is not normal, which
has a population standard deviation of 15?
3. Construct a 90 percent confidence interval around a sample mean of 16.3 taken
from a population that is not normally distributed with a population standard
deviation of 1.8 using a sample of size 10.
4. The following sample of size 30 was taken from a population that is not normally distributed:
10

Construct a 90 percent confidence interval around the mean.
5. The following sample of size 12 was taken from a population that is normally
distributed and that has a population standard deviation of 12.7:
37

Construct a 94 percent confidence interval around the mean.
6. The following sample of size 11 was taken from a population that is normally
distributed:
121 136 102 115 126 106 115 132 125 108 130
Construct a 98 percent confidence interval around the mean.
7. The following sample of size 11 was taken from a population that is not normally distributed:
87

Construct a 99 percent confidence interval around the mean.

>O`b!( 7\TS`S\bWOZAbObWabWQa
8. A sample of 200 light bulbs was tested, and it was found that 11 were defective.
Calculate a 95 percent confidence interval around this sample proportion.
9. What sample size would you need to construct a 96 percent confidence interval
around the proportion for voter turnout during the next election that would provide a margin of error of 4 percent? Assume the population proportion has been
estimated at 55 percent.

BVS:SOabG]cO`b!( 7\TS`S\bWOZAbObWabWQa
The purpose of this particular chapter is just to introduce the basic concept of hypothesis testing. The following two chapters will then get into more specific examples of
how we put hypothesis testing to work. Stay tuned!

6g^]bVSaWaBSabW\U¾bVS0OaWQa
In the statistical world, a hypothesis is an assumption about a population parameter.
Examples of hypotheses (that’s plural for hypothesis) include the following:
U The average adult drinks 1.7 cups of coffee per day.
U Twelve percent of undergraduate students will go directly to graduate school

after graduation.
U No more than 2 percent of our products sold to customers are defective.

A hypothesis is an assumption
about a population parameter.

In each case, we have made a statement about the
population that may or may not be true. The purpose
of hypothesis testing is to make a statistical conclusion about accepting or not accepting such statements. To further explain this concept, I present the
following story.

I am man enough to admit that I am deathly afraid of snakes. That’s why I did not
hesitate to express my panic when Sam, Debbie’s oldest teenage son, brought home
a snake that he had caught and Debbie wholeheartedly agreed to let him keep in his
bedroom.
Well, my worst nightmare came true the following morning. The snake had pushed
off the top of the cage overnight and was loose somewhere in the house. I guess Sam
never heard the story of the mommy snake that once lifted a Volkswagen Beetle off
her baby snake to save it.
I won’t name names here, but somebody’s wife suggested that we put a mouse in Sam’s
room to attract the snake so we could catch it. I thought this was a very good joke
until a white mouse showed up in Sam’s room later that day posing as “snake bait.”
That night, I lay in bed under high alert (i.e., at least one eye always open and ears
finely tuned for a hissing noise) while Debbie lay calmly snoring next to me.
The next morning, I discovered that I had a new worst nightmare. The mouse had
chewed its way out of its container overnight, and it, too, was loose somewhere in the
house. I now had two wild animals roaming freely in the places where I eat, sleep, and

1VO^bS`#( 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U

watch TV. By this time I’m frantically looking through the phonebook for a motel
that specifically prohibits all snakes and mice. Debbie thought I was “overreacting.”
That night I lay in my bed in the fetal position to protect my vital organs and keep
my arms and legs away from the side of the bed while again Debbie lay calmly snoring
next to me.
Anyway, let’s try to tie this sci-fi tale to hypothesis testing. Let’s say that my hypothesis
is that it will take an average of six days to capture a loose snake in a house. In other
words, I would like to test my belief that the population mean, R, is equal to six days.
I do this by gathering a sample of people who have had a loose snake in their home
and calculate the average number of days required to capture it. Suppose the sample
average is 6.1 days. The hypothesis test will then tell me whether or not 6.1 days is
significantly different from 6.0 days or if the difference is merely due to chance. More
details to follow!

BVS a specific value. In this example,
my alternative hypothesis would be stated as:
H 1 : M x 6.0 days
The following table shows the three valid
combinations of the null and alternative
hypothesis.

The null hypothesis, denoted by
H 0 , represents the status quo
and involves stating the belief
that the mean of the population
is f, =, or v a specific value.
The alternative hypothesis,
denoted by H1 , represents the
opposite of the null hypothesis
and holds true if the null hypothesis is found to be false.

>O`b!( 7\TS`S\bWOZAbObWabWQa
Null Hypothesis

Alternative Hypothesis

H 0 : M 6.0
H 0 : M r 6.0
H 0 : M b 6.0

H 1 : M x 6.0
H 1 : M 6.0
H 1 : M 6.0

Random Thoughts
Some textbooks will use the convention that the null hypothesis will always be stated
as = and will never use f or v. Choosing either method of stating your hypothesis will
not affect the statistical analysis. Just be consistent with the convention you decide to
use.

Note that the alternative hypothesis is never associated with f, =, or v. Selecting the
proper combination is the topic of the next section.

AbObW\UbVSO`b!( 7\TS`S\bWOZAbObWabWQa
Because there are two rejection regions in this figure, we have a two-tail hypothesis test. We will discuss how to determine the boundaries for the rejection regions
shortly.

Wrong Number
The only two statements that we can make about the null hypothesis are that
we …
U Reject the null hypothesis.
U Do not reject the null hypothesis.

Because our conclusions are based on a sample, we will never have enough evidence to accept the null hypothesis. It’s a much safer statement to say that we do not
have enough evidence to reject H 0 . We can use the analogy of the legal system to
explain. If a jury finds a defendant “not guilty,” they are not saying the defendant is
innocent. Rather, they are saying that there is not enough evidence to prove guilt.

=\SBOWZ6g^]bVSaWaBSab
A one-tail hypothesis test involves the alternative hypothesis being stated as < or >. My
golf ball example results in a one-tail test because the alternative hypothesis is being
expressed as H 1 : M 20 and is shown in Figure 15.2.

4WUc`S#
One-tail hypothesis test.

Do Not Reject H0

Reject H0

12
H

Mean Increase in Yards off the Tee

The one-tail hypothesis test
is used when the alternative
hypothesis is being stated as <
or >.

Here, there is only one rejection region, which is the
shaded area on the right tail of the distribution. We
follow the same procedure outlined for the two-tail
test and plot the sample mean, which represents the
average increase in distance from the tee with my
new golf ball. Two possible scenarios exist.

1VO^bS`#( 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U
U If the sample mean falls within the

white region, we do not reject
H 0 . That is, we do not have enough
evidence to support H1, the alternative hypothesis, which states that my
golf ball increased distance off the tee
by more than 20 yards. There goes
my fortune down the drain!
U If the sample mean falls in the rejection

region, we reject H 0 . That is, we have
enough evidence to support H 1, which
confirms my claim that my new golf
ball will increase distance off the tee by
more than 20 yards. Early retirement,
here I come!

Bob’s Basics
For a one-tail hypothesis
test, the rejection region will
always be consistent with the
direction of the inequality for
H1 . For H1 : M 20 , the rejection region will be in the right
tail of the sampling distribution.
For H1 : M 20, the rejection
region will be in the left tail.

Now that we have covered the basics of hypothesis testing, we need to consider errors
that can occur due to sampling.

Bg^S7O\RBg^S773``]`a
Remember that the purpose of the hypothesis test is to verify the validity of a claim
about a population based on a single sample. Because we are relying on a sample, we
expose ourselves to the risk that our conclusions about the population will be wrong.
Using the golf ball example, suppose that my sample falls within the “Reject H 0 ” region
of the last figure. That is, according to the sample, my golf ball increases distance off
the tee by more than 20 yards. But what if the true population mean is actually much
less than 20 yards? This can occur primarily because of sampling error, which I discussed in Chapter 12. This type of error, when we reject H 0 when in reality it’s true,
is known as a Type I error. The probability of making a Type I error is known as F, the
level of significance, which I first introduced in Chapter 14.
We also can experience another type of error with hypothesis testing. Let’s say the
golf ball sample fell within the “Do Not Reject H 0 ” region of the last figure. That
is, according to the sample, my golf ball does not increase the distance off the tee by
more than 20 yards. But what if the true population mean is actually much more than
20 yards? This type of error, when we do not reject H 0 when in reality it’s false, is
known as a Type II error. The probability of making a Type II error is known as G.

>O`b!( 7\TS`S\bWOZAbObWabWQa
The following table summarizes the two types of hypothesis errors.

Reject H 0

H 0 Is True

H 0 Is False

Type I Error

Correct Outcome

P[Type I Error] = F
Do Not Reject H 0

Correct Outcome

Type II Error
P[Type II Error] = G

A Type I error occurs when the
null hypothesis is not accepted
when in reality it is true. A Type
II error occurs when we fail to
reject the null hypothesis when
in reality it is not true.

Normally, with hypothesis testing, we decide on a
value for F that is somewhere between 0.01 and 0.10
before we collect the sample. The value of G can then
be calculated, but that topic goes beyond the scope of
this book. Be grateful for this because that concept is
very complicated!
Let’s put these concepts to work now and do some
real hypothesis testing!

Random Thoughts
Ideally, we would like the values of F and G to be as small as possible. However,
for a given sample size, reducing the value of F will result in an increase in the value
of G. The opposite also holds true. The only way to reduce both F and G simultaneously is to increase the sample size. Once the sample size has been increased to the
size of the population, the values of F and G will be 0. However, as we discussed in
Chapter 12, this is not a recommended strategy.

3fO[^ZS]TOBe]BOWZ6g^]bVSaWaBSab
I stated the hypotheses for the snake example as:
H 0 : M 6.0 days
H 1 : M x 6.0 days
Where R = the mean number of days to catch a loose snake in a home.

1VO^bS`#( 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U

Let’s say that I know that the standard deviation of the population, X, is 0.5 days, and
my sample size to test the hypothesis, n, is 30 homes. (Please don’t ask me how I’m
going to find 30 homes with loose snakes. I’m making this up as I go along, so just
humor me.) We’ll also set F = 0.05, which means I’m willing to accept a 5 percent
chance of committing a Type I error. Our first step is to calculate the standard error of
the mean, S x . If you remember from Chapter 13, the equation is:
0.50
S

0.0913 days
n
30
Let’s assume the sample mean from the 30 homes is 6.1 days. What is our conclusion
about our estimate of the population mean, R?
Sx

To answer this, we next have to determine the critical z-score, which corresponds to
F = 0.05. Because this is a two-tail test, this area needs to be evenly divided between
both tails, with each tail receiving A 0.025. According to Figure 15.3, we need to
2
find the critical z-score that corresponds to the area 0.950 + 0.025 = 0.975. As you can
see, the 0.950 area is derived from 1 – F.

4WUc`S#!
Critical z-score for F = 0.05.

0.950

2 = 0.025

2 = 0.025
1–

-1.96

+1.96

Using Table 3 in Appendix B, we look for the closest value to 0.9750 in the body of
the table. We can find this value by looking across column 1.9 and down row 0.06 to
arrive at the z-score of +1.96 for the right tail and –1.96 for the left tail.

CaW\UbVSAQOZS]TbVS=`WUW\OZDO`WOPZS
Now let’s determine the rejection region using the scale of the original variable, which
in this case is the number of days. To calculate the upper and lower limits of the rejection region, we use the following equations. Recall from Chapter 14 that we use the
z-scores from the standard normal distribution when n r 30 and X is known.
Limits of rejection region = M H 0

zc S x

>O`b!( 7\TS`S\bWOZAbObWabWQa
where M H 0 = the population mean assumed by the null hypothesis.
For our snake example:
Upper limit = M H 0

zc S x 6.0

Lower limit = M H 0

zc S x 6.0

1.96 0.0913 6.18 days
1.96 0.0913 5.82 days

Because our sample mean is 6.1 days, this falls within the “Do Not Reject H 0 ” region
as shown in Figure 15.4. Our conclusion is that the difference between 6.1 days and
6.0 days is merely due to chance variation, and we have support that the population
mean is 6 days.

4WUc`S#"
Hypothesis test for the snake
example (original variable
scale).
Reject H0

Do Not Reject H0

5.82

6.0
H

6.1
x

Reject H0

6.18

Mean Number of Days to Catch a Snake

CaW\UbVSAbO\RO`RWhSR<]`[OZAQOZS
We can arrive at the same conclusion by setting up the boundaries for the rejection
region using the standardized normal scale. We do this by calculating the z-score that
corresponds to the sample mean as follows:
z

x M H0
Sx

6.1 6.0
1.09
0.0913

Bob’s Basics
Be sure to distinguish between the calculated z-score and the critical z-score. The
calculated z-score, z, represents the number of standard deviations between the
sample mean and MH0 , the population mean according to the null hypothesis. The
critical z-score, z c , is based on the significance level, F, and determines the boundary for the rejection region.

1VO^bS`#( 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U

Figure 15.5 shows this result graphically. Because the calculated z-score of +1.09 is
within the “Do Not Reject H 0 ” region, the conclusions of both techniques are consistent.

4WUc`S##
Hypothesis test for the snake
example (standardized scale).

Reject H0

Do Not Reject H0

-1.96
–zc

Reject H0

+1.09 +1.96
z
+zc

Number of Standard Deviations from the Mean

3fO[^ZS]TO=\SBOWZ6g^]bVSaWaBSab
Because I formulated the alternative hypothesis for the golf ball example as > 20, this
becomes a one-tail test. The hypothesis for this example is stated as:
H 0 : M b 20 yards
H 1 : M 20 yards
Where R = the mean increase in yards off the tee using my new golf ball.
Let’s say that I know that the standard deviation of the population, X, is 5.3 yards and
my sample size to test the hypothesis, n, is 40 golfers. For this example, we’ll set
F = 0.01. The standard error of the mean, S x , will now be equal to:
Sx

S
n

5.3
0.838 yards
40

Let’s assume the sample mean from the 40 golfers is 22.5 yards. What is our conclusion about our estimate of the population mean, R?
Once again, we next have to determine the critical z-score, which corresponds to F =
0.01. Because this is a one-tail test, this entire area needs to be in one rejection region
on the right side of the distribution. According to Figure 15.6, we need to find the zscore that corresponds to the area 0.99 or 1 – F.

>O`b!( 7\TS`S\bWOZAbObWabWQa
Using Table 3 in Appendix B, we look for the closest value to 0.9900 in the body of
the table, which results in a critical z-score of 2.33.

4WUc`S#$
Critical z-score for F = 0.01.

0.99

= 0.01

+2.33
+zc

Number of Standard Deviations from the Mean

To calculate the limit for this rejection region using the scale of the original variable,
we use:
Limit = M H 0

zc S x 20

2.33 0.838

21.95 yards

Because our sample mean is 22.5 yards, this falls within the “Reject H 0 ” region as
shown in Figure 15.7. Our conclusion is that we have enough evidence to support the
hypothesis that the mean increase in distance off the tee with my new balls exceeds 20
yards. I’m in business!

4WUc`S#%
Hypothesis test for the golf
ball example (original variable scale).
Do Not Reject H0

20
H

Reject H0

21.95

Mean Increase in Distance off the Tee in Yards

x = 22.5

1VO^bS`#( 7\b`]RcQbW]\b]6g^]bVSaWaBSabW\U

Random Thoughts
You might be asking yourself, “If the sample mean was 21.0 yards, shouldn’t that
provide conclusive evidence that the new ball increases distance by more than 20
yards?” According to the previous figure, the answer is no. Because we are basing our decision on a sample, an average of 21 is just too close to 20 to satisfy my
claim. The sample average would have to be 21.95 yards or more in order to reject
the null hypothesis.

As I mentioned earlier, the purpose of this chapter was to introduce the basic concepts
of hypothesis testing. The following two chapters will explore hypothesis testing in
even more loving detail. So hang in there—we’re just getting warmed up!

G]c`Bc`\
1. Formulate a hypothesis statement for the following claim: “The average adult
drinks 1.7 cups of coffee per day.” A sample of 35 adults drank an average of
1.95 cups per day. Assume the population standard deviation is 0.5 cups. Using
F = 0.10, test your hypothesis. What is your conclusion?
2. Formulate a hypothesis statement for the following claim: “The average age of
our customers is less than 40 years old.” A sample of 50 customers had an average age of 38.7 years. Assume the population standard deviation is 12.5 years.
Using F = 0.05, test your hypothesis. What is your conclusion?
3. Formulate a hypothesis statement for the following claim: “The average life of
our light bulbs is more than 1,000 hours.” A sample of 32 light bulbs had an
average life of 1,190 hours. Assume the population standard deviation is 325
hours. Using F = 0.02, test your hypothesis. What is your conclusion?
4. Formulate a hypothesis statement for the following claim: “The average delivery
time is less than 30 minutes.” A sample of 42 deliveries had an average time of
26.9 minutes. Assume the population standard deviation is 8 minutes. Using F =
0.01, test your hypothesis. What is your conclusion?
5. Formulate a hypothesis statement for the following claim: “Students graduating
from college have an average credit card debt of $2,700.” A sample of 40 college
graduates averaged $2,450 in credit card debt. Assume the population standard
deviation is $950. Using A 0.05, test your hypothesis. What is your conclusion?

>O`b!( 7\TS`S\bWOZAbObWabWQa

BVS:SOabG]c.
U A Type I error occurs when the null hypothesis is rejected when, in reality, it is

true. The probability of this error occurring is known as F, the level of significance.
U A Type II error occurs when the null hypothesis is accepted when, in reality, it is

not true. The probability of this error occurring is known as G.

1VO^bS`

6g^]bVSaWaBSabW\UeWbV=\S
AO[^ZS
7\BVWa1VO^bS`
U Testing the mean of a population using a large and small sample
U Examining the role of alpha (F) in hypothesis testing
U Using the p-value to test a hypothesis
U Testing the proportion of a population using a large sample

In Chapter 15, I introduced the concept of hypothesis testing to whet your
appetite. I have devoted this chapter to hypothesis testing that involves only
one population, whereas in Chapter 17 I will discuss testing that compares
two different populations to each other.
Hypothesis testing involving one population focuses on confirming claims
such as the population average is equal to a specific value. We will consider many different cases with this type of hypothesis testing in the following sections. This chapter relies on many of the concepts we explored
in Chapters 14 and 15, so be sure you are comfortable with that material
before you dive into this chapter.

>O`b!( 7\TS`S\bWOZAbObWabWQa

6g^]bVSaWaBSabW\UT]`bVS;SO\eWbV:O`USAO[^ZSa
When the sample size we use to test our hypothesis is large (n v 30), we can rely on
our old friend the central limit theorem which we met in Chapter 13. However, we
still have two cases to consider—whether X, the population standard deviation, is
known or unknown.

EVS\AWU[O7a9\]e\
To demonstrate this type of hypothesis test, I’ll use the following story.
One of the most feared phrases a husband can hear from his wife is, “Honey, let’s go
on a diet together.” I should have been suspicious of Debbie’s motives when she suggested we go on the low-carbohydrate diet, especially because she wears size 2 pants.
But I guess I could stand to lose a few pounds, so in a weak moment, I agreed. After
all, I figured we could turn this into a competition to make things more interesting.
After a few harrowing days without my beloved carbohydrates (who would have
guessed a grown man could dream about Cheez-its night after night), I began to wonder how Debbie was doing so well with the diet. I found the answer to this mystery
hidden deep in the trunk of her car—a half-eaten box of cinnamon rolls. I guess that
makes me the winner. The thrill of victory!
Anyway, let’s say that this particular diet claims that the average age of the person
who participates in this self-inflicted torture is less than 40 years old. We set up our
hypothesis as follows:
H 0 : M r 40 years old
H 1 : M 40 years old
We sample 60 people on the diet and find that their average age is 35.7 years. Given
that X, the population standard deviation, is 16 years, we’ll test the hypothesis at
A 0.05.

Bob’s Basics
Remember from Chapter 15 that F, the level of significance, represents the probability of making a Type I error. A Type I error occurs when we reject H0, when H0
is actually true. In this case, a Type I error would mean that we believe the claim
that the average person on the diet is less than 40 years old when, in reality, the
claim is not true. For this example there’s a 5 percent chance of this error happening.

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

Because the sample size is greater than 30 and we know the value of X, we calculate
the z-score from the standardized normal distribution as we did in Chapter 15.
z

x M H0

Sx
For our example, the standard error of the mean, S x , would be:
16
S

2.07 years
Sx
n
60
This results in a calculated z-score of:
z

M H0
Sx

35.7 40
2.08
2.07

Also recall from Chapter 15, the critical z-score, which defines the boundary for the
rejection region, is –1.64 for a one-tail (left side) test with F = 0.05. Figure 16.1 shows
this test graphically.

4WUc`S$
0.95
= 0.05
Reject H0

One-tail hypothesis test for
the diet example (standardized scale).

1–
Do Not Reject H0

-2.08 -1.64
z zc

Number of Standard Deviations from the Mean

As you can see in the figure, the calculated z-score of –2.08 falls within the “Reject
H0” region, which allows us to conclude that the claim that the average age of those
on this diet is less than 40 years old. I knew I was too old for this diet! In general,
we reject H0 if z zc , where z means the “absolute value of z.” For instance,
2.08 2.08.

EVS\AWU[O7aC\Y\]e\
Many times, we just don’t have enough information to know the value of X, the population standard deviation. However, as long as our sample size is 30 or more, we can
substitute s, the sample standard deviation for X. To illustrate this technique, let’s use
the following example.

>O`b!( 7\TS`S\bWOZAbObWabWQa
I don’t know about you, but it seems I spend too much time on the phone waiting
on hold for a live customer service representative. Let’s say a particular company has
claimed that the average time a customer waits on hold is less than five minutes. We’ll
assume we do not know the value of X. The following table represents the wait time in
minutes for a random sample of 30 customers.
Wait Time in Minutes

6.2 3.8 1.3 5.4 4.7 4.4 4.6 5.0 6.6 8.3
3.2 2.7 4.0 7.3 3.6 4.9 0.5 2.9 2.5 5.6
5.5 4.7 6.5 7.1 4.4 5.2 6.1 7.4 4.8 2.9
Using Excel, we can determine that x 4.74 minutes and s = 1.82 minutes. At first
glance, it appears the company’s claim is valid. But let’s put it through a hypothesis
test with A 0.02 to be sure.
State the hypothesis as:
H 0 : M r 5.0 minutes
H 1 : M 5.0 minutes
From Chapter 15, we know that the critical z-score for a one-tail (left side) hypothesis
test with F = 0.02 is –2.05.
As we did earlier in Chapter 14, we can approximate the standard error of the mean
by:
s
1.82

0.332 minutes
Sˆ x
n
30
Our calculated z-score using this particular sample would be:
x M H 0 4.74 5.0
z

0.78
0.332
Sˆ x
Figure 16.2 shows this test graphically.
According to our figure, we do not reject the null hypothesis. In other words, we do
not have enough evidence from this sample to support the company’s claim that the
average wait on hold is less than five minutes. Even though the sample average is
actually less than five minutes (4.74), it is too close to five minutes to say there is a
difference between the two values. Another way to state this is to say: “The difference
between 4.74 and 5.0 is not statistically significant in this case.”

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

4WUc`S$
0.98
1–
= 0.02
Reject H0

-2.05
zc

One-tail hypothesis test for
waiting on hold example
(standardized scale).

Do Not Reject H0

-0.78
z

Number of Standard Deviations from the Mean

BVS@]ZS]T/Z^VOW\6g^]bVSaWaBSabW\U
For all the examples in these last two chapters, I have just stated a value for F, the
level of significance. I’m sure you’re wondering what impact changing the value of F
will have on the hypothesis test. Great question!
Suppose that I am making a claim that the average grade for a person using this book
will be more than an 87. (I’m not really making this claim, so don’t get too excited!) I
would state the hypothesis test as follows:
H 0 : M b 87
H 1 : M 87
Now, it would be in my best interest if I could reject H0, which would validate my
claim. I can do so by choosing a fairly high value for F, say 0.10. This corresponds to
a critical z-score of +1.28, because we are using the right tail of a one-tail hypothesis
test.
Let’s say that X, the population standard deviation, is 12 and my sample mean is 90.6,
which was taken from a sample size of 32 students. For this example, the standard
error of the mean, S x , would be:
Sx

12
S

2.12
n
32

>O`b!( 7\TS`S\bWOZAbObWabWQa
This results in a calculated z-score of:
z

M H0
Sx

90.6 87
1.70
2.12

According to Figure 16.3, I have achieved my goal of rejecting H0, because the calculated z-score is within the shaded region. My book appears to have done the trick!

