Psych Manual

User Manual:

Open the PDF directly: View PDF PDF.
Page Count: 367 [warning: Documents this large are best viewed by clicking the View PDF Link!]

Package ‘psych’
June 28, 2016
Version 1.6.6
Date 2016-06-28
Title Procedures for Psychological, Psychometric, and Personality
Research
Author William Revelle <revelle@northwestern.edu>
Maintainer William Revelle <revelle@northwestern.edu>
Description A general purpose toolbox for personality, psychometric theory and experimental psy-
chology. Functions are primarily for multivariate analysis and scale construction using fac-
tor analysis, principal component analysis, cluster analysis and reliability analysis, although oth-
ers provide basic descriptive statistics. Item Response Theory is done using factor analy-
sis of tetrachoric and polychoric correlations. Functions for analyzing data at multi-levels in-
clude within and between group statistics, including correlations and factor analysis. Func-
tions for simulating particular item and test structures are included. Several func-
tions serve as a useful front end for structural equation modeling. Graphical displays of path dia-
grams, factor analysis and structural equation models are created using basic graph-
ics. Some of the functions are written to support a book on psychometrics as well as publica-
tions in personality research. For more information, see the personality-project.org/r webpage.
License GPL (>= 2)
Imports mnormt,parallel,stats,graphics,grDevices,methods
Suggests GPArotation, sem, lavaan, Rcsdp, graph, Rgraphviz
LazyData true
URL http://personality-project.org/r/psych
http://personality-project.org/r/psych-manual.pdf
NeedsCompilation no
Depends R (>= 2.10)
Repository CRAN
Date/Publication 2016-06-28 23:40:54
1
2Rtopics documented:
Rtopics documented:
00.psych........................................... 5
ability ............................................ 14
affect ............................................ 16
alpha............................................. 17
Bechtoldt .......................................... 21
bestScales.......................................... 24
bfi .............................................. 26
bi.bars............................................ 29
biplot.psych......................................... 30
block.random ........................................ 32
blot ............................................. 33
bock............................................. 34
burt ............................................. 35
circ.tests........................................... 37
cities............................................. 39
cluster.fit........................................... 40
cluster.loadings ....................................... 41
cluster.plot.......................................... 43
cluster2keys......................................... 45
cohen.kappa......................................... 46
comorbidity......................................... 49
cor.ci............................................. 50
cor.plot ........................................... 52
cor.smooth.......................................... 55
cor.wt ............................................ 57
cor2dist ........................................... 58
corFiml ........................................... 59
corr.test ........................................... 60
correct.cor.......................................... 62
cortest.bartlett........................................ 64
cortest.mat.......................................... 65
cosinor............................................ 67
count.pairwise........................................ 72
cta.............................................. 73
cubits ............................................ 76
cushny............................................ 77
densityBy .......................................... 78
describe........................................... 80
describeBy ......................................... 83
df2latex ........................................... 84
diagram ........................................... 87
draw.tetra .......................................... 89
dummy.code......................................... 91
Dwyer............................................ 92
eigen.loadings........................................ 92
ellipses ........................................... 93
Rtopics documented: 3
epi.............................................. 95
epi.bfi ............................................ 98
error.bars .......................................... 99
error.bars.by.........................................102
error.crosses.........................................105
errorCircles .........................................107
fa ..............................................109
fa.diagram..........................................119
fa.extension.........................................122
fa.multi ...........................................125
fa.parallel ..........................................128
fa.sort ............................................132
factor.congruence......................................133
factor.fit...........................................136
factor.model.........................................137
factor.residuals .......................................138
factor.rotate .........................................139
factor.scores.........................................141
factor.stats..........................................143
factor2cluster ........................................145
fisherz............................................147
galton ............................................148
geometric.mean.......................................149
glb.algebraic.........................................150
Gleser............................................153
Gorsuch...........................................154
Harman ...........................................155
Harman.5 ..........................................156
Harman.8 ..........................................157
Harman.political ......................................159
harmonic.mean .......................................160
headTail...........................................161
heights............................................162
ICC .............................................163
iclust.............................................165
ICLUST.cluster.......................................170
iclust.diagram........................................171
ICLUST.graph .......................................173
ICLUST.rgraph .......................................176
ICLUST.sort.........................................178
income............................................180
interp.median ........................................181
iqitems............................................182
irt.1p.............................................184
irt.fa.............................................186
irt.item.diff.rasch ......................................190
irt.responses.........................................191
kaiser ............................................193
4Rtopics documented:
KMO ............................................194
logistic............................................195
lowerUpper .........................................197
make.keys..........................................198
mardia............................................200
mat.sort ...........................................202
matrix.addition .......................................203
mediate ...........................................204
mixed.cor ..........................................207
msq .............................................209
mssd.............................................215
multi.hist ..........................................217
neo..............................................218
omega............................................220
omega.graph.........................................228
outlier............................................230
p.rep.............................................231
paired.r ...........................................233
pairs.panels .........................................234
parcels............................................237
partial.r ...........................................238
peas .............................................239
phi..............................................241
phi.demo ..........................................242
phi2tetra...........................................243
plot.psych..........................................244
polar.............................................247
polychor.matrix.......................................248
predict.psych ........................................249
principal...........................................251
print.psych .........................................255
Promax ...........................................256
psych.misc .........................................259
r.test.............................................262
rangeCorrection.......................................264
read.clipboard........................................265
rescale............................................267
residuals.psych .......................................268
reverse.code.........................................269
sat.act ............................................270
scaling.fits..........................................271
scatter.hist..........................................272
Schmid ...........................................273
schmid............................................274
score.alpha .........................................276
score.irt ...........................................278
score.multiple.choice ....................................280
scoreItems..........................................282
00.psych 5
scoreOverlap ........................................287
scrub.............................................290
SD..............................................291
setCor............................................292
sim..............................................296
sim.anova ..........................................301
sim.congeneric .......................................304
sim.hierarchical.......................................305
sim.item...........................................307
sim.multilevel........................................310
sim.structure.........................................312
sim.VSS...........................................314
simulation.circ .......................................315
smc .............................................317
spider ............................................318
splitHalf...........................................319
statsBy............................................323
structure.diagram ......................................328
structure.list.........................................331
superMatrix.........................................332
table2matrix.........................................333
test.psych ..........................................334
tetrachoric..........................................336
thurstone ..........................................341
tr...............................................342
Tucker............................................343
vegetables..........................................344
VSS.............................................345
VSS.parallel.........................................349
VSS.plot...........................................350
VSS.scree..........................................351
winsor............................................352
withinBetween .......................................353
Yule.............................................355
Index 358
00.psych A package for personality, psychometric, and psychological research
Description
Overview of the psych package.
The psych package has been developed at Northwestern University to include functions most use-
ful for personality and psychological research. Some of the functions (e.g., read.clipboard,
describe,pairs.panels,error.bars ) are useful for basic data entry and descriptive analyses.
600.psych
Use help(package="psych") for a list of all functions. Two vignettes are included as part of the
package. The overview provides examples of using psych in many applications.
Psychometric applications include routines (fa for principal axes (fm="pa"), minimum residual
(fm="minres"), maximum likelihood (fm="mle") and weighted least squares (fm="wls") factor
analysis as well as functions to do Schmid Leiman transformations (schmid) to transform a hi-
erarchical factor structure into a bifactor solution. Factor or components transformations to a target
matrix include the standard Promax transformation (Promax), a transformation to a cluster target,
or to any simple target matrix (target.rot) as well as the ability to call many of the GPAro-
tation functions. Functions for determining the number of factors in a data matrix include Very
Simple Structure (VSS) and Minimum Average Partial correlation (MAP). An alternative approach
to factor analysis is Item Cluster Analysis (ICLUST). Reliability coefficients alpha (score.items,
score.multiple.choice), beta (ICLUST) and McDonald’s omega (omega and omega.graph) as
well as Guttman’s six estimates of internal consistency reliability (guttman) and the six measures
of Intraclass correlation coefficients (ICC) discussed by Shrout and Fleiss are also available.
The scoreItems, and score.multiple.choice functions may be used to form single or multiple
scales from sets of dichotomous, multilevel, or multiple choice items by specifying scoring keys.
Additional functions make for more convenient descriptions of item characteristics. Functions un-
der development include 1 and 2 parameter Item Response measures. The tetrachoric,polychoric
and irt.fa functions are used to find 2 parameter descriptions of item functioning.
A number of procedures have been developed as part of the Synthetic Aperture Personality Assess-
ment (SAPA) project. These routines facilitate forming and analyzing composite scales equivalent
to using the raw data but doing so by adding within and between cluster/scale item correlations.
These functions include extracting clusters from factor loading matrices (factor2cluster), syn-
thetically forming clusters from correlation matrices (cluster.cor), and finding multiple ((mat.regress)
and partial ((partial.r) correlations from correlation matrices.
Functions to generate simulated data with particular structures include sim.circ (for circumplex
structures), sim.item (for general structures) and sim.congeneric (for a specific demonstration
of congeneric measurement). The functions sim.congeneric and sim.hierarchical can be used
to create data sets with particular structural properties. A more general form for all of these is
sim.structural for generating general structural models. These are discussed in more detail in
the vignette (psych_for_sem).
Functions to apply various standard statistical tests include p.rep and its variants for testing the
probability of replication, r.con for the confidence intervals of a correlation, and r.test to test
single, paired, or sets of correlations.
In order to study diurnal or circadian variations in mood, it is helpful to use circular statistics. Func-
tions to find the circular mean (circadian.mean), circular (phasic) correlations (circadian.cor)
and the correlation between linear variables and circular variables (circadian.linear.cor) sup-
plement a function to find the best fitting phase angle (cosinor) for measures taken with a fixed
period (e.g., 24 hours).
The most recent development version of the package is always available for download as a source
file from the repository at http://personality-project.org/r/src/contrib/.
Details
Two vignettes (overview.pdf) and psych_for_sem.pdf) are useful introductions to the package. They
may be found as vignettes in R or may be downloaded from http://personality-project.org/
r/book/overview.pdf and http://personality-project.org/r/book/psych_for_sem.pdf.
00.psych 7
The psych package was originally a combination of multiple source files maintained at the http:
//personality-project.org/r repository: “useful.r", VSS.r., ICLUST.r, omega.r, etc.“useful.r"
is a set of routines for easy data entry (read.clipboard), simple descriptive statistics (describe),
and splom plots combined with correlations (pairs.panels, adapted from the help files of pairs).
Those files have now been replaced with a single package.
The vss routines allow for testing the number of factors (vss), showing plots (VSS.plot) of good-
ness of fit, and basic routines for estimating the number of factors/components to extract by using
the MAP’s procedure, the examining the scree plot (VSS.scree) or comparing with the scree of an
equivalent matrix of random numbers (VSS.parallel).
In addition, there are routines for hierarchical factor analysis using Schmid Leiman tranformations
(omega,omega.graph) as well as Item Cluster analysis (ICLUST,ICLUST.graph).
The more important functions in the package are for the analysis of multivariate data, with an
emphasis upon those functions useful in scale construction of item composites.
When given a set of items from a personality inventory, one goal is to combine these into higher
level item composites. This leads to several questions:
1) What are the basic properties of the data? describe reports basic summary statistics (mean,
sd, median, mad, range, minimum, maximum, skew, kurtosis, standard error) for vectors, columns
of matrices, or data.frames. describeBy provides descriptive statistics, organized by one or more
grouping variables. pairs.panels shows scatter plot matrices (SPLOMs) as well as histograms and
the Pearson correlation for scales or items. error.bars will plot variable means with associated
confidence intervals. error.bars will plot confidence intervals for both the x and y coordinates.
corr.test will find the significance values for a matrix of correlations.
2) What is the most appropriate number of item composites to form? After finding either standard
Pearson correlations, or finding tetrachoric or polychoric correlations using a wrapper (poly.mat)
for John Fox’s hetcor function, the dimensionality of the correlation matrix may be examined. The
number of factors/components problem is a standard question of factor analysis, cluster analysis,
or principal components analysis. Unfortunately, there is no agreed upon answer. The Very Simple
Structure (VSS) set of procedures has been proposed as on answer to the question of the optimal
number of factors. Other procedures (VSS.scree,VSS.parallel,fa.parallel, and MAP) also
address this question.
3) What are the best composites to form? Although this may be answered using principal compo-
nents (principal), principal axis (factor.pa) or minimum residual (factor.minres) factor anal-
ysis (all part of the fa function) and to show the results graphically (fa.diagram), it is sometimes
more useful to address this question using cluster analytic techniques. Previous versions of ICLUST
(e.g., Revelle, 1979) have been shown to be particularly successful at forming maximally consistent
and independent item composites. Graphical output from ICLUST.graph uses the Graphviz dot
language and allows one to write files suitable for Graphviz. If Rgraphviz is available, these graphs
can be done in R.
Graphical organizations of cluster and factor analysis output can be done using cluster.plot
which plots items by cluster/factor loadings and assigns items to that dimension with the highest
loading.
4) How well does a particular item composite reflect a single construct? This is a question of relia-
bility and general factor saturation. Multiple solutions for this problem result in (Cronbach’s) alpha
(alpha,score.items), (Revelle’s) Beta (ICLUST), and (McDonald’s) omega (both omega hierar-
chical and omega total). Additional reliability estimates may be found in the guttman function.
800.psych
This can also be examined by applying irt.fa Item Response Theory techniques using factor
analysis of the tetrachoric or polychoric correlation matrices and converting the results into the
standard two parameter parameterization of item difficulty and item discrimination. Information
functions for the items suggest where they are most effective.
5) For some applications, data matrices are synthetically combined from sampling different items
for different people. So called Synthetic Aperture Personality Assessement (SAPA) techniques
allow the formation of large correlation or covariance matrices even though no one person has
taken all of the items. To analyze such data sets, it is easy to form item composites based upon the
covariance matrix of the items, rather than original data set. These matrices may then be analyzed
using a number of functions (e.g., cluster.cor,factor.pa,ICLUST,principal,mat.regress,
and factor2cluster.
6) More typically, one has a raw data set to analyze. alpha will report several reliablity estimates
as well as item-whole correlations for items forming a single scale, score.items will score data
sets on multiple scales, reporting the scale scores, item-scale and scale-scale correlations, as well
as coefficient alpha, alpha-1 and G6+. Using a ‘keys’ matrix (created by make.keys or by hand),
scales can have overlapping or independent items. score.multiple.choice scores multiple choice
items or converts multiple choice items to dichtomous (0/1) format for other functions.
An additional set of functions generate simulated data to meet certain structural properties. sim.anova
produces data simulating a 3 way analysis of variance (ANOVA) or linear model with or with out
repeated measures. sim.item creates simple structure data, sim.circ will produce circumplex
structured data, sim.dichot produces circumplex or simple structured data for dichotomous items.
These item structures are useful for understanding the effects of skew, differential item endorsement
on factor and cluster analytic soutions. sim.structural will produce correlation matrices and data
matrices to match general structural models. (See the vignette).
When examining personality items, some people like to discuss them as representing items in a
two dimensional space with a circumplex structure. Tests of circumplex fit circ.tests have been
developed. When representing items in a circumplex, it is convenient to view them in polar coor-
dinates.
Additional functions for testing the difference between two independent or dependent correlation
r.test, to find the phi or Yule coefficients from a two by table, or to find the confidence interval
of a correlation coefficient.
Ten data sets are included: bfi represents 25 personality items thought to represent five factors of
personality, iqitems has 14 multiple choice iq items. sat.act has data on self reported test scores
by age and gender. galton Galton’s data set of the heights of parents and their children. peas
recreates the original Galton data set of the genetics of sweet peas. heights and cubits provide
even more Galton data, vegetables provides the Guilford preference matrix of vegetables. cities
provides airline miles between 11 US cities (demo data for multidimensional scaling).
Package: psych
Type: Package
Version: 1.4.3
Date: 2014–March–25
License: GPL version 2 or newer
Index:
00.psych 9
psych A package for personality, psychometric, and psychological research.
Useful data entry and descriptive statistics
read.clipboard shortcut for reading from the clipboard
read.clipboard.csv shortcut for reading comma delimited files from clipboard
read.clipboard.lower shortcut for reading lower triangular matrices from the clipboard
read.clipboard.upper shortcut for reading upper triangular matrices from the clipboard
describe Basic descriptive statistics useful for psychometrics
describe.by Find summary statistics by groups
statsBy Find summary statistics by a grouping variable, including within and between correlation matrices.
headtail combines the head and tail functions for showing data sets
pairs.panels SPLOM and correlations for a data matrix
corr.test Correlations, sample sizes, and p values for a data matrix
cor.plot graphically show the size of correlations in a correlation matrix
multi.hist Histograms and densities of multiple variables arranged in matrix form
skew Calculate skew for a vector, each column of a matrix, or data.frame
kurtosi Calculate kurtosis for a vector, each column of a matrix or dataframe
geometric.mean Find the geometric mean of a vector or columns of a data.frame
harmonic.mean Find the harmonic mean of a vector or columns of a data.frame
error.bars Plot means and error bars
error.bars.by Plot means and error bars for separate groups
error.crosses Two way error bars
interp.median Find the interpolated median, quartiles, or general quantiles.
rescale Rescale data to specified mean and standard deviation
table2df Convert a two dimensional table of counts to a matrix or data frame
Data reduction through cluster and factor analysis
fa Combined function for principal axis, minimum residual, weighted least squares,
and maximum likelihood factor analysis
factor.pa Do a principal Axis factor analysis (deprecated)
factor.minres Do a minimum residual factor analysis (deprecated)
factor.wls Do a weighted least squares factor analysis (deprecated)
fa.graph Show the results of a factor analysis or principal components analysis graphically
fa.diagram Show the results of a factor analysis without using Rgraphviz
fa.sort Sort a factor or principal components output
fa.extension Apply the Dwyer extension for factor loadingss
principal Do an eigen value decomposition to find the principal components of a matrix
fa.parallel Scree test and Parallel analysis
fa.parallel.poly Scree test and Parallel analysis for polychoric matrices
factor.scores Estimate factor scores given a data matrix and factor loadings
guttman 8 different measures of reliability (6 from Guttman (1945)
irt.fa Apply factor analysis to dichotomous items to get IRT parameters
iclust Apply the ICLUST algorithm
10 00.psych
ICLUST.graph Graph the output from ICLUST using the dot language
ICLUST.rgraph Graph the output from ICLUST using rgraphviz
kaiser Apply kaiser normalization before rotating
polychoric Find the polychoric correlations for items and find item thresholds
poly.mat Find the polychoric correlations for items (uses J. Fox’s hetcor)
omega Calculate the omega estimate of factor saturation (requires the GPArotation package)
omega.graph Draw a hierarchical or Schmid Leiman orthogonalized solution (uses Rgraphviz)
partial.r Partial variables from a correlation matrix
predict Predict factor/component scores for new data
schmid Apply the Schmid Leiman transformation to a correlation matrix
score.items Combine items into multiple scales and find alpha
score.multiple.choice Combine items into multiple scales and find alpha and basic scale statistics
set.cor Find Cohen’s set correlation between two sets of variables
smc Find the Squared Multiple Correlation (used for initial communality estimates)
tetrachoric Find tetrachoric correlations and item thresholds
polyserial Find polyserial and biserial correlations for item validity studies
mixed.cor Form a correlation matrix from continuous, polytomous, and dichotomous items
VSS Apply the Very Simple Structure criterion to determine the appropriate number of factors.
VSS.parallel Do a parallel analysis to determine the number of factors for a random matrix
VSS.plot Plot VSS output
VSS.scree Show the scree plot of the factor/principal components
MAP Apply the Velicer Minimum Absolute Partial criterion for number of factors
Functions for reliability analysis (some are listed above as well).
alpha Find coefficient alpha and Guttman Lambda 6 for a scale (see also score.items)
guttman 8 different measures of reliability (6 from Guttman (1945)
omega Calculate the omega estimates of reliability (requires the GPArotation package)
omegaSem Calculate the omega estimates of reliability using a Confirmatory model (requires the sem package)
ICC Intraclass correlation coefficients
score.items Combine items into multiple scales and find alpha
glb.algebraic The greates lower bound found by an algebraic solution (requires Rcsdp). Written by Andreas Moeltner
Procedures particularly useful for Synthetic Aperture Personality Assessment
alpha Find coefficient alpha and Guttman Lambda 6 for a scale (see also score.items)
make.keys Create the keys file for score.items or cluster.cor
correct.cor Correct a correlation matrix for unreliability
count.pairwise Count the number of complete cases when doing pair wise correlations
cluster.cor find correlations of composite variables from larger matrix
cluster.loadings find correlations of items with composite variables from a larger matrix
eigen.loadings Find the loadings when doing an eigen value decomposition
fa Do a minimal residual or principal axis factor analysis and estimate factor scores
fa.extension Extend a factor analysis to a set of new variables
factor.pa Do a Principal Axis factor analysis and estimate factor scores
00.psych 11
factor2cluster extract cluster definitions from factor loadings
factor.congruence Factor congruence coefficient
factor.fit How well does a factor model fit a correlation matrix
factor.model Reproduce a correlation matrix based upon the factor model
factor.residuals Fit = data - model
factor.rotate “hand rotate" factors
guttman 8 different measures of reliability
mat.regress standardized multiple regression from raw or correlation matrix input
polyserial polyserial and biserial correlations with massive missing data
tetrachoric Find tetrachoric correlations and item thresholds
Functions for generating simulated data sets
sim The basic simulation functions
sim.anova Generate 3 independent variables and 1 or more dependent variables for demonstrating ANOVA
and lm designs
sim.circ Generate a two dimensional circumplex item structure
sim.item Generate a two dimensional simple structure with particular item characteristics
sim.congeneric Generate a one factor congeneric reliability structure
sim.minor Simulate nfact major and nvar/2 minor factors
sim.structural Generate a multifactorial structural model
sim.irt Generate data for a 1, 2, 3 or 4 parameter logistic model
sim.VSS Generate simulated data for the factor model
phi.demo Create artificial data matrices for teaching purposes
sim.hierarchical Generate simulated correlation matrices with hierarchical or any structure
sim.spherical Generate three dimensional spherical data (generalization of circumplex to 3 space)
Graphical functions (require Rgraphviz) – deprecated
structure.graph Draw a sem or regression graph
fa.graph Draw the factor structure from a factor or principal components analysis
omega.graph Draw the factor structure from an omega analysis(either with or without the Schmid Leiman transformation)
ICLUST.graph Draw the tree diagram from ICLUST
Graphical functions that do not require Rgraphviz
diagram A general set of diagram functions.
structure.diagram Draw a sem or regression graph
fa.diagram Draw the factor structure from a factor or principal components analysis
omega.diagram Draw the factor structure from an omega analysis(either with or without the Schmid Leiman transformation)
ICLUST.diagram Draw the tree diagram from ICLUST
plot.psych A call to plot various types of output (e.g. from irt.fa, fa, omega, iclust
12 00.psych
cor.plot A heat map display of correlations
spider Spider and radar plots (circular displays of correlations)
Circular statistics (for circadian data analysis)
circadian.cor Find the correlation with e.g., mood and time of day
circadian.linear.cor Correlate a circular value with a linear value
circadian.mean Find the circular mean of each column of a a data set
cosinor Find the best fitting phase angle for a circular data set
Miscellaneous functions
comorbidity Convert base rate and comorbity to phi, Yule and tetrachoric
df2latex Convert a data.frame or matrix to a LaTeX table
dummy.code Convert categorical data to dummy codes
fisherz Apply the Fisher r to z transform
fisherz2r Apply the Fisher z to r transform
ICC Intraclass correlation coefficients
cortest.mat Test for equality of two matrices (see also cortest.normal, cortest.jennrich )
cortest.bartlett Test whether a matrix is an identity matrix
paired.r Test for the difference of two paired or two independent correlations
r.con Confidence intervals for correlation coefficients
r.test Test of significance of r, differences between rs.
p.rep The probability of replication given a p, r, t, or F
phi Find the phi coefficient of correlation from a 2 x 2 table
phi.demo Demonstrate the problem of phi coefficients with varying cut points
phi2poly Given a phi coefficient, what is the polychoric correlation
phi2poly.matrix Given a phi coefficient, what is the polychoric correlation (works on matrices)
polar Convert 2 dimensional factor loadings to polar coordinates.
scaling.fits Compares alternative scaling solutions and gives goodness of fits
scrub Basic data cleaning
tetrachor Finds tetrachoric correlations
thurstone Thurstone Case V scaling
tr Find the trace of a square matrix
wkappa weighted and unweighted versions of Cohen’s kappa
Yule Find the Yule Q coefficient of correlation
Yule.inv What is the two by two table that produces a Yule Q with set marginals?
Yule2phi What is the phi coefficient corresponding to a Yule Q with set marginals?
Yule2tetra Convert one or a matrix of Yule coefficients to tetrachoric coefficients.
Functions that are under development and not recommended for casual use
00.psych 13
irt.item.diff.rasch IRT estimate of item difficulty with assumption that theta = 0
irt.person.rasch Item Response Theory estimates of theta (ability) using a Rasch like model
Data sets included in the psych package
bfi represents 25 personality items thought to represent five factors of personality
Thurstone 8 different data sets with a bifactor structure
cities The airline distances between 11 cities (used to demonstrate MDS)
epi.bfi 13 personality scales
iqitems 14 multiple choice iq items
msq 75 mood items
sat.act Self reported ACT and SAT Verbal and Quantitative scores by age and gender
Tucker Correlation matrix from Tucker
galton Galton’s data set of the heights of parents and their children
heights Galton’s data set of the relationship between height and forearm (cubit) length
cubits Galton’s data table of height and forearm length
peas Galton‘s data set of the diameters of 700 parent and offspring sweet peas
vegetables Guilford‘s preference matrix of vegetables (used for thurstone)
A debugging function that may also be used as a demonstration of psych.
test.psych Run a test of the major functions on 5 different data sets. Primarily for development purposes.
Although the output can be used as a demo of the various functions.
Note
Development versions (source code) of this package are maintained at the repository http://
personality-project.org/r along with further documentation. Specify that you are download-
ing a source package.
Some functions require other packages. Specifically, omega and schmid require the GPArotation
package, ICLUST.rgraph and fa.graph require Rgraphviz but have alternatives using the diagram
functions. i.e.:
function requires
omega GPArotation
schmid GPArotation
poly.mat polychor
phi2poly polychor
polychor.matrix polychor
ICLUST.rgraph Rgraphviz
fa.graph Rgraphviz
structure.graph Rgraphviz
glb.algebraic Rcsdp
14 ability
Author(s)
William Revelle
Department of Psychology
Northwestern University
Evanston, Illiniois
http://personality-project.org/revelle.html
Maintainer: William Revelle <revelle@northwestern.edu>
References
A general guide to personality theory and research may be found at the personality-project http:
//personality-project.org. See also the short guide to R at http://personality-project.
org/r. In addition, see
Revelle, W. (in preparation) An Introduction to Psychometric Theory with applications in R. Springer.
at http://personality-project.org/r/book/
Examples
#See the separate man pages
#to test most of the psych package run the following
#test.psych()
ability 16 ability items scored as correct or incorrect.
Description
16 multiple choice ability items 1525 subjects taken from the Synthetic Aperture Personality As-
sessment (SAPA) web based personality assessment project are saved as iqitems. Those data are
shown as examples of how to score multiple choice tests and analyses of response alternatives.
When scored correct or incorrect, the data are useful for demonstrations of tetrachoric based factor
analysis irt.fa and finding tetrachoric correlations.
Usage
data(iqitems)
Format
A data frame with 1525 observations on the following 16 variables. The number following the name
is the item number from SAPA.
reason.4 Basic reasoning questions
reason.16 Basic reasoning question
reason.17 Basic reasoning question
ability 15
reason.19 Basic reasoning question
letter.7 In the following alphanumeric series, what letter comes next?
letter.33 In the following alphanumeric series, what letter comes next?
letter.34 In the following alphanumeric series, what letter comes next
letter.58 In the following alphanumeric series, what letter comes next?
matrix.45 A matrix reasoning task
matrix.46 A matrix reasoning task
matrix.47 A matrix reasoning task
matrix.55 A matrix reasoning task
rotate.3 Spatial Rotation of type 1.2
rotate.4 Spatial Rotation of type 1.2
rotate.6 Spatial Rotation of type 1.1
rotate.8 Spatial Rotation of type 2.3
Details
16 items were sampled from 80 items given as part of the SAPA (http://sapa-project.org)
project (Revelle, Wilt and Rosenthal, 2009; Condon and Revelle, 2014) to develop online measures
of ability. These 16 items reflect four lower order factors (verbal reasoning, letter series, matrix
reasoning, and spatial rotations. These lower level factors all share a higher level factor (’g’).
This data set may be used to demonstrate item response functions, tetrachoric correlations, or
irt.fa as well as omega estimates of of reliability and hierarchical structure.
In addition, the data set is a good example of doing item analysis to examine the empirical response
probabilities of each item alternative as a function of the underlying latent trait. When doing this,
it appears that two of the matrix reasoning problems do not have monotonically increasing trace
lines for the probability correct. At moderately high ability (theta = 1) there is a decrease in the
probability correct from theta = 0 and theta = 2.
Source
The example data set is taken from the Synthetic Aperture Personality Assessment personality
and ability test at http://sapa-project.org. The data were collected with David Condon from
8/08/12 to 8/31/12.
References
Revelle, William, Wilt, Joshua, and Rosenthal, Allen (2010) Personality and Cognition: The Personality-
Cognition Link. In Gruszka, Alexandra and Matthews, Gerald and Szymura, Blazej (Eds.) Hand-
book of Individual Differences in Cognition: Attention, Memory and Executive Control, Springer.
Condon, David and Revelle, William, (2014) The International Cognitive Ability Resource: Devel-
opment and initial validation of a public-domain measure. Intelligence, 43, 52-64.
16 affect
Examples
data(ability)
#not run
# ability.irt <- irt.fa(ability)
# ability.scores <- score.irt(ability.irt,ability)
affect Two data sets of affect and arousal scores as a function of personality
and movie conditions
Description
A recurring question in the study of affect is the proper dimensionality and the relationship to
various personality dimensions. Here is a data set taken from two studies of mood and arousal
using movies to induce affective states.
Usage
data(affect)
Details
These are data from two studies conducted in the Personality, Motivation and Cognition Laboratory
at Northwestern University. Both studies used a similar methodology:
Collection of pretest data using 5 scales from the Eysenck Personality Inventory and items taken
from the Motivational State Questionnaire (see msq. In addition, state and trait anxiety measures
were given. In the “maps" study, the Beck Depression Inventory was given also.
Then subjects were randomly assigned to one of four movie conditions: 1: Frontline. A documen-
tary about the liberation of the Bergen-Belsen concentration camp. 2: Halloween. A horror film. 3:
National Geographic, a nature film about the Serengeti plain. 4: Parenthood. A comedy. Each film
clip was shown for 9 minutes. Following this the MSQ was given again.
Data from the MSQ were scored for Energetic and Tense Arousal (EA and TA) as well as Positive
and Negative Affect (PA and NA).
Study flat had 170 participants, study maps had 160.
These studies are described in more detail in various publications from the PMC lab. In particular,
Revelle and Anderson, 1997 and Rafaeli and Revelle (2006). An analysis of these data has also
appeared in Smillie et al. (2012).
Source
Data collected at the Personality, Motivation, and Cognition Laboratory, Northwestern University.
alpha 17
References
Revelle, William and Anderson, Kristen Joan (1997) Personality, motivation and cognitive perfor-
mance: Final report to the Army Research Institute on contract MDA 903-93-K-0008
Rafaeli, Eshkol and Revelle, William (2006), A premature consensus: Are happiness and sadness
truly opposite affects? Motivation and Emotion, 30, 1, 1-12.
Smillie, Luke D. and Cooper, Andrew and Wilt, Joshua and Revelle, William (2012) Do Extraverts
Get More Bang for the Buck? Refining the Affective-Reactivity Hypothesis of Extraversion. Jour-
nal of Personality and Social Psychology, 103 (2), 206-326.
Examples
data(affect)
describeBy(affect[-1],group="Film")
pairs.panels(affect[14:17],bg=c("red","black","white","blue")[affect$Film],pch=21,
main="Affect varies by movies ")
errorCircles("EA2","TA2",data=affect,group="Film",labels=c("Sad","Fear","Neutral","Humor")
, main="Enegetic and Tense Arousal by Movie condition")
errorCircles(x="PA2",y="NA2",data=affect,group="Film",labels=c("Sad","Fear","Neutral","
Humor"), main="Positive and Negative Affect by Movie condition")
alpha Find two estimates of reliability: Cronbach’s alpha and Guttman’s
Lambda 6.
Description
Internal consistency measures of reliability range from ωhto αto ωt. This function reports two
estimates: Cronbach’s coefficient αand Guttman’s λ6. Also reported are item - whole correlations,
αif an item is omitted, and item means and standard deviations.
Usage
alpha(x, keys=NULL,cumulative=FALSE, title=NULL, max=10,na.rm = TRUE,
check.keys=FALSE,n.iter=1,delete=TRUE,use="pairwise",warnings=TRUE,n.obs=NULL)
Arguments
xA data.frame or matrix of data, or a covariance or correlation matrix
keys If some items are to be reversed keyed, then either specify the direction of all
items or just a vector of which items to reverse
title Any text string to identify this run
cumulative should means reflect the sum of items or the mean of the items. The default
value is means.
max the number of categories/item to consider if reporting category frequencies. De-
faults to 10, passed to link{response.frequencies}
18 alpha
na.rm The default is to remove missing values and find pairwise correlations
check.keys if TRUE, then find the first principal component and reverse key items with
negative loadings. Give a warning if this happens.
n.iter Number of iterations if bootstrapped confidence intervals are desired
delete Delete items with no variance and issue a warning
use Options to pass to the cor function: "everything", "all.obs", "complete.obs",
"na.or.complete", or "pairwise.complete.obs". The default is "pairwise"
warnings By default print a warning and a message that items were reversed. Suppress the
message if warnings = FALSE
n.obs If using correlation matrices as input, by specify the number of observations, we
can find confidence intervals
Details
Alpha is one of several estimates of the internal consistency reliability of a test.
Surprisingly, more than a century after Spearman (1904) introduced the concept of reliability to psy-
chologists, there are still multiple approaches for measuring it. Although very popular, Cronbach’s
α(1951) underestimates the reliability of a test and over estimates the first factor saturation.
α(Cronbach, 1951) is the same as Guttman’s λ3 (Guttman, 1945) and may be found by
λ3=n
n−11−tr(~
V)x
Vx=n
n−1
Vx−tr(~
Vx)
Vx
=α
Perhaps because it is so easy to calculate and is available in most commercial programs, alpha is
without doubt the most frequently reported measure of internal consistency reliability. Alpha is the
mean of all possible spit half reliabilities (corrected for test length). For a unifactorial test, it is a
reasonable estimate of the first factor saturation, although if the test has any microstructure (i.e., if
it is “lumpy") coefficients β(Revelle, 1979; see ICLUST) and ωh(see omega) are more appropriate
estimates of the general factor saturation. ωt(see omega) is a better estimate of the reliability of the
total test.
Guttman’s Lambda 6 (G6) considers the amount of variance in each item that can be accounted for
the linear regression of all of the other items (the squared multiple correlation or smc), or more
precisely, the variance of the errors, e2
j, and is
λ6= 1 −Pe2
j
Vx
= 1 −P(1 −r2
smc)
Vx
.
The squared multiple correlation is a lower bound for the item communality and as the number of
items increases, becomes a better estimate.
G6 is also sensitive to lumpyness in the test and should not be taken as a measure of unifactorial
structure. For lumpy tests, it will be greater than alpha. For tests with equal item loadings, alpha >
G6, but if the loadings are unequal or if there is a general factor, G6 > alpha. alpha is a generaliza-
tion of an earlier estimate of reliability for tests with dichotomous items developed by Kuder and
Richardson, known as KR20, and a shortcut approximation, KR21. (See Revelle, in prep).
Alpha and G6 are both positive functions of the number of items in a test as well as the average
intercorrelation of the items in the test. When calculated from the item variances and total test
alpha 19
variance, as is done here, raw alpha is sensitive to differences in the item variances. Standardized
alpha is based upon the correlations rather than the covariances.
A useful index of the quality of the test that is linear with the number of items and the average
correlation is the Signal/Noise ratio where
s/n =n¯r
1−n¯r
(Cronbach and Gleser, 1964; Revelle and Condon (in press)).
More complete reliability analyses of a single scale can be done using the omega function which
finds ωhand ωtbased upon a hierarchical factor analysis.
Alternative functions score.items and cluster.cor will also score multiple scales and report
more useful statistics. “Standardized" alpha is calculated from the inter-item correlations and will
differ from raw alpha.
Four alternative item-whole correlations are reported, three are conventional, one unique. raw.r
is the correlation of the item with the entire scale, not correcting for item overlap. std.r is the
correlation of the item with the entire scale, if each item were standardized. r.drop is the correlation
of the item with the scale composed of the remaining items. Although each of these are conventional
statistics, they have the disadvantage that a) item overlap inflates the first and b) the scale is different
for each item when an item is dropped. Thus, the fourth alternative, r.cor, corrects for the item
overlap by subtracting the item variance but then replaces this with the best estimate of common
variance, the smc. This is similar to a suggestion by Cureton (1966).
If some items are to be reversed keyed then they can be specified by either item name or by item
location. (Look at the 3rd and 4th examples.) Automatic reversal can also be done, and this is
based upon the sign of the loadings on the first principal component (Example 5). This requires the
check.keys option to be TRUE. Previous versions defaulted to have check.keys=TRUE, but some
users complained that this made it too easy to find alpha without realizing that some items had been
reversed (even though a warning was issued!). Thus, I have set the default to be check.keys=FALSE
with a warning that some items need to be reversed (if this is the case). To suppress these warnings,
set warnings=FALSE.
Scores are based upon the simple averages (or totals) of the items scored. Reversed items are
subtracted from the maximum + minimum item response for all the items.
When using raw data, standard errors for the raw alpha are calculated using equation 2 and 3 from
Duhhachek and Iacobucci (2004). This is problematic because some simulations suggest these
values are too small. It is probably better to use bootstrapped value
Bootstrapped resamples are found if n.iter > 1. These are returned as the boot object. They may be
plotted or described.
Value
total a list containing
raw_alpha alpha based upon the covariances
std.alpha The standarized alpha based upon the correlations
G6(smc) Guttman’s Lambda 6 reliability
average_r The average interitem correlation
mean For data matrices, the mean of the scale formed by summing the items
20 alpha
sd For data matrices, the standard deviation of the total score
alpha.drop A data frame with all of the above for the case of each item being removed one
by one.
item.stats A data frame including
nnumber of complete cases for the item
raw.r The correlation of each item with the total score, not corrected for item overlap.
std.r The correlation of each item with the total score (not corrected for item overlap)
if the items were all standardized
r.cor Item whole correlation corrected for item overlap and scale reliability
r.drop Item whole correlation for this item against the scale without this item
mean for data matrices, the mean of each item
sd For data matrices, the standard deviation of each item
response.freq For data matrices, the frequency of each item response (if less than 20)
boot a 6 column by n.iter matrix of boot strapped resampled values
Unidim An index of unidimensionality
Fit The fit of the off diagonal matrix
Note
By default, items that correlate negatively with the overall scale will be reverse coded. This option
may be turned off by setting check.keys = FALSE. If items are reversed, then each item is subtracted
from the minimum item response + maximum item response where min and max are taken over all
items. Thus, if the items intentionally differ in range, the scores will be off by a constant. See
scoreItems for a solution.
Two experimental measures of Goodness of Fit are returned in the output: Unidim and Fit. They
are not printed or displayed, but are available for analysis. The first is an index of how well the
modeled average correlations actually reproduce the original correlation matrix. The second is how
well the modeled correlations reproduce the off diagonal elements of the matrix. Both are indices
of squared residuals compared to the squared original correlations. These two measures are under
development and might well be modified or dropped in subsequent versions.
Author(s)
William Revelle
References
Cronbach, L.J. (1951) Coefficient alpha and the internal strucuture of tests. Psychometrika, 16,
297-334.
Cureton, E. (1966). Corrected item-test correlations. Psychometrika, 31(1):93-96.
Cronbach, L.J. and Gleser G.C. (1964)The signal/noise ratio in the comparison of reliability coeffi-
cients. Educational and Psychological Measurement, 24 (3) 467-480.
Duhachek, A. and Iacobucci, D. (2004). Alpha’s standard error (ase): An accurate and precise
confidence interval estimate. Journal of Applied Psychology, 89(5):792-808.
Bechtoldt 21
Guttman, L. (1945). A basis for analyzing test-retest reliability. Psychometrika, 10 (4), 255-282.
Revelle, W. (in preparation) An introduction to psychometric theory with applications in R. Springer.
(Available online at http://personality-project.org/r/book).
Revelle, W. Hierarchical Cluster Analysis and the Internal Structure of Tests. Multivariate Behav-
ioral Research, 1979, 14, 57-74.
Revelle, W. and Condon, D.C. Reliability. In Irwing, P., Booth, T. and Hughes, D. (Eds). the
Wiley-Blackwell Handbook of Psychometric Testing (in press).
Revelle, W. and Zinbarg, R. E. (2009) Coefficients alpha, beta, omega and the glb: comments on
Sijtsma. Psychometrika, 74 (1) 1145-154.
See Also
omega,ICLUST,guttman,scoreItems,cluster.cor
Examples
set.seed(42) #keep the same starting values
#four congeneric measures
r4 <- sim.congeneric()
alpha(r4)
#nine hierarchical measures -- should actually use omega
r9 <- sim.hierarchical()
alpha(r9)
# examples of two independent factors that produce reasonable alphas
#this is a case where alpha is a poor indicator of unidimensionality
two.f <- sim.item(8)
#specify which items to reverse key by name
alpha(two.f,keys=c("V1","V2","V7","V8"))
#by location
alpha(two.f,keys=c(1,2,7,8))
#automatic reversal base upon first component
alpha(two.f)
#an example with discrete item responses -- show the frequencies
items <- sim.congeneric(N=500,short=FALSE,low=-2,high=2,
categorical=TRUE) #500 responses to 4 discrete items with 5 categories
a4 <- alpha(items$observed) #item response analysis of congeneric measures
a4
#summary just gives Alpha
summary(a4)
Bechtoldt Seven data sets showing a bifactor solution.
22 Bechtoldt
Description
Holzinger-Swineford (1937) introduced the bifactor model of a general factor and uncorrelated
group factors. The Holzinger data sets are original 14 * 14 matrix from their paper as well as a 9
*9 matrix used as an example by Joreskog. The Thurstone correlation matrix is a 9 * 9 matrix of
correlations of ability items. The Reise data set is 16 * 16 correlation matrix of mental health items.
The Bechtholdt data sets are both 17 x 17 correlation matrices of ability tests.
Usage
data(Thurstone)
data(Thurstone.33)
data(Holzinger)
data(Holzinger.9)
data(Bechtoldt)
data(Bechtoldt.1)
data(Bechtoldt.2)
data(Reise)
Details
Holzinger and Swineford (1937) introduced the bifactor model (one general factor and several group
factors) for mental abilities. This is a nice demonstration data set of a hierarchical factor structure
that can be analyzed using the omega function or using sem. The bifactor model is typically used in
measures of cognitive ability.
There are several ways to analyze such data. One is to use the omega function to do a hierarchical
factoring using the schmid-leiman transformation. Another is to a regular factor analysis and use
either a bifactor or biquartimin rotation. These latter two functions implement the Jennrich and
Bentler (2011) bifactor and biquartimin transformations.
The 14 variables are ordered to reflect 3 spatial tests, 3 mental speed tests, 4 motor speed tests, and
4 verbal tests. The sample size is 355.
Another data set from Holzinger (Holzinger.9) represents 9 cognitive abilities (Holzinger, 1939)
and is used as an example by Karl Joreskog (2003) for factor analysis by the MINRES algorithm
and also appears in the LISREL manual as example NPV.KM.
Another classic data set is the 9 variable Thurstone problem which is discussed in detail by R.
P. McDonald (1985, 1999) and and is used as example in the sem package as well as in the PROC
CALIS manual for SAS. These nine tests were grouped by Thurstone and Thurstone, 1941 (based on
other data) into three factors: Verbal Comprehension, Word Fluency, and Reasoning. The original
data came from Thurstone and Thurstone (1941) but were reanalyzed by Bechthold (1961) who
broke the data set into two. McDonald, in turn, selected these nine variables from the larger set of
17 found in Bechtoldt.2. The sample size is 213.
Another set of 9 cognitive variables attributed to Thurstone (1933) is the data set of 4,175 students
reported by Professor Brigham of Princeton to the College Entrance Examination Board. This set
does not show a clear bifactor solution but is included as a demonstration of the differences between
a maximimum likelihood factor analysis solution versus a principal axis factor solution.
More recent applications of the bifactor model are to the measurement of psychological status. The
Reise data set is a correlation matrix based upon >35,000 observations to the Consumer Assess-
Bechtoldt 23
ment of Health Care Provideers and Systems survey instrument. Reise, Morizot, and Hays (2007)
describe a bifactor solution based upon 1,000 cases.
The five factors from Reise et al. reflect Getting care quickly (1-3), Doctor communicates well (4-
7), Courteous and helpful staff (8,9), Getting needed care (10-13), and Health plan customer service
(14-16).
The two Bechtoldt data sets are two samples from Thurstone and Thurstone (1941). They include 17
variables, 9 of which were used by McDonald to form the Thurstone data set. The sample sizes are
212 and 213 respectively. The six proposed factors reflect memory, verbal, words, space, number
and reasoning with three markers for all expect the rote memory factor. 9 variables from this set
appear in the Thurstone data set.
Two more data sets with similar structures are found in the Harman data set.
• Bechtoldt.1: 17 x 17 correlation matrix of ability tests, N = 212.
• Bechtoldt.2: 17 x 17 correlation matrix of ability tests, N = 213.
• Holzinger: 14 x 14 correlation matrix of ability tests, N = 355
• Holzinger.9: 9 x 9 correlation matrix of ability tests, N = 145
• Reise: 16 x 16 correlation matrix of health satisfaction items. N = 35,000
• Thurstone: 9 x 9 correlation matrix of ability tests, N = 213
• Thurstone.33: Another 9 x 9 correlation matrix of ability items, N=4175
Source
Holzinger: Holzinger and Swineford (1937)
Reise: Steve Reise (personal communication)
sem help page (for Thurstone)
References
Bechtoldt, Harold, (1961). An empirical study of the factor analysis stability hypothesis. Psy-
chometrika, 26, 405-432.
Holzinger, Karl and Swineford, Frances (1937) The Bi-factor method. Psychometrika, 2, 41-54
Holzinger, K., & Swineford, F. (1939). A study in factor analysis: The stability of a bifactor
solution. Supplementary Educational Monograph, no. 48. Chicago: University of Chicago Press.
McDonald, Roderick P. (1999) Test theory: A unified treatment. L. Erlbaum Associates. Mahwah,
N.J.
Reise, Steven and Morizot, Julien and Hays, Ron (2007) The role of the bifactor model in resolving
dimensionality issues in health outcomes measures. Quality of Life Research. 16, 19-31.
Thurstone, Louis Leon (1933) The theory of multiple factors. Edwards Brothers, Inc. Ann Arbor
Thurstone, Louis Leon and Thurstone, Thelma (Gwinn). (1941) Factorial studies of intelligence.
The University of Chicago Press. Chicago, Il.
24 bestScales
Examples
if(!require(GPArotation)) {message("I am sorry, to run omega requires GPArotation")
} else {
#holz <- omega(Holzinger,4, title = "14 ability tests from Holzinger-Swineford")
#bf <- omega(Reise,5,title="16 health items from Reise")
#omega(Reise,5,labels=colnames(Reise),title="16 health items from Reise")
thur.om <- omega(Thurstone,title="9 variables from Thurstone") #compare with
thur.bf <- fa(Thurstone,3,rotate="biquartimin")
factor.congruence(thur.om,thur.bf)
}
bestScales A set of functions for factorial and empirical scale construction
Description
When constructing scales through rational, factorial, or empirical means, it is useful to examine the
content of the items that relate most highly to each other (e.g., the factor loadings of fa.lookup
of a set of items) , or to some specific set of criteria (e.g., bestScales). Given a dictionary of
item content, these routines will sort by factor loading or criteria correlations and display the item
content.
Usage
bestScales(x, criteria, cut = 0.1, n.item = 10, overlap = FALSE,
dictionary = NULL, digits = 2)
bestItems(x,criteria=1,cut=.3, abs=TRUE, dictionary=NULL,cor=TRUE,digits=2)
lookup(x,y,criteria=NULL)
fa.lookup(f,dictionary,digits=2)
item.lookup(f,m, dictionary,cut=.3, digits = 2)
Arguments
xA data matrix or data frame depending upon the function.
yA data matrix or data frame or a vector
criteria Which variables (by name or location) should be the empirical target for bestScales
and bestItems
fThe object returned from either a factor analysis (fa) or a principal components
analysis (principal)
cut Return all values in abs(x[,c1]) > cut.
abs if TRUE, sort by absolute value in bestItems
dictionary a data.frame with rownames corresponding to rownames in the f$loadings ma-
trix or colnames of the data matrix or correlation matrix, and entries (may be
multiple columns) of item content.
bestScales 25
mA data frame of item means
cor if x is not a square matrix, should correlations be found?
n.item How many items make up an empirical scale
overlap Are the correlations with other criteria fair game for bestScales
digits round to digits
Details
bestItems and lookup are simple helper functions to summarize correlation matrices or factor
loading matrices. bestItems will sort the specified column (criteria) of x on the basis of the (ab-
solute) value of the column. The return as a default is just the rowname of the variable with those
absolute values > cut. If there is a dictionary of item content and item names, then include the
contents as a two column matrix with rownames corresponding to the item name and then as many
fields as desired for item content. (See the example dictionary bfi.dictionary).
lookup is used by bestItems and will find values in c1 of y that match those in x. It returns those
rows of y of that match x. Suppose that you have a "dictionary" of the many variables in a study
but you want to consider a small subset of them in a data set x. Then, you can find the entries in the
dictionary corresponding to x by lookup(rownames(x),y) If the column is not specified, then it will
match by rownames(y).
fa.lookup is used when examining the output of a factor analysis and one wants the correspond-
ing variable names and contents. The returned object may then be printed in LaTex by using the
df2latex function with the char option set to TRUE.
Similarly, given a correlation matrix, r, of the x variables, if you want to find the items that most
correlate with another item or scale, and then show the contents of that item from the dictionary,
bestItems(r,c1=column number or name of x, contents = y)
bestScales will find up to n.items that have absolute correlations with a criterion greater than cut.
If the overlap option is FALSE (default) the other criteria are not used.
item.lookup combines the output from a factor analysis fa with simple descriptive statistics (a
data frame of means) with a dictionary. Items are grouped by factor loadings > cut, and then sorted
by item mean. This allows a better understanding of how a scale works, in terms of the meaning of
the item endorsements.
Value
bestScales returns the correlation of the empirically constructed scale with each criteria and the
items used in the scale. If a dictionary is specified, it also returns a list (value) that shows the item
content. Also returns the keys list so that scales can be found using cluster.cor or scoreItems.
bestItems returns a sorted list of factor loadings or correlations with the labels as provided in the
dictionary.
lookup is a very simple implementation of the match function.
fa.lookup takes a factor/cluster analysis object (or just a keys like matrix), sorts it using fa.sort
and then matches by row.name to the corresponding dictionary entries.
26 bfi
Note
To create a dictionary, create an object with row names as the item numbers, and the columns as the
item content. See the link{bfi.dictionary} as an example.
Note
Although empirical scale construction is appealing, it has the basic problem of capitalizing on
chance. Thus, be careful of over interpreting the results unless working with large samples.
Author(s)
William Revelle
References
Revelle, W. (in preparation) An introduction to psychometric theory with applications in R. Springer.
(Available online at http://personality-project.org/r/book).
See Also
fa,iclust,principal
Examples
bs <- bestScales(bfi,criteria=c("gender","education","age"),dictionary=bfi.dictionary)
bs
f5 <- fa(bfi,5)
m <- colMeans(bfi,na.rm=TRUE)
item.lookup(f5,m,dictionary=bfi.dictionary[2])
fa.lookup(f5,dictionary=bfi.dictionary[2]) #just show the item content, not the source of the items
bfi 25 Personality items representing 5 factors
Description
25 personality self report items taken from the International Personality Item Pool (ipip.ori.org)
were included as part of the Synthetic Aperture Personality Assessment (SAPA) web based per-
sonality assessment project. The data from 2800 subjects are included here as a demonstration
set for scale construction, factor analysis, and Item Response Theory analysis. Three additional
demographic variables (sex, education, and age) are also included.
Usage
data(bfi)
data(bfi.dictionary)
bfi 27
Format
A data frame with 2800 observations on the following 28 variables. (The q numbers are the SAPA
item numbers).
A1 Am indifferent to the feelings of others. (q_146)
A2 Inquire about others’ well-being. (q_1162)
A3 Know how to comfort others. (q_1206)
A4 Love children. (q_1364)
A5 Make people feel at ease. (q_1419)
C1 Am exacting in my work. (q_124)
C2 Continue until everything is perfect. (q_530)
C3 Do things according to a plan. (q_619)
C4 Do things in a half-way manner. (q_626)
C5 Waste my time. (q_1949)
E1 Don’t talk a lot. (q_712)
E2 Find it difficult to approach others. (q_901)
E3 Know how to captivate people. (q_1205)
E4 Make friends easily. (q_1410)
E5 Take charge. (q_1768)
N1 Get angry easily. (q_952)
N2 Get irritated easily. (q_974)
N3 Have frequent mood swings. (q_1099
N4 Often feel blue. (q_1479)
N5 Panic easily. (q_1505)
O1 Am full of ideas. (q_128)
O2 Avoid difficult reading material.(q_316)
O3 Carry the conversation to a higher level. (q_492)
O4 Spend time reflecting on things. (q_1738)
O5 Will not probe deeply into a subject. (q_1964)
gender Males = 1, Females =2
education 1 = HS, 2 = finished HS, 3 = some college, 4 = college graduate 5 = graduate degree
age age in years
Details
The first 25 items are organized by five putative factors: Agreeableness, Conscientiousness, Ex-
traversion, Neuroticism, and Opennness. The scoring key is created using make.keys, the scores
are found using score.items.
These five factors are a useful example of using irt.fa to do Item Response Theory based latent
factor analysis of the polychoric correlation matrix. The endorsement plots for each item, as well
as the item information functions reveal that the items differ in their quality.
28 bfi
The item data were collected using a 6 point response scale: 1 Very Inaccurate 2 Moderately Inac-
curate 3 Slightly Inaccurate 4 Slightly Accurate 5 Moderately Accurate 6 Very Accurate
as part of the Synthetic Apeture Personality Assessment (SAPA http://sapa-project.org) project.
To see an example of the data collection technique, visit http://SAPA-project.org. The items
given were sampled from the International Personality Item Pool of Lewis Goldberg using the sam-
pling technique of SAPA. This is a sample data set taken from the much larger SAPA data bank.
Source
The items are from the ipip (Goldberg, 1999). The data are from the SAPA project (Revelle, Wilt
and Rosenthal, 2010) , collected Spring, 2010 ( http://sapa-project.org).
References
Goldberg, L.R. (1999) A broad-bandwidth, public domain, personality inventory measuring the
lower-level facets of several five-factor models. In Mervielde, I. and Deary, I. and De Fruyt, F. and
Ostendorf, F. (eds) Personality psychology in Europe. 7. Tilburg University Press. Tilburg, The
Netherlands.
Revelle, W., Wilt, J., and Rosenthal, A. (2010) Personality and Cognition: The Personality-Cognition
Link. In Gruszka, A. and Matthews, G. and Szymura, B. (Eds.) Handbook of Individual Differences
in Cognition: Attention, Memory and Executive Control, Springer.
See Also
bi.bars to show the data by age and gender, irt.fa for item factor analysis applying the irt model.
Examples
data(bfi)
describe(bfi)
keys.list <-
list(agree=c("-A1","A2","A3","A4","A5"),conscientious=c("C1","C2","C3","-C4","-C5"),
extraversion=c("-E1","-E2","E3","E4","E5"),neuroticism=c("N1","N2","N3","N4","N5"),
openness = c("O1","-O2","O3","O4","-O5"))
keys <- make.keys(bfi,keys.list)
scores <- scoreItems(keys[1:27,],bfi[1:27]) #don't score age
scores
#show the use of the fa.lookup with a dictionary
fa.lookup(keys,bfi.dictionary[,1:4])
bi.bars 29
bi.bars Draw pairs of bargraphs based on two groups
Description
When showing e.g., age or education distributions for two groups, it is convenient to plot them back
to back. bi.bars will do so.
Usage
bi.bars(x,grp,horiz,color,label=NULL,...)
Arguments
xThe data to be drawn
grp a grouping variable.
horiz horizontal (default) or vertical bars
color colors for the two groups – defaults to blue and red
label If specified, labels for the dependent axis
... Further parameters to pass to the graphing program
Details
A trivial, if useful, function to draw back to back histograms/barplots. One for each group.
Value
a graphic
Author(s)
William Revelle
Examples
data(bfi)
with(bfi,{bi.bars(age,gender,ylab="Age",main="Age by males and females")
bi.bars(education,gender,xlab="Education",main="Education by gender",horiz=FALSE)})
30 biplot.psych
biplot.psych Draw biplots of factor or component scores by factor or component
loadings
Description
Extends the biplot function to the output of fa,fa.poly or principal. Will plot factor scores and
factor loadings in the same graph. If the number of factors > 2, then all pairs of factors are plotted.
Factor score histograms are plotted on the diagonal. The input is the resulting object from fa,
principal, or }code{linkfa.poly with the scores=TRUE option. Points may be colored according
to other criteria.
Usage
## S3 method for class 'psych'
biplot(x, labels=NULL,cex=c(.75,1),main="Biplot from fa",
hist.col="cyan",xlim.s=c(-3,3),ylim.s=c(-3,3),xlim.f=c(-1,1),ylim.f=c(-1,1),
maxpoints=100,adjust=1.2,col,pos, arrow.len = 0.1,pch=16,choose=NULL,
cuts=1,cutl=.0,group=NULL,...)
Arguments
xThe output from fa,fa.poly or principal with the scores=TRUE option
labels if NULL, draw the points with the plot character (pch) specified. To identify the
data points, specify labels= 1:n where n is the number of observations, or labels
=rownames(data) where data was the data set analyzed by the factor analysis.
cex A vector of plot sizes of the data labels and of the factor labels
main A main title for a two factor biplot
hist.col If plotting more than two factors, the color of the histogram of the factor scores
xlim.s x limits of the scores. Defaults to plus/minus three sigma
ylim.s y limits of the scores.Defaults to plus/minus three sigma
xlim.f x limits of the factor loadings.Defaults to plus/minus 1.0
ylim.f y limits of the factor loadings.Defaults to plus/minus 1.0
maxpoints When plotting 3 (or more) dimensions, at what size should we switch from
plotting "o" to plotting "."
adjust an adjustment factor in the histogram
col a vector of colors for the data points and for the factor loading labels
pos If plotting labels, what position should they be in? 1=below, 2=left, 3 top, 4
right. If missing, then the assumption is that labels should be printed instead of
data points.
arrow.len the length of the arrow head
pch The plotting character to use. pch=16 gives reasonable size dots. pch="." gives
tiny points. If adding colors, use pch between 21 and 25. (see examples).
biplot.psych 31
choose Plot just the specified factors
cuts Do not label cases with abs(factor scores) < cuts) (Actually, the distance of the
x and y scores from 0)
cutl Do not label variables with communalities in the two space < cutl
group A vector of a grouping variable for the scores. Show a different color and symbol
for each group.
... more options for graphics
Details
Uses the generic biplot function to take the output of a factor analysis fa,fa.poly or principal com-
ponents analysis principal and plot the factor/component scores along with the factor/component
loadings.
This is an extension of the generic biplot function to allow more control over plotting points in a
two space and also to plot three or more factors (two at time).
This will work for objects produced by fa,fa.poly or principal if they applied to the original
data matrix. If however, one has a correlation matrix based upon the output from tetrachoric or
polychoric, and has done either fa or principal on the correlations, then obviously, we can not
do a biplot. However, both of those functions produce a weights matrix, which, in combination
with the original data can be used to find the scores by using factor.scores. Since biplot.psych is
looking for two elements of the x object: x$loadings and x$scores, you can create the appropriate
object to plot. See the third example.
Author(s)
William Revelle
See Also
fa,fa.poly,principal,fa.plot,pairs.panels
Examples
#the standard example
data(USArrests)
fa2 <- fa(USArrests,2,scores=TRUE)
biplot(fa2,labels=rownames(USArrests))
# plot the 3 factor solution
data(bfi)
fa3 <- fa(bfi[1:200,1:15],3,scores=TRUE)
biplot(fa3)
#just plot factors 1 and 3 from that solution
biplot(fa3,choose=c(1,3))
#
fa2 <- fa(bfi[16:25],2) #factor analysis
fa2$scores <- fa2$scores[1:100,] #just take the first 100
#now plot with different colors and shapes for males and females
32 block.random
biplot(fa2,pch=c(24,21)[bfi[1:100,"gender"]],group =bfi[1:100,"gender"],
main="Biplot of Conscientiousness and Neuroticism by gender")
r <- cor(bfi[1:200,1:10], use="pairwise") #find the correlations
f2 <- fa(r,2)
x <- list()
x$scores <- factor.scores(bfi[1:200,1:10],f2)
x$loadings <- f2$loadings
class(x) <- c('psych','fa')
biplot(x,main="biplot from correlation matrix and factor scores")
block.random Create a block randomized structure for n independent variables
Description
Random assignment of n subjects with an equal number in all of N conditions may done by block
randomization, where the block size is the number of experimental conditions. The number of
Independent Variables and the number of levels in each IV are specified as input. The output is a
the block randomized design.
Usage
block.random(n, ncond = NULL)
Arguments
nThe number of subjects to randomize. Must be a multiple of the number of
experimental conditions
ncond The number of conditions for each IV. Defaults to 2 levels for one IV. If more
than one IV, specify as a vector. If names are provided, they are used, otherwise
the IVs are labeled as IV1 ... IVn
Value
blocks A matrix of subject numbers, block number, and randomized levels for each IV
Note
Prepared for a course on Research Methods in Psychology http://personality-project.org/
revelle/syllabi/205/205.syllabus.html
Author(s)
William Revelle
blot 33
Examples
br <- block.random(n=24,c(2,3))
pairs.panels(br)
br <- block.random(96,c(time=4,drug=3,sex=2))
pairs.panels(br)
blot Bond’s Logical Operations Test – BLOT
Description
35 items for 150 subjects from Bond’s Logical Operations Test. A good example of Item Response
Theory analysis using the Rasch model. One parameter (Rasch) analysis and two parameter IRT
analyses produce somewhat different results.
Usage
data(blot)
Format
A data frame with 150 observations on 35 variables. The BLOT was developed as a paper and
pencil test for children to measure Logical Thinking as discussed by Piaget and Inhelder.
Details
Bond and Fox apply Rasch modeling to a variety of data sets. This one, Bond’s Logical Operations
Test, is used as an example of Rasch modeling for dichotomous items. In their text (p 56), Bond and
Fox report the results using WINSTEPS. Those results are consistent (up to a scaling parameter)
with those found by the rasch function in the ltm package. The WINSTEPS seem to produce
difficulty estimates with a mean item difficulty of 0, whereas rasch from ltm has a mean difficulty
of -1.52. In addition, rasch seems to reverse the signs of the difficulty estimates when reporting the
coefficients and is effectively reporting "easiness".
However, when using a two parameter model, one of the items (V12) behaves very differently.
This data set is useful when comparing 1PL, 2PL and 2PN IRT models.
Source
The data are taken (with kind permission from Trevor Bond) from the webpage http://homes.jcu.edu.au/~edtgb/book/data/Bond87.txt
and read using read.fwf.
References
T.G. Bond. BLOT:Bond’s Logical Operations Test. Townsville, Australia: James Cook Univer-
sity. (Original work published 1976), 1995.
T. Bond and C. Fox. (2007) Applying the Rasch model: Fundamental measurement in the human
sciences. Lawrence Erlbaum, Mahwah, NJ, US, 2 edition.
34 bock
See Also
See also the irt.fa and associated plot functions.
Examples
data(blot)
#not run
#library(ltm)
#bblot.rasch <- rasch(blot, constraint = cbind(ncol(blot) + 1, 1)) #a 1PL model
#blot.2pl <- ltm(blot~z1) #a 2PL model
#do the same thing with functions in psych
#blot.fa <- irt.fa(blot) # a 2PN model
#plot(blot.fa)
bock Bock and Liberman (1970) data set of 1000 observations of the LSAT
Description
An example data set used by McDonald (1999) as well as other discussions of Item Response
Theory makes use of a data table on 10 items (two sets of 5) from the Law School Admissions Test
(LSAT). Included in this data set is the original table as well as the reponses for 1000 subjects on
the first set (Figure Classification) and second set (Debate).
Usage
data(bock)
Format
A data frame with 32 observations on the following 8 variables.
index 32 response patterns
Q1 Responses to item 1
Q2 Responses to item 2
Q3 Responses to item 3
Q4 Responses to item 4
Q5 Responses to item 5
Ob6 count of observations for the section 6 test
Ob7 count of observations for the section 7 test
Two other data sets are derived from the bock dataset. These are converted using the table2df
function.
lsat6 reponses to 5 items for 1000 subjects on section 6
lsat7 reponses to 5 items for 1000 subjects on section 7
burt 35
Details
The lsat6 data set is analyzed in the ltm package as well as by McDonald (1999). lsat7 is another
1000 subjects on part 7 of the LSAT. Both sets are described by Bock and Lieberman (1970).
Both sets are useful examples of testing out IRT procedures and showing the use of tetrachoric
correlations and item factor analysis using the irt.fa function.
Source
R. Darrell Bock and M. Lieberman (1970). Fitting a response model for dichotomously scored
items. Psychometrika, 35(2):179-197.
References
R.P. McDonald. Test theory: A unified treatment. L. Erlbaum Associates, Mahwah, N.J., 1999.
Examples
data(bock)
responses <- table2df(bock.table[,2:6],count=bock.table[,7],
labs= paste("lsat6.",1:5,sep=""))
describe(responses)
## maybe str(bock.table) ; plot(bock.table) ...
burt 11 emotional variables from Burt (1915)
Description
Cyril Burt reported an early factor analysis with a circumplex structure of 11 emotional variables in
1915. 8 of these were subsequently used by Harman in his text on factor analysis. Unfortunately, it
seems as if Burt made a mistake for the matrix is not positive definite. With one change from .87 to
.81 the matrix is positive definite.
Usage
data(burt)
Format
A correlation matrix based upon 172 "normal school age children aged 9-12".
Sociality Sociality
Sorrow Sorrow
Tenderness Tenderness
Joy Joy
Wonder Wonder
36 burt
Elation Elation
Disgust Disgust
Anger Anger
Sex Sex
Fear Fear
Subjection Subjection
Details
The Burt data set is interesting for several reasons. It seems to be an early example of the orga-
nizaton of emotions into an affective circumplex, a subset of it has been used for factor analysis
examples (see Harman.Burt, and it is an example of how typos affect data. The original data matrix
has one negative eigenvalue. With the replacement of the correlation between Sorrow and Tender-
ness from .87 to .81, the matrix is positive definite.
Alternatively, using cor.smooth, the matrix can be made positive definite as well, although cor.smooth
makes more (but smaller) changes.
Source
(retrieved from the web at http://www.biodiversitylibrary.org/item/95822#790) Following a sugges-
tion by Jan DeLeeuw.
References
Burt, C.General and Specific Factors underlying the Primary Emotions. Reports of the British As-
sociation for the Advancement of Science, 85th meeting, held in Manchester, September 7-11, 1915.
London, John Murray, 1916, p. 694-696 (retrieved from the web at http://www.biodiversitylibrary.org/item/95822#790)
See Also
Harman.Burt in the Harman dataset and cor.smooth
Examples
data(burt)
eigen(burt)$values #one is negative!
burt.new <- burt
burt.new[2,3] <- burt.new[3,2] <- .81
eigen(burt.new)$values #all are positive
bs <- cor.smooth(burt)
round(burt.new - bs,3)
circ.tests 37
circ.tests Apply four tests of circumplex versus simple structure
Description
Rotations of factor analysis and principal components analysis solutions typically try to represent
correlation matrices as simple structured. An alternative structure, appealing to some, is a cir-
cumplex structure where the variables are uniformly spaced on the perimeter of a circle in a two
dimensional space. Generating these data is straightforward, and is useful for exploring alternative
solutions to affect and personality structure.
Usage
circ.tests(loads, loading = TRUE, sorting = TRUE)
Arguments
loads A matrix of loadings loads here
loading Are these loadings or a correlation matrix loading
sorting Should the variables be sorted sorting
Details
“A common model for representing psychological data is simple structure (Thurstone, 1947). Ac-
cording to one common interpretation, data are simple structured when items or scales have non-
zero factor loadings on one and only one factor (Revelle & Rocklin, 1979). Despite the common-
place application of simple structure, some psychological models are defined by a lack of simple
structure. Circumplexes (Guttman, 1954) are one kind of model in which simple structure is lack-
ing.