4WUc`S$!
Hypothesis test for grade
example, F = 0.10.
0.90
1–
= 0.10
Reject H0
Do Not Reject H0
0

1.28 1.7
zc
z

Number of Standard Deviations from the Mean

However, I must admit, I chose a pretty “wimpy” value of F = 0.10 in an effort to help
prove my claim. In this case, I am willing to accept a 10 percent chance of a Type I
error. A more impressive test would be to set alpha lower, say F = 0.01. Now that’s a
“real man’s alpha.” The level of significance corresponds to a critical z-score of +2.33.
Figure 16.4 shows the impact of this change.

4WUc`S$"
Hypothesis test for grade
example, F = 0.01.

0.99
1–
= 0.01
Reject H0

Do Not Reject H0

1.7
z

2.33
zc

Number of Standard Deviations from the Mean

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

As you can see, to my horror, the shaded region no longer includes my calculated
z-score of +1.7. Therefore, I do not reject H0 and cannot claim the average grade of
those using my book exceeds an 87. In general, a hypothesis test that rejects H0 is most
impressive with a low value of F.

7\b`]RcQW\UbVS]DOZcS
Just when you thought it was safe to get back in the water, along comes another shark!
This is the perfect opportunity to throw another concept at you. You might feel like
grumbling a little right now, but in the end
you’ll be thanking me.
The p-value is the smallest level of significance at which the null hypothesis will be
rejected, assuming the null hypothesis is
true. The p-value is sometimes referred to
as the observed level of significance. I know this
may sound like a lot of mumbo-jumbo right
now, but an illustration will help make this
clear.

The observed level of significance is the smallest level
of significance at which the
null hypothesis will be rejected,
assuming the null hypothesis is
true. It is also known as the pvalue.

BVS]DOZcST]`O=\SBOWZBSab
Using the previous grade example (over 87 if using this book), the p-value is represented by the shaded area to the right of the calculated z-score of +1.7. This is shown
in Figure 16.5.

4WUc`S$#
p-value for the grade
example.
0.9554
p – value
0.0446

Do Not Reject H0

1.7
z

2.33
zc

Number of Standard Deviations from the Mean

>O`b!( 7\TS`S\bWOZAbObWabWQa
Bob’s Basics
Recall that P[z 1.7] 1 P[z b 1.7] 1 0.9554 0.0446 . See Chapter 11
if you need a refresher on using the standardized normal z table.

Using our standardized normal z table (Table 3 in Appendix B), we can confirm that
the shaded area in the right tail is equal to P[ z 1.7] 0.0446.

Bob’s Basics
We can use the p-value to
determine whether or not to
reject the null hypothesis. In
general …
U If p -value f F, we reject

the null hypothesis.
U If p -value # F, we do not

reject the null hypothesis.

Because our p-value of 0.0446 is more than the value
of F (set at 0.01), we do not reject H0. Most statistical
software packages (including Excel) provide p-values
with the analysis.
Another way to describe this p-value is to say, in a
very scholarly voice, “Our results are significant at
the 0.0446 level.” This means that as long as the
value of F is 0.0446 or larger, we will reject H0 , which
is normally good news for researchers trying to validate their findings.
Calculating the p-value for a two-tail hypothesis test
is slightly different, and I’ll show you how in the next
section.

BVS]DOZcST]`OBe]BOWZBSab
Recall that you use a two-tail hypothesis test when the null hypothesis is stated as an
equality. For example, let’s test a claim that states the average number of miles driven
by a passenger vehicle in a year equals 11,500 miles. I have serious reservations about
this claim after spending half the day being a taxi driver to the kids. We would state
the hypotheses as follows:
H0 : R" 11,500 miles
H1 : R| 11,500 miles
Let’s assume X " 3000 miles, and we want to set F " 0.05. We sample 80 drivers and
determine the average number of miles driven is 11,900. What is our p-value, and
what do we conclude about the hypothesis?

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

For this example, the standard error of the mean, S x , would be:
Sx

3000
S

335.41 miles
n
80

This results in a calculated z-score of:
z

M H0
Sx

11, 900 11, 500
1.19
335.41

The critical z-score for a two-tail test with F" 0.05 is t 1.96. The shaded area in
Figure 16.6 shows the p-value for this test.
p-Value for a Two-Tail Hyphothesis Test

4WUc`S$$
p-value for the miles driven
per year example.

0.766

p-Value Equals the Sum
of the Shaded Regions
0.117

p-Value = 0.117 + 0.117
= 0.234
0.117

-1.96 -1.19
zc

+1.19 +1.96
z
zc

0.8830

According to Table 3 in Appendix B, the P[ z b 1.19 ] 0.8830. This means the shaded
region in the right tail of Figure 16.6 is P[ z 1.19 ] 1 0.8830 0.117. Because this
is a two-tail test, we need to double this area to arrive at our p-value. According to our
figure, the p-value is the total area of both shaded regions, which is 2 s 0.117 0.234 .
Because p A , we do not reject the null hypothesis. Our data supports the claim that
the average number of miles driven per year by a passenger vehicle is 11,500.
In general, the smaller the p-value, the more confident we are about rejecting the null
hypothesis. In most cases a researcher is attempting to find support for the alternative
hypothesis. A low p-value provides support that brings joy to his or her heart.

>O`b!( 7\TS`S\bWOZAbObWabWQa

6g^]bVSaWaBSabW\UT]`bVS;SO\eWbVA[OZZAO[^ZSa
Recall, from Chapter 14, that with a small sample size, we lose the use of the central
limit theorem, so, therefore, we need to assume that the population is normally distributed for all cases in this section. The first case that we’ll examine is when we know
X, the population standard deviation.

EVS\AWU[O7a9\]e\
When X is known, the hypothesis test reverts back to the large sample size case. We
can do this because we are now assuming the population is normally distributed. We
can demonstrate this method with the following example.
Opening up my monthly cell phone bill lately has become a nerve-wracking experience. As I warily open the envelope, I wonder what surprises await me. With several
users on our family “share plan,” I can often count on somebody having discovered a
new feature that has nothing to do with talking to another person on the phone and
having used this new-found discovery over and over and over again. Occasionally, after
digging through countless pages full of numbers and codes, I breathe a sigh of relief
and say a silent prayer of thanks. Most months, however, I end up clutching my chest
and screaming “AIEEEEEEEE!” It’s like playing a subtle form of Russian roulette
with the phone company.
Anyway, let’s say the phone company claims that the average monthly cell phone bill
for their customers is $92 (I wish). We can test this claim by stating our hypothesis as:
H0 : R" $92
H1 : R| $92

Bob’s Basics
Recall from Chapter 14 that
because we know X and
we assumed the population
is normally distributed, we can
use the z-scores from the normal
probability distribution to test
this hypothesis.

We’ll assume that X " $22.50 and that the population is normally distributed. We select 18 phone bills
randomly and determine the sample average equals
$107. Using F " 0.02, what do we conclude?
For this example, the standard error of the mean,
S x , would be:
Sx

$22.50
S

$5.30
n
18

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

This results in a calculated z-score of:
z

M H0

$107 $92
2.83
$5.30

The critical z-score for a two-tail test with F " 0.02 is t 2.33. Figure 16.7 shows this
test graphically.
As you can see in Figure 16.7, the calculated z-score of +2.83 is with the “Reject H0”
region. We, therefore, conclude that the average cell phone bill is not equal to $92. I
didn’t think so!

4WUc`S$%
Hypothesis test for cell phone
bills.

0.98
1–

/2 = 0.01

Reject H0

Do Not Reject H0

-2.33
zc

/2 = 0.01
Reject H0

+2.33
zc

+2.83
z

Number of Standard Deviations from the Mean

EVS\AWU[O7aC\Y\]e\
As we did in Chapter 14, when X is unknown for a small sample size taken from a
normally distributed population, we use the Student’s t-distribution. This particular
distribution allows us to substitute s, the sample standard deviation for X.
As an example, suppose my son John claims his average golf score is less than 88. Not
to be one to doubt him, I can test this claim with the following hypothesis:
H0 : Rv 88
H1 : R! 88
We will assume that we do not know X and that John’s scores follow a normal distribution. The following represents a random sample of 10 golf scores from John.

>O`b!( 7\TS`S\bWOZAbObWabWQa

8]V\Âa5]ZTAQ]`Sa
86

Using Excel, we can determine that x 86.3 and s = 2.58 for this sample. Recall from
Chapter 14, we can approximate the standard error of the mean using the following
equation:
s
2.58

0.816
Sˆ x
n
10
We can then determine the calculated t-score using the following equation:
t

M H 0 86.3 88

2.08
0.816
Sˆ x

We’ll test this hypothesis using F " 0.05. To find the corresponding critical t-score, we
use Table 4 from Appendix B. Here is an excerpt of this table.

AbcRS\bÂab2Wab`WPcbW]\BOPZS
Selected right-tail areas with confidence levels underneath
Alpha
0.2000 0.1500 0.1000 0.0500 0.0250 0.0100 0.0050
Conf lev 0.6000 0.7000 0.8000 0.9000 0.9500 0.9800 0.9900
d.f.

0.0010
0.9980

0.0005
0.9990

1.376

1.963

3.078

6.314

12.706 31.821 63.657

318.31

636.62

1.061

1.386

1.886

2.920

4.303

6.965

9.925

22.327

31.599

0.978

1.250

1.638

2.353

3.182

4.541

5.841

10.215

12.924

0.941

1.190

1.533

2.132

2.776

3.747

4.604

7.173

8.610

0.920

1.156

1.476

2.015

2.571

3.365

4.032

5.893

6.869

0.906

1.134

1.440

1.943

2.447

3.143

3.707

5.208

5.959

0.896

1.119

1.415

1.895

2.365

2.998

3.499

4.785

5.408

0.889

1.108

1.397

1.860

2.306

2.896

3.355

4.501

5.041

0.883

1.100

1.383

1.833

2.262

2.821

3.250

4.297

4.781

0.879

1.093

1.372

1.812

2.228

2.764

3.169

4.144

4.587

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

You recall from Chapter 14, we need to determine the number of degrees of freedom,
which is equal to n 1 10 1 9 for this example. Because this is a one-tail (left side)
test, we look under the A 0.05 column resulting in a critical t-score, t c , equal to
–1.833, which is underlined. Figure 16.8 shows this test graphically.

4WUc`S$&
Hypothesis test for John’s golf
scores.
0.95
1–
= 0.05
Reject H0

Do Not Reject H0

0
t = –2.08

tc = –1.833

As we can see in the figure, the calculated
t-score of –2.08 falls within the shaded
“Reject H0” region. Therefore, we can
conclude that John’s average golf score is
indeed lower than 88. So that explains why
he usually beats me! In general, we reject H0
if t t c .
Let’s take another example to demonstrate
a two-tail hypothesis test using the tdistribution. I would like to test a claim
that the average speed of cars passing a
specific spot on the interstate is 65 miles
per hour. We can express the hypothesis
test as follows:
H0 : R" 65 miles per hour
H1 : R| 65 miles per hour
We will assume that we do not know X and
that speeds follow a normal distribution.
The following represents a random sample
of the speed of seven cars.

Bob’s Basics
Because John’s golf score
example is a one-tail test on
the left side of the distribution, we use a negative critical
t-score. Had this been a one-tail
test on the right side, we would
use a positive critical t-score.

Bob’s Basics
It is not possible to determine
the p-value for a hypothesis
test when using the Student’s
t-distribution table in Appendix
B. However, most statistical software will provide the p-value as
part of the standard analysis.
We’ll see this in later chapters
as we use Excel.

>O`b!( 7\TS`S\bWOZAbObWabWQa

1O`A^SSRa
62

Using Excel, we can determine that x 66.9 mph and s = 4.16 mph for this sample.
We can approximate the standard error of the mean:
Sˆ x

s
n

4.16
7

1.57 mph

We can then determine the calculated t-score:
t

M H0
Sˆ x

66.9 65
1.21
1.57

We’ll test this hypothesis using F " 0.05. To find the corresponding critical t-score, we
use Table 4 from Appendix B. Here is an excerpt of this table.

AbcRS\bÂab2Wab`WPcbW]\BOPZS
Selected right-tail areas with confidence levels underneath
Alpha
0.2000 0.1500 0.1000 0.0500 0.0250
Conf lev 0.6000 0.7000 0.8000 0.9000 0.9500
d.f.

0.0100 0.0050
0.9800 0.9900

0.0010
0.9980

0.0005
0.9990

1.376

1.963

3.078

6.314

12.706

31.821 63.657

318.31

636.62

1.061

1.386

1.886

2.920

4.303

6.965

9.925

22.327

31.599

0.978

1.250

1.638

2.353

3.182

4.541

5.841

10.215

12.924

0.941

1.190

1.533

2.132

2.776

3.747

4.604

7.173

8.610

0.920

1.156

1.476

2.015

2.571

3.365

4.032

5.893

6.869

0.906

1.134

1.440

1.943

2.447

3.143

3.707

5.208

5.959

0.896

1.119

1.415

1.895

2.365

2.998

3.499

4.785

5.408

The number of degrees of freedom for this example equals n – 1 = 7 – 1 = 6. Because
this is a two-tail test, we need to divide F " 0.05 into two equal portions, one on
the right side of the distribution, the other on the left. We then look under the
A 0.025 column resulting in a critical t-score, t , equal to t 2.447, which is underc
2
lined. This test is shown graphically in Figure 16.9.

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

4WUc`S$'

Hypothesis Test for Car Speeds
(Two Tail t-Distribution)

Hypothesis test for car speed.

0.95
1–

2 = 0.025
Reject H0

-2.447
tc

2 = 0.025
Reject H0

Do Not Reject H0

1.21
t

+2.447
tc

As we can see in the figure, the calculated t-score of +1.21 falls within the “Do Not
Reject H0” region. Therefore, we can conclude that the average speed past this spot on
the interstate averages 65 miles per hour.

CaW\U3fQSZÂaB7O`b!( 7\TS`S\bWOZAbObWabWQa
Cell A1 contains the Excel formula =TINV(0.05, 6) with the result being 2.447. This
probability is underlined in the previous table.
A one-tail test requires a slight modification. We need to multiply the probability in
the TINV function by two because this parameter is based on a two-tail test. Figure
16.11 shows the TINV function being used to determine the critical t-score for F "
0.05 and d.f. = 9 from our earlier one-tail test example with John’s golf scores.

4WUc`S$
Excel’s TINV function for a
one-tail test.

Cell A1 contains the Excel formula =TINV(2*0.05, 9) with the result being 1.833.
This is consistent with the result from our previous example.

6g^]bVSaWaBSabW\UT]`bVS>`]^]`bW]\eWbV:O`US
AO[^ZSa
You can perform hypothesis testing for the proportion of a population as long as the
sample size is large enough. Recall from Chapter 13, that proportion data follows the
binomial distribution, which can be approximated by the normal distribution under
the following conditions:
np r 5 and nq r 5
where:
p = the probability of a success in the population
q = the probability of a failure in the population (q = 1 – p)
We will examine both one-tail and two-tail hypothesis testing for the proportion in
the following sections.

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

=\SBOWZ6g^]bVSaWaBSabT]`bVS>`]^]`bW]\
Let’s say we would like to test the hypothesis that more than 30 percent of U.S.
households have Internet access. We would state the hypothesis as:
H0 : pv 0.30
H1 : p! 0.30
where p = the proportion of U.S. households with Internet access.
We collect a sample of 150 households and
find that 38 percent of these have Internet
access. What can we conclude at the
F " 0.05 level?

Wrong Number
Be careful not to confuse
this definition of p with the
p-value that we talked about
earlier.

Our first step is to calculate S p , the standard error of the proportion, which was
described in Chapter 13 using the following
equation:
Sp

pH 0 1 pH 0
n

where pH 0 = the proportion assumed by the null hypothesis. For our example:
Sp

0.30 1
150

0.30

0.37

Next, we can determine the calculated z-score using:
z

pH 0
Sp

where p = the sample proportion. For our example:
z

pH 0
Sp

0.38 0.30
2.16
0.037

The critical z-score for a one-tail test with F " 0.05 is +1.64. This hypothesis test is
shown graphically in Figure 16.12.

>O`b!( 7\TS`S\bWOZAbObWabWQa

4WUc`S$
Hypothesis test for the
Internet access example.
0.95
1– a
a = 0.05
Reject H0

Do Not Reject H0

1.64 2.16
zc
z

Number of Standard Deviations from the Mean

As you can see in Figure 16.12, the calculated z-score of +2.16 is within the “Reject
H0” region. Therefore, we conclude that the proportion of U.S. households with
Internet access exceeds 30 percent.
We can show the p-value for this test graphically in Figure 16.13.

4WUc`S$!
p-value for the Internet access
example.
0.9846
p-Value
0.0154

2.16
z

Number of Standard Deviations from the Mean

Using our standardized normal z table (Table 3 in Appendix B), we can confirm that
the shaded area in the right tail is equal to:
P[ z 2.16 ] 1 P[ z b 2.16 ]
P[ z 2.16 ] 1 0.9846 0.0154
Therefore, our results are significant at the 0.0154 level. As long as F v 0.0154, we
will be able to reject H0.

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

Be]BOWZ6g^]bVSaWaBSabT]`bVS>`]^]`bW]\
We’ll wrap this chapter up with one final two-tail example. Here, we want to test a
hypothesis for a company that claims 50 percent of their customers are of the male
gender. We state our hypothesis as:
H0 : p" 0.50
H1 : p| 0.50
We randomly select 256 customers and find that 47 percent are male. What can we
conclude at the F " 0.05 level?
We need to determine S p , the standard error of the proportion:

pH 0 1 pH 0

0.50 1

0.50

256

0.0312

Next, we can determine the calculated z-score:
z

pH 0
Sp

0.47 0.50
0.96
0.0312

The critical z-score for a two-tail test with
F " 0.05 is t 1.96. This hypothesis test is
shown graphically in Figure 16.14.
As you can see in Figure 16.14, the calculated z-score of –0.86 is within the “Do Not
Reject H0” region. There, we conclude that
the proportion of male customers is equal to
50 percent for this company.

Bob’s Basics
In general, we reject H0 if
z z c or t t c .
Also, we do not reject H0 if
z b z c or t b t c .

4WUc`S$"
Hypothesis test for the percentage of males example.
0.95
1– a

a 2 = 0.025
Reject H0

Do Not Reject H0

-1.96
zc

-0.86
z

a 2 = 0.025
Reject H0

+1.96
zc

Number of Standard Deviations from the Mean

>O`b!( 7\TS`S\bWOZAbObWabWQa
Figure 16.15 graphically shows the p-value for this test.

4WUc`S$#
p-value for the percentage of
males example.
p-Value Equals the Sum
of the Shaded Regions

p-Value = 0.1685 + 0.1685
= 0.337

0.663

0.1685
-1.96
zc

0.1685

-0.86
z

+0.86

+1.96
zc

0.8315

Using our standardized normal z table (Table 3 in Appendix B), we can confirm that
the shaded area in the left tail is equal to:
P[ z b 0.86 ] 1 P[ z b 0.86 ]
P[ z b 0.86 ] 1 0.8315 0.1685
Because this is a two-tail test, the p-value would be 2 s 0.1685 0.337 , which represents the total area in both shaded regions.

G]c`Bc`\
1. Test the claim that the average SAT score for graduating high school students
is equal to 1100. A random sample of 70 students was selected, and the average
SAT score was 1035. Assume X " 310 and use F " 0.10. What is the p-value for
this sample?
2. A student organization at a small business college claims that the average class
size is greater than 35 students. Test this claim at F " 0.02, using the following
sample of class size:
42

Assume the population is normally distributed and that X is unknown.

1VO^bS`$( 6g^]bVSaWaBSabW\UeWbV=\SAO[^ZS

3. Test the claim that the average gasoline consumption per car in the United
States is more than 7 liters per day. (We’re going metric here!) Use the random
sample here, which represents daily gasoline usage for one car:
9

Assume the population is normally distributed, and that X is unknown. Use
F " 0.05 and determine the p-value for this sample.
4. Test the claim that the proportion of Republican voters in a particular city is less
than 40 percent. A random sample of 175 voters was selected and found to consist of 30 percent Republicans. Use F " 0.01 and determine the p-value for this
sample.
5. Test the claim that the proportion of teenage cell phone users exceeding their
allotted monthly minutes equals 65 percent. A random sample of 225 teenagers
was selected and found to consist of 69 percent exceeding their minutes. Use
F " 0.05 and determine the p-value for this sample.
6. Test the claim that the mean number of hours that undergraduate students work
at a particular college is less than 15 hours per week. A random sample of 60 students was selected, and the average number of working hours was 13.5 hours per
week. Assume X " 5 hours, and use F " 0.10. What is the p-value for this sample?

BVS:SOabG]cO`b!( 7\TS`S\bWOZAbObWabWQa

BVS1]\QS^b]TBSabW\UBe]>]^cZObW]\a
Many statistical studies involve comparing the same parameter, such as a mean,
between two different populations. For example:
U Is there a difference in average SAT scores between males and females?
U Do “long-life” light bulbs really outlast standard light bulbs?
U Does the average selling price of a house in

The sampling distribution
for the difference in means
describes the probability of
observing various intervals for
the difference between two
sample means.

Newark differ from the average selling price for
a house in Wilmington?
To answer such questions, we need to explore a new
sampling distribution. (I promise this will be the last.)
This one has the fanciest name of them all—the sampling distribution for the difference in means. (Dramatic
background music brings us to the edge of our seats.)

AO[^ZW\U2Wab`WPcbW]\T]`bVS2WTTS`S\QSW\;SO\a
The sampling distribution for the difference in means can best be described in Figure
17.1.

4WUc`S%
The sampling distribution for
the difference in means.

Population 1

Population 2

Sampling Distribution for the Mean
(Population 1)

x 1

Sampling Distribution for the Mean
(Population 2)

x 2

Sampling Distribution for the Difference in Means

x 1 – x 2

x 1– x 2

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

As an example, let’s consider testing for a difference in SAT scores for male and female
students. We’ll assign female students as Population 1 and male students as Population
2. Graph 1 in Figure 17.1 represents the distribution of SAT scores for the female
students with mean R1 and standard deviation X1. Graph 2 represents the same for the
male population.
Graph 3 represents the sampling distribution for the mean for the female students.
This graph is the result of taking samples of size n1 and plotting the distribution of
sample means. Recall that we discussed this distribution of sample means back in
Chapter 13. The mean of this distribution would be:
M x 1 M1
This is according to the central limit theorem from Chapter 13. The same logic holds
true for Graph 4 for the male population.
Graph 5 in Figure 17.1 shows the distribution that represents the difference of sample
means from the female and male populations. This is the sampling distribution for the
difference in means, which has the following mean:
M x1

M x1

M x2

In other words, the mean of this distribution, shown in Graph 5, is the difference
between the means of Graphs 3 and 4.
The standard deviation for the Graph 5 is known as the standard error of the difference
between two means and is calculated with:
S x1

S12
n1

S 22
n2

where:
S12 , S 22 = the variance for Populations
1 and 2
n1, n2 = the sample size from
Populations 1 and 2
Now before you pull the rest of your hair
out, let’s put these guys to work in the following section.

The standard error of the difference between two means
describes the variation in the
difference between two sample
means and is calculated using:
S x1

S12
n1

S 22
n2

>O`b!( 7\TS`S\bWOZAbObWabWQa

BSabW\UT]`2WTTS`S\QSa0SbeSS\;SO\aeWbV:O`US
AO[^ZSAWhSa
When the sample sizes from both populations of interest are greater than 30, the
central limit theorem allows us to use the normal distribution to approximate the sampling distribution for the difference in means. Let’s demonstrate this technique with
the following example.
Studies have been done to investigate the effects of stimulation on the brain development of rats. I guess the logic being what’s good for rats can’t be all that bad for us
humans. Two samples were randomly selected from the same rat population.
The first sample, we’ll call these the “lucky rats” (Population 1), was surrounded with
every luxury a rat could imagine. I can envision a country club atmosphere, complete
with a golf course (and tiny golf carts), tennis courts, and a five-star restaurant where
our lucky rats could feast on imported cheese and French wine while they discussed
the state of the rat economy.
The second sample, we’ll call them the “less-fortunate rats” (Population 2), didn’t
have it quite so good. These guys were locked in a barren cage and were forced to eat
Cheez Whiz from a can and watch reruns of reality TV shows. Animal rights activists
protested against this experiment, claiming the involuntary use of Cheez Whiz was
“inhumane.”
After spending three months in each of these environments, the size of each rat brain
was measured by weight for development. I’ll spare you the details as to how this was
done, but I will tell you that Harvey the Rat mysteriously failed to show for his 8 A.M.
tee time. His group went off without him.
The following table summarizes these gruesome findings.