“A number of elementary requirements can be teased out of the idea of circumplex structure. First,
circumplex structure implies minimally that variables are interrelated; random noise does not a
circumplex make. Second, circumplex structure implies that the domain in question is optimally
represented by two and only two dimensions. Third, circumplex structure implies that variables do
not group or clump along the two axes, as in simple structure, but rather that there are always inter-
stitial variables between any orthogonal pair of axes (Saucier, 1992). In the ideal case, this quality
will be reflected in equal spacing of variables along the circumference of the circle (Gurtman, 1994;
Wiggins, Steiger, & Gaelick, 1981). Fourth, circumplex structure implies that variables have a con-
stant radius from the center of the circle, which implies that all variables have equal communality on
the two circumplex dimensions (Fisher, 1997; Gurtman, 1994). Fifth, circumplex structure implies
that all rotations are equally good representations of the domain (Conte & Plutchik, 1981; Larsen
& Diener, 1992). (Acton and Revelle, 2004)
Acton and Revelle reviewed the effectiveness of 10 tests of circumplex structure and found that
four did a particularly good job of discriminating circumplex structure from simple structure, or
circumplexes from ellipsoidal structures. Unfortunately, their work was done in Pascal and is not
easily available. Here we release R code to do the four most useful tests:
1 The Gap test of equal spacing
38 circ.tests
2 Fisher’s test of equality of axes
3 A test of indifference to Rotation
4 A test of equal Variance of squared factor loadings across arbitrary rotations.
To interpret the values of these various tests, it is useful to compare the particular solution to simu-
lated solutions representing pure cases of circumplex and simple structure. See the example output
from circ.simulation and compare these plots with the results of the circ.test.
Value
A list of four items is returned. These are the gap, fisher, rotation and variance test results.
gaps gap.test
fisher fisher.test
RT rotation.test
VT variance.test
Note
Of the 10 criterion discussed in Acton and Revelle (2004), these tests operationalize the four most
useful.
Author(s)
William Revelle
References
Acton, G. S. and Revelle, W. (2004) Evaluation of Ten Psychometric Criteria for Circumplex Struc-
ture. Methods of Psychological Research Online, Vol. 9, No. 1 http://personality-project.
org/revelle/publications/acton.revelle.mpr110_10.pdf
See Also
To understand the results of the circ.tests it it best to compare it to simulated values. Thus, see
circ.simulation,sim.circ
Examples
circ.data <- circ.sim(24,500)
circ.fa <- fa(circ.data,2)
plot(circ.fa,title="Circumplex Structure")
ct <- circ.tests(circ.fa)
#compare with non-circumplex data
simp.data <- item.sim(24,500)
simp.fa <- fa(simp.data,2)
plot(simp.fa,title="Simple Structure")
st <- circ.tests(simp.fa)
res <- rbind(ct[1:4],st[1:4])
rownames(res) <- c("circumplex","Simple")
cities 39
print(res,digits=2)
cities Distances between 11 US cities
Description
Airline distances between 11 US cities may be used as an example for multidimensional scaling or
cluster analysis.
Usage
data(cities)
Format
A data frame with 11 observations on the following 11 variables.
ATL Atlana, Georgia
BOS Boston, Massachusetts
ORD Chicago, Illinois
DCA Washington, District of Columbia
DEN Denver, Colorado
LAX Los Angeles, California
MIA Miami, Florida
JFK New York, New York
SEA Seattle, Washington
SFO San Francisco, California
MSY New Orleans, Lousianna
Details
An 11 x11 matrix of distances between major US airports. This is a useful demonstration of multiple
dimensional scaling.
city.location is a dataframe of longitude and latitude for those cities.
Note that the 2 dimensional MDS solution does not perfectly capture the data from these city dis-
tances. Boston, New York and Washington, D.C. are located slightly too far west, and Seattle and
LA are slightly too far south.
Source
http://www.timeanddate.com/worldclock/distance.html
40 cluster.fit
Examples
data(cities)
city.location[,1] <- -city.location[,1]
#not run
#an overlay map can be added if the package maps is available
#
#
#libary(maps)
#map("usa")
#title("MultiDimensional Scaling of US cities")
#points(city.location)
plot(city.location, xlab="Dimension 1", ylab="Dimension 2",
main ="Multidimensional scaling of US cities")
city.loc <- cmdscale(cities, k=2) #ask for a 2 dimensional solution round(city.loc,0)
city.loc <- -city.loc
city.loc <- rescale(city.loc,apply(city.location,2,mean),apply(city.location,2,sd))
points(city.loc,type="n")
text(city.loc,labels=names(cities))
cluster.fit cluster Fit: fit of the cluster model to a correlation matrix
Description
How well does the cluster model found by ICLUST fit the original correlation matrix? A similar
algorithm factor.fit is found in VSS. This function is internal to ICLUST but has more general
use as well.
In general, the cluster model is a Very Simple Structure model of complexity one. That is, every
item is assumed to represent only one factor/cluster. Cluster fit is an analysis of how well this model
reproduces a correlation matrix. Two measures of fit are given: cluster fit and factor fit. Cluster fit
assumes that variables that define different clusters are orthogonal. Factor fit takes the loadings
generated by a cluster model, finds the cluster loadings on all clusters, and measures the degree of
fit of this somewhat more complicated model. Because the cluster loadings are similar to, but not
identical to factor loadings, the factor fits found here and by factor.fit will be similar.
Usage
cluster.fit(original, load, clusters, diagonal = FALSE)
Arguments
original The original correlation matrix being fit
load Cluster loadings – that is, the correlation of individual items with the clusters,
corrected for item overlap
clusters The cluster structure
diagonal Should we fit the diagonal as well?
cluster.loadings 41
Details
The cluster model is similar to the factor model: R is fitted by C’C. Where C <- Cluster definition
matrix x the loading matrix. How well does this model approximate the original correlation matrix
and how does this compare to a factor model?
The fit statistic is a comparison of the original (squared) correlations to the residual correlations.
Fit = 1 - r*2/r2 where r* is the residual correlation of data - model and model = C’C.
Value
clusterfit The cluster model is a reduced form of the factor loading matrix. That is, it is
the product of the elements of the cluster matrix * the loading matrix.
factorfit How well does the complete loading matrix reproduce the correlation matrix?
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
References
http://personality-project.org/r/r.ICLUST.html
See Also
VSS,ICLUST,factor2cluster,cluster.cor,factor.fit
Examples
r.mat<- Harman74.cor$cov
iq.clus <- ICLUST(r.mat,nclusters =2)
fit <- cluster.fit(r.mat,iq.clus$loadings,iq.clus$clusters)
fit
cluster.loadings Find item by cluster correlations, corrected for overlap and reliability
Description
Given a n x n correlation matrix and a n x c matrix of -1,0,1 cluster weights for those n items on c
clusters, find the correlation of each item with each cluster. If the item is part of the cluster, correct
for item overlap. Part of the ICLUST set of functions, but useful for many item analysis problems.
Usage
cluster.loadings(keys, r.mat, correct = TRUE,SMC=TRUE)
42 cluster.loadings
Arguments
keys Cluster keys: a matrix of -1,0,1 cluster weights
r.mat A correlation matrix
correct Correct for reliability
SMC Use the squared multiple correlation as a communality estimate, otherwise use
the greatest correlation for each variable
Details
Given a set of items to be scored as (perhaps overlapping) clusters and the intercorrelation matrix
of the items, find the clusters and then the correlations of each item with each cluster. Correct for
item overlap by replacing the item variance with its average within cluster inter-item correlation.
Although part of ICLUST, this may be used in any SAPA (http://sapa-project.org) application
where we are interested in item- whole correlations of items and composite scales.
These loadings are particularly interpretable when sorted by absolute magnitude for each cluster
(see ICLUST.sort).
Value
loadings A matrix of item-cluster correlations (loadings)
cor Correlation matrix of the clusters
corrected Correlation matrix of the clusters, raw correlations below the diagonal, alpha on
diagonal, corrected for reliability above the diagonal
sd Cluster standard deviations
alpha alpha reliabilities of the clusters
G6 G6* Modified estimated of Guttman Lambda 6
count Number of items in the cluster
Note
Although part of ICLUST, this may be used in any SAPA application where we are interested in
item- whole correlations of items and composite scales.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
References
ICLUST: http://personality-project.org/r/r.ICLUST.html
See Also
ICLUST,factor2cluster,cluster.cor
cluster.plot 43
Examples
r.mat<- Harman74.cor$cov
clusters <- matrix(c(1,1,1,rep(0,24),1,1,1,1,rep(0,17)),ncol=2)
cluster.loadings(clusters,r.mat)
cluster.plot Plot factor/cluster loadings and assign items to clusters by their high-
est loading.
Description
Cluster analysis and factor analysis are procedures for grouping items in terms of a smaller number
of (latent) factors or (observed) clusters. Graphical presentations of clusters typically show tree
structures, although they can be represented in terms of item by cluster correlations.
Cluster.plot plots items by their cluster loadings (taken, e.g., from ICLUST) or factor loadings (taken,
eg., from fa). Cluster membership may be assigned apriori or may be determined in terms of the
highest (absolute) cluster loading for each item.
If the input is an object of class "kmeans", then the cluster centers are plotted.
Usage
cluster.plot(ic.results, cluster = NULL, cut = 0, labels=NULL,
title = "Cluster plot",pch=18,pos,show.points=TRUE,choose=NULL,...)
fa.plot(ic.results, cluster = NULL, cut = 0, labels=NULL,title,
jiggle=FALSE,amount=.02,pch=18,pos,show.points=TRUE,choose=NULL,...)
factor.plot(ic.results, cluster = NULL, cut = 0, labels=NULL,title,jiggle=FALSE,
amount=.02,pch=18,pos,show.points=TRUE,...) #deprecated
Arguments
ic.results A factor analysis or cluster analysis output including the loadings, or a matrix
of item by cluster correlations. Or the output from a kmeans cluster analysis.
cluster A vector of cluster membership
cut Assign items to clusters if the absolute loadings are > cut
labels If row.names exist they will be added to the plot, or, if they don’t, labels can
be specified. If labels =NULL, and there are no row names, then variables are
labeled by row number.)
title Any title
44 cluster.plot
jiggle When plotting with factor loadings that are almost identical, it is sometimes
useful to "jiggle" the points by jittering them. The default is to not jiggle.
amount if jiggle=TRUE, then how much should the points be jittered?
pch factor and clusters are shown with different pch values, starting at pch+1
pos Position of the text for labels for two dimensional plots. 1=below, 2 = left, 3 =
above, 4= right
show.points When adding labels to the points, should we show the points as well as the
labels. For many points, better to not show them, just the labels.
choose Specify the factor/clusters to plot
... Further options to plot
Details
Results of either a factor analysis or cluster analysis are plotted. Each item is assigned to its highest
loading factor, and then identified by variable name as well as cluster (by color). The cluster as-
signments can be specified to override the automatic clustering by loading. Both of these functions
may be called directly or by calling the generic plot function. (see example).
Value
Graphical output is presented.
Author(s)
William Revelle
See Also
ICLUST,ICLUST.graph,fa.graph,plot.psych
Examples
circ.data <- circ.sim(24,500)
circ.fa <- fa(circ.data,2)
plot(circ.fa,cut=.5)
f5 <- fa(bfi[1:25],5)
plot(f5,labels=colnames(bfi)[1:25],show.points=FALSE)
plot(f5,labels=colnames(bfi)[1:25],show.points=FALSE,choose=c(1,2,4))
cluster2keys 45
cluster2keys Convert a cluster vector (from e.g., kmeans) to a keys matrix suitable
for scoring item clusters.
Description
The output of the kmeans clustering function produces a vector of cluster membership. The score.items
and cluster.cor functions require a matrix of keys. cluster2keys does this.
May also be used to take the output of an ICLUST analysis and find a keys matrix. (By doing a call
to the factor2cluster function.
Usage
cluster2keys(c)
Arguments
cA vector of cluster assignments or an object of class “kmeans" that contains a
vector of clusters.
Details
Note that because kmeans will not reverse score items, the clusters defined by kmeans will not
necessarily match those of ICLUST with the same number of clusters extracted.
Value
keys A matrix of keys suitable for score.items or cluster.cor
Author(s)
William Revelle
See Also
cluster.cor,score.items,factor2cluster,make.keys
Examples
test.data <- Harman74.cor$cov
kc <- kmeans(test.data,4)
keys <- cluster2keys(kc)
keys #these match those found by ICLUST
cluster.cor(keys,test.data)
46 cohen.kappa
cohen.kappa Find Cohen’s kappa and weighted kappa coefficients for correlation
of two raters
Description
Cohen’s kappa (Cohen, 1960) and weighted kappa (Cohen, 1968) may be used to find the agreement
of two raters when using nominal scores.
weighted.kappa is (probability of observed matches - probability of expected matches)/(1 - prob-
ability of expected matches). Kappa just considers the matches on the main diagonal. Weighted
kappa considers off diagonal elements as well.
Usage
cohen.kappa(x, w=NULL,n.obs=NULL,alpha=.05)
wkappa(x, w = NULL) #deprectated
Arguments
xEither a two by n data with categorical values from 1 to p or a p x p table. If a
data array, a table will be found.
wA p x p matrix of weights. If not specified, they are set to be 0 (on the diagonal)
and (distance from diagonal) off the diagonal)^2.
n.obs Number of observations (if input is a square matrix.
alpha Probability level for confidence intervals
Details
When cateogorical judgments are made with two cateories, a measure of relationship is the phi
coefficient. However, some categorical judgments are made using more than two outcomes. For
example, two diagnosticians might be asked to categorize patients three ways (e.g., Personality
disorder, Neurosis, Psychosis) or to categorize the stages of a disease. Just as base rates affect
observed cell frequencies in a two by two table, they need to be considered in the n-way table
(Cohen, 1960).
Kappa considers the matches on the main diagonal. A penalty function (weight) may be applied to
the off diagonal matches. If the weights increase by the square of the distance from the diagonal,
weighted kappa is similar to an Intra Class Correlation (ICC).
Derivations of weighted kappa are sometimes expressed in terms of similarities, and sometimes in
terms of dissimilarities. In the latter case, the weights on the diagonal are 1 and the weights off the
diagonal are less than one. In this, if the weights are 1 - squared distance from the diagonal / k, then
the result is similar to the ICC (for any positive k).
cohen.kappa may use either similarity weighting (diagonal = 0) or dissimilarity weighting (diagonal
= 1) in order to match various published examples.
The input may be a two column data.frame or matrix with columns representing the two judges and
rows the subjects being rated. Alternatively, the input may be a square n x n matrix of counts or
cohen.kappa 47
proportion of matches. If proportions are used, it is necessary to specify the number of observations
(n.obs) in order to correctly find the confidence intervals.
The confidence intervals are based upon the variance estimates discussed by Fleiss, Cohen, and
Everitt who corrected the formulae of Cohen (1968) and Blashfield.
Value
kappa Unweighted kappa
weighted.kappa
The default weights are quadratric.
var.kappa Variance of kappa
var.weighted Variance of weighted kappa
n.obs number of observations
weight The weights used in the estimation of weighted kappa
confid The alpha/2 confidence intervals for unweighted and weighted kappa
plevel The alpha level used in determining the confidence limits
Note
As is true of many R functions, there are alternatives in other packages. The Kappa function in
the vcd package estimates unweighted and weighted kappa and reports the variance of the estimate.
The input is a square matrix. The ckappa and wkappa functions in the psy package take raw data
matrices.
To avoid confusion with Kappa (from vcd) or the kappa function from base, the function was orig-
inally named wkappa. With additional features modified from psy::ckappa to allow input with a
different number of categories, the function has been renamed cohen.kappa.
Unfortunately, to make it more confusing, the weights described by Cohen are a function of the
reciprocals of those discucssed by Fleiss and Cohen. The cohen.kappa function uses the appropriate
formula for Cohen or Fleiss-Cohen weights.
Author(s)
William Revelle
References
Banerjee, M., Capozzoli, M., McSweeney, L and Sinha, D. (1999) Beyond Kappa: A review of
interrater agreement measures The Canadian Journal of Statistics / La Revue Canadienne de Statis-
tique, 27, 3-23
Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and Psychological
Measurement, 20 37-46
Cohen, J. (1968). Weighted kappa: Nominal scale agreement provision for scaled disagreement or
partial credit. Psychological Bulletin, 70, 213-220.
Fleiss, J. L., Cohen, J. and Everitt, B.S. (1969) Large sample standard errors of kappa and weighted
kappa. Psychological Bulletin, 72, 332-327.
Zwick, R. (1988) Another look at interrater agreement. Psychological Bulletin, 103, 374 - 378.
48 cohen.kappa
Examples
#rating data (with thanks to Tim Bates)
rater1 = c(1,2,3,4,5,6,7,8,9) # rater one's ratings
rater2 = c(1,3,1,6,1,5,5,6,7) # rater one's ratings
cohen.kappa(x=cbind(rater1,rater2))
#data matrix taken from Cohen
cohen <- matrix(c(
0.44, 0.07, 0.09,
0.05, 0.20, 0.05,
0.01, 0.03, 0.06),ncol=3,byrow=TRUE)
#cohen.weights weight differences
cohen.weights <- matrix(c(
0,1,3,
1,0,6,
3,6,0),ncol=3)
cohen.kappa(cohen,cohen.weights,n.obs=200)
#cohen reports .492 and .348
#another set of weights
#what if the weights are non-symmetric
wc <- matrix(c(
0,1,4,
1,0,6,
2,2,0),ncol=3,byrow=TRUE)
cohen.kappa(cohen,wc)
#Cohen reports kw = .353
cohen.kappa(cohen,n.obs=200) #this uses the squared weights
fleiss.cohen <- 1 - cohen.weights/9
cohen.kappa(cohen,fleiss.cohen,n.obs=200)
#however, Fleiss, Cohen and Everitt weight similarities
fleiss <- matrix(c(
106, 10,4,
22,28, 10,
2, 12, 6),ncol=3,byrow=TRUE)
#Fleiss weights the similarities
weights <- matrix(c(
1.0000, 0.0000, 0.4444,
0.0000, 1.0000, 0.6667,
0.4444, 0.6667, 1.0000),ncol=3)
cohen.kappa(fleiss,weights,n.obs=200)
#another example is comparing the scores of two sets of twins
comorbidity 49
#data may be a 2 column matrix
#compare weighted and unweighted
#also look at the ICC for this data set.
twins <- matrix(c(
1, 2,
2, 3,
3, 4,
5, 6,
6, 7), ncol=2,byrow=TRUE)
cohen.kappa(twins)
#data may be explicitly categorical
x <- c("red","yellow","blue","red")
y <- c("red", "blue", "blue" ,"red")
xy.df <- data.frame(x,y)
ck <- cohen.kappa(xy.df)
ck
ck$agree
#finally, input can be a data.frame of ratings from more than two raters
ratings <- matrix(rep(1:5,4),ncol=4)
ratings[1,2] <- ratings[2,3] <- ratings[3,4] <- NA
ratings[2,1] <- ratings[3,2] <- ratings[4,3] <- 1
cohen.kappa(ratings)
comorbidity Convert base rates of two diagnoses and their comorbidity into phi,
Yule, and tetrachorics
Description
In medicine and clinical psychology, diagnoses tend to be categorical (someone is depressed or not,
someone has an anxiety disorder or not). Cooccurrence of both of these symptoms is called comor-
bidity. Diagnostic categories vary in their degree of comorbidity with other diagnostic categories.
From the point of view of correlation, comorbidity is just a name applied to one cell in a four fold
table. It is thus possible to analyze comorbidity rates by considering the probability of the separate
diagnoses and the probability of the joint diagnosis. This gives the two by two table needed for a
phi, Yule, or tetrachoric correlation.
Usage
comorbidity(d1, d2, com, labels = NULL)
Arguments
d1 Proportion of diagnostic category 1
d2 Proportion of diganostic category 2
com Proportion of comorbidity (diagnostic category 1 and 2)
labels Names of categories 1 and 2
50 cor.ci
Value
twobytwo The two by two table implied by the input
phi Phi coefficient of the two by two table
Yule Yule coefficient of the two by two table
tetra Tetrachoric coefficient of the two by two table
Author(s)
William Revelle
See Also
phi,Yule
Examples
comorbidity(.2,.15,.1,c("Anxiety","Depression"))
cor.ci Bootstrapped confidence intervals for raw and composite correlations
Description
Although normal theory provides confidence intervals for correlations, this is particularly problem-
atic with Synthetic Aperture Personality Assessment (SAPA) data where the individual items are
Massively Missing at Random. Bootstrapped confidence intervals are found for Pearson, Spearman,
Kendall, tetrachoric, or polychoric correlations and for scales made from those correlations.
Usage
cor.ci(x, keys = NULL, n.iter = 100, p = 0.05,overlap = FALSE,
poly = FALSE, method = "pearson", plot=TRUE,...)
corCi(x, keys = NULL, n.iter = 100, p = 0.05,overlap = FALSE,
poly = FALSE, method = "pearson", plot=TRUE,...)
Arguments
xThe raw data
keys If NULL, then the confidence intervals of the raw correlations are found. Oth-
erwise, composite scales are formed from the keys applied to the correlation
matrix (in a logic similar to cluster.cor but without the bells and whistles)
and the confidence of those composite scales intercorrelations.
n.iter The number of iterations to bootstrap over. This will be very slow if using
tetrachoric/or polychoric correlations.
cor.ci 51
pThe upper and lower confidence region will include 1-p of the distribution.
overlap If true, the correlation between overlapping scales is corrected for item overlap.
poly if FALSE, then find the correlations using the method specified (defaults to Pear-
son). If TRUE, the polychoric correlations will be found (slowly). Because the
polychoric function uses multicores (if available), and cor.ci does as well, the
number of cores used is options("mc.cores")^2.
method "pearson","spearman", "kendall"
plot Show the correlation plot with correlations scaled by the probability values. To
show the matrix in terms of the confidence intervals, use cor.plot.upperLowerCi.
... Other parameters for axis (e.g., cex.axis to change the font size, srt to rotate the
numbers in the plot)
Details
The original data are and correlations are found. If keys are specified (the normal case), then
composite scales based upon the correlations are found and reported. This is the same procedure as
done using cluster.cor or scoreItems.
Then, n.iter times, the data are recreated by sampling subjects (rows) with replacement and the
correlations (and composite scales) are found again (and again and again). Mean and standard
deviations of these values are calculated based upon the Fisher Z transform of the correlations.
Summary statistics include the original correlations and their confidence intervals. For those who
want the complete set of replications, those are available as an object in the resulting output.
Although particularly useful for SAPA (http://sapa-project.org) type data, this will work for
any normal data set as well.
Although the correlations are shown automatically as a cor.plot, it is possible to show the upper
and lower confidence intervals by using cor.plot.upperLowerCi. This will also return, invisibly,
a matrix for printing with the lower and upper bounds of the correlations shown below and above
the diagonal.
Value
rho The original (composite) correlation matrix.
means Mean (Fisher transformed) correlation
sds Standard deviation of Fisher transformed correlations
ci Mean +/- alpha/2 of the z scores as well as the alpha/2 and 1-alpha/2 quan-
tiles. These are labeled as lower.emp(ircal), lower.norm(al), upper.norm and
upper.emp.
replicates The observed replication values so one can do one’s own estimates
Author(s)
William Revelle
52 cor.plot
References
For SAPA type data, see Revelle, W., Wilt, J., and Rosenthal, A. (2010) Personality and Cognition:
The Personality-Cognition Link. In Gruszka, A. and Matthews, G. and Szymura, B. (Eds.) Hand-
book of Individual Differences in Cognition: Attention, Memory and Executive Control, Springer.
See Also
make.keys,cluster.cor, and scoreItems for forming synthetic correlation matrices from com-
posites of item correlations. See scoreOverlap for correcting for item overlap in scales. See
also corr.test for standard significance testing of correlation matrices. See also lowerCor for
finding and printing correlation matrices, as well as lowerMat for displaying them. Also see
cor.plot.upperLowerCi for displaying the confidence intervals graphically.
Examples
cor.ci(bfi[1:200,1:10]) # just the first 10 variables
#The keys have overlapping scales
keys.list <- list(agree=c("-A1","A2","A3","A4","A5"), conscientious= c("C1",
"C2","C3","-C4","-C5"),extraversion=c("-E1","-E2","E3","E4","E5"), neuroticism=
c("N1", "N2", "N3","N4","N5"), openness = c("O1","-O2","O3","O4","-O5"),
alpha=c("-A1","A2","A3","A4","A5","C1","C2","C3","-C4","-C5","N1","N2","N3","N4","N5"),
beta = c("-E1","-E2","E3","E4","E5","O1","-O2","O3","O4","-O5") )
keys <- make.keys(bfi,keys.list)
#do not correct for item overlap
rci <- cor.ci(bfi[1:200,],keys,n.iter=10,main="correlation with overlapping scales")
#also shows the graphic -note the overlap
#correct for overlap
rci <- cor.ci(bfi[1:200,],keys,overlap=TRUE, n.iter=10,main="Correct for overlap")
#show the confidence intervals
ci <- cor.plot.upperLowerCi(rci) #to show the upper and lower confidence intervals
ci #print the confidence intervals in matrix form
cor.plot Create an image plot for a correlation or factor matrix
Description
Correlation matrices may be shown graphically by using the image function to emphasize struc-
ture. This is a particularly useful tool for showing the structure of correlation matrices with a clear
structure. Partially meant for the pedagogical value of the graphic for teaching or discussing factor
analysis and other multivariate techniques.
Usage
corPlot(r,numbers=FALSE,colors=TRUE,n=51,main=NULL,zlim=c(-1,1),
show.legend=TRUE, labels=NULL,n.legend=10,keep.par=TRUE,select=NULL,
pval=NULL,cuts=c(.001,.01),cex,MAR,upper=TRUE,diag=TRUE,...)
cor.plot 53
cor.plot(r,numbers=FALSE,colors=TRUE,n=51,main=NULL,zlim=c(-1,1),
show.legend=TRUE, labels=NULL,n.legend=10,keep.par=TRUE,select=NULL,
pval=NULL,cuts=c(.001,.01),cex,MAR,upper=TRUE,diag=TRUE,...)
cor.plot.upperLowerCi(R,numbers=TRUE,cuts=c(.001,.01,.05),select=NULL,
main="Upper and lower confidence intervals of correlations",...)
Arguments
rA correlation matrix or the output of fa,principal or omega.
RThe object returned from cor.ci
numbers Display the numeric value of the correlations. Defaults to FALSE.
colors Defaults to TRUE and colors use colors from the colorRampPalette from red
through white to blue, but colors=FALSE will use a grey scale
nThe number of levels of shading to use. Defaults to 51
main A title. Defaults to "correlation plot"
zlim The range of values to color – defaults to -1 to 1
show.legend A legend (key) to the colors is shown on the right hand side
labels if NULL, use column and row names, otherwise use labels
n.legend How many categories should be labelled in the legend?
keep.par restore the graphic parameters when exiting
pval scale the numbers by their pvals, categorizing them based upon the values of
cuts
cuts Scale the numbers by the categories defined by pval < cuts
select Select the subset of variables to plot
cex Character size. Should be reduced a bit for large numbers of variables.
MAR Allows for adjustment of the margins if using really long labels or big fonts
upper Should the upper off diagonal matrix be drawn, or left blank?
diag Should we show the diagonal?
... Other parameters for axis (e.g., cex.axis to change the font size, srt to rotate the
numbers in the plot)
Details
When summarizing the correlations of large data bases or when teaching about factor analysis or
cluster analysis, it is useful to graphically display the structure of correlation matrices. This is a
simple graphical display using the image function.
The difference between mat.plot with a regular image plot is that the primary diagonal goes from
the top left to the lower right. zlim defines how to treat the range of possible values. -1 to 1 and
the color choice is more reasonable. Setting it as c(0,1) will lead to negative correlations treated as
zero. This is advantageous when showing general factor structures, because it makes the 0 white.
The default shows a legend for the color coding on the right hand side of the figure.
54 cor.plot
Inspired, in part, by a paper by S. Dray (2008) on the number of components problem.
Modified following suggestions by David Condon and Josh Wilt to use a more meaningful color
choice ranging from dark red (-1) through white (0) to dark blue (1). Further modified to include
the numerical value of the correlation. (Inspired by the corrplot package). These values may be
scaled according the the probability values found in cor.ci or corr.test.
Unless specified, the font size is dynamically scaled to have a cex = 10/max(nrow(r),ncol(r). This
can produce fairly small fonts for large problems. The font size of the labels may be adjusted using
cex.axis which defaults to one.
By default cor.ci calls cor.plot.upperLowerCi and scales the correlations based upon "signifi-
cance" values. The correlations plotted are the upper and lower confidence boundaries. To show the
correlations themselves, call cor.plot directly.
If using the output of corr.test, the upper off diagonal will be scaled by the corrected probability,
the lower off diagonal the scaling is the uncorrected probabilities.
If using the output of corr.test or cor.ci as input to cor.plot.upperLowerCi, the upper off
diagonal will be the upper bounds and the lower off diagonal the lower bounds of the confidence
intervals.
Author(s)
William Revelle
References
Dray, Stephane (2008) On the number of principal components: A test of dimensionality based on
measurements of similarity between matrices. Computational Statistics \& Data Analysis. 52, 4,
2228-2237.
See Also
fa,mat.sort,cor.ci,corr.test.
Examples
cor.plot(Thurstone,main="9 cognitive variables from Thurstone")
#just blue implies positive manifold
#select just some variables to plot
cor.plot(Thurstone, zlim=c(0,1),main="9 cognitive variables from Thurstone",select=1:4)
#now red means less than .5
cor.plot(mat.sort(Thurstone),TRUE,zlim=c(0,1),
main="9 cognitive variables from Thurstone (sorted by factor loading) ")
simp <- sim.circ(24)
cor.plot(cor(simp),main="24 variables in a circumplex")
#scale by raw and adjusted probabilities
rs <- corr.test(sat.act[1:200,] ) #find the probabilities of the correlations
cor.plot(r=rs$r,numbers=TRUE,pval=rs$p,main="Correlations scaled by probability values")
#Show the upper and lower confidence intervals
cor.plot.upperLowerCi(R=rs,numbers=TRUE)
cor.smooth 55
cor.smooth Smooth a non-positive definite correlation matrix to make it positive
definite
Description
Factor analysis requires positive definite correlation matrices. Unfortunately, with pairwise deletion
of missing data or if using tetrachoric or polychoric correlations, not all correlation matrices
are positive definite. cor.smooth does a eigenvector (principal components) smoothing. Negative
eigen values are replaced with 100 * eig.tol, the matrix is reproduced and forced to a correlation
matrix using cov2cor.
Usage
cor.smooth(x,eig.tol=10^-12)
cor.smoother(x,cut=.01)
Arguments
xA correlation matrix or a raw data matrix.
eig.tol the minimum acceptable eigenvalue.
cut Report all abs(residuals) > cut
Details
The smoothing is done by eigen value decomposition. eigen values < eig.tol are changed to 100
* eig.tol. The positive eigen values are rescaled to sum to the number of items. The matrix is re-
computed (eigen.vectors %*% diag(eigen.values) %*% t(eigen.vectors) and forced to a correlation
matrix using cov2cor. (See Bock, Gibbons and Muraki, 1988 and Wothke, 1993).
This does not implement the Knol and ten Berge (1989) solution, nor do nearcor and posdefify in
sfmsmisc, not does nearPD in Matrix. As Martin Maechler puts it in the posdedify function, "there
are more sophisticated algorithms to solve this and related problems."
cor.smoother examines all of nvar minors of rank nvar-1 by systematically dropping one variable at
a time and finding the eigen value decomposition. It reports those variables, which, when dropped,
produce a positive definite matrix. It also reports the number of negative eigenvalues when each
variable is dropped. Finally, it compares the original correlation matrix to the smoothed correlation
matrix and reports those items with absolute deviations great than cut. These are all hints as to what
might be wrong with a correlation matrix.
Value
The smoothed matrix with a warning reporting that smoothing was necessary (if smoothing was in
fact necessary).
Author(s)
William Revelle
56 cor.smooth
References
R. Darrell Bock, Robert Gibbons and Eiji Muraki (1988) Full-Information Item Factor Analysis.
Applied Psychological Measurement, 12 (3), 261-280.
Werner Wothke (1993), Nonpositive definite matrices in structural modeling. In Kenneth A. Bollen
and J. Scott Long (Editors),Testing structural equation models, Sage Publications, Newbury Park.
D.L. Knol and JMF ten Berge (1989) Least squares approximation of an improper correlation matrix
by a proper one. Psychometrika, 54, 53-61.
See Also
tetrachoric,polychoric,fa and irt.fa, and the burt data set.
See also nearcor and posdefify in the sfsmisc package and nearPD in the Matrix package.
Examples
bs <- cor.smooth(burt) #burt data set is not positive definite
plot(burt[lower.tri(burt)],bs[lower.tri(bs)],ylab="smoothed values",xlab="original values")
abline(0,1,lty="dashed")
round(burt - bs,3)
fa(burt,2) #this throws a warning that the matrix yields an improper solution
#Smoothing first throws a warning that the matrix was improper,
#but produces a better solution
fa(cor.smooth(burt),2)
#this next example is a correlation matrix from DeLeuw used as an example
#in Knol and ten Berge.
#the example is also used in the nearcor documentation
cat("pr is the example matrix used in Knol DL, ten Berge (1989)\n")
pr <- matrix(c(1, 0.477, 0.644, 0.478, 0.651, 0.826,
0.477, 1, 0.516, 0.233, 0.682, 0.75,
0.644, 0.516, 1, 0.599, 0.581, 0.742,
0.478, 0.233, 0.599, 1, 0.741, 0.8,
0.651, 0.682, 0.581, 0.741, 1, 0.798,
0.826, 0.75, 0.742, 0.8, 0.798, 1),
nrow = 6, ncol = 6)
sm <- cor.smooth(pr)
resid <- pr - sm
# several goodness of fit tests
# from Knol and ten Berge
tr(resid %*% t(resid)) /2
# from nearPD
sum(resid^2)/2
cor.wt 57
cor.wt The sample size weighted correlation may be used in correlating ag-
gregated data
Description
If using aggregated data, the correlation of the means does not reflect the sample size used for each
mean. cov.wt in RCore does this and returns a covariance matrix or the correlation matrix. The
cor.wt function weights by sample size or by standard errors and by default return correlations.
Usage
cor.wt(data,vars=NULL, w=NULL,sds=NULL, cor=TRUE)
Arguments
data A matrix or data frame
vars Variables to analyze
wA set of weights (e.g., the sample sizes)
sds Standard deviations of the samples (used if weighting by standard errors)
cor Report correlations (the default) or covariances
Details
A weighted correlation is just rij =P(wtk(xik −xj k )
pwtik P(x2
ik )wtjkP(x2
jk )where xik is a deviation from the
weighted mean.
The weighted correlation is appropriate for correlating aggregated data, where individual data points
might reflect the means of a number of observations. In this case, each point is weighted by its
sample size (or alternatively, by the standard error). If the weights are all equal, the correlation is
just a normal Pearson correlation.
Used when finding correlations of group means found using statsBy.
Value
cor The weighted correlation
xwt The data as weighted deviations from the weighted mean
wt The weights used (calculated from the sample sizes).
mean The weighted means
xc Unweighted, centered deviation scores from the weighted mean
xs Deviation scores weighted by the standard error of each sample mean
Note
A generalization of cov.wt in core R
58 cor2dist
Author(s)
William Revelle
See Also
See Also as cov.wt,statsBy
Examples
means.by.age <- statsBy(sat.act,"age")
wt.cors <- cor.wt(means.by.age)
lowerMat(wt.cors$r) #show the weighted correlations
unwt <- lowerCor(means.by.age$mean)
mixed <- lowerUpper(unwt,wt.cors$r) #combine both results
cor.plot(mixed,TRUE,main="weighted versus unweighted correlations")
diff <- lowerUpper(unwt,wt.cors$r,TRUE)
cor.plot(diff,TRUE,main="differences of weighted versus unweighted correlations")
cor2dist Convert correlations to distances (necessary to do multidimensional
scaling of correlation data)
Description
A minor helper function to convert correlations (ranging from -1 to 1) to distances (ranging from 0
to 2). d=p(2(1 −r)).
Usage
cor2dist(x)
Arguments
xIf square, then assumed to be a correlation matrix, otherwise the correlations are
found first.
Value
dist: a square matrix of distances.
Note
For an example of doing multidimensional scaling on data that are normally factored, see Revelle
(in prep)
Author(s)
William Revelle
corFiml 59
References
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
corFiml Find a Full Information Maximum Likelihood (FIML) correlation or
covariance matrix from a data matrix with missing data
Description
Makes use of functions adapted from the lavaan package to find FIML covariance/correlation ma-
trices. FIML can be much slower than the normal pairwise deletion option of cor, but provides
slightly more precise estimates.
Usage
corFiml(x, covar = FALSE,show=FALSE)
Arguments
xA data.frame or data matrix
covar By default, just return the correlation matrix. If covar is TRUE, return a list
containing the covariance matrix and the ML fit function.
show If show=TRUE, then just show the patterns of missingness, but don’t do the
FIML. Useful for understanding the process of fiml.
Details
In the presence of missing data, Full Information Maximum Likelihood (FIML) is an alternative
to simply using the pairwise correlations. The implementation in the lavaan package for structural
equation modeling has been adapted for the simpler case of just finding the correlations or covari-
ances.
The pairwise solution for any pair of variables is insensitive to other variables included in the matrix.
On the other hand, the ML solution depends upon the entire set of items being correlated. This will
lead to slightly different solutions for different subsets of variables.
The basic FIML algorithm is to find the pairwise ML solution for covariances and means for ev-
ery pattern of missingness and then to weight the solution by the size of every unique pattern of
missingness.
Value
cor The correlation matrix found using FIML
cov The covariance matrix found using FIML
fx The ML fit function
60 corr.test
Note
The functions used in lavaan are not exported and so have been copied (and simplified) to the psych
package.
Author(s)
Wiliam Revelle
See Also
To use the resulting correlations, see fa. To see the pairwise pattern of missingness, see count.pairwise.
Examples
rML <- corFiml(bfi[20:27])
rpw <- cor(bfi[20:27],use="pairwise")
round(rML - rpw,3)
mp <- corFiml(bfi[20:27],show=TRUE)
mp
corr.test Find the correlations, sample sizes, and probability values between
elements of a matrix or data.frame.
Description
Although the cor function finds the correlations for a matrix, it does not report probability values.
corr.test uses cor to find the correlations for either complete or pairwise data and reports the sample
sizes and probability values as well. For symmetric matrices, raw probabilites are reported below
the diagonal and correlations adjusted for multiple comparisons above the diagonal. In the case of
different x and ys, the default is to adjust the probabilities for multiple tests.
Usage
corr.test(x, y = NULL, use = "pairwise",method="pearson",adjust="holm", alpha=.05,ci=TRUE)
corr.p(r,n,adjust="holm",alpha=.05)
Arguments
xA matrix or dataframe
yA second matrix or dataframe with the same number of rows as x
use use="pairwise" is the default value and will do pairwise deletion of cases. use="complete"
will select just complete cases.
method method="pearson" is the default value. The alternatives to be passed to cor are
"spearman" and "kendall"
corr.test 61
adjust What adjustment for multiple tests should be used? ("holm", "hochberg", "hom-
mel", "bonferroni", "BH", "BY", "fdr", "none"). See p.adjust for details about
why to use "holm" rather than "bonferroni").
alpha alpha level of confidence intervals
rA correlation matrix
nNumber of observations if using corr.p. May be either a matrix (as returned from
corr.test, or a scaler. Set to n- np if finding the significance of partial correlations.
(See below).
ci By default, confidence intervals are found. However, this leads to a great slow-
down of speed. So, for just the rs, ts and ps, set ci=FALSE
Details
corr.test uses the cor function to find the correlations, and then applies a t-test to the individual
correlations using the formula
t=r∗p(n−2)
p(1 −r2)
se =p(1−r2
n−2)
The t and Standard Errors are returned as objects in the result, but are not normally displayed.
Confidence intervals are found and printed if using the print(short=FALSE) option. These are found
by using the fisher z transform of the correlation, and the standard error of the z transforms is
se =p(1
n−3)
.
The probability values may be adjusted using the Holm (or other) correction. If the matrix is
symmetric (no y data), then the original p values are reported below the diagonal and the adjusted
above the diagonal. Otherwise, all probabilities are adjusted (unless adjust="none"). This is made
explicit in the output.
corr.p may be applied to the results of partial.r if n is set to n - s (where s is the number of
variables partialed out) Fisher, 1924.
Value
rThe matrix of correlations
nNumber of cases per correlation
tvalue of t-test for each correlation
ptwo tailed probability of t for each correlation. For symmetric matrices, p values
adjusted for multiple tests are reported above the diagonal.
se standard error of the correlation
ci the alpha/2 lower and upper values
62 correct.cor
Note
For very large matrices (> 200 x 200), there is a noticeable speed improvement if confidence inter-
vals are not found.
See Also
cor.test for tests of a single correlation, Hmisc::rcorr for an equivalant function, r.test to test the
difference between correlations, and cortest.mat to test for equality of two correlation matrices.
Also see cor.ci for bootstrapped confidence intervals of Pearson, Spearman, Kendall, tetrachoric
or polychoric correlations. In addition cor.ci will find bootstrapped estimates of composite scales
based upon a set of correlations (ala cluster.cor).
In particular, see p.adjust for a discussion of p values associated with multiple tests.
Other useful functions related to finding and displaying correlations include lowerCor for finding
the correlations and then displaying the lower off diagonal using the lowerMat function. lowerUpper
to compare two correlation matrices.
Examples
ct <- corr.test(attitude) #find the correlations and give the probabilities
ct #show the results
corr.test(attitude[1:3],attitude[4:6]) #reports all values corrected for multiple tests
#corr.test(sat.act[1:3],sat.act[4:6],adjust="none") #don't adjust the probabilities
#take correlations and show the probabilities as well as the confidence intervals
print(corr.p(cor(attitude[1:4]),30),short=FALSE)
#don't adjust the probabilities
print(corr.test(sat.act[1:3],sat.act[4:6],adjust="none"),short=FALSE)
correct.cor Find dis-attenuated correlations given correlations and reliabilities
Description
Given a raw correlation matrix and a vector of reliabilities, report the disattenuated correlations
above the diagonal.
Usage
correct.cor(x, y)
Arguments
xA raw correlation matrix
yVector of reliabilities
correct.cor 63
Details
Disattenuated correlations may be thought of as correlations between the latent variables measured
by a set of observed variables. That is, what would the correlation be between two (unreliable)
variables be if both variables were measured perfectly reliably.
This function is mainly used if importing correlations and reliabilities from somewhere else. If the
raw data are available, use score.items, or cluster.loadings or cluster.cor.
Examples of the output of this function are seen in cluster.loadings and cluster.cor
Value
Raw correlations below the diagonal, reliabilities on the diagonal, disattenuated above the diagonal.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
References
http://personality-project.org/revelle/syllabi/405.syllabus.html
See Also
cluster.loadings and cluster.cor
Examples
# attitude from the datasets package
#example 1 is a rather clunky way of doing things
a1 <- attitude[,c(1:3)]
a2 <- attitude[,c(4:7)]
x1 <- rowSums(a1) #find the sum of the first 3 attitudes
x2 <- rowSums(a2) #find the sum of the last 4 attitudes
alpha1 <- alpha(a1)
alpha2 <- alpha(a2)
x <- matrix(c(x1,x2),ncol=2)
x.cor <- cor(x)
alpha <- c(alpha1$total$raw_alpha,alpha2$total$raw_alpha)
round(correct.cor(x.cor,alpha),2)
#
#much better - although uses standardized alpha
clusters <- matrix(c(rep(1,3),rep(0,7),rep(1,4)),ncol=2)
cluster.loadings(clusters,cor(attitude))
# or
clusters <- matrix(c(rep(1,3),rep(0,7),rep(1,4)),ncol=2)
cluster.cor(clusters,cor(attitude))
#
#best
scores <- score.items(matrix(c(rep(1,3),rep(0,7),rep(1,4)),ncol=2),attitude)
64 cortest.bartlett
scores$corrected
cortest.bartlett Bartlett’s test that a correlation matrix is an identity matrix
Description
Bartlett (1951) proposed that -ln(det(R)*(N-1 - (2p+5)/6) was distributed as chi square if R were an
identity matrix. A useful test that residuals correlations are all zero.
Usage
cortest.bartlett(R, n = NULL,diag=TRUE)
Arguments
RA correlation matrix. (If R is not square, correlations are found and a warning
is issued.
nSample size (if not specified, 100 is assumed).
diag Will replace the diagonal of the matrix with 1s to make it a correlation matrix.
Details
More useful for pedagogical purposes than actual applications. The Bartlett test is asymptotically
chi square distributed.
Note that if applied to residuals from factor analysis (fa) or principal components analysis (principal)
that the diagonal must be replaced with 1s. This is done automatically if diag=TRUE. (See exam-
ples.)
Value
chisq Assymptotically chisquare
p.value Of chi square
df The degrees of freedom
Author(s)
William Revelle
References
Bartlett, M. S., (1951), The Effect of Standardization on a chi square Approximation in Factor
Analysis, Biometrika, 38, 337-344.
cortest.mat 65
See Also
cortest.mat,cortest.normal,cortest.jennrich
Examples
set.seed(42)
x <- matrix(rnorm(1000),ncol=10)
r <- cor(x)
cortest.bartlett(r) #random data don't differ from an identity matrix
data(bfi)
cortest.bartlett(bfi[1:200,1:10]) #not an identity matrix
f3 <- fa(Thurstone,3)
f3r <- f3$resid
cortest.bartlett(f3r,n=213,diag=FALSE) #incorrect
cortest.bartlett(f3r,n=213,diag=TRUE) #correct (by default)
cortest.mat Chi square tests of whether a single matrix is an identity matrix, or a
pair of matrices are equal.
Description
Steiger (1980) pointed out that the sum of the squared elements of a correlation matrix, or the Fisher
z score equivalents, is distributed as chi square under the null hypothesis that the values are zero
(i.e., elements of the identity matrix). This is particularly useful for examining whether correlations
in a single matrix differ from zero or for comparing two matrices. Jennrich (1970) also examined
tests of differences between matrices.
Usage
cortest.normal(R1, R2 = NULL, n1 = NULL, n2 = NULL, fisher = TRUE) #the steiger test
cortest(R1,R2=NULL,n1=NULL,n2 = NULL, fisher = TRUE,cor=TRUE) #same as cortest.normal
cortest.jennrich(R1,R2,n1=NULL, n2=NULL) #the Jennrich test
cortest.mat(R1,R2=NULL,n1=NULL,n2 = NULL) #an alternative test
Arguments
R1 A correlation matrix. (If R1 is not rectangular, and cor=TRUE, the correlations
are found).
R2 A correlation matrix. If R2 is not rectangular, and cor=TRUE, the correlations
are found. If R2 is NULL, then the test is just whether R1 is an identity matrix.
n1 Sample size of R1
n2 Sample size of R2
fisher Fisher z transform the correlations?
66 cortest.mat
cor By default, if the input matrices are not symmetric, they are converted to cor-
relation matrices. That is, they are treated as if they were the raw data. If
cor=FALSE, then the input matrices are taken to be correlation matrices.
Details
There are several ways to test if a matrix is the identity matrix. The most well known is the chi
square test of Bartlett (1951) and Box (1949). A very straightforward test, discussed by Steiger
(1980) is to find the sum of the squared correlations or the sum of the squared Fisher transformed
correlations. Under the null hypothesis that all the correlations are equal, this sum is distributed as
chi square. This is implemented in cortest and cortest.normal
Yet another test, is the Jennrich(1970) test of the equality of two matrices. This compares the
differences between two matrices to the averages of two matrices using a chi square test. This is
implemented in cortest.jennrich.
Yet another option cortest.mat is to compare the two matrices using an approach analogous to
that used in evaluating the adequacy of a factor model. In factor analysis, the maximum likelihood
fit statistic is
f=log(trace((F F 0+U2)−1R)−log(|(F F 0+U2)−1R|)−n.items.
This in turn is converted to a chi square
χ2= (n.obs −1−(2 ∗p+ 5)/6−(2 ∗factors)/3)) ∗f(see fa.)
That is, the model (M = FF’ + U2) is compared to the original correlation matrix (R) by a function
of M−1R. By analogy, in the case of two matrices, A and B, cortest.mat finds the chi squares
associated with A−1Band AB−1. The sum of these two χ2will also be a χ2but with twice the
degrees of freedom.
Value
chi2 The chi square statistic
df Degrees of freedom for the Chi Square
prob The probability of observing the Chi Square under the null hypothesis.
Note
Both the cortest.jennrich and cortest.normal are probably overly stringent. The ChiSquare values
for pairs of random samples from the same population are larger than would be expected. This is a
good test for rejecting the null of no differences.
Author(s)
William Revelle
References
Steiger, James H. (1980) Testing pattern hypotheses on correlation matrices: alternative statistics
and some empirical results. Multivariate Behavioral Research, 15, 335-352.
Jennrich, Robert I. (1970) An Asymptotic χ2Test for the Equality of Two Correlation Matrices.
Journal of the American Statistical Association, 65, 904-912.
cosinor 67
See Also
cortest.bartlett
Examples
x <- matrix(rnorm(1000),ncol=10)
cortest.normal(x) #just test if this matrix is an identity
x <- sim.congeneric(loads =c(.9,.8,.7,.6,.5),N=1000,short=FALSE)
y <- sim.congeneric(loads =c(.9,.8,.7,.6,.5),N=1000,short=FALSE)
cortest.normal(x$r,y$r,n1=1000,n2=1000) #The Steiger test
cortest.jennrich(x$r,y$r,n1=100,n2=1000) # The Jennrich test
cortest.mat(x$r,y$r,n1=1000,n2=1000) #twice the degrees of freedom as the Jennrich
cosinor Functions for analysis of circadian or diurnal data
Description
Circadian data are periodic with a phase of 24 hours. These functions find the best fitting phase
angle (cosinor), the circular mean, circular correlation with circadian data, and the linear by circular
correlation
Usage
cosinor(angle,x=NULL,code=NULL,data=NULL,hours=TRUE,period=24,
plot=FALSE,opti=FALSE,na.rm=TRUE)
cosinor.plot(angle,x=NULL,data = NULL, IDloc=NULL, ID=NULL,hours=TRUE, period=24,
na.rm=TRUE,ylim=NULL,ylab="observed",xlab="Time (double plotted)",
main="Cosine fit",add=FALSE,multi=FALSE,typ="l",...)
cosinor.period(angle,x=NULL,code=NULL,data=NULL,hours=TRUE,period=seq(23,26,1),
plot=FALSE,opti=FALSE,na.rm=TRUE)
circadian.phase(angle,x=NULL,code=NULL,data=NULL,hours=TRUE,period=24,
plot=FALSE,opti=FALSE,na.rm=TRUE)
circadian.mean(angle,data=NULL, hours=TRUE,na.rm=TRUE)
circadian.sd(angle,data=NULL,hours=TRUE,na.rm=TRUE)
circadian.stats(angle,data=NULL,hours=TRUE,na.rm=TRUE)
circadian.F(angle,group,data=NULL,hours=TRUE,na.rm=TRUE)
circadian.reliability(angle,x=NULL,code=NULL,data = NULL,min=16,
oddeven=FALSE, hours=TRUE,period=24,plot=FALSE,opti=FALSE,na.rm=TRUE)
circular.mean(angle,na.rm=TRUE) #angles in radians
circadian.cor(angle,data=NULL,hours=TRUE,na.rm=TRUE) #angles in radians
circular.cor(angle,na.rm=TRUE) #angles in radians
circadian.linear.cor(angle,x=NULL,data=NULL,hours=TRUE)
68 cosinor
Arguments
angle A data frame or matrix of observed values with the time of day as the first value
(unless specified in code) angle can be specified either as hours or as radians)
code A subject identification variable
data A matrix or data frame of data. If specified, then angle and code are variable
names (or locations). See examples.
group If doing comparisons by groups, specify the group code.
min The minimum number of observations per subject to use when finding split half
reliabilities.
oddeven Reliabilities are based upon odd and even items (TRUE) or first vs. last half
(FALSE). Default is first and last half.
period Although time of day is assumed to have a 24 hour rhythm, other rhythms may
be fit. If calling cosinor.period, a range may be specified.
IDloc Which column number is the ID field
ID What specific subject number should be plotted for one variable
plot if TRUE, then plot the first variable (angle)
opti opti=TRUE: iterative optimization (slow) or opti=FALSE: linear fitting (fast)
hours If TRUE, measures are in 24 hours to the day, otherwise, radians
xA set of external variables to correlate with the phase angles
na.rm Should missing data be removed?
ylim Specify the range of the y axis if the defaults don’t work
ylab The label of the yaxis
xlab Labels for the x axis
main the title of the graphic
add If doing multiple (spagetti) plots, set add = TRUE for the second and beyond
plots
multi If doing multiple (spagetti) plots, set multi=TRUE for the first and subsequent
plots
typ Pass the line type to graphics
... any other graphic parameters to pass
Details
When data represent angles (such as the hours of peak alertness or peak tension during the day), we
need to apply circular statistics rather than the more normal linear statistics (see Jammalamadaka
(2006) for a very clear set of examples of circular statistics). The generalization of the mean to
circular data is to convert each angle into a vector, average the x and y coordinates, and convert the
result back to an angle. A statistic that represents the compactness of the observations is R which is
the (normalized) vector length found by adding all of the observations together. This will achieve a
maximum value (1) when all the phase angles are the same and a minimum (0) if the phase angles
are distributed uniformly around the clock.
cosinor 69
The generalization of Pearson correlation to circular statistics is straight forward and is implemented
in cor.circular in the circular package and in circadian.cor here. Just as the Pearson r is a ratio
of covariance to the square root of the product of two variances, so is the circular correlation. The
circular covariance of two circular vectors is defined as the average product of the sines of the
deviations from the circular mean. The variance is thus the average squared sine of the angular
deviations from the circular mean. Circular statistics are used for data that vary over a period (e.g.,
one day) or over directions (e.g., wind direction or bird flight). Jammalamadaka and Lund (2006)
give a very good example of the use of circular statistics in calculating wind speed and direction.
The code from CircStats and circular was adapted to allow for analysis of data from various studies
of mood over the day. Those two packages do not seem to handle missing data, nor do they take
matrix input, but rather emphasize single vectors.
The cosinor function will either iteratively fit cosines of the angle to the observed data (opti=TRUE)
or use the circular by linear regression to estimate the best fitting phase angle. If cos.t <- cos(time)
and sin.t = sin(time) (expressed in hours), then beta.c and beta.s may be found by regression and
the phase is sign(beta.c)∗acos(beta.c/p(beta.c2+beta.s2)) ∗12/pi
Simulations (see examples) suggest that with incomplete times, perhaps the optimization procedure
yields slightly better fits with the correct phase than does the linear model, but the differences are
very small. In the presence of noisey data, these advantages seem to reverse. The recommendation
thus seems to be to use the linear model approach (the default). The fit statistic reported for cosinor
is the correlation of the data with the model [ cos(time - acrophase) ].
The circadian.reliability function splits the data for each subject into a first and second half
(by default, or into odd and even items) and then finds the best fitting phase for each half. These are
then correlated (using circadian.cor) and this correlation is then adjusted for test length using the
conventional Spearman-Brown formula. Returned as object in the output are the statistics for the
first and second part, as well as an ANOVA to compare the two halves.
circular.mean and circular.cor are just circadian.mean and circadian.cor but with input
given in radians rather than hours.
The circadian.linear.cor function will correlate a set of circular variables with a set of linear
variables. The first (angle) variables are circular, the second (x) set of variables are linear.
The circadian.F will compare 2 or more groups in terms of their mean position. This is adapted
from the equivalent function in the circular pacakge. This is clearly a more powerful test the more
each group is compact around its mean (large values of R).
Value
phase The phase angle that best fits the data (expressed in hours if hours=TRUE).
fit Value of the correlation of the fit. This is just the correlation of the data with the
phase adjusted cosine.
mean.angle A vector of mean angles
n,mean,sd The appropriate circular statistic.
correl A matrix of circular correlations or linear by circular correlations
RR is the vector length (0-1) of the mean vector when finding circadian statistics
using circadian.stats
z,p z is the number of observations x R^2. p is the probability of a z.
70 cosinor
phase.rel The reliability of the phase measures. This is the circular correlation between
the two halves adjusted using the Spearman-Brown correction.
fit.rel The split half reliability of the fit statistic.
split.F Do the two halves differ from each other? One would hope not.
group1,group2 The statistics from each half
splits The individual data from each half.
Note
These functions have been adapted from the circular package to allow for ease of use with circadian
data, particularly for data sets with missing data and multiple variables of interest.
Author(s)
William Revelle
References
See circular statistics Jammalamadaka, Sreenivasa and Lund, Ulric (2006),The effect of wind di-
rection on ozone levels: a case study, Environmental and Ecological Statistics, 13, 287-298.
See Also
See the circular and CircStats packages.
Examples
time <- seq(1:24) #create a 24 hour time
pure <- matrix(time,24,18)
colnames(pure) <- paste0("H",1:18)
pure <- data.frame(time,cos((pure - col(pure))*pi/12)*3 + 3)
#18 different phases but scaled to 0-6 match mood data
matplot(pure[-1],type="l",main="Pure circadian arousal rhythms",
xlab="time of day",ylab="Arousal")
op <- par(mfrow=c(2,2))
cosinor.plot(1,3,pure)
cosinor.plot(1,5,pure)
cosinor.plot(1,8,pure)
cosinor.plot(1,12,pure)
p <- cosinor(pure) #find the acrophases (should match the input)
#now, test finding the acrophases for different subjects on 3 variables
#They should be the first 3, second 3, etc. acrophases of pure
pp <- matrix(NA,nrow=6*24,ncol=4)
pure <- as.matrix(pure)
pp[,1] <- rep(pure[,1],6)
pp[1:24,2:4] <- pure[1:24,2:4]
pp[25:48,2:4] <- pure[1:24,5:7] *2 #to test different variances
pp[49:72,2:4] <- pure[1:24,8:10] *3
cosinor 71
pp[73:96,2:4] <- pure[1:24,11:13]
pp[97:120,2:4] <- pure[1:24,14:16]
pp[121:144,2:4] <- pure[1:24,17:19]
pure.df <- data.frame(ID = rep(1:6,each=24),pp)
colnames(pure.df) <- c("ID","Time",paste0("V",1:3))
cosinor("Time",3:5,"ID",pure.df)
op <- par(mfrow=c(2,2))
cosinor.plot(2,3,pure.df,IDloc=1,ID="1")
cosinor.plot(2,3,pure.df,IDloc=1,ID="2")
cosinor.plot(2,3,pure.df,IDloc=1,ID="3")
cosinor.plot(2,3,pure.df,IDloc=1,ID="4")
#now, show those in one panel as spagetti plots
op <- par(mfrow=c(1,1))
cosinor.plot(2,3,pure.df,IDloc=1,ID="1",multi=TRUE,ylim=c(0,20),ylab="Modeled")
cosinor.plot(2,3,pure.df,IDloc=1,ID="2",multi=TRUE,add=TRUE,lty="dotdash")
cosinor.plot(2,3,pure.df,IDloc=1,ID="3",multi=TRUE,add=TRUE,lty="dashed")
cosinor.plot(2,3,pure.df,IDloc=1,ID="4",multi=TRUE,add=TRUE,lty="dotted")
set.seed(42) #what else?
noisy <- pure
noisy[,2:19]<- noisy[,2:19] + rnorm(24*18,0,.2)
n <- cosinor(time,noisy) #add a bit of noise
small.pure <- pure[c(8,11,14,17,20,23),]
small.noisy <- noisy[c(8,11,14,17,20,23),]
small.time <- c(8,11,14,17,20,23)
cosinor.plot(1,3,small.pure,multi=TRUE)
cosinor.plot(1,3,small.noisy,multi=TRUE,add=TRUE,lty="dashed")
# sp <- cosinor(small.pure)
# spo <- cosinor(small.pure,opti=TRUE) #iterative fit
# sn <- cosinor(small.noisy) #linear
# sno <- cosinor(small.noisy,opti=TRUE) #iterative
# sum.df <- data.frame(pure=p,noisy = n, small=sp,small.noise = sn,
# small.opt=spo,small.noise.opt=sno)
# round(sum.df,2)
# round(circadian.cor(sum.df[,c(1,3,5,7,9,11)]),2) #compare alternatives
#
# #now, lets form three "subjects" and show how the grouping variable works
# mixed.df <- rbind(small.pure,small.noisy,noisy)
# mixed.df <- data.frame(ID=c(rep(1,6),rep(2,6),rep(3,24)),
# time=c(rep(c(8,11,14,17,20,23),2),1:24),mixed.df)
# group.df <- cosinor(angle="time",x=2:20,code="ID",data=mixed.df)
# round(group.df,2) #compare these values to the sp,sn,and n values done separately
72 count.pairwise
count.pairwise Count number of pairwise cases for a data set with missing (NA) data.
Description
When doing cor(x, use= "pairwise"), it is nice to know the number of cases for each pairwise
correlation. This is particularly useful when doing SAPA type analyses.
Usage
count.pairwise(x, y = NULL,diagonal=TRUE)
pairwiseDescribe(x,diagonal=FALSE)
Arguments
xAn input matrix, typically a data matrix ready to be correlated.
yAn optional second input matrix
diagonal if TRUE, then report the diagonal, else fill the diagonals with NA
Value
result = matrix of counts of pairwise observations
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
Examples
## Not run:
x <- matrix(rnorm(1000),ncol=6)
y <- matrix(rnorm(500),ncol=3)
x[x < 0] <- NA
y[y > 1] <- NA
count.pairwise(x)
count.pairwise(y)
count.pairwise(x,y)
count.pairwise(x,diagonal=FALSE)
pairwiseDescribe(x)
## End(Not run)
cta 73
cta Simulate the C(ues) T(endency) A(ction) model of motivation
Description
Dynamic motivational models such as the Dynamics of Action (Atkinson and Birch, 1970, Revelle,
1986) may be reparameterized as a simple pair of differential (matrix) equations (Revelle, 1986,
2008). This function simulates the dynamic aspects of the CTA. The CTA model is discussed in
detail in Revelle and Condon (2015).
Usage
cta (n=3,t=5000, cues = NULL, act=NULL, inhibit=NULL,expect = NULL, consume = NULL,
tendency = NULL,tstrength=NULL, type="both", fast=2,compare=FALSE,learn=TRUE,reward=NULL)
cta.15(n = 3, t = 5000, cues = NULL, act = NULL, inhibit = NULL, consume = NULL,
ten = NULL, type = "both", fast = 2)
Arguments
nnumber of actions to simuate
tlength of time to simulate
cues a vector of cue strengths
act matrix of associations between cues and action tendencies
inhibit inhibition matrix
consume Consummation matrix
ten Initial values of action tendencies
type show actions, tendencies, both, or state diagrams
fast display every fast time (skips
expect A matrix of expectations
tendency starting values of tendencies
tstrength a vector of starting value of tendencies
compare Allows a two x two graph to compare two plots
learn Allow the system to learn (self reinforce) over time
reward The strength of the reward for doing an action
74 cta
Details
A very thorough discussion of the CTA model is available from Revelle (2008). An application of
the model is discussed in Revelle and Condon (2015).
cta.15 is the version used to produce the figures and analysis in Revelle and Condon (2015). cta
is the most recent version and includes a learning function developed in collaboration with Luke
Smillie at the University of Melbourne.
The dynamics of action (Atkinson and Birch, 1970) was a model of how instigating forces elicited
action tendencies which in turn elicited actions. The basic concept was that action tendencies had
inertia. That is, a wish (action tendency) would persist until satisfied and would not change without
an instigating force. The consummatory strength of doing an action was thought in turn to reduce
the action tendency. Forces could either be instigating or inhibitory (leading to "negaction").
Perhaps the simplest example is the action tendency (T) to eat a pizza. The instigating forces (F)
to eat the pizza include the smell and look of the pizza, and once eating it, the flavor and texture.
However, if eating the pizza, there is also a consummatory force (C) which was thought to reflect
both the strength (gusto) of eating the pizza as well as some constant consummatory value of the
activity (c). If not eating the pizza, but in a pizza parlor, the smells and visual cues combine to
increase the tendency to eat the pizza. Once eating it, however, the consummatory effect is no
longer zero, and the change in action tendency will be a function of both the instigating forces and
the consummatory forces. These will achieve a balance when instigating forces are equal to the
consummatory forces. The asymptotic strength of eating the pizza reflects this balance and does not
require a “set point" or “comparator".
To avoid the problems of instigating and consummatory lags and the need for a decision mechanism,
it is possible to reparameterize the original DOA model in terms of action tendencies and actions
(Revelle, 1986). Rather than specifying inertia for action tendencies and a choice rule of always
expressing the dominant action tendency, it is useful to distinguish between action tendencies (t) and
the actions (a) themselves and to have actions as well as tendencies having inertial properties. By
separating tendencies from actions, and giving them both inertial properties, we avoid the necessity
of a lag parameter, and by making the decision rule one of mutual inhibition, the process is perhaps
easier to understand. In an environment which affords cues for action (c), cues enhance action
tendencies (t) which in turn strengthen actions (a). This leads to two differential equations, one
describing the growth and decay of action tendencies (t), the other of the actions themselves (a).
dt =Sc −Ca
and
da =Et −Ia
. (See Revelle and Condon (2015) for an extensive discussion of this model.)
cta simulates this model, with the addition of a learning parameter such that activities strengthen
the connection between cues and tendencies. The learning part of the cta model is still under
development. cta.15 represents the state of the cta model as described in the Revelle and Condon
(2015) article.
Value
graphical output unless type="none"
cues echo back the cue input
cta 75
inhibition echo back the inhibitory matrix
time time spent in each activity
frequency Frequency of each activity
tendencies average tendency strengths
actions average action strength
Author(s)
William Revelle
References
Atkinson, John W. and Birch, David (1970) The dynamics of action. John Wiley, New York, N.Y.
Revelle, William (1986) Motivation and efficiency of cognitive performance in Brown, Donald R.
and Veroff, Joe (ed). Frontiers of Motivational Psychology: Essays in honor of J. W. Atkinson.
Springer. (Available as a pdf at http://personality-project.org/revelle/publications/
dynamicsofmotivation.pdf.)
Revelle, W. (2008) Cues, Tendencies and Actions. The Dynamics of Action revisted. http://
personality-project.org/revelle/publications/cta.pdf
Revelle, W. and Condon, D. (2015) A model for personality at three levels. Journal of Research in
Personality http://www.sciencedirect.com/science/article/pii/S0092656615000318
Examples
#not run
#cta() #default values, running over time
#cta(type="state") #default values, in a state space of tendency 1 versus tendency 2
#these next are examples without graphic output
#not run
#two introverts
#c2i <- c(.95,1.05)
#cta(n=2,t=10000,cues=c2i,type="none")
#two extraverts
#c2e <- c(3.95,4.05)
#cta(n=2,t=10000,cues=c2e,type="none")
#three introverts
#c3i <- c(.95,1,1.05)
#cta(3,t=10000,cues=c3i,type="none")
#three extraverts
#c3i <- c(3.95,4, 4.05)
#cta(3,10000,c3i,type="none")
#mixed
#c3 <- c(1,2.5,4)
#cta(3,10000,c3,type="none")
76 cubits
cubits Galton’s example of the relationship between height and ’cubit’ or
forearm length
Description
Francis Galton introduced the ’co-relation’ in 1888 with a paper discussing how to measure the
relationship between two variables. His primary example was the relationship between height and
forearm length. The data table (cubits) is taken from Galton (1888). Unfortunately, there seem to
be some errors in the original data table in that the marginal totals do not match the table.
The data frame, heights, is converted from this table.
Usage
data(cubits)
Format
A data frame with 9 observations on the following 8 variables.
16.5 Cubit length < 16.5
16.75 16.5 <= Cubit length < 17.0
17.25 17.0 <= Cubit length < 17.5
17.75 17.5 <= Cubit length < 18.0
18.25 18.0 <= Cubit length < 18.5
18.75 18.5 <= Cubit length < 19.0
19.25 19.0 <= Cubit length < 19.5
19.75 19.5 <= Cubit length
Details
Sir Francis Galton (1888) published the first demonstration of the correlation coefficient. The re-
gression (or reversion to mediocrity) of the height to the length of the left forearm (a cubit) was
found to .8. There seem to be some errors in the table as published in that the row sums do not
agree with the actual row sums. These data are used to create a matrix using table2matrix for
demonstrations of analysis and displays of the data.
Source
Galton (1888)
References
Galton, Francis (1888) Co-relations and their measurement. Proceedings of the Royal Society.
London Series,45,135-145,
cushny 77
See Also
table2matrix,table2df,ellipses,heights,peas,galton
Examples
data(cubits)
cubits
heights <- table2df(cubits,labs = c("height","cubit"))
ellipses(heights,n=1,main="Galton's co-relation data set")
ellipses(jitter(heights$height,3),jitter(heights$cubit,3),pch=".",
main="Galton's co-relation data set",xlab="height",
ylab="Forearm (cubit)") #add in some noise to see the points
pairs.panels(heights,jiggle=TRUE,main="Galton's cubits data set")
cushny A data set from Cushny and Peebles (1905) on the effect of three drugs
on hours of sleep, used by Student (1908)
Description
The classic data set used by Gossett (publishing as Student) for the introduction of the t-test. The
design was a within subjects study with hours of sleep in a control condition compared to those in 3
drug conditions. Drug1 was 06mg of L Hscyamine, Drug 2L and Drug2R were said to be .6 mg of
Left and Right isomers of Hyoscine. As discussed by Zabell (2008) these were not optical isomers.
The detal1, delta2L and delta2R are changes from the baseline control.
Usage
data(cushny)
Format
A data frame with 10 observations on the following 7 variables.
Control Hours of sleep in a control condition
drug1 Hours of sleep in Drug condition 1
drug2L Hours of sleep in Drug condition 2
drug2R Hours of sleep in Drug condition 3 (an isomer of the drug in condition 2
delta1 Change from control, drug 1
delta2L Change from control, drug 2L
delta2R Change from control, drug 2R
Details
The original analysis by Student is used as an example for the t-test function, both as a paired t-test
and a two group t-test. The data are also useful for a repeated measures analysis of variance.
78 densityBy
Source
Cushny, A.R. and Peebles, A.R. (1905) The action of optical isomers: II hyoscines. The Journal of
Physiology 32, 501-510.
Student (1908) The probable error of the mean. Biometrika, 6 (1) , 1-25.
References
See also the data set sleep and the examples for the t.test
S. L. Zabell. On Student’s 1908 Article "The Probable Error of a Mean" Journal of the American
Statistical Association, Vol. 103, No. 481 (Mar., 2008), pp. 1- 20
Examples
data(cushny)
with(cushny, t.test(drug1,drug2L,paired=TRUE)) #within subjects
error.bars(cushny[1:4],within=TRUE,ylab="Hours of sleep",xlab="Drug condition",
main="95% confidence of within subject effects")
densityBy Create a ’violin plot’ or density plot of the distribution of a set of
variables
Description
Among the many ways to describe a data set, one is density plot or violin plot of the data. This is
similar to a box plot but shows the actual distribution. Median and 25th and 75th percentile lines
are added to the display. If a grouping variable is specified, densityBy will draw violin plots for
each variable and for each group.
Usage
densityBy(x,grp=NULL,grp.name=NULL,ylab="Observed",xlab="",main="Density plot",density=20,
restrict=TRUE,xlim=NULL,add=FALSE,col=NULL,pch=20, ...)
violinBy(x,grp=NULL,grp.name=NULL,ylab="Observed",xlab="",main="Density plot",density=20,
restrict=TRUE,xlim=NULL,add=FALSE,col=NULL,pch=20, ...)
Arguments
xA matrix or data.frame
grp A grouping variable
grp.name If the grouping variable is specified, the what names should be give to the group?
Defaults to 1:ngrp
ylab The y label
xlab The x label
densityBy 79
main Figure title
density How many lines per inch to draw
restrict Restrict the density to the observed max and min of the data
xlim if not specified, will be .5 beyond the number of variables
add Allows overplotting
col Allows for specification of colours. The default for 2 groups is blue and red, for
more group levels, rainbows.
pch The plot character for the mean is by default a small filled circle. To not show
the mean, use pch=NA
... Other graphic parameters
Details
Describe the data using a violin plot. Change density to modify the shading. density=NULL will
fill with col. The grp variable may be used to draw separate violin plots for each of multiple groups.
Value
The density plot of the data.
Note
Nothing yet
Author(s)
William Revelle
See Also
describe,describeBy and statsBy for descriptive statistics and error.bars and error.bars.by
for graphic displays
Examples
densityBy(bfi[1:5])
#not run
#violinBy(bfi[1:5],grp=bfi$gender,grp.name=c("M","F"))
#densityBy(sat.act[5:6],sat.act$education,col=rainbow(6))
80 describe
describe Basic descriptive statistics useful for psychometrics
Description
There are many summary statistics available in R; this function provides the ones most useful for
scale construction and item analysis in classic psychometrics. Range is most useful for the first pass
in a data set, to check for coding errors.
Usage
describe(x, na.rm = TRUE, interp=FALSE,skew = TRUE, ranges = TRUE,trim=.1,
type=3,check=TRUE,fast=NULL,quant=NULL,IQR=FALSE)
describeData(x,head=4,tail=4)
Arguments
xA data frame or matrix
na.rm The default is to delete missing data. na.rm=FALSE will delete the case.
interp Should the median be standard or interpolated
skew Should the skew and kurtosis be calculated?
ranges Should the range be calculated?
trim trim=.1 – trim means by dropping the top and bottom trim fraction
type Which estimate of skew and kurtosis should be used? (See details.)
check Should we check for non-numeric variables? Slower but helpful.
fast if TRUE, will do n, means, sds, ranges for an improvement in speed. If NULL,
will switch to fast mode for large (ncol * nrow > 10^7) problems, otherwise
defaults to fast = FALSE
quant if not NULL, will find the specified quantiles (e.g. quant=c(.25,.75) will find the
25th and 75th percentiles)
IQR If TRUE, show the interquartile range
head show the first 1:head cases for each variable in describeData
tail Show the last nobs-tail cases for each variable in describeData
Details
In basic data analysis it is vital to get basic descriptive statistics. Procedures such as summary
and hmisc::describe do so. The describe function in the psych package is meant to produce the
most frequently requested stats in psychometric and psychology studies, and to produce them in
an easy to read data.frame. The results from describe can be used in graphics functions (e.g.,
error.crosses).
The range statistics (min, max, range) are most useful for data checking to detect coding errors, and
should be found in early analyses of the data.
describe 81
Although describe will work on data frames as well as matrices, it is important to realize that for
data frames, descriptive statistics will be reported only for those variables where this makes sense
(i.e., not for alphanumeric data).
If the check option is TRUE, variables that are categorical or logical are converted to numeric and
then described. These variables are marked with an * in the row name. This is somewhat slower.
Note that in the case of categories or factors, the numerical ordering is not necessarily the one
expected. For instance, if education is coded "high school", "some college" , "finished college",
then the default coding will lead to these as values of 2, 3, 1. Thus, statistics for those variables
marked with * should be interpreted cautiously (if at all).
In a typical study, one might read the data in from the clipboard (read.clipboard), show the splom
plot of the correlations (pairs.panels), and then describe the data.
na.rm=FALSE is equivalent to describe(na.omit(x))
When finding the skew and the kurtosis, there are three different options available. These match the
choices available in skewness and kurtosis found in the e1071 package (see Joanes and Gill (1998)
for the advantages of each one).
If we define mr= [P(X−mx)r]/n then
Type 1 finds skewness and kurtosis by g1=m3/(m2)3/2and g2=m4/(m2)2−3.
Type 2 is G1 = g1∗pn∗(n−1)/(n−2) and G2=(n−1) ∗[(n+ 1)g2 + 6]/((n−2)(n−3)).
Type 3 is b1 = [(n−1)/n]3/2m3/m3/2
2and b2 = [(n−1)/n]3/2m4/m2
2).
The additional helper function describeData just scans the data array and reports on whether the
data are all numerical, logical/factorial, or categorical. This is a useful check to run if trying to get
descriptive statistics on very large data sets where to improve the speed, the check option is FALSE.
The fast=TRUE option will lead to a speed up of about 50% for larger problems by not finding all
of the statistics (see NOTE)
Value
A data.frame of the relevant statistics:
item name
item number
number of valid cases
mean
standard deviation
trimmed mean (with trim defaulting to .1)
median (standard or interpolated
mad: median absolute deviation (from the median)
minimum
maximum
skew
kurtosis
standard error
82 describe
Note
For very large data sets that are data.frames, describe can be rather slow. Converting the data to a
matrix first is recommended. However, if the data are of different types, (factors or logical), this is
not possible. If the data set includes columns of character data, it is also not possible. Thus, a quick
pass with describeData is recommended.
For the greatest speed, at the cost of losing information, do not ask for ranges or for skew and turn
off check. This is done automatically if the fast option is TRUE or for large data sets.
Note that by default, fast=NULL. But if the number of cases x number of variables exceeds (ncol
* nrow > 10^7), fast will be set to TRUE. This will provide just n, mean, sd, min, max, range, and
standard errors. To get all of the statistics (but at a cost of greater time) set fast=FALSE.
The problem seems to be a memory limitation in that the time taken is an accelerating function of
nvars * nobs. Thus, for a largish problem (72,000 cases with 1680 variables) which might take 330
seconds, doing it as two sets of 840 variable cuts the time down to 80 seconds.
Author(s)
http://personality-project.org/revelle.html
Maintainer: William Revelle <revelle@northwestern.edu>
References
Joanes, D.N. and Gill, C.A (1998). Comparing measures of sample skewness and kurtosis. The
Statistician, 47, 183-189.
See Also
describe.by,skew,kurtosi interp.median,read.clipboard. Then, for graphic output, see
error.crosses,pairs.panels,error.bars,error.bars.by and densityBy, or violinBy
Examples
data(sat.act)
describe(sat.act)
describe(sat.act,skew=FALSE)
describe(sat.act,IQR=TRUE) #show the interquartile Range
describe(sat.act,quant=c(.1,.25,.5,.75,.90) ) #find the 10th, 25th, 50th,
#75th and 90th percentiles
describeData(sat.act) #the fast version
describeBy 83
describeBy Basic summary statistics by group
Description
Report basic summary statistics by a grouping variable. Useful if the grouping variable is some ex-
perimental variable and data are to be aggregated for plotting. Partly a wrapper for by and describe
Usage
describeBy(x, group=NULL,mat=FALSE,type=3,digits=15,...)
describe.by(x, group=NULL,mat=FALSE,type=3,...) # deprecated
Arguments
xa data.frame or matrix. See note for statsBy.
group a grouping variable or a list of grouping variables
mat provide a matrix output rather than a list
type Which type of skew and kurtosis should be found
digits When giving matrix output, how many digits should be reported?
... parameters to be passed to describe
Details
To get descriptive statistics for several different grouping variables, make sure that group is a list. In
the case of matrix output with multiple grouping variables, the grouping variable values are added
to the output.
The type parameter specifies which version of skew and kurtosis should be found. See describe
for more details.
An alternative function (statsBy) returns a list of means, n, and standard deviations for each group.
This is particularly useful if finding weighted correlations of group means using cor.wt. More
importantly, it does a proper within and between group decomposition of the correlation.
Value
A data.frame of the relevant statistics broken down by group:
item name
item number
number of valid cases
mean
standard deviation
median
mad: median absolute deviation (from the median)
minimum
maximum
84 df2latex
skew
standard error
Author(s)
William Revelle
See Also
describe,statsBy,densityBy and violinBy as well as error.bars and error.bars.by for
other graphical displays.
Examples
data(sat.act)
describeBy(sat.act,sat.act$gender) #just one grouping variable
#describeBy(sat.act,list(sat.act$gender,sat.act$education)) #two grouping variables
des.mat <- describeBy(sat.act$age,sat.act$education,mat=TRUE) #matrix (data.frame) output
des.mat <- describeBy(sat.act$age,list(sat.act$education,sat.act$gender),
mat=TRUE,digits=2) #matrix output
df2latex Convert a data frame, correlation matrix, or factor analysis output to
a LaTeX table
Description
A set of handy helper functions to convert data frames or matrices to LaTeX tables. Although
Sweave is the preferred means of converting R output to LaTeX, it is sometimes useful to go directly
from a data.frame or matrix to a LaTeX table. cor2latex will find the correlations and then create
a lower (or upper) triangular matrix for latex output. fa2latex will create the latex commands for
showing the loadings and factor intercorrelations. As the default option, tables are prepared in an
approximation of APA format.
Usage
df2latex(x,digits=2,rowlabels=TRUE,apa=TRUE,short.names=TRUE,font.size ="scriptsize",
big.mark=NULL,drop.na=TRUE, heading="A table from the psych package in R",
caption="df2latex",label="default", char=FALSE,
stars=FALSE,silent=FALSE,file=NULL,append=FALSE,cut=0,big=0)
cor2latex(x,use = "pairwise", method="pearson", adjust="holm",stars=FALSE,
digits=2,rowlabels=TRUE,lower=TRUE,apa=TRUE,short.names=TRUE,
font.size ="scriptsize",
heading="A correlation table from the psych package in R.",
df2latex 85
caption="cor2latex",label="default",silent=FALSE,file=NULL,append=FALSE)
fa2latex(f,digits=2,rowlabels=TRUE,apa=TRUE,short.names=FALSE,cumvar=FALSE,
cut=0,big=.3,alpha=.05,font.size ="scriptsize",
heading="A factor analysis table from the psych package in R",
caption="fa2latex",label="default",silent=FALSE,file=NULL,append=FALSE)
omega2latex(f,digits=2,rowlabels=TRUE,apa=TRUE,short.names=FALSE,cumvar=FALSE,cut=.2,
font.size ="scriptsize",
heading="An omega analysis table from the psych package in R",
caption="omega2latex",label="default",silent=FALSE,file=NULL,append=FALSE)
irt2latex(f,digits=2,rowlabels=TRUE,apa=TRUE,short.names=FALSE,
font.size ="scriptsize", heading="An IRT factor analysis table from R",
caption="fa2latex",label="default",silent=FALSE,file=NULL,append=FALSE)
ICC2latex(icc,digits=2,rowlabels=TRUE,apa=TRUE,ci=TRUE,
font.size ="scriptsize",big.mark=NULL, drop.na=TRUE,
heading="A table from the psych package in R",
caption="ICC2latex",label="default",char=FALSE,silent=FALSE,file=NULL,append=FALSE)
Arguments
xA data frame or matrix to convert to LaTeX. If non-square, then correlations will
be found prior to printing in cor2latex
digits Round the output to digits of accuracy. NULL for formatting character data
rowlabels If TRUE, use the row names from the matrix or data.frame
short.names Name the columns with abbreviated rownames to save space
apa If TRUE formats table in APA style
cumvar For factor analyses, should we show the cumulative variance accounted for?
font.size e.g., "scriptsize", "tiny" or anyother acceptable LaTeX font size.
heading The label appearing at the top of the table
caption The table caption
lower in cor2latex, just show the lower triangular matrix
fThe object returned from a factor analysis using fa or irt.fa.
label The label for the table
big.mark Comma separate numbers large numbers (big.mark=",")
drop.na Do not print NA values
method When finding correlations, which method should be used (pearson)
use use="pairwise" is the default when finding correlations in cor2latex
adjust If showing probabilities, which adjustment should be used (holm)
stars Should probability asterixs be displayed in cor2latex (FALSE)
char char=TRUE allows printing tables with character information, but does not al-
low for putting in commas into numbers
cut In omega2latex, df2latex and fa2latex, do not print abs(values) < cut
big In fa2latex and df2latex boldface those abs(values) > big
86 df2latex
alpha If fa has returned confidence intervals, then what values of loadings should be
boldfaced?
icc Either the output of an ICC, or the data to be analyzed.
ci Should confidence intervals of the ICC be displayed
silent If TRUE, do not print any output, just return silently – useful if using Sweave
file If specified, write the output to this file
append If file is specified, then should we append (append=TRUE) or just write to the
file
Value
A LaTeX table. Note that if showing "stars" for correlations, then one needs to use the siunitx
package in LaTex. The entire LaTeX output is also returned invisibly. If using Sweave to create
tables, then the silent option should be set to TRUE and the returned object saved as a file. See the
last example.
Author(s)
William Revelle with suggestions from Jason French and David Condon and Davide Morselli
See Also
The many LaTeX conversion routines in Hmisc.
Examples
df2latex(Thurstone,rowlabels=FALSE,apa=FALSE,short.names=FALSE,
caption="Thurstone Correlation matrix")
df2latex(Thurstone,heading="Thurstone Correlation matrix in APA style")
df2latex(describe(sat.act)[2:10],short.names=FALSE)
cor2latex(Thurstone)
cor2latex(sat.act,short.names=FALSE)
fa2latex(fa(Thurstone,3),heading="Factor analysis from R in quasi APA style")
#If using Sweave to create a LateX table as a separate file then set silent=TRUE
#e.g.,
#LaTex preamble
#....
#<<print=FALSE,echo=FALSE>>=
#f3 <- fa(Thurstone,3)
#fa2latex(f3,silent=TRUE,file='testoutput.tex')
#@
#
#\input{testoutput.tex}
diagram 87
diagram Helper functions for drawing path model diagrams
Description
Path models are used to describe structural equation models or cluster analytic output. These func-
tions provide the primitives for drawing path models. Used as a substitute for some of the function-
ality of Rgraphviz.
Usage
diagram(fit,...)
dia.rect(x, y = NULL, labels = NULL, cex = 1, xlim = c(0, 1), ylim = c(0, 1), ...)
dia.ellipse(x, y = NULL, labels = NULL, cex=1,e.size=.05, xlim=c(0,1), ylim=c(0,1), ...)
dia.triangle(x, y = NULL, labels =NULL, cex = 1, xlim=c(0,1),ylim=c(0,1),...)
dia.ellipse1(x,y,e.size=.05,xlim=c(0,1),ylim=c(0,1),...)
dia.shape(x, y = NULL, labels = NULL, cex = 1,
e.size=.05, xlim=c(0,1), ylim=c(0,1), shape=1, ...)
dia.arrow(from,to,labels=NULL,scale=1,cex=1,adj=2,both=FALSE,pos=NULL,l.cex,gap.size,...)
dia.curve(from,to,labels=NULL,scale=1,...)
dia.curved.arrow(from,to,labels=NULL,scale=1,both=FALSE,...)
dia.self(location,labels=NULL,scale=.8,side=2,...)
dia.cone(x=0, y=-2, theta=45, arrow=TRUE,curves=TRUE,add=FALSE,labels=NULL,
xlim = c(-1, 1), ylim=c(-1,1),... )
Arguments
fit The results from a factor analysis fa, components analysis principal, omega
reliability analysis, omega, cluster analysis iclust or confirmatory factor anal-
ysis, cfa, or structural equation model,sem, using the lavaan package.
xx coordinate of a rectangle or ellipse
yy coordinate of a rectangle or ellipse
e.size The size of the ellipse (scaled by the number of variables
labels Text to insert in rectangle, ellipse, or arrow
cex adjust the text size
l.cex Adjust the text size in arrows, defaults to cex which in turn defaults to 1
gap.size Tweak the gap in an arrow to be allow the label to be in a gap
adj Where to put the label along the arrows (values are then divided by 4)
both Should the arrows have arrow heads on both ends?
scale modifies size of rectangle and ellipse as well as the curvature of curves. (For
curvature, positive numbers are concave down and to the left
from arrows and curves go from
to arrows and curves go to
88 diagram
location where is the rectangle?
shape Which shape to draw
xlim default ranges
ylim default ranges
side Which side of boxes should errors appear
theta Angle in degrees of vectors
arrow draw arrows for edges in dia.cone
add if TRUE, plot on previous plot
curves if TRUE, draw curves between arrows in dia.cone
pos The position of the text in dia.arrow. Follows the text positions of 1, 2, 3, 4 or
NULL
... Most graphic parameters may be passed here
Details
The diagram function calls fa.diagram,omega.diagram,ICLUST.diagram or lavaan.diagram
depending upon the class of the fit input. See those functions for particular parameter values.
The remaining functions are the graphic primitives used by fa.diagram,structure.diagram,
omega.diagram,ICLUST.diagram and het.diagram
They create rectangles, ellipses or triangles surrounding text, connect them to straight or curved
arrows, and can draw an arrow from and to the same rectangle.
Each shape (ellipse, rectangle or triangle) has a left, right, top and bottom and center coordinate that
may be used to connect the arrows.
Curves are double-headed arrows.
The helper functions were developed to get around the infelicities associated with trying to install
Rgraphviz and graphviz.
These functions form the core of fa.diagram,het.diagram.
Better documentation will be added as these functions get improved. Currently the helper functions
are just a work around for Rgraphviz.
dia.cone draws a cone with (optionally) arrows as sides and centers to show the problem of factor
indeterminacy.
Value
Graphic output
Author(s)
William Revelle
See Also
The diagram functions that use the dia functions: fa.diagram,structure.diagram,omega.diagram,
and ICLUST.diagram.
draw.tetra 89
Examples
#first, show the primitives
xlim=c(-2,10)
ylim=c(0,10)
plot(NA,xlim=xlim,ylim=ylim,main="Demonstration of diagram functions",axes=FALSE,xlab="",ylab="")
ul <- dia.rect(1,9,labels="upper left",xlim=xlim,ylim=ylim)
ml <- dia.rect(1,6,"middle left",xlim=xlim,ylim=ylim)
ll <- dia.rect(1,3,labels="lower left",xlim=xlim,ylim=ylim)
bl <- dia.rect(1,1,"bottom left",xlim=xlim,ylim=ylim)
lr <- dia.ellipse(7,3,"lower right",xlim=xlim,ylim=ylim,e.size=.07)
ur <- dia.ellipse(7,9,"upper right",xlim=xlim,ylim=ylim,e.size=.07)
mr <- dia.ellipse(7,6,"middle right",xlim=xlim,ylim=ylim,e.size=.07)
lm <- dia.triangle(4,1,"Lower Middle",xlim=xlim,ylim=ylim)
br <- dia.rect(9,1,"bottom right",xlim=xlim,ylim=ylim)
dia.curve(from=ul$left,to=bl$left,"double headed",scale=-1)
dia.arrow(from=lr,to=ul,labels="right to left")
dia.arrow(from=ul,to=ur,labels="left to right")
dia.curved.arrow(from=lr,to=ll,labels ="right to left")
dia.curved.arrow(to=ur,from=ul,labels ="left to right")
dia.curve(ll$top,ul$bottom,"right") #for rectangles, specify where to point
dia.curve(ll$top,ul$bottom,"left",scale=-1) #for rectangles, specify where to point
dia.curve(mr,ur,"up") #but for ellipses, you may just point to it.
dia.curve(mr,lr,"down")
dia.curve(mr,ur,"up")
dia.curved.arrow(mr,ur,"up") #but for ellipses, you may just point to it.
dia.curved.arrow(mr,lr,"down") #but for ellipses, you may just point to it.
dia.curved.arrow(ur$right,mr$right,"3")
dia.curve(ml,mr,"across")
dia.curve(ur,lr,"top down")
dia.curved.arrow(br$top,lr$bottom,"up")
dia.curved.arrow(bl,br,"left to right")
dia.curved.arrow(br,bl,"right to left",scale=-1)
dia.arrow(bl,ll$bottom)
dia.curved.arrow(ml,ll$right)
dia.curved.arrow(mr,lr$top)
#now, put them together in a factor analysis diagram
v9 <- sim.hierarchical()
f3 <- fa(v9,3,rotate="cluster")
fa.diagram(f3,error=TRUE,side=3)
draw.tetra Draw a correlation ellipse and two normal curves to demonstrate
tetrachoric correlation
90 draw.tetra
Description
A graphic of a correlation ellipse divided into 4 regions based upon x and y cutpoints on two normal
distributions. This is also an example of using the layout function. Draw a bivariate density plot to
show how tetrachorics work.
Usage
draw.tetra(r, t1, t2,shade=TRUE)
draw.cor(r=.5,expand=10,theta=30,phi=30,N=101,nbcol=30,box=TRUE,
main="Bivariate density rho = ",cuts=NULL,all=TRUE,ellipses=TRUE,ze=.15)
Arguments
rthe underlying Pearson correlation defines the shape of the ellipse
t1 X is cut at tau
t2 Y is cut at Tau
shade shade the diagram (default is TRUE)
expand The relative height of the z axis
theta The angle to rotate the x-y plane
phi The angle above the plane to view the graph
NThe grid resolution
nbcol The color resolution
box Draw the axes
main The main title
cuts Should the graphic show cuts (e.g., cuts=c(0,0))
all Show all four parts of the tetrachoric
ellipses Draw a correlation ellipse
ze height of the ellipse if requested
Details
A graphic demonstration of the tetrachoric correlation. Used for teaching purposes. The default
values are for a correlation of .5 with cuts at 1 and 1. Any other values are possible. The code is
also a demonstration of how to use the layout function for complex graphics using base graphics.
Author(s)
William Revelle
See Also
tetrachoric to find tetrachoric correlations, irt.fa and fa.poly to use them in factor analyses,
scatter.hist to show correlations and histograms.
dummy.code 91
Examples
#if(require(mvtnorm)) {
#draw.tetra(.5,1,1)
#draw.tetra(.8,2,1)} else {print("draw.tetra requires the mvtnorm package")
#draw.cor(.5,cuts=c(0,0))}
draw.tetra(.5,1,1)
draw.tetra(.8,2,1)
draw.cor(.5,cuts=c(0,0))
dummy.code Create dummy coded variables
Description
Given a variable x with n distinct values, create n new dummy coded variables coded 0/1 for pres-
ence (1) or absence (0) of each variable. A typical application would be to create dummy coded
college majors from a vector of college majors.
Usage
dummy.code(x)
Arguments
xA vector to be transformed into dummy codes
Details
When coding demographic information, it is typical to create one variable with multiple categorical
values (e.g., ethnicity, college major, occupation). dummy.code will convert these categories into n
distinct dummy coded variables.
If using dummy coded variables as predictors, remember to use n-1 variables.
Value
A matrix of dummy coded variables
Author(s)
William Revelle
Examples
new <- dummy.code(sat.act$education)
new.sat <- data.frame(new,sat.act)
round(cor(new.sat,use="pairwise"),2)
92 eigen.loadings
Dwyer 8 cognitive variables used by Dwyer for an example.
Description
Dwyer (1937) introduced a technique for factor extension and used 8 cognitive variables from Thur-
stone. This is the example data set used in his paper.
Usage
data(Dwyer)
Format
The format is: num [1:8, 1:8] 1 0.58 -0.28 0.01 0.36 0.38 0.61 0.15 0.58 1 ... - attr(*, "dim-
names")=List of 2 ..$ : chr [1:8] "V1" "V2" "V3" "V4" ... ..$ : chr [1:8] "V1" "V2" "V3" "V4"
...
Source
Data matrix retyped from the original publication.
References
Dwyer, Paul S. (1937), The determination of the factor loadings of a given test from the known
factor loadings of other tests. Psychometrika, 3, 173-178
Examples
data(Dwyer)
Ro <- Dwyer[1:7,1:7]
Roe <- Dwyer[1:7,8]
fo <- fa(Ro,2,rotate="none")
fa.extension(Roe,fo)
eigen.loadings Convert eigen vectors and eigen values to the more normal (for psy-
chologists) component loadings
Description
The default procedures for principal component returns values not immediately equivalent to the
loadings from a factor analysis. eigen.loadings translates them into the more typical metric of eigen
vectors multiplied by the squareroot of the eigenvalues. This lets us find pseudo factor loadings if
we have used princomp or eigen.
If we use principal to do our principal components analysis, then we do not need this routine.
ellipses 93
Usage
eigen.loadings(x)
Arguments
xthe output from eigen or a list of class princomp derived from princomp
Value
A matrix of Principal Component loadings more typical for what is expected in psychometrics. That
is, they are scaled by the square root of the eigenvalues.
Note
Useful for SAPA analyses
Author(s)
< revelle@northwestern.edu >
http://personality-project.org/revelle.html
Examples
x <- eigen(Harman74.cor$cov)
x$vectors[1:8,1:4] #as they appear from eigen
y <- princomp(covmat=Harman74.cor$cov)
y$loadings[1:8,1:4] #as they appear from princomp
eigen.loadings(x)[1:8,1:4] # rescaled by the eigen values
ellipses Plot data and 1 and 2 sigma correlation ellipses
Description
For teaching correlation, it is useful to draw ellipses around the mean to reflect the correlation. This
variation of the ellipse function from John Fox’s car package does so. Input may be either two
vectors or a matrix or data.frame. In the latter cases, if the number of variables >2, then the ellipses
are done in the pairs.panels function. Ellipses may be added to existing plots. The minkowski
function is included as a generalized ellipse.
Usage
ellipses(x, y = NULL, add = FALSE, smooth=TRUE, lm=FALSE,data=TRUE, n = 2,
span=2/3, iter=3, col = "red", xlab =NULL,ylab= NULL, ...)
minkowski(r=2,add=FALSE,main=NULL,xl=1,yl=1)
94 ellipses
Arguments
xa vector,matrix, or data.frame
yOptional second vector
add Should a new plot be created, or should it be added to?
smooth smooth = TRUE -> draw a loess fit
lm lm=TRUE -> draw the linear fit
data data=TRUE implies draw the data points
nShould 1 or 2 ellipses be drawn
span averaging window parameter for the lowess fit
iter iteration parameter for lowess
col color of ellipses (default is red
xlab label for the x axis
ylab label for the y axis
... Other parameters for plotting
rr=1 draws a city block, r=2 is a Euclidean circle, r > 2 tends towards a square
main title to use when drawing Minkowski circles
xl stretch the x axis
yl stretch the y axis
Details
Ellipse dimensions are calculated from the correlation between the x and y variables and are scaled
as sqrt(1+r) and sqrt(1-r).
Value
A single plot (for 2 vectors or data frames with fewer than 3 variables. Otherwise a call is made to
pairs.panels.
Note
Adapted from John Fox’s ellipse and data.ellipse functions.
Author(s)
William Revelle
References
Galton, Francis (1888), Co-relations and their measurement. Proceedings of the Royal Society.
London Series, 45, 135-145.
See Also
pairs.panels
epi 95
Examples
data(galton)
ellipses(galton,lm=TRUE)
ellipses(galton$parent,galton$child,xlab="Mid Parent Height",
ylab="Child Height") #input are two vectors
data(sat.act)
ellipses(sat.act) #shows the pairs.panels ellipses
minkowski(2,main="Minkowski circles")
minkowski(1,TRUE)
minkowski(4,TRUE)
epi Eysenck Personality Inventory (EPI) data for 3570 participants
Description
The EPI is and has been a very frequently administered personality test with 57 measuring two broad
dimensions, Extraversion-Introversion and Stability-Neuroticism, with an additional Lie scale. De-
veloped by Eysenck and Eysenck, 1964. Eventually replaced with the EPQ which measures three
broad dimensions. This data set represents 3570 observations collected in the early 1990s at the
Personality, Motivation and Cognition lab at Northwestern. The data are included here as demon-
stration of scale construction.
Usage
data(epi)
data(epi.dictionary)
Format
A data frame with 3570 observations on the following 57 variables.
V1 a numeric vector
V2 a numeric vector
V3 a numeric vector
V4 a numeric vector
V5 a numeric vector
V6 a numeric vector
V7 a numeric vector
V8 a numeric vector
V9 a numeric vector
V10 a numeric vector
V11 a numeric vector
96 epi
V12 a numeric vector
V13 a numeric vector
V14 a numeric vector
V15 a numeric vector
V16 a numeric vector
V17 a numeric vector
V18 a numeric vector
V19 a numeric vector
V20 a numeric vector
V21 a numeric vector
V22 a numeric vector
V23 a numeric vector
V24 a numeric vector
V25 a numeric vector
V26 a numeric vector
V27 a numeric vector
V28 a numeric vector
V29 a numeric vector
V30 a numeric vector
V31 a numeric vector
V32 a numeric vector
V33 a numeric vector
V34 a numeric vector
V35 a numeric vector
V36 a numeric vector
V37 a numeric vector
V38 a numeric vector
V39 a numeric vector
V40 a numeric vector
V41 a numeric vector
V42 a numeric vector
V43 a numeric vector
V44 a numeric vector
V45 a numeric vector
V46 a numeric vector
V47 a numeric vector
V48 a numeric vector
epi 97
V49 a numeric vector
V50 a numeric vector
V51 a numeric vector
V52 a numeric vector
V53 a numeric vector
V54 a numeric vector
V55 a numeric vector
V56 a numeric vector
V57 a numeric vector
Details
The original data were collected in a group testing framework for screening participants for sub-
sequent studies. The participants were enrolled in an introductory psychology class between Fall,
1991 and Spring, 1995.
The structure of the E scale has been shown by Rocklin and Revelle (1981) to have two subcompo-
nents, Impulsivity and Sociability. These were subsequently used by Revelle, Humphreys, Simon
and Gilliland to examine the relationship between personality, caffeine induced arousal, and cogni-
tive performance.
Source
Data from the PMC laboratory at Northwestern.
References
Eysenck, H.J. and Eysenck, S. B.G. (1968). Manual for the Eysenck Personality Inventory.Educational
and Industrial Testing Service, San Diego, CA.
Rocklin, T. and Revelle, W. (1981). The measurement of extraversion: A comparison of the
Eysenck Personality Inventory and the Eysenck Personality Questionnaire. British Journal of Social
Psychology, 20(4):279-284.
Examples
data(epi)
epi.keys <- make.keys(epi,list(E = c(1, 3, -5, 8, 10, 13, -15, 17, -20, 22, 25, 27,
-29, -32, -34, -37, 39, -41, 44, 46, 49, -51, 53, 56),
N=c(2, 4, 7, 9, 11, 14, 16, 19, 21, 23, 26, 28, 31, 33, 35, 38, 40,
43, 45, 47, 50, 52, 55, 57),
L = c(6, -12, -18, 24, -30, 36, -42, -48, -54),
I =c(1, 3, -5, 8, 10, 13, 22, 39, -41),
S = c(-11, -15, 17, -20, 25, 27, -29, -32, -37, 44, 46, -51, 53)))
scores <- scoreItems(epi.keys,epi)
N <- epi[abs(epi.keys[,"N"]) >0]
E <- epi[abs(epi.keys[,"E"]) >0]
fa.lookup(epi.keys[,1:3],epi.dictionary) #show the items and keying information
98 epi.bfi
epi.bfi 13 personality scales from the Eysenck Personality Inventory and Big
5 inventory
Description
A small data set of 5 scales from the Eysenck Personality Inventory, 5 from a Big 5 inventory,
a Beck Depression Inventory, and State and Trait Anxiety measures. Used for demonstrations of
correlations, regressions, graphic displays.
Usage
data(epi.bfi)
Format
A data frame with 231 observations on the following 13 variables.
epiE EPI Extraversion
epiS EPI Sociability (a subset of Extraversion items
epiImp EPI Impulsivity (a subset of Extraversion items
epilie EPI Lie scale
epiNeur EPI neuroticism
bfagree Big 5 inventory (from the IPIP) measure of Agreeableness
bfcon Big 5 Conscientiousness
bfext Big 5 Extraversion
bfneur Big 5 Neuroticism
bfopen Big 5 Openness
bdi Beck Depression scale
traitanx Trait Anxiety
stateanx State Anxiety
Details
Self report personality scales tend to measure the “Giant 2" of Extraversion and Neuroticism or the
“Big 5" of Extraversion, Neuroticism, Agreeableness, Conscientiousness, and Openness. Here is a
small data set from Northwestern University undergraduates with scores on the Eysenck Personality
Inventory (EPI) and a Big 5 inventory taken from the International Personality Item Pool.
Source
Data were collected at the Personality, Motivation, and Cognition Lab (PMCLab) at Northwestern
by William Revelle)
error.bars 99
References
http://personality-project.org/pmc.html
Examples
data(epi.bfi)
pairs.panels(epi.bfi[,1:5])
describe(epi.bfi)
error.bars Plot means and confidence intervals
Description
One of the many functions in R to plot means and confidence intervals. Can be done using barplots
if desired. Can also be combined with such functions as boxplot to summarize distributions. Means
and standard errors are calculated from the raw data using describe. Alternatively, plots of means
+/- one standard deviation may be drawn.
Usage
error.bars(x,stats=NULL, ylab = "Dependent Variable",xlab="Independent Variable",
main=NULL,eyes=TRUE, ylim = NULL, xlim=NULL,alpha=.05,sd=FALSE, labels = NULL,
pos = NULL, arrow.len = 0.05,arrow.col="black", add = FALSE,bars=FALSE,within=FALSE,
col="blue",...)
error.bars.tab(t,way="columns",raw=FALSE,col=c('blue','red'),...)
Arguments
xA data frame or matrix of raw data
tA table of frequencies
stats Alternatively, a data.frame of descriptive stats from (e.g., describe)
ylab y label
xlab x label
main title for figure
ylim if specified, the limits for the plot, otherwise based upon the data
xlim if specified, the x limits for the plot, otherwise c(.5,nvar + .5)
eyes should ’cats eyes’ plots be drawn
alpha alpha level of confidence interval – defaults to 95% confidence interval
sd if TRUE, draw one standard deviation instead of standard errors at the alpha
level
labels X axis label
pos where to place text: below, left, above, right
100 error.bars
arrow.len How long should the top of the error bars be?
arrow.col What color should the error bars be?
add add=FALSE, new plot, add=TRUE, just points and error bars
bars bars=TRUE will draw a bar graph if you really want to do that
within should the error variance of a variable be corrected by 1-SMC?
col color(s) of the catseyes. Defaults to blue.
way Percentages are based upon the row totals (default) column totals, or grand total
of the data Table
raw If raw is FALSE, display the graphs in terms of probability, raw TRUE displays
the data in terms of raw counts
... other parameters to pass to the plot function, e.g., typ="b" to draw lines, lty="dashed"
to draw dashed lines
Details
Drawing the mean +/- a confidence interval is a frequently used function when reporting experi-
mental results. By default, the confidence interval is 1.96 standard errors of the t-distribution.
If within=TRUE, the error bars are corrected for the correlation with the other variables by reducing
the variance by a factor of (1-smc). This allows for comparisons between variables.
The error bars are normally calculated from the data using the describe function. If, alternatively,
a matrix of statistics is provided with column headings of values, means, and se, then those values
will be used for the plot (using the stats option). If n is included in the matrix of statistics, then
the distribution is drawn for a t distribution for n-1 df. If n is omitted (NULL) or is NA, then the
distribution will be a normal distribution.
If sd is TRUE, then the error bars will represent one standard deviation from the mean rather than
be a function of alpha and the standard errors.
See the last two examples for the case of plotting data with statistics from another function.
Alternatively, error.bars.tab will take tabulated data and convert to either row, column or overall
percentages, and then plot these as percentages with the equivalent standard error (based upon
sqrt(pq/N)).
Value
Graphic output showing the means + x
These confidence regions are based upon normal theory and do not take into account any skew in
the variables. More accurate confidence intervals could be found by resampling.
The error.bars.tab function will return (invisibly) the cell means and standard errors.
Author(s)
William Revelle
error.bars 101
See Also
error.crosses for two way error bars, error.bars.by for error bars for different groups
In addition, as pointed out by Jim Lemon on the R-help news group, error bars or confidence
intervals may be drawn using
function package
bar.err (agricolae)
plotCI (gplots)
xYplot (Hmisc)
dispersion (plotrix)
plotCI (plotrix)
For advice why not to draw bar graphs with error bars, see http://biostat.mc.vanderbilt.
edu/wiki/Main/DynamitePlots
Examples
x <- replicate(20,rnorm(50))
boxplot(x,notch=TRUE,main="Notched boxplot with error bars")
error.bars(x,add=TRUE)
abline(h=0)
#show 50% confidence regions and color each variable separately
error.bars(attitude,alpha=.5,
main="50 percent confidence limits",col=rainbow(ncol(attitude)) )
error.bars(attitude,bar=TRUE) #show the use of bar graphs
#combine with a strip chart and boxplot
stripchart(attitude,vertical=TRUE,method="jitter",jitter=.1,pch=19,
main="Stripchart with 95 percent confidence limits")
boxplot(attitude,add=TRUE)
error.bars(attitude,add=TRUE,arrow.len=.2)
#use statistics from somewhere else
#by specifying n, we are using the t distribution for confidences
#The first example allows the variables to be spaced along the x axis
my.stats <- data.frame(values=c(1,2,8),mean=c(10,12,18),se=c(2,3,5),n=c(5,10,20))
error.bars(stats=my.stats,type="b",main="data with confidence intervals")
#don't connect the groups
my.stats <- data.frame(values=c(1,2,8),mean=c(10,12,18),se=c(2,3,5),n=c(5,10,20))
error.bars(stats=my.stats,main="data with confidence intervals")
#by not specifying value, the groups are equally spaced
my.stats <- data.frame(mean=c(10,12,18),se=c(2,3,5),n=c(5,10,20))
rownames(my.stats) <- c("First", "Second","Third")
error.bars(stats=my.stats,xlab="Condition",ylab="Score")
102 error.bars.by
#Consider the case where we get stats from describe
temp <- describe(attitude)
error.bars(stats=temp)
#show these do not differ from the other way by overlaying the two
error.bars(attitude,add=TRUE,col="red")
#n is omitted
#the error distribution is a normal distribution
my.stats <- data.frame(mean=c(2,4,8),se=c(2,1,2))
rownames(my.stats) <- c("First", "Second","Third")
error.bars(stats=my.stats,xlab="Condition",ylab="Score")
#n is specified
#compare this with small n which shows larger confidence regions
my.stats <- data.frame(mean=c(2,4,8),se=c(2,1,2),n=c(10,10,3))
rownames(my.stats) <- c("First", "Second","Third")
error.bars(stats=my.stats,xlab="Condition",ylab="Score")
#example of arrest rates (as percentage of condition)
arrest <- data.frame(Control=c(14,21),Treated =c(3,23))
rownames(arrest) <- c("Arrested","Not Arrested")
error.bars.tab(arrest,ylab="Probability of Arrest",xlab="Control vs Treatment",
main="Probability of Arrest varies by treatment")
#Show the raw rates
error.bars.tab(arrest,raw=TRUE,ylab="Number Arrested",xlab="Control vs Treatment",
main="Count of Arrest varies by treatment")
error.bars.by Plot means and confidence intervals for multiple groups
Description
One of the many functions in R to plot means and confidence intervals. Meant mainly for demon-
stration purposes for showing the probabilty of replication from multiple samples. Can also be
combined with such functions as boxplot to summarize distributions. Means and standard errors for
each group are calculated using describe.by.
Usage
error.bars.by(x,group,by.var=FALSE,x.cat=TRUE,ylab =NULL,xlab=NULL,main=NULL,ylim= NULL,
xlim=NULL, eyes=TRUE,alpha=.05,sd=FALSE,labels=NULL, v.labels=NULL, pos=NULL,
arrow.len=.05,add=FALSE,bars=FALSE,within=FALSE,colors=c("black","blue","red"),
lty,lines=TRUE, legend=0,pch,density=-10,...)
error.bars.by 103
Arguments
xA data frame or matrix
group A grouping variable
by.var A different line for each group (default) or each variable
x.cat Is the grouping variable categorical (TRUE) or continuous (FALSE
ylab y label
xlab x label
main title for figure
ylim if specified, the y limits for the plot, otherwise based upon the data
xlim if specified, the x limits for the plot, otherwise based upon the data
eyes Should ’cats eyes’ be drawn’
alpha alpha level of confidence interval. Default is 1- alpha =95% confidence interval
sd sd=TRUE will plot Standard Deviations instead of standard errors
labels X axis label
v.labels For a bar plot legend, these are the variable labels
pos where to place text: below, left, above, right
arrow.len How long should the top of the error bars be?
add add=FALSE, new plot, add=TRUE, just points and error bars
bars Draw a barplot with error bars rather than a simple plot of the means
within Should the s.e. be corrected by the correlation with the other variables?
colors groups will be plotted in different colors (mod n.groups). See the note for how
to make them transparent.
lty line type may be specified in the case of not plotting by variables
lines By default, when plotting different groups, connect the groups with a line of
type = lty. If lines is FALSE, then do not connect the groups
legend Where should the legend be drawn: 0 (do not draw it), 1= lower right corner, 2
= bottom, 3 ... 8 continue clockwise, 9 is the center
pch The first plot symbol to use. Subsequent groups are pch + group
density How many lines/inch should fill the cats eyes. If missing, non-transparent colors
are used. If negative, transparent colors are used.
... other parameters to pass to the plot function e.g., lty="dashed" to draw dashed
lines
Details
Drawing the mean +/- a confidence interval is a frequently used function when reporting experimen-
tal results. By default, the confidence interval is 1.96 standard errors (adjusted for the t-distribution).
This function was originally just a wrapper for error.bars but has been written to allow groups to
be organized either as the x axis or as separate lines.
104 error.bars.by
If desired, a barplot with error bars can be shown. Many find this type of plot to be uninformative
(e.g., http://biostat.mc.vanderbilt.edu/DynamitePlots ) and recommend the more standard dot plot.
Note in particular, if choosing to draw barplots, the starting value is 0.0 and setting the ylim param-
eter can lead to some awkward results if 0 is not included in the ylim range. Did you really mean to
draw a bar plot in this case?
For up to three groups, the colors are by default "black", "blue" and "red". For more than 3 groups,
they are by default rainbow colors with an alpha factor (transparency) of .5.
To make colors semitransparent, set the density to a negative number. See the last example.
Value
Graphic output showing the means + x% confidence intervals for each group. For ci=1.96, and
normal data, this will be the 95% confidence region. For ci=1, the 68% confidence region.
These confidence regions are based upon normal theory and do not take into account any skew in
the variables. More accurate confidence intervals could be found by resampling.
See Also
See Also as error.crosses,error.bars
Examples
data(sat.act)
#The generic plot of variables by group
error.bars.by(sat.act[1:4],sat.act$gender,legend=7)
#a bar plot
error.bars.by(sat.act[5:6],sat.act$gender,bars=TRUE,labels=c("male","female"),
main="SAT V and SAT Q by gender",ylim=c(0,800),colors=c("red","blue"),
legend=5,v.labels=c("SATV","SATQ")) #draw a barplot
#a bar plot of SAT by age -- not recommended, see the next plot
error.bars.by(sat.act[5:6],sat.act$education,bars=TRUE,xlab="Education",
main="95 percent confidence limits of Sat V and Sat Q", ylim=c(0,800),
v.labels=c("SATV","SATQ"),legend=5,colors=c("red","blue") )
#a better graph uses points not bars
#plot SAT V and SAT Q by education
error.bars.by(sat.act[5:6],sat.act$education,TRUE, xlab="Education",
legend=5,labels=colnames(sat.act[5:6]),ylim=c(525,700),
main="self reported SAT scores by education")
#make the cats eyes semi-transparent by specifying a negative density
error.bars.by(sat.act[5:6],sat.act$education,TRUE, xlab="Education",
legend=5,labels=colnames(sat.act[5:6]),ylim=c(525,700),
main="self reported SAT scores by education",density=-10)
#now for a more complicated examples using 25 big 5 items scored into 5 scales
#and showing age trends by decade
#this shows how to convert many levels of a grouping variable (age) into more manageable levels.
data(bfi) #The Big 5 data
#first create the keys
keys.list <- list(Agree=c(-1,2:5),Conscientious=c(6:8,-9,-10),
error.crosses 105
Extraversion=c(-11,-12,13:15),Neuroticism=c(16:20),Openness = c(21,-22,23,24,-25))
keys <- make.keys(bfi,keys.list)
#then create the scores for those older than 10 and less than 80
bfis <- subset(bfi,((bfi$age > 10) & (bfi$age < 80)))
scores <- scoreItems(keys,bfis,min=1,max=6) #set the right limits for item reversals
#now draw the results by age
error.bars.by(scores$scores,round(bfis$age/10)*10,by.var=TRUE,
main="BFI age trends",legend=3,labels=colnames(scores$scores),
xlab="Age",ylab="Mean item score")
error.bars.by(scores$scores,round(bfis$age/10)*10,by.var=TRUE,
main="BFI age trends",legend=3,labels=colnames(scores$scores),
xlab="Age",ylab="Mean item score",density=-10)
error.crosses Plot x and y error bars
Description
Given two vectors of data (X and Y), plot the means and show standard errors in both X and Y
directions.
Usage
error.crosses(x,y,labels=NULL,main=NULL,xlim=NULL,ylim= NULL,
xlab=NULL,ylab=NULL,pos=NULL,offset=1,arrow.len=.2,alpha=.05,sd=FALSE,add=FALSE,...)
Arguments
xA vector of data or summary statistics (from Describe)
yA second vector of data or summary statistics (also from Describe)
labels the names of each pair – defaults to rownames of x
main The title for the graph
xlim xlim values if desired– defaults to min and max mean(x) +/- 2 se
ylim ylim values if desired – defaults to min and max mean(y) +/- 2 se
xlab label for x axis – grouping variable 1
ylab label for y axis – grouping variable 2
pos Labels are located where with respect to the mean?
offset Labels are then offset from this location
arrow.len Arrow length
alpha alpha level of error bars
sd if sd is TRUE, then draw means +/- 1 sd)
add if TRUE, overlay the values with a prior plot
... Other parameters for plot
106 error.crosses
Details
For an example of two way error bars describing the effects of mood manipulations upon positive
and negative affect, see http://personality-project.org/revelle/publications/happy-sad-appendix/
FIG.A-6.pdf
The second example shows how error crosses can be done for multiple variables where the grouping
variable is found dynamically. The errorCircles example shows how to do this in one step.
Author(s)
William Revelle
<revelle@northwestern.edu>
See Also
To draw error bars for single variables error.bars, or by groups error.bars.by, or to find
descriptive statistics describe or descriptive statistics by a grouping variable describeBy and
statsBy.
A much improved version is now called errorCircles.
Examples
#just draw one pair of variables
desc <- describe(attitude)
x <- desc[1,]
y <- desc[2,]
error.crosses(x,y,xlab=rownames(x),ylab=rownames(y))
#now for a bit more complicated plotting
data(bfi)
desc <- describeBy(bfi[1:25],bfi$gender) #select a high and low group
error.crosses(desc$'1',desc$'2',ylab="female scores",xlab="male scores",main="BFI scores by gender")
abline(a=0,b=1)
#do it from summary statistics (using standard errors)
g1.stats <- data.frame(n=c(10,20,30),mean=c(10,12,18),se=c(2,3,5))
g2.stats <- data.frame(n=c(15,20,25),mean=c(6,14,15),se =c(1,2,3))
error.crosses(g1.stats,g2.stats)
#Or, if you prefer to draw +/- 1 sd. instead of 95% confidence
g1.stats <- data.frame(n=c(10,20,30),mean=c(10,12,18),sd=c(2,3,5))
g2.stats <- data.frame(n=c(15,20,25),mean=c(6,14,15),sd =c(1,2,3))
error.crosses(g1.stats,g2.stats,sd=TRUE)
#and seem even fancy plotting: This is taken from a study of mood
#four films were given (sad, horror, neutral, happy)
#with a pre and post test
data(affect)
colors <- c("black","red","white","blue")
films <- c("Sad","Horror","Neutral","Happy")
errorCircles 107
affect.mat <- describeBy(affect[10:17],affect$Film,mat=TRUE)
error.crosses(affect.mat[c(1:4,17:20),],affect.mat[c(5:8,21:24),],
labels=films[affect.mat$group1],xlab="Energetic Arousal",
ylab="Tense Arousal",col=colors[affect.mat$group1],pch=16,cex=2)
errorCircles Two way plots of means, error bars, and sample sizes
Description
Given a matrix or data frame, data, find statistics based upon a grouping variable and then plot x
and y means with error bars for each value of the grouping variable. If the data are paired (e.g. by
gender), then plot means and error bars for the two groups on all variables.
Usage
errorCircles(x, y, data, ydata = NULL, group=NULL, paired = FALSE, labels = NULL,
main = NULL, xlim = NULL, ylim = NULL, xlab = NULL, ylab = NULL,add=FALSE, pos = NULL,
offset = 1, arrow.len = 0.2, alpha = 0.05, sd = FALSE, bars = TRUE, circles = TRUE, ...)
Arguments
xThe x variable (by name or number) to plot
yThe y variable (name or number) to plot
data The matrix or data.frame to use for the x data
ydata If plotting data from two data.frames, then the y variable of the ydata frame will
be used.
group If specified, then statsBy is called first to find the statistics by group
paired If TRUE, plot all x and y variables for the two values of the grouping variable.
labels Variable names
main Main title for plot
xlim xlim values if desired– defaults to min and max mean(x) +/- 2 se
ylim ylim values if desired – defaults to min and max mean(y) +/- 2 se
xlab label for x axis – grouping variable 1
ylab label for y axis – grouping variable 2
add If TRUE, add to the prior plot
pos Labels are located where with respect to the mean?
offset Labels are then offset from this location
arrow.len Arrow length
alpha alpha level of error bars
sd if sd is TRUE, then draw means +/- 1 sd)
bars Should error.bars be drawn for both x and y
circles Should circles representing the relative sample sizes be drawn?
... Other parameters for plot
108 errorCircles
Details
When visualizing the effect of an experimental manipulation or the relationship of multiple groups,
it is convenient to plot their means as well as their confidence regions in a two dimensional space.
Value
If the group variable is specified, then the statistics from statsBy are (invisibly) returned.
Note
Basically this is a combination (and improvement) of statsBy with error.crosses. Can also
serve some of the functionality of error.bars.by (see the last example).
Author(s)
William Revelle
See Also
statsBy,describeBy,error.crosses
Examples
#BFI scores for males and females
errorCircles(1:25,1:25,data=bfi,group="gender",paired=TRUE,ylab="female scores",
xlab="male scores",main="BFI scores by gender")
abline(a=0,b=1)
#drop the circles since all samples are the same sizes
errorCircles(1:25,1:25,data=bfi,group="gender",paired=TRUE,circles=FALSE,
ylab="female scores",xlab="male scores",main="BFI scores by gender")
abline(a=0,b=1)
data(affect)
colors <- c("black","red","white","blue")
films <- c("Sad","Horror","Neutral","Happy")
affect.stats <- errorCircles("EA2","TA2",data=affect[-c(1,20)],group="Film",labels=films,
xlab="Energetic Arousal",ylab="Tense Arousal",ylim=c(10,22),xlim=c(8,20),
pch=16,cex=2,col=colors, main ="EA and TA pre and post affective movies")
#now, use the stats from the prior run
errorCircles("EA1","TA1",data=affect.stats,labels=films,pch=16,cex=2,col=colors,add=TRUE)
#Can also provide error.bars.by functionality
errorCircles(2,5,group=2,data=sat.act,circles=FALSE,pch=16,col="blue",
ylim= c(200,800),main="SATV by education",labels="")
#just do the breakdown and then show the points
# errorCircles(3,5,group=3,data=sat.act,circles=FALSE,pch=16,col="blue",
# ylim= c(200,800),main="SATV by age",labels="",bars=FALSE)
fa 109
fa Exploratory Factor analysis using MinRes (minimum residual) as well
as EFA by Principal Axis, Weighted Least Squares or Maximum Like-
lihood
Description
Among the many ways to do latent variable exploratory factor analysis (EFA), one of the better is to
use Ordinary Least Squares (OLS) to find the minimum residual (minres) solution. This produces
solutions very similar to maximum likelihood even for badly behaved matrices. A variation on
minres is to do weighted least squares (WLS). Perhaps the most conventional technique is principal
axes (PAF). An eigen value decomposition of a correlation matrix is done and then the communali-
ties for each variable are estimated by the first n factors. These communalities are entered onto the
diagonal and the procedure is repeated until the sum(diag(r)) does not vary. Yet another estimate
procedure is maximum likelihood. For well behaved matrices, maximum likelihood factor analysis
(either in the fa or in the factanal function) is probably preferred. Bootstrapped confidence intervals
of the loadings and interfactor correlations are found by fa with n.iter > 1.
Usage
fa(r,nfactors=1,n.obs = NA,n.iter=1, rotate="oblimin", scores="regression",
residuals=FALSE, SMC=TRUE, covar=FALSE,missing=FALSE,impute="median",
min.err = 0.001, max.iter = 50,symmetric=TRUE, warnings=TRUE, fm="minres",
alpha=.1,p=.05,oblique.scores=FALSE,np.obs,use="pairwise",cor="cor",weight=NULL,...)
fac(r,nfactors=1,n.obs = NA, rotate="oblimin", scores="tenBerge", residuals=FALSE,
SMC=TRUE, covar=FALSE,missing=FALSE,impute="median",min.err = 0.001,
max.iter=50,symmetric=TRUE,warnings=TRUE,fm="minres",alpha=.1,
oblique.scores=FALSE,np.obs,use="pairwise",cor="cor",weight=NULL,...)
fa.poly(x,nfactors=1,n.obs = NA, n.iter=1, rotate="oblimin", SMC=TRUE, missing=FALSE,
impute="median", min.err = .001, max.iter=50, symmetric=TRUE, warnings=TRUE,
fm="minres",alpha=.1, p =.05,scores="regression", oblique.scores=TRUE,
weight=NULL,global=TRUE,...) #deprecated
factor.minres(r, nfactors=1, residuals = FALSE, rotate = "varimax",n.obs = NA,
scores = FALSE,SMC=TRUE, missing=FALSE,impute="median",min.err = 0.001, digits = 2,
max.iter = 50,symmetric=TRUE,warnings=TRUE,fm="minres") #deprecated
factor.wls(r,nfactors=1,residuals=FALSE,rotate="varimax",n.obs = NA,
scores=FALSE,SMC=TRUE,missing=FALSE,impute="median", min.err = .001,
digits=2,max.iter=50,symmetric=TRUE,warnings=TRUE,fm="wls") #deprecated
Arguments
rA correlation or covariance matrix or a raw data matrix. If raw data, the corre-
lation matrix will be found using pairwise deletion. If covariances are supplied,
they will be converted to correlations unless the covar option is TRUE.
110 fa
xFor fa.poly.ci, only raw data may be used
nfactors Number of factors to extract, default is 1
n.obs Number of observations used to find the correlation matrix if using a correlation
matrix. Used for finding the goodness of fit statistics. Must be specified if using
a correlaton matrix and finding confidence intervals.
np.obs The pairwise number of observations. Used if using a correlation matrix and
asking for a minchi solution.
rotate "none", "varimax", "quartimax", "bentlerT", "equamax", "varimin", "geominT"
and "bifactor" are orthogonal rotations. "promax", "oblimin", "simplimax",
"bentlerQ, "geominQ" and "biquartimin" and "cluster" are possible oblique trans-
formations of the solution. The default is to do a oblimin transformation, al-
though versions prior to 2009 defaulted to varimax.
n.iter Number of bootstrap interations to do in fa or fa.poly
residuals Should the residual matrix be shown
scores the default="regression" finds factor scores using regression. Alternatives for
estimating factor scores include simple regression ("Thurstone"), correlaton pre-
serving ("tenBerge") as well as "Anderson" and "Bartlett" using the appropriate
algorithms ( factor.scores). Although scores="tenBerge" is probably pre-
ferred for most solutions, it will lead to problems with some improper correla-
tion matrices.
SMC Use squared multiple correlations (SMC=TRUE) or use 1 as initial communality
estimate. Try using 1 if imaginary eigen values are reported. If SMC is a vector
of length the number of variables, then these values are used as starting values
in the case of fm=’pa’.
covar if covar is TRUE, factor the covariance matrix, otherwise factor the correlation
matrix
missing if scores are TRUE, and missing=TRUE, then impute missing values using either
the median or the mean
impute "median" or "mean" values are used to replace missing values
min.err Iterate until the change in communalities is less than min.err
digits How many digits of output should be returned– deprecated – now specified in
the print function
max.iter Maximum number of iterations for convergence
symmetric symmetric=TRUE forces symmetry by just looking at the lower off diagonal
values
warnings warnings=TRUE => warn if number of factors is too many
fm factoring method fm="minres" will do a minimum residual (OLS), fm="wls"
will do a weighted least squares (WLS) solution, fm="gls" does a generalized
weighted least squares (GLS), fm="pa" will do the principal factor solution,
fm="ml" will do a maximum likelihood factor analysis. fm="minchi" will min-
imize the sample size weighted chi square when treating pairwise correlations
with different number of subjects per pair.
alpha alpha level for the confidence intervals for RMSEA
fa 111
pif doing iterations to find confidence intervals, what probability values should
be found for the confidence intervals
oblique.scores When factor scores are found, should they be based on the structure matrix (de-
fault) or the pattern matrix (oblique.scores=TRUE).
weight If not NULL, a vector of length n.obs that contains weights for each observation.
The NULL case is equivalent to all cases being weighted 1.
use How to treat missing data, use="pairwise" is the default". See cor for other
options.
cor How to find the correlations: "cor" is Pearson", "cov" is covariance, "tet" is
tetrachoric, "poly" is polychoric, "mixed" uses mixed cor for a mixture of tetra-
chorics, polychorics, Pearsons, biserials, and polyserials, Yuleb is Yulebonett,
Yuleq and YuleY are the obvious Yule coefficients as appropriate
global should overall taus be used in polychoric or should they be found for each pair.
Necessary to be set to false in the case of different number of alternatives for
each item.
... additional parameters, specifically, keys may be passed if using the target rota-
tion, or delta if using geominQ, or whether to normalize if using Varimax
Details
Factor analysis is an attempt to approximate a correlation or covariance matrix with one of lesser
rank. The basic model is that nRn≈nFkkF0
n+U2where k is much less than n. There are many
ways to do factor analysis, and maximum likelihood procedures are probably the most preferred
(see factanal ). The existence of uniquenesses is what distinguishes factor analysis from principal
components analysis (e.g., principal). If variables are thought to represent a “true" or latent part
then factor analysis provides an estimate of the correlations with the latent factor(s) representing
the data. If variables are thought to be measured without error, then principal components provides
the most parsimonious description of the data.
The fa function will do factor analyses using one of four different algorithms: minimum residual
(minres), principal axes, weighted least squares, or maximum likelihood.
Principal axes factor analysis has a long history in exploratory analysis and is a straightforward
procedure. Successive eigen value decompositions are done on a correlation matrix with the diago-
nal replaced with diag (FF’) until P(diag(F F 0)) does not change (very much). The current limit
of max.iter =50 seems to work for most problems, but the Holzinger-Harmon 24 variable problem
needs about 203 iterations to converge for a 5 factor solution.
Not all factor programs that do principal axes do iterative solutions. The example from the SAS
manual (Chapter 26) is such a case. To achieve that solution, it is necessary to specify that the
max.iterations = 1. Comparing that solution to an iterated one (the default) shows that iterations
improve the solution. In addition, fm="minres" or fm="mle" produces even better solutions for this
example.
Principal axes may be used in cases when maximum likelihood solutions fail to converge, although
fm="minres" will also do that and tends to produce better (smaller residuals) solutions.
The fm="minchi" option is a variation on the "minres" (ols) solution and minimizes the sample size
weighted residuals rather than just the residuals. This was developed to handle the problem of data
that Massively Missing Completely at Random (MMCAR) which a condition that happens in the
SAPA project.
112 fa
A problem in factor analysis is to find the best estimate of the original communalities. Using the
Squared Multiple Correlation (SMC) for each variable will underestimate the communalities, using
1s will over estimate. By default, the SMC estimate is used. In either case, iterative techniques will
tend to converge on a stable solution. If, however, a solution fails to be achieved, it is useful to try
again using ones (SMC =FALSE). Alternatively, a vector of starting values for the communalities
may be specified by the SMC option.
The iterated principal axes algorithm does not attempt to find the best (as defined by a maximum
likelihood criterion) solution, but rather one that converges rapidly using successive eigen value
decompositions. The maximum likelihood criterion of fit and the associated chi square value are
reported, and will be worse than that found using maximum likelihood procedures.
The minimum residual (minres) solution is an unweighted least squares solution that takes a slightly
different approach. It uses the optim function and adjusts the diagonal elements of the correlation
matrix to mimimize the squared residual when the factor model is the eigen value decomposition of
the reduced matrix. MINRES and PA will both work when ML will not, for they can be used when
the matrix is singular. At least on a number of test cases, the MINRES solution is slightly more
similar to the ML solution than is the PA solution. To a great extent, the minres and wls solutions
follow ideas in the factanal function.
The weighted least squares (wls) solution weights the residual matrix by 1/ diagonal of the inverse
of the correlation matrix. This has the effect of weighting items with low communalities more than
those with high communalities.
The generalized least squares (gls) solution weights the residual matrix by the inverse of the corre-
lation matrix. This has the effect of weighting those variables with low communalities even more
than those with high communalities.
The maximum likelihood solution takes yet another approach and finds those communality values
that minimize the chi square goodness of fit test. The fm="ml" option provides a maximum likeli-
hood solution following the procedures used in factanal but does not provide all the extra features
of that function.
Test cases comparing the output to SPSS suggest that the PA algorithm matches what SPSS calls uls,
and that the wls solutions are equivalent in their fits. The wls and gls solutions have slightly larger
eigen values, but slightly worse fits of the off diagonal residuals than do the minres or maximum
likelihood solutions. Comparing the results to the examples in Harman (76), the PA solution with no
iterations matches what Harman calls Principal Axes (as does SAS), while the iterated PA solution
matches his minres solution. The minres solution found in psych tends to have slightly smaller off
diagonal residuals (as it should) than does the iterated PA solution.
Although for items, it is typical to find factor scores by scoring the salient items (using, e.g.,
scoreItems) factor scores can be estimated by regression as well as several other means. There
are multiple approaches that are possible (see Grice, 2001) and the one taken here was developed
by tenBerge et al. (see factor.scores. The alternative, which will match factanal is to find
the scores using regression – Thurstone’s least squares regression where the weights are found by
W=R(−1)Swhere R is the correlation matrix of the variables ans S is the structure matrix.
Then, factor scores are just F s =XW .
In the oblique case, the factor loadings are referred to as Pattern coefficients and are related to the
Structure coefficients by S=PΦand thus P=SΦ−1. When estimating factor scores, fa and
factanal differ in that fa finds the factors from the Structure matrix while factanal seems to do
it from the Pattern matrix. Thus, although in the orthogonal case, fa and factanal agree perfectly in
fa 113
their factor score estimates, they do not agree in the case of oblique factors. Setting oblique.scores
= TRUE will produce factor score estimate that match those of factanal.
It is sometimes useful to extend the factor solution to variables that were not factored. This may
be done using fa.extension. Factor extension is typically done in the case where some variables
were not appropriate to factor, but factor loadings on the original factors are still desired.
For dichotomous items or polytomous items, it is recommended to analyze the tetrachoric or
polychoric correlations rather than the Pearson correlations. This may be done by specifying
cor="poly" or cor="tet" or cor="mixed" if the data have a mixture of dichotomous, polytomous,
and continous variables.
Analysis of dichotomous or polytomous data may also be done by using irt.fa or fa.poly func-
tions. In the first case, the factor analysis results are reported in Item Response Theory (IRT) terms,
although the original factor solution is returned in the results. In the later case, a typical factor load-
ings matrix is returned, but the tetrachoric/polychoric correlation matrix and item statistics are saved
for reanalysis by irt.fa. (See also the mixed.cor function to find correlations from a mixture of
continuous, dichotomous, and polytomous items.)
Of the various rotation/transformation options, varimax, Varimax, quartimax, bentlerT, geominT,
and bifactor do orthogonal rotations. Promax transforms obliquely with a target matix equal to the
varimax solution. oblimin, quartimin, simplimax, bentlerQ, geominQ and biquartimin are oblique
transformations. Most of these are just calls to the GPArotation package. The “cluster” option does
a targeted rotation to a structure defined by the cluster representation of a varimax solution. With
the optional "keys" parameter, the "target" option will rotate to a target supplied as a keys matrix.
(See target.rot.)
Two additional target rotation options are available through calls to GPArotation. These are the
targetQ (oblique) and targetT (orthogonal) target rotations of Michael Browne. See target.rot
for more documentation.
The "bifactor" rotation implements the Jennrich and Bentler (2011) bifactor rotation by calling the
GPForth function in the GPArotation package and using two functions adapted from the MatLab
code of Jennrich and Bentler.
There are two varimax rotation functions. One, Varimax, in the GPArotation package does not by
default apply Kaiser normalization. The other, varimax, in the stats package, does. It appears that
the two rotation functions produce slightly different results even when normalization is set. For
consistency with the other rotation functions, Varimax is probably preferred.
The rotation matrix (rot.mat) is returned from all of these options. This is the inverse of the Th
(theta?) object returned by the GPArotation package. The correlations of the factors may be found
by Φ = θ0θ
There are three ways to handle dichotomous or polytomous responses: fa with the cor="poly"
option, fa.poly which will return the tetrachoric or polychoric correlation matrix, as well as the
normal factor analysis output, and irt.fa which returns a two parameter irt analysis as well as the
normal fa output.
When factor analyzing items with dichotomous or polytomous responses, the irt.fa function pro-
vides an Item Response Theory representation of the factor output. The factor analysis results are
available, however, as an object in the irt.fa output.
fa.poly is appropriate if the data are categorical (but just setting the cor="poly" option works as
well). It will produce normal factor analysis output but also will save the polychoric matrix (rho)
and items difficulties (tau) for subsequent irt analyses. fa.poly will, by default, find factor scores
114 fa
if the data are available. The correlations are found using either tetrachoric or polychoric and
then this matrix is factored. Weights from the factors are then applied to the original data to estimate
factor scores.
The function fa will repeat the analysis n.iter times on a bootstrapped sample of the data (if they
exist) or of a simulated data set based upon the observed correlation matrix. The mean estimate and
standard deviation of the estimate are returned and will print the original factor analysis as well as
the alpha level confidence intervals for the estimated coefficients. The bootstrapped solutions are
rotated towards the original solution using target.rot. The factor loadings are z-transformed, aver-
aged and then back transformed. This leads to an error in the case of Heywood cases. The probably
better alternative is to just find the mean bootstrapped value and find the confidence intervals based
upon the observed range of values. The default is to have n.iter =1 and thus not do bootstrapping.
fa.poly will find confidence intervals for a factor solution for dichotomous or polytomous items
(set n.iter > 1 to do so). But, so will fa with the cor="poly" option. Perhaps more useful is to find
the Item Response Theory parameters equivalent to the factor loadings reported in fa.poly by using
the irt.fa function.
Some correlation matrices that arise from using pairwise deletion or from tetrachoric or polychoric
matrices will not be proper. That is, they will not be positive semi-definite (all eigen values >= 0).
The cor.smooth function will adjust correlation matrices (smooth them) by making all negative
eigen values slightly greater than 0, rescaling the other eigen values to sum to the number of vari-
ables, and then recreating the correlation matrix. See cor.smooth for an example of this problem
using the burt data set.
For those who like SPSS type output, the measure of factoring adequacy known as the Kaiser-
Meyer-Olkin KMO test may be found from the correlation matrix or data matrix using the KMO func-
tion. Similarly, the Bartlett’s test of Sphericity may be found using the cortest.bartlett function.
For those who want to have an object of the variances accounted for, this is returned invisibly by
the print function. (e.g., p <- print(fa(ability))$Vaccounted )
The output from the print.psych.fa function displays the factor loadings (from the pattern matrix,
the h2 (communalities) the u2 (the uniquenesses), com (the complexity of the factor loadings for
that variable (see below). In the case of an orthogonal solution, h2 is merely the row sum of the
squared factor loadings. But for an oblique solution, it is the row sum of the orthogonal factor
loadings (remember, that rotations or transformations do not change the communality).