Ac[[O`WhSR2ObOT]`@Ob3f^S`W[S\b
Population

Average Brain
Weight in Grams
_
x

Sample Standard
Deviation
s

Sample
Size
n

Lucky (1)

2.4

0.6

Less-Fortunate (2)

2.1

0.8

For this hypothesis test, we need to assume that the two samples are independent
of each other. In other words, there is no relationship between the rats in the lucky

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

sample and the rats in the less-fortunate sample. The hypothesis statement for this
two-sample test would be as follows:
H0 : R1f R2
H1 : R1# R2
where:
R1 = the mean brain weight of the lucky rat population
R2 = the mean brain weight of the less-fortunate rat population
The hypothesis can also be expressed as:
H0 : R1R2 f 0
H1 : R1R2# 0
The alternative hypothesis supports the claim that the lucky rats will have heavier
brains. Seems to me this could lead to neck problems for these rats—but I’ll leave that
question for another study. We’ll test this hypothesis at the F" 0.05 level.
If X1 or X2 are not known, then we can use s1 or s2, the standard deviation from the
samples of populations 1 and 2 as an approximation, as long as n v 30 for both populations, as shown here:
Ŝ s
With this assumption, we can approximate the standard error of the difference
between two means using:
Sˆ x1

Sˆ 12
n1

Sˆ 22
n2

Because we do not know X1 or X2 in our rat example, we set:
Ŝ1 s1 and Ŝ 2 s 2

Sˆ x1

Sˆ 12
n1

Sˆ 22

0.6
50

0.8
60

0.134 grams

>O`b!( 7\TS`S\bWOZAbObWabWQa
We are now ready to determine the calculated zscore using the following equation:

Bob’s Basics

The term M1

refers

to the hypothesized difference between the two
population means. When the
null hypothesis is testing that
there is no difference between
population means, then the term
M1 M 2 H is set to 0.

M
Ŝ x1

For the rat example, our calculated z-score becomes:
z

Ŝ x1

2.4

2.1
0.134

2.24

Figure 17.2 shows the results of this hypothesis test.

The critical z-score for a one-tail (right side) test
with F" 0.05 is +1.64. According to Figure 17.2, this
places the calculated z-score of +2.24 in the “Reject
H0” region, which leads to our conclusion that the lucky rats have heavier brains than
the less-fortunate rats.

4WUc`S%
Hypothesis test for rat
example.
0.95

= 0.05
Reject H0

1–
Do Not Reject H0

1.64 2.24
zc z

Number of Standard Deviations from the Mean

Random Thoughts
The conditions that are necessary for the hypothesis test for differences between means
with large sample sizes are as follows:
U The samples are independent of each other.
U The size of each sample must be greater than or equal to 30.
U If the population standard deviations are unknown, we can use the sample stan-

dard deviations to approximate them.

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

We can find the p-value for this sample by using the normal z-score table found in
Appendix B as follows:
P[ z 2.24 ] 1 P[ z b 2.24 ]
P[ z 2.24 ] 1 0.9875 0.0125

Bob’s Basics
We can also apply this technique to hypothesis tests that involve sample sizes less
than 30. However, to do so, the following conditions must be met:
U Both populations must be normally distributed.
U Both population standard deviations must be known.

The results of our rat study can greatly improve the lives of many. When your spouse
catches you sneaking off to the golf course on Saturday morning, you can tell him or
her with a straight face that you are just trying to improve your mind. We now have
the statistics to support you. But be warned, you might develop a sore neck with all
that extra brain weight.

BSabW\UO2WTTS`S\QS=bVS`BVO\HS`]
In the previous example, we were just testing whether or not there was any difference
between the two populations. We can also test whether the difference exceeds a certain value. As an example, suppose we want to test the hypothesis that the average salary of a mathematician in New Jersey exceeds the average salary in Virginia by more
than $5,000. We would state the hypotheses as follows:
H0 : R1R2 f 5000
H1 : R1R2# 5000
where:
R1 = the mean salary of a mathematician in New Jersey
R2 = the mean salary of a mathematician in Virginia
We’ll assume that X1 = $8100 and X2 = $7600, and we’ll test this hypothesis at the
F = 0.10 level.

>O`b!( 7\TS`S\bWOZAbObWabWQa
A sample of 42 mathematicians from New Jersey had a mean salary of $51,500,
whereas a sample of 54 mathematicians from Virginia had a mean salary of $45,400.
The standard error of the difference between two means is:
S x1

S12
n1

S 22

8100 2 7600 2
42

$1622.3

Our calculated z-score becomes:
z

x 2 M1 M 2
$51, 500 $45, 400

S x1 x2
1622.3

$55000

0.68

The results of this hypothesis test are shown in Figure 17.3.

4WUc`S%!
Hypothesis test for the salary
example.
0.90

= 0.10
Reject H0

1–
Do Not Reject H0

0.68 1.28
z
zc

Number of Standard Deviations from the Mean

The critical z-score for a one-tail (right side) test with F = 0.10 is +1.28. According
to Figure 17.3, this places the calculated z-score of +0.68 in the “Do Not Reject H0”
region, which leads to our conclusion that the difference in salaries between the two
states does not exceed $5,000.

BSabW\UT]`2WTTS`S\QSa0SbeSS\;SO\aeWbVA[OZZ
AO[^ZSAWhSaO\RC\Y\]e\AWU[O
This section addresses the situation where the population standard deviation, X, is not
known and the sample sizes are small. If one or both of our sample sizes is less than
30, the population needs to be normally distributed to use any of the following techniques. We made the same assumption for small sample sizes back in Chapters 14 and
16.

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

The sampling distribution for the difference between sample means for this scenario
follows the Student’s t-distribution. Also for small sample sizes, the equation for the
standard error of the difference between two means, S x1 x 2 , depends on whether or
not the standard deviations (or the variances) of the two populations are equal. The
first example will deal with equal standard deviations.

3_cOZ>]^cZObW]\AbO\RO`R2SdWObW]\a
We have a very mysterious occurrence in our household—batteries seem to vanish
into thin air. So I started buying them in 24-packs at the warehouse store, naively
thinking that “these will last a long time.” Wrong again—the more I buy, the faster
they disappear. Maybe it has something to do with certain teenagers listening to music
on their portable CD players at a “brain-numbing” volume into the wee hours of the
morning. Just a thought. So if I ever hear about a new “longer-lasting battery,” I’m all
over it. Let’s say a company is promoting one of these batteries, claiming that its life is
significantly longer than regular batteries. The hypothesis statement would be:
H0 : R1f R2
H1 : R1# R2
where:
R1 = the mean life of the long-lasting batteries
R2 = the mean life of the regular batteries
We’ll test this hypothesis at the F = 0.01 level. The following data was collected measuring the battery life in hours for both types of batteries:

@Oe2ObOT]`0ObbS`g3fO[^ZS
Long-Lasting Battery (Population 1):
51 44 58 36 48 53 57 40

Regular Battery (Population 2):
42 29 51 38 39 44 35

Using Excel, we can summarize this data in the following table.

>O`b!( 7\TS`S\bWOZAbObWabWQa

Ac[[O`WhSR0ObbS`g2ObO
Sample
Mean
_
x

Sample Standard
Deviation
s

Sample
Size
n

Long-lasting (1)

49.2

7.31

Regular (2)

41.1

6.40

Population in Hours

In this example, we are assuming that X1 = X2, but that the values of X1 and X2 are
unknown. Under these conditions, we calculate a pooled estimate of the standard deviation
using the following equation:

1 s12 n2 1 s 22
n1 n2 2

The pooled estimate of the standard deviation combines two sample variances into
one variance and is calculated using s p

1 s12

n1 n2

1 s22

Don’t panic just yet. This equation looks a whole lot better with numbers plugged in.

sp
sp

1 s12 n2 1 s 22

n1 n2 2

1 7.31
10 1 6.40
12 10 2

956.44
6.92
20

We can now approximate the standard error of the difference between two means
using:
Ŝ x1

1
n1

1
n2

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

Let’s apply our example to this fellow.
1
n1

Sˆ x1

6.92

1
6.92
n2

1 1
12 10

0.1833 2.96 hours

We are now ready to determine our calculated t-score using the following equation:
t

M
Ŝ x1

49.2

41.1
2.96

2.73

The number of degrees of freedom for this test are:
d . f . n1 n2 2 12 10 2 20
The critical t-score, taken from Table 4 in Appendix B, for a one-tail (right) test using
F = 0.01 with d.f. = 20 is +2.528. This hypothesis test is shown graphically in Figure
17.4.

4WUc`S%"
Hypothesis test for the battery
example.
0.99
1–

= 0.01
Reject H0

Do Not Reject H0

0
tc = 2.528

t = 2.73

According to Figure 17.4, our calculated t-score of +2.73 is found in the “Reject H0”
region, which leads to our conclusion that the long-lasting batteries do indeed have a
longer life than the regular batteries. Now that has my attention!
This procedure was based on the assumption that the standard deviations of the populations were equal. What if this assumption is not true? I’m glad you asked!

>O`b!( 7\TS`S\bWOZAbObWabWQa
Random Thoughts
The conditions that are necessary for the hypothesis test for differences between means
with small sample sizes are as follows:
U The samples are independent of each other.
U The population must be normally distributed.
U If X1 and X2 are known, use the normal distribution to determine the rejection

region.
U If X1 and X2 are unknown, approximate them with s1 and s2 and use the Student’s

t-distribution to determine the rejection region.

C\S_cOZ>]^cZObW]\AbO\RO`R2SdWObW]\a
We’ll investigate this scenario using the same battery example, but now we will
assume that X1 | X2. The procedure is identical to the previous method except for two
changes.
The first difference involves the standard error of the difference between two means.
The equation used for this scenario is as follows:
Ŝ x1

s 22
n2

s12
n1

For the battery example, our result is:
Sˆ x1

7.31 2 6.40 2
12

4.45 4.10

2.92

We are now ready to determine our calculated t-score using the following equation:
x1 x 2 M1 M 2 H0 49.2 41.1 0
2.77
t

Ŝ x1 x2
2.92
The second difference (hold on to your hat) is the method for determining the number of degrees of freedom for the Student’s t-distribution.

d. f .

¥ s12
¦§ n

1
2

s 22 ´
n2 µ¶

¥ s12 ´
¦§ n µ¶

¥ s 22 ´
¦§ n µ¶

n1 1

n2 1

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

Before you have a seizure, let me demonstrate that this animal’s bark is worse than its
bite. First, recognize that for our battery example:
2

s12 7.31
s 2 6.40
4.45 and 2

n1
n2
10
12

4.10

We can now plug these values into the above equation as follows:

4.45
d . f . ª̈
4.45 2
11

4.10 ·¹
4.10 2

73.10
19.92
1.80 1.87

Because the number of degrees of freedom must be an integer, we round this result to
20. The critical t-score, taken from Table 4 in Appendix B, for a one-tail (right) test
using F = 0.01 with d.f. = 20 is +2.528. Because t t c , we reject H0.

:SbbW\U3fQSZ2]bVS5`c\bE]`Y
Excel performs many of the hypothesis tests that we’ve discussed in this chapter. So let
me explain how to perform the previous battery example using this nifty tool. Follow
these steps:
1. Open a blank Excel sheet and enter the data from the battery example in
Columns A and B as shown in Figure 17.5.
2. From the Tools menu, choose Data Analysis and select t-Test: Two-Sample
Assuming Unequal Variances. (Refer to the section “Installing the Data Analysis
Add-in” from Chapter 2 if you don’t see the Data Analysis command on the
Tools menu.)

4WUc`S%#
Data entry for the battery
example.

>O`b!( 7\TS`S\bWOZAbObWabWQa
3. Click OK.
4. In the t-Test: Two-Sample Assuming Unequal Variances dialog box, choose cells
B1:B12 for Variable 1 Range and cells A1:A10 for Variable 2 Range. Set the
Hypothesized Mean Difference to 0, Alpha to 0.01, and Output Range to cell
D1, as shown in Figure 17.6.

4WUc`S%$
The t-test: Two-Sample
Assuming Unequal Variances
dialog box.

5. Click OK. The t-test output is shown in Figure 17.7.

4WUc`S%%
t-test output.

According to Figure 17.7, the calculated t-score of 2.758 is found in cell E9, which
differs slightly from what we calculated in the previous section (2.77) due to the
rounding of numbers. The p-value of 0.006 is found in cell E10. Because p-value b A ,
we reject the null hypothesis.

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

BSabW\UT]`2WTTS`S\QSa0SbeSS\;SO\aeWbV2S^S\RS\b
AO[^ZSa
Up to this point, all the samples that we have used in the chapter have been independent samples. Samples are independent if they are not related in any way with each
other. This is in contrast to dependent samples, where each observation of one sample is
related to an observation in another.
An example of a dependent sample would be
a weight-loss study. Each person is weighed
at the beginning (Population 1) and end
(Population 2) of the program. The change
in weight of each person is calculated by
subtracting the Population 2 weights from
the Population 1 weights. Each observation
from Population 1 is matched to an observation in Population 2. Dependent samples are
tested differently than independent samples.

With independent samples,
there is no relationship in the
observations between the samples. With dependent samples,
the observation from one sample is related to an observation
from another sample.

To demonstrate testing dependent samples, let’s revisit my golf ball example from
Chapter 15. If you remember, I dreamed I had invented a golf ball that I claimed
would increase the distance off the tee by more than 20 yards. To test my claim,
suppose we had nine golfers hit my golf ball and the same golfers hit a regular golf
ball. The following table shows these results. The letter “d” refers to the difference
between my ball and the other ball.

2WabO\QSW\GO`RaT]`5]ZT0OZZ3fO[^ZS
Golfer

My ball

215

228

256

264

248

255

239

218

239

Other ball

201

213

230

233

218

226

212

195

208

196

225

676

961

900

841

729

529

961

d
d

For future calculations, we will need:

¤ d 14 15 26 31
¤ d 196 225 676
2

30 29 27 23 31 226
961 900 841 729 529 961 6018

>O`b!( 7\TS`S\bWOZAbObWabWQa
The distances using my golf ball will be considered Population 1, and the distances
with the other golf ball will be labeled Population 2. Because the same golfer hit both
balls in each instance in the preceding table, these two samples are considered dependent.
My hypothesis statement for my claim would look like:
H0 : R1R2 f 20
H1 : R1R2# 20
where:
R1 = the average distance off the tee with my new golf ball
R2 = the average distance off the tee with the other golf ball
However, because we are only interested in the difference between the two populations, we can rewrite this statement as a single sample hypothesis as follows:
H0 : Rdf 20
H1 : Rd# 20
where Rd is the mean of the difference between the two populations.
We will test this hypothesis using F = 0.05.
Our next step is to calculate the mean difference, d , and the standard deviation of the
difference, sd , between the two samples as follows:
d

¤ d 226 25.11
9

¤d

¤ d

yards

n
n 1

226 2

6018

9
8

342.89
6.55 yards.
8

The equation for sd is the same standard deviation equation that you learned in
Chapter 5.

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

If both populations follow the normal distribution, we use the Student’s t-distribution
because both sample sizes are less than 30 and X1 and X2 are unknown. The calculated
t-score is found using:
t

d M d 25.11 20 5.11

2.34
sd
6.55
2.18
9
n

The number of degrees of freedom for this test is:
d. f . n 1 9 1 8
The critical t-score, taken from Table 4 in Appendix B, for a one-tail (right) test using
F = 0.05 with d.f. = 8 is +1.86. This hypothesis test is shown graphically in Figure
17.8.

4WUc`S%&
Hypothesis test for the golf
ball example.
0.95
1–

= 0.05
Reject H0

Do Not Reject H0

0
tc = 1.86

t = 2.34

According to Figure 17.8, our calculated t-score of +2.34 is found in the “Reject H0”
region, which leads to our conclusion that my golf ball increases the distance off the
tee by more than 20 yards. Too bad this was only a dream!

BSabW\UT]`2WTTS`S\QSa0SbeSS\>`]^]`bW]\aeWbV
7\RS^S\RS\bAO[^ZSa
We can perform hypothesis testing to examine the difference between proportions of
two populations as long as the sample size is large enough. Recall from Chapter 13,
proportion data follow the binomial distribution, which can be approximated by the
normal distribution under the following conditions.

>O`b!( 7\TS`S\bWOZAbObWabWQa
np v 5 and nq v 5
where:
p = the probability of a success in the population
q = the probability of a failure in the population (q = 1 – p)
Let’s say that I want to test the claim that the proportion of males and females
between the ages of 13 and 19 who use instant messages (IM) on the Internet every
week are the same. My hypothesis would be stated as:
H0 : p1" p2
H1 : p1| p2
where:
p1 = the proportion of 13- to 19-year-old males who use IMs every week
p2 = the proportion of 13- to 19-year-old females who use IMs every week
The following table summarizes the data from the IM samples:

Ac[[O`WhSR2ObOT]`7;AO[^ZSa
Population

Number of Successes
x

Sample Size
n

Male
Female

207
266

300
350

What can we conclude at the F = 0.10 level?
Our sample proportion of male IM users, p1 , and female users, p 2 , can be found by:
p1

x1 207
x
266

0.69 and p 2 2
0.76
n1 300
n2 350

To determine the calculated z-score, we need to know the standard error of the difference between two proportions (that’s a mouthful), S p p , which is found using:
1

p1 1 p1
n1

p2 1 p2
n2

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

Our problem is that we don’t know the values of p1 and p2, the actual population proportions of male and female IM users. The next best thing is to calculate the estimated
standard error of the difference between two proportions, Ŝ p p , using the following
1
2
equation:
¥ 1 1´
Sˆ p p pˆ 1 pˆ ¦
1
2
§ n1 n2 µ¶
where p̂ , the estimated overall proportion of two populations, is found using the
following equation:
pˆ

x1 x 2 207 266

0.728
n1 n2 300 350

For our IM example, the estimated standard error of the difference between two proportions is:
Sˆ p1

0.728 1

¥ 1
0.728 ¦
§ 300

1 ´
0.035
350 µ¶

Bob’s Basics

The term p1 p2 H refers to the hypothesized difference between the two
0
population proportions. When the null hypothesis is testing that there is no difference between population proportions, then the term p1 p2 H is set to 0.

Now we can finally determine the calculated z-score using:
z

Ŝ p1

For the IM example, our calculated z-score becomes:
z

Sˆ p1

0.69

0.76
0.035

2.00

The critical z-scores for a two-tail test with F = 0.10 are +1.64 and –1.64. This
hypothesis test is shown graphically in Figure 17.9.

>O`b!( 7\TS`S\bWOZAbObWabWQa
As you can see in Figure 17.9, the calculated z-score of –2.00 is within the “Reject H0”
region. There, we conclude that the proportions of male and female IM users between
13 and 19 years old are not equal to each other.

4WUc`S%'

Hypothesis Test for the Difference in Proportions

Hypothesis test for the IM
example.
0.90

2 = 0.05
Reject H0

1–
Do Not Reject H0

-2.0 -1.64
z
zc

+1.64
zc

Number of Standard Deviations from the Mean

Bob’s Basics
The standard error of the difference between two proportions describes the variation in the difference between two sample proportions and is calculated using
Sp

p1 1 p1

p2 1 p2

. The estimated standard error of the difference

between two proportions approximates the variation in the difference between two
¥ 1 1´
pˆ ¦
µ . The esti§ n1 n2 ¶
mated overall proportion of two populations is the weighted average of two sample
x x2
.
proportions and is calculated using p̂ 1
n1 n2
ˆ
sample proportions and is calculated using S
p1

pˆ 1

The p-value for these samples can be found using the normal z-score table found in
Appendix B as follows:
2(P[z # +2.00]) = 2(1–P[z f +2.00])
2(P[z # +2.00]) = 2(1–0.9772) = 0.0456
This also confirms that we reject H0 because the p-value b A .

1VO^bS`%( 6g^]bVSaWaBSabW\UeWbVBe]AO[^ZSa

This completes our invigorating journey through the land of hypothesis testing. Don’t
be too sad, though. We’ll have the pleasure of revisiting this technique in Part 4 of
this book—Advanced Inferential Statistics. I just bet you can’t wait.

G]c`Bc`\
1. Test the hypothesis that the average SAT math scores from students in
Pennsylvania and Ohio are different. A sample of 45 students from Pennsylvania
had an average score of 552, whereas a sample of 38 Ohio students had an average score of 530. Assume the population standard deviations for Pennsylvania
and Ohio are 105 and 114 respectively. Test at the F = 0.05 level. What is the
p-value for these samples?
2. A company tracks satisfaction scores based on customer feedback from individual
stores on a scale of 0 to 100. The following data represents the customer scores
from Stores 1 and 2.
Store 1:
90

Store 2:
82

Assume population standard deviations are equal but unknown and that the
population is normally distributed. Test the hypothesis using F = 0.10.
3. A new diet program claims that participants will lose more than 15 pounds after
completion of the program. The following data represents the before and after
weights of nine individuals who completed the program. Test the claim at the
F = 0.05 level.
Before:

221 215 206 185 202 197 244 188 218

After:

200 192 195 166 187 177 227 165 201

4. Test the hypothesis that the proportion of home ownership in the state of
Florida exceeds the national proportion at the F = 0.01 level using the following
data.

>O`b!( 7\TS`S\bWOZAbObWabWQa
Population

Number of Successes

Sample Size

Florida
Nation

272
390

400
600

What is the p-value for these samples?
5. Test the hypothesis that the average hourly wage for City A is more than $0.50 per
hour above the average hourly wage in City B using the following sample data:
City

Average Wage

Sample Standard

Sample Size
Deviation

A
B

$9.80
$9.10

$2.25
$2.70

60
80

Test at the F = 0.05 level. What is the p-value for this test?
6. Test the hypothesis that the average number of days that a home is on the market in City A is different from City B using the following sample data:
City A: 12 8 19 10 26 4 15 20 18 25 7 11
City B: 15 31 14 5 18 20 10 7 25 20 27
Assume population standard deviations are unequal and that the population is
normally distributed. Test the hypothesis using F = 0.10.

BVS:SOabG]cO`b

/RdO\QSR7\TS`S\bWOZ
AbObWabWQa

We covered a lot of ground so far in the first three parts of this book. What
could possibly be left? Well, the last few topics focus on the more advanced
statistical methods (don’t worry, you can handle it) of chi-square tests,
analysis of variance, and simple regression. Armed with these techniques,
we can determine whether two categorical variables are related (chi-square),
compare three or more populations (analysis of variance), and describe the
strength and direction of the relationship between two variables (simple
regression). After you have mastered these concepts, the sky is the limit!

1VO^bS`

BVS1VWA_cO`S>`]POPWZWbg
2Wab`WPcbW]\
7\BVWa1VO^bS`

U Performing a goodness-of-fit test with the chi-square distribution
U Performing a test of independence with the chi-square distribution
U Using contingency tables to display frequency distributions

In the last three chapters, we explored the wonderful world of hypothesis
testing as we compared means and proportions of one and two populations,
making an educated conclusion about our initial claims. With that technique under our belt, we are now ready for bigger and better things.
In this chapter, we will compare two or more proportions using a new
probability distribution: the chi-square. With this new test, we can confirm
whether a set of data follows a specific probability distribution, such as the
binomial or Poisson. (Remember those? They’re back!) We can also use
this distribution to determine whether two variables are statistically independent. It’s actually a lot of fun—really it is!

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

@SdWSe]T2ObO;SOac`S[S\bAQOZSa
In Chapter 2, we discussed the different type of data measurement scales, which were
nominal, ordinal, interval, and ratio. Here is a brief refresher of each:
U Nominal level of measurement deals strictly with qualitative data. Observations

are simply assigned to predetermined categories. One example is gender of the
respondent with the categories being male and female.
U Ordinal measurement is the next level up. It has all the properties of nominal

data with the added feature that we can rank order the values from highest to
lowest. An example would be ranking a movie as great, good, fair, or poor.
U Interval level of measurement involves strictly quantitative data. Here we can use

the mathematical operations of addition and subtraction when comparing values.
For this data, the difference between the different categories can be measured
with actual numbers and also provides meaningful information. Temperature
measurement in degrees Fahrenheit is a common example here.
U Ratio level is the highest measurement scale. Now we can perform all four math-

ematical operations to compare values. Examples of this type of data are age,
weight, height, and salary. Ratio data has all the features of interval data with the
added benefit of a “true zero point,” meaning that a zero data value indicates the
absence of the object being measured.

The chi-square distribution
is used to perform hypothesis
testing on nominal and ordinal
data.

The hypothesis testing that we covered in the last
three chapters strictly used interval and ratio data.
However, the chi-square distribution in this chapter
will allow us to perform hypothesis testing on nominal and ordinal data.
The two major techniques that we will learn about
are using the chi-square distribution to perform a
goodness-of-fit test and to test for the independence
of two variables. So let’s get started!

BVS1VWA_cO`S5]]R\Saa]T4WbBSab
One of the many uses of the chi-square distribution is to perform a goodness-of-fit
test, which uses a sample to test whether a frequency distribution fits the predicted

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\
distribution. As an example, let’s say that a new movie in the making has an expected
distribution of ratings summarized in the following table.

3f^SQbSR;]dWS@ObW\U2Wab`WPcbW]\
Number of Stars

Percentage

5
4
3
2
1
Total

40%
30%
20%
5%
5%
100%

After its debut, a sample of 400 moviegoers were asked to rate the movie, with the
results shown in the following table.

=PaS`dSR;]dWS@ObW\U2Wab`WPcbW]\
Number of Stars

Number of Observations

5
4
3
2
1
Total

145
128
73
32
22
400

Can we conclude that the expected movie
ratings are true based on the observed ratings of 400 people?

The goodness-of-fit test uses
a sample to test whether a
frequency distribution fits the
predicted distribution.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

AbObW\UbVS`]POPWZWbg2Wab`WPcbW]\
2

400

0.05(400) = 20

400

0.05(400) = 20

100%

400

Total

We are now ready to calculate the chi-square statistic.

1OZQcZObW\UbVS1VWA_cO`SAbObWabWQ
The chi-square statistic is found using the following equation:
C2 ¤

E
E

where:
O = the number of observed frequencies for each category
E = the number of expected frequencies for each category
The calculation using this equation is shown in the following table.

BVS1OZQcZObSR1VWA_cO`SAQ]`ST]`bVS;]dWS3fO[^ZS
Movie
Rating

Five
Four
Three
Two
One

145
128
73
32
22

160
120
80
20
20

–15
8
–7
12
2

Total

225
64
49
144
4
C2 ¤

E
E

1.41
0.53
0.61
7.20
0.20

E
E

9.95

2SbS`[W\W\UbVS1`WbWQOZ1VWA_cO`SAQ]`S
2
The critical chi-square score, C c , depends on the number of degrees of freedom,
which for this test would be:

d.f. = k – 1

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
where k equals the number of categories in the frequency distribution. For the movie
example, there are 5 categories, so d.f. = k – 1 = 5 – 1 = 4.
The critical chi-square score is read from the chi-square table found on Table 5 in
Appendix B of this book. Here is an excerpt of this table.

1`WbWQOZ1VWA_cO`SDOZcSa
Selected right tail areas
d.f. 0.3000 0.2000 0.1500 0.1000

0.0500

0.0250 0.0100

1.074

1.642

2.072

2.706

3.841

5.024

6.635

7.879

10.828

2.408

3.219

3.794

4.605

5.991

7.378

9.210

10.597

13.816

3.665

4.642

5.317

6.251

7.815

9.348

11.345

12.838

16.266

4.878

5.989

6.745

7.779

9.488

11.143

13.277

14.860

18.467

6.064

7.289

8.115

9.236

11.070

12.833

15.086

16.750

20.515

7.231

8.558

9.446

10.645

12.592

14.449

16.812

18.548

22.458

0.0050

0.0010

2
For F " 0.10 and d.f. = 4, the critical chi-square score, C c 7.779, is indicated in the
underlined part of the table. Figure 18.1 shows the results of our hypothesis test.

4WUc`S&
Chi-square test for the movie
example.

0.90

= 0.10

1–

Reject H0

Do Not Reject H0
0

7.779
df=4

9.95
x

According to Figure 18.1, the calculated chi-square score of 9.95 is within the “Reject
H0” region, which leads us to the conclusion that the actual movie-rating frequency
distribution differs from the expected distribution. We will always reject H0 as long as
C c2 b C 2.
Also, because the calculated chi-square score for the goodness-of-fit test can only be
positive, the hypothesis test will always be a one-tail with the rejection region on the
right side.

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\

CaW\U3fQSZÂa1677O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

4WUc`S&!
Family of chi-square
distributions.

df=1

df=2

df=3
df=5

d f = 10

0
0

9 10 11 12 13 14 15 16 17 18 19 20

Chi-Square Values

/5]]R\Saa]T4WbBSabeWbVbVS0W\][WOZ2Wab`WPcbW]\
In past chapters, we have occasionally made assumptions that a population follows a
specific distribution such as the normal or binomial. In this section, we can demonstrate how to verify this claim.
As an example, suppose that a certain major league baseball player claims the probability that he will get a hit at any given time is 30 percent. The following table is a frequency distribution of the number of hits per game over the last 100 games. Assume
he has come to bat four times in each of the games.

2ObOT]`bVS0OaSPOZZ>ZOgS`
Number of Hits

Number of Games

0
1
2
3
4
Total

26
34
30
7
3
100

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\

In other words, in 26 games he had 0 hits, in 34 games he had 1 hit, etc. Test the
claim that this distribution follows a binomial distribution with p = 0.30 using
F " 0.05.
The hypothesis statement would look like the following:
H0: The distribution of hits by the baseball player can be described with the
binomial probability distribution using p = 0.30.
H1: The distribution differs from the binomial probability distribution using
p = 0.30.
Our first step is to calculate the frequency distribution for the expected number of hits
per game. To do this, we need to look up the binomial probabilities in Table 1 from
Appendix B for n = 4 (the number of trials per game) and p = 0.30 (the probability of a
success). These probabilities, along with the calculations for the expected frequencies,
are shown in the following table.

3f^SQbSR4`S_cS\Qg1OZQcZObW]\aT]`0OaSPOZZ>ZOgS`
Number of Hits
per Game

Binomial
Probabilities

Number of
Games

Expected
Frequency

0.2401

100

24.01

0.4116

100

41.16
26.46

0.2646

100

0.0756

100

7.56

0.0081

100

0.81

Total

1.0000

Before continuing, we need to make one
adjustment to the expected frequencies.
When using the chi-square test, we need
at least five observations in each of the
expected frequency categories. If there are
less than five, we need to combine categories. In the previous table, we will combine
3 and 4 hits per game into one category to
meet this requirement.

100.00

Bob’s Basics
Expected frequencies do not
have to be integer numbers
because they only represent
theoretical values.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Now we are ready to determine the calculated chi-square score using the following
table:

BVS1OZQcZObSR1VWA_cO`SAQ]`ST]`bVS0OaSPOZZ3fO[^ZS
Hits

0
1
2
3-4

26
34
30
10*

24.01
41.16
26.46
8.37**

1.99
-7.16
3.54
1.63

3.96
51.27
12.53
2.66

0.16
1.25
0.47
0.32

C2 ¤

Total

E
E

2.20

* 7 + 3 = 10
** 7.56 + 0.81 = 8.37

According to Table 5 in Appendix B, the critical chi-square score for F " 0.05 and
d.f. = k = 1 = 4 – 1 = 3 is 7.815. This test is shown in Figure 18.4.

4WUc`S&"
Chi-square test for the
baseball example.

0.95

= 0.05

1–

Reject H0

Do Not Reject H0
0

2.20
x

7.815
df=3

According to Figure 18.4, the calculated chi-square score of 2.20 is within the “Do
Not Reject H0” region, which leads us to the conclusion that the baseball player’s hitting distribution can be described with the binomial distribution using p = 0.30.

1VWA_cO`SBSabT]`7\RS^S\RS\QS
In addition to the goodness-of-fit test, the chi-square distribution can also test for
independence between variables. To demonstrate this technique, I’m going to revisit
the tennis example from Chapter 7.

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\

If you recall, Debbie felt that a short warm-up period before playing our match was
hurting her chances of beating me. After examining the conditional probabilities, I
had to admit there was some evidence supporting Debbie’s claim. However, I’m not
one to take this sitting down. I demand justice, I demand further evidence, I demand
a recount. (Oh, wait a minute, this isn’t Florida.) I demand … a hypothesis test using
the chi-square distribution!
Unbeknownst to Debbie, I have meticulously collected data from our 50 previous
matches. The following table represents the number of wins for each of us according
to the length of the warm-up period.

=PaS`dSR4`S_cS\QWSaT]`BS\\Wa3fO[^ZS
0–10
Min
Debbie wins

11–20
Min
10

More than
20 Min

Total

Bob wins

Total

This is known as a contingency table, which shows the observed frequencies of two
variables. In this case, the variables are warm-up time and tennis player. The table is
organized into r rows and c columns. For our table, r = 2 and c = 3. An intersection of
a row and column is known as a cell. A contingency table has r z c cells, which in our
case, would be 6.
The chi-square test of independence will
determine whether the proportion of times
that Debbie wins is the same for all three
warm-up periods. If the outcome of the
hypothesis test is that the proportions are
not the same, we conclude that the length
of warm-up does impact the performance of
the players. But I have my doubts.
First we state the hypotheses as:
H0 : Warm-up time is independent of
performance
H1 : Warm-up time affects performance

A contingency table shows the
observed frequencies of two
variables. An intersection of a
row and column in a contingency table is known as a cell.
A contingency table has r z c
cells.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
We will test this hypothesis at F " 0.10 level.
Our next step is to determine the expected frequency of each cell in the contingency
table under the assumption that the two variables are independent. We do this using
the following equation:
Er , c

Total of Row r Total of Column c
Total Number of Observations

where Er , c = the expected frequency of the cell that corresponds to the intersection of
Row r and Column c.
The following table applies this notation to our tennis example.
Row/Column

Category

Total Observations

r=1
r=2
c=1
c=2
c=3

Debbie Wins
Bob Wins
0-10 Minute Warm-up
11-20 Minute Warm-up
More than 20 Minute Warm-up

23
27
18
19
13

The total number of observations for this example is 50, which we can confirm by
adding 23 + 27 or 18 + 19 + 13. We can now determine the expected frequencies for
each cell:
E1,1
E2,1

23 18
50

27 18
50

8.28

E1,2

9.72

E2,2

23 19
50

27 19
50

8.74

E1,3

10.26

E2,3

23 13
50

27 13
50

5.98

7.02

The following table summarizes these findings.

3f^SQbSR4`S_cS\QWSaT]`BS\\Wa3fO[^ZS
0–10
Min

11–20
Min

More Than
20 Min

Total

Debbie wins

8.28

8.74

5.98

Bob wins

9.72

10.26

7.02

Total

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\
We now need to determine the calculated
chi-square score using:
C2 ¤

E
E

Bob’s Basics
Notice that the expected
frequencies for a contingency
table add up to the row and
column totals from the observed
frequencies.

This calculation is summarized in the following table.

1VWA_cO`S1OZQcZObW]\T]`bVSBS\\Wa3fO[^ZS
Row

1
1
1
2
2
2

Column

1
2
3
1
2
3

4
10
9
14
9
4

8.28
8.74
5.98
9.72
10.26
7.02

-4.28
1.26
3.02
4.28
-1.26
-3.02

18.32
1.59
9.12
18.32
1.59
9.12
C2 ¤

E
E

2.21
0.18
1.53
1.88
0.15
1.30

E
E

7.25

To determine the critical chi-square score, we need to know the number of degrees of
freedom, which for the independence test would be:
d.f. = (r – 1)(c – 1)
For this example, we have (r – 1)(c – 1) = (2 – 1)(3 – 1) = 2 degrees of freedom.
According to Table 5 in Appendix B, the critical chi-square score for F " 0.10 and
d.f. = 2 is 4.605. This test is shown in Figure 18.5.
According to Figure 18.5, the calculated chi-square score of 7.25 is within the “Reject
H0” region, which leads us to the conclusion that there is a relationship between
warm-up time and performance when Debbie and I play tennis. Darn it—once again,
Debbie is right. Boy, does that have a familiar ring to it.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

4WUc`S&#
Chi-square test for the tennis
example.

0.90

= 0.10

1–

Reject H0

Do Not Reject H0
0

4.605
df=2

7.25
x

However, I do have one consolation. The chi-square test of independence only
investigates whether a relationship exists between two variables. It does not conclude
anything about the direction of the relationship. In other words, from a statistical perspective, Debbie cannot claim that she is disadvantaged by the short warm-up time.
She can only claim that warm-up time has some effect on her performance. We statisticians always leave ourselves a way out!

G]c`Bc`\
1. A company believes that the distribution of customer arrivals during the week
are as follows:
Day

Expected Percentage of Customers

Monday
Tuesday
Wednesday
Thursday
Friday
Saturday
Total

10
10
15
15
20
30
100

A week was randomly chosen and the number of customers each day was
counted. The results were: Monday—31, Tuesday—18, Wednesday—36,
Thursday—23, Friday—47, Saturday—60. Use this sample to test the expected
distribution using F " 0.05.

1VO^bS`&( BVS1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\

2. An e-commerce site would like to test the hypothesis that the number of hits per
minute on their site follows the Poisson distribution with Q = 3. The following
data was collected:
Number of Hits
Per Minute

7 or More

Frequency

Test the hypothesis using F " 0.01.
3. An English professor would like to test the relationship between an English
grade and the number of hours per week a student reads. A survey of 500 students resulted in the following frequency distribution.
Numbers of Hours
Reading per Week

Grade
A
B

Total

Less than 2
2–4
More than 4
Total

36
27
32
95

81
50
24
155

63
25
6
94

10
10
8
28

265
140
95
500

75
28
25
128

Test the hypothesis using F " 0.05.
4. John Armstrong, salesman for the Dillard Paper Company, has five accounts to
visit each day. It is suggested that the random variable, successful sales visits by
Mr. Armstrong, may be described by the binomial distribution, with the probability of a successful visit being 0.4. Given the following frequency distribution
of Mr. Armstrong’s number of successful sales visits per day, can we conclude
that the data actually follows the binomial distribution? Use F = 0.05.
Number of Successful
Visits per Day:

Observed Frequency:

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

BVS:SOabG]cO`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

=\SEOg/\OZgaWa]TDO`WO\QS
If you want to compare the means for three or more populations, ANOVA is the test
for you. Let’s say I’m interested in determining whether there is a difference in consumer satisfaction ratings between three fast-food chains. I would collect a sample of
satisfaction ratings from each chain and test to see whether there is a significant difference between the sample means. Suppose my data look like the following:
Population

Fast-Food Chain

Sample Mean Rating

1
2
3

McDoogles
Burger Queen
Windy’s

7.8
8.2
8.3

My hypothesis statement would look like the following:
H 0 : M1 M 2 M 3
H 1 : not all M’s are equal
Essentially, I’m testing to see whether the variations in customer ratings from the
previous table are due to the fast-food chains or whether the variations are purely
random. In other words, do customers perceive any differences in satisfaction
between the three chains? If I reject the null hypothesis, however, my only conclusion is that a difference does exist. Analysis of variance does not allow me to compare
population means to one another to determine which is greater. That task requires
further analysis.

Bob’s Basics
To use one-way ANOVA, the following conditions must be present:
U The populations of interest must be normally distributed.
U The samples must be independent of each other.
U Each population must have the same variance.

A factor in ANOVA describes the cause of the variation in the data. In the previous
example, the factor would be the fast-food chain. This would be considered a one-way
ANOVA because we are considering only one factor. More complex types of ANOVA
can examine multiple factors, but that topic goes beyond the scope of this book.

1VO^bS`'( /\OZgaWa]TDO`WO\QS

A level in ANOVA describes the number of categories within the factor of interest.
For our example, we have three levels based on the three different fast-food chains
being examined.
To demonstrate one-way ANOVA, I’ll use the following example. I admit, much to
Debbie’s chagrin, that I am clueless when it comes to lawn care. My motto is, “If it’s
green, it’s good.” Debbie, on the other hand, knows exactly what type of fertilizer
to get and when to apply it during the year. I hate spreading this stuff on the lawn
because it apparently makes the grass grow faster, which means I have to cut it more
often.
To make matters worse, we have a neighbor,
Bill, whose yard puts my yard to shame. Mr.
“Perfect Lawn” is out every weekend, meticulously manicuring his domain until it looks
like the home field for the National Lawn
Bowling Association. This gives Debbie a
serious case of “lawn envy.” Bill even has one
of those cute little carts that he pulls on the
back of his tractor. I asked Debbie if I could
get one for my tractor, but she said based on
my “Lawn IQ” I would probably just injure
myself.

A factor in ANOVA describes
the cause of the variation in the
data. When only one factor
is being considered, the procedure is known as one-way
ANOVA. A level in ANOVA
describes the number of categories within the factor of interest.

Anyway, there are several different types of analysis of variance, and covering them all
would take a book unto itself. So throughout the remainder of this chapter, we’ll use
my lawn-care topic to describe two basic ANOVA procedures.

1][^ZSbSZg@O\R][WhSR/<=D/
The simplest type of ANOVA is known as
completely randomized one-way ANOVA, which
involves an independent random selection
of observations for each level of one factor. Now that’s a mouthful! To help explain
this, let’s say I’m interested in comparing
the effectiveness of three lawn fertilizers.
Suppose I select 18 random patches of my
precious lawn and apply either Fertilizer 1,
2, or 3 to each of them. After a week, I mow
the patches and weigh the grass clippings.

The simplest type of ANOVA is
known as completely randomized one-way ANOVA, which
involves an independent random selection of observations
for each level of one factor.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
The factor in this example is fertilizer. There are three levels, representing the three
types of fertilizer we are testing. The table that follows indicates the weight of the
clippings in pounds from each patch. The mean and variance of each level are also
shown.

2ObOT]`:Oe\1ZW^^W\Ua
Fertilizer 1

Fertilizer 2

Fertilizer 3

10.2

11.6

8.5

12.0

9.0

8.4

9.2

10.7

10.5

10.3

9.1

9.0

9.9

10.5
9.5

8.1

12.5

Mean

9.12

10.92

9.48

Variance

1.01

1.70

0.96

We’ll refer to the data for each type of fertilizer as a sample. From the previous table,
we have three samples, each consisting of six observations. The hypotheses statement
can be stated as:
H 0 : M1 M 2 M 3
H 1 : not all M’s are equal
where R1, R2, and R3 are the true population means for the pounds of grass clippings
for each type of fertilizer.

>O`bWbW]\W\UbVSAc[]TA_cO`Sa
The hypothesis test for ANOVA compares two types of variations from the samples.
We first need to recognize that the total variation in the data from our samples can be
divided, or as statisticians like to say, “partitioned,” into two parts.
The first part is the variation within each sample, which is officially known as the sum
of squares within (SSW). This can be found using the following equation:
k

SSW ¤ ni 1 s i2
i 1

1VO^bS`'( /\OZgaWa]TDO`WO\QS
where k = the number of samples (or levels). For the fertilizer example, k = 3 and:
s12 1.01

s 22 1.70

s32 0.96

n1 = 6

n2 = 6

n3 = 6

The sum of squares within can now be calculated as:
SSW = (6 – 1)1.01 + (6 – 1)1.70 + (6 – 1)0.96 = 18.35
Some textbooks will also refer to this value as the error sum of squares (SSE).
The second partition is the variation among the
samples, which is known as the sum of squares
between (SSB). This can be found by:
k

SSB ¤ ni x i
i 1

Random Thoughts
Some textbooks will also refer
to this SSB value as the treatment sum of squares (SSTR).

where x is the grand mean or the average value
of all the observations. For the fertilizer example:
x 1 9.12

x 2 10.92

x 3 9.48

We find x , the grand mean, using:
x

¤x

N
where N = the total number of observations from all samples.
For the fertilizer example:
10.2 8.5 8.4 10.5 ... 9.1 10.5 9.5
9.83
18
We can now calculate the sum of squares between:
x

SSB ¤ ni x i
i 1

SSB 6 9.12 9.83

6 10.92 9.83

6 9.48 9.83 10.86

Random Thoughts
ANOVA does not require that all the sample sizes are equal, as they are in the fertilizer example. See Problem 1 in the “Your Turn” section as an example of unequal
sample sizes.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Finally, the total variation of all the observations is known as the total sum of squares
(SST) and can be found by:
k

SST ¤ ¤ x ij
i 1 j 1

This equation may look nasty, but it is just the difference between each observation
and the grand mean squared and then totaled over all of the observations. This is
clarified more in the following table.
x ij

10.2
8.5
8.4
10.5
9.0
8.1
11.6
12.0
9.2
10.3
9.9
12.5
8.1
9.0
10.7
9.1
10.5
9.5

9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83
9.83

0.37
-1.33
-1.43
0.67
-0.83
-1.73
1.77
2.17
-0.63
0.47
0.07
2.67
-1.73
-0.83
0.87
-0.73
0.67
-0.33

2
ij

0.14
1.77
2.04
0.45
0.69
2.99
3.13
4.71
0.40
0.22
0.01
7.13
2.99
0.69
0.76
0.53
0.45
0.11
k

SST ¤ ¤ x ij
i 1 j 1

This total sum of squares calculation can be confirmed recognizing that:
SST SSW

SSB

SST = 18.35 + 10.86 = 29.21

29.21

1VO^bS`'( /\OZgaWa]TDO`WO\QS

Note that we can determine the variance of the original 18 observations, s2, by:
29.21
SST

1.72
N 1 18 1
This result can be confirmed by using the variance equation that we discussed in
Chapter 5 or by using Excel.
s2

2SbS`[W\W\UbVS1OZQcZObSR4AbObWabWQ
To test the hypothesis for ANOVA, we need to compare the calculated test statistic to
a critical test statistic using the F-distribution. The calculated F-statistic can be found
using the equation:
MSB
F
MSW
where MSB is the mean square between, found by:
SSB
k 1
and MSW is the mean square within, found by:
SSW
MSW
N k
Now, let’s put these guys to work with our fertilizer example.
MSB

MSB

SSB 10.86

5.43
k 1 3 1

MSW
F

SSW 18.35

1.22
N k 18 3

MSB 5.43

4.45
MSW 1.22

If the variation between the samples (MSB) is much greater than the variation within
the samples (MSW), we will tend to reject the null hypothesis and conclude that there
is a difference between population means. To complete our test for this hypothesis, we
need to introduce the F-distribution.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Bob’s Basics
The mean square between (MSB) is a measure of variation between the sample
means. The mean square within (MSW) is a measure of variation within each
sample. A large MSB variation, relative to the MSW variation, indicates that the
sample means are not very close to one another. This condition will result in a large
value of F, the calculated F-statistic. The larger the value of F, the more likely it will
exceed the critical F-statistic (to be determined shortly), leading us to conclude there is
a difference between population means.

2SbS`[W\W\UbVS1`WbWQOZ4AbObWabWQ
We use the F-distribution to determine the critical F-statistic, which is compared
to the calculated F-statistic for the ANOVA hypothesis test. The critical F-statistic,
FA,k 1,N k , depends on two different degrees of freedom, which are determined by:
v1 = k – 1 and v2 = N – k
For our fertilizer example:
v1 = 3 – 1 = 2 and v2 = 18 – 3 = 15
The critical F-statistic is read from the F-distribution table found in Table 6 in
Appendix B of this book. Here is an excerpt of this table.

BOPZS]T1`WbWQOZ4AbObWabWQa
F = 0.05
\ v1 1
v2

1 161.448 199.500 215.707 224.583 230.162 233.986 236.768 238.882 240.543 241.882
2

18.513

19.000

19.164

19.247

19.296

19.330

19.353

19.371

19.385

19.396

10.128

9.552

9.277

9.117

9.013

8.941

8.887

8.845

8.812

8.786

7.709

6.944

6.591

6.388

6.256

6.163

6.094

6.041

5.999

5.964

6.608

5.786

5.409

5.192

5.050

4.950

4.876

4.818

4.772

4.735

5.987

5.143

4.757

4.534

4.387

4.284

4.207

4.147

4.099

4.060

5.591

4.737

4.347

4.120

3.972

3.866

3.787

3.726

3.677

3.637

5.318

4.459

4.066

3.838

3.687

3.581

3.500

3.438

3.388

3.347

1VO^bS`'( /\OZgaWa]TDO`WO\QS

5.117

4.256

3.863

3.633

3.482

3.374

3.293

3.230

3.179

3.137

4.965

4.103

3.708

3.478

3.326

3.217

3.135

3.072

3.020

2.978

4.844

3.982

3.587

3.357

3.204

3.095

3.012

2.948

2.896

2.854

4.747

3.885

3.490

3.259

3.106

2.996

2.913

2.849

2.796

2.753

4.667

3.806

3.411

3.179

3.025

2.915

2.832

2.767

2.714

2.671

4.600

3.739

3.344

3.112

2.958

2.848

2.764

2.699

2.646

2.602

4.543

3.682

3.287

3.056

2.901

2.790

2.707

2.641

2.588

2.544

4.494

3.634

3.239

3.007

2.852

2.741

2.657

2.591

2.538

2.494

Note that this table is based only on F = 0.05. Other values of F will require a different table. For v1 = 2 and v2 = 15, the critical F-statistic, F.05,2,15 = 3.682, as indicated in
the underlined part of the table. Figure 19.1 shows the results of our hypothesis test.