Value
values Eigen values of the common factor solution
e.values Eigen values of the original matrix
communality Communality estimates for each item. These are merely the sum of squared
factor loadings for that item.
rotation which rotation was requested?
n.obs number of observations specified or found
loadings An item by factor (pattern) loading matrix of class “loadings" Suitable for use
in other programs (e.g., GPA rotation or factor2cluster. To show these by sorted
order, use print.psych with sort=TRUE
complexity Hoffman’s index of complexity for each item. This is just (Σa2
i)2
Σa4
i
where a_i
is the factor loading on the ith factor. From Hofmann (1978), MBR. See also
Pettersson and Turkheimer (2010).
fa 115
Structure An item by factor structure matrix of class “loadings". This is just the loadings
(pattern) matrix times the factor intercorrelation matrix.
fit How well does the factor model reproduce the correlation matrix. This is just
Σr2
ij −Σr∗2
ij
Σr2
ij
(See VSS,ICLUST, and principal for this fit statistic.
fit.off how well are the off diagonal elements reproduced?
dof Degrees of Freedom for this model. This is the number of observed correlations
minus the number of independent parameters. Let n=Number of items, nf =
number of factors then
dof =n∗(n−1)/2−n∗nf +nf ∗(nf −1)/2
objective Value of the function that is minimized by a maximum likelihood procedures.
This is reported for comparison purposes and as a way to estimate chi square
goodness of fit. The objective function is
f=log(trace((F F 0+U2)−1R)−log(|(F F 0+U2)−1R|)−n.items. When
using MLE, this function is minimized. When using OLS (minres), although
we are not minimizing this function directly, we can still calculate it in order to
compare the solution to a MLE fit.
STATISTIC If the number of observations is specified or found, this is a chi square based
upon the objective function, f (see above). Using the formula from factanal(which
seems to be Bartlett’s test) :
χ2= (n.obs −1−(2 ∗p+ 5)/6−(2 ∗factors)/3)) ∗f
PVAL If n.obs > 0, then what is the probability of observing a chisquare this large or
larger?
Phi If oblique rotations (using oblimin from the GPArotation package or promax)
are requested, what is the interfactor correlation.
communality.iterations
The history of the communality estimates (For principal axis only.) Probably
only useful for teaching what happens in the process of iterative fitting.
residual The matrix of residual correlations after the factor model is applied. To display
it conveniently, use the residuals command.
chi When normal theory fails (e.g., in the case of non-positive definite matrices), it
useful to examine the empirically derived χ2based upon the sum of the squared
residuals * N. This will differ slightly from the MLE estimate which is based
upon the fitting function rather than the actual residuals.
rms This is the sum of the squared (off diagonal residuals) divided by the degrees
of freedom. Comparable to an RMSEA which, because it is based upon χ2,
requires the number of observations to be specified. The rms is an empirical
value while the RMSEA is based upon normal theory and the non-central χ2
distribution. That is to say, if the residuals are particularly non-normal, the rms
value and the associated χ2and RMSEA can differ substantially.
crms rms adjusted for degrees of freedom
RMSEA The Root Mean Square Error of Approximation is based upon the non-central
χ2distribution and the χ2estimate found from the MLE fitting function. With
normal theory data, this is fine. But when the residuals are not distributed ac-
cording to a noncentral χ2, this can give very strange values. (And thus the
116 fa
confidence intervals can not be calculated.) The RMSEA is a conventional in-
dex of goodness (badness) of fit but it is also useful to examine the actual rms
values.
TLI The Tucker Lewis Index of factoring reliability which is also known as the non-
normed fit index.
BIC Based upon χ2with the assumption of normal theory and using the χ2found
using the objective function defined above. This is just χ2−2df
eBIC When normal theory fails (e.g., in the case of non-positive definite matrices), it
useful to examine the empirically derived eBIC based upon the empirical χ2- 2
df.
R2 The multiple R square between the factors and factor score estimates, if they
were to be found. (From Grice, 2001). Derived from R2 is is the minimum
correlation between any two factor estimates = 2R2-1.
r.scores The correlations of the factor score estimates using the specified model, if they
were to be found. Comparing these correlations with that of the scores them-
selves will show, if an alternative estimate of factor scores is used (e.g., the
tenBerge method), the problem of factor indeterminacy. For these correlations
will not necessarily be the same.
weights The beta weights to find the factor score estimates. These are also used by the
predict.psych function to find predicted factor scores for new cases.
scores The factor scores as requested. Note that these scores reflect the choice of the
way scores should be estimated (see scores in the input). That is, simple regres-
sion ("Thurstone"), correlaton preserving ("tenBerge") as well as "Anderson"
and "Bartlett" using the appropriate algorithms (see factor.scores). The cor-
relation between factor score estimates (r.scores) is based upon using the regres-
sion/Thurstone approach. The actual correlation between scores will reflect the
rotation algorithm chosen and may be found by correlating those scores.
valid The validity coffiecient of course coded (unit weighted) factor score estimates
(From Grice, 2001)
score.cor The correlation matrix of course coded (unit weighted) factor score estimates, if
they were to be found, based upon the loadings matrix rather than the weights
matrix.
rot.mat The rotation matrix as returned from GPArotation.
Note
Thanks to Erich Studerus for some very helpful suggestions about various rotation and factor scor-
ing algorithms, and to Gumundur Arnkelsson for suggestions about factor scores for singular ma-
trices.
The fac function is the original fa function which is now called by fa repeatedly to get confidence
intervals.
SPSS will sometimes use a Kaiser normalization before rotating. This will lead to different solutions
than reported here. To get the Kaiser normalized loadings, use kaiser.
The communality for a variable is the amount of variance accounted for by all of the factors. That
is to say, for orthogonal factors, it is the sum of the squared factor loadings (rowwise). The com-
munality is insensitive to rotation. However, if an oblique solution is found, then the communality
fa 117
is not the sum of squared pattern coefficients. In both cases (oblique or orthogonal) the communal-
ity is the diagonal of the reproduced correlation matrix where nRn=nPkk ΦkkP0
nwhere P is the
pattern matrix and Φis the factor intercorrelation matrix. This is the same, of course to multiplying
the pattern by the structure: R=P S0R = PS’ where the Structure matrix is S= ΦP. Similarly,
the eigen values are the diagonal of the product kΦkkP0
nnPk.
A frequently asked question is why are the factor names of the rotated solution not in ascending
order? That is, for example, if factoring the 25 items of the bfi, the factor names are MR2 MR3
MR5 MR1 MR4, rather than the seemingly more logical "MR1" "MR2" "MR3" "MR4" "MR5".
This is for pedagogical reasons, in that factors as extracted are orthogonal and are in order of
amount of variance accounted for. But when rotated (orthogonally) or transformed (obliquely) the
simple structure solution does not preserve that order. The factor names are, of course, arbitrary,
and are kept with the original names to show the effect of rotation/transformation. To give them
names associated with their ordinal position, simply paste("F", 1:nf, sep="") where nf is the number
of factors. See the last example.
Correction to documentation: as of September, 2014, the oblique.scores option is correctly ex-
plained. (It had been backwards.) The default (oblique.scores=FALSE) finds scores based upon the
Structure matrix, while oblique.scores=TRUE finds them based upon the pattern matrix. The latter
case matches factanal. This error was detected by Mark Seeto.
Author(s)
William Revelle
References
Gorsuch, Richard, (1983) Factor Analysis. Lawrence Erlebaum Associates.
Grice, James W. (2001), Computing and evaluating factor scores. Psychological Methods, 6, 430-
450
Harman, Harry and Jones, Wayne (1966) Factor analysis by minimizing residuals (minres), Psy-
chometrika, 31, 3, 351-368.
Hofmann, R. J. ( 1978 ) . Complexity and simplicity as objective indices descriptive of factor
solutions. Multivariate Behavioral Research, 13, 247-250.
Pettersson E, Turkheimer E. (2010) Item selection, evaluation, and simple structure in personality
data. Journal of research in personality, 44(4), 407-420.
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
See Also
principal for principal components analysis (PCA). PCA will give very similar solutions to factor
analysis when there are many variables. The differences become more salient as the number vari-
ables decrease. The PCA and FA models are actually very different and should not be confused.
One is a model of the observed variables, the other is a model of latent variables.
irt.fa for Item Response Theory analyses using factor analysis, using the two parameter IRT
equivalent of loadings and difficulties.
VSS will produce the Very Simple Structure (VSS) and MAP criteria for the number of factors,
nfactors to compare many different factor criteria.
118 fa
ICLUST will do a hierarchical cluster analysis alternative to factor analysis or principal components
analysis.
predict.psych to find predicted scores based upon new data, fa.extension to extend the factor
solution to new variables, omega for hierarchical factor analysis with one general factor. codefa.multi
for hierarchical factor analysis with an arbitrary number of higher order factors.
fa.sort will sort the factor loadings into echelon form. fa.organize will reorganize the factor
pattern matrix into any arbitrary order of factors and items.
KMO and cortest.bartlett for various tests that some people like.
factor2cluster will prepare unit weighted scoring keys of the factors that can be used with
scoreItems.
fa.lookup will print the factor analysis loadings matrix along with the item “content" taken from
a dictionary of items. This is useful when examining the meaning of the factors.
anova.psych allows for testing the difference between two (presumably nested) factor models .
Examples
#using the Harman 24 mental tests, compare a principal factor with a principal components solution
pc <- principal(Harman74.cor$cov,4,rotate="varimax") #principal components
pa <- fa(Harman74.cor$cov,4,fm="pa" ,rotate="varimax") #principal axis
uls <- fa(Harman74.cor$cov,4,rotate="varimax") #unweighted least squares is minres
wls <- fa(Harman74.cor$cov,4,fm="wls") #weighted least squares
#to show the loadings sorted by absolute value
print(uls,sort=TRUE)
#then compare with a maximum likelihood solution using factanal
mle <- factanal(covmat=Harman74.cor$cov,factors=4)
factor.congruence(list(mle,pa,pc,uls,wls))
#note that the order of factors and the sign of some of factors may differ
#finally, compare the unrotated factor, ml, uls, and wls solutions
wls <- fa(Harman74.cor$cov,4,rotate="none",fm="wls")
pa <- fa(Harman74.cor$cov,4,rotate="none",fm="pa")
minres <- factanal(factors=4,covmat=Harman74.cor$cov,rotation="none")
mle <- fa(Harman74.cor$cov,4,rotate="none",fm="mle")
uls <- fa(Harman74.cor$cov,4,rotate="none",fm="uls")
factor.congruence(list(minres,mle,pa,wls,uls))
#in particular, note the similarity of the mle and min res solutions
#note that the order of factors and the sign of some of factors may differ
#an example of where the ML and PA and MR models differ is found in Thurstone.33.
#compare the first two factors with the 3 factor solution
Thurstone.33 <- as.matrix(Thurstone.33)
mle2 <- fa(Thurstone.33,2,rotate="none",fm="mle")
mle3 <- fa(Thurstone.33,3 ,rotate="none",fm="mle")
pa2 <- fa(Thurstone.33,2,rotate="none",fm="pa")
pa3 <- fa(Thurstone.33,3,rotate="none",fm="pa")
mr2 <- fa(Thurstone.33,2,rotate="none")
fa.diagram 119
mr3 <- fa(Thurstone.33,3,rotate="none")
factor.congruence(list(mle2,mr2,pa2,mle3,pa3,mr3))
#f5 <- fa(bfi[1:25],5)
#f5 #names are not in ascending numerical order (see note)
#colnames(f5$loadings) <- paste("F",1:5,sep="")
#f5
fa.diagram Graph factor loading matrices
Description
Factor analysis or principal components analysis results are typically interpreted in terms of the
major loadings on each factor. These structures may be represented as a table of loadings or graph-
ically, where all loadings with an absolute value > some cut point are represented as an edge (path).
fa.diagram uses the various diagram functions to draw the diagram. fa.graph generates dot
code for external plotting. fa.rgraph uses the Rgraphviz package (if available) to draw the graph.
het.diagram will draw "heterarchy" diagrams of factor/scale solutions at different levels.
Usage
fa.diagram(fa.results,Phi=NULL,fe.results=NULL,sort=TRUE,labels=NULL,cut=.3,
simple=TRUE, errors=FALSE,g=FALSE,digits=1,e.size=.05,rsize=.15,side=2,
main,cex=NULL,marg=c(.5,.5,1,.5),adj=1, ...)
het.diagram(r,levels,cut=.3,digits=2,both=TRUE,
main="Heterarchy diagram",l.cex,gap.size,...)
fa.graph(fa.results,out.file=NULL,labels=NULL,cut=.3,simple=TRUE,
size=c(8,6), node.font=c("Helvetica", 14),
edge.font=c("Helvetica", 10), rank.direction=c("RL","TB","LR","BT"),
digits=1,main="Factor Analysis", ...)
fa.rgraph(fa.results,out.file=NULL,labels=NULL,cut=.3,simple=TRUE,
size=c(8,6), node.font=c("Helvetica", 14),
edge.font=c("Helvetica", 10), rank.direction=c("RL","TB","LR","BT"),
digits=1,main="Factor Analysis",graphviz=TRUE, ...)
Arguments
fa.results The output of factor analysis, principal components analysis, or ICLUST analy-
sis. May also be a factor loading matrix from anywhere.
Phi Normally not specified (it is is found in the FA, pc, or ICLUST, solution), this
may be given if the input is a loadings matrix.
fe.results the results of a factor extension analysis (if any)
out.file If it exists, a dot representation of the graph will be stored here (fa.graph)
labels Variable labels
120 fa.diagram
cut Loadings with abs(loading) > cut will be shown
simple Only the biggest loading per item is shown
gDoes the factor matrix reflect a g (first) factor. If so, then draw this to the left of
the variables, with the remaining factors to the right of the variables. It is useful
to turn off the simple parameter in this case.
rA correlation matrix for the het.diagram function
levels A list of the elements in each level
both Should arrows have double heads (in het.diagram)
size graph size
sort sort the factor loadings before showing the diagram
errors include error estimates (as arrows)
e.size size of ellipses
rsize size of rectangles
side on which side should error arrows go?
cex modify font size
l.cex modify the font size in arrows, defaults to cex
gap.size The gap in the arrow for the label. Can be adjusted to compensate for variations
in cex or l.cex
marg sets the margins to be wider than normal, returns them to the normal size upon
exit
adj how many different positions (1-3) should be used for the numeric labels. Useful
if they overlap each other.
node.font what font should be used for nodes in fa.graph
edge.font what font should be used for edges in fa.graph
rank.direction parameter passed to Rgraphviz– which way to draw the graph
digits Number of digits to show as an edgelable
main Graphic title, defaults to "factor analyis" or "factor analysis and extension"
graphviz Should we try to use Rgraphviz for output?
... other parameters
Details
Path diagram representations have become standard in confirmatory factor analysis, but are not
yet common in exploratory factor analysis. Representing factor structures graphically helps some
people understand the structure.
fa.diagram does not use Rgraphviz and is the preferred function. fa.graph generates dot code to be
used by an external graphics program. It does not have all the bells and whistles of fa.diagram, but
these may be done in the external editor.
Hierarchical (bifactor) models may be drawn by specifying the g parameter as TRUE. This allows
for an graphical displays of various factor transformations with a bifactor structure (e.g., bifactor
and biquartimin. See omega for an alternative way to find these structures.
fa.diagram 121
The het.diagram function will show the case of a hetarchical structure at multiple levels. It can
also be used to show the patterns of correlations between sets of scales (e.g., EPI, NEO, BFI). The
example is for showing the relationship between 3 sets of 4 variables from the Thurstone data set.
The parameters l.cex and gap.size are used to adjust the font size of the labels and the gap in the
lines.
In fa.rgraph although a nice graph is drawn for the orthogonal factor case, the oblique factor drawing
is acceptable, but is better if cleaned up outside of R or done using fa.diagram.
The normal input is taken from the output of either fa or ICLUST. This latter case displays the
ICLUST results in terms of the cluster loadings, not in terms of the cluster structure. Actually an
interesting option.
It is also possible to just give a factor loading matrix as input. In this case, supplying a Phi matrix
of factor correlations is also possible.
It is possible, using fa.graph, to export dot code for an omega solution. fa.graph should be applied
to the schmid$sl object with labels specified as the rownames of schmid$sl. The results will need
editing to make fully compatible with dot language plotting.
To specify the model for a structural equation confirmatory analysis of the results, use structure.diagram
instead.
Value
fa.diagram: A path diagram is drawn without using Rgraphviz. This is probably the more useful
function.
fa.rgraph: A graph is drawn using rgraphviz. If an output file is specified, the graph instructions are
also saved in the dot language.
fa.graph: the graph instructions are saved in the dot language.
Note
fa.rgraph requires Rgraphviz. Because there are occasional difficulties installing Rgraphviz from
Bioconductor in that some libraries are misplaced and need to be relinked, it is probably better to
use fa.diagram or fa.graph.
Author(s)
William Revelle
See Also
omega.graph,ICLUST.graph,structure.diagram to convert the factor diagram to sem modeling
code.
Examples
test.simple <- fa(item.sim(16),2,rotate="oblimin")
#if(require(Rgraphviz)) {fa.graph(test.simple) }
fa.diagram(test.simple)
f3 <- fa(Thurstone,3,rotate="cluster")
122 fa.extension
fa.diagram(f3,cut=.4,digits=2)
f3l <- f3$loadings
fa.diagram(f3l,main="input from a matrix")
Phi <- f3$Phi
fa.diagram(f3l,Phi=Phi,main="Input from a matrix")
fa.diagram(ICLUST(Thurstone,2,title="Two cluster solution of Thurstone"),main="Input from ICLUST")
het.diagram(Thurstone,levels=list(1:4,5:8,3:7))
fa.extension Apply Dwyer’s factor extension to find factor loadings for extended
variables
Description
Dwyer (1937) introduced a method for finding factor loadings for variables not included in the
original analysis. This is basically finding the unattenuated correlation of the extension variables
with the factor scores. An alternative, which does not correct for factor reliability was proposed by
Gorsuch (1997). Both options are an application of exploratory factor analysis with extensions to
new variables.
Usage
fa.extension(Roe,fo,correct=TRUE)
fa.extend(r,nfactors=1,ov=NULL,ev=NULL,n.obs = NA, np.obs=NULL,
correct=TRUE,rotate="oblimin",SMC=TRUE, warnings=TRUE,
fm="minres",alpha=.1,omega=FALSE, ...)
Arguments
Roe The correlations of the original variables with the extended variables
fo The output from the fa or omega functions applied to the original variables.
correct correct=TRUE produces Dwyer’s solution, correct=FALSE produces Gorsuch’s
solution
rA correlation or data matrix with all of the variables to be analyzed by fa.extend
ov The original variables to factor
ev The extension variables
nfactors Number of factors to extract, default is 1
n.obs Number of observations used to find the correlation matrix if using a correlation
matrix. Used for finding the goodness of fit statistics. Must be specified if using
a correlaton matrix and finding confidence intervals.
np.obs Pairwise number of observations. Required if using fm="minchi", suggested in
other cases to estimate the empirical goodness of fit.
fa.extension 123
rotate "none", "varimax", "quartimax", "bentlerT", "geominT" and "bifactor" are or-
thogonal rotations. "promax", "oblimin", "simplimax", "bentlerQ, "geominQ"
and "biquartimin" and "cluster" are possible rotations or transformations of the
solution. The default is to do a oblimin transformation, although versions prior
to 2009 defaulted to varimax.
SMC Use squared multiple correlations (SMC=TRUE) or use 1 as initial communality
estimate. Try using 1 if imaginary eigen values are reported. If SMC is a vector
of length the number of variables, then these values are used as starting values
in the case of fm=’pa’.
warnings warnings=TRUE => warn if number of factors is too many
fm factoring method fm="minres" will do a minimum residual (OLS), fm="wls"
will do a weighted least squares (WLS) solution, fm="gls" does a generalized
weighted least squares (GLS), fm="pa" will do the principal factor solution,
fm="ml" will do a maximum likelihood factor analysis. fm="minchi" will min-
imize the sample size weighted chi square when treating pairwise correlations
with different number of subjects per pair.
alpha alpha level for the confidence intervals for RMSEA
omega Do the extension analysis for an omega type analysis
... additional parameters, specifically, keys may be passed if using the target rota-
tion, or delta if using geominQ, or whether to normalize if using Varimax
Details
It is sometimes the case that factors are derived from a set of variables (the Fo factor loadings)
and we want to see what the loadings of an extended set of variables (Fe) would be. Given the
original correlation matrix Ro and the correlation of these original variables with the extension
variables of Roe, it is a straight forward calculation to find the loadings Fe of the extended variables
on the original factors. This technique was developed by Dwyer (1937) for the case of adding
new variables to a factor analysis without doing all the work over again. But, as discussed by
Horn (1973) factor extension is also appropriate when one does not want to include the extension
variables in the original factor analysis, but does want to see what the loadings would be anyway.
This could be done by estimating the factor scores and then finding the covariances of the extension
variables with the factor scores. But if the original data are not available, but just the covariance or
correlation matrix is, then the use of fa.extension is most appropriate.
The factor analysis results from either fa or omega functions applied to the original correlation
matrix is extended to the extended variables given the correlations (Roe) of the extended variables
with the original variables.
fa.extension assumes that the original factor solution was found by the fa function.
For a very nice discussion of the relationship between factor scores, correlation matrices, and the
factor loadings in a factor extension, see Horn (1973).
The fa.extend function may be thought of as a "seeded" factor analysis. That is, the variables in
the original set are factored, this solution is then extended to the extension set, and the resulting
output is presented as if both the original and extended variables were factored together. This may
also be done for an omega analysis.
The example of codefa.extend compares the extended solution to a direct solution of all of the
variables using factor.congruence.
124 fa.extension
Value
Factor Loadings of the exended variables on the original factors
Author(s)
William Revelle
References
Paul S. Dwyer (1937) The determination of the factor loadings of a given test from the known factor
loadings of other tests. Psychometrika, 3, 173-178
Gorsuch, Richard L. (1997) New procedure for extension analysis in exploratory factor analysis,
Educational and Psychological Measurement, 57, 725-740
Horn, John L. (1973) On extension analysis and its relation to correlations between variables and
factor scores. Multivariate Behavioral Research, 8, (4), 477-489.
See Also
See Also as fa,principal,Dwyer
Examples
#The Dwyer Example
Ro <- Dwyer[1:7,1:7]
Roe <- Dwyer[1:7,8]
fo <- fa(Ro,2,rotate="none")
fe <- fa.extension(Roe,fo)
#an example from simulated data
set.seed(42)
d <- sim.item(12) #two orthogonal factors
R <- cor(d)
Ro <- R[c(1,2,4,5,7,8,10,11),c(1,2,4,5,7,8,10,11)]
Roe <- R[c(1,2,4,5,7,8,10,11),c(3,6,9,12)]
fo <- fa(Ro,2)
fe <- fa.extension(Roe,fo)
fa.diagram(fo,fe=fe)
#create two correlated factors
fx <- matrix(c(.9,.8,.7,.85,.75,.65,rep(0,12),.9,.8,.7,.85,.75,.65),ncol=2)
Phi <- matrix(c(1,.6,.6,1),2)
sim.data <- sim.structure(fx,Phi,n=1000,raw=TRUE)
R <- cor(sim.data$observed)
Ro <- R[c(1,2,4,5,7,8,10,11),c(1,2,4,5,7,8,10,11)]
Roe <- R[c(1,2,4,5,7,8,10,11),c(3,6,9,12)]
fo <- fa(Ro,2)
fe <- fa.extension(Roe,fo)
fa.diagram(fo,fe=fe)
#now show how fa.extend works with the same data set
#note that we have to make sure that the variables are in the order to do the factor congruence
fa.multi 125
fe2 <- fa.extend(R,2,ov=c(1,2,4,5,7,8,10,11),ev=c(3,6,9,12),n.obs=1000)
fa.diagram(fe2,main="factor analysis with extension variables")
fa2 <- fa(sim.data$observed[,c(1,2,4,5,7,8,10,11,3,6,9,12)],2)
factor.congruence(fe2,fa2)
summary(fe2)
#an example of extending an omega analysis
fload <- matrix(c(c(c(.9,.8,.7,.6),rep(0,20)),c(c(.9,.8,.7,.6),rep(0,20)),c(c(.9,.8,.7,.6),
rep(0,20)),c(c(c(.9,.8,.7,.6),rep(0,20)),c(.9,.8,.7,.6))),ncol=5)
gload <- matrix(rep(.7,5))
five.factor <- sim.hierarchical(gload,fload,500,TRUE) #create sample data set
ss <- c(1,2,3,5,6,7,9,10,11,13,14,15,17,18,19)
Ro <- cor(five.factor$observed[,ss])
Re <- cor(five.factor$observed[,ss],five.factor$observed[,-ss])
om5 <-omega(Ro,5) #the omega analysis
fa.extension(Re,om5) #the extension analysis
fa.multi Multi level (hierarchical) factor analysis
Description
Some factor analytic solutions produce correlated factors which may in turn be factored. If the
solution has one higher order, the omega function is most appropriate. But, in the case of multi
higher order factors, then the faMulti function will do a lower level factoring and then factor the
resulting correlation matrix. Multi level factor diagrams are also shown.
Usage
fa.multi(r, nfactors = 3, nfact2 = 1, n.obs = NA, n.iter = 1, rotate = "oblimin",
scores = "regression", residuals = FALSE, SMC = TRUE, covar = FALSE, missing =
FALSE,impute = "median", min.err = 0.001, max.iter = 50, symmetric = TRUE, warnings
=TRUE, fm = "minres", alpha = 0.1, p = 0.05, oblique.scores = FALSE, np.obs = NULL,
use ="pairwise", cor = "cor", ...)
fa.multi.diagram(multi.results,sort=TRUE,labels=NULL,flabels=NULL,cut=.2,gcut=.2,
simple=TRUE,errors=FALSE,
digits=1,e.size=.1,rsize=.15,side=3,main=NULL,cex=NULL,color.lines=TRUE
,marg=c(.5,.5,1.5,.5),adj=2, ...)
Arguments
The arguments match those of the fa function.
A correlation matrix or raw data matrix
126 fa.multi
rnfactors The desired number of factors for the lower level
nfact2 The desired number of factors for the higher level
n.obs Number of observations used to find the correlation matrix if using a correlation
matrix. Used for finding the goodness of fit statistics. Must be specified if using
a correlaton matrix and finding confidence intervals.
np.obs The pairwise number of observations. Used if using a correlation matrix and
asking for a minchi solution.
rotate "none", "varimax", "quartimax", "bentlerT", "equamax", "varimin", "geominT"
and "bifactor" are orthogonal rotations. "promax", "oblimin", "simplimax",
"bentlerQ, "geominQ" and "biquartimin" and "cluster" are possible oblique trans-
formations of the solution. The default is to do a oblimin transformation, al-
though versions prior to 2009 defaulted to varimax.
n.iter Number of bootstrap interations to do in fa or fa.poly
residuals Should the residual matrix be shown
scores the default="regression" finds factor scores using regression. Alternatives for
estimating factor scores include simple regression ("Thurstone"), correlaton pre-
serving ("tenBerge") as well as "Anderson" and "Bartlett" using the appropriate
algorithms (see factor.scores). Although scores="tenBerge" is probably pre-
ferred for most solutions, it will lead to problems with some improper corre-
lation matrices.
SMC Use squared multiple correlations (SMC=TRUE) or use 1 as initial communality
estimate. Try using 1 if imaginary eigen values are reported. If SMC is a vector
of length the number of variables, then these values are used as starting values
in the case of fm=’pa’.
covar if covar is TRUE, factor the covariance matrix, otherwise factor the correlation
matrix
missing if scores are TRUE, and missing=TRUE, then impute missing values using either
the median or the mean
impute "median" or "mean" values are used to replace missing values
min.err Iterate until the change in communalities is less than min.err
max.iter Maximum number of iterations for convergence
symmetric symmetric=TRUE forces symmetry by just looking at the lower off diagonal
values
warnings warnings=TRUE => warn if number of factors is too many
fm factoring method fm="minres" will do a minimum residual (OLS), fm="wls"
will do a weighted least squares (WLS) solution, fm="gls" does a generalized
weighted least squares (GLS), fm="pa" will do the principal factor solution,
fm="ml" will do a maximum likelihood factor analysis. fm="minchi" will min-
imize the sample size weighted chi square when treating pairwise correlations
with different number of subjects per pair.
alpha alpha level for the confidence intervals for RMSEA
pif doing iterations to find confidence intervals, what probability values should
be found for the confidence intervals
fa.multi 127
oblique.scores When factor scores are found, should they be based on the structure matrix (de-
fault) or the pattern matrix (oblique.scores=TRUE).
use How to treat missing data, use="pairwise" is the default". See cor for other
options.
cor How to find the correlations: "cor" is Pearson", "cov" is covariance, "tet" is
tetrachoric, "poly" is polychoric, "mixed" uses mixed cor for a mixture of tetra-
chorics, polychorics, Pearsons, biserials, and polyserials, Yuleb is Yulebonett,
Yuleq and YuleY are the obvious Yule coefficients as appropriate
multi.results The results from fa.multi
labels variable labels
flabels Labels for the factors (not counting g)
size size of graphics window
digits Precision of labels
cex control font size
color.lines Use black for positive, red for negative
marg The margins for the figure are set to be wider than normal by default
adj Adjust the location of the factor loadings to vary as factor mod 4 + 1
main main figure caption
... additional parameters, specifically, keys may be passed if using the target rota-
tion, or delta if using geominQ, or whether to normalize if using Varimax. In
addition, for fa.multi.diagram, other options to pass into the graphics packages
e.size the size to draw the ellipses for the factors. This is scaled by the number of
variables.
cut Minimum path coefficient to draw
gcut Minimum general factor path to draw
simple draw just one path per item
sort sort the solution before making the diagram
side on which side should errors be drawn?
errors show the error estimates
rsize size of the rectangles
Details
See fa and omega for a discussion of factor analysis and of the case of one higher order factor.
Value
f1 The standard output from a factor analysis from fa for the raw variables
f2 The standard output from a factor analysis from fa for the correlation matrix of
the level 1 solution.
128 fa.parallel
Note
This is clearly an early implementation (Feb 14 2016) which might be improved.
Author(s)
William Revelle
References
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
See Also
fa,omega
Examples
f31 <- fa.multi(Thurstone,3,1) #compare with \code{\link{omega}}
f31
fa.multi.diagram(f31)
fa.parallel Scree plots of data or correlation matrix compared to random “paral-
lel" matrices
Description
One way to determine the number of factors or components in a data matrix or a correlation matrix
is to examine the “scree" plot of the successive eigenvalues. Sharp breaks in the plot suggest the
appropriate number of components or factors to extract. “Parallel" analyis is an alternative tech-
nique that compares the scree of factors of the observed data with that of a random data matrix of
the same size as the original. fa.parallel.poly does this for tetrachoric or polychoric analyses.
Usage
fa.parallel(x,n.obs=NULL,fm="minres",fa="both",main="Parallel Analysis Scree Plots",
n.iter=20,error.bars=FALSE,se.bars=TRUE,SMC=FALSE,ylabel=NULL,show.legend=TRUE,
sim=TRUE,quant=.95,cor="cor",use="pairwise")
fa.parallel.poly(x ,n.iter=10,SMC=TRUE, fm = "minres",correct=TRUE,sim=FALSE,
fa="both",global=TRUE)
## S3 method for class 'poly.parallel'
plot(x,show.legend=TRUE,fa="both",...)
fa.parallel 129
Arguments
xA data.frame or data matrix of scores. If the matrix is square, it is assumed to
be a correlation matrix. Otherwise, correlations (with pairwise deletion) will be
found
n.obs n.obs=0 implies a data matrix/data.frame. Otherwise, how many cases were
used to find the correlations.
fm What factor method to use. (minres, ml, uls, wls, gls, pa) See fa for details.
fa show the eigen values for a principal components (fa="pc") or a principal axis
factor analysis (fa="fa") or both principal components and principal factors (fa="both")
main a title for the analysis
n.iter Number of simulated analyses to perform
use How to treat missing data, use="pairwise" is the default". See cor for other
options.
cor How to find the correlations: "cor" is Pearson", "cov" is covariance, "tet" is
tetrachoric, "poly" is polychoric, "mixed" uses mixed cor for a mixture of tetra-
chorics, polychorics, Pearsons, biserials, and polyserials, Yuleb is Yulebonett,
Yuleq and YuleY are the obvious Yule coefficients as appropriate. This matches
the call to fa
correct For tetrachoric correlations, should a correction for continuity be applied. (See
tetrachoric)
sim For continuous data, the default is to resample as well as to generate random
normal data. If sim=FALSE, then just show the resampled results. These two
results are very similar. This does not make sense in the case of correlation
matrix, in which case resampling is impossible. In the case of polychoric or
tetrachoric data, in addition to randomizing the real data, should we compare
the solution to random simulated data. This will double the processing time, but
will yiedl basically show the same result.
error.bars Should error.bars be plotted (default = FALSE)
se.bars Should the error bars be standard errors (the default) or 1 standard deviation
(se.bars=FALSE). With many iterations, the standard errors are very small and
some prefer to see the broader range.
SMC SMC=TRUE finds eigen values after estimating communalities by using SMCs.
smc = FALSE finds eigen values after estimating communalities with the first
factor.
ylabel Label for the y axis – defaults to “eigen values of factors and components", can
be made empty to show many graphs
show.legend the default is to have a legend. For multiple panel graphs, it is better to not show
the legend
quant if nothing is specified, the empirical eigen values are compared to the mean of
the resampled or simulated eigen values. If a value (e.g., quant=.95) is specified,
then the eigen values are compared against the matching quantile of the simu-
lated data. Clearly the larger the value of quant, the few factors/components will
be identified.
130 fa.parallel
global If doing polychoric analyses (fa.parallel.poly) and the number of alternatives
differ across items, it is necessary to turn off the global option
... additional plotting parameters, for plot.poly.parallel
Details
Cattell’s “scree" test is one of most simple tests for the number of factors problem. Horn’s (1965)
“parallel" analysis is an equally compelling procedure. Other procedures for determining the most
optimal number of factors include finding the Very Simple Structure (VSS) criterion (VSS) and
Velicer’s MAP procedure (included in VSS). Both the VSS and the MAP criteria are included in
the link{nfactors} function which also reports the mean item complexity and the BIC for each
of multiple solutions. fa.parallel plots the eigen values for a principal components and the factor
solution (minres by default) and does the same for random matrices of the same size as the original
data matrix. For raw data, the random matrices are 1) a matrix of univariate normal data and 2)
random samples (randomized across rows) of the original data.
fa.parallel.poly will do parallel analysis for polychoric and tetrachoric factors. If the data
are dichotomous, fa.parallel.poly will find tetrachoric correlations for the real and simulated
data, otherwise, if the number of categories is less than 10, it will find polychoric correlations.
Note that fa.parallel.poly is slower than fa.parallel because of the complexity of calculating the
tetrachoric/polychoric correlations. The functionality of fa.parallel.poly is now included in
fa.parallel with cor=poly option (etc.) option.
fa.parallel now will do tetrachorics or polychorics directly if the cor option is set to "tet" or
"poly". As with fa.parallel.poly this will take longer.
The means of (ntrials) random solutions are shown. Error bars are usually very small and are
suppressed by default but can be shown if requested. If the sim option is set to TRUE (default),
then parallel analyses are done on resampled data as well as random normal data. In the interests of
speed, the parallel analyses are done just on resampled data if sim=FALSE. Both procedures tend
to agree.
As of version 1.5.4, I added the ability to specify the quantile of the simulated/resampled data, and
to plot standard deviations or standard errors.
Alternative ways to estimate the number of factors problem are discussed in the Very Simple Struc-
ture (Revelle and Rocklin, 1979) documentation (VSS) and include Wayne Velicer’s MAP algorithm
(Veicer, 1976).
Parallel analysis for factors is actually harder than it seems, for the question is what are the appro-
priate communalities to use. If communalities are estimated by the Squared Multiple Correlation
(SMC) smc, then the eigen values of the original data will reflect major as well as minor factors (see
sim.minor to simulate such data). Random data will not, of course, have any structure and thus the
number of factors will tend to be biased upwards by the presence of the minor factors.
By default, fa.parallel estimates the communalities based upon a one factor minres solution. Al-
though this will underestimate the communalities, it does seem to lead to better solutions on simu-
lated or real (e.g., the bfi or Harman74) data sets.
For comparability with other algorithms (e.g, the paran function in the paran package), setting
smc=TRUE will use smcs as estimates of communalities. This will tend towards identifying more
factors than the default option.
Printing the results will show the eigen values of the original data that are greater than simulated
values.
fa.parallel 131
A sad observation about parallel analysis is that it is sensitive to sample size. That is, for large data
sets, the eigen values of random data are very close to 1. This will lead to different estimates of the
number of factors as a function of sample size. Consider factor structure of the bfi data set (the first
25 items are meant to represent a five factor model). For samples of 200 or less, parallel analysis
suggests 5 factors, but for 1000 or more, six factors and components are indicated. This is not due
to an instability of the eigen values of the real data, but rather the closer approximation to 1 of the
random data as n increases.
When simulating dichotomous data in fa.parallel.poly, the simulated data have the same difficulties
as the original data. This functionally means that the simulated and the resampled results will
be very similar. Note that fa.parallel.poly has functionally been replaced with fa.parallel with the
cor="poly" option.
As with many psych functions, fa.parallel has been changed to allow for multicore processing.
For running a large number of iterations, it is obviously faster to increase the number of cores
to the maximum possible (using the options("mc.cores=n) command where n is determined from
detectCores().
Value
A plot of the eigen values for the original data, ntrials of resampling of the original data, and of a
equivalent size matrix of random normal deviates. If the data are a correlation matrix, specify the
number of observations.
Also returned (invisibly) are:
fa.values The eigen values of the factor model for the real data.
fa.sim The descriptive statistics of the simulated factor models.
pc.values The eigen values of a principal components of the real data.
pc.sim The descriptive statistics of the simulated principal components analysis.
nfact Number of factors with eigen values > eigen values of random data
ncomp Number of components with eigen values > eigen values of random data
values The simulated values for all simulated trials
Note
Although by default the test is applied to the mean eigen values, this can be modified by setting the
quant parameter to any particular quantile. The actual simulated data are also returned (invisibly)
in the value object. Thus, it is possible to do descriptive statistics on those to choose a preferred
comparison. See the last example (not run)
Author(s)
William Revelle
References
Floyd, Frank J. and Widaman, Keith. F (1995) Factor analysis in the development and refinement
of clinical assessment instruments. Psychological Assessment, 7(3):286-299, 1995.
132 fa.sort
Horn, John (1965) A rationale and test for the number of factors in factor analysis. Psychometrika,
30, 179-185.
Humphreys, Lloyd G. and Montanelli, Richard G. (1975), An investigation of the parallel analysis
criterion for determining the number of common factors. Multivariate Behavioral Research, 10,
193-205.
Revelle, William and Rocklin, Tom (1979) Very simple structure - alternative procedure for estimat-
ing the optimal number of interpretable factors. Multivariate Behavioral Research, 14(4):403-414.
Velicer, Wayne. (1976) Determining the number of components from the matrix of partial correla-
tions. Psychometrika, 41(3):321-327, 1976.
See Also
fa,nfactors,VSS,VSS.plot,VSS.parallel,sim.minor
Examples
#test.data <- Harman74.cor$cov #The 24 variable Holzinger - Harman problem
#fa.parallel(test.data,n.obs=145)
fa.parallel(Thurstone,n.obs=213) #the 9 variable Thurstone problem
#set.seed(123)
#minor <- sim.minor(24,4,400) #4 large and 12 minor factors
#ffa.parallel(minor$observed) #shows 5 factors and 4 components -- compare with
#fa.parallel(minor$observed,SMC=FALSE) #which shows 6 and 4 components factors
#a demonstration of parallel analysis of a dichotomous variable
#fp <- fa.parallel(ability) #use the default Pearson correlation
#fpt <- fa.parallel(ability,cor="tet") #do a tetrachoric correlation
#fpt <- fa.parallel(ability,cor="tet",quant=.95) #do a tetrachoric correlation and
#use the 95th percentile of the simulated results
#apply(fp$values,2,function(x) quantile(x,.95)) #look at the 95th percentile of values
#apply(fpt$values,2,function(x) quantile(x,.95)) #look at the 95th percentile of values
#describe(fpt$values) #look at all the statistics of the simulated values
fa.sort Sort factor analysis or principal components analysis loadings
Description
Although the print.psych function will sort factor analysis loadings, sometimes it is useful to do this
outside of the print function. fa.sort takes the output from the fa or principal functions and sorts
the loadings for each factor. Items are located in terms of their greatest loading. The new order is
returned as an element in the fa list.
Usage
fa.sort(fa.results,polar=FALSE)
fa.organize(fa.results,o=NULL,i=NULL,cn=NULL)
factor.congruence 133
Arguments
fa.results The output from a factor analysis or principal components analysis using fa or
principal.
polar Sort by polar coordinates of first two factors (FALSE)
oThe order in which to order the factors
iThe order in which to order the items
cn new factor names
Details
The fa.results$loadings are replaced with sorted loadings.
fa.organize takes a factor analysis or components output and reorganizes the factors in the o order.
Items are organized in the i order. This is useful when comparing alternative factor solutions.
Value
A sorted factor analysis, principal components analysis, or omega loadings matrix.
These sorted values are used internally by the various diagram functions.
The values returned are the same as fa, except in sorted order. In addition, the order is returned as
an additional element in the fa list.
Author(s)
William Revelle
See Also
See Also as fa,print.psych,fa.diagram,
Examples
test.simple <- fa(sim.item(16),2)
fa.sort(test.simple)
fa.organize(test.simple,c(2,1)) #the factors but not the items have been rearranged
factor.congruence Coefficient of factor congruence
Description
Given two sets of factor loadings, report their degree of congruence (vector cosine). Although first
reported by Burt (1948), this is frequently known as the Tucker index of factor congruence.
134 factor.congruence
Usage
factor.congruence(x, y=NULL,digits=2,use=NULL)
fa.congruence(x, y=NULL,digits=2,use=NULL)
Arguments
xA matrix of factor loadings or a list of matrices of factor loadings
yA second matrix of factor loadings (if x is a list, then y may be empty)
digits Round off to digits
use If NULL, then no loading matrices may contain missing values. If use="complete"
then variables with any missing loadings are dropped (with a warning)
Details
Find the coefficient of factor congruence between two sets of factor loadings.
Factor congruences are the cosines of pairs of vectors defined by the loadings matrix and based at
the origin. Thus, for loadings that differ only by a scaler (e.g. the size of the eigen value), the factor
congruences will be 1.
For factor loading vectors of F1 and F2 the measure of factor congruence, phi, is
φ=PF1F2
pP(F2
1)P(F2
2).
It is an interesting exercise to compare factor congruences with the correlations of factor loadings.
Factor congruences are based upon the raw cross products, while correlations are based upon cen-
tered cross products. That is, correlations of factor loadings are cosines of the vectors based at the
mean loading for each factor.
φ=P(F1−a)(F2−b)
pP((F1−a)2)P((F2−b)2).
.
For congruence coefficients, a = b= 0. For correlations a=mean F1, b= mean F2.
Input may either be matrices or factor analysis or principal components analyis output (which in-
cludes a loadings object), or a mixture of the two.
To compare more than two solutions, x may be a list of matrices, all of which will be compared.
Normally, all factor loading matrices should be complete (have no missing loadings). In the case
where some loadings are missing, if the use option is specified, then variables with missing loadings
are dropped.
Value
A matrix of factor congruences.
Author(s)
<revelle@northwestern.edu>
http://personality-project.org/revelle.html
factor.congruence 135
References
Burt, Cyril (1948) The factorial study of temperamental traits. British Journal of Statistical Psy-
chology, 1(3) 178-203.
Lorenzo-Seva, U. and ten Berge, J. M. F. (2006). Tucker’s congruence coefficient as a meaningful
index of factor similarity. Methodology: European Journal of Research Methods for the Behavioral
and Social Sciences, 2(2):57-64.
Gorsuch, Richard, (1983) Factor Analysis. Lawrence Erlebaum Associates.
Revelle, W. (In preparation) An Introduction to Psychometric Theory with applications in R (http:
//personality-project.org/r/book/)
See Also
principal,fa
Examples
#factor congruence of factors and components, both rotated
#fa <- fa(Harman74.cor$cov,4)
#pc <- principal(Harman74.cor$cov,4)
#factor.congruence(fa,pc)
# RC1 RC3 RC2 RC4
#MR1 0.98 0.41 0.28 0.32
#MR3 0.35 0.96 0.41 0.31
#MR2 0.23 0.16 0.95 0.28
#MR4 0.28 0.38 0.36 0.98
#factor congruence without rotation
#fa <- fa(Harman74.cor$cov,4,rotate="none")
#pc <- principal(Harman74.cor$cov,4,rotate="none")
#factor.congruence(fa,pc) #just show the beween method congruences
# PC1 PC2 PC3 PC4
#MR1 1.00 -0.04 -0.06 -0.01
#MR2 0.15 0.97 -0.01 -0.15
#MR3 0.31 0.05 0.94 0.11
#MR4 0.07 0.21 -0.12 0.96
#factor.congruence(list(fa,pc)) #this shows the within method congruence as well
# MR1 MR2 MR3 MR4 PC1 PC2 PC3 PC4
#MR1 1.00 0.11 0.25 0.06 1.00 -0.04 -0.06 -0.01
#MR2 0.11 1.00 0.06 0.07 0.15 0.97 -0.01 -0.15
#MR3 0.25 0.06 1.00 0.01 0.31 0.05 0.94 0.11
#MR4 0.06 0.07 0.01 1.00 0.07 0.21 -0.12 0.96
#PC1 1.00 0.15 0.31 0.07 1.00 0.00 0.00 0.00
#PC2 -0.04 0.97 0.05 0.21 0.00 1.00 0.00 0.00
#PC3 -0.06 -0.01 0.94 -0.12 0.00 0.00 1.00 0.00
#PC4 -0.01 -0.15 0.11 0.96 0.00 0.00 0.00 1.00
136 factor.fit
#pa <- fa(Harman74.cor$cov,4,fm="pa")
# factor.congruence(fa,pa)
# PA1 PA3 PA2 PA4
#Factor1 1.00 0.61 0.46 0.55
#Factor2 0.61 1.00 0.50 0.60
#Factor3 0.46 0.50 1.00 0.57
#Factor4 0.56 0.62 0.58 1.00
#compare with
#round(cor(fa$loading,pc$loading),2)
# RC1 RC3 RC2 RC4
#MR1 0.99 -0.18 -0.33 -0.34
#MR3 -0.33 0.96 -0.16 -0.43
#MR2 -0.29 -0.46 0.98 -0.21
#MR4 -0.44 -0.30 -0.22 0.98
factor.fit How well does the factor model fit a correlation matrix. Part of the
VSS package
Description
The basic factor or principal components model is that a correlation or covariance matrix may be
reproduced by the product of a factor loading matrix times its transpose: F’F or P’P. One simple
index of fit is the 1 - sum squared residuals/sum squared original correlations. This fit index is used
by VSS,ICLUST, etc.
Usage
factor.fit(r, f)
Arguments
ra correlation matrix
fA factor matrix of loadings.
Details
There are probably as many fit indices as there are psychometricians. This fit is a plausible estimate
of the amount of reduction in a correlation matrix given a factor model. Note that it is sensitive to
the size of the original correlations. That is, if the residuals are small but the original correlations
are small, that is a bad fit.
Let
R∗=R−F F 0
fit = 1 −P(R∗2)
P(R2)
factor.model 137
.
The sums are taken for the off diagonal elements.
Value
fit
Author(s)
William Revelle
See Also
VSS,ICLUST
Examples
## Not run:
#compare the fit of 4 to 3 factors for the Harman 24 variables
fa4 <- factanal(x,4,covmat=Harman74.cor$cov)
round(factor.fit(Harman74.cor$cov,fa4$loading),2)
#[1] 0.9
fa3 <- factanal(x,3,covmat=Harman74.cor$cov)
round(factor.fit(Harman74.cor$cov,fa3$loading),2)
#[1] 0.88
## End(Not run)
factor.model Find R = F F’ + U2 is the basic factor model
Description
The basic factor or principal components model is that a correlation or covariance matrix may be
reproduced by the product of a factor loading matrix times its transpose. Find this reproduced
matrix. Used by factor.fit,VSS,ICLUST, etc.
Usage
factor.model(f,Phi=NULL,U2=TRUE)
Arguments
fA matrix of loadings.
Phi A matrix of factor correlations
U2 Should the diagonal be model by ff’ (U2 = TRUE) or replaced with 1’s (U2 =
FALSE)
138 factor.residuals
Value
A correlation or covariance matrix.
Author(s)
<revelle@northwestern.edu >
http://personality-project.org/revelle.html
References
Gorsuch, Richard, (1983) Factor Analysis. Lawrence Erlebaum Associates.
Revelle, W. In preparation) An Introduction to Psychometric Theory with applications in R (http:
//personality-project.org/r/book/)
See Also
ICLUST.graph,ICLUST.cluster,cluster.fit ,VSS,omega
Examples
f2 <- matrix(c(.9,.8,.7,rep(0,6),.6,.7,.8),ncol=2)
mod <- factor.model(f2)
round(mod,2)
factor.residuals R* = R- F F’
Description
The basic factor or principal components model is that a correlation or covariance matrix may be
reproduced by the product of a factor loading matrix times its transpose. Find the residuals of the
original minus the reproduced matrix. Used by factor.fit,VSS,ICLUST, etc.
Usage
factor.residuals(r, f)
Arguments
rA correlation matrix
fA factor model matrix or a list of class loadings
factor.rotate 139
Details
The basic factor equation is nRn≈nFkkF0
n+U2. Residuals are just R* = R - F’F. The residuals
should be (but in practice probably rarely are) examined to understand the adequacy of the factor
analysis. When doing Factor analysis or Principal Components analysis, one usually continues to
extract factors/components until the residuals do not differ from those expected from a random
matrix.
Value
rstar is the residual correlation matrix.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
See Also
fa,principal,VSS,ICLUST
Examples
fa2 <- fa(Harman74.cor$cov,2,rotate=TRUE)
fa2resid <- factor.residuals(Harman74.cor$cov,fa2)
fa2resid[1:4,1:4] #residuals with two factors extracted
fa4 <- fa(Harman74.cor$cov,4,rotate=TRUE)
fa4resid <- factor.residuals(Harman74.cor$cov,fa4)
fa4resid[1:4,1:4] #residuals with 4 factors extracted
factor.rotate “Hand" rotate a factor loading matrix
Description
Given a factor or components matrix, it is sometimes useful to do arbitrary rotations of particular
pairs of variables. This supplements the much more powerful rotation package GPArotation and is
meant for specific requirements to do unusual rotations.
Usage
factor.rotate(f, angle, col1=1, col2=2,plot=FALSE,...)
140 factor.rotate
Arguments
foriginal loading matrix or a data frame (can be output from a factor analysis
function
angle angle (in degrees!) to rotate
col1 column in factor matrix defining the first variable
col2 column in factor matrix defining the second variable
plot plot the original (unrotated) and rotated factors
... parameters to pass to fa.plot
Details
Partly meant as a demonstration of how rotation works, factor.rotate is useful for those cases that
require specific rotations that are not available in more advanced packages such as GPArotation. If
the plot option is set to TRUE, then the original axes are shown as dashed lines.
The rotation is in degrees counter clockwise.
Value
the resulting rotated matrix of loadings.
Note
For a complete rotation package, see GPArotation
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu >
References
http://personality-project.org/r/book
Examples
#using the Harman 24 mental tests, rotate the 2nd and 3rd factors 45 degrees
f4<- fa(Harman74.cor$cov,4,rotate="TRUE")
f4r45 <- factor.rotate(f4,45,2,3)
f4r90 <- factor.rotate(f4r45,45,2,3)
print(factor.congruence(f4,f4r45),digits=3) #poor congruence with original
print(factor.congruence(f4,f4r90),digits=3) #factor 2 and 3 have been exchanged and 3 flipped
#a graphic example
data(Harman23.cor)
f2 <- fa(Harman23.cor$cov,2,rotate="none")
op <- par(mfrow=c(1,2))
cluster.plot(f2,xlim=c(-1,1),ylim=c(-1,1),title="Unrotated ")
f2r <- factor.rotate(f2,-33,plot=TRUE,xlim=c(-1,1),ylim=c(-1,1),title="rotated -33 degrees")
factor.scores 141
op <- par(mfrow=c(1,1))
factor.scores Various ways to estimate factor scores for the factor analysis model
Description
A fundamental problem with factor analysis is that although the model is defined at the structural
level, it is indeterminate at the data level. This problem of factor indeterminancy leads to alternative
ways of estimating factor scores, none of which is ideal. Following Grice (2001) four different
methods are available here.
Usage
factor.scores(x, f, Phi = NULL, method = c("Thurstone", "tenBerge", "Anderson",
"Bartlett", "Harman","components"),rho=NULL)
Arguments
xEither a matrix of data if scores are to be found, or a correlation matrix if just
the factor weights are to be found.
fThe output from the fa function, or a factor loading matrix.
Phi If a pattern matrix is provided, then what were the factor intercorrelations. Does
not need to be specified if f is the output from the fa function.
method Which of four factor score estimation procedures should be used. Defaults to
"Thurstone" or regression based weights. See details below for the other four
methods.
rho If x is a set of data and rho is specified, then find scores based upon the fa results
and the correlations reported in rho. Used when scoring fa.poly results.
Details
Although the factor analysis model is defined at the structural level, it is undefined at the data level.
This is a well known but little discussed problem with factor analysis.
Factor scores represent estimates of common part of the variables and should not be thought of
as identical to the factors themselves. If a factor scores is thought of as a chop stick stuck into
the center of an ice cream cone and factor scores are represented by straws anywhere along the
edge of the cone the problem of factor indeterminacy becomes clear, for depending on the shape
of the cone, two straws can be negatively correlated with each other. (The imagery is taken from
Niels Waller, adapted from Stanley Mulaik). In a very clear discussion of the problem of factor
score indeterminacy, Grice (2001) reviews several alternative ways of estimating factor scores and
considers weighting schemes that will produce uncorrelated factor score estimates as well as the
effect of using course coded (unit weighted) factor weights.
factor.scores uses four different ways of estimate factor scores. In all cases, the factor score esti-
mates are based upon the data matrix, X, times a weighting matrix, W, which weights the observed
variables.
142 factor.scores
• method="Thurstone" finds the regression based weights: W=R−1Fwhere R is the correla-
tion matrix and F is the factor loading matrix.
• method="tenBerge" finds weights such that the correlation between factors for an oblique
solution is preserved. Note that formula 8 in Grice has a typo in the formula for C and should
be: L=FΦ(1/2) C=R(−1/2)L(L0R(−1)L)(−1/2) W=R(−1/2)CΦ(1/2)
• method="Anderson" finds weights such that the factor scores will be uncorrelated: W=
U−2F(F0U−2RU−2F)−1/2where U is the diagonal matrix of uniquenesses. The Anderson
method works for orthogonal factors only, while the tenBerge method works for orthogonal
or oblique solutions.
• method = "Bartlett" finds weights given W=U−2F(F0U−2F)−1
• method="Harman" finds weights based upon socalled "idealized" variables: W=F(t(F)F)−1.
• method="components" uses weights that are just component loadings.
Value
• scores (the factor scores if the raw data is given)
• weights (the factor weights)
Author(s)
William Revelle
References
Grice, James W.,2001, Computing and evaluating factor scores, Psychological Methods, 6,4, 430-
450. (note the typo in equation 8)
ten Berge, Jos M.F., Wim P. Krijnen, Tom Wansbeek and Alexander Shapiro (1999) Some new
results on correlation-preserving factor scores prediction methods. Linear Algebra and its Applica-
tions, 289, 311-318.
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
See Also
fa,factor.stats
Examples
f3 <- fa(Thurstone)
f3$weights #just the scoring weights
f5 <- fa(bfi,5)
round(cor(f5$scores,use="pairwise"),2)
#compare to the f5 solution
factor.stats 143
factor.stats Find various goodness of fit statistics for factor analysis and principal
components
Description
Chi square and other goodness of fit statistics are found based upon the fit of a factor or components
model to a correlation matrix. Although these statistics are normally associated with a maximum
likelihood solution, they can be found for minimal residual (OLS), principal axis, or principal com-
ponent solutions as well. Primarily called from within these functions, factor.stats can be used by
itself. Measures of factorial adequacy and validity follow the paper by Grice, 2001.
Usage
fa.stats(r=NULL,f,phi=NULL,n.obs=NA,np.obs=NULL,alpha=.1,fm=NULL)
factor.stats(r=NULL,f,phi=NULL,n.obs=NA,np.obs=NULL,alpha=.1,fm=NULL)
Arguments
rA correlation matrix or a data frame of raw data
fA factor analysis loadings matrix or the output from a factor or principal com-
ponents analysis. In which case the r matrix need not be specified.
phi A factor intercorrelation matrix if the factor solution was oblique.
n.obs The number of observations for the correlation matrix. If not specified, and a
correlation matrix is used, chi square will not be reported. Not needed if the
input is a data matrix.
np.obs The pairwise number of subjects for each pair in the correlation matrix. This is
used for finding observed chi square.
alpha alpha level of confidence intervals for RMSEA
fm flag if components are being given statistics
Details
Combines the goodness of fit tests used in fa and principal into one function. If the matrix is
singular, will smooth the correlation matrix before finding the fit functions. Now will find the
RMSEA (root mean square error of approximation) and the alpha confidence intervals similar to a
SEM function. Also reports the root mean square residual.
Chi square is found two ways. The first (STATISTIC) applies the goodness of fit test from Maximum
Likelihood objective function (see below). This assumes multivariate normality. The second is the
empirical chi square based upon the observed residual correlation matrix and the observed sample
size for each correlation. This is found by summing the squared residual correlations time the
sample size.
144 factor.stats
Value
fit How well does the factor model reproduce the correlation matrix. (See VSS,
ICLUST, and principal for this fit statistic.
fit.off how well are the off diagonal elements reproduced? This is just 1 - the rela-
tive magnitude of the squared off diagonal residuals to the squared off diagonal
original values.
dof Degrees of Freedom for this model. This is the number of observed correlations
minus the number of independent parameters. Let n=Number of items, nf =
number of factors then
dof =n∗(n−1)/2−n∗nf +nf ∗(nf −1)/2
objective value of the function that is minimized by maximum likelihood procedures. This
is reported for comparison purposes and as a way to estimate chi square good-
ness of fit. The objective function is
f=log(trace((F F 0+U2)−1R)−log(|(F F 0+U2)−1R|)−n.items.
STATISTIC If the number of observations is specified or found, this is a chi square based
upon the objective function, f. Using the formula from factanal(which seems
to be Bartlett’s test) :
χ2= (n.obs −1−(2 ∗p+ 5)/6−(2 ∗factors)/3)) ∗f
Note that this is different from the chi square reported by the sem package which
seems to use χ2= (n.obs −1−(2 ∗p+ 5)/6−(2 ∗factors)/3)) ∗f
PVAL If n.obs > 0, then what is the probability of observing a chisquare this large or
larger?
Phi If oblique rotations (using oblimin from the GPArotation package or promax)
are requested, what is the interfactor correlation.
R2 The multiple R square between the factors and factor score estimates, if they
were to be found. (From Grice, 2001)
r.scores The correlations of the factor score estimates, if they were to be found.
weights The beta weights to find the factor score estimates
valid The validity coffiecient of course coded (unit weighted) factor score estimates
(From Grice, 2001)
score.cor The correlation matrix of course coded (unit weighted) factor score estimates, if
they were to be found, based upon the loadings matrix.
RMSEA The Root Mean Square Error of Approximation and the alpha confidence in-
tervals. Based upon the chi square non-centrality parameter. This is found as
pf/dof −1(/−1)
rms The empirically found square root of the squared residuals. This does not require
sample size to be specified nor does it make assumptions about normality.
crms While the rms uses the number of correlations to find the average, the crms
uses the number of degrees of freedom. Thus, there is a penalty for having too
complex a model.
Author(s)
William Revelle
factor2cluster 145
References
Grice, James W.,2001, Computing and evaluating factor scores, Psychological Methods, 6,4, 430-
450.
See Also
fa with fm="pa" for principal axis factor analysis, fa with fm="minres" for minimum residual
factor analysis (default). factor.pa also does principal axis factor analysis, but is deprecated, as is
factor.minres for minimum residual factor analysis. See principal for principal components.
Examples
v9 <- sim.hierarchical()
f3 <- fa(v9,3)
factor.stats(v9,f3,n.obs=500)
f3o <- fa(v9,3,fm="pa",rotate="Promax")
factor.stats(v9,f3o,n.obs=500)
factor2cluster Extract cluster definitions from factor loadings
Description
Given a factor or principal components loading matrix, assign each item to a cluster corresponding
to the largest (signed) factor loading for that item. Essentially, this is a Very Simple Structure
approach to cluster definition that corresponds to what most people actually do: highlight the largest
loading for each item and ignore the rest.
Usage
factor2cluster(loads, cut = 0)
Arguments
loads either a matrix of loadings, or the result of a factor analysis/principal compo-
nents analyis with a loading component
cut Extract items with absolute loadings > cut
Details
A factor/principal components analysis loading matrix is converted to a cluster (-1,0,1) definition
matrix where each item is assigned to one and only one cluster. This is a fast way to extract
items that will be unit weighted to form cluster composites. Use this function in combination with
cluster.cor to find the corrleations of these composite scores.
146 factor2cluster
A typical use in the SAPA project is to form item composites by clustering or factoring (see ICLUST,
principal), extract the clusters from these results (factor2cluster), and then form the composite
correlation matrix using cluster.cor. The variables in this reduced matrix may then be used in
multiple R procedures using mat.regress.
The input may be a matrix of item loadings, or the output from a factor analysis which includes a
loadings matrix.
Value
a matrix of -1,0,1 cluster definitions for each item.
Author(s)
http://personality-project.org/revelle.html
Maintainer: William Revelle < revelle@northwestern.edu >
References
http://personality-project.org/r/r.vss.html
See Also
cluster.cor,factor2cluster,fa,principal,ICLUST
Examples
## Not run:
f <- factanal(x,4,covmat=Harman74.cor$cov)
factor2cluster(f)
## End(Not run)
# Factor1 Factor2 Factor3 Factor4
#VisualPerception 0 1 0 0
#Cubes 0 1 0 0
#PaperFormBoard 0 1 0 0
#Flags 0 1 0 0
#GeneralInformation 1 0 0 0
#PargraphComprehension 1 0 0 0
#SentenceCompletion 1 0 0 0
#WordClassification 1 0 0 0
#WordMeaning 1 0 0 0
#Addition 0 0 1 0
#Code 0 0 1 0
#CountingDots 0 0 1 0
#StraightCurvedCapitals 0 0 1 0
#WordRecognition 0 0 0 1
#NumberRecognition 0 0 0 1
#FigureRecognition 0 0 0 1
#ObjectNumber 0 0 0 1
#NumberFigure 0 0 0 1
fisherz 147
#FigureWord 0 0 0 1
#Deduction 0 1 0 0
#NumericalPuzzles 0 0 1 0
#ProblemReasoning 0 1 0 0
#SeriesCompletion 0 1 0 0
#ArithmeticProblems 0 0 1 0
fisherz Fisher r to z and z to r and confidence intervals
Description
Convert a correlation to a z score or z to r using the Fisher transformation or find the confidence
intervals for a specified correlation. r2d converts a correlation to an effect size (Cohen’s d) and d2r
converts a d into an r.
Usage
fisherz(rho)
fisherz2r(z)
r.con(rho,n,p=.95,twotailed=TRUE)
r2t(rho,n)
r2d(rho)
d2r(d)
Arguments
rho a Pearson r
zA Fisher z
nSample size for confidence intervals
pConfidence interval
twotailed Treat p as twotailed p
dan effect size (Cohen’s d)
Value
z value corresponding to r (fisherz) \ r corresponding to z (fisherz2r) \ lower and upper p confidence
intervals (r.con) \ t with n-2 df (r2t) r corresponding to effect size d or d corresponding to r.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu >
148 galton
Examples
cors <- seq(-.9,.9,.1)
zs <- fisherz(cors)
rs <- fisherz2r(zs)
round(zs,2)
n <- 30
r <- seq(0,.9,.1)
rc <- matrix(r.con(r,n),ncol=2)
t <- r*sqrt(n-2)/sqrt(1-r^2)
p <- (1-pt(t,n-2))/2
r.rc <- data.frame(r=r,z=fisherz(r),lower=rc[,1],upper=rc[,2],t=t,p=p)
round(r.rc,2)
galton Galton’s Mid parent child height data
Description
Two of the earliest examples of the correlation coefficient were Francis Galton’s data sets on the
relationship between mid parent and child height and the similarity of parent generation peas with
child peas. This is the data set for the Galton height.
Usage
data(galton)
Format
A data frame with 928 observations on the following 2 variables.
parent Mid Parent heights (in inches)
child Child Height
Details
Female heights were adjusted by 1.08 to compensate for sex differences. (This was done in the
original data set)
Source
This is just the galton data set from UsingR, slightly rearranged.
geometric.mean 149
References
Stigler, S. M. (1999). Statistics on the Table: The History of Statistical Concepts and Methods.
Harvard University Press. Galton, F. (1886). Regression towards mediocrity in hereditary stature.
Journal of the Anthropological Institute of Great Britain and Ireland, 15:246-263. Galton, F. (1869).
Hereditary Genius: An Inquiry into its Laws and Consequences. London: Macmillan.
Wachsmuth, A.W., Wilkinson L., Dallal G.E. (2003). Galton’s bend: A previously undiscovered
nonlinearity in Galton’s family stature regression data. The American Statistician, 57, 190-192.
See Also
The other Galton data sets: heights,peas,cubits
Examples
data(galton)
describe(galton)
#show the scatter plot and the lowess fit
pairs.panels(galton,main="Galton's Parent child heights")
#but this makes the regression lines look the same
pairs.panels(galton,lm=TRUE,main="Galton's Parent child heights")
#better is to scale them
pairs.panels(galton,lm=TRUE,xlim=c(62,74),ylim=c(62,74),main="Galton's Parent child heights")
geometric.mean Find the geometric mean of a vector or columns of a data.frame.
Description
The geometric mean is the nth root of n products or e to the mean log of x. Useful for describing
non-normal, i.e., geometric distributions.
Usage
geometric.mean(x,na.rm=TRUE)
Arguments
xa vector or data.frame
na.rm remove NA values before processing
Details
Useful for teaching how to write functions, also useful for showing the different ways of estimating
central tendency.
Value
geometric mean(s) of x or x.df.
150 glb.algebraic
Note
Not particularly useful if there are elements that are <= 0.
Author(s)
William Revelle
See Also
harmonic.mean,mean
Examples
x <- seq(1,5)
x2 <- x^2
x2[2] <- NA
X <- data.frame(x,x2)
geometric.mean(x)
geometric.mean(x2)
geometric.mean(X)
geometric.mean(X,na.rm=FALSE)
glb.algebraic Find the greatest lower bound to reliability.
Description
The greatest lower bound solves the “educational testing problem". That is, what is the reliability
of a test? (See guttman for a discussion of the problem). Although there are many estimates of a
test reliability (Guttman, 1945) most underestimate the true reliability of a test.
For a given covariance matrix of items, C, the function finds the greatest lower bound to reliability
of the total score using the csdp function from the Rcsdp package.
Usage
glb.algebraic(Cov, LoBounds = NULL, UpBounds = NULL)
Arguments
Cov A p * p covariance matrix. Positive definiteness is not checked.
LoBounds A vector l= (l1, . . . , lp)of length p with lower bounds to the diagonal elements
xi. The default l=(0, . . . , 0) does not imply any constraint, because positive
semidefiniteness of the matrix ˜
C+Diag(x)implies 0≤xi
UpBounds A vector u =(u1, . . . , up) of length p with upper bounds to the diagonal
elements xi. The default is u = v.
glb.algebraic 151
Details
If C is a p * p-covariance matrix, v = diag(C) its diagonal (i. e. the vector of variances vi=cii),
˜
C=C−Diag(v)is the covariance matrix with 0s substituted in the diagonal and x = the vector
x1, . . . , xnthe educational testing problem is (see e. g., Al-Homidan 2008)
p
X
i=1
xi→min
s.t. ˜
C+Diag(x)≥0
(i.e. positive semidefinite) and xi≤vi, i = 1, . . . , p. This is the same as minimizing the trace of
the symmetric matrix
˜
C+diag(x) =
x1c12 . . . c1p
c12 x2. . . c2p
.
.
..
.
.....
.
.
c1pc2p. . . xp
s. t. ˜
C+Diag(x)is positive semidefinite and xi≤vi.
The greatest lower bound to reliability is
Pij ¯cij +Pixi
Pij cij
Additionally, function glb.algebraic allows the user to change the upper bounds xi≤vito xi≤ui
and add lower bounds li≤xi.
The greatest lower bound to reliability is applicable for tests with non-homogeneous items. It gives
a sharp lower bound to the reliability of the total test score.
Caution: Though glb.algebraic gives exact lower bounds for exact covariance matrices, the esti-
mates from empirical matrices may be strongly biased upwards for small and medium sample sizes.
glb.algebraic is wrapper for a call to function csdp of package Rcsdp (see its documentation).
If Cov is the covariance matrix of subtests/items with known lower bounds, rel, to their reliabilities
(e. g. Cronbachs α), LoBounds can be used to improve the lower bound to reliability by setting
LoBounds <- rel*diag(Cov).
Changing UpBounds can be used to relax constraints xi≤vior to fix xi-values by setting LoBounds[i]
< -z; UpBounds[i] <- z.
Value
glb The algebraic greatest lower bound
solution The vector x of the solution of the semidefinite program. These are the elements
on the diagonal of C.
status Status of the solution. See documentation of csdp in package Rcsdp. If status is
2 or greater or equal than 4, no glb and solution is returned. If status is not 0, a
warning message is generated.
Call The calling string
152 glb.algebraic
Author(s)
Andreas Moltner
Center of Excellence for Assessment in Medicine/Baden-Wurttemberg
University of Heidelberg
William Revelle
Department of Psychology
Northwestern University Evanston, Illiniois
http://personality-project.org/revelle.html
References
Al-Homidan S (2008). Semidefinite programming for the educational testing problem. Central
European Journal of Operations Research, 16:239-249.