4WUc`S'
0.95

= 0.05

1–

Reject H0

ANOVA test for the fertilizer
example.

Do Not Reject H0
0
v1 = 2

v2 = 15

3.682

4.45

According to Figure 19.1, the calculated F-statistic of 4.45 is within the “Reject H0”
region, which leads us to the conclusion that the population means are not equal. We
will always reject H0 as long as FF,k–1,N–k f F.

Bob’s Basics
The F-distribution has the following characteristics:
U It is not symmetrical but rather has a positive skew.
U The shape of the F-distribution will change with the degrees of freedom speci-

fied by the values of v1 and v2.
U As v1 and v2 increase in size, the shape of the F-distribution becomes more

symmetrical.
U The total area under the curve is equal to 1.
U The F-distribution mean is approximately equal to 1.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Our final conclusion is that one or more of those darn fertilizers is making the grass
grow faster than the others. Sounds like trouble to me.

Wrong Number
Even though we have rejected H0 and concluded that the population means are
not all equal, ANOVA does not allow us to make comparisons between means.
In other words, we do not have enough evidence to conclude that Fertilizer 2 produces more grass clippings than Fertilizer 1. This requires another test known as pairwise comparisons, which we’ll address later in this chapter.

Now let’s explore how Excel can take some of the burden from all these nasty calculations.

CaW\U3fQSZb]>S`T]`[=\SEOg/<=D/
I’m sure you’ve come to the conclusion that calculating ANOVA manually is a lot of
work, and I think you’ll be amazed how easy this procedure is using Excel.
1. Start by placing the fertilizer data in Columns A, B, and C in a blank sheet.
2. Go to the Tools menu and select Data Analysis. (Refer to the section “Installing
the Data Analysis Add-in” from Chapter 2 if you don’t see the Data Analysis
command on the Tools menu.)
3. From the Data Analysis dialog box, select Anova: Single Factor as shown in
Figure 19.2 and click OK.

4WUc`S'
Setting up the one-way
ANOVA in Excel.

1VO^bS`'( /\OZgaWa]TDO`WO\QS
4.

Set up the Anova: Single Factor dialog box according to Figure 19.3.

4WUc`S'!
The ANOVA: Single Factor
dialog box.

Click OK. Figure 19.4 shows the final ANOVA results.

4WUc`S'"
Final results of the one-way
ANOVA in Excel.

These results are consistent with what we found doing it the hard way in the previous sections. Notice that the p-value = 0.0305 for this test, meaning we can reject H0,
because this p-value f F. If you remember, we had set F = 0.05 when we stated the
hypothesis test.

>OW`eWaS1][^O`Wa]\a
Once we have rejected H0 using ANOVA, we can determine which of the sample
means are different using the Scheffé test. This test compares each pair of sample
means from the ANOVA procedure. For the fertilizer example, we would compare x 1
versus x 2 , x 1 versus x 3 , and x 2 versus x 3 to see whether any differences exist.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Bob’s Basics

After rejecting H0 using
ANOVA, we can determine
which of the sample means
are different using the Scheffé
test.

First, the following test statistic for the Scheffé
test, FS, is calculated for each of the pairs of sample
means:

SSW
k

¤ n

¨1
©
n
1 ª a

1·
¸
nb ¹

where:
x a , x b = the sample means being compared
SSW = the sum of squares within from the ANOVA procedure
na, nb = the samples sizes
k = the number of samples (or levels)
Comparing x 1 and x 2 , we have:
FS

9.12

10.92
3.24
8.048

18.35 ¨ 1 1 · 1.22; 0.33=
5 5 5 ©ª 6 6 ¸¹

Comparing x 1 and x 3 , we have:
FS

9.12

9.48
0.13

0.323
18.35 ¨ 1 1 · 1.2
22; 0.33=
5 5 5 ©ª 6 6 ¸¹

Comparing x 2 and x 3 , we have:
FS

10.92

9.48
2.07

5.142
18.35 ¨ 1 1 · 1.22; 0.33=
5 5 5 ©ª 6 6 ¸¹

Next the critical value for the Scheffé test, FSC , is determined by multiplying the critical F-statistic from the ANOVA test by k – 1 as follows:
FSC k 1 FA,k

1, N k

1VO^bS`'( /\OZgaWa]TDO`WO\QS

For the fertilizer example:
F.05,2,15 = 3.682
FSC = (3 – 1)(3.682) = 7.364
If FS f FSC, we conclude there is no difference between the sample means; otherwise
there is a difference. The following table summarizes these results.

Ac[[O`g]TbVSAQVSTT{BSab
Sample Pair

FSC

Conclusion

x 1 and x 2
x 1 and x 3
x 2 and x 3

8.048
0.323
5.142

7.364
7.364
7.364

Difference
No Difference
No Difference

According to our results, the only statistically significant difference is between
Fertilizer 1 and Fertilizer 2. It appears that Fertilizer 2 is more effective in making
grass grow faster when compared to Fertilizer 1. I better keep Debbie away from this
brand.

1][^ZSbSZg@O\R][WhSR0Z]QY/<=D/
Now let’s modify the original fertilizer example: rather than select 18 random samples
from my lawn, we are going to collect 3 random samples from 6 different lawns.
Using the original data, the samples look as follows:
Lawn

Fertilizer 1 Fertilizer 2

Fertilizer 3

Block
Mean

10.2

11.6

8.1

9.97

8.5

12.0

9.0

9.83

8.4

9.2

10.7

9.43

10.5

10.3

9.1

9.97

9.0

9.9

10.5

9.80

8.1

12.5

9.5

10.03

Fertilizer
Mean

9.12

10.92

9.48

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
One concern in this scenario is that the variations in the lawns will account for some
of the variation in the three fertilizers, which may interfere with our hypothesis test.
We can control for this possibility by using a completely randomized block ANOVA,
which is used in the previous table. The type of fertilizer is still the factor, and the
lawns are called blocks.

Completely randomized block
ANOVA controls for variations from other sources than
the factors of interest. This is
accomplished by grouping the
samples using a blocking variable.

There are two hypotheses for the completely randomized block ANOVA. The first (primary) hypothesis tests the equality of the population means, just
like we did earlier with one-way ANOVA:
H 0 : M1 M 2 M 3
H 1 : not all M’s are equal
The secondary hypothesis tests the effectiveness of
the blocking variable as follows:
H 0 ’ : the block means are all equal
H 1 ’ : the blo
ock means are not all equa

The blocking variable would be an effective contributor to our ANOVA model if we
can reject H 0 ’ and claim that the block means are not equal to each other.

>O`bWbW]\W\UbVSAc[]TA_cO`Sa
For the completely randomized block ANOVA, the sum of squares total is partitioned
into three parts according to the following equation:
SST = SSW + SSB + SSBL
where:
SSW = sum of squares within
SSB = sum of squares between
SSBL = sum of squares for the blocking variable (lawns)
Fortunately for us, the calculations for SST and SSW are identical to the one-way
ANOVA procedure that we’ve already discussed, so those values remain unchanged
(SST = 29.21 and SSB = 10.86). We can find the sum of squares block (SSBL) by using
the equation:

1VO^bS`'( /\OZgaWa]TDO`WO\QS
b

SSBL ¤ k x j
j 1

where:
x j the average observation of each blocking level
b = the number of blocking levels (b = 6 for our example)
Using the values from the previous table, we have:
SSBL 3 9.97 9.83
3 9.97 9.83

3 9.83 9.83

39.80 9.83

3 9.43 9.83

310.03 9.83

SSBL 0.72
That leaves us with the sum of squares within (SSW ), which we can find using:
SSW = SST – SSB – SSBL
SSW = 29.21 – 10.86 – 0.72 = 17.63
Almost done!

2SbS`[W\W\UbVS1OZQcZObSR4AbObWabWQ
Since we have two hypothesis tests for the completely randomized block ANOVA, we
have two calculated F-statistics. The F-statistic to test the equality of the population
means (the original hypothesis) is found using:
MSB
MSW
where MSB is the means square between, found by:
F

SSB
k 1
and MSW is the mean square within, found by:
SSW
MSW
k 1 b 1
MSB

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Inserting our fertilizer values into these equations looks like this:
SSB 10.86

5.43
k 1 3 1
SSW
17.63
MSW

1.76
k 1 b 1 3 1 6 1
MSB

MSB 5.43

3.09
MSW 1.76

The second F-statistic will test the significance of the blocking variable (the second
hypothesis) and will be denoted F’ . We will determine this statistic using:
MSBL
MSW
where MSBL is the (can you guess?) mean square blocking, found by:
F’

SSBL
b 1
Plugging our numbers into these guys results in:
MSBL

MSBL

SSBL 0.72

0.14
b 1 6 1

MSBL 0.14

0.08
MSW 1.76
We now need to sit back, catch our breath, and figure out what all these numbers
mean.
F’

B]0Z]QY]`<]bb]0Z]QYBVOb7abVS?cSabW]\
First, we will examine the primary hypothesis, H0, that all population means are equal
using F = 0.05. The degrees of freedom for this critical F-statistic would be:
v1 k 1 3 1 2
v2 k 1 b 1 3 1 6 1 10
The critical F-statistic from Appendix B is F0.05, 2, 10 = 4.103. Since the calculated
F-statistic equals 3.09 and is less than this critical F-statistic, we fail to reject H0 and
cannot conclude that the fertilizer means are different.

1VO^bS`'( /\OZgaWa]TDO`WO\QS

We next examine the secondary hypothesis, H 0 ’, concerning the effectiveness of the
blocking variable, also using F = 0.05. The degrees of freedom for this critical F-statistic would be:
v1 ’ b 1 6 1 5
v2 ’ k 1 b 1 3 1 6 1 10
The critical F-statistic from Appendix B is F0.05, 5, 10 = 3.326. Since the calculated F-statistic, F’, equals 0.08 and is less than this critical F-statistic, we fail to reject H 0 ’ and
cannot conclude that the block means are different.
What does all this mean? Since we failed to reject H 0 ’, the hypothesis that states the
blocking means are equal, the blocking variable (lawns) proved not to be effective and
should not be included in the model. Including an ineffective blocking variable in the
ANOVA increases the chance of a Type II error in the primary hypothesis, H0. The
conclusion of the primary hypothesis in this example would be more precise without
the blocking variable. In fact, this is what essentially happened when we included the
blocking variable with the randomized block design. With the blocking variable present in the model, we failed to discover a difference in the population means. Now go
back to the beginning of the chapter. When we tested the population means using
one-way ANOVA (without a blocking variable), we concluded that the population
means were indeed different.
In summary (It’s about time!), if you feel there is a variable present in your model that
could contribute undesirable variation, such as taking samples from different lawns,
use the randomized block ANOVA. First test H 0 ’, the blocking hypothesis.
U If you reject H 0 ’, the blocking procedure was effective. Proceed to test H0, the

primary hypothesis concerning the population means, and draw your conclusions.
U If you fail to reject H 0 ’, the blocking procedure was not effective. Redo the anal-

ysis using one-way ANOVA (without blocking) and draw your conclusions.
U If all else fails, take two aspirin and call me in the morning.

G]c`Bc`\
1. A consumer group is testing the gas mileage of three different models of cars.
Several cars of each model were driven 500 miles and the mileage was recorded
as follows.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Car 1

Car 2

Car 3

22.5
20.8
22.0
23.6
21.3
22.5

18.7
19.8
20.4
18.0
21.4
19.7

17.2
18.0
21.1
19.8
18.6

Note that the size of each sample does not have to be equal for ANOVA.

Test for a difference between sample means using F = 0.05.
2. Perform a pairwise comparison test on the sample means from Problem 1.
3. A vice president would like to determine whether there is a difference between
the average number of customers per day between four different stores using the
following data.
Store 1

Store 2

Store 3

Store 4

36
48
32
28
31
55

35
20
31
22
19
42
29

26
20
38
32
37
15
21

26
52
37
36
18
30

Note that the size of each sample does not have to be equal for ANOVA.

Test for a difference between sample means using F = 0.05.
4. A certain unnamed statistics author and his two sons played golf at four different
courses with the following scores:

1VO^bS`'( /\OZgaWa]TDO`WO\QS

Course 1
Course 2
Course 3
Course 4

Dad

Brian

John

93
98
89
90

85
87
82
80

80
88
84
82

Using completely randomized block ANOVA, test for the difference of golf
score means using F = 0.05 and using the courses as the blocking variable.

BVS:SOabG]cO`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

7\RS^S\RS\bDS`aca2S^S\RS\bDO`WOPZSa
Suppose I would like to investigate the relationship between the number of hours that
a student studies for a statistics exam and the grade for that exam (uh-oh). The following table shows sample data from six students whom I randomly chose.

2ObOT]`AbObWabWQa3fO[
Hours Studied

Exam Grade

3
5
4
4
2
3

86
95
92
83
78
82

The independent variable (x)
causes variation in the dependent variable (y).

Wrong Number
Exercise caution when
deciding which variable
is independent and which is
dependent. Examine the relationship from both directions to
see which one makes the most
sense. The wrong choice will
lead to meaningless results.

Obviously, we would expect the number of hours
studying to affect the grade. The Hours Studied variable is considered the independent variable (x) because
it causes the observed variation in the Exam Grade,
which is considered the dependent variable (y). The
data from the previous table is considered ordered
pairs of (x,y) values, such as (3,86) and (5,95).
This “causal relationship” between independent and
dependent variables only exists in one direction, as
shown here:
Independent variable (x) m Dependent
variable (y)
This relationship does not work in reverse. For
instance, we would not expect that the exam grade
variable would cause the student to study a certain
number of hours in our previous example.

Other examples of independent and dependent variables are shown in the following
table.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

3fO[^ZSa]T7\RS^S\RS\bO\R2S^S\RS\bDO`WOPZSa
Independent Variable

Dependent Variable

Size of TV
Level of advertising
Size of sports team payroll

Selling price of TV
Volume of sales
Number of games won

Now, let’s focus on describing the relationship between the x and y variables using
inferential statistics.

1]``SZObW]\
Correlation measures both the strength and direction of the relationship between x
and y. Figure 20.1 illustrates the different types of correlation in a series of scatter
plots, which graphs each ordered pair of (x,y). The convention is to place the x variable on the horizontal axis and the y variable on the vertical axis.
y

4WUc`S
Different types of correlation.

X
(A) Positive Linear Correlation
y

X
(C) No Correlation

X
(B) Negative Linear Correlation

X
(D) Nonlinear Correlation

Graph A in Figure 20.1 shows an example of positive linear correlation where, as x
increases, y also tends to increase in a linear (straight line) fashion. Graph B shows a
negative linear correlation where, as x increases, y tends to decrease linearly. Graph
C indicates no correlation between x and y. This set of variables appears to have no
impact on each other. And finally, Graph D is an example of a nonlinear relationship

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
between variables. As x increases, y decreases at first and then changes direction and
increases.
For the remainder of this chapter, we will focus on linear relationships between the
independent and dependent variables. Nonlinear relationships can be very disagreeable and go beyond the scope of this book.

1]``SZObW]\1]STTWQWS\b
The correlation coefficient, r, provides us with both the strength and direction of
the relationship between the independent and dependent variables. Values of r range
between –1.0 and +1.0. When r is positive, the relationship between x and y is positive (Graph A from Figure 20.1), and when r is negative, the relationship is negative
(Graph B). A correlation coefficient close to 0 is evidence that there is no relationship
between x and y (Graph C).
The strength of the relationship between x and y is measured by how close the correlation coefficient is to +1.0 or –1.0 and can be viewed in Figure 20.2.

4WUc`S

The Strength of the Relationship

The strength of the
relationship.

r = +1.0

r = -1.0

X
A

X
B

y
r = +0.60

r = -0.60

X
C

X
D

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\
Graph A illustrates a perfect positive correlation between x and y with r = +1.0. Graph
B shows a perfect negative correlation
between x and y with r = –1.0. Graphs C
and D are examples of weaker relationships
between the independent and dependent
variables.

The correlation coefficient, r,
indicates both the strength and
direction of the relationship
between the independent and
dependent variables. Values
of r range from –1.0, a strong
negative relationship, to +1.0,
a strong positive relationship.
When r = 0, there is no relationship between variables x
and y.

We can calculate the actual correlation coefficient using the following equation:
r

n ¤ xy
¨n x
©ª ¤

¤ x ¤ y

¤ x

· ¨n y2
¹̧ ©ª ¤

¤ y

·
¹̧

Wow! I know this looks overwhelming, but before we panic, let’s try out our exam
grade example on this. The following table will help break down the calculations and
make them more manageable.
Hours of
Study
x

Exam Grade
y

3
5
4
4
2
3

86
95
92
83
78
82

9
25
16
16
4
9

¤ x 21

¤ y 516

258
475
368
332
156
246
¤ xy 1835

7396
8464
9025
6889
6084
6724
¤ y 2 44582

¤x

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Using these values along with n = 6, the number of ordered pairs, we have:
r

¤ x ¤ y

n¤ xy

¤ x

¨n x 2
ª© ¤

· ¨n y2
¹̧ ©ª ¤

¤ y

·
¹̧

6 1835
¨6 79
ª
174

33 1236

21 516
· ¨6 44582 516 2 ·
¹ª
¹

0.862

As you can see, we have a fairly strong positive correlation between hours of study and
the exam grade. That’s good news for us teachers.

Wrong Number
Be careful to distinguish between

¤x

and

¤ x

. With

¤x
¤ x

, we first
2

square each value of x and then add each squared term. With
, we first
add each value of x and then square this result. The answers between the two are
very different!

What is the benefit of establishing a relationship between two variables such as these?
That’s an excellent question. When we discover that a relationship does exist, we can
predict exam scores based on a particular number of hours of study. Simply put, the
stronger the relationship, the more accurate our prediction will be. You will learn how
to make such predictions later in this chapter when we discuss simple regression.

BSabW\UbVSAWU\WTWQO\QS]TbVS1]``SZObW]\1]STTWQWS\b
We can perform a hypothesis test to determine whether the population correlation
coefficient, p, is significantly different from 0 based on the value of the calculated correlation coefficient, r. We can state the hypotheses as:
H0 : p f 0
H1 : p # 0

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

This statement tests whether a positive correlation exists between x and y. I could also
choose a two-tail test that would investigate whether any correlation exists (either
positive or negative) by setting H0 : p " 0 and H1 : p | 0.
The test statistic for the correlation coefficient uses the Student’s t-distribution as
follows:
r
t
1 r2
n 2
where:
r = the calculated correlation coefficient from the ordered pairs
n = the number of ordered pairs
For the exam grade example, the calculated t-statistic becomes:
t

r
1 r2
n 2

0.862

6 2
0.862
t
3.401
0.257
4
The critical t-statistic is based on d.f. = n – 2 if we choose F = 0.05, tc = 2.132 from
Table 4 in Appendix B for a one-tail test. Because t > tc, we reject H0 and conclude that
there is indeed a positive correlation coefficient between hours of study and the exam
grade. Once again, statistics has proven that all is right in the world!

CaW\U3fQSZb]1OZQcZObS1]``SZObW]\1]STTWQWS\ba
After looking at the nasty calculations involved for the correlation coefficient, I’m
sure you’ll be relieved to know that Excel will do the work for you with the CORREL
function that has the following characteristics:
CORREL(array1, array2)
where:
array1 = the range of data for the first variable
array2 = the range of data for the second variable

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
For instance, Figure 20.3 shows the CORREL function being used to calculate the
correlation coefficient for the exam grade example.

4WUc`S !
CORREL function in
Excel with the exam grade
example.

Cell C1 contains the Excel formula =CORREL(A2:A7,B2:B7) with the result being
0.862.

AW[^ZS@SU`SaaW]\
The technique of simple regression enables us to describe a straight line that best fits a
series of ordered pairs (x,y). The equation for a straight line, known as a linear equation, takes the form:
ŷ a bx
where:
ŷ = the predicted value of y, given a value of x
The technique of simple regression enables us to describe a
straight line that best fits a series
of ordered pairs (x,y).

x = the independent variable
a = the y-intercept for the straight line
b = the slope of the straight line
Figure 20.4 illustrates this concept.

Figure 20.4 shows a line described by the equation yˆ 2 0.5x . The y-intercept is the
point where the line crosses the y-axis, which in this case is a = 2. The slope of the
line, b, is shown as the ratio of the rise of the line over the run of the line, shown as
b = 0.5. A positive slope indicates the line is rising from left to right. A negative slope,
you guessed it, moves lower from left to right. If b = 0, the line is horizontal, which
means there is no relationship between the independent and dependent variables. In
other words, a change in the value of x has no effect on the value of y.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\
6

4WUc`S "

y^ = 2 + 0.5x

5
4

y 3

Rise 1
=
= 0.5
2
Run

Equation for a straight line.

2
a=2

1
0
0 1

5 6

Students sometimes struggle with the distinction between ŷ and y. Figure 20.5 shows
six ordered pairs and a line that appears to fit the data described by the equation
yˆ 2 0.5x .
6

4WUc`S #

(2,4)

y^ = 2 + 0.5x

4
x=2
y=4
^
y=3

y 3
2
1

The difference between y
and ŷ.

0
0 1

5 6

Figure 20.5 shows a data point that corresponds to the ordered pair x = 2 and y =
4. Notice that the predicted value of y according to the line at x = 2 is ŷ = 3. We can
verify this using the equation as follows:
yˆ 2 0.5x 2 0.5 2 3
The value of y represents an actual data point, while the value of ŷ is the predicted
value of y using the linear equation, given a value for x.
Our next step is to find the linear equation that best fits a set of ordered pairs.

BVS:SOabA_cO`Sa;SbV]R
The least squares method is a mathematical procedure to identify the linear equation
that best fits a set of ordered pairs by finding values for a, the y-intercept; and b, the
slope. The goal of the least squares method is to minimize the total squared error
between the values of y and ŷ. If we define the error as y yˆ for each data point, the
least squares method will minimize:
n

¤ y

yˆ i

i 1

where n is the number of ordered pairs around the line that best fits the data.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
This concept is illustrated in Figure 20.6.

4WUc`S $
Minimizing the error.

y3– y^ 3

6
5
4

y4– y^4

y2– y^2

y 3
y1– y^ 1

2
1
0
0 1

5 6

According to Figure 20.6, the line that best fits the data, the regression line, will minimize the total squared error of the four data points. I’ll demonstrate how to determine this regression equation using the least squares method through the following
example.
Apparently, there has been a silent war raging in our
bathroom at home that has recently caught my attention. I’m of course referring to the space on our bathThe least squares method is
room countertop that, under an unwritten agreement,
a mathematical procedure to
Debbie and I are supposed to “share.” Over the past
identify the linear equation that
best fits a set of ordered pairs
few months, I have been keeping a wary eye on the
by finding values for a, the
increasing number of “things” on her side that are
y-intercept; and b, the slope.
growing in number at a rate faster than the federal
The goal of the least squares
deficit. I’m slowly being squeezed out of my end of
method is to minimize the total
the bathroom by containers with words such as “volusquared error between the valmizing fixative” and “soyagen complex.” Debbie’s
ues of y and ŷ . The regression
little army has even taken both electrical outlets, cutline is the line that best fits the
ting me off from any source of power. I might as well
data.
just raise the white towel in surrender and head off to
the teenagers’ bathroom in exile, a room I had vowed
never to step foot into because … well, I’ll just spare you the messy details. Just believe
me—you don’t ever want to go in there voluntarily.
Anyway, let’s say the following table shows the number of Debbie’s items on the bathroom counter for the past several months.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

0ObV`]][1]c\bS`2ObO
Month

Number of Items

Month

Number of Items

1
2
3
4
5

8
6
10
6
10

6
7
8
9
10

13
9
11
15
17

Because my goal is to investigate whether the number of items is increasing over time,
Month will be the independent variable and Number of Items will be the dependent
variable.
The least squares method finds the linear equation that best fits the data by determining the value for a, the y-intercept; and b, the slope, using the following equations:
b

¤ x ¤ y
n¤ x ¤ x

n¤ xy

a y bx
where:
x = the average value of x, the independent variable
y = the average value of y, the dependent variable
The following table summarizes the calculations necessary for these equations.

1OZQcZObW]\aT]`bVSAZ]^SO\R7\bS`QS^b
Month
x

Items
y

2
3
4

6
10
6

12
30
24

4
9
16

36
100
36

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Month
x

Items
y

5
6
7
8
9
10

10
13
9
11
15
17

¤ x 55

¤ y 105

x
b
b

50
78
63
88
135
170
¤ xy 658

25
36
49
64
81
100
¤ x 2 385

100
169
81
121
225
289
¤ y 2 1221

55
5.5
10

¤ x ¤ y
n¤ x ¤ x

n¤ xy

105
10.5
10

10 658

55 105
10 385 55
2

805
0.976
825

a y bx 10.5

0.976

5.5 5.13

The regression line for the bathroom counter example would be:
yˆ 5.13 0.976 x
Because the slope of this equation is a positive 0.976, I have evidence that the number
of items on the counter is increasing over time at an average rate of nearly one per
month. Figure 20.7 shows the regression line with the ordered pairs.
My prediction for the number of items on the counter in another six months (Month
16 from my data) will be:
yˆ 5.13 0.976 x 5.13 0.976 16 20.7 z 21 items
Look out, kids. Make room for Dad.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\
4WUc`S %

20
y^ = 5.13 + 0.976x

15
Items

Regression line for the bathroom counter example.

10
5
0
0

Month

1]\TWRS\QS7\bS`dOZT]`bVS@SU`SaaW]\:W\S
Just how accurate is my estimate for the number of items on the counter for a particular month? To answer this, we need to determine the standard error of the estimate, se ,
using the following formula:
se

¤y

a¤ y b ¤ xy

n 2
The standard error of the estimate measures the amount of dispersion of the observed
data around the regression line. If the data points are very close to the line, the standard error of the estimate is relatively low and vice versa. For our bathroom example:
se
se

¤y

a¤ y b ¤ xy
n 2

1221

5.13105 0.976 658
10 2

40.14
2.24
8

We are now ready to calculate a confidence interval (remember those from Chapter
14?) for the mean of y around a particular value of x. For Month 8 (x = 8) in the data,
Debbie has 11 items (y = 11) on the counter. The regression line predicted she would
have:
yˆ 5.13 0.976 x 5.13 0.976 8 12.9 items

The standard error of the estimate, se , measures the amount of dispersion of the
observed data around the regression line.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
In general, the confidence interval around the mean of y given a specific value of x can
be found by:
CI yˆ p t c s e

1
n

x
¤ x

¤ x

where:
tc = the critical t-statistic from the Students’ t-distribution
se = the standard error of the mean
n = the number of ordered pairs
Hold on to your hat while we dive into this one with our example. Suppose we would
like a 95 percent confidence interval around the mean of y for Month 8. To find our
critical t-statistic, we look to Table 4 in Appendix B. This procedure has n – 2 = 10 – 2
= 8 degrees of freedom, resulting in tc = 2.306 from Table 4 in Appendix B. Our confidence interval is then:
CI yˆ p t c s e

1
n

x
¤ x

CI 12.9 p 2.306 2.24

¤ x

n
1
10

8
385

5.5

CI 12.9 p 2.306 2.24 0.419 12.9 p 2.16
CI 10.74 and 15.06
This interval is shown graphically on Figure 20.8.
Our 95 percent confidence interval for the number of items on the counter in Month
8 is between 10.74 and 15.06 items. Sounds like a very crowded countertop to me.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\
4WUc`S &

20
y = 15.06

95 percent confidence interval
for x = 8.

15
Items

10
y = 10.74

5
0
0

Month

BSabW\UbVSAZ]^S]TbVS@SU`SaaW]\:W\S
Recall that if the slope of the regression line, b, is equal to 0, then there is no relationship between x and y. In our bathroom counter example, we found the slope of the
regression line to be 0.976. However, because this result was based on a sample of
observations, we need to test to see whether 0.976 is far enough away from 0 to claim
a relationship really does exist between the variables. If G is the slope of the true population, then our hypotheses statement would be:
H0 : G " 0
H1 : G | 0
If we reject the null hypothesis, we conclude
that a relationship does exist between the
independent and dependent variables based
on our sample. We’ll test this using
F = 0.01.
This hypothesis test requires the standard
error of the slope, sb, which is found with
the following equation:
sb

¤x

where se is the standard error of the estimate
that we calculated earlier.
For our bathroom example:
se
2.24
sb

2
2
385 10 5.5
¤ x nx
sb

2.24
247
0.2
82.5

Wrong Number
Just because a relationship
between two variables is
statistically significant doesn’t
necessarily mean that a causal
relationship truly exists. The
mathematical relationship could
be due to pure coincidence.
Always use your best judgment
when making these decisions.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
The test statistic for this hypothesis is:
b BH
t
sb
0

where B H is the value of the population slope according to the null hypothesis.
0

For this example, our calculated t-statistic is:
t

b BH

0.976 0
3.951
0.247

The critical t-statistic is taken from the Student’s t-distribution with n – 2 = 10 – 2 =
8 degrees of freedom. With a two-tail test and F = 0.01, tc= 3.355 according to Table
4 in Appendix B. Because t > tc, we reject the null hypothesis and conclude there is a
relationship between the month and the number of items on the bathroom countertop. I thought so!

BVS1]STTWQWS\b]T2SbS`[W\ObW]\
Another way of measuring the strength of a relationship is with the coefficient of
determination, r2..This represents the percentage of the variation in y that is explained
by the regression line. We find this value by simply
squaring r, the correlation coefficient. For the bathroom example, the correlation coefficient is:
The coefficient of determinan¤ xy ¤ x ¤ y
tion, r2, represents the percentr
age of the variation in y that
¨ n¤ x 2 ¤ x 2 · ¨ n¤ y 2 ¤ y 2 ·
ª
¹ª
¹
is explained by the regression
line.
10 658 55 105
r
¨10 385 55 2 · ¨10 1221 105 2 ·
ª
¹ª
¹
805
r
0.814
825 1185

The coefficient of determination becomes:
r 2 .0814

0.663

In other words, 66.3 percent of the variation in the number of items on the counter is
explained by the Month variable. If r2 = 1, all of the variation in y is explained by the
variable x. If r2 = 0, none of the variation in y is explained by the variable x.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

CaW\U3fQSZT]`AW[^ZS@SU`SaaW]\
Now that we have burned out our calculators with all these fancy equations, let me
show you how Excel does it all for us.
1. Start by placing the bathroom counter data in Columns A and B in a blank sheet.
2. Go to the Tools menu and select Data Analysis. (Refer to the section “Installing
the Data Analysis Add-in” from Chapter 2 if you don’t see the Data Analysis
command on the Tools menu.)
3. From the Data Analysis dialog box, select Regression as shown in Figure 20.9
and click OK.

4WUc`S '
Setting up simple regression
with Excel.

4. Set up the Regression dialog box according to Figure 20.10.

4WUc`S
The Regression dialog box.

5. Click OK. Figure 20.11 shows the final regression results.

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa

4WUc`S
Final results of the regression
analysis in Excel.

These results are consistent with what we found after grinding it out in the previous
sections. Because the p-value for the independent variable Month is shown as 0.00414,
which is less than F = 0.01, we can reject the null hypothesis and conclude that a relationship between the variables does exist. Debbie has to believe me now!

/AW[^ZS@SU`SaaW]\3fO[^ZSeWbVO`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
The correlation coefficient can be found using:
r

n ¤ xy

¤ x

¨n x 2
ª© ¤

¤ x ¤ y
2

· ¨n y2
¹̧ ©ª ¤

8 5100.1
¨8 28786.8 450
ª
2498.82

27794.4 392.76

¤ y

·
¹̧

450.1 96.2
· ¨8 1205.9 96.2 2 ·
¹ª
¹

0.756

The negative correlation indicates that as mileage (x) increases, the price (y) decreases
as we would expect. The coefficient of determination becomes:
2

r 2 0.756 0.572
Approximately 57 percent of the variation in price is explained by the variation in
mileage. The regression line is determined using:
b
b

¤ x ¤ y
n¤ x ¤ x

n¤ xy

8 5100.1

450.1 96.2
450.1 2

8 28786.8

2498.82
0.0902
27704.39

a y bx 12.025

0.0902 56.26 17.100

We can describe the regression line by the equation:
yˆ 17.1 0.0902x
This equation is shown graphically in Figure 20.12.

4WUc`S
Regression line for car
example.

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

What would the predicted price be for a car with 45,000 miles?
yˆ 17.1 0.0902 45.0 $13.041
The regression line would predict that a car with 45,000 miles would be priced at
$13,041. What would be the 90 percent confidence interval at x = 45,000? The standard error of the estimate would be:
se
se

¤y

a¤ y b ¤ xy
n 2

1205.9

17.196.2 0.0902 5100.1
8 2

1205.9 1645.02 460.03

1.867
6
The critical t-statistic for n – 2 = 8 – 2 = 6 degrees of freedom and a 90 percent confidence interval is tc = 1.943 from Table 4 in Appendix B. Our confidence interval is
then:
se

CI yˆ p t c s e

1
n

x
¤ x

¤ x

CI 13.041 p 1.934 1.867

n
1
8

45
28786.8

56.26

450.1 2
8

CI 13.041 p 1.934 1.867 0.402 13.041 p 1.452
CI 11.589 and 14.493
The 90 percent confidence interval for a car with 45,000 miles is between $11,589 and
$14,493.
Is the relationship between mileage and price statistically significant at the
F = 0.10 level? Our hypotheses’ statement is:
H0 : G " 0
H1 : G | 0

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
The standard error of the slope, sb, is found using:
se
1.867
sb

2
2
2
28786.8 8 56.26
¤ x nx
sb

1.867
3465.3

0.0317

The calculated test statistic for this hypothesis is:
b BH0
0.0902 0
t

2.845
0.0317
sb
The critical t-statistic is taken from the Student’s t-distribution with n – 2 = 8 – 2 = 6
degrees of freedom. With a two-tail test and F = 0.10 level, tc = 1.943 according to
Table 4 in Appendix B. Because t t c , we reject the null hypothesis and conclude
there is a relationship between the mileage and price variable. We use the absolute
values because the calculated t-statistic is in the left tail of the t-distribution with a
two-tail hypothesis test.

/aac[^bW]\aT]`AW[^ZS@SU`SaaW]\
For all these results to be valid, we need to make sure that the underlying assumptions
of simple regression are not violated. These assumptions are as follows:
U Individual differences between the data and the regression line,

yˆ i , are inde-

pendent of one another.
U The observed values of y are normally distributed around the predicted value, ŷ .
U The variation of y around the regression line is equal for all values of x.

Unfortunately (or fortunately), the techniques to test these assumptions go beyond the
level of this book.

AW[^ZSDS`aca;cZbW^ZS@SU`SaaW]\
Simple regression is limited to examining the relationship between a dependent variable
and only one independent variable. If more than one independent variable is involved in
the relationship, then we need to graduate to multiple regression. The regression equation for this method looks like this:
yˆ a b1 x1

b2 x 2 ... bn x n

1VO^bS` ( 1]``SZObW]\O\RAW[^ZS@SU`SaaW]\

As you can imagine, this technique gets really messy and goes beyond the scope of this
book. I’ll have to save this topic for The Complete Idiot’s Guide to Statistics, Part 3. Uhoh, I think I just heard Debbie faint.

G]c`Bc`\
1. The following table shows the payroll for 10 major league baseball teams (in millions) for the 2002 season, along with the number of wins for that year.
Payroll

Wins

Payroll

Wins

$171
$108
$119
$43
$58

103
75
92
55
56

$56
$62
$43
$57
$75

62
84
78
73
67

Calculate the correlation coefficient. Test to see whether the correlation coefficient is not equal to 0 at the 0.05 level.
2. Using the data from Problem 1, answer the following questions:
a) What is the regression line that best fits the data?
b) Is the relationship between payroll and wins statistically significant at the
0.05 level?
c) What is the predicted number of wins with a $70 million payroll?
d) What is the 99 percent confidence interval around the mean number of
wins for a $70 million payroll?
e) What percent of the variation in wins is explained by the payroll?
3. The following table shows the grade point average (GPA) for five students along
with their entrance exam scores for MBA programs (GMAT). Develop a model
that would predict the GPA of a student based on his GMAT score. What would
be the predicted GPA for a student with a GMAT score of 600?

>O`b"( /RdO\QSR7\TS`S\bWOZAbObWabWQa
Student

GPA

GMAT

1
2
3
4
5

3.7
3.0
3.2
4.0
3.5

660
580
450
710
550

BVS:SOabG]c`]POPWZWbgBOPZSa
The following table provides the probability of exactly r successes in n trials
for various values of p.

!$&

/^^S\RWf0

BOPZS 0W\][WOZ>`]POPWZWbgBOPZSa
Values of p
n
2

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.8100

0.6400

0.4900

0.3600

0.2500

0.1600

0.0900

0.0400

0.0100

0.1800

0.3200

0.4200

0.4800

0.5000

0.4800

0.4200

0.3200

0.1800

0.0100

0.0400

0.0900

0.1600

0.2500

0.3600

0.4900

0.6400

0.8100

0.7290

0.5120

0.3430

0.2160

0.1250

0.0640

0.0270

0.0080

0.0010

0.2430

0.3840

0.4410

0.4320

0.3750

0.2880

0.1890

0.0960

0.0270

0.0960

0.1890

0.2880

0.3750

0.4320

0.4410

0.3840

0.2430

0.0010

0.0080

0.0270

0.0640

0.1250

0.2160

0.3430

0.5120

0.7290

0.6561

0.4096

0.2401

0.1296

0.0625

0.0256

0.0081

0.0016

0.0001

0.2916

0.4096

0.4116

0.3456

0.2500

0.1536

0.0756

0.0256

0.0036

0.0486

0.1536

0.2646

0.3456

0.3750

0.3456

0.2646

0.1536

0.0486

0.0036

0.0256

0.0756

0.1536

0.2500

0.3456

0.4116

0.4096

0.2916

0.0001

0.0016

0.0081

0.0256

0.0625

0.1296

0.2401

0.4096

0.6561

0.5905

0.3277

0.1681

0.0778

0.0313

0.0102

0.0024

0.0003

0.0000

0.3280

0.4096

0.3601

0.2592

0.1563

0.0768

0.0284

0.0064

0.0005

0.0729

0.2048

0.3087

0.3456

0.3125

0.2304

0.1323

0.0512

0.0081

0.0512

0.1323

0.2304

0.3125

0.3456

0.3087

0.2048

0.0729

0.0005

0.0064

0.0283

0.0768

0.1563

0.2592

0.3601

0.4096

0.3281

0.0000

0.0003

0.0024

0.0102

0.0313

0.0778

0.1681

0.3277

0.5905

0.5314

0.2621

0.1176

0.0467

0.0156

0.0041

0.0007

0.0001

0.0000

0.3543

0.3932

0.3025

0.1866

0.0938

0.0369

0.0102

0.0015

0.0001

0.0984

0.2458

0.3241

0.3110

0.2344

0.1382

0.0595

0.0154

0.0012

0.0146

0.0819

0.1852

0.2765

0.3125

0.2765

0.1852

0.0819

0.0146

0.0012

0.0154

0.0595

0.1382

0.2344

0.3110

0.3241

0.2458

0.0984

0.0001

0.0015

0.0102

0.0369

0.0938

0.1866

0.3025

0.3932

0.3543

0.0000

0.0001

0.0007

0.0041

0.0156

0.0467

0.1176

0.2621

0.5314

0.4783

0.2097

0.0824

0.0280

0.0078

0.0016

0.0002

0.0000

0.3720

0.3670

0.2471

0.1306

0.0547

0.0172

0.0036

0.0004

0.0000

0.1240

0.2753

0.3177

0.2613

0.1641

0.0774

0.0250

0.0043

0.0002

AbObWabWQOZBOPZSa

!$'

0.0230

0.1147

0.2269

0.2903

0.2734

0.1935

0.0972

0.0287

0.0026

0.0287

0.0972

0.1935

0.2734

0.2903

0.2269

0.1147

0.0230

0.0002

0.0043

0.0250

0.0774

0.1641

0.2613

0.3177

0.2753

0.1240

0.0000

0.0004

0.0036

0.0172

0.0547

0.1306

0.2471

0.3670

0.3720

0.0000

0.0002

0.0016

0.0078

0.0280

0.0824

0.2097

0.4783

0.4305

0.1678

0.0576

0.0168

0.0039

0.0007

0.0001

0.0000

0.3826

0.3355

0.1977

0.0896

0.0313

0.0079

0.0012

0.0001

0.0000

0.1488

0.2936

0.2965

0.2090

0.1094

0.0413

0.0100

0.0011

0.0000

0.0331

0.1468

0.2541

0.2787

0.2188

0.1239

0.0467

0.0092

0.0004

0.0046

0.0459

0.1361

0.2322

0.2734

0.2322

0.1361

0.0459

0.0046

0.0004

0.0092

0.0467

0.1239

0.2188

0.2787

0.2541

0.1468

0.0331

0.0000

0.0011

0.0100

0.0413

0.1094

0.2090

0.2965

0.2936

0.1488

0.0000

0.0001

0.0012

0.0079

0.0313

0.0896

0.1977

0.3355

0.3826

0.0000

0.0001

0.0007

0.0039

0.0168

0.0576

0.1678

0.4305

>]Waa]\>`]POPWZWbgBOPZSa
This table provides the probability of exactly x number of occurrences for various values of R.

BOPZS >]Waa]\>`]POPWZWbgBOPZSa
Values of R
x

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

0.9048

0.8187

0.7408

0.6703

0.6065

0.5488

0.4966

0.4493

0.4066

0.3679

0.0905

0.1637

0.2222

0.2681

0.3033

0.3293

0.3476

0.3595

0.3659

0.3679

0.0045

0.0164

0.0333

0.0536

0.0758

0.0988

0.1217

0.1438

0.1647

0.1839

0.0002

0.0011

0.0033

0.0072

0.0126

0.0198

0.0284

0.0383

0.0494

0.0613

0.0000

0.0001

0.0003

0.0007

0.0016

0.0030

0.0050

0.0077

0.0111

0.0153

0.0000

0.0001

0.0002

0.0004

0.0007

0.0012

0.0020

0.0031

0.0000

0.0001

0.0002

0.0003

0.0005

/^^S\RWf0
Values of R
x

1.1

0
1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

2.0

0.3329 0.3012 0.2725

0.2466

0.2231

0.2019

0.1827

0.1653

0.1496

0.1353

0.3662 0.3614 0.3543

0.3452

0.3347

0.3230

0.3106

0.2975

0.2842

0.2707

0.2014 0.2169 0.2303

0.2417

0.2510

0.2584

0.2640

0.2678

0.2700

0.2707

0.0738 0.0867 0.0998

0.1128

0.1255

0.1378

0.1496

0.1607

0.1710

0.1804

0.0203 0.0260 0.0324

0.0395

0.0471

0.0551

0.0636

0.0723

0.0812

0.0902

0.0045 0.0062 0.0084

0.0111

0.0141

0.0176

0.0216

0.0260

0.0309

0.0361

0.0008 0.0012 0.0018

0.0026

0.0035

0.0047

0.0061

0.0078

0.0098

0.0120

0.0001 0.0002 0.0003

0.0005

0.0008

0.0011

0.0015

0.0020

0.0027

0.0034

0.0000 0.0000 0.0001

0.0001

0.0002

0.0003

0.0005

0.0006

0.0009

0.0000 0.0000 0.0000

0.0000

0.0001

0.0002

2.4

2.5

2.6

2.7

2.8

2.9

3.0

Values of R
x

2.1

2.2

2.3

0.1225 0.1108 0.1003

0.0907

0.0821

0.0743

0.0672

0.0608

0.0550

0.0498

0.2572 0.2438 0.2306

0.2177

0.2052

0.1931

0.1815

0.1703

0.1596

0.1494

0.2700 0.2681 0.2652

0.2613

0.2565

0.2510

0.2450

0.2384

0.2314

0.2240

0.1890 0.1966 0.2033

0.2090

0.2138

0.2176

0.2205

0.2225

0.2237

0.2240

0.0992 0.1082 0.1169

0.1254

0.1336

0.1414

0.1488

0.1557

0.1622

0.1680

0.0417 0.0476 0.0538

0.0602

0.0668

0.0735

0.0804

0.0872

0.0940

0.1008

0.0146 0.0174 0.0206

0.0241

0.0278

0.0319

0.0362

0.0407

0.0455

0.0504

0.0044 0.0055 0.0068

0.0083

0.0099

0.0118

0.0139

0.0163

0.0188

0.0216

0.0011 0.0015 0.0019

0.0025

0.0031

0.0038

0.0047

0.0057

0.0068

0.0081

0.0003 0.0004 0.0005

0.0007

0.0009

0.0011

0.0014

0.0018

0.0022

0.0027

0.0001 0.0001 0.0001

0.0002

0.0003

0.0004

0.0005

0.0006

0.0008

0.0000 0.0000 0.0000

0.0000

0.0001

0.0002

AbObWabWQOZBOPZSa

Values of R
x

3.2

0
1

3.4

3.6

3.8

4.0

4.2

4.4

4.6

4.8

5.0

0.0408 0.0334 0.0273

0.0224

0.0183

0.0150

0.0123

0.0101

0.0082

0.0067

0.1304 0.1135 0.0984

0.0850

0.0733

0.0630

0.0540

0.0462

0.0395

0.0337

0.2087 0.1929 0.1771

0.1615

0.1465

0.1323

0.1188

0.1063

0.0948

0.0842

0.2226 0.2186 0.2125

0.2046

0.1954

0.1852

0.1743

0.1631

0.1517

0.1404

0.1781 0.1858 0.1912

0.1944

0.1954

0.1944

0.1917

0.1875

0.1820

0.1755

0.1140 0.1264 0.1377

0.1477

0.1563

0.1633

0.1687

0.1725

0.1747

0.1755

0.0608 0.0716 0.0826

0.0936

0.1042

0.1143

0.1237

0.1323

0.1398

0.1462

0.0278 0.0348 0.0425

0.0508

0.0595

0.0686

0.0778

0.0869

0.0959

0.1044

0.0111 0.0148 0.0191

0.0241

0.0298

0.0360

0.0428

0.0500

0.0575

0.0653

0.0040 0.0056 0.0076

0.0102

0.0132

0.0168

0.0209

0.0255

0.0307

0.0363

0.0013 0.0019 0.0028

0.0039

0.0053

0.0071

0.0092

0.0118

0.0147

0.0181

0.0004 0.0006 0.0009

0.0013

0.0019

0.0027

0.0037

0.0049

0.0064

0.0082

0.0001 0.0002 0.0003

0.0004

0.0006

0.0009

0.0013

0.0019

0.0026

0.0034

0.0000 0.0000 0.0001

0.0001

0.0002

0.0003

0.0005

0.0007

0.0009

0.0013

0.0000 0.0000 0.0000

0.0000

0.0001

0.0002

0.0003

0.0005

0.0000 0.0000 0.0000

0.0000

0.0001

0.0002

<]`[OZ>`]POPWZWbgBOPZSa
Table 3 provides the area to the left of the corresponding z-score for the standard normal distribution.