Bentler PM (1972) A lower-bound method for the dimension-free measurement of internal consis-
tency. Soc Sci Res 1:343-357.
Fletcher R (1981) A nonlinear programming problem in statistics (educational testing). SIAM J Sci
Stat Comput 2:257-267.
Shapiro A, ten Berge JMF (2000). The asymptotic bias of minimum trace factor analysis, with
applications to the greatest lower bound to reliability. Psychometrika, 65:413-425.
ten Berge, Socan G (2004). The greatest bound to reliability of a test and the hypothesis of unidi-
mensionality. Psychometrika, 69:613-625.
See Also
For an alternative estimate of the greatest lower bound, see glb.fa. For multiple estimates of
reliablity, see guttman
Examples
Cv<-matrix(c(215, 64, 33, 22,
64, 97, 57, 25,
33, 57,103, 36,
22, 25, 36, 77),ncol=4)
Cv # covariance matrix of a test with 4 subtests
Cr<-cov2cor(Cv) # Correlation matrix of tests
if(!require(Rcsdp)) {print("Rcsdp must be installed to find the glb.algebraic")} else {
glb.algebraic(Cv) # glb of total score
glb.algebraic(Cr) # glb of sum of standardized scores
w<-c(1,2,2,1) # glb of weighted total score
glb.algebraic(diag(w) %*% Cv %*% diag(w))
alphas <- c(0.8,0,0,0) # Internal consistency of first test is known
glb.algebraic(Cv,LoBounds=alphas*diag(Cv))
# Fix all diagonal elements to 1 but the first:
Gleser 153
lb <- glb.algebraic(Cr,LoBounds=c(0,1,1,1),UpBounds=c(1,1,1,1))
lb$solution[1] # should be the same as the squared mult. corr.
smc(Cr)[1]
}
Gleser Example data from Gleser, Cronbach and Rajaratnam (1965) to show
basic principles of generalizability theory.
Description
Gleser, Cronbach and Rajaratnam (1965) discuss the estimation of variance components and their
ratios as part of their introduction to generalizability theory. This is a adaptation of their "illustrative
data for a completely matched G study" (Table 3). 12 patients are rated on 6 symptoms by two
judges. Components of variance are derived from the ANOVA.
Usage
data(Gleser)
Format
A data frame with 12 observations on the following 12 variables. J item by judge:
J11 a numeric vector
J12 a numeric vector
J21 a numeric vector
J22 a numeric vector
J31 a numeric vector
J32 a numeric vector
J41 a numeric vector
J42 a numeric vector
J51 a numeric vector
J52 a numeric vector
J61 a numeric vector
J62 a numeric vector
Details
Generalizability theory is the application of a components of variance approach to the analysis of
reliability. Given a G study (generalizability) the components are estimated and then may be used
in a D study (Decision). Different ratios are formed as appropriate for the particular D study.
154 Gorsuch
Source
Gleser, G., Cronbach, L., and Rajaratnam, N. (1965). Generalizability of scores influenced by mul-
tiple sources of variance. Psychometrika, 30(4):395-418. (Table 3, rearranged to show increasing
patient severity and increasing item severity.
References
Gleser, G., Cronbach, L., and Rajaratnam, N. (1965). Generalizability of scores influenced by
multiple sources of variance. Psychometrika, 30(4):395-418.
Examples
#Find the MS for each component:
#First, stack the data
data(Gleser)
stack.g <- stack(Gleser)
st.gc.df <- data.frame(stack.g,Persons=rep(letters[1:12],12),
Items=rep(letters[1:6],each=24),Judges=rep(letters[1:2],each=12))
#now do the ANOVA
anov <- aov(values ~ (Persons*Judges*Items),data=st.gc.df)
summary(anov)
Gorsuch Example data set from Gorsuch (1997) for an example factor exten-
sion.
Description
Gorsuch (1997) suggests an alternative to the classic Dwyer (1937) factor extension technique. This
data set is taken from that article. Useful for comparing link{fa.extension} with and without
the correct=TRUE option.
Usage
data(Gorsuch)
Details
Gorsuc (1997) suggested an alternative model for factor extension. His method is appropriate for
the case of repeated variables. This is handled in link{fa.extension} with correct=FALSE
Source
Richard L. Gorsuch (1997) New Procedure for Extension Analysis in Exploratory Factor Analysis.
Educational and Psychological Measurement, 57, 725-740.
Harman 155
References
Dwyer, Paul S. (1937), The determination of the factor loadings of a given test from the known
factor loadings of other tests. Psychometrika, 3, 173-178
Examples
data(Gorsuch)
Ro <- Gorsuch[1:6,1:6]
Roe <- Gorsuch[1:6,7:10]
fo <- fa(Ro,2,rotate="none")
fa.extension(Roe,fo,correct=FALSE)
Harman Two data sets from Harman (1967). 9 cognitive variables from
Holzinger and 8 emotional variables from Burt
Description
Two classic data sets reported by Harman (1967) are 9 psychological (cognitive) variables taken
from Holzinger and 8 emotional variables taken from Burt. Both of these are used for tests and
demonstrations of various factoring algortithms.
Usage
data(Harman)
Details
• Harman.Holzinger: 9 x 9 correlation matrix of ability tests, N = 696.
• Harman.Burt: a 8 x 8 correlation matrix of “emotional" items. N = 172
Harman.Holzinger. The nine psychological variables from Harman (1967, p 244) are taken from
unpublished class notes of K.J. Holzinger with 696 participants. This is a subset of 12 tests with
4 factors. It is yet another nice example of a bifactor solution. Bentler (2007) uses this data set to
discuss reliablity analysis. The data show a clear bifactor structure and are a nice example of the
various estimates of reliability included in the omega function. Should not be confused with the
Holzinger or Holzinger.9 data sets in bifactor.
Harman.Burt. Eight “emotional" variables are taken from Harman (1967, p 164) who in turn
adapted them from Burt (1939). They are said be from 172 normal children aged nine to twelve.
As pointed out by Harman, this correlation matrix is singular and has squared multiple correlations
> 1. Because of this problem, it is a nice test case for various factoring algorithms. (For instance,
omega will issue warning messages for fm="minres" or fm="pa" but will fail for fm="ml".)
The Burt data set probably has a typo in the original correlation matrix. Changing the Sorrow-
Tenderness correlation from .87 to .81 makes the correlation positive definite.
As pointed out by Jan DeLeeuw, the Burt data set is a subset of 8 variables from the original 11
reported by Burt in 1915. That matrix has the same problem. See burt.
156 Harman.5
Other example data sets that are useful demonstrations of factor analysis are the seven bifactor
examples in bifactor and the 24 ability measures in Harman74.cor
There are several other Harman examples in the psych package (i.e., Harman.8) as well as in the
dataseta and GPArotation packages. The Harman 24 mental tests problem is in the basic datasets
package at Harman74.cor.
Source
Harman (1967 p 164 and p 244.)
References
Harman, Harry Horace (1967), Modern factor analysis. Chicago, University of Chicago Press.
P.Bentler. Covariance structure models for maximal reliability of unit-weighted composites. In
Handbook of latent variable and related models, pages 1–17. North Holland, 2007.
Burt, C.General and Specific Factors underlying the Primary Emotions. Reports of the British As-
sociation for the Advancement of Science, 85th meeting, held in Manchester, September 7-11, 1915.
London, John Murray, 1916, p. 694-696 (retrieved from the web at http://www.biodiversitylibrary.org/item/95822#790)
See Also
See also the original burt data set
Examples
data(Harman)
cor.plot(Harman.Holzinger)
cor.plot(Harman.Burt)
smc(Harman.Burt) #note how this produces impossible results
Harman.5 5 socio-economic variables from Harman (1967)
Description
Harman (1967) uses 5 socio-economic variables for demonstrations of principal components and
factor analysis. This example is used in the SAS manual for Proc Factor as well.
Usage
data(Harman.5)
Harman.8 157
Format
A data frame with 12 observations on the following 5 variables.
population a numeric vector
schooling a numeric vector
employment a numeric vector
professional a numeric vector
housevalue a numeric vector
Details
Harman reports that the data "were taken (not entirely arbitrarily) from a study of the Los Angeles
Standard Metropolitan Statistical Area. The twelve individuals are used in the examples are census
tracts." (p 13).
Source
Harman, Harry Horace (1967), Modern factor analysis. Chicago, University of Chicago Press.
References
SAS users manual, chapter 26: pages 1123-1192
Examples
data(Harman.5)
if(require('GPArotation')){
pc2 <- principal(Harman.5,2,scores=TRUE)
pc2$residual
biplot(pc2,main="Biplot of the Harman 5 socio-demographic variables") }
Harman.8 Correlations of eight physical variables (from Harman, 1966)
Description
A classic data set from Harman (1976) reporting the correlations of eight physical variables. Used
by Harman for demonstrations of factor analysis (both principal axis and minimum residual).
Usage
data(Harman.8)
Format
The format is: num [1:8, 1:8] 1 0.846 0.805 0.859 0.473 0.398 0.301 0.382 0.846 1 ... - attr(*,
"dimnames")=List of 2 ..$ : chr [1:8] "Height" "Arm span" "Length of forearm" "Length of lower
leg" ... ..$ : chr [1:8] "V1" "V2" "V3" "V4" ...
158 Harman.8
Details
The Eight Physical Variables problem is taken from Harman (1976) and represents the correlations
between eight physical variables for 305 girls. The two correlated clusters represent four measures
of "lankiness" and then four measures of "stockiness". The original data were selected from 17
variables reported in an unpublished dissertation by Mullen (1939).
Variable 6 ("Bitrochanteric diamter") is the distance between the outer points of the hips.
The row names match the original Harman paper, the column names have been abbreviated.
The fa solution for principal axes (fm="pa") matches the reported minres solution, while the
fm="minres" does not.
For those interested in teaching examples using various body measurements, see the body data set
in the gclus package.
There are several other Harman examples in the psych package as well as in the dataseta and and
GPArotation packages. The Harman 24 mental tests problem is in the basic datasets package at
Harman74.cor.
Source
H. Harman and W.Jones. (1966) Factor analysis by minimizing residuals (minres). Psychometrika,
31(3):351-368.
References
Harman, Harry Horace (1976) Modern factor analysis, 3d ed., rev, University of Chicago Press.
Chicago.
Harman, Harry Horace and Jones, W. (1966) Factor analysis by minimizing residuals (minres).
Psychometrika, 31(3):351-368.
See Also
Harman,Harman.political and Harman74.cor
Examples
data(Harman.8)
cor.plot(Harman.8)
fa(Harman.8,2,rotate="none") #the minres solution
fa(Harman.8,2,rotate="none",fm="pa") #the principal axis solution
Harman.political 159
Harman.political Eight political variables used by Harman (1967) as example 8.17
Description
Another one of the many Harman (1967) data sets. This contains 8 political variables taken over
147 election areas. The principal factor method with SMCs as communalities match those of table
8.18. The data are used by Dziubian and Shirkey as an example of the Kaiser-Meyer-Olkin test of
factor adequacy.
Usage
data(Harman.political)
Format
The format is: num [1:8, 1:8] 1 0.84 0.62 -0.53 0.03 0.57 -0.33 -0.63 0.84 1 ... - attr(*, "dim-
names")=List of 2 ..$ : chr [1:8] "Lewis" "Roosevelt" "Party Voting" "Median Rental" ... ..$ : chr
[1:8] "Lewis" "Roosevelt" "Party Voting" "Median Rental" ...
Details
The communalities from the original table are not included. They are .52, 1.00, .78, .82, .36, .80,
.63, and .97
Source
Harman, Harry Horace (1976) Modern factor analysis, 3d ed., rev, University of Chicago Press.
Chicago. p 166.
References
Dziuban, Charles D. and Shirkey, Edwin C. (1974) When is a correlation matrix appropriate for
factor analysis? Some decision rules. Psychological Bulletin, 81 (6) 358 - 361.
Examples
data(Harman.political)
KMO(Harman.political)
160 harmonic.mean
harmonic.mean Find the harmonic mean of a vector, matrix, or columns of a
data.frame
Description
The harmonic mean is merely the reciprocal of the arithmetic mean of the reciprocals.
Usage
harmonic.mean(x,na.rm=TRUE)
Arguments
xa vector, matrix, or data.frame
na.rm na.rm=TRUE remove NA values before processing
Details
Included as an example for teaching about functions. As well as for a discussion of how to estimate
central tendencies.
Value
The harmonic mean(s)
Note
Included as a simple demonstration of how to write a function
Examples
x <- seq(1,5)
x2 <- x^2
x2[2] <- NA
X <- data.frame(x,x2)
harmonic.mean(x)
harmonic.mean(x2)
harmonic.mean(X)
harmonic.mean(X,FALSE)
headTail 161
headTail Combine calls to head and tail
Description
A quick way to show the first and last n lines of a data.frame, matrix, or a text object. Just a pretty
call to head and tail
Usage
headTail(x,hlength=4,tlength=4,digits=2,ellipsis=TRUE)
headtail(x,hlength=4,tlength=4,digits=2,ellipsis=TRUE)
topBottom(x,hlength=4,tlength=4,digits=2)
Arguments
xA matrix or data frame or free text
hlength The number of lines at the beginning to show
tlength The number of lines at the end to show
digits Round off the data to digits
ellipsis Separate the head and tail with dots (ellipsis)
Value
The first hlength and last tlength lines of a matrix or data frame with an ellipsis in between. If the
input is neither a matrix nor data frame, the output will be the first hlength and last tlength lines.
topBottom is just a call to headTail with ellipsis = FALSE and returning a matrix output.
See Also
head and tail
Examples
headTail(iqitems[1:5],4,8)
162 heights
heights A data.frame of the Galton (1888) height and cubit data set.
Description
Francis Galton introduced the ’co-relation’ in 1888 with a paper discussing how to measure the
relationship between two variables. His primary example was the relationship between height and
forearm length. The data table (cubits) is taken from Galton (1888). Unfortunately, there seem to
be some errors in the original data table in that the marginal totals do not match the table.
The data frame, heights, is converted from this table using table2df.
Usage
data(heights)
Format
A data frame with 348 observations on the following 2 variables.
height Height in inches
cubit Forearm length in inches
Details
Sir Francis Galton (1888) published the first demonstration of the correlation coefficient. The re-
gression (or reversion to mediocrity) of the height to the length of the left forearm (a cubit) was
found to .8. The original table cubits is taken from Galton (1888). There seem to be some errors
in the table as published in that the row sums do not agree with the actual row sums. These data
are used to create a matrix using table2matrix for demonstrations of analysis and displays of the
data.
Source
Galton (1888)
References
Galton, Francis (1888) Co-relations and their measurement. Proceedings of the Royal Society.
London Series,45,135-145,
See Also
table2matrix,table2df,cubits,ellipses,galton
Examples
data(heights)
ellipses(heights,n=1,main="Galton's co-relation data set")
ICC 163
ICC Intraclass Correlations (ICC1, ICC2, ICC3 from Shrout and Fleiss)
Description
The Intraclass correlation is used as a measure of association when studying the reliability of raters.
Shrout and Fleiss (1979) outline 6 different estimates, that depend upon the particular experimental
design. All are implemented and given confidence limits.
Usage
ICC(x,missing=TRUE,alpha=.05)
Arguments
xa matrix or dataframe of ratings
missing if TRUE, remove missing data – work on complete cases only
alpha The alpha level for significance for finding the confidence intervals
Details
Shrout and Fleiss (1979) consider six cases of reliability of ratings done by k raters on n targets.
ICC1: Each target is rated by a different judge and the judges are selected at random. (This is a
one-way ANOVA fixed effects model and is found by (MSB- MSW)/(MSB+ (nr-1)*MSW))
ICC2: A random sample of k judges rate each target. The measure is one of absolute agreement in
the ratings. Found as (MSB- MSE)/(MSB + (nr-1)*MSE + nr*(MSJ-MSE)/nc)
ICC3: A fixed set of k judges rate each target. There is no generalization to a larger population of
judges. (MSB - MSE)/(MSB+ (nr-1)*MSE)
Then, for each of these cases, is reliability to be estimated for a single rating or for the average of
k ratings? (The 1 rating case is equivalent to the average intercorrelation, the k rating case to the
Spearman Brown adjusted reliability.)
ICC1 is sensitive to differences in means between raters and is a measure of absolute agreement.
ICC2 and ICC3 remove mean differences between judges, but are sensitive to interactions of raters
by judges. The difference between ICC2 and ICC3 is whether raters are seen as fixed or random
effects.
ICC1k, ICC2k, ICC3K reflect the means of k raters.
The intraclass correlation is used if raters are all of the same “class". That is, there is no logical way
of distinguishing them. Examples include correlations between pairs of twins, correlations between
raters. If the variables are logically distinguishable (e.g., different items on a test), then the more
typical coefficient is based upon the inter-class correlation (e.g., a Pearson r) and a statistic such as
alpha or omega might be used.
164 ICC
Value
results A matrix of 6 rows and 8 columns, including the ICCs, F test, p values, and
confidence limits
summary The anova summary table
stats The anova statistics
MSW Mean Square Within based upon the anova
Note
The results for the Lower and Upper Bounds for ICC(2,k) do not match those of SPSS 9 or 10,
but do match the definitions of Shrout and Fleiss. SPSS seems to have been using the formula in
McGraw and Wong, but not the errata on p 390. They seem to have fixed it in more recent releases
(15).
Starting with psych 1.4.2, the confidence intervals are based upon (1-alpha)% at both tails of the
confidence interval. This is in agreement with Shrout and Fleiss. Prior to 1.4.2 the confidence
intervals were (1-alpha/2)%.
Author(s)
William Revelle
References
Shrout, Patrick E. and Fleiss, Joseph L. Intraclass correlations: uses in assessing rater reliability.
Psychological Bulletin, 1979, 86, 420-3428.
McGraw, Kenneth O. and Wong, S. P. (1996), Forming inferences about some intraclass correlation
coefficients. Psychological Methods, 1, 30-46. + errata on page 390.
Revelle, W. (in prep) An introduction to psychometric theory with applications in R. Springer.
(working draft available at http://personality-project.org/r/book/
Examples
sf <- matrix(c(9, 2, 5, 8,
6, 1, 3, 2,
8, 4, 6, 8,
7, 1, 2, 6,
10, 5, 6, 9,
6, 2, 4, 7),ncol=4,byrow=TRUE)
colnames(sf) <- paste("J",1:4,sep="")
rownames(sf) <- paste("S",1:6,sep="")
sf #example from Shrout and Fleiss (1979)
ICC(sf)
iclust 165
iclust iclust: Item Cluster Analysis – Hierarchical cluster analysis using psy-
chometric principles
Description
A common data reduction technique is to cluster cases (subjects). Less common, but particularly
useful in psychological research, is to cluster items (variables). This may be thought of as an
alternative to factor analysis, based upon a much simpler model. The cluster model is that the
correlations between variables reflect that each item loads on at most one cluster, and that items
that load on those clusters correlate as a function of their respective loadings on that cluster and
items that define different clusters correlate as a function of their respective cluster loadings and the
intercluster correlations. Essentially, the cluster model is a Very Simple Structure factor model of
complexity one (see VSS).
This function applies the iclust algorithm to hierarchically cluster items to form composite scales.
Clusters are combined if coefficients alpha and beta will increase in the new cluster.
Alpha, the mean split half correlation, and beta, the worst split half correlation, are estimates of
the reliability and general factor saturation of the test. (See also the omega function to estimate
McDonald’s coeffients ωhand ωt)
Usage
iclust(r.mat, nclusters=0, alpha=3, beta=1, beta.size=4, alpha.size=3,
correct=TRUE,correct.cluster=TRUE, reverse=TRUE, beta.min=.5, output=1,
digits=2,labels=NULL,cut=0, n.iterations =0, title="ICLUST", plot=TRUE,
weighted=TRUE,cor.gen=TRUE,SMC=TRUE,purify=TRUE,diagonal=FALSE)
ICLUST(r.mat, nclusters=0, alpha=3, beta=1, beta.size=4, alpha.size=3,
correct=TRUE,correct.cluster=TRUE, reverse=TRUE, beta.min=.5, output=1,
digits=2,labels=NULL,cut=0,n.iterations = 0,title="ICLUST",plot=TRUE,
weighted=TRUE,cor.gen=TRUE,SMC=TRUE,purify=TRUE,diagonal=FALSE)
#iclust(r.mat) #use all defaults
#iclust(r.mat,nclusters =3) #use all defaults and if possible stop at 3 clusters
#ICLUST(r.mat, output =3) #long output shows clustering history
#ICLUST(r.mat, n.iterations =3) #clean up solution by item reassignment
Arguments
r.mat A correlation matrix or data matrix/data.frame. (If r.mat is not square i.e, a
correlation matrix, the data are correlated using pairwise deletion.
nclusters Extract clusters until nclusters remain (default will extract until the other criteria
are met or 1 cluster, whichever happens first). See the discussion below for
alternative techniques for specifying the number of clusters.
166 iclust
alpha Apply the increase in alpha criterion (0) never or for (1) the smaller, 2) the
average, or 3) the greater of the separate alphas. (default = 3)
beta Apply the increase in beta criterion (0) never or for (1) the smaller, 2) the aver-
age, or 3) the greater of the separate betas. (default =1)
beta.size Apply the beta criterion after clusters are of beta.size (default = 4)
alpha.size Apply the alpha criterion after clusters are of size alpha.size (default =3)
correct Correct correlations for reliability (default = TRUE)
correct.cluster
Correct cluster -sub cluster correlations for reliability of the sub cluster , default
is TRUE))
reverse Reverse negative keyed items (default = TRUE
beta.min Stop clustering if the beta is not greater than beta.min (default = .5)
output 1) short, 2) medium, 3 ) long output (default =1)
labels vector of item content or labels. If NULL, then the colnames are used. If
FALSE, then labels are V1 .. Vn
cut sort cluster loadings > absolute(cut) (default = 0)
n.iterations iterate the solution n.iterations times to "purify" the clusters (default = 0)
digits Precision of digits of output (default = 2)
title Title for this run
plot Should ICLUST.rgraph be called automatically for plotting (requires Rgraphviz
default=TRUE)
weighted Weight the intercluster correlation by the size of the two clusters (TRUE) or do
not weight them (FALSE)
cor.gen When correlating clusters with subclusters, base the correlations on the general
factor (default) or general + group (cor.gen=FALSE)
SMC When estimating cluster-item correlations, use the smcs as the estimate of an
item communality (SMC=TRUE) or use the maximum correlation (SMC=FALSE).
purify Should clusters be defined as the original groupings (purify = FAlSE) or by the
items with the highest loadings on those original clusters? (purify = TRUE)
diagonal Should the diagonal be included in the fit statistics. The default is not to include
it. Prior to 1.2.8, the diagonal was included.
Details
Extensive documentation and justification of the algorithm is available in the original MBR 1979
http://personality-project.org/revelle/publications/iclust.pdf paper. Further dis-
cussion of the algorithm and sample output is available on the personality-project.org web page:
http://personality-project.org/r/r.ICLUST.html
The results are best visualized using ICLUST.graph, the results of which can be saved as a dot file
for the Graphviz program. http://www.graphviz.org/. The iclust.diagram is called automatically
to produce cluster diagrams. The resulting diagram is not quite as pretty as what can be achieved
in dot code but is quite adequate if you don’t want to use an external graphics program. With the
installation of Rgraphviz, ICLUST can also provide cluster graphs.
iclust 167
A common problem in the social sciences is to construct scales or composites of items to measure
constructs of theoretical interest and practical importance. This process frequently involves admin-
istering a battery of items from which those that meet certain criteria are selected. These criteria
might be rational, empirical,or factorial. A similar problem is to analyze the adequacy of scales
that already have been formed and to decide whether the putative constructs are measured properly.
Both of these problems have been discussed in numerous texts, as well as in myriad articles. Pro-
ponents of various methods have argued for the importance of face validity, discriminant validity,
construct validity, factorial homogeneity, and theoretical importance.
Revelle (1979) proposed that hierachical cluster analysis could be used to estimate a new coefficient
(beta) that was an estimate of the general factor saturation of a test. More recently, Zinbarg, Revelle,
Yovel and Li (2005) compared McDonald’s Omega to Chronbach’s alpha and Revelle’s beta. They
conclude that ωhhierarchical is the best estimate. An algorithm for estimating omega is available
as part of this package.
Revelle and Zinbarg (2009) discuss alpha, beta, and omega, as well as other estimates of reliability.
The original ICLUST program was written in FORTRAN to run on CDC and IBM mainframes and
was then modified to run in PC-DOS. The R version of iclust is a completely new version written
for the psych package. Please email me if you want help with this version of iclust or if you desire
more features.
A requested feature (not yet available) is to specify certain items as forming a cluster. That is, to do
confirmatory cluster analysis.
The program currently has three primary functions: cluster, loadings, and graphics.
In June, 2009, the option of weighted versus unweighted beta was introduced. Unweighted beta
calculates beta based upon the correlation between two clusters, corrected for test length using the
Spearman-Brown prophecy formala, while weighted beta finds the average interitem correlation
between the items within two clusters and then finds beta from this. That is, for two clusters A and
B of size N and M with between average correlation rb, weighted beta is (N+M)^2 rb/(Va +Vb +
2Cab). Raw (unweighted) beta is 2rab/(1+rab) where rab = Cab/sqrt(VaVb). Weighted beta seems a
more appropriate estimate and is now the default. Unweighted beta is still available for consistency
with prior versions.
Also modified in June, 2009 was the way of correcting for item overlap when calculating the cluster-
subcluster correlations for the graphic output. This does not affect the final cluster solution, but
does produce slightly different path values. In addition, there are two ways to solve for the cluster -
subcluster correlation.
Given the covariance between two clusters, Cab with average rab = Cab/(N*M), and cluster vari-
ances Va and Vb with Va = N + N*(N-1)*ra then the correlation of cluster A with the combined
cluster AB is either
a) ((N^2)ra + Cab)/sqrt(Vab*Va) (option cor.gen=TRUE) or b) (Va - N + Nra + Cab)/sqrt(Vab*Va)
(option cor.gen=FALSE)
The default is to use cor.gen=TRUE.
Although iclust will give what it thinks is the best solution in terms of the number of clusters to
extract, the user will sometimes disagree. To get more clusters than the default solution, just set
the nclusters parameter to the number desired. However, to get fewer than meet the alpha and beta
criteria, it is sometimes necessary to set alpha=0 and beta=0 and then set the nclusters to the desired
number.
Clustering 24 tests of mental ability
168 iclust
A sample output using the 24 variable problem by Harman can be represented both graphically and
in terms of the cluster order. The default is to produce graphics using the diagram functions. An
alternative is to use the Rgraphviz package (from BioConductor). Because this package is some-
times hard to install, there is an alternative option (ICLUST.graph to write dot language instructions
for subsequent processing. This will create a graphic instructions suitable for any viewing program
that uses the dot language. ICLUST.rgraph produces the dot code for Graphviz. Somewhat lower
resolution graphs with fewer options are available in the ICLUST.rgraph function which requires
Rgraphviz. Dot code can be viewed directly in Graphviz or can be tweaked using commercial
software packages (e.g., OmniGraffle)
Note that for the Harman 24 variable problem, with the default parameters, the data form one large
cluster. (This is consistent with the Very Simple Structure (VSS) output as well, which shows a clear
one factor solution for complexity 1 data.)
An alternative solution is to ask for a somewhat more stringent set of criteria and require an increase
in the size of beta for all clusters greater than 3 variables. This produces a 4 cluster solution.
It is also possible to use the original parameter settings, but ask for a 4 cluster solution.
At least for the Harman 24 mental ability measures, it is interesting to compare the cluster pattern
matrix with the oblique rotation solution from a factor analysis. The factor congruence of a four
factor oblique pattern solution with the four cluster solution is > .99 for three of the four clusters
and > .97 for the fourth cluster. The cluster pattern matrix (returned as an invisible object in the
output)
In September, 2012, the fit statistics (pattern fit and cluster fit) were slightly modified to (by default)
not consider the diagonal (diagonal=FALSE). Until then, the diagonal was included in the cluster fit
statistics. The pattern fit is analogous to factor analysis and is based upon the model = P x Structure
where Structure is Pattern * Phi. Then R* = R - model and fit is the ratio of sum(r*^2)/sum(r^2) for
the off diagonal elements.
Value
title Name of this analysis
results A list containing the step by step cluster history, including which pair was
grouped, what were the alpha and betas of the two groups and of the combined
group.
Note that the alpha values are “standardized alphas” based upon the correlation
matrix, rather than the raw alphas that will come from scoreItems
The print.psych and summary.psych functions will print out just the must im-
portant results.
corrected The raw and corrected for alpha reliability cluster intercorrelations.
clusters a matrix of -1,0, and 1 values to define cluster membership.
purified A list of the cluster definitions and cluster loadings of the purified solution.
These are sorted by importance (the eigenvalues of the clusters). The cluster
membership from the original (O) and purified (P) clusters are indicated along
with the cluster structure matrix. These item loadings are the same as those
found by the scoreItems function and are found by correcting the item-cluster
correlation for item overlap by summing the item-cluster covariances with all
except that item and then adding in the smc for that item. These resulting corre-
lations are then corrected for scale reliability.
iclust 169
To show just the most salient items, use the cutoff option in print.psych
cluster.fit, structure.fit, pattern.fit
There are a number of ways to evaluate how well any factor or cluster matrix
reproduces the original matrix. Cluster fit considers how well the clusters fit if
only correlations with clusters are considered. Structure fit evaluates R = CC’
while pattern fit evaluate R = C inverse (phi) C’ where C is the cluster loading
matrix, and phi is the intercluster correlation matrix.
pattern The pattern matrix loadings. Pattern is just C inverse (Phi). The pattern matrix is
conceptually equivalent to that of a factor analysis, in that the pattern coefficients
are b weights of the cluster to the variables, while the normal cluster loadings
are correlations of the items with the cluster. The four cluster and four factor
pattern matrices for the Harman problem are very similar.
Note
iclust draws graphical displays with or without using Rgraphiviz. Because of difficulties installing
Rgraphviz on many systems, the default it not even try using it. With the introduction of the diagram
functions, iclust now draws using iclust.diagram which is not as pretty as using Rgraphviz, but
more stable. However, Rgraphviz can be used by using ICLUST.rgraph to produces slightly better
graphics. It is also possible to export dot code in the dot language for further massaging of the
graphic. This may be done using ICLUST.graph. This last option is probably preferred for nice
graphics which can be massaged in any dot code program (e.g., graphviz (http://graphviz.org) or a
commercial program such as OmniGraffle.
To view the cluster structure more closely, it is possible to save the graphic output as a pdf and then
magnify this using a pdf viewer. This is useful when clustering a large number of variables.
In order to sort the clusters by cluster loadings, use iclust.sort.
Author(s)
William Revelle
References
Revelle, W. Hierarchical Cluster Analysis and the Internal Structure of Tests. Multivariate Behav-
ioral Research, 1979, 14, 57-74.
Revelle, W. and Zinbarg, R. E. (2009) Coefficients alpha, beta, omega and the glb: comments on
Sijtsma. Psychometrika, 2009.
http://personality-project.org/revelle/publications/iclust.pdf
See also more extensive documentation at http://personality-project.org/r/r.ICLUST.html
and
Revelle, W. (in prep) An introduction to psychometric theory with applications in R. To be published
by Springer. (working draft available at http://personality-project.org/r/book/
See Also
iclust.sort,ICLUST.graph,ICLUST.cluster,cluster.fit ,VSS,omega
170 ICLUST.cluster
Examples
test.data <- Harman74.cor$cov
ic.out <- iclust(test.data,title="ICLUST of the Harman data")
summary(ic.out)
#use all defaults and stop at 4 clusters
ic.out4 <- iclust(test.data,nclusters =4,title="Force 4 clusters")
summary(ic.out4)
ic.out1 <- iclust(test.data,beta=3,beta.size=3) #use more stringent criteria
ic.out #more complete output
plot(ic.out4) #this shows the spatial representation
#use a dot graphics viewer on the out.file
dot.graph <- ICLUST.graph(ic.out,out.file="test.ICLUST.graph.dot")
#show the equivalent of a factor solution
fa.diagram(ic.out4$pattern,Phi=ic.out4$Phi,main="Pattern taken from iclust")
ICLUST.cluster Function to form hierarchical cluster analysis of items
Description
The guts of the ICLUST algorithm. Called by ICLUST See ICLUST for description.
Usage
ICLUST.cluster(r.mat, ICLUST.options,smc.items)
Arguments
r.mat A correlation matrix
ICLUST.options A list of options (see ICLUST)
smc.items passed from the main program to speed up processing
Details
See ICLUST
Value
A list of cluster statistics, described more fully in ICLUST
comp1 Description of ’comp1’
comp2 Description of ’comp2’
...
iclust.diagram 171
Note
Although the main code for ICLUST is here in ICLUST.cluster, the more extensive documentation
is for ICLUST.
Author(s)
William Revelle
References
Revelle, W. 1979, Hierarchical Cluster Analysis and the Internal Structure of Tests. Multivariate
Behavioral Research, 14, 57-74. http://personality-project.org/revelle/publications/
iclust.pdf
See also more extensive documentation at http://personality-project.org/r/r.ICLUST.html
See Also
ICLUST.graph,ICLUST,cluster.fit ,VSS,omega
iclust.diagram Draw an ICLUST hierarchical cluster structure diagram
Description
Given a cluster structure determined by ICLUST, create a graphic structural diagram using graphic
functions in the psych package To create dot code to describe the ICLUST output with more pre-
cision, use ICLUST.graph. If Rgraphviz has been successfully installed, the alternative is to use
ICLUST.rgraph.
Usage
iclust.diagram(ic, labels = NULL, short = FALSE, digits = 2, cex = NULL, min.size = NULL,
e.size =1,colors=c("black","blue"),
main = "ICLUST diagram",cluster.names=NULL,marg=c(.5,.5,1.5,.5))
Arguments
ic Output from ICLUST
labels labels for variables (if not specified as rownames in the ICLUST output
short if short=TRUE, variable names are replaced with Vn
digits Round the path coefficients to digits accuracy
cex The standard graphic control parameter for font size modifications. This can be
used to make the labels bigger or smaller than the default values.
min.size Don’t provide statistics for clusters less than min.size
e.size size of the ellipses with the cluster statistics.
172 iclust.diagram
colors postive and negative
main The main graphic title
cluster.names Normally, clusters are named sequentially C1 ... Cn. If cluster.names are speci-
fied, then these values will be used instead.
marg Sets the margins to be narrower than the default values. Resets them upon return
Details
iclust.diagram provides most of the power of ICLUST.rgraph without the difficulties involved in
installing Rgraphviz. It is called automatically from ICLUST.
Following a request by Michael Kubovy, cluster.names may be specified to replace the normal C1
... Cn names.
If access to a dot language graphics program is available, it is probably better to use the iclust.graph
function to get dot output for offline editing.
Value
Graphical output summarizing the hierarchical cluster structure. The graph is drawn using the dia-
gram functions (e.g., dia.curve,dia.arrow,dia.rect,dia.ellipse ) created as a work around
to Rgraphviz.
Note
Suggestions for improving the graphic output are welcome.
Author(s)
William Revelle
References
Revelle, W. Hierarchical Cluster Analysis and the Internal Structure of Tests. Multivariate Behav-
ioral Research, 1979, 14, 57-74.
See Also
ICLUST
Examples
v9 <- sim.hierarchical()
v9c <- ICLUST(v9)
test.data <- Harman74.cor$cov
ic.out <- ICLUST(test.data)
#now show how to relabel clusters
ic.bfi <- iclust(bfi[1:25],beta=3) #find the clusters
cluster.names <- rownames(ic.bfi$results) #get the old names
#change the names to the desired ones
cluster.names[c(16,19,18,15,20)] <- c("Neuroticism","Extra-Open","Agreeableness",
ICLUST.graph 173
"Conscientiousness","Open")
#now show the new names
iclust.diagram(ic.bfi,cluster.names=cluster.names,min.size=4,e.size=1.75)
ICLUST.graph create control code for ICLUST graphical output
Description
Given a cluster structure determined by ICLUST, create dot code to describe the ICLUST output.
To use the dot code, use either http://www.graphviz.org/ Graphviz or a commercial viewer (e.g.,
OmniGraffle). This function parallels ICLUST.rgraph which uses Rgraphviz.
Usage
ICLUST.graph(ic.results, out.file,min.size=1, short = FALSE,labels=NULL,
size = c(8, 6), node.font = c("Helvetica", 14), edge.font = c("Helvetica", 12),
rank.direction=c("RL","TB","LR","BT"), digits = 2, title = "ICLUST", ...)
Arguments
ic.results output list from ICLUST
out.file name of output file (defaults to console)
min.size draw a smaller node (without all the information) for clusters < min.size – useful
for large problems
short if short==TRUE, don’t use variable names
labels vector of text labels (contents) for the variables
size size of output
node.font Font to use for nodes in the graph
edge.font Font to use for the labels of the arrows (edges)
rank.direction LR or RL
digits number of digits to show
title any title
... other options to pass
Details
Will create (or overwrite) an output file and print out the dot code to show a cluster structure. This
dot file may be imported directly into a dot viewer (e.g., http://www.graphviz.org/). The "dot"
language is a powerful graphic description language that is particulary appropriate for viewing
cluster output. Commercial graphics programs (e.g., OmniGraffle) can also read (and clean up) dot
files.
174 ICLUST.graph
ICLUST.graph takes the output from ICLUST results and processes it to provide a pretty picture
of the results. Original variables shown as rectangles and ordered on the left hand side (if rank
direction is RL) of the graph. Clusters are drawn as ellipses and include the alpha, beta, and size of
the cluster. Edges show the cluster intercorrelations.
It is possible to trim the output to not show all cluster information. Clusters < min.size are shown
as small ovals without alpha, beta, and size information.
Although it would be nice to process the dot code directly in R, the Rgraphviz package is difficult
to use on all platforms and thus the dot code is written directly.
Value
Output is a set of dot commands written either to console or to the output file. These commands
may then be used as input to any "dot" viewer, e.g., Graphviz.
Author(s)
<revelle@northwestern.edu >
http://personality-project.org/revelle.html
References
ICLUST: http://personality-project.org/r/r.ICLUST.html
See Also
VSS.plot,ICLUST
Examples
## Not run:
test.data <- Harman74.cor$cov
ic.out <- ICLUST(test.data)
out.file <- file.choose(new=TRUE) #create a new file to write the plot commands to
ICLUST.graph(ic.out,out.file)
now go to graphviz (outside of R) and open the out.file you created
print(ic.out,digits=2)
## End(Not run)
#test.data <- Harman74.cor$cov
#my.iclust <- ICLUST(test.data)
#ICLUST.graph(my.iclust)
#
#
#digraph ICLUST {
# rankdir=RL;
# size="8,8";
# node [fontname="Helvetica" fontsize=14 shape=box, width=2];
# edge [fontname="Helvetica" fontsize=12];
# label = "ICLUST";
ICLUST.graph 175
# fontsize=20;
#V1 [label = VisualPerception];
#V2 [label = Cubes];
#V3 [label = PaperFormBoard];
#V4 [label = Flags];
#V5 [label = GeneralInformation];
#V6 [label = PargraphComprehension];
#V7 [label = SentenceCompletion];
#V8 [label = WordClassification];
#V9 [label = WordMeaning];
#V10 [label = Addition];
#V11 [label = Code];
#V12 [label = CountingDots];
#V13 [label = StraightCurvedCapitals];
#V14 [label = WordRecognition];
#V15 [label = NumberRecognition];
#V16 [label = FigureRecognition];
#V17 [label = ObjectNumber];
#V18 [label = NumberFigure];
#V19 [label = FigureWord];
#V20 [label = Deduction];
#V21 [label = NumericalPuzzles];
#V22 [label = ProblemReasoning];
#V23 [label = SeriesCompletion];
#V24 [label = ArithmeticProblems];
#node [shape=ellipse, width ="1"];
#C1-> V9 [ label = 0.78 ];
#C1-> V5 [ label = 0.78 ];
#C2-> V12 [ label = 0.66 ];
#C2-> V10 [ label = 0.66 ];
#C3-> V18 [ label = 0.53 ];
#C3-> V17 [ label = 0.53 ];
#C4-> V23 [ label = 0.59 ];
#C4-> V20 [ label = 0.59 ];
#C5-> V13 [ label = 0.61 ];
#C5-> V11 [ label = 0.61 ];
#C6-> V7 [ label = 0.78 ];
#C6-> V6 [ label = 0.78 ];
#C7-> V4 [ label = 0.55 ];
#C7-> V1 [ label = 0.55 ];
#C8-> V16 [ label = 0.5 ];
#C8-> V14 [ label = 0.49 ];
#C9-> C1 [ label = 0.86 ];
#C9-> C6 [ label = 0.86 ];
#C10-> C4 [ label = 0.71 ];
#C10-> V22 [ label = 0.62 ];
#C11-> V21 [ label = 0.56 ];
#C11-> V24 [ label = 0.58 ];
#C12-> C10 [ label = 0.76 ];
#C12-> C11 [ label = 0.67 ];
#C13-> C8 [ label = 0.61 ];
#C13-> V15 [ label = 0.49 ];
#C14-> C2 [ label = 0.74 ];
176 ICLUST.rgraph
#C14-> C5 [ label = 0.72 ];
#C15-> V3 [ label = 0.48 ];
#C15-> C7 [ label = 0.65 ];
#C16-> V19 [ label = 0.48 ];
#C16-> C3 [ label = 0.64 ];
#C17-> V8 [ label = 0.62 ];
#C17-> C12 [ label = 0.8 ];
#C18-> C17 [ label = 0.82 ];
#C18-> C15 [ label = 0.68 ];
#C19-> C16 [ label = 0.66 ];
#C19-> C13 [ label = 0.65 ];
#C20-> C19 [ label = 0.72 ];
#C20-> C18 [ label = 0.83 ];
#C21-> C20 [ label = 0.87 ];
#C21-> C9 [ label = 0.76 ];
#C22-> 0 [ label = 0 ];
#C22-> 0 [ label = 0 ];
#C23-> 0 [ label = 0 ];
#C23-> 0 [ label = 0 ];
#C1 [label = "C1\n alpha= 0.84\n beta= 0.84\nN= 2"] ;
#C2 [label = "C2\n alpha= 0.74\n beta= 0.74\nN= 2"] ;
#C3 [label = "C3\n alpha= 0.62\n beta= 0.62\nN= 2"] ;
#C4 [label = "C4\n alpha= 0.67\n beta= 0.67\nN= 2"] ;
#C5 [label = "C5\n alpha= 0.7\n beta= 0.7\nN= 2"] ;
#C6 [label = "C6\n alpha= 0.84\n beta= 0.84\nN= 2"] ;
#C7 [label = "C7\n alpha= 0.64\n beta= 0.64\nN= 2"] ;
#C8 [label = "C8\n alpha= 0.58\n beta= 0.58\nN= 2"] ;
#C9 [label = "C9\n alpha= 0.9\n beta= 0.87\nN= 4"] ;
#C10 [label = "C10\n alpha= 0.74\n beta= 0.71\nN= 3"] ;
#C11 [label = "C11\n alpha= 0.62\n beta= 0.62\nN= 2"] ;
#C12 [label = "C12\n alpha= 0.79\n beta= 0.74\nN= 5"] ;
#C13 [label = "C13\n alpha= 0.64\n beta= 0.59\nN= 3"] ;
#C14 [label = "C14\n alpha= 0.79\n beta= 0.74\nN= 4"] ;
#C15 [label = "C15\n alpha= 0.66\n beta= 0.58\nN= 3"] ;
#C16 [label = "C16\n alpha= 0.65\n beta= 0.57\nN= 3"] ;
#C17 [label = "C17\n alpha= 0.81\n beta= 0.71\nN= 6"] ;
#C18 [label = "C18\n alpha= 0.84\n beta= 0.75\nN= 9"] ;
#C19 [label = "C19\n alpha= 0.74\n beta= 0.65\nN= 6"] ;
#C20 [label = "C20\n alpha= 0.87\n beta= 0.74\nN= 15"] ;
#C21 [label = "C21\n alpha= 0.9\n beta= 0.77\nN= 19"] ;
#C22 [label = "C22\n alpha= 0\n beta= 0\nN= 0"] ;
#C23 [label = "C23\n alpha= 0\n beta= 0\nN= 0"] ;
#{ rank=same;
#V1;V2;V3;V4;V5;V6;V7;V8;V9;V10;V11;V12;V13;V14;V15;V16;V17;V18;V19;V20;V21;V22;V23;V24;}}
#
#copy the above output to Graphviz and draw it
#see \url{http://personality-project.org/r/r.ICLUST.html} for an example.
ICLUST.rgraph Draw an ICLUST graph using the Rgraphviz package
ICLUST.rgraph 177
Description
Given a cluster structure determined by ICLUST, create a rgraphic directly using Rgraphviz. To cre-
ate dot code to describe the ICLUST output with more precision, use ICLUST.graph. As an option,
dot code is also generated and saved in a file. To use the dot code, use either http://www.graphviz.org/
Graphviz or a commercial viewer (e.g., OmniGraffle).
Usage
ICLUST.rgraph(ic.results, out.file = NULL, min.size = 1, short = FALSE,
labels = NULL, size = c(8, 6), node.font = c("Helvetica", 14),
edge.font = c("Helvetica", 10), rank.direction=c("RL","TB","LR","BT"),
digits = 2, title = "ICLUST",label.font=2, ...)
Arguments
ic.results output list from ICLUST
out.file File name to save optional dot code.
min.size draw a smaller node (without all the information) for clusters < min.size – useful
for large problems
short if short==TRUE, don’t use variable names
labels vector of text labels (contents) for the variables
size size of output
node.font Font to use for nodes in the graph
edge.font Font to use for the labels of the arrows (edges)
rank.direction LR or TB or RL
digits number of digits to show
title any title
label.font The variable labels can be a different size than the other nodes. This is particu-
larly helpful if the number of variables is large or the labels are long.
... other options to pass
Details
Will create (or overwrite) an output file and print out the dot code to show a cluster structure. This
dot file may be imported directly into a dot viewer (e.g., http://www.graphviz.org/). The "dot"
language is a powerful graphic description language that is particulary appropriate for viewing
cluster output. Commercial graphics programs (e.g., OmniGraffle) can also read (and clean up) dot
files.
ICLUST.rgraph takes the output from ICLUST results and processes it to provide a pretty picture
of the results. Original variables shown as rectangles and ordered on the left hand side (if rank
direction is RL) of the graph. Clusters are drawn as ellipses and include the alpha, beta, and size of
the cluster. Edges show the cluster intercorrelations.
It is possible to trim the output to not show all cluster information. Clusters < min.size are shown
as small ovals without alpha, beta, and size information.
178 ICLUST.sort
Value
Output is a set of dot commands written either to console or to the output file. These commands
may then be used as input to any "dot" viewer, e.g., Graphviz.
ICLUST.rgraph is a version of ICLUST.graph that uses Rgraphviz to draw on the screen as well.
Additional output is drawn to main graphics screen.
Note
Requires Rgraphviz
Author(s)
<revelle@northwestern.edu >
http://personality-project.org/revelle.html
References
ICLUST: http://personality-project.org/r/r.ICLUST.html
See Also
VSS.plot,ICLUST
Examples
test.data <- Harman74.cor$cov
ic.out <- ICLUST(test.data) #uses iclust.diagram instead
ICLUST.sort Sort items by absolute size of cluster loadings
Description
Given a cluster analysis or factor analysis loadings matrix, sort the items by the (absolute) size of
each column of loadings. Used as part of ICLUST and SAPA analyses. The columns are rearranged
by the
Usage
ICLUST.sort(ic.load, cut = 0, labels = NULL,keys=FALSE,clustsort=TRUE)
ICLUST.sort 179
Arguments
ic.load The output from a factor or principal components analysis, or from ICLUST, or
a matrix of loadings.
cut Do not include items in clusters with absolute loadings less than cut
labels labels for each item.
keys should cluster keys be returned? Useful if clusters scales are to be scored.
clustsort TRUE will will sort the clusters by their eigenvalues
Details
When interpreting cluster or factor analysis outputs, is is useful to group the items in terms of
which items have their biggest loading on each factor/cluster and then to sort the items by size of
the absolute factor loading.
A stable cluster solution will be one in which the output of these cluster definitions does not vary
when clusters are formed from the clusters so defined.
With the keys=TRUE option, the resulting cluster keys may be used to score the original data or the
correlation matrix to form clusters from the factors.
Value
sorted A data.frame of item numbers, item contents, and item x factor loadings.
cluster A matrix of -1, 0, 1s defining each item by the factor/cluster with the row wise
largest absolute loading.
...
Note
Although part of the ICLUST set of programs, this is also more useful for factor or principal com-
ponents analysis.
Author(s)
William Revelle
References
http://personality-project.org/r/r.ICLUST.html
See Also
ICLUST.graph,ICLUST.cluster,cluster.fit ,VSS,factor2cluster
180 income
income US family income from US census 2008
Description
US census data on family income from 2008
Usage
data(income)
Format
A data frame with 44 observations on the following 4 variables.
value lower boundary of the income group
count Number of families within that income group
mean Mean of the category
prop proportion of families
Details
The distribution of income is a nice example of a log normal distribution. It is also an interesting
example of the power of graphics. It is quite clear when graphing the data that income statistics are
bunched to the nearest 5K. That is, there is a clear sawtooth pattern in the data.
The all.income set is interpolates intervening values for 100-150K, 150-200K and 200-250K
Source
US Census: Table HINC-06. Income Distribution to $250,000 or More for Households: 2008
http://www.census.gov/hhes/www/cpstables/032009/hhinc/new06_000.htm
Examples
data(income)
with(income[1:40,], plot(mean,prop, main="US family income for 2008",xlab="income",
ylab="Proportion of families",xlim=c(0,100000)))
with (income[1:40,], points(lowess(mean,prop,f=.3),typ="l"))
describe(income)
with(all.income, plot(mean,prop, main="US family income for 2008",xlab="income",
ylab="Proportion of families",xlim=c(0,250000)))
with (all.income[1:50,], points(lowess(mean,prop,f=.25),typ="l"))
#curve(100000* dlnorm(x, 10.8, .8), x = c(0,250000),ylab="Proportion")
interp.median 181
interp.median Find the interpolated sample median, quartiles, or specific quantiles
for a vector, matrix, or data frame
Description
For data with a limited number of response categories (e.g., attitude items), it is useful treat each
response category as range with width, w and linearly interpolate the median, quartiles, or any
quantile value within the median response.
Usage
interp.median(x, w = 1,na.rm=TRUE)
interp.quantiles(x, q = .5, w = 1,na.rm=TRUE)
interp.quartiles(x,w=1,na.rm=TRUE)
interp.boxplot(x,w=1,na.rm=TRUE)
interp.values(x,w=1,na.rm=TRUE)
interp.qplot.by(y,x,w=1,na.rm=TRUE,xlab="group",ylab="dependent",
ylim=NULL,arrow.len=.05,typ="b",add=FALSE,...)
Arguments
xinput vector
qquantile to estimate ( 0 < q < 1
wcategory width
yinput vector for interp.qplot.by
na.rm should missing values be removed
xlab x label
ylab Y label
ylim limits for the y axis
arrow.len length of arrow in interp.qplot.by
typ plot type in interp.qplot.by
add add the plot or not
... additional parameters to plotting function
Details
If the total number of responses is N, with median, M, and the number of responses at the median
value, Nm >1, and Nb= the number of responses less than the median, then with the assumption
that the responses are distributed uniformly within the category, the interpolated median is M - .5w
+ w*(N/2 - Nb)/Nm.
The generalization to 1st, 2nd and 3rd quartiles as well as the general quantiles is straightforward.
A somewhat different generalization allows for graphic presentation of the difference between in-
terpolated and non-interpolated points. This uses the interp.values function.
If the input is a matrix or data frame, quantiles are reported for each variable.
182 iqitems
Value
im interpolated median(quantile)
vinterpolated values for all data points
See Also
median
Examples
interp.median(c(1,2,3,3,3)) # compare with median = 3
interp.median(c(1,2,2,5))
interp.quantiles(c(1,2,2,5),.25)
x <- sample(10,100,TRUE)
interp.quartiles(x)
#
x <- c(1,1,2,2,2,3,3,3,3,4,5,1,1,1,2,2,3,3,3,3,4,5,1,1,1,2,2,3,3,3,3,4,2)
y <- c(1,2,3,3,3,3,4,4,4,4,4,1,2,3,3,3,3,4,4,4,4,5,1,5,3,3,3,3,4,4,4,4,4)
x <- x[order(x)] #sort the data by ascending order to make it clearer
y <- y[order(y)]
xv <- interp.values(x)
yv <- interp.values(y)
barplot(x,space=0,xlab="ordinal position",ylab="value")
lines(1:length(x)-.5,xv)
points(c(length(x)/4,length(x)/2,3*length(x)/4),interp.quartiles(x))
barplot(y,space=0,xlab="ordinal position",ylab="value")
lines(1:length(y)-.5,yv)
points(c(length(y)/4,length(y)/2,3*length(y)/4),interp.quartiles(y))
data(galton)
interp.median(galton)
interp.qplot.by(galton$child,galton$parent,ylab="child height"
,xlab="Mid parent height")
iqitems 16 multiple choice IQ items
Description
16 multiple choice ability items taken from the Synthetic Aperture Personality Assessment (SAPA)
web based personality assessment project. The data from 1525 subjects are included here as a
demonstration set for scoring multiple choice inventories and doing basic item statistics. For more
information on the development of an open source measure of cognitive ability, consult the readings
available at the personality-project.org.
Usage
data(iqitems)
iqitems 183
Format
A data frame with 1525 observations on the following 16 variables. The number following the name
is the item number from SAPA.
reason.4 Basic reasoning questions
reason.16 Basic reasoning question
reason.17 Basic reasoning question
reason.19 Basic reasoning question
letter.7 In the following alphanumeric series, what letter comes next?
letter.33 In the following alphanumeric series, what letter comes next?
letter.34 In the following alphanumeric series, what letter comes next
letter.58 In the following alphanumeric series, what letter comes next?
matrix.45 A matrix reasoning task
matrix.46 A matrix reasoning task
matrix.47 A matrix reasoning task
matrix.55 A matrix reasoning task
rotate.3 Spatial Rotation of type 1.2
rotate.4 Spatial Rotation of type 1.2
rotate.6 Spatial Rotation of type 1.1
rotate.8 Spatial Rotation of type 2.3
Details
16 items were sampled from 80 items given as part of the SAPA (http://sapa-project.org)
project (Revelle, Wilt and Rosenthal, 2009; Condon and Revelle, 2014) to develop online measures
of ability. These 16 items reflect four lower order factors (verbal reasoning, letter series, matrix
reasoning, and spatial rotations. These lower level factors all share a higher level factor (’g’).
This data set and the associated data set (ability based upon scoring these multiple choice items
and converting them to correct/incorrect may be used to demonstrate item response functions,
tetrachoric correlations, or irt.fa as well as omega estimates of of reliability and hierarchi-
cal structure.
In addition, the data set is a good example of doing item analysis to examine the empirical response
probabilities of each item alternative as a function of the underlying latent trait. When doing this,
it appears that two of the matrix reasoning problems do not have monotonically increasing trace
lines for the probability correct. At moderately high ability (theta = 1) there is a decrease in the
probability correct from theta = 0 and theta = 2.
Source
The example data set is taken from the Synthetic Aperture Personality Assessment personality
and ability test at http://sapa-project.org. The data were collected with David Condon from
8/08/12 to 8/31/12.
184 irt.1p
References
Revelle, William, Wilt, Joshua, and Rosenthal, Allen (2010) Personality and Cognition: The Personality-
Cognition Link. In Gruszka, Alexandra and Matthews, Gerald and Szymura, Blazej (Eds.) Hand-
book of Individual Differences in Cognition: Attention, Memory and Executive Control, Springer.
Condon, David and Revelle, William, (2014) The International Cognitive Ability Resource: Devel-
opment and initial validation of a public-domain measure. Intelligence, 43, 52-64.
Examples
## Not run:
data(iqitems)
iq.keys <- c(4,4,4, 6, 6,3,4,4, 5,2,2,4, 3,2,6,7)
score.multiple.choice(iq.keys,iqitems) #this just gives summary statisics
#convert them to true false
iq.scrub <- scrub(iqitems,isvalue=0) #first get rid of the zero responses
iq.tf <- score.multiple.choice(iq.keys,iq.scrub,score=FALSE)
#convert to wrong (0) and correct (1) for analysis
describe(iq.tf)
#see the ability data set for these analyses
#now, for some item analysis
#iq.irt <- irt.fa(iq.tf) #do a basic irt
#iq.sc <-score.irt(iq.irt,iq.tf) #find the scores
#op <- par(mfrow=c(4,4))
#irt.responses(iq.sc[,1], iq.tf)
#op <- par(mfrow=c(1,1))
## End(Not run)
irt.1p Item Response Theory estimate of theta (ability) using a Rasch (like)
model
Description
Item Response Theory models individual responses to items by estimating individual ability (theta)
and item difficulty (diff) parameters. This is an early and crude attempt to capture this modeling
procedure. A better procedure is to use irt.fa.
Usage
irt.person.rasch(diff, items)
irt.0p(items)
irt.1p(delta,items)
irt.2p(delta,beta,items)
irt.1p 185
Arguments
diff A vector of item difficulties –probably taken from irt.item.diff.rasch
items A matrix of 0,1 items nrows = number of subjects, ncols = number of items
delta delta is the same as diff and is the item difficulty parameter
beta beta is the item discrimination parameter found in irt.discrim
Details
A very preliminary IRT estimation procedure. Given scores xij for ith individual on jth item
Classical Test Theory ignores item difficulty and defines ability as expected score : abilityi = theta(i)
= x(i.) A zero parameter model rescales these mean scores from 0 to 1 to a quasi logistic scale
ranging from - 4 to 4 This is merely a non-linear transform of the raw data to reflect a logistic
mapping.
Basic 1 parameter (Rasch) model considers item difficulties (delta j): p(correct on item j for the ith
subject |theta i, deltaj) = 1/(1+exp(deltaj - thetai)) If we have estimates of item difficulty (delta),
then we can find theta i by optimization
Two parameter model adds item sensitivity (beta j): p(correct on item j for subject i |thetai, deltaj,
betaj) = 1/(1+exp(betaj *(deltaj- theta i))) Estimate delta, beta, and theta to maximize fit of model
to data.
The procedure used here is to first find the item difficulties assuming theta = 0 Then find theta given
those deltas Then find beta given delta and theta.
This is not an "official" way to do IRT, but is useful for basic item development. See irt.fa and
score.irt for far better options.
Value
a data.frame with estimated ability (theta) and quality of fit. (for irt.person.rasch)
a data.frame with the raw means, theta0, and the number of items completed
Note
Not recommended for serious use. This code is under development. Much better functions are in
the ltm and eRm packages. Similar analyses can be done using irt.fa and score.irt.
Author(s)
William Revelle
See Also
sim.irt,sim.rasch,logistic,irt.fa,tetrachoric,irt.item.diff.rasch
186 irt.fa
irt.fa Item Response Analysis by Exploratory Factor Analysis of tetra-
choric/polychoric correlations
Description
Although exploratory factor analysis and Item Response Theory seem to be very different models
of binary data, they can provide equivalent parameter estimates of item difficulty and item discrim-
ination. Tetrachoric or polychoric correlations of a data set of dichotomous or polytomous items
may be factor analysed using a minimum residual or maximum likelihood factor analysis and the
result loadings transformed to item discrimination parameters. The tau parameter from the tetra-
choric/polychoric correlations combined with the item factor loading may be used to estimate item
difficulties.
Usage
irt.fa(x,nfactors=1,correct=TRUE,plot=TRUE,n.obs=NULL,rotate="oblimin",fm="minres",
sort=TRUE,...)
irt.select(x,y)
fa2irt(f,rho,plot=TRUE,n.obs=NULL)
Arguments
xA data matrix of dichotomous or discrete items, or the result of tetrachoric or
polychoric
nfactors Defaults to 1 factor
correct If true, then correct the tetrachoric correlations for continuity. (See tetrachoric).
plot If TRUE, automatically call the plot.irt or plot.poly functions.
ythe subset of variables to pick from the rho and tau output of a previous irt.fa
analysis to allow for further analysis.
n.obs The number of subjects used in the initial analysis if doing a second analysis of
a correlation matrix. In particular, if using the fm="minchi" option, this should
be the matrix returned by count.pairwise.
rotate The default rotation is oblimin. See fa for the other options.
fm The default factor extraction is minres. See fa for the other options.
fThe object returned from fa
rho The object returned from polychoric or tetrachoric. This will include both
a correlation matrix and the item difficulty levels.
sort Should the factor loadings be sorted before preparing the item information ta-
bles. Defaults to TRUE as this is more useful for tabular output.
... Additional parameters to pass to the factor analysis function
irt.fa 187
Details
irt.fa combines several functions into one to make the process of item response analysis easier.
Correlations are found using either tetrachoric or polychoric. Exploratory factor analyeses with
all the normal options are then done using fa. The results are then organized to be reported in terms
of IRT parameters (difficulties and discriminations) as well as the more conventional factor analysis
output. In addition, because the correlation step is somewhat slow, reanalyses may be done using
the correlation matrix found in the first step. In this case, if it is desired to use the fm="minchi"
factoring method, the number of observations needs to be specified as the matrix resulting from
count.pairwise.
The tetrachoric correlation matrix of dichotomous items may be factored using a (e.g.) minimum
residual factor analysis function fa and the resulting loadings, λiare transformed to discriminations
by α=λi
√1−λ2
i
.
The difficulty parameter, δis found from the τparameter of the tetrachoric or polychoric
function.
δi=τi
√1−λ2
i
Similar analyses may be done with discrete item responses using polychoric correlations and distinct
estimates of item difficulty (location) for each item response.
The results may be shown graphically using link{plot.irt} for dichotomous items or link{plot.poly}
for polytomous items. These called by plotting the irt.fa output, see the examples). For plotting there
are three options: type = "ICC" will plot the item characteristic response function. type = "IIC" will
plot the item information function, and type= "test" will plot the test information function. Invisible
output from the plot function will return tables of item information as a function of several levels of
the trait, as well as the standard error of measurement and the reliability at each of those levels.
The normal input is just the raw data. If, however, the correlation matrix has already been found
using tetrachoric,polychoric, or a previous analysis using irt.fa then that result can be pro-
cessed directly. Because irt.fa saves the rho and tau matrices from the analysis, subsequent
analyses of the same data set are much faster if the input is the object returned on the first run. A
similar feature is available in omega.
The output is best seen in terms of graphic displays. Plot the output from irt.fa to see item and test
information functions.
The print function will print the item location and discriminations. The additional factor analysis
output is available as an object in the output and may be printed directly by specifying the $fa
object.
The irt.select function is a helper function to allow for selecting a subset of a prior analysis for
further analysis. First run irt.fa, then select a subset of variables to be analyzed in a subsequent irt.fa
analysis. Perhaps a better approach is to just plot and find the information for selected items.
The plot function for an irt.fa object will plot ICC (item characteristic curves), IIC (item information
curves), or test information curves. In addition, by using the "keys" option, these three kinds of
plots can be done for selected items. This is particularly useful when trying to see the information
characteristics of short forms of tests based upon the longer form factor analysis.
The plot function will also return (invisibly) the informaton at multiple levels of the trait, the average
information (area under the curve) as well as the location of the peak information for each item.
These may be then printed or printed in sorted order using the sort option in print.
188 irt.fa
Value
irt A list of Item location (difficulty) and discrimination
fa A list of statistics for the factor analyis
rho The tetrachoric/polychoric correlation matrix
tau The tetrachoric/polychoric cut points
Note
In comparing irt.fa to the ltm function in the ltm package or to the analysis reported in Kamata and
Bauer (2008) the discrimination parameters are not identical, because the irt.fa reports them in units
of the normal curve while ltm and Kamata and Bauer report them in logistic units. In addition,
Kamata and Bauer do their factor analysis using a logistic error model. Their results match the irt.fa
results (to the 2nd or 3rd decimal) when examining their analyses using a normal model. (With
thanks to Akihito Kamata for sharing that analysis.)
irt.fa reports parameters in normal units. To convert them to conventional IRT parameters, mul-
tiply by 1.702. In addition, the location parameter is expressed in terms of difficulty (high positive
scores imply lower frequency of response.)
The results of irt.fa can be used by score.irt for irt based scoring. First run irt.fa and then
score the results using a two parameter model using score.irt.
Author(s)
William Revelle
References
Kamata, Akihito and Bauer, Daniel J. (2008) A Note on the Relation Between Factor Analytic and
Item Response Theory Models Structural Equation Modeling, 15 (1) 136-153.
McDonald, Roderick P. (1999) Test theory: A unified treatment. L. Erlbaum Associates.
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
See Also
fa,sim.irt,tetrachoric,polychoric as well as plot.psych for plotting the IRT item curves.
See also score.irt for scoring items based upon these parameter estimates. irt.responses will
plot the empirical response curves for the alternative response choices for multiple choice items.
Examples
## Not run:
set.seed(17)
d9 <- sim.irt(9,1000,-2.5,2.5,mod="normal") #dichotomous items
test <- irt.fa(d9$items)
test
op <- par(mfrow=c(3,1))
plot(test,type="ICC")
irt.fa 189
plot(test,type="IIC")
plot(test,type="test")
par(op)
set.seed(17)
items <- sim.congeneric(N=500,short=FALSE,categorical=TRUE) #500 responses to 4 discrete items
d4 <- irt.fa(items$observed) #item response analysis of congeneric measures
d4 #show just the irt output
d4$fa #show just the factor analysis output
op <- par(mfrow=c(2,2))
plot(d4,type="ICC")
par(op)
#using the iq data set for an example of real items
#first need to convert the responses to tf
data(iqitems)
iq.keys <- c(4,4,4, 6, 6,3,4,4, 5,2,2,4, 3,2,6,7)
iq.tf <- score.multiple.choice(iq.keys,iqitems,score=FALSE) #just the responses
iq.irt <- irt.fa(iq.tf)
print(iq.irt,short=FALSE) #show the IRT as well as factor analysis output
p.iq <- plot(iq.irt) #save the invisible summary table
p.iq #show the summary table of information by ability level
#select a subset of these variables
small.iq.irt <- irt.select(iq.irt,c(1,5,9,10,11,13))
small.irt <- irt.fa(small.iq.irt)
plot(small.irt)
#find the information for three subset of iq items
keys <- make.keys(16,list(all=1:16,some=c(1,5,9,10,11,13),others=c(1:5)))
plot(iq.irt,keys=keys)
## End(Not run)
#compare output to the ltm package or Kamata and Bauer -- these are in logistic units
ls <- irt.fa(lsat6)
#library(ltm)
# lsat.ltm <- ltm(lsat6~z1)
# round(coefficients(lsat.ltm)/1.702,3) #convert to normal (approximation)
#
# Dffclt Dscrmn
#Q1 -1.974 0.485
#Q2 -0.805 0.425
#Q3 -0.164 0.523
#Q4 -1.096 0.405
#Q5 -1.835 0.386
#Normal results ("Standardized and Marginal")(from Akihito Kamata )
#Item discrim tau
# 1 0.4169 -1.5520
# 2 0.4333 -0.5999
# 3 0.5373 -0.1512
190 irt.item.diff.rasch
# 4 0.4044 -0.7723
# 5 0.3587 -1.1966
#compare to ls
#Normal results ("Standardized and conditional") (from Akihito Kamata )
#item discrim tau
# 1 0.3848 -1.4325
# 2 0.3976 -0.5505
# 3 0.4733 -0.1332
# 4 0.3749 -0.7159
# 5 0.3377 -1.1264
#compare to ls$fa and ls$tau
#Kamata and Bauer (2008) logistic estimates
#1 0.826 2.773
#2 0.723 0.990
#3 0.891 0.249
#4 0.688 1.285
#5 0.657 2.053
irt.item.diff.rasch Simple function to estimate item difficulties using IRT concepts
Description
Steps toward a very crude and preliminary IRT program. These two functions estimate item diffi-
culty and discrimination parameters. A better procedure is to use irt.fa or the ltm package.
Usage
irt.item.diff.rasch(items)
irt.discrim(item.diff,theta,items)
Arguments
items a matrix of items
item.diff a vector of item difficulties (found by irt.item.diff)
theta ability estimate from irt.person.theta
irt.responses 191
Details
Item Response Theory (aka "The new psychometrics") models individual responses to items with a
logistic function and an individual (theta) and item difficulty (diff) parameter.
irt.item.diff.rasch finds item difficulties with the assumption of theta=0 for all subjects and that all
items are equally discriminating.
irt.discrim takes those difficulties and theta estimates from irt.person.rasch to find item discrim-
ination (beta) parameters.
A far better package with these features is the ltm package. The IRT functions in the psych-package
are for pedagogical rather than production purposes. They are believed to be accurate, but are not
guaranteed. They do seem to be slightly more robust to missing data structures associated with
SAPA data sets than the ltm package.
The irt.fa function is also an alternative. This will find tetrachoric or polychoric correlations
and then convert to IRT parameters using factor analysis (fa).
Value
a vector of item difficulties or item discriminations.
Note
Under development. Not recommended for public consumption. See irt.fa and score.irt for
far better options.
Author(s)
William Revelle
See Also
irt.fa,irt.person.rasch
irt.responses Plot probability of multiple choice responses as a function of a latent
trait
Description
When analyzing ability tests, it is important to consider how the distractor alternatives vary as a
function of the latent trait. The simple graphical solution is to plot response endorsement frequen-
cies against the values of the latent trait found from multiple items. A good item is one in which the
probability of the distractors decrease and the keyed answer increases as the latent trait increases.
Usage
irt.responses(theta,items, breaks = 11,show.missing=FALSE, show.legend=TRUE,
legend.location="topleft", colors=NULL,...)