BOPZS! <]`[OZ>`]POPWZWbgBOPZSa
Second digit of z
z

0.00

0.0
0.1

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.5000 0.5040 0.5080

0.5120

0.5160

0.5199

0.5239

0.5279

0.5319

0.5359

0.5398 0.5438 0.5478

0.5517

0.5557

0.5596

0.5636

0.5675

0.5714

0.5753

0.2

0.5793 0.5832 0.5871

0.5910

0.5948

0.5987

0.6026

0.6064

0.6103

0.6141

0.3

0.6179 0.6217 0.6255

0.6293

0.6331

0.6368

0.6406

0.6443

0.6480

0.6517

0.4

0.6554 0.6591 0.6628

0.6664

0.6700

0.6736

0.6772

0.6808

0.6844

0.6879

/^^S\RWf0
z

0.00

0.5
0.6

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.6915 0.6950 0.6985

0.7019

0.7054

0.7088

0.7123

0.7157

0.7190

0.7224

0.7257 0.7291 0.7324

0.7357

0.7389

0.7422

0.7454

0.7486

0.7517

0.7549

0.7

0.7580 0.7611 0.7642

0.7673

0.7704

0.7734

0.7764

0.7794

0.7823

0.7852

0.8

0.7881 0.7910 0.7939

0.7967

0.7995

0.8023

0.8051

0.8078

0.8106

0.8133

0.9

0.8159 0.8186 0.8212

0.8238

0.8264

0.8289

0.8315

0.8340

0.8365

0.8389

1.0

0.8413 0.8438 0.8461

0.8485

0.8508

0.8531

0.8554

0.8577

0.8599

0.8621

1.1

0.8643 0.8665 0.8686

0.8708

0.8729

0.8749

0.8770

0.8790

0.8810

0.8830

1.2

0.8849 0.8869 0.8888

0.8907

0.8925

0.8944

0.8962

0.8980

0.8997

0.9015

1.3

0.9032 0.9049 0.9066

0.9082

0.9099

0.9115

0.9131

0.9147

0.9162

0.9177

1.4

0.9192 0.9207 0.9222

0.9236

0.9251

0.9265

0.9279

0.9292

0.9306

0.9319

1.5

0.9332 0.9345 0.9357

0.9370

0.9382

0.9394

0.9406

0.9418

0.9429

0.9441

1.6

0.9452 0.9463 0.9474

0.9484

0.9495

0.9505

0.9515

0.9525

0.9535

0.9545

1.7

0.9554 0.9564 0.9573

0.9582

0.9591

0.9599

0.9608

0.9616

0.9625

0.9633

1.8

0.9641 0.9649 0.9656

0.9664

0.9671

0.9678

0.9686

0.9693

0.9699

0.9706

1.9

0.9713 0.9719 0.9726

0.9732

0.9738

0.9744

0.9750

0.9756

0.9761

0.9767

2.0

0.9772 0.9778 0.9783

0.9788

0.9793

0.9798

0.9803

0.9808

0.9812

0.9817

2.1

0.9821 0.9826 0.9830

0.9834

0.9838

0.9842

0.9846

0.9850

0.9854

0.9857

2.2

0.9861 0.9864 0.9868

0.9871

0.9875

0.9878

0.9881

0.9884

0.9887

0.9890

2.3

0.9893 0.9896 0.9898

0.9901

0.9904

0.9906

0.9909

0.9911

0.9913

0.9916

2.4

0.9918 0.9920 0.9922

0.9925

0.9927

0.9929

0.9931

0.9932

0.9934

0.9936

2.5

0.9938 0.9940 0.9941

0.9943

0.9945

0.9946

0.9948

0.9949

0.9951

0.9952

2.6

0.9953 0.9955 0.9956

0.9957

0.9959

0.9960

0.9961

0.9962

0.9963

0.9964

2.7

0.9965 0.9966 0.9967

0.9968

0.9969

0.9970

0.9971

0.9972

0.9973

0.9974

2.8

0.9974 0.9975 0.9976

0.9977

0.9978

0.9979

0.9980

0.9981

2.9

0.9981 0.9982 0.9982

0.9983

0.9984

0.9985

0.9986

3.0

0.9987 0.9987 0.9987

0.9988

0.9989

0.9990

3.1

0.9990 0.9991 0.9991

0.9991

0.9992

0.9993

3.2

0.9993 0.9993 0.9994

0.9994

0.9995

3.3

0.9995 0.9995 0.9995

0.9996

0.9997

3.4

0.9997 0.9997 0.9997

0.9997

0.9998

3.5

0.9998 0.9998 0.9998

0.9998

AbObWabWQOZBOPZSa

AbcRS\bÂab2Wab`WPcbW]\
Table 4 provides the t-statistic for the corresponding value of alpha or confidence
interval and the number of degrees of freedom.

BOPZS" AbcRS\bÂab2Wab`WPcbW]\
Selected right-tail areas with confidence levels underneath
alpha 0.2000 0.1500 0.1000 0.0500 0.0250 0.0100 0.0050 0.0010 0.0005
conf
lev 0.6000 0.7000 0.8000 0.9000 0.9500 0.9800 0.9900 0.9980 0.9990
d.f.
1

1.376

1.963

3.078

6.314

12.706

31.821

1.061

1.386

1.886

2.920

4.303

6.965

63.657 318.31
9.925

22.327

636.62
31.599

0.978

1.250

1.638

2.353

3.182

4.541

5.841

10.215

12.924

0.941

1.190

1.533

2.132

2.776

3.747

4.604

7.173

8.610

0.920

1.156

1.476

2.015

2.571

3.365

4.032

5.893

6.869

0.906

1.134

1.440

1.943

2.447

3.143

3.707

5.208

5.959

0.896

1.119

1.415

1.895

2.365

2.998

3.499

4.785

5.408

0.889

1.108

1.397

1.860

2.306

2.896

3.355

4.501

5.041

0.883

1.100

1.383

1.833

2.262

2.821

3.250

4.297

4.781

0.879

1.093

1.372

1.812

2.228

2.764

3.169

4.144

4.587

0.876

1.088

1.363

1.796

2.201

2.718

3.106

4.025

4.437

0.873

1.083

1.356

1.782

2.179

2.681

3.055

3.930

4.318

0.870

1.079

1.350

1.771

2.160

2.650

3.012

3.852

4.221

0.868

1.076

1.345

1.761

2.145

2.624

2.977

3.787

4.140

0.866

1.074

1.341

1.753

2.131

2.602

2.947

3.733

4.073

0.865

1.071

1.337

1.746

2.120

2.583

2.921

3.686

4.015

0.863

1.069

1.333

1.740

2.110

2.567

2.898

3.646

3.965

0.862

1.067

1.330

1.734

2.101

2.552

2.878

3.610

3.922

0.861

1.066

1.328

1.729

2.093

2.539

2.861

3.579

3.883

0.860

1.064

1.325

1.725

2.086

2.528

2.845

3.552

3.850

0.859

1.063

1.323

1.721

2.080

2.518

2.831

3.527

3.819

0.858

1.061

1.321

1.717

2.074

2.508

2.819

3.505

3.792

!%!

!%"

/^^S\RWf0
alpha 0.2000 0.1500 0.1000 0.0500 0.0250 0.0100 0.0050 0.0010 0.0005
conf
lev 0.6000 0.7000 0.8000 0.9000 0.9500 0.9800 0.9900 0.9980 0.9990
d.f.
23

0.858

1.060

1.319

1.714

2.069

2.500

2.807

3.485

3.768

0.857

1.059

1.318

1.711

2.064

2.492

2.797

3.467

3.745

0.856

1.058

1.316

1.708

2.060

2.485

2.787

3.450

3.725

0.856

1.058

1.315

1.706

2.056

2.479

2.779

3.435

3.707

0.855

1.057

1.314

1.703

2.052

2.473

2.771

3.421

3.690

0.855

1.056

1.313

1.701

2.048

2.467

2.763

3.408

3.674

0.854

1.055

1.311

1.699

2.045

2.462

2.756

3.396

3.659

0.854

1.055

1.310

1.697

2.042

2.457

2.750

3.385

3.646

1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\
Table 5 provides the chi-square for the corresponding value of alpha and the number
of degrees of freedom.

BOPZS# 1VWA_cO`S>`]POPWZWbg2Wab`WPcbW]\
Selected right-tail areas
d.f.

0.3000 0.2000 0.1500 0.1000 0.0500 0.0250 0.0100 0.0050 0.0010

1.074

1.642

2.072

2.706

3.841

5.024

2.408

3.219

3.794

4.605

5.991

3.665

4.642

5.317

6.251

7.815

4.878

5.989

6.745

7.779

6.064

7.289

8.115

7.231

8.558

8.383

6.635

7.879

10.828

7.378

9.210

10.597

13.816

9.348

11.345

12.838

16.266

9.488

11.143

13.277

14.860

18.467

9.236

11.070

12.833

15.086

16.750

20.515

9.446

10.645

12.592

14.449

16.812

18.548

22.458

9.803 10.748

12.017

14.067

16.013

18.475

20.278

24.322

9.524 11.030 12.027

13.362

15.507

17.535

20.090

21.955

26.124

10.656 12.242 13.288

14.684

16.919

19.023

21.666

23.589

27.877

11.781 13.442 14.534

15.987

18.307

20.483

23.209

25.188

29.588

12.899 14.631 15.767

17.275

19.675

21.920

24.725

26.757

31.264

14.011 15.812 16.989

18.549

21.026

23.337

26.217

28.300

32.909

AbObWabWQOZBOPZSa

!%#

42Wab`WPcbW]\
Table 6 provides the F-statistic for the corresponding degrees of freedom v1 and v2
using a value of alpha equal to 0.05.

BOPZS$ 42Wab`WPcbW]\
F = 0.05
\ v1 1
v2
1

161.448 199.500 215.707 224.583 230.162 233.986 236.768 238.882 240.543 241.882

18.513 19.000 19.164

19.247

19.296

19.330

19.353

10.128

9.552

9.277

9.117

9.013

8.941

8.887

19.371 19.385
8.845

8.812

19.396
8.786

7.709

6.944

6.591

6.388

6.256

6.163

6.094

6.041

5.999

5.964

6.608

5.786

5.409

5.192

5.050

4.950

4.876

4.818

4.772

4.735

5.987

5.143

4.757

4.534

4.387

4.284

4.207

4.147

4.099

4.060

5.591

4.737

4.347

4.120

3.972

3.866

3.787

3.726

3.677

3.637

5.318

4.459

4.066

3.838

3.687

3.581

3.500

3.438

3.388

3.347

5.117

4.256

3.863

3.633

3.482

3.374

3.293

3.230

3.179

3.137

4.965

4.103

3.708

3.478

3.326

3.217

3.135

3.072

3.020

2.978

4.844

3.982

3.587

3.357

3.204

3.095

3.012

2.948

2.896

2.854

4.747

3.885

3.490

3.259

3.106

2.996

2.913

2.849

2.796

2.753

4.667

3.806

3.411

3.179

3.025

2.915

2.832

2.767

2.714

2.671

4.600

3.739

3.344

3.112

2.958

2.848

2.764

2.699

2.646

2.602

4.543

3.682

3.287

3.056

2.901

2.790

2.707

2.641

2.588

2.544

4.494

3.634

3.239

3.007

2.852

2.741

2.657

2.591

2.538

2.494

4.451

3.592

3.197

2.965

2.810

2.699

2.614

2.548

2.494

2.450

4.414

3.555

3.160

2.928

2.773

2.661

2.577

2.510

2.456

2.412

4.381

3.522

3.127

2.895

2.740

2.628

2.544

2.477

2.423

2.378

4.351

3.493

3.098

2.866

2.711

2.599

2.514

2.447

2.393

2.348

/^^S\RWf

5Z]aaO`g
Addition Rule of Probabilities
of two or more events.

Determines the probability of the union

Alternative Hypothesis Denoted by H1, represents the opposite of the
null hypothesis and holds true if the null hypothesis is found to be false.
Analysis of Variance (ANOVA) A procedure to test the difference
between more than two population means.
Bar Chart A data display where the value of the observation is proportional to the height of the bar on the graph.
Bayes’ Theorem A theorem used to calculate P [B/A] from information
about P[A/B]. The term P [A/B] refers to the probability of Event A, given
that Event B has occurred.
Biased Sample A sample that does not represent the intended population
and can lead to distorted findings.
Binomial Experiment An experiment that has only two possible outcomes for each trial. The probability of success and failure is constant.
Each trial of the experiment is independent of any other trial.
Binomial Probability Distribution A method used to calculate the probability of a specific number of successes for a certain number of trials.
Central Limit Theorem A theorem that states as the sample size, n, gets
larger, the sample means tend to follow a normal probability distribution.

!%&

/^^S\RWf1
Class

The interval in a frequency distribution.

Classical Probability Reference to situations when we know the number of possible
outcomes of the event of interest.
Cluster Sample A simple random sample of groups, or clusters, of the population.
Each member of the chosen clusters would be part of the final sample.
Coefficient of Determination, r2 Term represents the percentage of the variation
in y that is explained by the regression line.
Combinations The number of different ways in which objects can be arranged
without regard to order.
Completely Randomized One-Way ANOVA An analysis of variance procedure
that involves the independent random selection of observations for each level of one
factor.
Conditional Probability The probability of Event A, knowing that Event B has
already occurred.
Confidence Interval A range of values used to estimate a population parameter and
associated with a specific confidence level.
Confidence Level The probability that the interval estimate will include the population parameter.
Contingency Table A table which shows the actual or relative frequency of two
types of data at the same time in a table.
Continuous Random Variable A variable that can assume any numerical value
within an interval as a result of measuring the outcome of an experiment.
Correlation Coefficient Indicates the strength and direction of the linear relationship between the independent and dependent variables.
Cumulative Frequency Distribution Indicates the percentage of observations that
are less than or equal to the current class.
Data The value assigned to an observation or a measurement and the building block
to statistical analysis.
Degrees of Freedom The number of values that are free to be varied given information, such as the sample mean, is known.
Dependent Sample The observation from one sample is related to an observation
from another sample.

5Z]aaO`g

!%'

Dependent Variable The variable denoted by y in the regression equation that is
suspected to be influenced by the independent variable.
Descriptive Statistics
obtain an overview.
Direct Observation
ral environment.

Used to summarize or display data so that we can quickly
Gathering data while the subjects of interest are in their natu-

Discrete Probability Distribution A listing of all the possible outcomes of an
experiment for a discrete random variable along with the relative frequency or probability.
Discrete Random Variable A variable that is limited to assuming only specific integer values as a result of counting the outcome of an experiment.
Empirical Probability Type of probability that observes the number of occurrences
of an event through an experiment and calculates the probability from a relative frequency distribution.
Empirical Rule If a distribution follows a bell-shaped, symmetrical curve centered
around the mean, we would expect approximately 68, 95, and 99.7 percent of the
values to fall within one, two, and three standard deviations around the mean respectively.
Expected Frequencies The number of observations that would be expected for
each category of a frequency distribution, assuming the null hypothesis is true with
chi-squared analysis.
Experiment The process of measuring or observing an activity for the purpose of
collecting data.
Event One or more outcomes that are of interest for the experiment and which is/
are a subset of the sample space.
Factor

Describes the cause of the variation in the data for analysis of variance.

Frequency Distribution A table that shows the number of data observations that
fall into specific intervals.
Focus Group An observational technique where the subjects are aware that data is
being collected. Businesses use this type of group to gather information in a group
setting that is controlled by a moderator.
Fundamental Counting Principle A concept that states if one event can occur in
m ways and a second event can occur in n ways, the total number of ways both events
can occur together is m • n ways.

/^^S\RWf1
Goodness-of-Fit Test Uses a sample to test whether a frequency distribution fits
the predicted distribution.
Histogram A bar graph showing the number of observations in each class as the
height of each bar.
Hypothesis An assumption about a population parameter.
Independent Event The occurrence of Event B has no effect on the probability of
Event A.
Independent Sample The observation from one sample is not related to any observations from another sample.
Independent Variable The variable denoted by x in the regression equation is suspected to influence the dependent variable.
Inferential Statistics Used to make claims or conclusions about a population based
on a sample of data from that population.
Interquartile Range Measures the spread of the center half of the data set and is
used to identify outliers.
Intersection Two or more events occurring at the same time.
Interval Estimate Provides a range of values that best describe the population.
Interval Level of Measurement Level of data that allows the use of addition and
subtraction when comparing values, but the zero point is arbitrary.
Joint Probability

The probability of the intersection of two events.

Law of Large Numbers This law states that when an experiment is conducted a
large number of times, the empirical probabilities of the process will converge to the
classical probabilities.
Least Squares Method A mathematical procedure to identify the linear equation
that best fits a set of ordered pairs by finding values for a, the y-intercept; and b, the
slope. The goal of the least squares method is to minimize the total squared error
between the values of y and ŷ .
Level The number of categories within the factor of interest in the analysis of variance procedure.
Level of Significance (F)

Probability of making a Type I error.

5Z]aaO`g

Line Chart A display where ordered pair data points are connected together with a
line.
Margin of Error Concept determines the width of a confidence interval and is calculated using zc S x .
Mean Measure is calculated by adding all the values in the data set and then dividing
this result by the number of observations.
Mean Square Between (MSB)

A measure of variation between the sample means.

Mean Square Within (MSW )

A measure of variation within each sample.

Measure of Central Tendency Describes the center point of our data set with a
single value.
Measure of Relative Position
point.

Describes the percentage of the data below a certain

Median The value in the data set for which half the observations are higher and half
the observations are lower.
Mode

The observation in the data set that occurs most frequently.

Multiplication Rule of Probabilities This rule determines the probability of the
intersection of two or more events.
Mutually Exclusive Events When two events cannot occur at the same time during
an experiment.
Nominal Level of Measurement
identify a group or category.

Lowest level of data where numbers are used to

Null Hypothesis Denoted by H0, this represents the status quo and involves stating
the belief that the mean of the population is f, =, or v a specific value.
Observed Frequencies The number of actual observations noted for each category
of a frequency distribution with chi-squared analysis.
Observed Level of Significance The smallest level of significance at which the null
hypothesis will be rejected, assuming the null hypothesis is true. It is also known as
the p-value.
One-Tail Hypothesis Test
being stated as < or >.
One-Way ANOVA
being considered.

This test is used when the alternative hypothesis is

An analysis of variance procedure where only one factor is

/^^S\RWf1
Ordinal Level of Measurement This measurement has all the properties of nominal data with the added feature that we can rank the values from highest to lowest.
Outcome

A particular result of an experiment.

Outliers Extreme values in a data set that should be discarded before analysis.
p-Value The smallest level of significance at which the null hypothesis will be
rejected, assuming the null hypothesis is true.
Parameter Data that describes a characteristic about a population.
Percentiles Measures of the relative position of the data values from dividing the
data set into 100 equal segments.
Permutations The number of different ways in which objects can be arranged in
order.
Pie Chart Chart used to describe data from relative frequency distributions with a
circle divided into portions whose area is equal to the relative frequency distribution.
Point Estimate A single value that best describes the population of interest, the
sample mean being the most common.
Poisson Probability Distribution A measurement that is used to calculate the
probability that a certain number of events will occur over a specific period of time.
Pooled Estimate of the Standard Deviation A weighted average of two sample
variances.
Population A number which represents all possible outcomes or measurements of
interest.
Primary Data

Data that is collected by the person who eventually uses the data.

Probability The likelihood that a particular event will occur.
Probability Distribution A listing of all the possible outcomes of an experiment
along with the relative frequency or probability of each outcome.
Qualitative Data Information which uses descriptive terms to measure or classify
something of interest.
Quantitative Data
of interest.

Information which uses numerical values to describe something

Quartiles Measures the relative position of the data values by dividing the data set
into four equal segments.

5Z]aaO`g
Random Variable
ment.

!&!

A variable that takes on a numerical value as a result of an experi-

Randomized Block ANOVA Analysis of variance procedure that controls for variations from other sources than the factors of interest.
Range Obtained by subtracting the smallest measurement from the largest measurement of a sample.
Ratio Level of Measurement Level of data that allows the use of all four mathematical operations to compare values and has a true zero point.
Relative Frequency Distribution Displays the percentage of observations of each
class relative to the total number of observations.
Sample

A subset of a population.

Sample Space

All the possible outcomes of an experiment.

Sampling Distribution for the Difference in Means Describes the probability of
observing various intervals for the difference between two sample means.
Sampling Distribution of the Mean The pattern of the sample means that will
occur as samples are drawn from the population at large.
Sampling Error An error which occurs when the sample measurement is different
from the population measurement.
Standard Error of the Difference between Two Means The error describes the
variation in the difference between two sample means.
Standard Error of the Estimate (se) Measures the amount of dispersion of the
observed data around the regression line.
Scheffé Test This test is used to determine which of the sample means are different
after rejecting the null hypothesis using analysis of variance.
Secondary Data
ers to use.

Data that somebody else has collected and made available for oth-

Simple Random Sample A sample where every element in the population has a
chance at being selected.
Simple Regression A procedure that describes a straight line that best fits a series
of ordered pairs (x,y).
Standard Deviation
the variance.

A measure of variation calculated by taking the square root of

!&"

/^^S\RWf1
Standard Error of the Mean The standard deviation of sample means.
Standard Error of the Proportion
tions.

The standard deviation of the sample propor-

Statistic Data that describes a characteristic about a sample.
Statistics The science that deals with the collection, tabulation, and systematic classification of quantitative data, especially as a basis for inference and induction.
Stem and Leaf Display This chart displays the frequency distribution by splitting
the data values into leaves (the last digit in the value) and stems (the remaining digits
in the value).
Stratified Sample A sample that is obtained by dividing the population into mutually exclusive groups, or strata, and randomly sampling from each of these groups.
Subjective Probability This probability is estimated based on experience and intuition.
Sum of Squares Between (SSB)
variance.

The variation among the samples in analysis of

Sum of Squares Block (SSBL)
ance.

The variation among the blocks in analysis of vari-

Sum of Squares Within (SSW )
variance.

The variation within the samples in analysis of

Surveys
tions.

Data collection that involves directly asking the subject a series of ques-

Systematic Sample A sample where every kth member of the population is chosen
for the sample, with value of k being approximately N , where N equals the size of
the population and n equals the size of the sample. n
Test Statistic A quantity from a sample used to decide whether or not to reject the
null hypothesis.
Total Sum of Squares The total variation in analysis of variance that is obtained by
adding the sum of squares between (SSB) and the sum of squares within (SSW).
Two-Tail Hypothesis Test
expressed as |.

This test is used whenever the alternative hypothesis is

Type I Error Occurs when the null hypothesis is rejected when, in reality, it is true.

5Z]aaO`g
Type II Error
true.
Union

!&#

Occurs when the null hypothesis is accepted when, in reality, it is not

At least one of a number of possible events occur.

Variance A measure of dispersion that describes the relative distance between the
data points in the set and the mean of the data set.
Weighted Mean Measure which allows the assignment of more weight to certain
values and less weight to others when calculating an average.