192 irt.responses
Arguments
theta The estimated latent trait (found, for example by using score.irt).
items A matrix or data frame of the multiple choice item responses.
breaks The number of levels of the theta to use to form the probability estimates. May
be increased if there are enough cases.
show.legend Show the legend
show.missing For some SAPA data sets, there are a very large number of missing responses.
In general, we do not want to show their frequency.
legend.location
Choose among c("bottomright", "bottom", "bottomleft", "left", "topleft", "top",
"topright", "right", "center","none"). The default is "topleft".
colors if NULL, then use the default colors, otherwise, specify the color choices. The
basic color palette is c("black", "blue", "red", "darkgreen", "gold2", "gray50",
"cornflowerblue", "mediumorchid2").
... Other parameters for plots and points
Details
This function is a convenient way to analyze the quality of item alternatives in a multiple choice
ability test. The typical use is to first score the test (using, e.g., score.multiple.choice according
to some scoring key and to then find the score.irt based scores. Response frequencies for each
alternative are then plotted against total score. An ideal item is one in which just one alternative
(the correct one) has a monotonically increasing response probability.
Because of the similar pattern of results for IRT based or simple sum based item scoring, the func-
tion can be run on scores calculated either by score.irt or by score.multiple.choice. In the
latter case, the number of breaks should not exceed the number of possible score alternatives.
Value
Graphic output
Author(s)
William Revelle
References
Revelle, W. An introduction to psychometric theory with applications in R (in prep) Springer. Draft
chapters available at http://personality-project.org/r/book/
See Also
score.multiple.choice,score.irt
kaiser 193
Examples
data(iqitems)
iq.keys <- c(4,4,4, 6,6,3,4,4, 5,2,2,4, 3,2,6,7)
scores <- score.multiple.choice(iq.keys,iqitems,score=TRUE,short=FALSE)
#note that for speed we can just do this on simple item counts rather
# than IRT based scores.
op <- par(mfrow=c(2,2)) #set this to see the output for multiple items
irt.responses(scores$scores,iqitems[1:4],breaks=11)
op <- par(op)
kaiser Apply the Kaiser normalization when rotating factors
Description
Kaiser (1958) suggested normalizing factor loadings before rotating them, and then denormalizing
them after rotation. The GPArotation package does not (by default) normalize, nor does the fa
function. Then, to make it more confusing, varimax in stats does,Varimax in GPArotation does not.
kaiser will take the output of a non-normalized solution and report the normalized solution.
Usage
kaiser(f, rotate = "oblimin")
Arguments
fA factor analysis output from fa or a factor loading matrix.
rotate Any of the standard rotations avaialable in the GPArotation package.
Details
Best results if called from an unrotated solution. Repeated calls using a rotated solution will produce
incorrect estimates of the correlations between the factors.
Value
See the values returned by GPArotation functions
Note
Prepared in response to a question about why fa oblimin results are different from SPSS.
Author(s)
William Revelle
194 KMO
References
Kaiser, H. F. (1958) The varimax criterion for analytic rotation in factor analysis. Psychometrika
23, 187-200.
See Also
fa
Examples
f3 <- fa(Thurstone,3)
f3n <- kaiser(fa(Thurstone,3,rotate="none"))
factor.congruence(f3,f3n)
KMO Find the Kaiser, Meyer, Olkin Measure of Sampling Adequacy
Description
Henry Kaiser (1970) introduced an Measure of Sampling Adequacy (MSA) of factor analytic data
matrices. Kaiser and Rice (1974) then modified it. This is just a function of the squared elements of
the ‘image’ matrix compared to the squares of the original correlations. The overall MSA as well
as estimates for each item are found. The index is known as the Kaiser-Meyer-Olkin (KMO) index.
Usage
KMO(r)
Arguments
rA correlation matrix or a data matrix (correlations will be found)
Details
Let S2=diag(R−1)−1and Q=SR−1S. Then Q is said to the be the anti-image intercorrelation
matrix. Let sumr2 = PR2and sumq2 = PQ2for all off diagonal elements of R and Q, then
SMA =sumr2)/(sumr2 + sumq2). Although originally MSA was 1 - sumq2/sumr2 (Kaiser,
1970), this was modified in Kaiser and Rice, (1974) to be SM A =sumr2)/(sumr2 + sumq2).
This is the formula used by Dziuban and Shirkey (1974) and by SPSS.
Value
• MSAThe overall Measure of Sampling Adequacy
• MSAiThe measure of sampling adequacy for each item itemImageThe Image correlation ma-
trix (Q)
logistic 195
Author(s)
William Revelle
References
H.~F. Kaiser. (1970) A second generation little jiffy. Psychometrika, 35(4):401–415.
H.~F. Kaiser and J.~Rice. (1974) Little jiffy, mark iv. Educational and Psychological Measurement,
34(1):111–117.
Dziuban, Charles D. and Shirkey, Edwin C. (1974) When is a correlation matrix appropriate for
factor analysis? Some decision rules. Psychological Bulletin, 81 (6) 358 - 361.
See Also
See Also as fa,Harman.political.
Examples
KMO(Thurstone)
KMO(Harman.political) #compare to the results in Dziuban and Shirkey (1974)
logistic Logistic transform from x to p and logit transform from p to x
Description
The logistic function (1/(1+exp(-x)) and logit function (log(p/(1-p)) are fundamental to Item Re-
sponse Theory. Although just one line functions, they are included here for ease of demonstrations
and in drawing IRT models. Also included is the logistic.grm for a graded response model.
Usage
logistic(x,d=0, a=1,c=0, z=1)
logit(p)
logistic.grm( x,d=0,a=1.5,c=0,z=1,r=2,s=c(-1.5,-.5,.5,1.5))
Arguments
xAny integer or real value
dItem difficulty or delta parameter
aThe slope of the curve at x=0 is equivalent to the discrimination parameter in
2PL models or alpha parameter. Is either 1 in 1PL or 1.702 in 1PN approxima-
tions.
cLower asymptote = guessing parameter in 3PL models or gamma
zThe upper asymptote — in 4PL models
196 logistic
pProbability to be converted to logit value
rThe response category for the graded response model
sThe response thresholds
Details
These three functions are provided as simple helper functions for demonstrations of Item Response
Theory. The one parameter logistic (1PL) model is also known as the Rasch model. It assumes items
differ only in difficulty. 1PL, 2PL, 3PL and 4PL curves may be drawn by choosing the appropriate
d (delta or item difficulty), a (discrimination or slope), c (gamma or guessing) and z (zeta or upper
asymptote).
logit is just the inverse of logistic.
logistic.grm will create the responses for a graded response model for the rth category where cut-
points are in s.
Value
plogistic returns the probability associated with x
xlogit returns the real number associated with p
Author(s)
William Revelle
Examples
curve(logistic(x,a=1.702),-3,3,ylab="Probability of x",
main="Logistic transform of x",xlab="z score units")
#logistic with a=1.702 is almost the same as pnorm
curve(pnorm(x),add=TRUE,lty="dashed")
curve(logistic(x),add=TRUE)
text(2,.8, expression(alpha ==1))
text(2,1.0,expression(alpha==1.7))
curve(logistic(x),-4,4,ylab="Probability of x",
main = "Logistic transform of x in logit units",xlab="logits")
curve(logistic(x,d=-1),add=TRUE)
curve(logistic(x,d=1),add=TRUE)
curve(logistic(x,c=.2),add=TRUE,lty="dashed")
text(1.3,.5,"d=1")
text(.3,.5,"d=0")
text(-1.5,.5,"d=-1")
text(-3,.3,"c=.2")
#demo of graded response model
curve(logistic.grm(x,r=1),-4,4,ylim=c(0,1),main="Five level response scale",
ylab="Probability of endorsement",xlab="Latent attribute on logit scale")
curve(logistic.grm(x,r=2),add=TRUE)
curve(logistic.grm(x,r=3),add=TRUE)
curve(logistic.grm(x,r=4),add=TRUE)
curve(logistic.grm(x,r=5),add=TRUE)
lowerUpper 197
text(-2.,.5,1)
text(-1.,.4,2)
text(0,.4,3)
text(1.,.4,4)
text(2.,.4,5)
lowerUpper Combine two square matrices to have a lower off diagonal for one,
upper off diagonal for the other
Description
When reporting correlation matrices for two samples (e.g., males and females), it is convenient to
show them as one matrix, with entries below the diagonal representing one matrix, and entries above
the diagonal the other matrix. It is also useful to compare a correlation matrix with the residuals
from a fitted (e.g., factor) model.
Usage
lowerUpper(lower, upper=NULL, diff=FALSE)
Arguments
lower A square matrix
upper A square matrix of the same size as the first (if omitted, then the matrix is con-
verted to two symmetric matrices).
diff Find the difference between the first and second matrix and put the results in the
above the diagonal entries.
Details
If just one matrix is provided (i.e., upper is missing), it is decomposed into two square matrices,
one equal to the lower off diagonal entries, the other to the upper off diagonal entries. In the normal
case two symmetric matrices are provided and combined into one non-symmetric matrix with the
lower off diagonals representing the lower matrix and the upper off diagonals representing the upper
matrix.
If diff is true, the upper off diagonal matrix reflects the differences between the two matrices.
Value
Either one matrix or a list of two
Author(s)
William Revelle
198 make.keys
See Also
read.clipboard.lower,cor.plot
Examples
b1 <- Bechtoldt.1
b2 <- Bechtoldt.2
b12 <- lowerUpper(b1,b2)
cor.plot(b12)
diff12 <- lowerUpper(b1,b2,diff=TRUE)
cor.plot(t(diff12),numbers=TRUE,main="Bechtoldt1 and the differences from Bechtoldt2")
make.keys Create a keys matrix for use by score.items or cluster.cor
Description
When scoring items by forming composite scales either from the raw data using score.items or
from the correlation matrix using cluster.cor, it is necessary to create a keys matrix. This is
just a short cut for doing so. The keys matrix is a nvar x nscales matrix of -1,0, 1 and defines the
membership for each scale. Items can be specified by location or by name.
Usage
make.keys(nvars, keys.list, item.labels = NULL, key.labels = NULL)
Arguments
nvars Number of variables items to be scored
keys.list A list of the scoring keys,one element for each scale
item.labels Typically, just the colnames of the items data matrix.
key.labels Labels for the scales can be specified here, or in the key.list
Details
There are two ways to create keys for the scoreItems and scoreOverlap functions. One is to
laboriously do it in a spreadsheet and then copy them into R. The other is to just specify them by
item number in a list. Make keys allows one to specify items by name or by location or a mixture
of both.
To address items by name it is necessary to specify item names, either by using the item.labels
value, or by putting the name of the data file or the colnames of the data file to be scored into the
first (nvars) position.
If specifying by number, then nvars is the total number of items in the object to be scored, not just
the number of items used.
make.keys 199
See the examples for the various options.
Note that make.keys was revised in Sept, 2013 to allow for keying by name.
It is also possible to do several make.keys operations and then combine them using superMatrix.
Value
keys a nvars x nkeys matrix of -1, 0, or 1s describing how to score each scale. nkeys
is the length of the keys.list
See Also
scoreItems,scoreOverlap,cluster.cor superMatrix
Examples
data(attitude) #specify the items by location
key.list <- list(all=c(1,2,3,4,-5,6,7),
first=c(1,2,3),
last=c(4,5,6,7))
keys <- make.keys(7,key.list,item.labels = colnames(attitude))
keys
#scores <- score.items(keys,attitude)
#scores
data(bfi)
#first create the keys by location (the conventional way)
keys.list <- list(agree=c(-1,2:5),conscientious=c(6:8,-9,-10),
extraversion=c(-11,-12,13:15),neuroticism=c(16:20),openness = c(21,-22,23,24,-25))
keys <- make.keys(25,keys.list,item.labels=colnames(bfi)[1:25])
#alternatively, create by a mixture of names and locations
keys.list <- list(agree=c("-A1","A2","A3","A4","A5"),
conscientious=c("C1","C2","C2","-C4","-C5"),extraversion=c("-E1","-E2","E3","E4","E5"),
neuroticism=c(16:20),openness = c(21,-22,23,24,-25))
keys <- make.keys(bfi,keys.list) #specify the data file to be scored (bfi)
#or
keys <- make.keys(colnames(bfi),keys.list) #specify the names of the variables to be used
#or
#specify the number of variables to be scored and their names in all cases
keys <- make.keys(28,keys.list,colnames(bfi))
scores <- score.items(keys,bfi)
summary(scores)
200 mardia
mardia Calculate univariate or multivariate (Mardia’s test) skew and kurtosis
for a vector, matrix, or data.frame
Description
Find the skew and kurtosis for each variable in a data.frame or matrix. Unlike skew and kurtosis in
e1071, this calculates a different skew for each variable or column of a data.frame/matrix. mardia
applies Mardia’s tests for multivariate skew and kurtosis
Usage
skew(x, na.rm = TRUE,type=3)
kurtosi(x, na.rm = TRUE,type=3)
mardia(x,na.rm = TRUE,plot=TRUE)
Arguments
xA data.frame or matrix
na.rm how to treat missing data
type See the discussion in describe.
plot Plot the expected normal distribution values versus the Mahalanobis distance of
the subjects.
Details
given a matrix or data.frame x, find the skew or kurtosis for each column (for skew and kurtosis) or
the multivariate skew and kurtosis in the case of mardia.
As of version 1.2.3,when finding the skew and the kurtosis, there are three different options avail-
able. These match the choices available in skewness and kurtosis found in the e1071 package (see
Joanes and Gill (1998) for the advantages of each one).
If we define mr= [P(X−mx)r]/n then
Type 1 finds skewness and kurtosis by g1=m3/(m2)3/2and g2=m4/(m2)2−3.
Type 2 is G1 = g1∗pn∗(n−1)/(n−2) and G2=(n−1) ∗[(n+ 1)g2 + 6]/((n−2)(n−3)).
Type 3 is b1 = [(n−1)/n]3/2m3/m3/2
2and b2 = [(n−1)/n]3/2m4/m2
2).
For consistency with e1071 and with the Joanes and Gill, the types are now defined as above.
However, from revision 1.0.93 to 1.2.3, kurtosi by default gives an unbiased estimate of the kurtosis
(DeCarlo, 1997). Prior versions used a different equation which produced a biased estimate. (See
the kurtosis function in the e1071 package for the distinction between these two formulae. The
default, type 1 gave what is called type 2 in e1071. The other is their type 3.) For comparison
with previous releases, specifying type = 2 will give the old estimate. These type numbers are now
changed.
mardia 201
Value
skew if input is a matrix or data.frame, skew is a vector of skews
kurtosi if input is a matrix or data.frame, kurtosi is a vector of kurtosi
bp1 Mardia’s bp1 estimate of multivariate skew
bp2 Mardia’s bp2 estimate of multivariate kurtosis
skew Mardia’s skew statistic
small.skew Mardia’s small sample skew statistic
p.skew Probability of skew
p.small Probability of small.skew
kurtosis Mardia’s multivariate kurtosis statistic
p.kurtosis Probability of kurtosis statistic
DMahalanobis distance of cases from centroid
Note
The mean function supplies means for the columns of a data.frame, but the overall mean for a
matrix. Mean will throw a warning for non-numeric data, but colMeans stops with non-numeric
data. Thus, the function uses either mean (for data frames) or colMeans (for matrices). This is true
for skew and kurtosi as well.
Author(s)
William Revelle
References
Joanes, D.N. and Gill, C.A (1998). Comparing measures of sample skewness and kurtosis. The
Statistician, 47, 183-189.
L.DeCarlo. 1997) On the meaning and use of kurtosis, Psychological Methods, 2(3):292-307,
K.V. Mardia (1970). Measures of multivariate skewness and kurtosis with applications. Biometrika,
57(3):pp. 519-30, 1970.
See Also
describe,describe.by, mult.norm in QuantPsyc, Kurt in QuantPsyc
Examples
round(skew(attitude),2) #type 3 (default)
round(kurtosi(attitude),2) #type 3 (default)
#for the differences between the three types of skew and kurtosis:
round(skew(attitude,type=1),2) #type 1
round(skew(attitude,type=2),2) #type 2
mardia(attitude)
x <- matrix(rnorm(1000),ncol=10)
describe(x)
mardia(x)
202 mat.sort
mat.sort Sort the elements of a correlation matrix to reflect factor loadings
Description
To see the structure of a correlation matrix, it is helpful to organize the items so that the similar
items are grouped together. One such grouping technique is factor analysis. mat.sort will sort the
items by a factor model (if specified), or any other order, or by the loadings on the first factor (if
unspecified)
Usage
mat.sort(m, f = NULL)
Arguments
mA correlation matrix
fA factor analysis output (i.e., one with a loadings matrix) or a matrix of weights
Details
The factor analysis output is sorted by size of the largest factor loading for each variable and then
the matrix items are organized by those loadings. The default is to sort by the loadings on the first
factor. Alternatives allow for ordering based upon any vector or matrix.
Value
A sorted correlation matrix, suitable for showing with cor.plot.
Author(s)
William Revelle
See Also
fa,cor.plot
Examples
data(Bechtoldt.1)
sorted <- mat.sort(Bechtoldt.1,fa(Bechtoldt.1,5))
cor.plot(sorted)
matrix.addition 203
matrix.addition A function to add two vectors or matrices
Description
It is sometimes convenient to add two vectors or matrices in an operation analogous to matrix
multiplication. For matrices nXm and mYp, the matrix sum of the i,jth element of nSp = sum(over
m) of iXm + mYj.
Usage
x %+% y
Arguments
xa n by m matrix (or vector if m= 1)
ya m by p matrix (or vector if m = 1)
Details
Used in such problems as Thurstonian scaling. Although not technically matrix addition, as pointed
out by Krus, there are many applications where the sum or difference of two vectors or matrices is
a useful operation. An alternative operation for vectors is outer(x ,y , FUN="+") but this does not
work for matrices.
Value
a n by p matix of sums
Author(s)
William Revelle
References
Krus, D. J. (2001) Matrix addition. Journal of Visual Statistics, 1, (February, 2001).
Examples
x <- seq(1,4)
z <- x %+% -t(x)
x
z
#compare with outer(x,-x,FUN="+")
x <- matrix(seq(1,6),ncol=2)
y <- matrix(seq(1,10),nrow=2)
z <- x %+% y
204 mediate
x
y
z
#but compare this with outer(x ,y,FUN="+")
mediate Estimate and display direct and indirect effects of mediators and mod-
erator in path models
Description
Find the direct and indirect effects of a predictor in path models of mediation and moderation.
Bootstrap confidence intervals for the indirect effects. Mediation models are just extended regres-
sion models making explicit the effect of particular covariates in the model. Moderation is done
by multiplication of the predictor variables. This function supplies basic mediation/moderation
analyses for some of the classic problem types.
Usage
mediate(y, x, m, data, mod = NULL, n.obs = NULL, use = "pairwise", n.iter = 5000,
alpha = 0.05, std = FALSE,plot=TRUE)
mediate.diagram(medi,digits=2,ylim=c(3,7),xlim=c(-1,10),show.c=TRUE,
main="Mediation model",...)
moderate.diagram(medi,digits=2,ylim=c(2,8),main="Moderation model",...)
Arguments
yThe dependent variable (or a formula suitable for a linear model)
xOne or more predictor variables
mOne (or more) mediating variables
data A data frame holding the data or a correlation or covariance matrix.
mod A moderating variable, if desired
n.obs If the data are from a correlation or covariance matrix, how many observations
were used. This will lead to simulated data for the bootstrap.
use use="pairwise" is the default when finding correlations or covariances
n.iter Number of bootstrap resamplings to conduct
alpha Set the width of the confidence interval to be 1 - alpha
std standardize the covariances to find the standardized betas
plot Plot the resulting paths
digits The number of digits to report in the mediate.diagram.
medi The output from mediate may be imported into mediate.diagram
ylim The limits for the y axis in the mediate and moderate diagram functions
mediate 205
xlim The limits for the x axis. Make the minimum more negative if the x by x corre-
lations do not fit.
show.c If FALSE, do not draw the c lines, just the partialed (c’) lines
main The title for the mediate and moderate functions
... Additional graphical parameters to pass to mediate.diagram
Details
When doing linear modeling, it is frequently convenient to estimate the direct effect of a predictor
controlling for the indirect effect of a mediator. See Preacher and Hayes (2004) for a very thor-
ough discussion of mediation. The mediate function will do some basic mediation and moderation
models, with bootstrapped confidence intervals for the mediation/moderation effects.
Functionally, this is just regular linear regression and partial correlation with some different output.
In the case of being provided just a correlation matrix, the bootstrapped values are based upon
bootstrapping from data matching the original covariance/correlation matrix with the addition of
normal errors. This allows us to test the mediation/moderation effect even if not given raw data.
The function has been tested against some of the basic cases and examples in Hayes (2013) and the
associated data sets.
For fine tuning the size of the graphic output, xlim and ylim can be specified in the mediate.diagram
function. Otherwise, the graphics produced by mediate and moderate use the default xlim and ylim
values.
Value
total The total direct effect of x on y (c)
direct The beta effects of x (c’) and m (b) on y
indirect The indirect effect of x through m on y (c-ab)
mean.boot mean bootstrapped value of indirect effect
sd.boot Standard deviation of bootstrapped values
ci.quant The upper and lower confidence intervals based upon the quantiles of the boot-
strapped distribution.
boot The bootstrapped values themselves.
aThe effect of x on m
bThe effect of m on y
b.int The interaction of x and mod (if specified)
Note
There are a number of other packages that do mediation analysis (e.g., sem and lavaan) and they are
probably preferred. This function is supplied for the more basic cases, with 1..k y variables, 1..n x
variables, and 1 ..j mediators. It will not do two step mediation.
Author(s)
William Revelle
206 mediate
References
Hayes, Andrew F. (2013) Introduction to mediation, moderation, and conditional process analysis:
A regression-based approach. Guilford Press.
Preacher, Kristopher J and Hayes, Andrew F (2004) SPSS and SAS procedures for estimating indi-
rect effects in simple mediation models. Behavior Research Methods, Instruments, \& Computers
36, (4) 717-731.
Data from Hayes (2013), Preacher and Hayes (2004), and from Kerchoff (1974)
See Also
setCor and setCor.diagram
Examples
#data from Preacher and Hayes (2004)
sobel <- structure(list(SATIS = c(-0.59, 1.3, 0.02, 0.01, 0.79, -0.35,
-0.03, 1.75, -0.8, -1.2, -1.27, 0.7, -1.59, 0.68, -0.39, 1.33,
-1.59, 1.34, 0.1, 0.05, 0.66, 0.56, 0.85, 0.88, 0.14, -0.72,
0.84, -1.13, -0.13, 0.2), THERAPY = structure(c(0, 1, 1, 0, 1,
1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 1, 0, 1, 1, 1, 0, 1,
1, 1, 1, 0), value.labels = structure(c(1, 0), .Names = c("cognitive",
"standard"))), ATTRIB = c(-1.17, 0.04, 0.58, -0.23, 0.62, -0.26,
-0.28, 0.52, 0.34, -0.09, -1.09, 1.05, -1.84, -0.95, 0.15, 0.07,
-0.1, 2.35, 0.75, 0.49, 0.67, 1.21, 0.31, 1.97, -0.94, 0.11,
-0.54, -0.23, 0.05, -1.07)), .Names = c("SATIS", "THERAPY", "ATTRIB"
), row.names = c(NA, -30L), class = "data.frame", variable.labels = structure(c("Satisfaction",
"Therapy", "Attributional Positivity"), .Names = c("SATIS", "THERAPY",
"ATTRIB")))
#n.iter set to 50 (instead of default of 5000) for speed of example
mediate(1,2,3,sobel,n.iter=50) #The example in Preacher and Hayes
#the pmi covariance matrix from Hayes. 2013.
#data set from Hayes, 2013 has 123 cases instead of the covariance matrix used here
C.pmi <- structure(c(0.251232840197254, 0.119718779155005, 0.157470345195255,
0.124533519925363, 0.03052112488338, 0.0734039717446355, 0.119718779155005,
1.74573503931761, 0.647207783553245, 0.914575836332134, 0.0133613221378115,
-0.0379181660669066, 0.157470345195255, 0.647207783553245, 3.01572704251633,
1.25128282020525, -0.0224576835932294, 0.73973743835799, 0.124533519925363,
0.914575836332134, 1.25128282020525, 2.40342196454751, -0.0106624017059843,
-0.752990470478475, 0.03052112488338, 0.0133613221378115, -0.0224576835932294,
-0.0106624017059843, 0.229241636678662, 0.884479541516727, 0.0734039717446355,
-0.0379181660669066, 0.73973743835799, -0.752990470478475, 0.884479541516727,
33.6509729441557), .Dim = c(6L, 6L), .Dimnames = list(c("cond",
"pmi", "import", "reaction", "gender", "age"), c("cond", "pmi",
"import", "reaction", "gender", "age")))
#n.iter set to 50 (instead of default of 5000) for speed of example
mediate(y="reaction",x = "cond",m=c("pmi","import"),data=C.pmi,n.obs=123,n.iter=50)
mixed.cor 207
#Data from sem package taken from Kerckhoff (and in turn, from Lisrel manual)
R.kerch <- structure(list(Intelligence = c(1, -0.1, 0.277, 0.25, 0.572,
0.489, 0.335), Siblings = c(-0.1, 1, -0.152, -0.108, -0.105,
-0.213, -0.153), FatherEd = c(0.277, -0.152, 1, 0.611, 0.294,
0.446, 0.303), FatherOcc = c(0.25, -0.108, 0.611, 1, 0.248, 0.41,
0.331), Grades = c(0.572, -0.105, 0.294, 0.248, 1, 0.597, 0.478
), EducExp = c(0.489, -0.213, 0.446, 0.41, 0.597, 1, 0.651),
OccupAsp = c(0.335, -0.153, 0.303, 0.331, 0.478, 0.651, 1
)), .Names = c("Intelligence", "Siblings", "FatherEd", "FatherOcc",
"Grades", "EducExp", "OccupAsp"), class = "data.frame", row.names = c("Intelligence",
"Siblings", "FatherEd", "FatherOcc", "Grades", "EducExp", "OccupAsp"
))
#n.iter set to 50 (instead of default of 5000) for speed of demo
mod.k <- mediate("OccupAsp","Intelligence",m= c(2:5),data=R.kerch,n.obs=767,n.iter=50)
mediate.diagram(mod.k)
#Compare the following solution to the path coefficients found by the sem package
mod.k2 <- mediate(y="OccupAsp",x=c("Intelligence","Siblings","FatherEd","FatherOcc"),
m= c(5:6),data=R.kerch,n.obs=767,n.iter=50)
mediate.diagram(mod.k2,show.c=FALSE) #simpler output
mixed.cor Find correlations for mixtures of continuous, polytomous, and dichoto-
mous variables
Description
For data sets with continuous, polytomous and dichotmous variables, the absolute Pearson correla-
tion is downward biased from the underlying latent correlation. mixed.cor finds Pearson correlations
for the continous variables, polychorics for the polytomous items, tetrachorics for the dichoto-
mous items, and the polyserial or biserial correlations for the various mixed variables. Results
include the complete correlation matrix, as well as the separate correlation matrices and difficulties
for the polychoric and tetrachoric correlations.
Usage
mixed.cor(x = NULL, p = NULL, d=NULL,smooth=TRUE, correct=.5,global=TRUE,
ncat=8,use="pairwise",method="pearson",weight=NULL)
Arguments
xA set of continuous variables (may be missing) or, if p and d are missing, the
variables to be analyzed.
pA set of polytomous items (may be missing)
dA set of dichotomous items (may be missing)
208 mixed.cor
smooth If TRUE, then smooth the correlation matix if it is non-positive definite
correct When finding tetrachoric correlations, what value should be used to correct for
continuity?
global For polychorics, should the global values of the tau parameters be used, or
should the pairwise values be used. Set to local if errors are occurring.
ncat The number of categories beyond which a variable is considered "continuous".
use The various options to the cor function include "everything", "all.obs", "com-
plete.obs", "na.or.complete", or "pairwise.complete.obs". The default here is
"pairwise"
method The correlation method to use for the continuous variables. "pearson" (default),
"kendall", or "spearman"
weight If specified, this is a vector of weights (one per participant) to differentially
weight participants. The NULL case is equivalent of weights of 1 for all cases.
Details
This function is particularly useful as part of the Synthetic Apeture Personality Assessment (SAPA)
(http://sapa-project.org) data sets where continuous variables (age, SAT V, SAT Q, etc) and
mixed with polytomous personality items taken from the International Personality Item Pool (IPIP)
and the dichotomous experimental IQ items that have been developed as part of SAPA (see, e.g.,
Revelle, Wilt and Rosenthal, 2010).
This is a very computationally intensive function which can be speeded up considerably by using
multiple cores and using the parallel package. The number of cores to use when doing polychoric
or tetrachoric. The greatest step in speed is going from 1 core to 2. This is about a 50% savings.
Going to 4 cores seems to have about at 66% savings, and 8 a 75% savings. The number of parallel
processes defaults to 2 but can be modified by using the options command: options("mc.cores"=4)
will set the number of cores to 4.
Item response analyses using irt.fa may be done separately on the polytomous and dichotomous
items in order to develop internally consistent scales. These scale may, in turn, be correlated with
each other using the complete correlation matrix found by mixed.cor and using the score.items
function.
This function is not quite as flexible as the hetcor function in John Fox’s polychor package.
Note that the variables may be organized by type of data: first continuous, then polytomous, then
dichotomous. This is done by simply specifying x, p, and d. This is advantageous in the case of
some continuous variables having a limited number of categories because of subsetting.
Value
rho The complete matrix
rx The Pearson correlation matrix for the continuous items
poly the polychoric correlation (poly$rho) and the item difficulties (poly$tau)
tetra the tetrachoric correlation (tetra$rho) and the item difficulties (tetra$tau)
msq 209
Note
mixed.cor was designed for the SAPA project (http://sapa-project.org) with large data sets
with a mixture of continuous, dichotomous, and polytomous data. For smaller data sets, it is some-
times the case that the global estimate of the tau parameter will lead to unstable solutions. This may
be corrected by setting the global parameter = FALSE.
When finding correlations between dummy coded SAPA data (e.g., of occupations), the real cor-
relations are all slightly less than zero because of the ipsatized nature of the data. This leads to a
non-positive definite correlation matrix because the matrix is no longer of full rank. Smoothing will
correct this, even though this might not be desired. Turn off smoothing in this case.
Note that the variables no longer need to be organized by type of data: first continuous, then poly-
tomous, then dichotomous. However, this automatic detection will lead to problems if the variables
such as age are limited to less than 8 categories but those category values differ from the polytomous
items. The fall back is to specify x, p, and d.
Author(s)
William Revelle
References
W.Revelle, J.Wilt, and A.Rosenthal. Personality and cognition: The personality-cognition link. In
A.Gruszka, G. Matthews, and B. Szymura, editors, Handbook of Individual Differences in Cogni-
tion: Attention, Memory and Executive Control, chapter 2, pages 27-49. Springer, 2010.
See Also
polychoric,tetrachoric,score.items,score.irt
Examples
data(bfi)
r <- mixed.cor(bfi[,c(1:5,26,28)])
r
#compare to raw Pearson
#note that the biserials and polychorics are not attenuated
rp <- cor(bfi[c(1:5,26,28)],use="pairwise")
lowerMat(rp)
msq 75 mood items from the Motivational State Questionnaire for 3896
participants
210 msq
Description
Emotions may be described either as discrete emotions or in dimensional terms. The Motivational
State Questionnaire (MSQ) was developed to study emotions in laboratory and field settings. The
data can be well described in terms of a two dimensional solution of energy vs tiredness and tension
versus calmness. Additional items include what time of day the data were collected and a few
personality questionnaire scores.
Usage
data(msq)
Format
A data frame with 3896 observations on the following 92 variables.
active a numeric vector
afraid a numeric vector
alert a numeric vector
angry a numeric vector
anxious a numeric vector
aroused a numeric vector
ashamed a numeric vector
astonished a numeric vector
at.ease a numeric vector
at.rest a numeric vector
attentive a numeric vector
blue a numeric vector
bored a numeric vector
calm a numeric vector
cheerful a numeric vector
clutched.up a numeric vector
confident a numeric vector
content a numeric vector
delighted a numeric vector
depressed a numeric vector
determined a numeric vector
distressed a numeric vector
drowsy a numeric vector
dull a numeric vector
elated a numeric vector
energetic a numeric vector
msq 211
enthusiastic a numeric vector
excited a numeric vector
fearful a numeric vector
frustrated a numeric vector
full.of.pep a numeric vector
gloomy a numeric vector
grouchy a numeric vector
guilty a numeric vector
happy a numeric vector
hostile a numeric vector
idle a numeric vector
inactive a numeric vector
inspired a numeric vector
intense a numeric vector
interested a numeric vector
irritable a numeric vector
jittery a numeric vector
lively a numeric vector
lonely a numeric vector
nervous a numeric vector
placid a numeric vector
pleased a numeric vector
proud a numeric vector
quiescent a numeric vector
quiet a numeric vector
relaxed a numeric vector
sad a numeric vector
satisfied a numeric vector
scared a numeric vector
serene a numeric vector
sleepy a numeric vector
sluggish a numeric vector
sociable a numeric vector
sorry a numeric vector
still a numeric vector
strong a numeric vector
surprised a numeric vector
212 msq
tense a numeric vector
tired a numeric vector
tranquil a numeric vector
unhappy a numeric vector
upset a numeric vector
vigorous a numeric vector
wakeful a numeric vector
warmhearted a numeric vector
wide.awake a numeric vector
alone a numeric vector
kindly a numeric vector
scornful a numeric vector
EA Thayer’s Energetic Arousal Scale
TA Thayer’s Tense Arousal Scale
PA Positive Affect scale
NegAff Negative Affect scale
Extraversion Extraversion from the Eysenck Personality Inventory
Neuroticism Neuroticism from the Eysenck Personality Inventory
Lie Lie from the EPI
Sociability The sociability subset of the Extraversion Scale
Impulsivity The impulsivity subset of the Extraversions Scale
MSQ_Time Time of day the data were collected
MSQ_Round Rounded time of day
TOD a numeric vector
TOD24 a numeric vector
ID subject ID
condition What was the experimental condition after the msq was given
scale a factor with levels msq r original or revised msq
exper Which study were the data collected: a factor with levels AGES BING BORN CART CITY COPE
EMIT FAST Fern FILM FLAT Gray imps item knob MAPS mite pat-1 pat-2 PATS post RAFT
Rim.1 Rim.2 rob-1 rob-2 ROG1 ROG2 SALT sam-1 sam-2 SAVE/PATS sett swam swam-2 TIME
VALE-1 VALE-2 VIEW
Details
The Motivational States Questionnaire (MSQ) is composed of 72 items, which represent the full af-
fective range (Revelle & Anderson, 1998). The MSQ consists of 20 items taken from the Activation-
Deactivation Adjective Check List (Thayer, 1986), 18 from the Positive and Negative Affect Sched-
ule (PANAS, Watson, Clark, & Tellegen, 1988) along with the items used by Larsen and Diener
(1992). The response format was a four-point scale that corresponds to Russell and Carroll’s (1999)
msq 213
"ambiguous–likely-unipolar format" and that asks the respondents to indicate their current standing
(“at this moment") with the following rating scale:
0—————-1—————-2—————-3
Not at all A little Moderately Very much
The original version of the MSQ included 72 items. Intermediate analyses (done with 1840 subjects)
demonstrated a concentration of items in some sections of the two dimensional space, and a paucity
of items in others. To begin correcting this, 3 items from redundantly measured sections (alone,
kindly, scornful) were removed, and 5 new ones (anxious, cheerful, idle, inactive, and tranquil) were
added. Thus, the correlation matrix is missing the correlations between items anxious, cheerful,
idle, inactive, and tranquil with alone, kindly, and scornful.
Procedure. The data were collected over nine years, as part of a series of studies examining the
effects of personality and situational factors on motivational state and subsequent cognitive perfor-
mance. In each of 38 studies, prior to any manipulation of motivational state, participants signed a
consent form and filled out the MSQ. (The procedures of the individual studies are irrelevant to this
data set and could not affect the responses to the MSQ, since this instrument was completed before
any further instructions or tasks). Some MSQ post test (after manipulations) is available in affect.
The EA and TA scales are from Thayer, the PA and NA scales are from Watson et al. (1988). Scales
and items:
Energetic Arousal: active, energetic, vigorous, wakeful, wide.awake, full.of.pep, lively, -sleepy,
-tired, - drowsy (ADACL)
Tense Arousal: Intense, Jittery, fearful, tense, clutched up, -quiet, -still, - placid, - calm, -at rest
(ADACL)
Positive Affect: active, alert, attentive, determined, enthusiastic, excited, inspired, interested, proud,
strong (PANAS)
Negative Affect: afraid, ashamed, distressed, guilty, hostile, irritable , jittery, nervous, scared, upset
(PANAS)
The PA and NA scales can in turn can be thought of as having subscales: (See the PANAS-X) Fear:
afraid, scared, nervous, jittery (not included frightened, shaky) Hostility: angry, hostile, irritable,
(not included: scornful, disgusted, loathing guilt: ashamed, guilty, (not included: blameworthy,
angry at self, disgusted with self, dissatisfied with self) sadness: alone, blue, lonely, sad, (not
included: downhearted) joviality: cheerful, delighted, energetic, enthusiastic, excited, happy, lively,
(not included: joyful) self-assurance: proud, strong, confident, (not included: bold, daring, fearless
) attentiveness: alert, attentive, determined (not included: concentrating)
The next set of circumplex scales were taken (I think) from Larsen and Diener (1992). High ac-
tivation: active, aroused, surprised, intense, astonished Activated PA: elated, excited, enthusiastic,
lively Unactivated NA : calm, serene, relaxed, at rest, content, at ease PA: happy, warmhearted,
pleased, cheerful, delighted Low Activation: quiet, inactive, idle, still, tranquil Unactivated PA:
dull, bored, sluggish, tired, drowsy NA: sad, blue, unhappy, gloomy, grouchy Activated NA: jittery,
anxious, nervous, fearful, distressed.
Keys for these separate scales are shown in the examples.
In addition to the MSQ, there are 5 scales from the Eysenck Personality Inventory (Extraversion,
Impulsivity, Sociability, Neuroticism, Lie). The Imp and Soc are subsets of the the total extraversion
scale.
214 msq
Source
Data collected at the Personality, Motivation, and Cognition Laboratory, Northwestern University.
References
Rafaeli, Eshkol and Revelle, William (2006), A premature consensus: Are happiness and sadness
truly opposite affects? Motivation and Emotion, 30, 1, 1-12.
Revelle, W. and Anderson, K.J. (1998) Personality, motivation and cognitive performance: Final re-
port to the Army Research Institute on contract MDA 903-93-K-0008. (http://www.personality-project.
org/revelle/publications/ra.ari.98.pdf).
Thayer, R.E. (1989) The biopsychology of mood and arousal. Oxford University Press. New York,
NY.
Watson,D., Clark, L.A. and Tellegen, A. (1988) Development and validation of brief measures of
positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology,
54(6):1063-1070.
See Also
affect for an example of the use of some of these adjectives in a mood manipulation study.
make.keys,scoreItems and scoreOverlap for instructions on how to score multiple scales with
and without item overlap. Also see fa and fa.extension for instructions on how to do factor
analyses or factor extension.
Examples
data(msq)
if(FALSE){ #not run in the interests of time
#basic descriptive statistics
describe(msq)
}
#score them for 20 short scales -- note that these have item overlap
#The first 2 are from Thayer
#The next 2 are classic positive and negative affect
#The next 9 are circumplex scales
#the last 7 are msq estimates of PANASX scales (missing some items)
keys <- make.keys(msq[1:75],list(
EA = c("active", "energetic", "vigorous", "wakeful", "wide.awake", "full.of.pep",
"lively", "-sleepy", "-tired", "-drowsy"),
TA =c("intense", "jittery", "fearful", "tense", "clutched.up", "-quiet", "-still",
"-placid", "-calm", "-at.rest") ,
PA =c("active", "excited", "strong", "inspired", "determined", "attentive",
"interested", "enthusiastic", "proud", "alert"),
NAf =c("jittery", "nervous", "scared", "afraid", "guilty", "ashamed", "distressed",
"upset", "hostile", "irritable" ),
HAct = c("active", "aroused", "surprised", "intense", "astonished"),
aPA = c("elated", "excited", "enthusiastic", "lively"),
uNA = c("calm", "serene", "relaxed", "at.rest", "content", "at.ease"),
pa = c("happy", "warmhearted", "pleased", "cheerful", "delighted" ),
LAct = c("quiet", "inactive", "idle", "still", "tranquil"),
mssd 215
uPA =c( "dull", "bored", "sluggish", "tired", "drowsy"),
naf = c( "sad", "blue", "unhappy", "gloomy", "grouchy"),
aNA = c("jittery", "anxious", "nervous", "fearful", "distressed"),
Fear = c("afraid" , "scared" , "nervous" , "jittery" ) ,
Hostility = c("angry" , "hostile", "irritable", "scornful" ),
Guilt = c("guilty" , "ashamed" ),
Sadness = c( "sad" , "blue" , "lonely", "alone" ),
Joviality =c("happy","delighted", "cheerful", "excited", "enthusiastic", "lively", "energetic"),
Self.Assurance=c( "proud","strong" , "confident" , "-fearful" ),
Attentiveness = c("alert" , "determined" , "attentive" )
#acquiscence = c("sleepy" , "wakeful" , "relaxed","tense")
))
msq.scores <- scoreItems(keys,msq[1:75])
#show a circumplex structure for the non-overlapping items
fcirc <- fa(msq.scores$scores[,5:12],2)
fa.plot(fcirc,labels=colnames(msq.scores$scores)[5:12])
#now, find the correlations corrected for item overlap
msq.overlap <- scoreOverlap(keys,msq[1:75])
f2 <- fa(msq.overlap$cor,2)
fa.plot(f2,labels=colnames(msq.overlap$cor),title="2 dimensions of affect, corrected for overlap")
if(FALSE) {
#extend this solution to EA/TA NA/PA space
fe <- fa.extension(cor(msq.scores$scores[,5:12],msq.scores$scores[,1:4]),fcirc)
fa.diagram(fcirc,fe=fe,main="Extending the circumplex structure to EA/TA and PA/NA ")
#show the 2 dimensional structure
f2 <- fa(msq[1:72],2)
fa.plot(f2,labels=colnames(msq)[1:72],title="2 dimensions of affect at the item level")
#sort them by polar coordinates
round(polar(f2),2)
}
mssd Find von Neuman’s Mean Square of Successive Differences
Description
Von Neuman et al. (1941) discussed the Mean Square of Successive Differences as a measure of
variability that takes into account gradual shifts in mean. This is appropriate when studying errors
in ballistics or variability and stability in mood when studying affect. For random data, this will be
twice the variance, but for data with a sequential order and a positive autocorrelation, this will be
much smaller. This is just an application of the diff an ny functions
216 mssd
Usage
mssd(x,group=NULL, lag = 1,na.rm=TRUE)
rmssd(x,group=NULL, lag=1, na.rm=TRUE)
Arguments
xa vector, data.frame or matrix
lag the lag to use when finding diff
group A column of the x data.frame to be used for grouping
na.rm Should missing data be removed?
Details
When examining multiple measures within subjects, it is sometimes useful to consider the variabil-
ity of trial by trial observations in addition to the over all between trial variation. The Mean Square
of Successive Differences (mssd) and root mean square of successive differences (rmssd) find the
variance or standard deviation of the trial to trial differences.
σ2= Σ(xi−xi+1)2/(n−1)
In the case of multiple subjects (groups) with multiple observations per subject (group), specify the
grouping variable will produce output for each group.
Similar functions are available in the matrixStats package. This is just the variance and standard
deviation applied to the result from the diff function.
Value
The variance (mssd) or standard deviation (rmssd) of the lagged differences.
Author(s)
William Revelle
References
Von Neumann, J., Kent, R., Bellinson, H., and Hart, B. (1941). The mean square successive differ-
ence. The Annals of Mathematical Statistics, pages 153-162.
See Also
See Also rmssd for the standard deviation or describe for more conventional statistics. describeBy
and statsBy give group level statistics.
Examples
t <- seq(-pi, pi, .1)
trial <- 1: length(t)
gr <- trial %% 8
c <- cos(t)
ts <- sample(t,length(t))
multi.hist 217
cs <- cos(ts)
x.df <- data.frame(trial,gr,t,c,ts,cs)
rmssd(x.df)
rmssd(x.df,gr)
describe(x.df)
#pairs.panels(x.df)
multi.hist Multiple histograms with density and normal fits on one page
Description
Given a matrix or data.frame, produce histograms for each variable in a "matrix" form. Include
normal fits and density distributions for each plot.
The number of rows and columns may be specified, or calculated. May be used for single variables.
Usage
multi.hist(x,nrow=NULL,ncol=NULL,density=TRUE,freq=FALSE,bcol="white",
dcol=c("black","black"),dlty=c("dashed","dotted"),
main="Histogram, Density, and Normal Fit",...)
histBy(x,var,group,density=TRUE,alpha=.5,breaks=21,col,xlab,
main="Histograms by group",...)
Arguments
xmatrix or data.frame
var The variable in x to plot in histBy
group The name of the variable in x to use as the grouping variable
nrow number of rows in the plot
ncol number of columns in the plot
density density=TRUE, show the normal fits and density distributions
freq freq=FALSE shows probability densities and density distribution, freq=TRUE
shows frequencies
bcol Color for the bars
dcol The color(s) for the normal and the density fits. Defaults to black.
dlty The line type (lty) of the normal and density fits. (specify the optional graphic
parameter lwd to change the line size)
main title for each panel
xlab Label for the x variable
breaks The number of breaks in histBy (see hist)
alpha The degree of transparency of the overlapping bars in histBy
col A vector of colors in histBy (defaults to the rainbow)
... additional graphic parameters (e.g., col)
218 neo
Author(s)
William Revelle
See Also
bi.bars for drawing pairwise histograms
Examples
multi.hist(sat.act)
multi.hist(sat.act,bcol="red")
multi.hist(sat.act,dcol="blue") #make both lines blue
multi.hist(sat.act,dcol= c("blue","red"),dlty=c("dotted", "solid"))
multi.hist(sat.act,freq=TRUE) #show the frequency plot
multi.hist(sat.act,nrow=2)
histBy(sat.act,"SATQ","gender")
neo NEO correlation matrix from the NEO_PI_R manual
Description
The NEO.PI.R is a widely used personality test to assess 5 broad factors (Neuroticism, Extraver-
sion, Openness, Agreeableness and Conscientiousness) with six facet scales for each factor. The
correlation matrix of the facets is reported in the NEO.PI.R manual for 1000 subjects.
Usage
data(neo)
Format
A data frame of a 30 x 30 correlation matrix with the following 30 variables.
N1 Anxiety
N2 AngryHostility
N3 Depression
N4 Self-Consciousness
N5 Impulsiveness
N6 Vulnerability
E1 Warmth
E2 Gregariousness
E3 Assertiveness
E4 Activity
E5 Excitement-Seeking
neo 219
E6 PositiveEmotions
O1 Fantasy
O2 Aesthetics
O3 Feelings
O4 Ideas
O5 Actions
O6 Values
A1 Trust
A2 Straightforwardness
A3 Altruism
A4 Compliance
A5 Modesty
A6 Tender-Mindedness
C1 Competence
C2 Order
C3 Dutifulness
C4 AchievementStriving
C5 Self-Discipline
C6 Deliberation
Details
The past thirty years of personality research has led to a general consensus on the identification of
major dimensions of personality. Variously known as the “Big 5" or the “Five Factor Model", the
general solution represents 5 broad domains of personal and interpersonal experience. Neuroticism
and Extraversion are thought to reflect sensitivity to negative and positive cues from the environment
and the tendency to withdraw or approach. Openness is sometimes labeled as Intellect and reflects
an interest in new ideas and experiences. Agreeableness and Conscientiousness reflect tendencies
to get along with others and to want to get ahead.
The factor structure of the NEO suggests five correlated factors as well as two higher level factors.
The NEO was constructed with 6 “facets" for each of the five broad factors.
Source
Costa, Paul T. and McCrae, Robert R. (1992) (NEO PI-R) professional manual. Psychological
Assessment Resources, Inc. Odessa, FL. (with permission of the author and the publisher)
References
Digman, John M. (1990) Personality structure: Emergence of the five-factor model. Annual Review
of Psychology. 41, 417-440.
John M. Digman (1997) Higher-order factors of the Big Five. Journal of Personality and Social
Psychology, 73, 1246-1256.
220 omega
McCrae, Robert R. and Costa, Paul T., Jr. (1999) A Five-Factor theory of personality. In Pervin,
Lawrence A. and John, Oliver P. (eds) Handbook of personality: Theory and research (2nd ed.)
139-153. Guilford Press, New York. N.Y.
Revelle, William (1995), Personality processes, Annual Review of Psychology, 46, 295-328.
Joshua Wilt and William Revelle (2009) Extraversion and Emotional Reactivity. In Mark Leary and
Rick H. Hoyle (eds). Handbook of Individual Differences in Social Behavior. Guilford Press, New
York, N.Y.
Examples
data(neo)
n5 <- fa(neo,5)
neo.keys <- make.keys(30,list(N=c(1:6),E=c(7:12),O=c(13:18),A=c(19:24),C=c(25:30)))
n5p <- target.rot(n5,neo.keys) #show a targeted rotation for simple structure
n5p
omega Calculate McDonald’s omega estimates of general and total factor
saturation
Description
McDonald has proposed coefficient omega as an estimate of the general factor saturation of a test.
One way to find omega is to do a factor analysis of the original data set, rotate the factors obliquely,
do a Schmid Leiman transformation, and then find omega. This function estimates omega as sug-
gested by McDonald by using hierarchical factor analysis (following Jensen). A related option is
to define the model using omega and then perform a confirmatory factor analysis using the sem
package. This is done by omegaSem and omegaFromSem.
Usage
omega(m,nfactors=3,fm="minres",n.iter=1,p=.05,poly=FALSE,key=NULL,
flip=TRUE,digits=2, title="Omega",sl=TRUE,labels=NULL,
plot=TRUE,n.obs=NA,rotate="oblimin",Phi=NULL,option="equal",covar=FALSE, ...)
omegaSem(m,nfactors=3,fm="minres",key=NULL,flip=TRUE,digits=2,title="Omega",
sl=TRUE,labels=NULL, plot=TRUE,n.obs=NA,rotate="oblimin",
Phi = NULL, option="equal",...)
omegah(m,nfactors=3,fm="minres",key=NULL,flip=TRUE,
digits=2,title="Omega",sl=TRUE,labels=NULL, plot=TRUE,
n.obs=NA,rotate="oblimin",Phi = NULL,option="equal",covar=FALSE,...)
omega 221
Arguments
mA correlation matrix, or a data.frame/matrix of data, or (if Phi is specified, an
oblique factor pattern matrix
nfactors Number of factors believed to be group factors
n.iter How many replications to do in omega for bootstrapped estimates
fm factor method (the default is minres) fm="pa" for principal axes, fm="minres"
for a minimum residual (OLS) solution, fm="pc" for principal components (see
note), or fm="ml" for maximum likelihood.
poly should the correlation matrix be found using polychoric/tetrachoric or normal
Pearson correlations
key a vector of +/- 1s to specify the direction of scoring of items. The default is
to assume all items are positively keyed, but if some items are reversed scored,
then key should be specified.
flip If flip is TRUE, then items are automatically flipped to have positive correlations
on the general factor. Items that have been reversed are shown with a - sign.
pprobability of two tailed conference boundaries
digits if specified, round the output to digits
title Title for this analysis
sl If plotting the results, should the Schmid Leiman solution be shown or should
the hierarchical solution be shown? (default sl=TRUE)
labels If plotting, what labels should be applied to the variables? If not specified, will
default to the column names.
plot plot=TRUE (default) calls omega.diagram, plot =FALSE does not. If Rgraphviz
is available, then omega.graph may be used separately.
n.obs Number of observations - used for goodness of fit statistic
rotate What rotation to apply? The default is oblimin, the alternatives include simpli-
max, Promax, cluster and target. target will rotate to an optional keys matrix
(See target.rot)
Phi If specified, then omega is found from the pattern matrix (m) and the factor
intercorrelation matrix (Phi).
option In the two factor case (not recommended), should the loadings be equal, empha-
size the first factor, or emphasize the second factor. See in particular the option
parameter in schmid for treating the case of two group factors.
covar defaults to FALSE and the correlation matrix is found (standardized variables.)
If TRUE, the do the calculations on the unstandardized variables and use covari-
ances.
... Allows additional parameters to be passed through to the factor routines.
Details
“Many scales are assumed by their developers and users to be primarily a measure of one latent
variable. When it is also assumed that the scale conforms to the effect indicator model of measure-
ment (as is almost always the case in psychological assessment), it is important to support such an
222 omega
interpretation with evidence regarding the internal structure of that scale. In particular, it is impor-
tant to examine two related properties pertaining to the internal structure of such a scale. The first
property relates to whether all the indicators forming the scale measure a latent variable in common.
The second internal structural property pertains to the proportion of variance in the scale scores
(derived from summing or averaging the indicators) accounted for by this latent variable that is
common to all the indicators (Cronbach, 1951; McDonald, 1999; Revelle, 1979). That is, if an
effect indicator scale is primarily a measure of one latent variable common to all the indicators
forming the scale, then that latent variable should account for the majority of the variance in the
scale scores. Put differently, this variance ratio provides important information about the sampling
fluctuations when estimating individuals’ standing on a latent variable common to all the indicators
arising from the sampling of indicators (i.e., when dealing with either Type 2 or Type 12 sampling,
to use the terminology of Lord, 1956). That is, this variance proportion can be interpreted as the
square of the correlation between the scale score and the latent variable common to all the indicators
in the infinite universe of indicators of which the scale indicators are a subset. Put yet another way,
this variance ratio is important both as reliability and a validity coefficient. This is a reliability
issue as the larger this variance ratio is, the more accurately one can predict an individual’s relative
standing on the latent variable common to all the scale’s indicators based on his or her observed
scale score. At the same time, this variance ratio also bears on the construct validity of the scale
given that construct validity encompasses the internal structure of a scale." (Zinbarg, Yovel, Revelle,
and McDonald, 2006).
McDonald has proposed coefficient omega_hierarchical (ωh) as an estimate of the general factor
saturation of a test. Zinbarg, Revelle, Yovel and Li (2005) http://personality-project.org/
revelle/publications/zinbarg.revelle.pmet.05.pdf compare McDonald’s ωhto Cronbach’s
αand Revelle’s β. They conclude that ωhis the best estimate. (See also Zinbarg et al., 2006 and
Revelle and Zinbarg (2009)).
One way to find ωhis to do a factor analysis of the original data set, rotate the factors obliquely,
factor that correlation matrix, do a Schmid-Leiman (schmid) transformation to find general factor
loadings, and then find ωh. Here we present code to do that.
ωhdiffers as a function of how the factors are estimated. Four options are available, three use the
fa function but with different factoring methods: the default does a minres factor solution, fm="pa"
does a principle axes factor analysis fm="mle" does a maximum likelihood solution; fm="pc" does
a principal components analysis using (principal).
For ability items, it is typically the case that all items will have positive loadings on the general
factor. However, for non-cognitive items it is frequently the case that some items are to be scored
positively, and some negatively. Although probably better to specify which directions the items are
to be scored by specifying a key vector, if flip =TRUE (the default), items will be reversed so that
they have positive loadings on the general factor. The keys are reported so that scores can be found
using the scoreItems function. Arbitrarily reversing items this way can overestimate the general
factor. (See the example with a simulated circumplex).
β, an alternative to ωh, is defined as the worst split half reliability (Revelle, 1979). It can be esti-
mated by using ICLUST (a hierarchical clustering algorithm originally developed for main frames
and written in Fortran and that is now part of the psych package. (For a very complimentary review
of why the ICLUST algorithm is useful in scale construction, see Cooksey and Soutar, 2005).
The omega function uses exploratory factor analysis to estimate the ωhcoefficient. It is important
to remember that “A recommendation that should be heeded, regardless of the method chosen to
estimate ωh, is to always examine the pattern of the estimated general factor loadings prior to esti-
mating ωh. Such an examination constitutes an informal test of the assumption that there is a latent
omega 223
variable common to all of the scale’s indicators that can be conducted even in the context of EFA. If
the loadings were salient for only a relatively small subset of the indicators, this would suggest that
there is no true general factor underlying the covariance matrix. Just such an informal assumption
test would have afforded a great deal of protection against the possibility of misinterpreting the
misleading ωhestimates occasionally produced in the simulations reported here." (Zinbarg et al.,
2006, p 137).
A simple demonstration of the problem of an omega estimate reflecting just one of two group factors
can be found in the last example.
Diagnostic statistics that reflect the quality of the omega solution include a comparison of the rela-
tive size of the g factor eigen value to the other eigen values, the percent of the common variance
for each item that is general factor variance (p2), the mean of p2, and the standard deviation of p2.
Further diagnostics can be done by describing (describe) the $schmid$sl results.
Although omega_h is uniquely defined only for cases where 3 or more subfactors are extracted, it
is sometimes desired to have a two factor solution. By default this is done by forcing the schmid
extraction to treat the two subfactors as having equal loadings.
There are three possible options for this condition: setting the general factor loadings between the
two lower order factors to be "equal" which will be the sqrt(oblique correlations between the factors)
or to "first" or "second" in which case the general factor is equated with either the first or second
group factor. A message is issued suggesting that the model is not really well defined. This solution
discussed in Zinbarg et al., 2007. To do this in omega, add the option="first" or option="second" to
the call.
Although obviously not meaningful for a 1 factor solution, it is of course possible to find the sum
of the loadings on the first (and only) factor, square them, and compare them to the overall matrix
variance. This is done, with appropriate complaints.
In addition to ωh, another of McDonald’s coefficients is ωt. This is an estimate of the total reliability
of a test.
McDonald’s ωt, which is similar to Guttman’s λ6,guttman but uses the estimates of uniqueness
(u2) from factor analysis to find e2
j. This is based on a decomposition of the variance of a test score,
Vxinto four parts: that due to a general factor, ~g, that due to a set of group factors, ~
f, (factors
common to some but not all of the items), specific factors, ~s unique to each item, and ~e, random
error. (Because specific variance can not be distinguished from random error unless the test is given
at least twice, some combine these both into error).
Letting ~x =~cg +~
Af +~
Ds +~e then the communality of itemj, based upon general as well as group
factors, h2
j=c2
j+Pf2
ij and the unique variance for the item u2
j=σ2
j(1 −h2
j)may be used to
estimate the test reliability. That is, if h2
jis the communality of itemj, based upon general as well
as group factors, then for standardized items, e2
j= 1 −h2
jand
ωt=~
1~
cc0~
1 + ~
1~
AA0~
10
Vx
= 1 −P(1 −h2
j)
Vx
= 1 −Pu2
Vx
Because h2
j≥r2
smc,ωt≥λ6.
It is important to distinguish here between the two ωcoefficients of McDonald, 1978 and Equation
6.20a of McDonald, 1999, ωtand ωh. While the former is based upon the sum of squared loadings
on all the factors, the latter is based upon the sum of the squared loadings on the general factor.
ωh=~
1~
cc0~
1
Vx
224 omega
Another estimate reported is the omega for an infinite length test with a structure similar to the
observed test (omega H asymptotic). This is found by
ωlimit =~
1~
cc0~
1
~
1~
cc0~
1 + ~
1~
AA0~
10
.
Following suggestions by Steve Reise, the Explained Common Variance (ECV) is also reported.
This is the ratio of the general factor eigen value to the sum of all of the eigen values. As such,
it is a better indicator of unidimensionality than of the amount of test variance accounted for by a
general factor.
The input to omega may be a correlation matrix or a raw data matrix, or a factor pattern matrix with
the factor intercorrelations (Phi) matrix.
omega is an exploratory factor analysis function that uses a Schmid-Leiman transformation. omegaSem
first calls omega and then takes the Schmid-Leiman solution, converts this to a confirmatory sem
model and then calls the sem package to conduct a confirmatory model. ωhis then calculated from
the CFA output. Although for well behaved problems, the efa and cfa solutions will be practically
identical, the CFA solution will not always agree with the EFA solution. In particular, the estimated
R2will sometimes exceed 1. (An example of this is the Harman 24 cognitive abilities problem.)
In addition, not all EFA solutions will produce workable CFA solutions. Model misspecifications
will lead to very strange CFA estimates.
omegaFromSem takes the output from a sem model and uses it to find ωh. The estimate of factor
indeterminacy, found by the multiple R2of the variables with the factors, will not match that found
by the EFA model. In particular, the estimated R2will sometimes exceed 1. (An example of this is
the Harman 24 cognitive abilities problem.)
The notion of omega may be applied to the individual factors as well as the overall test. A typical
use of omega is to identify subscales of a total inventory. Some of that variability is due to the
general factor of the inventory, some to the specific variance of each subscale. Thus, we can find a
number of different omega estimates: what percentage of the variance of the items identified with
each subfactor is actually due to the general factor. What variance is common but unique to the
subfactor, and what is the total reliable variance of each subfactor. These results are reported in
omega.group object and in the last few lines of the normal output.
The summary of the omega object is a reduced set of the most useful output.
The various objects returned from omega include:
Value
omega hierarchical
The ωhcoefficient
omega.lim The limit of ωhas the test becomes infinitly large
omega total The omegatcoefficient
alpha Cronbach’s α
schmid The Schmid Leiman transformed factor matrix and associated matrices
schmid$sl The g factor loadings as well as the residualized factors
schmid$orthog Varimax rotated solution of the original factors
omega 225
schmid$oblique The oblimin or promax transformed factors
schmid$phi the correlation matrix of the oblique factors
schmid$gloading
The loadings on the higher order, g, factor of the oblimin factors
key A vector of -1 or 1 showing which direction the items were scored.
model a matrix suitable to be given to the sem function for structure equation models
sem The output from a sem analysis
omega.group The summary statistics for the omega total, omega hierarchical (general) and
omega within each group.
scores Factor score estimates are found for the Schmid-Leiman solution. To get scores
for the hierarchical model see the note.
various fit statistics
various fit statistics, see output
Note
Requires the GPArotation package.
The default rotation uses oblimin from the GPArotation package. Alternatives include the simpli-
max function, as well as Promax.
If the factor solution leads to an exactly orthogonal solution (probably only for demonstration data
sets), then use the rotate="Promax" option to get a solution.
omegaSem requires the sem package. omegaFromSem uses the output from the sem package.
omega may be run on raw data (finding either Pearson or tetrachoric/polychoric corrlations, de-
pending upon the poly option) a correlation matrix, a polychoric correlation matrix (found by e.g.,
polychoric), or the output of a previous omega run. This last case is particularly useful when
working with categorical data using the poly=TRUE option. For in this case, most of the time is
spent in finding the correlation matrix. The matrix is saved as part of the omega output and may
be used as input for subsequent runs. A similar feature is found in irt.fa where the output of one
analysis can be taken as the input to the subsequent analyses.
However, simulations based upon tetrachoric and polychoric correlations suggest that although the
structure is better defined, that the estimates of omega are inflated over the true general factor
saturation.
Omega returns factor scores based upon the Schmid-Leiman transformation. To get the hierarchical
factor scores, it is necessary to do this outside of omega. See the example (not run).
Consider the case of the raw data in an object data. Then
f3 <- fa(data,3,scores="tenBerge", oblique.rotation=TRUE f1 <- fa(f3$scores) hier.scores <- data.frame(f1$scores,f3$scores)
When doing fm="pc", principal components are done for the original correlation matrix, but minres
is used when examining the intercomponent correlations. A warning is issued that the method was
changed to minres for the higher order solution. omega is a factor model, and finding loadings using
principal components will overestimate the resulting solution. This is particularly problematic for
the amount of group saturation, and thus the omega.group statistics are overestimates.
The last three lines of omega report "Total, General and Subset omega for each subset". These are
available as the omega.group object in the output.
226 omega
The last of these (omega group) is effectively what Steve Reise calls omegaS for the subset omega.
The omega general is the amount of variance in the group that is accounted for by the general factor,
the omega total is the amount of variance in the group accounted for by general + group.
This is based upon a cluster solution (that is to say, every item is assigned to one group) and this
is why for first column the omega general and group do not add up to omega total. Some of the
variance is found in the cross loadings between groups.
Reise and others like to report the ratio of the second line to the first line (what portion of the reliable
variance is general factor) and the third row to the first (what portion of the reliable variance is
within group but not general. This may be found by using the omega.group object that is returned
by omega. (See the last example.)
Author(s)
http://personality-project.org/revelle.html
Maintainer: William Revelle < revelle@northwestern.edu >
References
http://personality-project.org/r/r.omega.html
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
Revelle, W. (1979). Hierarchical cluster analysis and the internal structure of tests. Multivariate
Behavioral Research, 14, 57-74. (http://personality-project.org/revelle/publications/
iclust.pdf)
Revelle, W. and Zinbarg, R. E. (2009) Coefficients alpha, beta, omega and the glb: comments
on Sijtsma. Psychometrika, 74, 1, 145-154. (http://personality-project.org/revelle/
publications/rz09.pdf
Zinbarg, R.E., Revelle, W., Yovel, I., & Li. W. (2005). Cronbach’s Alpha, Revelle’s Beta, Mc-
Donald’s Omega: Their relations with each and two alternative conceptualizations of reliabil-
ity. Psychometrika. 70, 123-133. http://personality-project.org/revelle/publications/
zinbarg.revelle.pmet.05.pdf
Zinbarg, R., Yovel, I. & Revelle, W. (2007). Estimating omega for structures containing two group
factors: Perils and prospects. Applied Psychological Measurement. 31 (2), 135-157.
Zinbarg, R., Yovel, I., Revelle, W. & McDonald, R. (2006). Estimating generalizability to a
universe of indicators that all have one attribute in common: A comparison of estimators for
omega. Applied Psychological Measurement, 30, 121-144. DOI: 10.1177/0146621605278814
http://apm.sagepub.com/cgi/reprint/30/2/121
See Also
omega.graph ICLUST,ICLUST.graph,VSS,schmid ,make.hierarchical
omega 227
Examples
## Not run:
test.data <- Harman74.cor$cov
# if(!require(GPArotation)) {message("Omega requires GPA rotation" )} else {
my.omega <- omega(test.data)
print(my.omega,digits=2)
#}
#create 9 variables with a hierarchical structure
v9 <- sim.hierarchical()
#with correlations of
round(v9,2)
#find omega
v9.omega <- omega(v9,digits=2)
v9.omega
#create 8 items with a two factor solution, showing the use of the flip option
sim2 <- item.sim(8)
omega(sim2) #an example of misidentification-- remember to look at the loadings matrices.
omega(sim2,2) #this shows that in fact there is no general factor
omega(sim2,2,option="first") #but, if we define one of the two group factors
#as a general factor, we get a falsely high omega
#apply omega to analyze 6 mental ability tests
data(ability.cov) #has a covariance matrix
omega(ability.cov$cov)
#om <- omega(Thurstone)
#round(om$omega.group,2)
#round(om$omega.group[2]/om$omega.group[1],2) #fraction of reliable that is general variance
# round(om$omega.group[3]/om$omega.group[1],2) #fraction of reliable that is group variance
#To find factor score estimates for the hierarchical model it is necessary to
#do two extra steps.
#Consider the case of the raw data in an object data. (An example from simulation)
# set.seed(42)
# gload <- matrix(c(.9,.8,.7),nrow=3)
# fload <- matrix(c(.8,.7,.6,rep(0,9),.7,.6,.5,rep(0,9),.7,.6,.4), ncol=3)
# data <- sim.hierarchical(gload=gload,fload=fload, n=100000, raw=TRUE)
#
# f3 <- fa(data$observed,3,scores="tenBerge", oblique.scores=TRUE)
# f1 <- fa(f3$scores)
# om <- omega(data$observed,sl=FALSE) #draw the hierarchical figure
# The scores from om are based upon the schmid-leiman factors and although the g factor
# is identical, the group factors are not.
# This is seen in the following correlation matrix
# hier.scores <- cbind(om$scores,f1$scores,f3$scores)
# lowerCor(hier.scores)
## End(Not run)
228 omega.graph
omega.graph Graph hierarchical factor structures
Description
Hierarchical factor structures represent the correlations between variables in terms of a smaller set
of correlated factors which themselves can be represented by a higher order factor.
Two alternative solutions to such structures are found by the omega function. The correlated factors
solutions represents the effect of the higher level, general factor, through its effect on the correlated
factors. The other representation makes use of the Schmid Leiman transformation to find the direct
effect of the general factor upon the original variables as well as the effect of orthogonal residual
group factors upon the items.
Graphic presentations of these two alternatives are helpful in understanding the structure. omega.graph
and omega.diagram draw both such structures. Graphs are drawn directly onto the graphics window
or expressed in “dot" commands for conversion to graphics using implementations of Graphviz (if
using omega.graph).
Using Graphviz allows the user to clean up the Rgraphviz output. However, if Graphviz and
Rgraphviz are not available, use omega.diagram.
See the other structural diagramming functions, fa.diagram and structure.diagram.
In addition
Usage
omega.diagram(om.results,sl=TRUE,sort=TRUE,labels=NULL,flabels=NULL,cut=.2,
gcut=.2,simple=TRUE, errors=FALSE, digits=1,e.size=.1,rsize=.15,side=3,
main=NULL,cex=NULL,color.lines=TRUE,marg=c(.5,.5,1.5,.5),adj=2, ...)
omega.graph(om.results, out.file = NULL, sl = TRUE, labels = NULL, size = c(8, 6),
node.font = c("Helvetica", 14), edge.font = c("Helvetica", 10),
rank.direction=c("RL","TB","LR","BT"), digits = 1, title = "Omega", ...)
Arguments
om.results The output from the omega function
out.file Optional output file for off line analysis using Graphviz
sl Orthogonal clusters using the Schmid-Leiman transform (sl=TRUE) or oblique
clusters
labels variable labels
flabels Labels for the factors (not counting g)
size size of graphics window
node.font What font to use for the items
edge.font What font to use for the edge labels
rank.direction Defaults to left to right
digits Precision of labels
omega.graph 229
cex control font size
color.lines Use black for positive, red for negative
marg The margins for the figure are set to be wider than normal by default
adj Adjust the location of the factor loadings to vary as factor mod 4 + 1
title Figure title
main main figure caption
... Other options to pass into the graphics packages
e.size the size to draw the ellipses for the factors. This is scaled by the number of
variables.
cut Minimum path coefficient to draw
gcut Minimum general factor path to draw
simple draw just one path per item
sort sort the solution before making the diagram
side on which side should errors be drawn?
errors show the error estimates
rsize size of the rectangles
Details
While omega.graph requires the Rgraphviz package, omega.diagram does not. codeomega requires
the GPArotation package.
Value
clust.graph A graph object
sem A matrix suitable to be run throughe the sem function in the sem package.
Note
omega.graph requires rgraphviz. – omega requires GPArotation
Author(s)
http://personality-project.org/revelle.html
Maintainer: William Revelle < revelle@northwestern.edu >
References
http://personality-project.org/r/r.omega.html
Revelle, W. (in preparation) An Introduction to Psychometric Theory with applications in R. http:
//personality-project.org/r/book
Revelle, W. (1979). Hierarchical cluster analysis and the internal structure of tests. Multivariate
Behavioral Research, 14, 57-74. (http://personality-project.org/revelle/publications/
iclust.pdf)
230 outlier
Zinbarg, R.E., Revelle, W., Yovel, I., & Li. W. (2005). Cronbach’s Alpha, Revelle’s Beta, Mc-
Donald’s Omega: Their relations with each and two alternative conceptualizations of reliabil-
ity. Psychometrika. 70, 123-133. http://personality-project.org/revelle/publications/
zinbarg.revelle.pmet.05.pdf
Zinbarg, R., Yovel, I., Revelle, W. & McDonald, R. (2006). Estimating generalizability to a
universe of indicators that all have one attribute in common: A comparison of estimators for
omega. Applied Psychological Measurement, 30, 121-144. DOI: 10.1177/0146621605278814
http://apm.sagepub.com/cgi/reprint/30/2/121
See Also
omega,make.hierarchical,ICLUST.rgraph
Examples
#24 mental tests from Holzinger-Swineford-Harman
if(require(GPArotation) ) {om24 <- omega(Harman74.cor$cov,4) } #run omega
#
#example hierarchical structure from Jensen and Weng
if(require(GPArotation) ) {jen.omega <- omega(make.hierarchical())}
outlier Find and graph Mahalanobis squared distances to detect outliers
Description
The Mahalanobis distance is D2= (x−µ)0Σ−1(x−µ)where Σis the covariance of the x matrix.
D2 may be used as a way of detecting outliers in distribution. Large D2 values, compared to the
expected Chi Square values indicate an unusual response pattern. The mahalanobis function in stats
does not handle missing data.
Usage
outlier(x, plot = TRUE, bad = 5,na.rm = TRUE, xlab, ylab, ...)
Arguments
xA data matrix or data.frame
plot Plot the resulting QQ graph
bad Label the bad worst values
na.rm Should missing data be deleted
xlab Label for x axis
ylab Label for y axis
... More graphic parameters, e.g., cex=.8
p.rep 231
Details
Adapted from the mahalanobis function and help page from stats.
Value
The D2 values for each case
Author(s)
William Revelle
References
Yuan, Ke-Hai and Zhong, Xiaoling, (2008) Outliers, Leverage Observations, and Influential Cases
in Factor Analysis: Using Robust Procedures to Minimize Their Effect, Sociological Methodology,
38, 329-368.
See Also
mahalanobis
Examples
#first, just find and graph the outliers
d2 <- outlier(sat.act)
#combine with the data frame and plot it with the outliers highlighted in blue
sat.d2 <- data.frame(sat.act,d2)
pairs.panels(sat.d2,bg=c("yellow","blue")[(d2 > 25)+1],pch=21)
p.rep Find the probability of replication for an F, t, or r and estimate effect
size
Description
The probability of replication of an experimental or correlational finding as discussed by Peter
Killeen (2005) is the probability of finding an effect in the same direction upon an exact replication.
For articles submitted to Psychological Science, p.rep needs to be reported.
F, t, p and r are all estimates of the size of an effect. But F, t, and p also are also a function of
the sample size. Effect size, d prime, may be expressed as differences between means compared to
within cell standard deviations, or as a correlation coefficient. These functions convert p, F, and t to
d prime and the r equivalent.
Usage
p.rep(p = 0.05, n=NULL,twotailed = FALSE)
p.rep.f(F,df2,twotailed=FALSE)
p.rep.r(r,n,twotailed=TRUE)
p.rep.t(t,df,df2=NULL,twotailed=TRUE)
232 p.rep
Arguments
pconventional probability of statistic (e.g., of F, t, or r)
FThe F statistic
df Degrees of freedom of the t-test, or of the first group if unequal sizes
df2 Degrees of freedom of the denominator of F or the second group in an unequal
sizes t test
rCorrelation coefficient
nTotal sample size if using r
tt-statistic if doing a t-test or testing significance of a regression slope
twotailed Should a one or two tailed test be used?
Details
The conventional Null Hypothesis Significance Test (NHST) is the likelihood of observing the data
given the null hypothesis of no effect. But this tells us nothing about the probability of the null
hypothesis. Peter Killeen (2005) introduced the probability of replication as a more useful measure.
The probability of replication is the probability that an exact replication study will find a result in
the same direction as the original result.
p.rep is based upon a 1 tailed probability value of the observed statistic.
Other frequently called for statistics are estimates of the effect size, expressed either as Cohen’s d,
Hedges g, or the equivalent value of the correlation, r.
For p.rep.t, if the cell sizes are unequal, the effect size estimates are adjusted by the ratio of the
mean cell size to the harmonic mean cell size (see Rownow et al., 2000).
Value
p.rep Probability of replication
dprime Effect size (Cohen‘s d) if more than just p is specified
prob Probability of F, t, or r. Note that this can be either the one-tailed or two tailed
probability value.
r.equivalent For t-tests, the r equivalent to the t (see Rosenthal and Rubin(2003), Rosnow,
Rosenthal, and Rubin, 2000))
.
Note
The p.rep value is the one tailed probability value of obtaining a result in the same direction.
References
Cummings, Geoff (2005) Understanding the average probability of replication: comment on Killeen
2005). Psychological Science, 16, 12, 1002-1004).
paired.r 233
Killeen, Peter H. (2005) An alternative to Null-Hypothesis Significance Tests. Psychological Sci-
ence, 16, 345-353
Rosenthal, R. and Rubin, Donald B.(2003), r-sub(equivalent): A Simple Effect Size Indicator. Psy-
chological Methods, 8, 492-496.
Rosnow, Ralph L., Rosenthal, Robert and Rubin, Donald B. (2000) Contrasts and correlations in
effect-size estimation, Psychological Science, 11. 446-453.
Examples
p.rep(.05) #probability of replicating a result if the original study had a p = .05
p.rep.f(9.0,98) #probability of replicating a result with F = 9.0 with 98 df
p.rep.r(.4,50) #probability of replicating a result if r =.4 with n = 50
p.rep.t(3,98) #probability of replicating a result if t = 3 with df =98
p.rep.t(2.14,84,14) #effect of equal sample sizes (see Rosnow et al.)
paired.r Test the difference between (un)paired correlations
Description
Test the difference between two (paired or unpaired) correlations. Given 3 variables, x, y, z, is the
correlation between xy different than that between xz? If y and z are independent, this is a simple
t-test of the z transformed rs. But, if they are dependent, it is a bit more complicated.
Usage
paired.r(xy, xz, yz=NULL, n, n2=NULL,twotailed=TRUE)
Arguments
xy r(xy)
xz r(xz)
yz r(yz)
nNumber of subjects for first group
n2 Number of subjects in second group (if not equal to n)
twotailed Calculate two or one tailed probability values
234 pairs.panels
Details
To find the z of the difference between two independent correlations, first convert them to z scores
using the Fisher r-z transform and then find the z of the difference between the two correlations.
The default assumption is that the group sizes are the same, but the test can be done for different
size groups by specifying n2.
If the correlations are not independent (i.e., they are from the same sample) then the correlation with
the third variable r(yz) must be specified. Find a t statistic for the difference of thee two dependent
correlations.
Value
a list containing the calculated t or z values and the associated two (or one) tailed probability.
tt test of the difference between two dependent correlations
pprobability of the t or of the z
zz test of the difference between two independent correlations
Author(s)
William Revelle
See Also
r.test for more tests of independent as well as dependent (paired) tests. p.rep.r for the probabil-
ity of replicating a particular correlation. cor.test from stats for testing a single correlation and
corr.test for finding the values and probabilities of multiple correlations. See also set.cor to do
multiple correlations from matrix input.
Examples
paired.r(.5,.3, .4, 100) #dependent correlations
paired.r(.5,.3,NULL,100) #independent correlations same sample size
paired.r(.5,.3,NULL, 100, 64) # independent correlations, different sample sizes
pairs.panels SPLOM, histograms and correlations for a data matrix
Description
Adapted from the help page for pairs, pairs.panels shows a scatter plot of matrices (SPLOM), with
bivariate scatter plots below the diagonal, histograms on the diagonal, and the Pearson correlation
above the diagonal. Useful for descriptive statistics of small data sets. If lm=TRUE, linear regres-
sion fits are shown for both y by x and x by y. Correlation ellipses are also shown. Points may be
given different colors depending upon some grouping variable.
pairs.panels 235
Usage
## S3 method for class 'panels'
pairs(x, smooth = TRUE, scale = FALSE, density=TRUE,ellipses=TRUE,
digits = 2,method="pearson", pch = 20, lm=FALSE,cor=TRUE,jiggle=FALSE,factor=2,
hist.col="cyan",show.points=TRUE,rug=TRUE, breaks = "Sturges",cex.cor=1,wt=NULL, ...)
Arguments
xa data.frame or matrix
smooth TRUE draws loess smooths
scale TRUE scales the correlation font by the size of the absolute correlation.
density TRUE shows the density plots as well as histograms
ellipses TRUE draws correlation ellipses
lm Plot the linear fit rather than the LOESS smoothed fits.
digits the number of digits to show
method method parameter for the correlation ("pearson","spearman","kendall")
pch The plot character (defaults to 20 which is a ’.’).
cor If plotting regressions, should correlations be reported?
jiggle Should the points be jittered before plotting?
factor factor for jittering (1-5)
hist.col What color should the histogram on the diagonal be?
show.points If FALSE, do not show the data points, just the data ellipses and smoothed func-
tions
rug if TRUE (default) draw a rug under the histogram, if FALSE, don’t draw the rug
breaks If specified, allows control for the number of breaks in the histogram (see the
hist function)
cex.cor If this is specified, this will change the size of the text in the correlations. this
allows one to also change the size of the points in the plot by specifying the
normal cex values. If just specifying cex, it will change the character size, if
cex.cor is specified, then cex will function to change the point size.
wt If specified, then weight the correlations by a weights matrix (see note for some
comments)
... other options for pairs
Details
Shamelessly adapted from the pairs help page. Uses panel.cor, panel.cor.scale, and panel.hist, all
taken from the help pages for pairs. Also adapts the ellipse function from John Fox’s car package.
pairs.panels is most useful when the number of variables to plot is less than about 6-10. It is
particularly useful for an initial overview of the data.
To show different groups with different colors, use a plot character (pch) between 21 and 25 and
then set the background color to vary by group. (See the second example).
236 pairs.panels
When plotting more than about 10 variables, it is useful to set the gap parameter to something less
than 1 (e.g., 0). Alternatively, consider using cor.plot
In addition, when plotting more than about 100-200 cases, it is useful to set the plotting character
to be a point. (pch=".")
Sometimes it useful to draw the correlation ellipses and best fitting lowess without the points.
(points.false=TRUE).
Value
A scatter plot matrix (SPLOM) is drawn in the graphic window. The lower off diagonal draws
scatter plots, the diagonal histograms, the upper off diagonal reports the Pearson correlation (with
pairwise deletion).
If lm=TRUE, then the scatter plots are drawn above and below the diagonal, each with a linear
regression fit. Useful to show the difference between regression lines.
Note
If the data are either categorical or character, this is flagged with an astrix for the variable name. If
character, they are changed to factors before plotting.
The wt parameter allows for scatter plots of the raw data while showing the weighted correlation
matrix (found by using cor.wt). The current implementation uses the first two columns of the
weights matrix for all analyses. This is useful, but not perfect. The use of this option would be to
plot the means from a statsBy analysis and then display the weighted correlations by specifying
the means and ns from the statsBy run. See the final (not run) example.
See Also
pairs which is the base from which pairs.panels is derived, cor.plot to do a heat map of correla-
tions, and scatter.hist to draw a single correlation plot with histograms and best fitted lines.
To find the probability "significance" of the correlations using normal theory, use corr.test. To
find confidence intervals using boot strapping procedures, use cor.ci. To graphically show confi-
dence intervals, see cor.plot.upperLowerCi.
Examples
pairs.panels(attitude) #see the graphics window
data(iris)
pairs.panels(iris[1:4],bg=c("red","yellow","blue")[iris$Species],
pch=21,main="Fisher Iris data by Species") #to show color grouping
pairs.panels(iris[1:4],bg=c("red","yellow","blue")[iris$Species],
pch=21+as.numeric(iris$Species),main="Fisher Iris data by Species",hist.col="red")
#to show changing the diagonal
#demonstrate not showing the data points
data(sat.act)
pairs.panels(sat.act,show.points=FALSE)
#better yet is to show the points as a period
parcels 237
pairs.panels(sat.act,pch=".")
#show many variables with 0 gap between scatterplots
# data(bfi)
# pairs.panels(bfi,show.points=FALSE,gap=0)
#plot raw data points and then the weighted correlations.
#output from statsBy
sb <- statsBy(sat.act,"education")
pairs.panels(sb$mean,wt=sb$n) #report the weighted correlations
#compare with
pairs.panels(sb$mean) #unweighed correlations
parcels Find miniscales (parcels) of size 2 or 3 from a set of items
Description
Given a set of n items, form n/2 or n/3 mini scales or parcels of the most similar pairs or triplets
of items. These may be used as the basis for subsequent scale construction or multivariate (e.g.,
factor) analysis.
Usage
parcels(x, size = 3, max = TRUE, flip=TRUE,congruence = FALSE)
keysort(keys)
Arguments
xA matrix/dataframe of items or a correlation/covariance matrix of items
size Form parcels of size 2 or size 3
flip if flip=TRUE, negative correlations lead to at least one item being negatively
scored
max Should item correlation/covariance be adjusted for their maximum correlation
congruence Should the correlations be converted to congruence coefficients?
keys Sort a matrix of keys to reflect item order as much as possible
Details
Items used in measuring ability or other aspects of personality are typically not very reliable. One
suggestion has been to form items into homogeneous item composites (HICs), Factorially Homoge-
neous Item Dimensions (FHIDs) or mini scales (parcels). Parcelling may be done rationally, facto-
rially, or empirically based upon the structure of the correlation/covariance matrix. link{parcels}
facilitates the finding of parcels by forming a keys matrix suitable for using in score.items. These
keys represent the n/2 most similar pairs or the n/3 most similar triplets.
238 partial.r
The algorithm is straightforward: For size = 2, the correlation matrix is searched for the highest
correlation. These two items form the first parcel and are dropped from the matrix. The procedure
is repeated until there are no more pairs to form.
For size=3, the three items with the greatest sum of variances and covariances with each other is
found. This triplet is the first parcel. All three items are removed and the procedure then identifies
the next most similar triplet. The procedure repeats until n/3 parcels are identified.
Value
keys A matrix of scoring keys to be used to form mini scales (parcels) These will be
in order of importance, that is, the first parcel (P1) will reflect the most similar
pair or triplet. The keys may also be sorted by average item order by using the
keysort function.
Author(s)
William Revelle
References
Cattell, R. B. (1956). Validation and intensification of the sixteen personality factor questionnaire.
Journal of Clinical Psychology , 12 (3), 205 -214.
See Also
score.items to score the parcels or iclust for an alternative way of forming item clusters.
Examples
parcels(Thurstone)
keys <- parcels(bfi)
keys <- keysort(keys)
score.items(keys,bfi)
partial.r Find the partial correlations for a set (x) of variables with set (y)
removed.
Description
A straightforward application of matrix algebra to remove the effect of the variables in the y set
from the x set. Input may be either a data matrix or a correlation matrix. Variables in x and y are
specified by location.
Usage
partial.r(m, x, y)
peas 239
Arguments
mA data or correlation matrix
xThe variable numbers associated with the X set.
yThe variable numbers associated with the Y set
Details
It is sometimes convenient to partial the effect of a number of variables (e.g., sex, age, education)
out of the correlations of another set of variables. This could be done laboriously by finding the
residuals of various multiple correlations, and then correlating these residuals. The matrix algebra
alternative is to do it directly. To find the confidence intervals and "significance" of the correlations,
use the corr.p function with n = n - s where s is the numer of covariates.
Value
The matrix of partial correlations.
Author(s)
William Revelle
References
Revelle, W. (in prep) An introduction to psychometric theory with applications in R. To be published
by Springer. (working draft available at http://personality-project.org/r/book/
See Also
mat.regress for a similar application for regression
Examples
jen <- make.hierarchical() #make up a correlation matrix
round(jen[1:5,1:5],2)
par.r <- partial.r(jen,c(1,3,5),c(2,4))
cp <- corr.p(par.r,n=98) #assumes the jen data based upon n =100.
print(cp,short=FALSE) #show the confidence intervals as well
peas Galton‘s Peas
Description
Francis Galton introduced the correlation coefficient with an analysis of the similarities of the parent
and child generation of 700 sweet peas.
240 peas
Usage
data(peas)
Format
A data frame with 700 observations on the following 2 variables.
parent The mean diameter of the mother pea for 700 peas
child The mean diameter of the daughter pea for 700 sweet peas
Details
Galton’s introduction of the correlation coefficient was perhaps the most important contribution to
the study of individual differences. This data set allows a graphical analysis of the data set. There
are two different graphic examples. One shows the regression lines for both relationships, the other
finds the correlation as well.
Source
Stanton, Jeffrey M. (2001) Galton, Pearson, and the Peas: A brief history of linear regression for
statistics intstructors, Journal of Statistics Education, 9. (retrieved from the web from http://www.amstat.org/publications/jse/v9n3/stanton.html)
reproduces the table from Galton, 1894, Table 2.
The data were generated from this table.
References
Galton, Francis (1877) Typical laws of heredity. paper presented to the weekly evening meeting
of the Royal Institution, London. Volume VIII (66) is the first reference to this data set. The data
appear in
Galton, Francis (1894) Natural Inheritance (5th Edition), New York: MacMillan).
See Also
The other Galton data sets: heights,galton,cubits
Examples
data(peas)
pairs.panels(peas,lm=TRUE,xlim=c(14,22),ylim=c(14,22),main="Galton's Peas")
describe(peas)
pairs.panels(peas,main="Galton's Peas")
phi 241
phi Find the phi coefficient of correlation between two dichotomous vari-
ables
Description
Given a 1 x 4 vector or a 2 x 2 matrix of frequencies, find the phi coefficient of correlation. Typical
use is in the case of predicting a dichotomous criterion from a dichotomous predictor.
Usage
phi(t, digits = 2)
Arguments
ta 1 x 4 vector or a 2 x 2 matrix
digits round the result to digits
Details
In many prediction situations, a dichotomous predictor (accept/reject) is validated against a di-
chotomous criterion (success/failure). Although a polychoric correlation estimates the underlying
Pearson correlation as if the predictor and criteria were continuous and bivariate normal variables,
and the tetrachoric correlation if both x and y are assumed to dichotomized normal distributions,
the phi coefficient is the Pearson applied to a matrix of 0’s and 1s.
The phi coefficient was first reported by Yule (1912), but should not be confused with the Yule Q
coefficient.
For a very useful discussion of various measures of association given a 2 x 2 table, and why one
should probably prefer the Yule Q coefficient, see Warren (2008).
Given a two x two table of counts
a b a+b (R1)
c d c+d (R2)
a+c(C1) b+d (C2) a+b+c+d (N)
convert all counts to fractions of the total and then \ Phi = [a- (a+b)*(a+c)]/sqrt((a+b)(c+d)(a+c)(b+d)
) =\ (a - R1 * C1)/sqrt(R1 * R2 * C1 * C2)
This is in contrast to the Yule coefficient, Q, where \ Q = (ad - bc)/(ad+bc) which is the same as \
[a- (a+b)*(a+c)]/(ad+bc)
Since the phi coefficient is just a Pearson correlation applied to dichotomous data, to find a matrix
of phis from a data set involves just finding the correlations using cor or lowerCor or corr.test.
Value
phi coefficient of correlation
242 phi.demo
Author(s)
William Revelle with modifications by Leo Gurtler
References
Warrens, Matthijs (2008), On Association Coefficients for 2x2 Tables and Properties That Do Not
Depend on the Marginal Distributions. Psychometrika, 73, 777-789.
Yule, G.U. (1912). On the methods of measuring the association between two attributes. Journal of
the Royal Statistical Society, 75, 579-652.
See Also
phi2tetra ,Yule,Yule.inv Yule2phi,tetrachoric and polychoric
Examples
phi(c(30,20,20,30))
phi(c(40,10,10,40))
x <- matrix(c(40,5,20,20),ncol=2)
phi(x)
phi.demo A simple demonstration of the Pearson, phi, and polychoric corelation
Description
A not very interesting demo of what happens if bivariate continuous data are dichotomized. Bas-
cially a demo of r, phi, and polychor.
Usage
phi.demo(n=1000,r=.6, cuts=c(-2,-1,0,1,2))
Arguments
nnumber of cases to simulate
rcorrelation between latent and observed
cuts form dichotomized variables at the value of cuts
Details
A demonstration of the problem of different base rates on the phi correlation, and how these are
partially solved by using the polychoric correlation. Not one of my more interesting demonstra-
tions. See http://personality-project.org/r/simulating-personality.html and http:
//personality-project.org/r/r.datageneration.html for better demonstrations of data gen-
eration.
phi2tetra 243
Value
a matrix of correlations and a graphic plot. The items above the diagonal are the tetrachoric corre-
lations, below the diagonal are raw correlations.
Author(s)
William Revelle
References
http://personality-project.org/r/simulating-personality.html and http://personality-project.
org/r/r.datageneration.html for better demonstrations of data generation.
See Also
VSS.simulate,item.sim
Examples
#demo <- phi.demo() #compare the phi (lower off diagonal and polychoric correlations
# (upper off diagonal)
#show the result from tetrachoric which corrects for zero entries by default
#round(demo$tetrachoric$rho,2)
#show the result from phi2poly
#tetrachorics above the diagonal, phi below the diagonal
#round(demo$phis,2)
phi2tetra Convert a phi coefficient to a tetrachoric correlation
Description
Given a phi coefficient (a Pearson r calculated on two dichotomous variables), and the marginal
frequencies (in percentages), what is the corresponding estimate of the tetrachoric correlation?
Given a two x two table of counts
a b
c d
The phi coefficient is (a - (a+b)*(a+c))/sqrt((a+b)(a+c)(b+d)(c+c)).
This function reproduces the cell entries for specified marginals and then calls the tetrachoric func-
tion. (Which was originally based upon John Fox’s polychor function.) The phi2poly name will
become deprecated in the future.
244 plot.psych
Usage
phi2tetra(ph,m,n=NULL,correct=TRUE)
phi2poly(ph,cp,cc,n=NULL,correct=TRUE) #deprecated
Arguments
ph phi
ma vector of the selection ratio and probability of criterion. In the case where ph
is a matrix, m is a vector of the frequencies of the selected cases
correct When finding tetrachoric correlations, should we correct for continuity for small
marginals. See tetrachoric for a discussion.
nIf the marginals are given as frequencies, what was the total number of cases?
cp probability of the predictor – the so called selection ratio
cc probability of the criterion – the so called success rate.
Details
used to require the mvtnorm package but this has been replaced with mnormt
Value
a tetrachoric correlation
Author(s)
William Revelle
See Also
tetrachoric,Yule2phi.matrix,phi2poly.matrix
Examples
phi2tetra(.3,c(.5,.5))
#phi2poly(.3,.3,.7)
plot.psych Plotting functions for the psych package of class “psych"
Description
Combines several plotting functions into one for objects of class “psych". This can be used to plot
the results of fa,irt.fa,VSS,ICLUST,omega,factor.pa, or principal.
plot.psych 245
Usage
## S3 method for class 'psych'
plot(x,labels=NULL,...)
## S3 method for class 'irt'
plot(x,xlab,ylab,main,D,type=c("ICC","IIC","test"),cut=.3,labels=NULL,
keys=NULL, xlim,ylim,y2lab,lncol="black",...)
## S3 method for class 'poly'
plot(x,D,xlab,ylab,xlim,ylim,main,type=c("ICC","IIC","test"),cut=.3,labels,
keys=NULL,y2lab,lncol="black",...)
## S3 method for class 'residuals'
plot(x,main,type=c("qq","chi","hist","cor"),std, bad=4,
numbers=TRUE, upper=FALSE,diag=FALSE,...)
Arguments
xThe object to plot
labels Variable labels
xlab Label for the x axis – defaults to Latent Trait
ylab Label for the y axis
xlim The limits for the x axis
ylim Specify the limits for the y axis
main Main title for graph
type "ICC" plots items, "IIC" plots item information, "test" plots test information, de-
faults to IIC.,"qq" does a quantile plot,"chi" plots chi square distributions,"hist"
shows the histogram,"cor" does a corPlot of the residuals.
DThe discrimination parameter
cut Only plot item responses with discrimiantion greater than cut
keys Used in plotting irt results from irt.fa.
y2lab ylab for test reliability, defaults to "reliability"
bad label the most 1.. bad items in residuals
numbers if using the cor option in plot residuals, show the numeric values
upper if using the cor option in plot residuals, show the upper off diagonal values
diag if using the cor option in plot residuals, show the diagonal values
std Standardize the resduals?
lncol The color of the lines in the IRT plots. Defaults to all being black, but it is
possible to specify lncol as a vector of colors to be used.
... other calls to plot
246 plot.psych
Details
Passes the appropriate values to plot. For plotting the results of irt.fa, there are three options:
type = "IIC" (default) will plot the item characteristic respone function. type = "IIC" will plot the
item information function, and type= "test" will plot the test information function.
Note that plotting an irt result will call either plot.irt or plot.poly depending upon the type of data
that were used in the original irt.fa call.
These are calls to the generic plot function that are intercepted for objects of type "psych". More
precise plotting control is available in the separate plot functions. plot may be used for psych objects
returned from fa,irt.fa,ICLUST,omega, as well as principal
A "jiggle" parameter is available in the fa.plot function (called from plot.psych when the type is
a factor or cluster. If jiggle=TRUE, then the points are jittered slightly (controlled by amount)
before plotting. This option is useful when plotting items with identical factor loadings (e.g., when
comparing hypothetical models).
Objects from irt.fa are plotted according to "type" (Item informations, item characteristics, or
test information). In addition, plots for selected items may be done if using the keys matrix. Plots
of irt information return three invisible objects, a summary of information for each item at levels of
the trait, the average area under the curve (the average information) for each item as well as where
the item is most informative.
If plotting multiple factor solutions in plot.poly, then main can be a vector of names, one for each
factor. The default is to give main + the factor number.
It is also possible to create irt like plots based upon just a scoring key and item difficulties, or from
a factor analysis and item difficulties. These are not true IRT type analyses, in that the parameters
are not estimated from the data, but are rather indications of item location and discrimination for
arbitrary sets of items. To do this, find irt.stats.like and then plot the results.
plot.residuals allows the user to graphically examine the residuals of models formed by fa,
irt.fa,omega, as well as principal and display them in a number of ways. "qq" will show
quantiles of standardized or unstandardized residuals, "chi" will show quantiles of the squared stan-
dardized or unstandardized residuals plotted against the expected chi square values, "hist" will draw
the histogram of the raw or standardized residuals, and "cor" will show a corPlot of the residual cor-
relations.
Value
Graphic output for factor analysis, cluster analysis and item response analysis.
Note
More precise plotting control is available in the separate plot functions.
Author(s)
William Revelle
See Also
VSS.plot and fa.plot,cluster.plot,fa,irt.fa,VSS,ICLUST,omega, or principal
polar 247
Examples
test.data <- Harman74.cor$cov
f4 <- fa(test.data,4)
plot(f4)
plot(resid(f4))
plot(resid(f4),main="Residuals from a 4 factor solution",qq=FALSE)
#not run
#data(bfi)
#e.irt <- irt.fa(bfi[11:15]) #just the extraversion items
#plot(e.irt) #the information curves
#
ic <- iclust(test.data,3) #shows hierarchical structure
plot(ic) #plots loadings
#
polar Convert Cartesian factor loadings into polar coordinates
Description
Factor and cluster analysis output typically presents item by factor correlations (loadings). Tables
of factor loadings are frequently sorted by the size of loadings. This style of presentation tends to
make it difficult to notice the pattern of loadings on other, secondary, dimensions. By converting to
polar coordinates, it is easier to see the pattern of the secondary loadings.
Usage
polar(f, sort = TRUE)
Arguments
fA matrix of loadings or the output from a factor or cluster analysis program
sort sort=TRUE: sort items by the angle of the items on the first pair of factors.
Details
Although many uses of factor analysis/cluster analysis assume a simple structure where items have
one and only one large loading, some domains such as personality or affect items have a more
complex structure and some items have high loadings on two factors. (These items are said to have
complexity 2, see VSS). By expressing the factor loadings in polar coordinates, this structure is more
readily perceived.
For each pair of factors, item loadings are converted to an angle with the first factor, and a vector
length corresponding to the amount of variance in the item shared with the two factors.
For a two dimensional structure, this will lead to a column of angles and a column of vector lengths.
For n factors, this leads to n* (n-1)/2 columns of angles and an equivalent number of vector lengths.
248 polychor.matrix
Value
polar A data frame of polar coordinates
Author(s)
William Revelle
References
Rafaeli, E. & Revelle, W. (2006). A premature consensus: Are happiness and sadness truly opposite
affects? Motivation and Emotion. \
Hofstee, W. K. B., de Raad, B., & Goldberg, L. R. (1992). Integration of the big five and circumplex
approaches to trait structure. Journal of Personality and Social Psychology, 63, 146-163.
See Also
ICLUST,cluster.plot,circ.tests,fa
Examples
circ.data <- circ.sim(24,500)
circ.fa <- fa(circ.data,2)
circ.polar <- round(polar(circ.fa),2)
circ.polar
#compare to the graphic
cluster.plot(circ.fa)
polychor.matrix Phi or Yule coefficient matrix to polychoric coefficient matrix
Description
A set of deprecated functions that have replaced by Yule2tetra and Yule2phi.
Some older correlation matrices were reported as matrices of Phi or of Yule correlations. That is,
correlations were found from the two by two table of counts:
a b
c d
Yule Q is (ad - bc)/(ad+bc).
With marginal frequencies of a+b, c+d, a+c, b+d.
Given a square matrix of such correlations, and the proportions for each variable that are in the a
predict.psych 249
+ b cells, it is possible to reconvert each correlation into a two by two table and then estimate the
corresponding polychoric correlation (using John Fox’s polychor function.
Usage
Yule2poly.matrix(x, v) #deprectated
phi2poly.matrix(x, v) #deprectated
Yule2phi.matrix(x, v) #deprectated
Arguments
xa matrix of phi or Yule coefficients
vA vector of marginal frequencies
Details
These functions call Yule2poly,Yule2phi or phi2poly for each cell of the matrix. See those
functions for more details. See phi.demo for an example.
Value
A matrix of correlations
Author(s)
William Revelle
Examples
#demo <- phi.demo()
#compare the phi (lower off diagonal and polychoric correlations (upper off diagonal)
#show the result from poly.mat
#round(demo$tetrachoric$rho,2)
#show the result from phi2poly
#tetrachorics above the diagonal, phi below the diagonal
#round(demo$phis,2)
predict.psych Prediction function for factor analysis or principal components
Description
Finds predicted factor/component scores from a factor analysis or components analysis of data set A
predicted to data set B. Predicted factor scores use the weights matrix used to find estimated factor
scores, predicted components use the loadings matrix. Scores are either standardized with respect
to the prediction sample or based upon the original data.
250 predict.psych
Usage
## S3 method for class 'psych'
predict(object, data,old.data,...)
Arguments
object the result of a factor analysis or principal components analysis of data set A
data Data set B, of the same number of variables as data set A.
old.data if specified, the data set B will be standardized in terms of values from the old
data. This is probably the preferred option.
... More options to pass to predictions
Value
Predicted factor/components scores. The scores are based upon standardized items where the stan-
dardization is either that of the original data (old.data) or of the prediction set. This latter case can
lead to confusion if just a small number of predicted scores are found.
Note
Thanks to Reinhold Hatzinger for the suggestion and request
Author(s)
William Revelle
See Also
fa,principal
Examples
set.seed(42)
x <- sim.item(12,500)
f2 <- fa(x[1:250,],2,scores="regression") # a two factor solution
p2 <- principal(x[1:250,],2,scores=TRUE) # a two component solution
round(cor(f2$scores,p2$scores),2) #correlate the components and factors from the A set
#find the predicted scores (The B set)
pf2 <- predict(f2,x[251:500,],x[1:250,])
#use the original data for standardization values
pp2 <- predict(p2,x[251:500,],x[1:250,])
#standardized based upon the first set
round(cor(pf2,pp2),2) #find the correlations in the B set
#test how well these predicted scores match the factor scores from the second set
fp2 <- fa(x[251:500,],2,scores=TRUE)
round(cor(fp2$scores,pf2),2)
pf2.n <- predict(f2,x[251:500,]) #Standardized based upon the new data set
round(cor(fp2$scores,pf2.n))
#predict factors of set two from factors of set 1, factor order is arbitrary
principal 251
#note that the signs of the factors in the second set are arbitrary
principal Principal components analysis (PCA)
Description
Does an eigen value decomposition and returns eigen values, loadings, and degree of fit for a spec-
ified number of components. Basically it is just doing a principal components analysis (PCA) for
n principal components of either a correlation or covariance matrix. Can show the residual correla-
tions as well. The quality of reduction in the squared correlations is reported by comparing residual
correlations to original correlations. Unlike princomp, this returns a subset of just the best nfactors.
The eigen vectors are rescaled by the sqrt of the eigen values to produce the component loadings
more typical in factor analysis.
Usage
principal(r, nfactors = 1, residuals = FALSE,rotate="varimax",n.obs=NA, covar=FALSE,
scores=TRUE,missing=FALSE,impute="median",oblique.scores=TRUE,method="regression",...)
Arguments
ra correlation matrix. If a raw data matrix is used, the correlations will be found
using pairwise deletions for missing values.
nfactors Number of components to extract
residuals FALSE, do not show residuals, TRUE, report residuals
rotate "none", "varimax", "quatimax", "promax", "oblimin", "simplimax", and "clus-
ter" are possible rotations/transformations of the solution. See fa for all rota-
tions avaiable.
n.obs Number of observations used to find the correlation matrix if using a correlation
matrix. Used for finding the goodness of fit statistics.
covar If false, find the correlation matrix from the raw data or convert to a correlation
matrix if given a square matrix as input.
scores If TRUE, find component scores
missing if scores are TRUE, and missing=TRUE, then impute missing values using either
the median or the mean
impute "median" or "mean" values are used to replace missing values
oblique.scores If TRUE (default), then the component scores are based upon the structure ma-
trix. If FALSE, upon the pattern matrix.
method Which way of finding component scores should be used. The default is "regres-
sion"
... other parameters to pass to functions such as factor.scores or the various rotation
functions.
252 principal
Details
Useful for those cases where the correlation matrix is improper (perhaps because of SAPA tech-
niques).
There are a number of data reduction techniques including principal components analysis (PCA) and
factor analysis (EFA). Both PC and FA attempt to approximate a given correlation or covariance
matrix of rank n with matrix of lower rank (p). nRn≈nFkk F0
n+U2where k is much less
than n. For principal components, the item uniqueness is assumed to be zero and all elements of
the correlation or covariance matrix are fitted. That is, nRn≈nFkkF0
nThe primary empirical
difference between a components versus a factor model is the treatment of the variances for each
item. Philosophically, components are weighted composites of observed variables while in the
factor model, variables are weighted composites of the factors.
For a n x n correlation matrix, the n principal components completely reproduce the correlation
matrix. However, if just the first k principal components are extracted, this is the best k dimensional
approximation of the matrix.
It is important to recognize that rotated principal components are not principal components (the
axes associated with the eigen value decomposition) but are merely components. To point this out,
unrotated principal components are labelled as PCi, while rotated PCs are now labeled as RCi (for
rotated components) and obliquely transformed components as TCi (for transformed components).
(Thanks to Ulrike Gromping for this suggestion.)
Rotations and transformations are either part of psych (Promax and cluster), of base R (varimax),
or of GPArotation (simplimax, quartimax, oblimin, etc.).
Of the various rotation/transformation options, varimax, Varimax, quartimax, bentlerT, geominT,
and bifactor do orthogonal rotations. Promax transforms obliquely with a target matix equal to the
varimax solution. oblimin, quartimin, simplimax, bentlerQ, geominQ and biquartimin are oblique
transformations. Most of these are just calls to the GPArotation package. The “cluster” option does
a targeted rotation to a structure defined by the cluster representation of a varimax solution. With
the optional "keys" parameter, the "target" option will rotate to a target supplied as a keys matrix.
(See target.rot.)
The rotation matrix (rot.mat) is returned from all of these options. This is the inverse of the Th
(theta?) object returned by the GPArotation package. The correlations of the factors may be found
by Φ = θ0θ
Some of the statistics reported are more appropriate for (maximum likelihood) factor analysis rather
than principal components analysis, and are reported to allow comparisons with these other models.
Although for items, it is typical to find component scores by scoring the salient items (using, e.g.,
score.items) component scores are found by regression where the regression weights are R−1λ
where λis the matrix of component loadings. The regression approach is done to be parallel with
the factor analysis function fa. The regression weights are found from the inverse of the correlation
matrix times the component loadings. This has the result that the component scores are standard
scores (mean=0, sd = 1) of the standardized input. A comparison to the scores from princomp
shows this difference. princomp does not, by default, standardize the data matrix, nor are the
components themselves standardized. The regression weights are found from the Structure matrix,
not the Pattern matrix. If the scores are found with the covar option = TRUE, then the scores are
not standardized but are just mean centered.
Jolliffe (2002) discusses why the interpretation of rotated components is complicated. The approach
used here is consistent with the factor analytic tradition. The correlations of the items with the
principal 253
component scores closely matches (as it should) the component loadings (as reported in the structure
matrix).
The output from the print.psych function displays the component loadings (from the pattern matrix),
the h2 (communalities) the u2 (the uniquenesses), com (the complexity of the component loadings
for that variable (see below). In the case of an orthogonal solution, h2 is merely the row sum of
the squared component loadings. But for an oblique solution, it is the row sum of the (squared)
orthogonal component loadings (remember, that rotations or transformations do not change the
communality).
Value
values Eigen Values of all components – useful for a scree plot
rotation which rotation was requested?
n.obs number of observations specified or found
communality Communality estimates for each item. These are merely the sum of squared
factor loadings for that item.
complexity Hoffman’s index of complexity for each item. This is just (Σa2
i)2
Σa4
i
where a_i
is the factor loading on the ith factor. From Hofmann (1978), MBR. See also
Pettersson and Turkheimer (2010).
loadings A standard loading matrix of class “loadings"
fit Fit of the model to the correlation matrix
fit.off how well are the off diagonal elements reproduced?
residual Residual matrix – if requested
dof Degrees of Freedom for this model. This is the number of observed correlations
minus the number of independent parameters (number of items * number of
factors - nf*(nf-1)/2. That is, dof = niI * (ni-1)/2 - ni * nf + nf*(nf-1)/2.
objective value of the function that is minimized by maximum likelihood procedures. This
is reported for comparison purposes and as a way to estimate chi square good-
ness of fit. The objective function is
f=log(trace((F F 0+U2)−1R)−log(|(F F 0+U2)−1R|)−n.items. Because
components do not minimize the off diagonal, this fit will be not as good as for
factor analysis.
STATISTIC If the number of observations is specified or found, this is a chi square based
upon the objective function, f. Using the formula from factanal:
χ2= (n.obs −1−(2 ∗p+ 5)/6−(2 ∗factors)/3)) ∗f
PVAL If n.obs > 0, then what is the probability of observing a chisquare this large or
larger?
phi If oblique rotations (using oblimin from the GPArotation package) are requested,
what is the interfactor correlation.
scores If scores=TRUE, then estimates of the factor scores are reported
weights The beta weights to find the principal components from the data
R2 The multiple R square between the factors and factor score estimates, if they
were to be found. (From Grice, 2001) For components, these are of course 1.0.
254 principal
valid The correlations of the component score estimates with the components, if they
were to be found and unit weights were used. (So called course coding).
rot.mat The rotation matrix used to produce the rotated component loadings.
Note
By default, the accuracy of the varimax rotation function seems to be less than the Varimax function.
This can be enhanced by specifying eps=1e-14 in the call to principal if using varimax rotation.
Furthermore, note that Varimax by default does not apply the Kaiser normalization, but varimax
does. Gottfried Helms compared these two rotations with those produced by SPSS and found
identical values if using the appropriate options. (See the last two examples.)
Author(s)
William Revelle
References
Grice, James W. (2001), Computing and evaluating factor scores. Psychological Methods, 6, 430-
450
Jolliffe, I. (2002) Principal Component Analysis (2nd ed). Springer.
Revelle, W. An introduction to psychometric theory with applications in R (in prep) Springer. Draft
chapters available at http://personality-project.org/r/book/
See Also
VSS (to test for the number of components or factors to extract), VSS.scree and fa.parallel
to show a scree plot and compare it with random resamplings of the data), factor2cluster
(for course coding keys), fa (for factor analysis), factor.congruence (to compare solutions),
predict.psych to find factor/component scores for a new data set based upon the weights from an
original data set.
Examples
#Four principal components of the Harman 24 variable problem
#compare to a four factor principal axes solution using factor.congruence
pc <- principal(Harman74.cor$cov,4,rotate="varimax")
mr <- fa(Harman74.cor$cov,4,rotate="varimax") #minres factor analysis
pa <- fa(Harman74.cor$cov,4,rotate="varimax",fm="pa") # principal axis factor analysis
round(factor.congruence(list(pc,mr,pa)),2)
pc2 <- principal(Harman.5,2,rotate="varimax")
pc2
round(cor(Harman.5,pc2$scores),2) #compare these correlations to the loadings
#now do it for unstandardized scores, and transform obliquely
pc2o <- principal(Harman.5,2,rotate="promax",covar=TRUE)
pc2o
round(cov(Harman.5,pc2o$scores),2)
pc2o$Structure #this matches the covariances with the scores
biplot(pc2,main="Biplot of the Harman.5 socio-economic variables",labels=paste0(1:12))
print.psych 255
#For comparison with SPSS (contributed by Gottfried Helms)
pc2v <- principal(iris[1:4],2,rotate="varimax",normalize=FALSE,eps=1e-14)
print(pc2v,digits=7)
pc2V <- principal(iris[1:4],2,rotate="Varimax",eps=1e-7)
print(pc2V,digits=7)
print.psych Print and summary functions for the psych class
Description
Give limited output (print) or somewhat more detailed (summary) for most of the functions in psych.
Usage
## S3 method for class 'psych'
print(x,digits=2,all=FALSE,cut=NULL,sort=FALSE,short=TRUE,lower=TRUE,...)
## S3 method for class 'psych'
summary(object,digits=2,items=FALSE,...)
## S3 method for class 'psych'
anova(object,object2,...)
Arguments
xOutput from a psych function (e.g., factor.pa, omega,ICLUST, score.items, clus-
ter.cor
object Output from a psych function
items items=TRUE (default) does not print the item whole correlations
digits Number of digits to use in printing
all if all=TRUE, then the object is declassed and all output from the function is
printed
cut Cluster loadings < cut will not be printed. For the factor analysis functions (fa
and factor.pa etc.), cut defaults to 0, for ICLUST to .3, for omega to .2.
sort Cluster loadings are in sorted order
short Controls how much to print
lower For square matrices, just print the lower half of the matrix
object2 Another object from fa or omega
... More options to pass to summary and print
256 Promax
Details
Most of the psych functions produce too much output. print.psych and summary.psych use generic
methods for printing just the highlights. To see what else is available, ask for the structure of the
particular object: (str(theobject).
Alternatively, to get complete output, unclass(theobject) and then print it. This may be done by
using the all=TRUE option.
As an added feature, if the promax function is applied to a factanal loadings matrix, the normal
output just provides the rotation matrix. print.psych will provide the factor correlations. (Following
a suggestion by John Fox and Uli Keller to the R-help list). The alternative is to just use the Promax
function directly on the factanal object.
Value
Various psych functions produce copious output. This is a way to summarize the most impor-
tant parts of the output of the score.items, cluster.scores, and ICLUST functions. See those (
score.items,cluster.cor,cluster.loadings, or ICLUST) for details on what is produced.
Note
See score.items,cluster.cor,cluster.loadings, or ICLUSTfor details on what is printed.
Author(s)
William Revelle
Examples
data(bfi)
keys.list <- list(agree=c(-1,2:5),conscientious=c(6:8,-9,-10),
extraversion=c(-11,-12,13:15),neuroticism=c(16:20),openness = c(21,-22,23,24,-25))
keys <- make.keys(25,keys.list,item.labels=colnames(bfi[1:25]))
scores <- score.items(keys,bfi[1:25])
scores
summary(scores)
Promax Perform bifactor, promax or targeted rotations and return the inter
factor angles.
Description
The bifactor rotation implements the rotation introduced by Jennrich and Bentler (2011) by call-
ing GPForth in the GPArotation package. promax is an oblique rotation function introduced by
Hendrickson and White (1964) and implemented in the promax function in the stats package. Un-
fortunately, promax does not report the inter factor correlations. Promax does. TargetQ does a target
rotation with elements that can be missing (NA), or numeric (e.g., 0, 1). It uses the GPArotation
package. target.rot does general target rotations to an arbitrary target matrix. The default target
Promax 257
rotation is for an independent cluster solution. equamax facilitates the call to GPArotation to do an
equamax rotation. Equamax, although available as a specific option within GPArotation is easier to
call by name if using equamax. The varimin rotation suggested by Ertl (2013) is implemented by
appropriate calls to GPArotation.
Usage
bifactor(L, Tmat=diag(ncol(L)), normalize=FALSE, eps=1e-5, maxit=1000)
biquartimin(L, Tmat=diag(ncol(L)), normalize=FALSE, eps=1e-5, maxit=1000)
TargetQ(L, Tmat=diag(ncol(L)), normalize=FALSE, eps=1e-5, maxit=1000,Target=NULL)
Promax(x, m = 4)
target.rot(x,keys=NULL)
varimin(L, Tmat = diag(ncol(L)), normalize = FALSE, eps = 1e-05, maxit = 1000)
vgQ.bimin(L) #called by bifactor
vgQ.targetQ(L,Target=NULL) #called by TargetQ
vgQ.varimin(L) #called by varimin
equamax(L, Tmat=diag(ncol(L)), eps=1e-5, maxit=1000)
Arguments
xA loadings matrix
mthe power to which to raise the varimax loadings (for Promax)
keys An arbitrary target matrix, can be composed of any weights, but probably -1,0,
1 weights. If missing, the target is the independent cluster structure determined
by assigning every item to it’s highest loaded factor.
LA loadings matrix
Target A matrix of values (mainly 0s, some 1s, some NAs) to which the matrix is
transformed.
Tmat An initial rotation matrix
normalize parameter passed to optimization routine (GPForth in the GPArotation package
eps parameter passed to optimization routine (GPForth in the GPArotation package)
maxit parameter passed to optimization routine (GPForth in the GPArotation package)
Details
The two most useful of these six functions is probably biquartimin which implements the oblique
bifactor rotation introduced by Jennrich and Bentler (2011). The second is TargetQ which allows
for missing NA values in the target. Next best is the orthogonal case, bifactor. None of these seem
to be implemented in GPArotation (yet).
The difference between biquartimin and bifactor is just that the latter is the orthogonal case which
is documented in Jennrich and Bentler (2011). It seems as if these two functions are sensitive to the
starting values and random restarts (modifying T) might be called for.
bifactor output for the 24 cognitive variable of Holzinger matches that of Jennrich and Bentler
as does output for the Chen et al. problem when fm="mle" is used and the Jennrich and Bentler
solution is rescaled from covariances to correlations.
258 Promax
Promax is a very direct adaptation of the stats::promax function. The addition is that it will return
the interfactor correlations as well as the loadings and rotation matrix.
varimin implements the varimin criterion proposed by Suitbert Ertl (2013). Rather than maximize
the varimax criterion, it minimizes it. For a discussion of the benefits of this procedure, consult
Ertel (2013).
In addition, these functions will take output from either the factanal, fa or earlier (factor.pa,
factor.minres or principal) functions and select just the loadings matrix for analysis.
equamax is just a call to GPArotation’s cFT function (for the Crawford Ferguson family of rotations.
TargetQ implements Michael Browne’s algorithm and allows specification of NA values. The Target
input is a list (see examples). It is interesting to note how powerful specifying what a factor isn’t
works in defining a factor. That is, by specifying the pattern of 0s and letting most other elements
be NA, the factor structure is still clearly defined.
The target.rot function is an adaptation of a function of Michael Browne’s to do rotations to arbitrary
target matrices. Suggested by Pat Shrout.
The default for target.rot is to rotate to an independent cluster structure (every items is assigned to
a group with its highest loading.)
target.rot will not handle targets that have linear dependencies (e.g., a pure bifactor model where
there is a g loading and a group factor for all variables).
Value
loadings Oblique factor loadings
rotmat The rotation matrix applied to the original loadings to produce the promax
soluion or the targeted matrix
Phi The interfactor correlation matrix
Note
A direct adaptation of the stats:promax function following suggestions to the R-help list by Ulrich
Keller and John Fox. Further modified to do targeted rotation similar to a function of Michael
Browne.
varimin is a direct application of the GPArotation GPForth function modified to do varimin.
Author(s)
William Revelle
References
Ertel, S. (2013). Factor analysis: healing an ailing model. Universitatsverlag Gottingen.
Hendrickson, A. E. and White, P. O, 1964, British Journal of Statistical Psychology, 17, 65-70.
Jennrich, Robert and Bentler, Peter (2011) Exploratory Bi-Factor Analysis. Psychometrika, 1-13
See Also
promax,fa, or principal for examples of data analysis and Holzinger or Bechtoldt for examples
of bifactor data. factor.rotate for ’hand rotation’.
psych.misc 259
Examples
jen <- sim.hierarchical()
f3 <- fa(jen,3,rotate="varimax")
f3 #not a very clean solution
Promax(f3)
target.rot(f3)
m3 <- fa(jen,nfactors=3)
Promax(m3) #example of taking the output from factanal
#compare this rotation with the solution from a targeted rotation aimed for
#an independent cluster solution
target.rot(m3)
#now try a bifactor solution
fb <-fa(jen,3,rotate="bifactor")
fq <- fa(jen,3,rotate="biquartimin")
#Suitbert Ertel has suggested varimin
fm <- fa(jen,3,rotate="varimin") #the Ertel varimin
fn <- fa(jen,3,rotate="none") #just the unrotated factors
#compare them
factor.congruence(list(f3,fb,fq,fm,fn))
# compare an oblimin with a target rotation using the Browne algorithm
#note that we are changing the factor #order (this is for demonstration only)
Targ <- make.keys(9,list(f1=1:3,f2=7:9,f3=4:6))
Targ <- scrub(Targ,isvalue=1) #fix the 0s, allow the NAs to be estimated
Targ <- list(Targ) #input must be a list
#show the target
Targ
fa(Thurstone,3,rotate="TargetQ",Target=Targ) #targeted rotation
#compare with oblimin
fa(Thurstone,3)
psych.misc Miscellaneous helper functions for the psych package
Description
This is a set of minor, if not trivial, helper functions. lowerCor finds the correlation of x variables
and then prints them using lowerMat which is a trivial, but useful, function to round off and print
the lower triangle of a matrix. reflect reflects the output of a factor analysis or principal components
analysis so that one or more factors is reflected. (Requested by Alexander Weiss.) progressBar
prints out ... as a calling routine (e.g., tetrachoric) works through a tedious calculation. shannon
finds the Shannon index (H) of diversity or of information. test.all tests all the examples in a
package. best.items sorts a factor matrix for absolute values and displays the expanded items names.
fa.lookup returns sorted factor analysis output with item labels.
Usage
psych.misc()
260 psych.misc
lowerCor(x,digits=2,use="pairwise",method="pearson")
lowerMat(R, digits = 2)
tableF(x,y)
reflect(f,flip=NULL)
progressBar(value,max,label=NULL)
shannon(x,correct=FALSE,base=2)
test.all(pl,package="psych",dependencies
= c("Depends", "Imports", "LinkingTo"),find=FALSE,skip=NULL)
Arguments
RA rectangular matrix or data frame (probably a correlation matrix)
xA data matrix or data frame or a vector depending upon the function.
yA data matrix or data frame or a vector
fThe object returned from either a factor analysis (fa) or a principal components
analysis (principal)
digits round to digits
use Should pairwise deletion be done, or one of the other options to cor
method "pearson", "kendall", "spearman"
value the current value of some looping variable
max The maximum value the loop will achieve
label what function is looping
flip The factor or components to be reversed keyed (by factor number)
correct Correct for the maximum possible information in this item
base What is the base for the log function (default=2, e implies base = exp(1))
pl The name of a package (or list of packages) to be activated and then have all the
examples tested.
package Find the dependencies for this package, e.g., psych
dependencies Which type of dependency to examine?
find Look up the dependencies, and then test all of their examples
skip Do not test these dependencies
Details
lowerCor prints out the lower off diagonal matrix rounded to digits with column names abbreviated
to digits + 3 characters, but also returns the full and unrounded matrix. By default, it uses pairwise
deletion of variables. It in turn calls
lowerMat which does the pretty printing.
It is important to remember to not call lowerCor when all you need is lowerMat!
psych.misc 261
Value
tableF is fast alternative to the table function for creating two way tables of numeric variables. It
does not have any of the elegant checks of the table function and thus is much faster. Used in the
tetrachoric and polychoric functions to maximize speed.
The lower triangle of a matrix, rounded to digits with titles abbreviated to digits + 3 (lowerMat) or
a series of dots (progressBar).
lowerCor prints the lower diagonal correlation matrix but returns (invisibly) the full correlation
matrix found with the use and method parameters. The default values are for pairwise deletion of
variables, and to print to 2 decimal places.
tableF (for tableFast) is a cut down version of table that does no error checking, nor returns pretty
output, but is significantly faster than table. It will just work on two integer vectors. This is used in
polychoric an tetrachoric for about a 50% speed improvement for large problems.
shannon finds Shannon’s H index of information. Used for estimating the complexity or diversity
of the distribution of responses in a vector or matrix.
H=−Xpilog(pi)
test.all allows one to test all the examples in specified package. This allows us to make sure
that those examples work when other packages (e.g., psych) are also loaded. This is used when
developing revisions to the psych package to make sure the the other packages work. Some pack-
ages will not work and/or crash the system (e.g., DeducerPlugInScaling requires Java and even
with Java, crashes when loaded, even if psych is not there!). Alternatively, if testing a long list of
dependencies, you can skip the first part by specifying them by name.
See Also
corr.test to find correlations, count the pairwise occurrences, and to give significance tests for
each correlation. r.test for a number of tests of correlations, including tests of the difference
between correlations. lowerUpper will display the differences between two matrices.
Examples
lowerMat(Thurstone)
lb <- lowerCor(bfi[1:10]) #finds and prints the lower correlation matrix,
# returns the square matrix.
#fiml <- corFiml(bfi[1:10]) #FIML correlations require lavaan package
#lowerMat(fiml) #to get pretty output
f3 <- fa(Thurstone,3)
f3r <- reflect(f3,2) #reflect the second factor
#find the complexity of the response patterns of the iqitems.
round(shannon(iqitems),2)
#test.all('BinNor') #Does the BinNor package work when we are using other packages
bestItems(lb,3,cut=.1)
#to make this a latex table
#df2latex(bestItems(lb,2,cut=.2))
#
data(bfi.dictionary)
f2 <- fa(bfi[1:10],2)
fa.lookup(f2,bfi.dictionary)
262 r.test
r.test Tests of significance for correlations
Description
Tests the significance of a single correlation, the difference between two independent correlations,
the difference between two dependent correlations sharing one variable (Williams’s Test), or the
difference between two dependent correlations with different variables (Steiger Tests).
Usage
r.test(n, r12, r34 = NULL, r23 = NULL, r13 = NULL, r14 = NULL, r24 = NULL,
n2 = NULL,pooled=TRUE, twotailed = TRUE)
Arguments
nSample size of first group
r12 Correlation to be tested
r34 Test if this correlation is different from r12, if r23 is specified, but r13 is not,
then r34 becomes r13
r23 if ra = r(12) and rb = r(13) then test for differences of dependent correlations
given r23
r13 implies ra =r(12) and rb =r(34) test for difference of dependent correlations
r14 implies ra =r(12) and rb =r(34)
r24 ra =r(12) and rb =r(34)
n2 n2 is specified in the case of two independent correlations. n2 defaults to n if if
not specified
pooled use pooled estimates of correlations
twotailed should a twotailed or one tailed test be used
Details
Depending upon the input, one of four different tests of correlations is done. 1) For a sample size
n, find the t value for a single correlation.
2) For sample sizes of n and n2 (n2 = n if not specified) find the z of the difference between the z
transformed correlations divided by the standard error of the difference of two z scores.
3) For sample size n, and correlations r12, r13 and r23 test for the difference of two dependent
correlations (r12 vs r13).
4) For sample size n, test for the difference between two dependent correlations involving different
variables.
For clarity, correlations may be specified by value. If specified by location and if doing the test
of dependent correlations, if three correlations are specified, they are assumed to be in the order
r12, r13, r23. Consider the example the example from Steiger: where Masculinity at time 1 (M1)
r.test 263
correlates with Verbal Ability .5 (r12), femininity at time 1 (F1) correlates with Verbal ability r13
=.4, and M1 correlates with F1 (r23= .1). Then, given the correlations: r12 = .4, r13 = .5, and r23 =
.1, t = -.89 for n =103, i.e., r.test(n=103, r12=.4, r13=.5,r23=.1)
Value
test Label of test done
zz value for tests 2 or 4
tt value for tests 1 and 3
pprobability value of z or t
Note
Steiger specifically rejects using the Hotelling T test to test the difference between correlated cor-
relations. Instead, he recommends Williams’ test. (See also Dunn and Clark, 1971). These tests
follow Steiger’s advice.
Author(s)
William Revelle
References
Olkin, I. and Finn, J. D. (1995). Correlations redux. Psychological Bulletin, 118(1):155-164.
Steiger, J.H. (1980), Tests for comparing elements of a correlation matrix, Psychological Bulletin,
87, 245-251.
Williams, E.J. (1959) Regression analysis. Wiley, New York, 1959.
See Also
See also corr.test which tests all the elements of a correlation matrix, and cortest.mat to com-
pare two matrices of correlations. r.test extends the tests in paired.r,r.con
Examples
n <- 30
r <- seq(0,.9,.1)
rc <- matrix(r.con(r,n),ncol=2)
test <- r.test(n,r)
r.rc <- data.frame(r=r,z=fisherz(r),lower=rc[,1],upper=rc[,2],t=test$t,p=test$p)
round(r.rc,2)
r.test(50,r)
r.test(30,.4,.6) #test the difference between two independent correlations
r.test(103,.4,.5,.1) #Steiger case A of dependent correlations
r.test(n=103, r12=.4, r13=.5,r23=.1)
#for complicated tests, it is probably better to specify correlations by name
r.test(n=103,r12=.5,r34=.6,r13=.7,r23=.5,r14=.5,r24=.8) #steiger Case B
264 rangeCorrection
rangeCorrection Correct correlations for restriction of range. (Thorndike Case 2)
Description
In applied settings, it is typical to find a correlation between a predictor and some criterion. Un-
fortunately, if the predictor is used to choose the subjects, the range of the predictor is seriously
reduced. This restricts the observed correlation to be less than would be observed in the full range
of the predictor. A correction for this problem is well known as Thorndike Case 2:
Let R the unrestricted correlaton, r the restricted correlation, S the unrestricted standard deviation,
s the restricted standard deviation, then
R = (rS/s)/ sqrt(1-r^2 + r^2(S^2/s^2)).
Several other cases of restriction were also considered by Thorndike and are implemented in rangeCorrection.
Usage
rangeCorrection(r,sdu,sdr,sdxu=NULL,sdxr=NULL,case=2)
Arguments
rThe observed correlation
sdu The unrestricted standard deviation)
sdr The restricted standard deviation
sdxu Unrestricted standard deviation for case 4
sdxr Restricted standard deviation for case 4
case Which of the four Thurstone/Stauffer cases to use
Details
When participants in a study are selected on one variable, that will reduce the variance of that
variable and the resulting correlation. Thorndike (1949) considered four cases of range restriction.
Others have continued this discussion but have changed the case numbers.
Can be used to find correlations in a restricted sample as well as the unrestricted sample. Not the
same as the correction to reliability for restriction of range.
Value
The corrected correlation.
Author(s)
William Revelle
read.clipboard 265
References
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
Stauffer, Joseph and Mendoza, Jorge. (2001) The proper sequence for correcting correlation coeffi-
cients for range restriction and unreliability. Psychometrika, 66, 63-68.
See Also
cRRr in the psychometric package.
Examples
rangeCorrection(.33,100.32,48.19) #example from Revelle (in prep) Chapter 4.
read.clipboard shortcut for reading from the clipboard
Description
Input from the clipboard is easy but a bit obscure, particularly for Mac users. This is just an easier
way to do so. Data may be copied to the clipboard from Exel spreadsheets, csv files, or fixed width
formatted files and then into a data.frame. Data may also be read from lower (or upper) triangular
matrices and filled out to square matrices.
Usage
read.clipboard(header = TRUE, ...) #assumes headers and tab or space delimited
read.clipboard.csv(header=TRUE,sep=',',...) #assumes headers and comma delimited
read.clipboard.tab(header=TRUE,sep='\t',...) #assumes headers and tab delimited
#read in a matrix given the lower off diagonal
read.clipboard.lower(diag=TRUE,names=FALSE,...)
read.clipboard.upper(diag=TRUE,names=FALSE,...)
#read in data using a fixed format width (see read.fwf for instructions)
read.clipboard.fwf(header=FALSE,widths=rep(1,10),...)
read.https(filename,header=TRUE)
Arguments
header Does the first row have variable labels
sep What is the designated separater between data fields?
diag for upper or lower triangular matrices, is the diagonal specified or not
names for read.clipboard.lower or upper, are colnames in the the first column
widths how wide are the columns in fixed width input. The default is to read 10 columns
of size 1.
266 read.clipboard
filename name or address of remote https file to read
... Other parameters to pass to read
Details
A typical session of R might involve data stored in text files, generated online, etc. Although
it is easy to just read from a file (particularly if using file.choose(), copying from the file to the
clipboard and then reading from the clipboard is also very convenient (and somewhat more intuitive
to the naive user). This is particularly convenient when copying from a text book or article and just
moving a section of text into R.)
Based upon a suggestion by Ken Knoblauch to the R-help listserve.
If the input file that was copied into the clipboard was an Excel file with blanks for missing data,
then read.clipboard.tab() will correctly replace the blanks with NAs. Similarly for a csv file with
blank entries, read.clipboard.csv will replace empty fields with NA.
read.clipboard.lower and read.clipboard.upper are adapted from John Fox’s read.moments function
in the sem package. They will read a lower (or upper) triangular matrix from the clipboard and
return a full, symmetric matrix for use by factanal, factor.pa ,ICLUST , etc. If the diagonal is
false, it will be replaced by 1.0s. These two function were added to allow easy reading of examples
from various texts and manuscripts with just triangular output.
Many articles will report lower triangular matrices with variable labels in the first column. read.clipboard.lower
will handle this case. Names must be in the first column if names=TRUE is specified.
Other articles will report upper triangular matrices with variable labels in the first row. read.clipboard.upper
will handle this. Note that labels in the first column will not work for read.clipboard.upper. The
names, if present, must be in the first row.
read.clipboard.fwf will read fixed format files from the clipboard. It includes a patch to read.fwf
which will not read from the clipboard or from remote file. See read.fwf for documentation of how
to specify the widths.
Value
the contents of the clipboard.
Author(s)
William Revelle
Examples
#my.data <- read.clipboad()
#my.data <- read.clipboard.csv()
#my.data <- read.clipboad(header=FALSE)
#my.matrix <- read.clipboard.lower()
rescale 267
rescale Function to convert scores to “conventional " metrics
Description
Psychologists frequently report data in terms of transformed scales such as “IQ" (mean=100, sd=15,
“SAT/GRE" (mean=500, sd=100), “ACT" (mean=18, sd=6), “T-scores" (mean=50, sd=10), or “Sta-
nines" (mean=5, sd=2). The rescale function converts the data to standard scores and then rescales
to the specified mean(s) and standard deviation(s).
Usage
rescale(x, mean = 100, sd = 15,df=TRUE)
Arguments
xA matrix or data frame
mean Desired mean of the rescaled scores- may be a vector
sd Desired standard deviation of the rescaled scores
df if TRUE, returns a data frame, otherwise a matrix
Value
A data.frame (default) or matrix of rescaled scores.
Author(s)
William Revelle
See Also
See Also scale
Examples
T <- rescale(attitude,50,10) #all put on same scale
describe(T)
T1 <- rescale(attitude,seq(0,300,50),seq(10,70,10)) #different means and sigmas
describe(T1)
268 residuals.psych
residuals.psych Extract residuals from various psych objects
Description
Residuals in the various psych functions are extracted and then may be "pretty" printed.
Usage
## S3 method for class 'psych'
residuals(object,...)
## S3 method for class 'psych'
resid(object,...)
Arguments
object The object returned by a psych function.
... Other parameters to be passed to residual (ignored but required by the generic
function)
Details
Currently implemented for fa,principal,omega,irt.fa, and fa.extension.
Value
residuals: a matrix of residual estimates
Author(s)
William Revelle
Examples
f3 <- fa(Thurstone,3)
residuals(f3)
reverse.code 269
reverse.code Reverse the coding of selected items prior to scale analysis
Description
Some IRT functions require all items to be coded in the same direction. Some data sets have items
that need to be reverse coded (e.g., 6 -> 1, 1 -> 6). reverse.code will flip items based upon a keys
vector of 1s and -1s. Reversed items are subtracted from the item max + item min. These may be
specified or may be calculated.
Usage
reverse.code(keys, items, mini = NULL, maxi = NULL)
Arguments
keys A vector of 1s and -1s. -1 implies reverse the item
items A data set of items
mini if NULL, the empirical minimum for each item. Otherwise, a vector of minima
maxi f NULL, the empirical maximum for each item. Otherwise, a vector of maxima
Details
Not a very complicated function, but useful in the case that items need to be reversed prior to using
IRT functions from the ltm or eRM packages. Most psych functions do not require reversing prior
to analysis, but will do so within the function.
Value
The corrected items.
Examples
original <- matrix(sample(6,50,replace=TRUE),10,5)
keys <- c(1,1,-1,-1,1) #reverse the 3rd and 4th items
new <- reverse.code(keys,original,mini=rep(1,5),maxi=rep(6,5))
original[1:3,]
new[1:3,]
270 sat.act
sat.act 3 Measures of ability: SATV, SATQ, ACT
Description
Self reported scores on the SAT Verbal, SAT Quantitative and ACT were collected as part of the
Synthetic Aperture Personality Assessment (SAPA) web based personality assessment project. Age,
gender, and education are also reported. The data from 700 subjects are included here as a demon-
stration set for correlation and analysis.
Usage
data(sat.act)
Format
A data frame with 700 observations on the following 6 variables.
gender males = 1, females = 2
education self reported education 1 = high school ... 5 = graduate work
age age
ACT ACT composite scores may range from 1 - 36. National norms have a mean of 20.
SATV SAT Verbal scores may range from 200 - 800.
SATQ SAT Quantitative scores may range from 200 - 800
Details
hese items were collected as part of the SAPA project (http://sapa-project.org)to develop
online measures of ability (Revelle, Wilt and Rosenthal, 2009). The score means are higher than
national norms suggesting both self selection for people taking on line personality and ability tests
and a self reporting bias in scores.
See also the iq.items data set.
Source
http://personality-project.org
References
Revelle, William, Wilt, Joshua, and Rosenthal, Allen (2009) Personality and Cognition: The Personality-
Cognition Link. In Gruszka, Alexandra and Matthews, Gerald and Szymura, Blazej (Eds.) Hand-
book of Individual Differences in Cognition: Attention, Memory and Executive Control, Springer.
Examples
data(sat.act)
describe(sat.act)
pairs.panels(sat.act)
scaling.fits 271
scaling.fits Test the adequacy of simple choice, logistic, or Thurstonian scaling.
Description
Given a matrix of choices and a vector of scale values, how well do the scale values capture the
choices? That is, what is size of the squared residuals given the model versus the size of the squared
choice values?
Usage
scaling.fits(model, data, test = "logit", digits = 2, rowwise = TRUE)
Arguments
model A vector of scale values
data A matrix or dataframe of choice frequencies
test "choice", "logistic", "normal"
digits Precision of answer
rowwise Are the choices ordered by column over row (TRUE) or row over column False)
Details
How well does a model fit the data is the classic problem of all of statistics. One fit statistic for
scaling is the just the size of the residual matrix compared to the original estimates.