7\RSf
/

Add-Ins dialog box, 25
addition rule of probabilities,
99-101
alpha role (one sample
hypothesis testing), 231-233
alternative hypothesis, 215
chi-square goodness-of-fit
test, 276
stating, 216-217
analysis of variance. See
ANOVA (analysis of variance), 289
Analysis ToolPak, 25
ANOVA (analysis of variance),
289
completely randomized
block ANOVA, 301
calculated F-statistic,
303-304
critical F-statistic,
304-305
partitioning the sum of
squares, 302-303
one-way ANOVA, 290
completely randomized
one-way ANOVA,
291-298
Excel application,
298-299
pairwise comparisons,
299-301
practice, 305-307
assumptions, simple regression, 330
average, 48-50

bar charts, 41-42
bar graphs, 34
Bayes, Thomas, 5
Bayes theorem, 102-103
bell-shape distribution, empirical rule, 69
Bernoulli, James, 5, 121
Bernoulli process.
See binomial probability
distribution
biased samples, 10, 167
BINOMDIST function
(Excel), 127
binomial distribution,
goodness-of-fit test, 280-282
binomial probability distribution (Bernoulli process),
121-126
approximation
normal distribution, 157,
160-161
Poisson distribution,
140-142
characteristics of
experiments, 122-123
Excel calculation, 127-129
mean, 129
practice, 129-130
standard deviation, 129
tables, 126-127
blocking variables, 302

calculated F-statistic
completely randomized
block ANOVA, 303-304
completely randomized
one-way ANOVA,
295-296
cells, 283
census data, 17
center angle (pie charts), 40
central limit theorem, 182-190
central tendency, measures
of, 48
Excel application, 56-58
mean, 48-50
mean of grouped data from
frequency distribution,
51-54
median, 54-55
mode, 55-56
practice, 58-60
selecting measure, 56
weighted mean, 50-51
characteristics
binomial probability
distribution, 122-123
chi-square distribution,
279-280
normal probability distribution, 146-148
Poisson probability
distribution, 132-133
Chart Wizard (Excel), 39
charts, 39
bar charts, 41-42
line charts, 43-44
pie charts, 39-41
Chebyshev’s theorem, 71-73

!&& BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
chi-square probability distribution, 273
characteristics, 279-280
chi-square statistic, 277
CHIINV function (Excel),
279
critical chi-square score,
277-279
data measurement scales,
274
goodness-of-fit test,
274-275
binomial distribution,
280-282
null and alternative
hypothesis, 276
observed versus expected
frequencies, 276-277
practice, 286-288
test for independence,
282-286
class frequencies, 31
classical probability, 82
cluster samples, 171
coefficient of determination,
324-325
combinations (probability),
109-112
complements (probability), 86
completely randomized block
ANOVA, 301
calculated F-statistic,
303-304
critical F-statistic, 304-305
partitioning the sum of
squares, 302-303
completely randomized
one-way ANOVA, 291
calculated F-statistic,
295-296
critical F-statistic, 296-298
partitioning the sum of
squares, 292-295
computer programs
Excel. See Excel (Microsoft)

performance of statistical
techniques, 7, 23-26
conditional probability, 94-97
CONFIDENCE function
(Excel), 203-204
confidence intervals, 195
large samples, 196
calculating intervals,
202-203
changing confidence
levels, 200-201
changing sample size,
201-202
CONFIDENCE function (Excel), 203-204
determining sample size
for the mean, 202
interpretation, 199-200
interval estimate,
196-198
point estimate, 196
proportion, 208-211
practice, 211-212
regression line, 321-323
small samples, 204-208
confidence levels, 197-198
construction, frequency
distributions, 31-32
contingency tables, 87, 283
continuous random variables,
113
CORREL function (Excel),
316
correlation, 311
correlation coefficient,
312-314
calculating with Excel,
315-316
significance, 314-315
practice, 331
counting principles, probability, 106
combinations, 109-111
Excel applications, 112

fundamental counting principle, 106-107
permutations, 107-109
criteria, discrete probability
distributions, 115
critical chi-square score,
277-279
critical F-statistic
completely randomized
block ANOVA, 304-305
completely randomized
one-way ANOVA,
296-298
cumulative frequency distribution, 33-34

2
d.f. (degrees of freedom), 205
data
defined, 15
importance of, 16-17
measurement classification
identification of, 26-27
interval level, 22
nominal level, 21
ordinal level, 21
ratio level, 22-23
ordered pairs. See ordered
pair data
presentation, 29
charts, 39-44
frequency distributions,
30-37
practice, 44
stem and leaf display,
37-39
qualitative, 20
quantitative, 20
sources, 17-18
direct observation, 19
experiments, 19
surveys, 20

7\RSf !&'
summarization, 47
measures of central
tendency, 48-60
measures of dispersion,
61-78
Data Analysis add-in, 24-26
data measurement scales,
chi-square probability
distribution, 274
degrees of freedom (d.f.), 205
Deming, W. Edwards (14
points), 6
Department of Commerce,
census data, 17
Department of Labor, labor
statistics, 18
Department of the Interior,
U.S. resource data, 18
dependent events
conditional probability,
96-97
testing difference between
means, 263-265
dependent variables, 310-311
descriptive statistics, 6-8
data presentation, 29
charts, 39-44
frequency distributions,
30-37
practice, 44
stem and leaf display,
37-39
data summarization, 47
central tendency, 48-60
dispersion, 61-78
identification of, 12
dialog boxes
Add-Ins, 25
ANOVA: Single Factor,
298
Histogram, 36
direct observation, as source
of data, 19
discrete probability distributions, 113

mean, 115-116
rules, 115
standard deviation, 116-118
variance, 116-118
discrete random variables, 113
discrete uniform probability
distribution, 179-180
dispersion, measures of, 61
Chebyshev’s theorem,
71-73
Excel calculation, 75-76
measures of relative position, 73-75
practice, 76-78
range, 62-63
standard deviation, 67-71
variance, 63-67
distributions
chi-square probability, 273
characteristics, 279-280
chi-square statistic, 277
CHIINV function
(Excel), 279
critical chi-square score,
277-279
data measurement scales,
274
goodness-of-fit test,
274-282
observed versus expected
frequencies, 276
practice, 286-288
test for independence,
282-286
probability. See probability
distributions
sampling, 177-178
central limit theorem,
182-190
mean, 178-180
practice, 193-194
proportion, 190-193
standard error of the
mean, 185-186

3
E (margin of error), 198
empirical probability, 83
empirical rule
normal probability distribution, 155-156
standard deviation, 69-71
equal population standard
deviations, 257-260
equations. See formulas
error sum of squares (SSE),
293
errors
hypothesis testing, 219-220
sampling, 173
estimators
interval estimate, 196-198
point estimate, 196
ethics, 10-12
events
mutually exclusive, 98-99
probability, 82
intersection of, 87-88
union of, 88-89
Excel (Microsoft)
calculations
binomial probabilities,
127-129
central tendency, 56-58
correlation coefficient,
315-316
frequency distributions,
34-37
measures of dispersion,
75-76
normal probabilities,
156-157
permutations and combinations, 112
Poisson probabilities,
139-140
Chart Wizard, 39
CHIINV function, 279

!' BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
confidence intervals,
203-204
one-way ANOVA, 298-299
performance of statistical
techniques, 7, 23-24
frequency distributions,
34-37
installation of Data
Analysis add-in, 24-26
simple regression, 325-326
TINV function, 241-242
expected frequencies, chisquare probability distribution, 276-277
experiments
as source of data, 19
binomial characteristics,
122-123
probability, 82

4
F-distribution (ANOVA), 289
completely randomized
block ANOVA, 301-305
completely randomized
one-way ANOVA,
291-298
Excel application, 298-299
one-way ANOVA, 290
pairwise comparisons,
299-301
practice, 305-307
factor (ANOVA), 290
focus groups, 19
formulas
classical probability, 82
mean of frequency distribution, 52-53
permutations, 108
population mean, 49
population variance, 65
range, 62
raw score method, 64

sample mean, 48
sample proportion, 191
standard deviation, 67-68
variance, 63
z-score for the proportion,
193
frequency distributions, 30
construction of, 31-32
cumulative frequency
distribution, 33-34
Excel application, 34-37
histograms, 34
relative frequency distribution, 32-33
fundamental counting principle, probability, 106-107

5½6
goodness-of-fit test
chi-square probability
distribution, 274-275
binomial distribution,
280-282
null and alternative
hypothesis, 276
Poisson process, 134
Gossett, William, 6, 205
graphs
data presentation
frequency distributions,
30-37
stem and leaf display,
37-39
deceptive presentation,
11-12
Histogram dialog box, 36
histograms, 34
history of statistics, 5-6
hypothesis testing, 213
alternative hypothesis,
215-216
null hypothesis, 215

one sample. See one sample
hypothesis testing
one-tail hypothesis test,
218-219, 223-225
practice, 225-226
stating the null and alternative hypothesis, 216-217
two samples. See two sample hypothesis testing
two-tail hypothesis test,
217-218
sample, 220-223
scale of the original
variable, 221-222
standardized normal
scale, 222-223
Type I and II errors,
219-220

7½8½9
independence test, chi-square
distribution, 282-286
independent events, conditional probability, 96-97
independent samples, 265-269
independent trials, 5
independent variables,
310-311
inference, 5
inferential statistics, 6-9
correlation, 311-316
hypothesis testing, 213
alternative hypothesis,
215-216
null hypothesis, 215
one sample, 227-247
one-tail hypothesis test,
218-219, 223-225
practice, 225-226
stating the null and
alternative hypothesis,
216-217
two samples, 250-270

7\RSf !'
two-tail hypothesis test,
217-223
Type I and II errors,
219-220
identification of, 12
independent versus dependent variables, 310-311
samples, 165
cluster, 171
errors, 173
poor techniques, 174
population versus,
166-167
practice, 176
random, 167-170
sampling distributions,
177-194
stratified, 172
systematic, 170
simple regression, 316
assumptions, 330
coefficient of determination, 324-325
Excel application,
325-326
least squares method,
317-321
multiple regression
versus, 330-331
negative correlation
example, 326-330
regression line confidence intervals,
321-323
regression line slope,
323-324
information, data versus, 17
Internet, misuse of statistics,
12
interquartile range (IQR),
74-75
intersection of events,
probability, 87-88
interval estimates, 196-198
interval level of measurement
(data), 22

intervals, frequency distributions, 31
IQR (interquartile range),
74-75

:
labor statistics, 18
least squares method (linear
equations), 317-321
level (ANOVA), 291
line charts, 43-44
linear relationships
correlation coefficient,
312-316
simple regression, 316
assumptions, 330
coefficient of determination, 324-325
Excel application,
325-326
least squares method,
317-321
multiple regression
versus, 330-331
negative correlation
example, 326-330
regression line
confidence intervals,
321-323
regression line slope,
323-324

;
margin of error (E), 198
mean, 48-50
binomial probability
distribution, 129
confidence intervals for
large samples, 196
calculating intervals,
202-203
changing confidence
levels, 200-201

changing sample size,
201-202
CONFIDENCE function (Excel), 203-204
determining sample size,
202
interpretation, 199-200
interval estimate,
196-198
point estimate, 196
confidence intervals for
small samples, 204-208
discrete probability distributions, 115-116
sampling distributions,
178-180
two sample hypothesis
testing, 252-255
dependent samples,
263-265
equal population
standard deviations,
257-260
small sample size and
unknown sigma, 256
unequal population
standard deviations,
260-263
mean square between (MSB),
296
mean square within (MSW),
296
measurements
central tendency, 48
Excel application, 56-58
mean, 48-50
mean of grouped data
from frequency distribution, 51-54
median, 54-55
mode, 55-56
practice, 58-60
selecting measure, 56
weighted mean, 50-51
data, 21

!' BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
identification of, 26-27
interval level, 22
nominal level, 21
ordinal level, 21
ratio level, 22-23
dispersion, 61
Chebyshev’s theorem,
71-73
Excel calculation, 75-76
measures of relative
position, 73-75
practice, 76-78
range, 62-63
standard deviation,
67-71
variance, 63-67
relative position, 73-75
median, 54-55
Microsoft Excel. See Excel
(Microsoft)
mode, 55-56
MSB (mean square between),
296
MSW (mean square within),
296
multiple regression, 330-331
multiplication rule of probabilities, 97-98
mutually exclusive classes (frequency distributions), 32
mutually exclusive events,
98-99

<
negative correlation, 326-330
negative linear correlation,
311
Nielsen Media Research, 166
nominal level of measurement
(data), 21
normal probability distribution, 145
approximating binomial
distribution, 157, 160-161

calculating probabilities,
148
empirical rule, 155-156
Excel, 156-157
standard normal table,
150-155
standard z-score,
148-150
characteristics, 146-148
practice, 161-162
NORMDIST function
(Excel), 156
null hypothesis, 215
chi-square goodness-of-fit
test, 276
stating, 216-217
numerical summarization of
data, 47
central tendency, 48
Excel application, 56-58
mean, 48-50
mean of grouped data
from frequency distribution, 51-54
median, 54-55
mode, 55-56
practice, 58-60
selecting measure, 56
weighted mean, 50-51
dispersion, 61
Chebyshev’s theorem,
71-73
Excel calculation, 75-76
measures of relative
position, 73-75
practice, 76-78
range, 62-63
standard deviation,
67-71
variance, 63-67

=
object order, permutations,
107-109

observation, direct, 19
observed frequencies, 276-277
observed level of significance,
233
one sample hypothesis testing,
227
large sample when sigma is
known, 228-229
large sample when sigma is
unknown, 229-231
one-tail hypothesis test for
proportion, 243-245
p-value, 233-236
practice, 246-247
proportion with large samples, 242
role of alpha, 231-233
small sample when sigma is
known, 236-237
small sample when sigma is
unknown, 237-241
TINV function (Excel),
241-242
two-tail hypothesis test for
proportion, 245-246
one-tail hypothesis test,
218-219
p-value, 233-234
sample, 223-225
one-way ANOVA, 290
completely randomized
one-way ANOVA, 291
calculated F-statistic,
295-296
critical F-statistic,
296-298
partitioning the sum of
squares, 292-295
Excel application, 298-299
ordered pair data, 310
correlation, 311-316
independent versus dependent variables, 310-311
simple regression, 316
assumptions, 330

7\RSf !'!
coefficient of determination, 324-325
Excel application,
325-326
least squares method,
317-321
multiple regression
versus, 330-331
negative correlation
example, 326-330
regression line confidence intervals,
321-323
regression line slope,
323-324
order of objects, permutations,
107-109
ordinal level of measurement
(data), 21

>
p-value, 233-236
pairwise comparisons, Scheffé
test, 299-301
parameters (population), 16,
197
partitioning the sum of
squares
completely randomized
block ANOVA, 302-303
completely randomized
one-way ANOVA,
292-295
permutations, probability,
107-109, 112
Petty, Sir William, 5
pie charts, 39-41
point estimate, 196
Poisson, Simeon, 131
POISSON function (Excel),
139
Poisson probability distribution, 131

approximating binomial
distribution, 140-142
calculating with Excel,
139-140
characteristics, 132-133
practice, 142-143
tables, 136-139
population
mean, 49
parameters, 16
sample versus, 8, 166-167
variance, 65-67
positive linear correlation, 311
posterior probabilities, 96
presentation of data, 29
charts, 39
bar charts, 41-42
line charts, 43-44
pie charts, 39-41
graphs
frequency distributions,
30-37
stem and leaf display,
37-39
practice, 44
primary sources of data, 17-20
prior probabilities, 95
probability, 101
addition rule, 99-101
Bayes theorem, 102-103
classical, 82
conditional, 94-97
counting principles, 106
combinations, 109-111
Excel applications, 112
fundamental counting
principle, 106-107
permutations, 107-109
defined, 82
distributions, 112
binomial experiments,
121-130
discrete, 113-118
normal distribution,
145-162

Poisson process, 131-143
practice, 118-119
random variables,
112-113
empirical, 83
intersection of events,
87-88
multiplication rule, 97-98
mutually exclusive events,
98-99
posterior probabilities, 96
practice, 89-91, 103-104
prior probabilities, 95
properties, 86-87
subjective, 85
union of events, 88-89
proportion
confidence intervals for
large samples, 208-211
one sample hypothesis
testing, 242
one-tail hypothesis test,
243-245
sampling distributions,
190-193
two sample hypothesis
testing, 265-269
two-tail hypothesis test,
245-246
purpose of statistics, 4

?½@
qualitative data, 20
quality control, 6
quantitative data, 20
quartiles, 73-74
r (correlation coefficient), 313
RAND function (Excel), 170
random number table, 169
random samples, 167-170
random variables
Poisson probability distribution, 132

!'" BVS1][^ZSbS7RW]ba5cWRSb]AbObWabWQaASQ]\R3RWbW]\
probability distributions,
distributions, 177
112-113
central limit theorem,
range (measure of dispersion),
182-190
62-63
mean, 178-180
ratio level of measurement
practice, 193-194
(data), 22-23
proportion, 190-193
raw score method (variance
standard error of the
calculations), 64-65
mean, 185-186
relative position, 73-75
errors, 173
regression
mean, 48
lines
poor techniques, 174
confidence intervals,
population versus, 8,
321-323
166-167
slope, 323-324
practice, 176
simple regression. See
random, 167-170
simple regression
stratified, 172
relative frequency distribution,
systematic, 170
32-33
SAS, 7
resources, data, 18
scale of the original variable,
role of alpha, one sample
221-222
hypothesis testing, 231-233
Scheffé test, 299-301
secondary sources of data,
17-18
A
significance, correlation coefficient, 314-315
samples
simple
random sampling,
biased, 10
168-170
clustered, 171
confidence intervals for the simple regression, 316
assumptions, 330
mean, 196
coefficient of determinacalculating intervals,
tion, 324-325
202-203
Excel application, 325-326
changing confidence
least squares method,
levels, 200-201
317-321
changing sample size,
multiple
regression versus,
201-202
330-331
CONFIDENCE funcnegative correlation
tion (Excel), 203-204
example, 326-330
determining sample size,
practice, 331
202
regression line, 321-324
interpretation, 199-200
slope, regression line, 323-324
interval estimate,
software
196-198
Excel. See Excel (Microsoft)
point estimate, 196
performance of statistical
practice, 211-212
techniques, 7, 23-26
proportion, 208-211
small samples, 204-208

sources of data, 17-20
SPSS, 7
SSB (sum of squares between),
293
SSBL (sum of squares block),
303
SSE (error sum of squares),
293
SST (total sum of squares),
294
SSW (sum of squares within),
303
standard deviation, 67
binomial probability distribution, 129
discrete probability distributions, 116-118
empirical rule, 69-71
grouped data calculation,
67-69
standard error of the mean,
185-186
standard error of the proportion, 192-193
standard normal distribution,
149
standard normal table (normal
probability distribution),
150-155
standard z-score (normal
probability distribution),
148-150
standardized normal scale,
222-223
Statistics Canada, 18
stem and leaf display, 37-39
stratified samples, 172
subjective probability, 85
summarization of data. See
numerical summarization of
data
surveys, as source of data, 20
symmetrical curve distribution, 69
systematic samples, 170

7\RSf !'#

B
t-distribution, 205
t-test, 6
tables
binomial probability,
126-127
contingency, 87
Poisson probability distribution, 136-139
random number, 169
standard normal distribution, 150-155
test for independence,
282-286
theoretical sampling distribution of the mean, 186-188
TINV function (Excel),
241-242
total sum of squares (SST),
294
trials, independent, 5
true zero point, 22
twice as much rule, 22
two sample hypothesis testing,
250
differences between means,
252
dependent samples,
263-265
equal population
standard deviations,
257-260
small sample size and
unknown sigma, 256
unequal population
standard deviations,
260-263
differences between proportions, 265-269
differences other than zero,
255-256
practice, 269-270

sampling distribution for
the difference in means,
250-252
two-tail hypothesis test,
217-218
p-value, 234-236
sample, 220-223
scale of the original
variable, 221-222
standardized normal scale,
222-223
Type I errors, 219-220
Type II errors, 219-220

C½D½E
U.S. resource data, 18
unequal population standard
deviations, 260-263
union of events, probability,
88-89
variables, 310-311
variance (measure of dispersion), 63
discrete probability distributions, 116-118
population variance, 65-67
raw score method, 64-65
weighted mean, 50-51

F½G½H
x-axis (line charts), 44
y-axis (line charts), 44

Dear Reader,
Welcome to my world of statistics! I want to commend you for seeking help
with this very challenging topic. Countless individuals out there like you are
struggling with statistics, and many of those don’t make the effort to seek
additional help. I, too, was nearly one of those statistics (sorry, I just love
to use that word!) back in my graduate school days. One of my required
courses was an advanced, theoretical statistics class with seven students that
was taught by a very nice professor who was a brilliant researcher with only
one minor flaw—the man couldn’t teach you how to lick a stamp. After two
classes, a feeling of panic started to set in as I saw my dreams of earning a
Ph.D. fading away. My predominant thought in class was, “What is this guy
talking about?”
Like you, I decided to seek help. Unfortunately for me, the Complete Idiot’s
Guide series hadn’t been invented yet. So I sought the help of a private tutor.
Eugene was an international graduate student with a limited ability to speak
English, but he had a phenomenal sense of explaining abstract concepts. I
quickly fell into the routine of leaving class in a complete fog, meeting with
Eugene, and then exclaiming, “Eureka, that’s what he was talking about!”
I went on to receive an “A” in this class, earned my degree, and the rest is
history. As a token of my appreciation to Eugene, I presented him with my
first-born male child. (I’m only kidding, Brian!)
Based on my own experiences, my advice to you is to either find a brilliant
international graduate student with a limited ability to speak English who
can explain abstract concepts with amazing clarity to personally tutor you, or
use this book.
Each statistical concept in the chapters that follow is explained in loving
detail with plenty of examples and, when appropriate, a little humor. In
writing this book, my goal has been to play the role of Eugene for you and
explain those messy concepts in a way that makes sense to you, so you can
say “Eureka!” Only I won’t cost as much as Eugene, and I hope my English
is a little better.

Bob Donnelly

/P]cbbVS/cbV]`
Robert A. Donnelly, Jr., Ph.D. (bob@stat-guide.com) is a professor at
Goldey-Beacom College in Wilmington, Delaware, with more than 20 years
of teaching experience. He teaches classes in statistics, operations management, management information systems, and database management at both
the undergraduate and graduate level. Bob earned an undergraduate degree
in chemical engineering from the University of Delaware, after which he
worked for several years as an engineer in a local chemical plant. Despite
success in this field, Bob felt drawn to pursue a career in education. It was
his desire to teach (or maybe he just had a bad day) that took him back to
school to earn his MBA and Ph.D. in operations research, also from the
University of Delaware. Go Blue Hens!
Bob’s working experience prior to his teaching career provides him with
many opportunities to incorporate real-life examples into classroom learning. His students appreciate his knowledge of the business world as well as
his mastery of the course subject matter. Many former students seek Bob’s
assistance in work-related issues that deal with his expertise. Typical student
comments focus on his genuine concern for their welfare and his desire to
help them succeed in reaching their goals. They also love when he cancels
class because the roads in his backwoods neighborhood have flooded.
While keeping teaching as his main focus, Bob performs consulting activities
through his firm, Partners for Strategic Solutions, which provides services
for businesses seeking management techniques to improve performance. He
recently completed a test bank for a new textbook on mathematical modeling using Excel for Prentice-Hall Publishers. Bob has also remained current
with today’s technology with CIW certification as Master CIW Designer.
You can reach him at bob@stat-guide.com.
It is obvious to anyone that Bob’s first love is teaching. His children can
attest to that when his eyes light up at the end of the day and he asks “Well,
does anybody need help with their math homework?” Sometimes they say
yes just to make him happy.

tear here

BVS1][^ZSbS7RW]bÂa5cWRSb]AbObWabWQa
@STS`S\QS1O`R
1]\TWRS\QS7\bS`dOZa
Type

Sample

Population

Confidence Interval

Mean

Any

Known

x p zc

S
n

Mean

Any

Unknown

x p zc

s
n

Mean

Must Be Normal

Known

x p zc

S
n

Mean

Must Be Normal

Unknown

Proportion

npr
nqr

Any

x p tc

s
n

p s p zc

d. f . n 1

ps 1 ps
n

AO[^ZSAWhST]`1]\TWRS\QS7\bS`dOZa
Type

Sample Size

Mean

¥ zS ´
n¦ µ
§ E¶

Proportion

¥z ´
n pq ¦ c µ
§ E¶

1`WbWQOZhAQ]`Sa
Alpha

Tail

Critical z-Score

0.01
0.01
0.02
0.02

One
Two
One
Two

±2.33
±2.57
±2.05
±2.33

Alpha

Tail

Critical z-Score

0.05
0.05
0.10
0.10

One
Two
One
Two

±1.64
±1.96
±1.28
±1.64

tear here

=\SAO[^ZS6g^]bVSaWaBSab
Type

Sample

Population

Test Statistic

Mean

Any

Known

Mean

Any

Unknown

Mean

Must Be Normal

Known

Mean

Must Be Normal

Unknown

Proportion

npr
nqr

Any

x M H0
S/ n
x M H0
s/ n
x M H0
S/ n

x M H0

d. f . n 1

s/ n
p

pH 0

pH 0 1 pH 0
n

Be]AO[^ZS6g^]bVSaWaBSab
Type

Sample

Population

S S

Mean

n 1 , n 2 r
Independent
Samples

Any

Known

n 1 , n 2 r
Independent
Samples

Any

n 1 , n 2
Independent
Samples

Must Be
Normal

Known

n 1 , n 2
Independent
Samples

Must Be
Normal

Unknown
and
Equal

n 1 , n 2
Independent
Samples

Must Be
Normal

Unknown
and
Unequal

x
t

Mean

Proportion

npr
nqr
Independent
Samples

Test Statistic
z

Unknown

x
n

Any
z

S12
n1

M
s12
n1

s 22
n2

s 22
n2
1

S 22
n2

1 s12 n2 1 s 22 1
n1 n2 2
n1
x2

S 22
n2

s12
n1

x2
S12
n1

d . f . n1 n2

1
n2

d. f .

¥ s12
¦§ n

1
2

¥
pˆ 1 pˆ ¦ n1
§

1´
n2 ¶µ

¥ s12 ´
¦§ n µ¶

¥ s 22 ´
¦§ n µ¶

n1 1

n2 1

s 22 ´
n2 µ¶

p̂

x1
n1

x2
n2

Source Exif Data:

File Type                       : PDF
File Type Extension             : pdf
MIME Type                       : application/pdf
PDF Version                     : 1.7
Linearized                      : Yes
Author                          : Robert A. Donnelly, Jr., Ph.D.
Create Date                     : 2007:11:13 15:25:32+05:30
Keywords                        : 142951390X
Modify Date                     : 2017:03:01 01:59:33+07:00
Subject                         : Penguin Group USA, Inc.
Has XFA                         : No
XMP Toolkit                     : Adobe XMP Core 5.6-c015 84.159810, 2016/09/10-02:41:30
Format                          : application/pdf
Creator                         : Robert A. Donnelly, Jr., Ph.D.
Description                     : Penguin Group USA, Inc.
Title                           : The Complete Idiot's Guide to Statistics
Metadata Date                   : 2017:03:01 01:59:33+07:00
Document ID                     : uuid:dc9b70f2-45d8-421f-93f0-c1df71036d7b
Instance ID                     : uuid:2ef1815b-0497-481c-b479-797b2b7ba19c
Page Layout                     : SinglePage
Page Mode                       : UseOutlines
Page Count                      : 421

EXIF Metadata provided by EXIF.tools

The Idiot's Guide To Statistics Jr., Ph.D., Robert A. Donnelly Statistics, 2nd Edition Alpha (

Navigation menu

Versions of this User Manual:

Views

Navigation