Value
GF Goodness of fit of the model
original Sum of squares for original data
resid Sum of squares for residuals given the data and the model
residual Residual matrix
Note
Mainly for demonstration purposes for a course on psychometrics
Author(s)
William Revelle
References
Revelle, W. (in preparation) Introduction to psychometric theory with applications in R, Springer.
http://personality-project.org/r/book
272 scatter.hist
See Also
thurstone,vegetables
scatter.hist Draw a scatter plot with associated X and Y histograms, densitie and
correlation
Description
Draw a X Y scatter plot with associated X and Y histograms with estimated densities. Partly a
demonstration of the use of layout. Also includes lowess smooth or linear model slope, as well as
correlation. Adapted from addicted to R example 78
Usage
scatter.hist(x,y=NULL,smooth=TRUE,ab=FALSE,correl=TRUE,density=TRUE,ellipse=TRUE,
digits=2, method,cex.cor=1,title="Scatter plot + histograms",xlab=NULL,ylab=NULL,...)
Arguments
xThe X vector, or the first column of a data.frame or matrix.
yThe Y vector, of if X is a data.frame or matrix, the second column of X
smooth if TRUE, then loess smooth it
ab if TRUE, then show the best fitting linear fit
correl TRUE: Show the correlation
density TRUE: Show the estimated densities
ellipse TRUE: draw 1 and 2 sigma ellipses and smooth
digits How many digits to use if showing the correlation
method Which method to use for correlation ("pearson","spearman","kendall") defaults
to "pearson"
cex.cor Adjustment for the size of the correlation
xlab Label for the x axis
ylab Label for the y axis
title An optional title
... Other parameters for graphics
Details
Just a straightforward application of layout and barplot, with some tricks taken from pairs.panels.
The various options allow for correlation ellipses (1 and 2 sigma from the mean), lowess smooths,
linear fits, density curves on the histograms, and the value of the correlation. ellipse = TRUE implies
smooth = TRUE )
Schmid 273
Note
Adapted from Addicted to R example 78
Author(s)
William Revelle
See Also
pairs.panels for multiple plots, multi.hist for multiple histograms.
Examples
data(sat.act)
with(sat.act,scatter.hist(SATV,SATQ))
#or for something a bit more splashy
scatter.hist(sat.act[5:6],pch=(19+sat.act$gender),col=c("blue","red")[sat.act$gender])
Schmid 12 variables created by Schmid and Leiman to show the Schmid-
Leiman Transformation
Description
John Schmid and John M. Leiman (1957) discuss how to transform a hierarchical factor structure
to a bifactor structure. Schmid contains the example 12 x 12 correlation matrix. schmid.leiman is a
12 x 12 correlation matrix with communalities on the diagonal. This can be used to show the effect
of correcting for attenuation. Two additional data sets are taken from Chen et al. (2006).
Usage
data(Schmid)
Details
Two artificial correlation matrices from Schmid and Leiman (1957). One real and one artificial
covariance matrices from Chen et al. (2006).
• Schmid: a 12 x 12 artificial correlation matrix created to show the Schmid-Leiman transfor-
mation.
• schmid.leiman: A 12 x 12 matrix with communalities on the diagonal. Treating this as a
covariance matrix shows the 6 x 6 factor solution
• Chen: An 18 x 18 covariance matrix of health related quality of life items from Chen et
al. (2006). Number of observations = 403. The first item is a measure of the quality of
life. The remaining 17 items form four subfactors: The items are (a) Cognition subscale:
“Have difficulty reasoning and solving problems?" “React slowly to things that were said or
done?"; “Become confused and start several actions at a time?" “Forget where you put things
274 schmid
or appointments?"; “Have difficulty concentrating?" (b) Vitality subscale: “Feel tired?" “Have
enough energy to do the things you want?" (R) “Feel worn out?" ; “Feel full of pep?" (R). (c)
Mental health subscale: “Feel calm and peaceful?"(R) “Feel downhearted and blue?"; “Feel
very happy"(R) ; “Feel very nervous?" ; “Feel so down in the dumps nothing could cheer
you up? (d) Disease worry subscale: “Were you afraid because of your health?"; “Were you
frustrated about your health?"; “Was your health a worry in your life?" .
• West: A 16 x 16 artificial covariance matrix from Chen et al. (2006).
Source
John Schmid Jr. and John. M. Leiman (1957), The development of hierarchical factor solu-
tions.Psychometrika, 22, 83-90.
F.F. Chen, S.G. West, and K.H. Sousa.(2006) A comparison of bifactor and second-order models of
quality of life. Multivariate Behavioral Research, 41(2):189-225, 2006.
References
Y.-F. Yung, D.Thissen, and L.D. McLeod. (1999) On the relationship between the higher-order
factor model and the hierarchical factor model. Psychometrika, 64(2):113-128, 1999.
Examples
data(Schmid)
cor.plot(Schmid,TRUE)
print(fa(Schmid,6,rotate="oblimin"),cut=0) #shows an oblique solution
round(cov2cor(schmid.leiman),2)
cor.plot(cov2cor(West),TRUE)
schmid Apply the Schmid Leiman transformation to a correlation matrix
Description
One way to find omega is to do a factor analysis of the original data set, rotate the factors obliquely,
do a Schmid Leiman transformation, and then find omega. Here is the code for Schmid Leiman. The
S-L transform takes a factor or PC solution, transforms it to an oblique solution, factors the oblique
solution to find a higher order (g ) factor, and then residualizes g out of the the group factors.
Usage
schmid(model, nfactors = 3, fm = "minres",digits=2,rotate="oblimin",
n.obs=NA,option="equal",Phi=NULL,covar=FALSE,...)
schmid 275
Arguments
model A correlation matrix
nfactors Number of factors to extract
fm the default is to do minres. fm="pa" for principal axes, fm="pc" for principal
components, fm = "minres" for minimum residual (OLS), pc="ml" for maxi-
mum likelihood
digits if digits not equal NULL, rounds to digits
rotate The default, oblimin, produces somewhat more correlated factors than the alter-
native, simplimax. The third option is the promax criterion
n.obs Number of observations, used to find fit statistics if specified. Will be calculated
if input is raw data
option When asking for just two group factors, option can be for "equal", "first" or
"second"
Phi If Phi is specified, then the analysis is done on a pattern matrix with the asso-
ciated factor intercorrelation (Phi) matrix. This allows for reanalysess of pub-
lished results
covar Defaults to FALSE and finds correlations. If set to TRUE, then do the calcula-
tions on the unstandardized variables.
... Allows additional parameters to be passed to the factoring routines
Details
Schmid Leiman orthogonalizations are typical in the ability domain, but are not seen as often in the
non-cognitive personality domain. S-L is one way of finding the loadings of items on the general
factor for estimating omega.
A typical example would be in the study of anxiety and depression. A general neuroticism factor
(g) accounts for much of the variance, but smaller group factors of tense anxiety, panic disorder,
depression, etc. also need to be considerd.
An alternative model is to consider hierarchical cluster analysis techniques such as ICLUST.
Requires the GPArotation package.
Although 3 factors are the minimum number necessary to define the solution uniquely, it is occa-
sionally useful to allow for a two factor solution. There are three possible options for this condition:
setting the general factor loadings between the two lower order factors to be "equal" which will be
the sqrt(oblique correlations between the factors) or to "first" or "second" in which case the general
factor is equated with either the first or second group factor. A message is issued suggesting that
the model is not really well defined.
A diagnostic tool for testing the appropriateness of a hierarchical model is p2 which is the percent
of the common variance for each variable that is general factor variance. In general, p2 should not
have much variance.
Value
sl loadings on g + nfactors group factors, communalities, uniqueness, percent of
g2 of h2
276 score.alpha
orthog original orthogonal factor loadings
oblique oblique factor loadings
phi correlations among the transformed factors
gload loadings of the lower order factors on g
...
Author(s)
William Revelle
References
http://personality-project.org/r/r.omega.html gives an example taken from Jensen and
Weng, 1994 of a S-L transformation.
See Also
omega,omega.graph,fa.graph,ICLUST,VSS
Examples
jen <- sim.hierarchical() #create a hierarchical demo
if(!require(GPArotation)) {
message("I am sorry, you must have GPArotation installed to use schmid.")} else {
p.jen <- schmid(jen,digits=2) #use the oblimin rotation
p.jen <- schmid(jen,rotate="promax") #use the promax rotation
}
score.alpha Score scales and find Cronbach’s alpha as well as associated statistics
Description
Given a matrix or data.frame of k keys for m items (-1, 0, 1), and a matrix or data.frame of items
scores for m items and n people, find the sum scores or average scores for each person and each
scale. In addition, report Cronbach’s alpha, the average r, the scale intercorrelations, and the item
by scale correlations. (Superseded by score.items).
Usage
score.alpha(keys, items, labels = NULL, totals=TRUE,digits = 2) #deprecated
Arguments
keys A matrix or dataframe of -1, 0, or 1 weights for each item on each scale
items Data frame or matrix of raw item scores
labels column names for the resulting scales
totals Find sum scores (default) or average score
digits Number of digits for answer (default =2)
score.alpha 277
Details
This function has been replaced with score.items (for multiple scales) and alpha for single scales.
The process of finding sum or average scores for a set of scales given a larger set of items is a
typical problem in psychometric research. Although the structure of scales can be determined from
the item intercorrelations, to find scale means, variances, and do further analyses, it is typical to
find the sum or the average scale score.
Various estimates of scale reliability include “Cronbach’s alpha", and the average interitem correla-
tion. For k = number of items in a scale, and av.r = average correlation between items in the scale,
alpha = k * av.r/(1+ (k-1)*av.r). Thus, alpha is an increasing function of test length as well as the
test homeogeneity.
Alpha is a poor estimate of the general factor saturation of a test (see Zinbarg et al., 2005) for it can
seriously overestimate the size of a general factor, and a better but not perfect estimate of total test
reliability because it underestimates total reliability. None the less, it is a useful statistic to report.
Value
scores Sum or average scores for each subject on the k scales
alpha Cronbach’s coefficient alpha. A simple (but non-optimal) measure of the inter-
nal consistency of a test. See also beta and omega.
av.r The average correlation within a scale, also known as alpha 1 is a useful index
of the internal consistency of a domain.
n.items Number of items on each scale
cor The intercorrelation of all the scales
item.cor The correlation of each item with each scale. Because this is not corrected for
item overlap, it will overestimate the amount that an item correlates with the
other items in a scale.
Author(s)
William Revelle
References
An introduction to psychometric theory with applications in R (in preparation). http://personality-project.
org/r/book
See Also
score.items,alpha,correct.cor,cluster.loadings,omega
Examples
y <- attitude #from the datasets package
keys <- matrix(c(rep(1,7),rep(1,4),rep(0,7),rep(-1,3)),ncol=3)
labels <- c("first","second","third")
x <- score.alpha(keys,y,labels) #deprecated
278 score.irt
score.irt Find Item Response Theory (IRT) based scores for dichotomous or
polytomous items
Description
irt.fa finds Item Response Theory (IRT) parameters through factor analysis of the tetrachoric
or polychoric correlations of dichtomous or polytomous items. score.irt uses these parameter
estimates of discrimination and location to find IRT based scores for the responses. As many factors
as found for the correlation matrix will be scored.
Usage
score.irt(stats=NULL, items, keys=NULL,cut = 0.3,bounds=c(-5,5),mod="logistic")
#the higher order call just calls one of the next two
#for dichotomous items
score.irt.2(stats, items,keys=NULL,cut = 0.3,bounds=c(-5,5),mod="logistic")
#for polytomous items
score.irt.poly(stats, items, keys=NULL, cut = 0.3,bounds=c(-5,5))
#to create irt like statistics for plotting
irt.stats.like(items,stats,keys=NULL,cut=.3)
irt.tau(x)
Arguments
stats Output from irt.fa is used for parameter estimates of location and discrimination.
Stats may also be the output from a normal factor analysis (fa)
items The raw data, may be either dichotomous or polytomous.
keys A keys matrix of which items should be scored for each factor
cut Only items with discrimination values > cut will be used for scoring.
xThe raw data to be used to find the tau parameter in irt.tau
bounds The lower and upper estimates for the fitting function
mod Should a logistic or normal model be used in estimating the scores?
Details
Although there are more elegant ways of finding subject scores given a set of item locations (diffi-
culties) and discriminations, simply finding that value of theta θthat best fits the equation P(x|θ) =
1/(1+exp(β(δ−θ)) for a score vector X, and location δand discrimination βprovides more infor-
mation than just total scores. With complete data, total scores and irt estimates are almost perfectly
correlated. However, the irt estimates provide much more information in the case of missing data.
score.irt 279
The bounds parameter sets the lower and upper limits to the estimate. This is relevant for the case
of a subject who gives just the lowest score on every item, or just the top score on every item. In this
case, the scores are estimated by finding the probability of missing every item taken, converting this
to a quantile score based upon the normal distribution, and then assigning a z value equivalent to 1/2
of that quantile. Similarly, if a person gets all the items they take correct, their score is defined as
the quantile of the z equivalent to the probability of getting all of the items correct, and then moving
up the distribution half way. If these estimates exceed either the upper or lower bounds, they are
adjusted to those boundaries.
There are several more elegant packages in R that provide Full Information Maximum Likeliood
IRT based estimates. The estimates from score.irt do not do so. However, the score.irt seems to do
a good job of recovering the basic structure.
The keys matrix is a matrix of 1s, 0s, and -1s reflecting whether an item should be scored or not
scored for a particular factor. See score.items or make.keys for details. The default case is to
score all items with absolute discriminations > cut.
If one wants to score scales taking advantage of differences in item location but not do a full irt
analysis, then find the item difficulties from the raw data using irt.tau or combine this information
with a scoring keys matrix (see score.items and codemake.keys and create quasi-irt statistics using
irt.stats.like.
There are conventionally two different metrics and models that are used. The logistic metric and
model and the normal metric and model. These are chosen using the mod parameter.
Value
scores A data frame of theta estimates, total scores based upon raw sums, and estimates
of fit.
Note
Still under development. Suggestions for improvement are most appreciated.
score.irt is just a wrapper to score.irt.poly and score.irt.2
Author(s)
William Revelle
References
Kamata, Akihito and Bauer, Daniel J. (2008) A Note on the Relation Between Factor Analytic and
Item Response Theory Models Structural Equation Modeling, 15 (1) 136-153.
McDonald, Roderick P. (1999) Test theory: A unified treatment. L. Erlbaum Associates.
Revelle, William. (in prep) An introduction to psychometric theory with applications in R. Springer.
Working draft available at http://personality-project.org/r/book/
See Also
irt.fa for finding the parameters. For more conventional scoring algorithms see score.items.
irt.responses will plot the empirical response patterns for the alternative response choices for
multiple choice items. For more conventional IRT estimations, see the ltm package.
280 score.multiple.choice
Examples
if(FALSE) { #not run in the interest of time, but worth doing
d9 <- sim.irt(9,1000,-2.5,2.5,mod="normal") #dichotomous items
test <- irt.fa(d9$items)
scores <- score.irt(test,d9$items)
scores.df <- data.frame(scores,true=d9$theta) #combine the estimates with the true thetas.
pairs.panels(scores.df,pch=".",
main="Comparing IRT and classical with complete data")
#with all the data, why bother ?
#now delete some of the data
d9$items[1:333,1:3] <- NA
d9$items[334:666,4:6] <- NA
d9$items[667:1000,7:9] <- NA
scores <- score.irt(test,d9$items)
scores.df <- data.frame(scores,true=d9$theta) #combine the estimates with the true thetas.
pairs.panels(scores.df, pch=".",
main="Comparing IRT and classical with random missing data")
#with missing data, the theta estimates are noticably better.
}
v9 <- sim.irt(9,1000,-2.,2.,mod="normal") #dichotomous items
items <- v9$items
test <- irt.fa(items)
total <- rowSums(items)
ord <- order(total)
items <- items[ord,]
#now delete some of the data - note that they are ordered by score
items[1:333,5:9] <- NA
items[334:666,3:7] <- NA
items[667:1000,1:4] <- NA
scores <- score.irt(test,items)
unitweighted <- score.irt(items=items,keys=rep(1,9)) #each item has a discrimination of 1
#combine the estimates with the true thetas.
scores.df <- data.frame(v9$theta[ord],scores,unitweighted)
colnames(scores.df) <- c("True theta","irt theta","total","fit","rasch","total","fit")
pairs.panels(scores.df,pch=".",main="Comparing IRT and classical with missing data")
#with missing data, the theta estimates are noticably better estimates
#of the generating theta than calling them all equal
score.multiple.choice Score multiple choice items and provide basic test statistics
Description
Ability tests are typically multiple choice with one right answer. score.multiple.choice takes a
scoring key and a data matrix (or data.frame) and finds total or average number right for each
score.multiple.choice 281
participant. Basic test statistics (alpha, average r, item means, item-whole correlations) are also
reported.
Usage
score.multiple.choice(key, data, score = TRUE, totals = FALSE, ilabels = NULL,
missing = TRUE, impute = "median", digits = 2,short=TRUE,skew=FALSE)
Arguments
key A vector of the correct item alternatives
data a matrix or data frame of items to be scored.
score score=FALSE, just convert to right (1) or wrong (0).
score=TRUE, find the totals or average scores and do item analysis
totals total=FALSE: find the average number correct
total=TRUE: find the total number correct
ilabels item labels
missing missing=TRUE: missing values are replaced with means or medians
missing=FALSE missing values are not scored
impute impute="median", replace missing items with the median score
impute="mean": replace missing values with the item mean
digits How many digits of output
short short=TRUE, just report the item statistics,
short=FALSE, report item statistics and subject scores as well
skew Should the skews and kurtosi of the raw data be reported? Defaults to FALSE
because what is the meaning of skew for a multiple choice item?
Details
Basically combines score.items with a conversion from multiple choice to right/wrong.
The item-whole correlation is inflated because of item overlap.
The example data set is taken from the Synthetic Aperture Personality Assessment personality and
ability test at http://test.personality-project.org.
Value
scores Subject scores on one scale
missing Number of missing items for each subject
item.stats scoring key, response frequencies, item whole correlations, n subjects scored,
mean, sd, skew, kurtosis and se for each item
alpha Cronbach’s coefficient alpha
av.r Average interitem correlation
Author(s)
William Revelle
282 scoreItems
See Also
score.items,omega
Examples
data(iqitems)
iq.keys <- c(4,4,4, 6,6,3,4,4, 5,2,2,4, 3,2,6,7)
score.multiple.choice(iq.keys,iqitems)
#just convert the items to true or false
iq.tf <- score.multiple.choice(iq.keys,iqitems,score=FALSE)
describe(iq.tf) #compare to previous results
scoreItems Score item composite scales and find Cronbach’s alpha, Guttman
lambda 6 and item whole correlations
Description
Given a matrix or data.frame of k keys for n items (-1, 0, 1), and a matrix or data.frame of items
scores for m items and N people, find the sum scores or average scores for each person and each
scale. In addition, report Cronbach’s alpha, Guttman’s Lambda 6, the average r, the scale intercor-
relations, and the item by scale correlations (raw and corrected for item overlap). Replace missing
values with the item median or mean if desired. Will adjust scores for reverse scored items. See
make.keys for a convenient way to make the keys file. If the input is a square matrix, then it is
assumed that the input is a covariance or correlation matix and scores are not found, but the item
statistics are reported. (Similar functionality to cluster.cor). response.frequencies reports
the frequency of item endorsements fore each response category for polytomous or multiple choice
items.
Usage
scoreItems(keys, items, totals = FALSE, ilabels = NULL,missing=TRUE, impute="median",
delete=TRUE, min = NULL, max = NULL, digits = 2)
score.items(keys, items, totals = FALSE, ilabels = NULL,missing=TRUE, impute="median",
delete=TRUE, min = NULL, max = NULL, digits = 2)
response.frequencies(items,max=10,uniqueitems=NULL)
Arguments
keys A matrix or dataframe of -1, 0, or 1 weights for each item on each scale. May
be created by hand, or by using make.keys
items Matrix or dataframe of raw item scores
totals if TRUE find total scores, if FALSE (default), find average scores
ilabels a vector of item labels.
missing missing = TRUE is the normal case and data are imputed according to the impute
option. missing=FALSE, only complete cases are scored.
scoreItems 283
impute impute="median" replaces missing values with the item median, impute = "mean"
replaces values with the mean response. impute="none" the subject’s scores are
based upon the average of the keyed, but non missing scores.
delete if delete=TRUE, automatically delete items with no variance (and issue a warn-
ing)
min May be specified as minimum item score allowed, else will be calculated from
data. min and max should be specified if items differ in their possible minima
or maxima. See notes for details.
max May be specified as maximum item score allowed, else will be calculated from
data. Alternatively, in response frequencies, it is maximum number of alterna-
tive responses to count.
uniqueitems If specified, the set of possible unique response categories
digits Number of digits to report
Details
The process of finding sum or average scores for a set of scales given a larger set of items is a typical
problem in applied psychometrics and in psychometric research. Although the structure of scales
can be determined from the item intercorrelations, to find scale means, variances, and do further
analyses, it is typical to find scores based upon the sum or the average item score. For some strange
reason, personality scale scores are typically given as totals, but attitude scores as averages. The
default for scoreItems is the average as it would seem to make more sense to report scale scores in
the metric of the item.
Various estimates of scale reliability include “Cronbach’s alpha", Guttman’s Lambda 6, and the
average interitem correlation. For k = number of items in a scale, and av.r = average correlation
between items in the scale, alpha = k * av.r/(1+ (k-1)*av.r). Thus, alpha is an increasing function of
test length as well as the test homeogeneity.
Surprisingly, more than a century after Spearman (1904) introduced the concept of reliability to psy-
chologists, there are still multiple approaches for measuring it. Although very popular, Cronbach’s
α(1951) underestimates the reliability of a test and over estimates the first factor saturation.
α(Cronbach, 1951) is the same as Guttman’s λ3(Guttman, 1945) and may be found by
λ3=n
n−11−tr(~
V)x
Vx=n
n−1
Vx−tr(~
Vx)
Vx
=α
Perhaps because it is so easy to calculate and is available in most commercial programs, alpha is
without doubt the most frequently reported measure of internal consistency reliability. Alpha is the
mean of all possible spit half reliabilities (corrected for test length). For a unifactorial test, it is a
reasonable estimate of the first factor saturation, although if the test has any microstructure (i.e.,
if it is “lumpy") coefficients β(Revelle, 1979; see ICLUST) and ωh(see omega) (McDonald, 1999;
Revelle and Zinbarg, 2009) are more appropriate estimates of the general factor saturation. ωt(see
omega) is a better estimate of the reliability of the total test.
Guttman’s Lambda 6 (G6) considers the amount of variance in each item that can be accounted for
the linear regression of all of the other items (the squared multiple correlation or smc), or more
precisely, the variance of the errors, e2
j, and is
λ6= 1 −Pe2
j
Vx
= 1 −P(1 −r2
smc)
Vx
.
284 scoreItems
The squared multiple correlation is a lower bound for the item communality and as the number of
items increases, becomes a better estimate.
G6 is also sensitive to lumpyness in the test and should not be taken as a measure of unifactorial
structure. For lumpy tests, it will be greater than alpha. For tests with equal item loadings, alpha >
G6, but if the loadings are unequal or if there is a general factor, G6 > alpha. Although it is normal
when scoring just a single scale to calculate G6 from just those items within the scale, logically it is
appropriate to estimate an item reliability from all items available. This is done here and is labeled
as G6* to identify the subtle difference.
Alpha and G6* are both positive functions of the number of items in a test as well as the average
intercorrelation of the items in the test. When calculated from the item variances and total test
variance, as is done here, raw alpha is sensitive to differences in the item variances. Standardized
alpha is based upon the correlations rather than the covariances. alpha is a generalization of an
earlier estimate of reliability for tests with dichotomous items developed by Kuder and Richardson,
known as KR20, and a shortcut approximation, KR21. (See Revelle, in prep; Revelle and Condon,
in press.).
A useful index is the ratio of reliable variance to unreliable variance and is known as the Sig-
nal/Noise ratio. This is just
s/n =n¯r
1−n¯r
(Cronbach and Gleser, 1964; Revelle and Condon (in press)).
Standard errors for unstandardized alpha are reported using the formula from Duhachek and Ia-
cobucci (2005).
More complete reliability analyses of a single scale can be done using the omega function which
finds ωhand ωtbased upon a hierarchical factor analysis. Alternative estimates of the Greatest
Lower Bound for the reliability are found in the guttman function.
Alpha is a poor estimate of the general factor saturation of a test (see Revelle and Zinbarg, 2009;
Zinbarg et al., 2005) for it can seriously overestimate the size of a general factor, and a better but
not perfect estimate of total test reliability because it underestimates total reliability. None the less,
it is a common statistic to report. In general, the use of alpha should be discouraged and the use of
more appropriate estimates (ωhand ωt) should be encouraged.
Correlations between scales are attenuated by a lack of reliability. Correcting correlations for re-
liability (by dividing by the square roots of the reliabilities of each scale) sometimes help show
structure.
By default, missing values are replaced with the corresponding median value for that item. Means
can be used instead (impute="mean"), or subjects with missing data can just be dropped (missing
= FALSE). For data with a great deal of missingness, yet another option is to just find the average
of the available responses (impute="none"). This is useful for findings means for scales for the
SAPA project (see https://sapa-project.org) where most scales are estimated from random
sub samples of the items from the scale. In this case, the alpha reliabilities are seriously overinflated
because they are based upon the total number of items in each scale. The "alpha observed" values
are based upon the average number of items answered in each scale using the standard form for
alpha a function of inter-item correlation and number of items.
scoreItems can be applied to correlation matrices to find just the reliability statistics. This will be
done automatically if the items matrix is square and none of the values in the matrix are less than
-1 or greater than 1.
scoreItems 285
Value
scores Sum or average scores for each subject on the k scales
alpha Cronbach’s coefficient alpha. A simple (but non-optimal) measure of the inter-
nal consistency of a test. See also beta and omega. Set to 1 for scales of length
1.
av.r The average correlation within a scale, also known as alpha 1, is a useful index
of the internal consistency of a domain. Set to 1 for scales with 1 item.
G6 Guttman’s Lambda 6 measure of reliability
G6* A generalization of Guttman’s Lambda 6 measure of reliability using all the
items to find the smc.
n.items Number of items on each scale
item.cor The correlation of each item with each scale. Because this is not corrected for
item overlap, it will overestimate the amount that an item correlates with the
other items in a scale.
cor The intercorrelation of all the scales based upon the interitem correlations (see
note for why these differ from the correlations of the observed scales them-
selves).
corrected The correlations of all scales (below the diagonal), alpha on the diagonal, and
the unattenuated correlations (above the diagonal)
item.corrected The item by scale correlations for each item, corrected for item overlap by re-
placing the item variance with the smc for that item
response.freq The response frequency (based upon number of non-missing responses) for each
alternative.
missing How many items were not answered for each scale
num.ob.item The average number of items with responses on a scale. Used in calculating the
alpha.observed– relevant for SAPA type data structures.
Note
It is important to recognize in the case of massively missing data (e.g., data from a Synthetic Aper-
ture Personality Assessment (https://sapa-project.org) study where perhaps only 10-50% of
the items per scale are given to any one subject)) that the number of items per scale, and hence
standardized alpha, is not the nominal value and hence alpha of the observed scales will be overes-
timated. For this case (impute="none"), an additional alpha (alpha.ob) is reported.
More importantly in this case of massively missing data, there is a difference between the correla-
tions of the composite scales based upon the correlations of the items and the correlations of the
scored scales based upon the observed data. That is, the cor object will have correlations as if all
items had been given, while the correlation of the scores object will reflect the actual correlation of
the scores. For SAPA data, it is recommended to use the cor object. Confidence of these correlations
may be found using the cor.ci function.
Further note that the inter-scale correlations are based upon the correlations of scales formed from
the covariance matrix of the items. This will differ from the correlation of scales based upon the
correlation of the items. Thus, although scoreItems will produce reliabilities and intercorrelations
from either the raw data or from a correlation matrix, these values will differ slightly. In addition,
286 scoreItems
with a great deal of missing data, the scale intercorrelations will differ from the correlations of the
scores produced, for the latter will be attenuated.
An alternative to classical test theory scoring is to use score.irt to find score estimates based
upon Item Response Theory. This is particularly useful in the case of SAPA data which tend to be
massively missing. It is also useful to find scores based upon polytomous items following a factor
analysis of the polychoric correlation matrix (see irt.fa).
When reverse scoring items from a set where items differ in their possible minima or maxima, it is
important to specify the min and max values. Items are reversed by subtracting them from max +
min. Thus, if items range from 1 to 6, items are reversed by subtracting them from 7. But, if the
data set includes other variables, (say an id field) that far exceeds the item min or max, then the max
id will incorrectly be used to reverse key. min and max can either be single values, or vectors for all
items.
Author(s)
William Revelle
References
Cronbach, L.J. and Gleser G.C. (1964)The signal/noise ratio in the comparison of reliability coeffi-
cients. Educational and Psychological Measurement, 24 (3) 467-480.
Duhachek, A. and Iacobucci, D. (2004). Alpha’s standard error (ase): An accurate and precise
confidence interval estimate. Journal of Applied Psychology, 89(5):792-808.
McDonald, R. P. (1999). Test theory: A unified treatment. L. Erlbaum Associates, Mahwah, N.J.
Revelle, W. (in preparation) An introduction to psychometric theory with applications in R. http:
//personality-project.org/r/book
Revelle, W. and Condon, D.C. Reliability. In Irwing, P., Booth, T. and Hughes, D. (Eds). the
Wiley-Blackwell Handbook of Psychometric Testing (in press).
Revelle W. and R.E. Zinbarg. (2009) Coefficients alpha, beta, omega and the glb: comments on
Sijtsma. Psychometrika, 74(1):145-154.
Zinbarg, R. E., Revelle, W., Yovel, I. and Li, W. (2005) Cronbach’s alpha, Revelle’s beta, and
McDonald’s omega h, Their relations with each other and two alternative conceptualizations of
reliability, Psychometrika, 70, 123-133.
See Also
make.keys for a convenient way to create the keys file, score.multiple.choice for multiple
choice items,
alpha,correct.cor,cluster.cor ,cluster.loadings,omega,guttman for item/scale analysis.
If scales are formed from overlapping sets of items, their correlations will be inflated. This is
corrected for when using the scoreOverlap function which, although it will not produce scores,
will report scale intercorrelations corrected for item overlap.
In addition, the irt.fa function provides an alternative way of examining the structure of a test and
emphasizes item response theory approaches to the information returned by each item and the total
test. Associated with these IRT parameters is the score.irt function for finding IRT based scores
as well as irt.responses to show response curves for the alternatives in a multiple choice test.
scoreOverlap 287
Examples
#see the example including the bfi data set
data(bfi)
keys.list <- list(agree=c("-A1","A2","A3","A4","A5"),
conscientious=c("C1","C2","C3","-C4","-C5"),extraversion=c("-E1","-E2","E3","E4","E5"),
neuroticism=c("N1","N2","N3","N4","N5"), openness = c("O1","-O2","O3","O4","-O5"))
keys <- make.keys(bfi,keys.list)
scores <- scoreItems(keys,bfi,min=1,max=6)
summary(scores)
#to get the response frequencies, we need to not use the age variable
scores <- scoreItems(keys[1:27,],bfi[1:27],min=1,max=6)
scores
#The scores themselves are available in the scores$scores object. I.e.,
describe(scores$scores)
#compare this output to that for the impute="none" option for SAPA type data
#first make many of the items missing in a missing pattern way
missing.bfi <- bfi
missing.bfi[1:1000,3:8] <- NA
missing.bfi[1001:2000,c(1:2,9:10)] <- NA
scores <- scoreItems(keys,missing.bfi,impute="none",min=1,max=6)
scores
describe(scores$scores) #the actual scores themselves
scoreOverlap Find correlations of composite variables (corrected for overlap) from
a larger matrix.
Description
Given a n x c cluster definition matrix of -1s, 0s, and 1s (the keys) , and a n x n correlation matrix,
or an N x n data matrix, find the correlations of the composite clusters. The keys matrix can
be entered by hand, copied from the clipboard (read.clipboard), or taken as output from the
factor2cluster or make.keys functions. Similar functionality to scoreItems which also gives
item by cluster correlations.
Usage
scoreOverlap(keys, r, correct = TRUE, SMC = TRUE, av.r = TRUE, item.smc = NULL,
impute = TRUE)
cluster.cor(keys, r.mat, correct = TRUE,SMC=TRUE,item.smc=NULL,impute=TRUE)
Arguments
keys A matrix of cluster keys
288 scoreOverlap
r.mat A correlation matrix
rEither a correlation matrix or a raw data matrix
correct TRUE shows both raw and corrected for attenuation correlations
SMC Should squared multiple correlations be used as communality estimates for the
correlation matrix?
item.smc the smcs of the items may be passed into the function for speed, or calculated if
SMC=TRUE
impute if TRUE, impute missing scale correlations based upon the average interitem
correlation, otherwise return NA.
av.r Should the average r be used in correcting for overlap? smcs otherwise.
Details
This are two of the functions used in the SAPA (http://sapa-project.org) procedures to form
synthetic correlation matrices. Given any correlation matrix of items, it is easy to find the correlation
matrix of scales made up of those items. This can also be done from the original data matrix
or from the correlation matrix using scoreItems which is probably preferred unless the keys are
overlapping.
In the case of overlapping keys, (items being scored on multiple scales), scoreOverlap will adjust
for this overlap by replacing the overlapping covariances (which are variances when overlapping)
with the corresponding best estimate of an item’s “true" variance using either the average correlation
or the smc estimate for that item. This parallels the operation done when finding alpha reliability.
This is similar to ideas suggested by Cureton (1966) and Bashaw and Anderson (1966) but uses the
smc or the average interitem correlation (default).
A typical use in the SAPA project is to form item composites by clustering or factoring (see fa,
ICLUST,principal), extract the clusters from these results (factor2cluster), and then form the
composite correlation matrix using cluster.cor. The variables in this reduced matrix may then be
used in multiple correlatin procedures using mat.regress.
The original correlation is pre and post multiplied by the (transpose) of the keys matrix.
If some correlations are missing from the original matrix this will lead to missing values (NA) for
scale intercorrelations based upon those lower level correlations. If impute=TRUE (the default),
a warning is issued and the correlations are imputed based upon the average correlations of the
non-missing elements of each scale.
Because the alpha estimate of reliability is based upon the correlations of the items rather than upon
the covariances, this estimate of alpha is sometimes called “standardized alpha". If the raw items are
available, it is useful to compare standardized alpha with the raw alpha found using scoreItems.
They will differ substantially only if the items differ a great deal in their variances.
scoreOverlap answers an important question when developing scales and related subscales, or
when comparing alternative versions of scales. For by removing the effect of item overlap, it gives
a better estimate the relationship between the latent variables estimated by the observed sum (mean)
scores.
Value
cor the (raw) correlation matrix of the clusters
scoreOverlap 289
sd standard deviation of the cluster scores
corrected raw correlations below the diagonal, alphas on diagonal, disattenuated above
diagonal
alpha The (standardized) alpha reliability of each scale.
G6 Guttman’s Lambda 6 reliability estimate is based upon the smcs for each item
in a scale. G6 uses the smc based upon the entire item domain.
av.r The average inter item correlation within a scale
size How many items are in each cluster?
Note
See SAPA Revelle, W., Wilt, J., and Rosenthal, A. (2010) Personality and Cognition: The Personality-
Cognition Link. In Gruszka, A. and Matthews, G. and Szymura, B. (Eds.) Handbook of Individual
Differences in Cognition: Attention, Memory and Executive Control, Springer.
The second example uses the msq data set of 72 measures of motivational state to examine the
overlap between four lower level scales and two higher level scales.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
References
Bashaw, W. and Anderson Jr, H. E. (1967). A correction for replicated error in correlation coeffi-
cients. Psychometrika, 32(4):435-441.
Cureton, E. (1966). Corrected item-test correlations. Psychometrika, 31(1):93-96.
See Also
factor2cluster,mat.regress,alpha, and most importantly, scoreItems, which will do all of
what cluster.cor does for most users. cluster.cor is an important helper function for iclust
Examples
#use the msq data set that shows the structure of energetic and tense arousal
small.msq <- msq[ c("active", "energetic", "vigorous", "wakeful", "wide.awake",
"full.of.pep", "lively", "sleepy", "tired", "drowsy","intense", "jittery", "fearful",
"tense", "clutched.up", "quiet", "still", "placid", "calm", "at.rest") ]
small.R <- cor(small.msq,use="pairwise")
keys <- make.keys(small.R,list(
EA = c("active", "energetic", "vigorous", "wakeful", "wide.awake", "full.of.pep",
"lively", "-sleepy", "-tired", "-drowsy"),
TA =c("intense", "jittery", "fearful", "tense", "clutched.up", "-quiet", "-still",
"-placid", "-calm", "-at.rest") ,
high.EA = c("active", "energetic", "vigorous", "wakeful", "wide.awake", "full.of.pep",
"lively"),
low.EA =c("sleepy", "tired", "drowsy"),
lowTA= c("quiet", "still", "placid", "calm", "at.rest"),
290 scrub
highTA = c("intense", "jittery", "fearful", "tense", "clutched.up")
))
adjusted.scales <- scoreOverlap(keys,small.R)
#compare with unadjusted
confounded.scales <- cluster.cor(keys,small.R)
summary(adjusted.scales)
summary(confounded.scales)
scrub A utility for basic data cleaning and recoding. Changes values outside
of minimum and maximum limits to NA.
Description
A tedious part of data analysis is addressing the problem of miscoded data that need to be converted
to NA or some other value. For a given data.frame or matrix, scrub will set all values of columns
from=from to to=to that are less than a set (vector) of min values or more than a vector of max
values to NA. Can also be used to do basic recoding of data for all values=isvalue to newvalue.
The length of the where, isvalue, and newvalues must either match, or be 1.
Usage
scrub(x, where, min, max,isvalue,newvalue)
Arguments
xa data frame or matrix
where The variables to examine. (Can be by name or by column number)
min a vector of minimum values that are acceptable
max a vector of maximum values that are acceptable
isvalue a vector of values to be converted to newvalue (one per variable)
newvalue a vector of values to replace those that match isvalue
Details
Solves a tedious problem that can be done directly but that is sometimes awkward. Will either
replace specified values with NA or
Value
The corrected data frame.
Note
Probably could be optimized to avoid one loop
SD 291
Author(s)
William Revelle
See Also
reverse.code,rescale for other simple utilities.
Examples
data(attitude)
x <- scrub(attitude,isvalue=55) #make all occurrences of 55 NA
x1 <- scrub(attitude, where=c(4,5,6), isvalue =c(30,40,50),
newvalue = c(930,940,950)) #will do this for the 4th, 5th, and 6th variables
x2 <- scrub(attitude, where=c(4,4,4), isvalue =c(30,40,50),
newvalue = c(930,940,950)) #will just do it for the 4th column
#get rid of a complicated set of cases and replace with missing values
y <- scrub(attitude,where=2:4,min=c(20,30,40),max= c(120,110,100),isvalue= c(32,43,54))
y1 <- scrub(attitude,where="learning",isvalue=55,newvalue=999) #change a column by name
y2 <- scrub(attitude,where="learning",min=45,newvalue=999) #change a column by name
y3 <- scrub(attitude,where="learning",isvalue=c(45,48),
newvalue=999) #change a column by name look for multiple values in that column
y4 <- scrub(attitude,where="learning",isvalue=c(45,48),
newvalue= c(999,-999)) #change values in one column to one of two different things
SD Find the Standard deviation for a vector, matrix, or data.frame - do
not return error if there are no cases
Description
Find the standard deviation of a vector, matrix, or data.frame. In the latter two cases, return the sd
of each column. Unlike the sd function, return NA if there are no observations rather than throw an
error.
Usage
SD(x, na.rm = TRUE) #deprecated
Arguments
xa vector, data.frame, or matrix
na.rm na.rm is assumed to be TRUE
Details
Finds the standard deviation of a vector, matrix, or data.frame. Returns NA if no cases.
Just an adaptation of the stats:sd function to return the functionality found in R < 2.7.0 or R >=
2.8.0 Because this problem seems to have been fixed, SD will be removed eventually.
292 setCor
Value
The standard deviation
Note
Until R 2.7.0, sd would return a NA rather than an error if no cases were observed. SD brings back
that functionality. Although unusual, this condition will arise when analyzing data with high rates
of missing values. This function will probably be removed as 2.7.0 becomes outdated.
Author(s)
William Revelle
See Also
These functions use SD rather than sd: describe.by,skew,kurtosi
Examples
data(attitude)
apply(attitude,2,sd) #all complete
attitude[,1] <- NA
SD(attitude) #missing a column
describe(attitude)
setCor Set Correlation and Multiple Regression from matrix or raw input
Description
Finds Cohen’s Set Correlation between a predictor set of variables (x) and a criterion set (y). Also
finds multiple correlations between x variables and each of the y variables. Will work with either
raw data or a correlation matrix. A set of covariates (z) can be partialled from the x and y sets.
Regression diagrams are automatically included.
Usage
setCor(y,x,data,z=NULL,n.obs=NULL,use="pairwise",std=TRUE,square=FALSE,
main="Regression Models",plot=TRUE)
setCor.diagram(sc,main="Regression model",digits=2,show=TRUE,...)
set.cor(y,x,data,z=NULL,n.obs=NULL,use="pairwise",std=TRUE,square=FALSE,
main="Regression Models",plot=TRUE) #an alias to setCor
mat.regress(y, x,data, z=NULL,n.obs=NULL,use="pairwise",square=FALSE)
matReg(x,y,C,n.obs=0)
setCor 293
Arguments
yeither the column numbers of the y set (e.g., c(2,4,6) or the column names of the
y set (e.g., c("Flags","Addition")
xeither the column numbers of the x set (e.g., c(1,3,5) or the column names of the
x set (e.g. c("Cubes","PaperFormBoard")
data a matrix or data.frame of correlations or, if not square, of raw data
CA variance/covariance matrix, or a correlation matrix
zthe column names or numbers of the set of covariates
n.obs If specified, then confidence intervals, etc. are calculated, not needed if raw data
are given
use find the correlations "pairwise" (default) or just use "complete" cases (to match
the lm function)
std Report standardized betas (based upon the correlations) or raw (based upon co-
variances)
main The title for setCor.diagram
square if FALSE, then square matrices are treated as correlation matrices not as data
matrices. In the rare case that one has as many cases as variables, then set
square=TRUE.
sc The output of setCor may be used for drawing diagrams
digits How many digits should be displayed in the setCor.diagram?
show Show the matrix correlation between the x and y sets?
plot By default, setCor makes a plot of the results, set to FALSE to suppress the plot
... Additional graphical parameters for setCor.diagram
Details
Although it is more common to calculate multiple regression and canonical correlations from the
raw data, it is, of course, possible to do so from a matrix of correlations or covariances. In this case,
the input to the function is a square covariance or correlation matrix, as well as the column numbers
(or names) of the x (predictor), y (criterion) variables, and if desired z (covariates). The function
will find the correlations if given raw data.
The output is a set of multiple correlations, one for each dependent variable in the y set, as well as
the set of canonical correlations.
An additional output is the R2 found using Cohen’s set correlation (Cohen, 1982). This is a measure
of how much variance and the x and y set share.
Cohen (1982) introduced the set correlation, a multivariate generalization of the multiple correlation
to measure the overall relationship between two sets of variables. It is an application of canoncial
correlation (Hotelling, 1936) and 1−Q(1 −ρ2
i)where ρ2
iis the squared canonical correlation. Set
correlation is the amount of shared variance (R2) between two sets of variables. With the addition
of a third, covariate set, set correlation will find multivariate R2, as well as partial and semi partial
R2. (The semi and bipartial options are not yet implemented.) Details on set correlation may be
found in Cohen (1982), Cohen (1988) and Cohen, Cohen, Aiken and West (2003).
294 setCor
R2 between two sets is just
R2= 1 −|Ryx|
|Ry||Rx|= 1 −Y(1 −ρ2
i)
where R is the complete correlation matrix of the x and y variables and Rx and Ry are the two sets
involved.
Unfortunately, the R2 is sensitive to one of the canonical correlations being very high. An alterna-
tive, T2, is the proportion of additive variance and is the average of the squared canonicals. (Cohen
et al., 2003), see also Cramer and Nicewander (1979). This average, because it includes some very
small canonical correlations, will tend to be too small. Cohen et al. admonition is appropriate:
"In the final analysis, however, analysts must be guided by their substantive and methodological
conceptions of the problem at hand in their choice of a measure of association." ( p613).
Yet another measure of the association between two sets is just the simple, unweighted correlation
between the two sets. That is,
Ruw =1Rxy10
(1Ryy10).5(1Rxx10).5
where Rxy is the matrix of correlations between the two sets. This is just the simple (unweighted)
sums of the correlations in each matrix. This technique exemplifies the robust beauty of linear
models and is particularly appropriate in the case of one dimension in both x and y, and will be a
drastic underestimate in the case of items where the betas differ in sign.
When finding the unweighted correlations, as is done in alpha, items are flipped so that they all are
positively signed.
A typical use in the SAPA project is to form item composites by clustering or factoring (see
fa,ICLUST,principal), extract the clusters from these results (factor2cluster), and then form
the composite correlation matrix using cluster.cor. The variables in this reduced matrix may
then be used in multiple R procedures using set.cor.
Although the overall matrix can have missing correlations, the correlations in the subset of the
matrix used for prediction must exist.
If the number of observations is entered, then the conventional confidence intervals, statistical sig-
nificance, and shrinkage estimates are reported.
If the input is rectangular, correlations or covariances are found from the data.
The print function reports t and p values for the beta weights, the summary function just reports the
beta weights.
matReg is primarily a helper function for mediate but is a general multiple regression function
given a covariance matrix and the specified x, and y variables. Its output includes betas, se, t, p and
R2. It does not work on data matrices, nor does it take formula input.
Value
beta the beta weights for each variable in X for each variable in Y
RThe multiple R for each equation (the amount of change a unit in the predictor
set leads to in the criterion set).
R2 The multiple R2 (% variance acounted for) for each equation
setCor 295
se Standard errors of beta weights (if n.obs is specified)
tt value of beta weights (if n.obs is specified)
Probability Probability of beta = 0 (if n.obs is specified)
shrunkenR2 Estimated shrunken R2 (if n.obs is specified)
setR2 The multiple R2 of the set correlation between the x and y sets
itemresidualThe residual correlation matrix of Y with x and z removed
ruw The unit weighted multiple correlation
Ruw The unit weighted set correlation
Note
As of April 30, 2011, the order of x and y was swapped in the call to be consistent with the general y
~ x syntax of the lm and aov functions. In addition, the primary name of the function was switched
to set.cor from mat.regress to reflect the estimation of the set correlation.
The denominator degrees of freedom for the set correlation does not match that reported by Cohen
et al., 2003 in the example on page 621 but does match the formula on page 615, except for the
typo in the estimation of F (see Cohen 1982). The difference seems to be that they are adding in a
correction factor of df 2 = df2 + df1.
Author(s)
William Revelle
Maintainer: William Revelle <revelle@northwestern.edu>
References
J. Cohen (1982) Set correlation as a general multivariate data-analytic method. Multivariate Behav-
ioral Research, 17(3):301-341.
J. Cohen, P. Cohen, S.G. West, and L.S. Aiken. (2003) Applied multiple regression/correlation
analysis for the behavioral sciences. L. Erlbaum Associates, Mahwah, N.J., 3rd ed edition.
H. Hotelling. (1936) Relations between two sets of variates. Biometrika 28(3/4):321-377.
E.Cramer and W. A. Nicewander (1979) Some symmetric, invariant measures of multivariate asso-
ciation. Psychometrika, 44:43-54.
See Also
cluster.cor,factor2cluster,principal,ICLUST,link{cancor} and cca in the yacca package.
Examples
#the Kelly data from Hoteling
kelly <- structure(list(speed = c(1, 0.4248, 0.042, 0.0215, 0.0573), power = c(0.4248,
1, 0.1487, 0.2489, 0.2843), words = c(0.042, 0.1487, 1, 0.6693,
0.4662), symbols = c(0.0215, 0.2489, 0.6693, 1, 0.6915), meaningless = c(0.0573,
296 sim
0.2843, 0.4662, 0.6915, 1)), .Names = c("speed", "power", "words",
"symbols", "meaningless"), class = "data.frame", row.names = c("speed",
"power", "words", "symbols", "meaningless"))
kelly
setCor(1:2,3:5,kelly)
#Hotelling reports canonical correlations of .3073 and .0583 or squared correlations of
# 0.09443329 and 0.00339889 vs. our values of 0.0946 0.0035,
setCor(y=c(7:9),x=c(1:6),data=Thurstone,n.obs=213)
#now try partialling out some variables
set.cor(y=c(7:9),x=c(1:3),z=c(4:6),data=Thurstone) #compare with the previous
#compare complete print out with summary printing
sc <- setCor(x=c("gender","education"),y=c("SATV","SATQ"),data=sat.act) # regression from raw data
sc
summary(sc)
sim Functions to simulate psychological/psychometric data.
Description
A number of functions in the psych package will generate simulated data with particular structures.
These functions include sim for a factor simplex, and sim.simplex for a data simplex, sim.circ
for a circumplex structure, sim.congeneric for a one factor factor congeneric model, sim.dichot
to simulate dichotomous items, sim.hierarchical to create a hierarchical factor model, sim.item
a more general item simulation, sim.minor to simulate major and minor factors, sim.omega to
test various examples of omega, sim.parallel to compare the efficiency of various ways of de-
terimining the number of factors, sim.rasch to create simulated rasch data, sim.irt to create
general 1 to 4 parameter IRT data by calling sim.npl 1 to 4 parameter logistic IRT or sim.npn
1 to 4 paramater normal IRT, sim.poly to create polytomous ideas by calling sim.poly.npn 1-
4 parameter polytomous normal theory items or sim.poly.npl 1-4 parameter polytomous logis-
tic items, and sim.poly.ideal which creates data following an ideal point or unfolding model
by calling sim.poly.ideal.npn 1-4 parameter polytomous normal theory ideal point model or
sim.poly.ideal.npl 1-4 parameter polytomous logistic ideal point model.
sim.structural a general simulation of structural models, and sim.anova for ANOVA and lm
simulations, and sim.VSS. Some of these functions are separately documented and are listed here
for ease of the help function. See each function for more detailed help.
Usage
sim(fx=NULL,Phi=NULL,fy=NULL,alpha=.8,lambda = 0,n=0,mu=NULL,raw=TRUE)
sim.simplex(nvar =12, alpha=.8,lambda=0,beta=1,mu=NULL, n=0)
sim.general(nvar=9,nfact =3, g=.3,r=.3,n=0)
sim.minor(nvar=12,nfact=3,n=0,g=NULL,fbig=NULL,fsmall = c(-.2,.2),bipolar=TRUE)
sim 297
sim.omega(nvar=12,nfact=3,n=500,g=NULL,sem=FALSE,fbig=NULL,fsmall =
c(-.2,.2),bipolar=TRUE,om.fact=3,flip=TRUE,option="equal",ntrials=10)
sim.parallel(ntrials=10,nvar = c(12,24,36,48),nfact = c(1,2,3,4,6),
n = c(200,400))
sim.rasch(nvar = 5,n = 500, low=-3,high=3,d=NULL, a=1,mu=0,sd=1)
sim.irt(nvar = 5, n = 500, low=-3, high=3,a=NULL,c=0,z=1,d=NULL,mu=0,sd=1,mod="logistic")
sim.npl(nvar = 5, n = 500, low=-3,high=3,a=NULL,c=0,z=1,d=NULL,mu=0,sd=1)
sim.npn(nvar = 5, n = 500, low=-3,high=3,a=NULL,c=0,z=1,d=NULL,mu=0,sd=1)
sim.poly(nvar = 5 ,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL,
mu=0,sd=1,cat=5,mod="logistic")
sim.poly.npn(nvar = 5 ,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL, mu=0,sd=1,cat=5)
sim.poly.npl(nvar = 5 ,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL, mu=0,sd=1,cat=5)
sim.poly.ideal(nvar = 5 ,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL,
mu=0,sd=1,cat=5,mod="logistic")
sim.poly.ideal.npn(nvar = 5,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL, mu=0,sd=1,cat=5)
sim.poly.ideal.npl(nvar = 5,n = 500,low=-2,high=2,a=NULL,c=0,z=1,d=NULL,
mu=0,sd=1,cat=5,theta=NULL)
sim.poly.mat(R,m,n)
Arguments
fx The measurement model for x. If NULL, a 4 factor model is generated
Phi The structure matrix of the latent variables
fy The measurement model for y
mu The means structure for the fx factors
nNumber of cases to simulate. If n=0 or NULL, the population matrix is returned.
raw if raw=TRUE, raw data are returned as well.
nvar Number of variables for a simplex structure
nfact Number of large factors to simulate in sim.minor,number of group factors in
sim.general,sim.omega
gGeneral factor correlations in sim.general and general factor loadings in sim.omega
and sim.minor
sem Should the sim.omega function do both an EFA omega as well as a CFA omega
using the sem package?
rgroup factor correlations in sim.general
alpha the base correlation for an autoregressive simplex
lambda the trait component of a State Trait Autoregressive Simplex
beta Test reliability of a STARS simplex
fbig Factor loadings for the main factors. Default is a simple structure with loadings
sampled from (.8,.6) for nvar/nfact variables and 0 for the remaining. If fbig is
specified, then each factor has loadings sampled from it.
bipolar if TRUE, then positive and negative loadings are generated from fbig
om.fact Number of factors to extract in omega
298 sim
flip In omega, should item signs be flipped if negative
option In omega, for the case of two factors, how to weight them?
fsmall nvar/2 small factors are generated with loadings sampled from (-.2,0,.2)
ntrials Number of replications per level
low lower difficulty for sim.rasch or sim.irt
high higher difficulty for sim.rasch or sim.irt
aif not specified as a vector, the descrimination parameter a = αwill be set to 1.0
for all items
dif not specified as a vector, item difficulties (d = δ) will range from low to high
cthe gamma parameter: if not specified as a vector, the guessing asymptote is set
to 0
zthe zeta parameter: if not specified as a vector, set to 1
sd the standard deviation for the underlying latent variable in the irt simulations
mod which IRT model to use, mod="logistic" simulates a logistic function, otherwise,
a normal function
cat Number of categories to simulate in sim.poly. If cat=2, then this is the same as
simulating t/f items and sim.poly is functionally equivalent to sim.irt
theta The underlying latent trait value for each simulated subject
RA correlation matrix to be simulated using the sim.poly.mat function
mThe matrix of marginals for all the items
Details
Simulation of data structures is a very useful tool in psychometric research and teaching. By know-
ing “truth" it is possible to see how well various algorithms can capture it. For a much longer
discussion of the use of simulation in psychometrics, see the accompany vignettes.
The simulations documented here are a miscellaneous set of functions that will be documented in
other help files eventually.
The default values for sim.structure is to generate a 4 factor, 12 variable data set with a simplex
structure between the factors. This, and the simplex of items (sim.simplex) can also be converted
in a STARS model with an autoregressive component (alpha) and a stable trait component (lambda).
Two data structures that are particular challenges to exploratory factor analysis are the simplex
structure and the presence of minor factors. Simplex structures sim.simplex will typically occur in
developmental or learning contexts and have a correlation structure of r between adjacent variables
and r^n for variables n apart. Although just one latent variable (r) needs to be estimated, the structure
will have nvar-1 factors.
An alternative version of the simplex is the State-Trait-Auto Regressive Structure (STARS) which
has both a simplex state structure, with autoregressive path alpha and a trait structure with path
lambda. This simulated in sim.simplex by specifying a non-zero lambda value.
Many simulations of factor structures assume that except for the major factors, all residuals are
normally distributed around 0. An alternative, and perhaps more realistic situation, is that the there
are a few major (big) factors and many minor (small) factors. The challenge is thus to identify the
sim 299
major factors. sim.minor generates such structures. The structures generated can be thought of as
havinga a major factor structure with some small correlated residuals. To make these simulations
complete, the possibility of a general factor is considered. For simplicity, sim.minor allows one to
specify a set of loadings to be sampled from for g, fmajor and fminor. Alternatively, it is possible
to specify the complete factor matrix.
Another structure worth considering is direct modeling of a general factor with several group fac-
tors. This is done using sim.general.
Although coefficient ωis a very useful indicator of the general factor saturation of a unifactorial
test (one with perhaps several sub factors), it has problems with the case of multiple, independent
factors. In this situation, one of the factors is labelled as “general” and the omega estimate is too
large. This situation may be explored using the sim.omega function with general left as NULL. If
there is a general factor, then results from sim.omega suggests that omega estimated either from
EFA or from SEM does a pretty good job of identifying it but that the EFA approach using Schmid-
Leiman transformation is somewhat more robust than the SEM approach.
The four irt simulations, sim.rasch, sim.irt, sim.npl and sim.npn, simulate dichotomous items fol-
lowing the Item Response model. sim.irt just calls either sim.npl (for logistic models) or sim.npn
(for normal models) depending upon the specification of the model.
The logistic model is
P(i, j) = γ+ζ−γ
1 + eα(δ−θ)
where γis the lower asymptote or guesssing parameter, ζis the upper asymptote (normally 1), αis
item discrimination and δis item difficulty. For the 1 Paramater Logistic (Rasch) model, gamma=0,
zeta=1, alpha=1 and item difficulty is the only free parameter to specify.
For the 2PL and 2PN models, a = αand d = δare specified.
For the 3PL or 3PN models, items also differ in their guessing parameter c =γ.
For the 4PL and 4PN models, the upper asymptote, z= ζis also specified.
(Graphics of these may be seen in the demonstrations for the logistic function.)
The normal model (irt.npn calculates the probability using pnorm instead of the logistic function
used in irt.npl, but the meaning of the parameters are otherwise the same. With the a = αparameter
= 1.702 in the logistic model the two models are practically identical.
In parallel to the dichotomous IRT simulations are the poly versions which simulate polytomous
item models. They have the additional parameter of how many categories to simulate. In addi-
tion, the sim.poly.ideal functions will simulate an ideal point or unfolding model in which the
response probability varies by the distance from each subject’s ideal point. Some have claimed
that this is a more appropriate model of the responses to personality questionnaires. It will lead to
simplex like structures which may be fit by a two factor model. The middle items form one factor,
the extreme a bipolar factor.
The previous functions all assume one latent trait. Alternatively, we can simulate dichotomous or
polytomous items with a particular structure using the sim.poly.mat function. This takes as input the
population correlation matrix, the population marginals, and the sample size. It returns categorical
items with the specified structure.
Other simulation functions in psych are:
sim.structure A function to combine a measurement and structural model into one data matrix.
Useful for understanding structural equation models. Combined with structure.diagram to see
the proposed structure.
300 sim
sim.congeneric A function to create congeneric items/tests for demonstrating classical test theory.
This is just a special case of sim.structure.
sim.hierarchical A function to create data with a hierarchical (bifactor) structure.
sim.item A function to create items that either have a simple structure or a circumplex structure.
sim.circ Create data with a circumplex structure.
sim.dichot Create dichotomous item data with a simple or circumplex structure.
sim.minor Create a factor structure for nvar variables defined by nfact major factors and nvar/2
“minor" factors for n observations.
Although the standard factor model assumes that K major factors (K « nvar) will account for the
correlations among the variables
R=F F 0+U2
where R is of rank P and F is a P x K matrix of factor coefficients and U is a diagonal matrix of
uniquenesses. However, in many cases, particularly when working with items, there are many small
factors (sometimes referred to as correlated residuals) that need to be considered as well. This leads
to a data structure such that
R=F F 0+MM 0+U2
where R is a P x P matrix of correlations, F is a P x K factor loading matrix, M is a P x P/2 matrix
of minor factor loadings, and U is a diagonal matrix (P x P) of uniquenesses.
Such a correlation matrix will have a poor χ2value in terms of goodness of fit if just the K factors
are extracted, even though for all intents and purposes, it is well fit.
sim.minor will generate such data sets with big factors with loadings of .6 to .8 and small factors
with loadings of -.2 to .2. These may both be adjusted.
sim.parallel Create a number of simulated data sets using sim.minor to show how parallel anal-
ysis works. The general observation is that with the presence of minor factors, parallel analysis is
probably best done with component eigen values rather than factor eigen values, even when using
the factor model.
sim.anova Simulate a 3 way balanced ANOVA or linear model, with or without repeated measures.
Useful for teaching research methods and generating teaching examples.
sim.multilevel To understand some of the basic concepts of multilevel modeling, it is useful to
create multilevel structures. The correlations of aggregated data is sometimes called an ’ecological
correlation’. That group level and individual level correlations are independent makes such infer-
ences problematic. This simulation allows for demonstrations that correlations within groups do
not imply, nor are implied by, correlations between group means.
Author(s)
William Revelle
References
Revelle, W. (in preparation) An Introduction to Psychometric Theory with applications in R. Springer.
at http://personality-project.org/r/book/
sim.anova 301
See Also
See above
Examples
simplex <- sim.simplex() #create the default simplex structure
lowerMat(simplex) #the correlation matrix
#create a congeneric matrix
congeneric <- sim.congeneric()
lowerMat(congeneric)
R <- sim.hierarchical()
lowerMat(R)
#now simulate categorical items with the hierarchical factor structure.
#Let the items be dichotomous with varying item difficulties.
marginals = matrix(c(seq(.1,.9,.1),seq(.9,.1,-.1)),byrow=TRUE,nrow=2)
X <- sim.poly.mat(R=R,m=marginals,n=1000)
lowerCor(X) #show the raw correlations
#lowerMat(tetrachoric(X)$rho) # show the tetrachoric correlations (not run)
#generate a structure
fx <- matrix(c(.9,.8,.7,rep(0,6),c(.8,.7,.6)),ncol=2)
fy <- c(.6,.5,.4)
Phi <- matrix(c(1,0,.5,0,1,.4,0,0,0),ncol=3)
R <- sim.structure(fx,Phi,fy)
cor.plot(R$model) #show it graphically
simp <- sim.simplex()
#show the simplex structure using cor.plot
cor.plot(simp,colors=TRUE,main="A simplex structure")
#Show a STARS model
simp <- sim.simplex(alpha=.8,lambda=.4)
#show the simplex structure using cor.plot
cor.plot(simp,colors=TRUE,main="State Trait Auto Regressive Simplex" )
sim.anova Simulate a 3 way balanced ANOVA or linear model, with or without
repeated measures.
Description
For teaching basic statistics, it is useful to be able to generate examples suitable for analysis of
variance or simple linear models. sim.anova will generate the design matrix of three independent
variables (IV1, IV2, IV3) with an arbitrary number of levels and effect sizes for each main effect
and interaction. IVs can be either continuous or categorical and can have linear or quadratic effects.
Either a single dependent variable or multiple (within subject) dependent variables are generated
according to the specified model. The repeated measures are assumed to be tau equivalent with a
specified reliability.
302 sim.anova
Usage
sim.anova(es1 = 0, es2 = 0, es3 = 0, es12 = 0, es13 = 0,
es23 = 0, es123 = 0, es11=0,es22=0, es33=0,n = 2,n1 = 2, n2 = 2, n3 = 2,
within=NULL,r=.8,factors=TRUE,center = TRUE,std=TRUE)
Arguments
es1 Effect size of IV1
es2 Effect size of IV2
es3 Effect size of IV3
es12 Effect size of the IV1 x IV2 interaction
es13 Effect size of the IV1 x IV3 interaction
es23 Effect size of the IV2 x IV3 interaction
es123 Effect size of the IV1 x IV2 * IV3 interaction
es11 Effect size of the quadratric term of IV1
es22 Effect size of the quadratric term of IV2
es33 Effect size of the quadratric term of IV3
nSample size per cell (if all variables are categorical) or (if at least one variable
is continuous), the total sample size
n1 Number of levels of IV1 (0) if continuous
n2 Number of levels of IV2
n3 Number of levels of IV3
within if not NULL, then within should be a vector of the means of any repeated mea-
sures.
rthe correlation between the repeated measures (if they exist). This can be thought
of as the reliablility of the measures.
factors report the IVs as factors rather than numeric
center center=TRUE provides orthogonal contrasts, center=FALSE adds the minimum
value + 1 to all contrasts
std Standardize the effect sizes by standardizing the IVs
Details
A simple simulation for teaching about ANOVA, regression and reliability. A variety of demonstra-
tions of the relation between anova and lm can be shown.
The default is to produce categorical IVs (factors). For more than two levels of an IV, this will show
the difference between the linear model and anova in terms of the comparisons made.
The within vector can be used to add congenerically equivalent dependent variables. These will
have intercorrelations (reliabilities) of r and means as specified as values of within.
To demonstrate the effect of centered versus non-centering, make factors = center=FALSE. The
default is to center the IVs. By not centering them, the lower order effects will be incorrect given
the higher order interaction terms.
sim.anova 303
Value
y.df is a data.frame of the 3 IV values as well as the DV values.
IV1 ... IV3 Independent variables 1 ... 3
DV If there is a single dependent variable
DV.1 ... DV.n If within is specified, then the n within subject dependent variables
Author(s)
William Revelle
See Also
The general set of simulation functions in the psych package sim
Examples
set.seed(42)
data.df <- sim.anova(es1=1,es2=.5,es13=1) # one main effect and one interaction
describe(data.df)
pairs.panels(data.df) #show how the design variables are orthogonal
#
summary(lm(DV~IV1*IV2*IV3,data=data.df))
summary(aov(DV~IV1*IV2*IV3,data=data.df))
set.seed(42)
#demonstrate the effect of not centering the data on the regression
data.df <- sim.anova(es1=1,es2=.5,es13=1,center=FALSE) #
describe(data.df)
#
#this one is incorrect, because the IVs are not centered
summary(lm(DV~IV1*IV2*IV3,data=data.df))
summary(aov(DV~IV1*IV2*IV3,data=data.df)) #compare with the lm model
#now examine multiple levels and quadratic terms
set.seed(42)
data.df <- sim.anova(es1=1,es13=1,n2=3,n3=4,es22=1)
summary(lm(DV~IV1*IV2*IV3,data=data.df))
summary(aov(DV~IV1*IV2*IV3,data=data.df))
pairs.panels(data.df)
#
data.df <- sim.anova(es1=1,es2=-.5,within=c(-1,0,1),n=10)
pairs.panels(data.df)
304 sim.congeneric
sim.congeneric Simulate a congeneric data set
Description
Classical Test Theory (CTT) considers four or more tests to be congenerically equivalent if all tests
may be expressed in terms of one factor and a residual error. Parallel tests are the special case where
(usually two) tests have equal factor loadings. Tau equivalent tests have equal factor loadings but
may have unequal errors. Congeneric tests may differ in both factor loading and error variances.
Usage
sim.congeneric(loads = c(0.8, 0.7, 0.6, 0.5),N = NULL, err=NULL, short = TRUE,
categorical=FALSE, low=-3,high=3,cuts=NULL)
Arguments
NHow many subjects to simulate. If NULL, return the population model
loads A vector of factor loadings for the tests
err A vector of error variances – if NULL then error = 1 - loading 2
short short=TRUE: Just give the test correlations, short=FALSE, report observed test
scores as well as the implied pattern matrix
categorical continuous or categorical (discrete) variables.
low values less than low are forced to low
high values greater than high are forced to high
cuts If specified, and categorical = TRUE, will cut the resulting continuous output at
the value of cuts
Details
When constructing examples for reliability analysis, it is convenient to simulate congeneric data
structures. These are the most simple of item structures, having just one factor. Mainly used for a
discussion of reliability theory as well as factor score estimates.
The implied covariance matrix is just pattern %*% t(pattern).
Value
model The implied population correlation matrix if N=NULL or short=FALSE, other-
wise the sample correlation matrix
pattern The pattern matrix implied by the loadings and error variances
rThe sample correlation matrix for long output
observed a matrix of test scores for n tests
latent The latent trait and error scores
sim.hierarchical 305
Author(s)
William Revelle
References
Revelle, W. (in prep) An introduction to psychometric theory with applications in R. To be published
by Springer. (working draft available at http://personality-project.org/r/book/
See Also
item.sim for other simulations, fa for an example of factor scores, irt.fa and polychoric for
the treatment of item data with discrete values.
Examples
test <- sim.congeneric(c(.9,.8,.7,.6)) #just the population matrix
test <- sim.congeneric(c(.9,.8,.7,.6),N=100) # a sample correlation matrix
test <- sim.congeneric(short=FALSE, N=100)
round(cor(test$observed),2) # show a congeneric correlation matrix
f1=fa(test$observed,scores=TRUE)
round(cor(f1$scores,test$latent),2)
#factor score estimates are correlated with but not equal to the factor scores
set.seed(42)
#500 responses to 4 discrete items
items <- sim.congeneric(N=500,short=FALSE,low=-2,high=2,categorical=TRUE)
d4 <- irt.fa(items$observed) #item response analysis of congeneric measures
sim.hierarchical Create a population or sample correlation matrix, perhaps with hier-
archical structure.
Description
Create a population orthogonal or hierarchical correlation matrix from a set of factor loadings and
factor intercorrelations. Samples of size n may be then be drawn from this population. Return either
the sample data, sample correlations, or population correlations. This is used to create sample data
sets for instruction and demonstration.
Usage
sim.hierarchical(gload=NULL, fload=NULL, n = 0, raw = FALSE,mu = NULL)
make.hierarchical(gload=NULL, fload=NULL, n = 0, raw = FALSE) #deprecated
306 sim.hierarchical
Arguments
gload Loadings of group factors on a general factor
fload Loadings of items on the group factors
nNumber of subjects to generate: N=0 => population values
raw raw=TRUE, report the raw data, raw=FALSE, report the sample correlation ma-
trix.
mu means for the individual variables
Details
Many personality and cognitive tests have a hierarchical factor structure. For demonstration pur-
poses, it is useful to be able to create such matrices, either with population values, or sample values.
Given a matrix of item factor loadings (fload) and of loadings of these factors on a general factor
(gload), we create a population correlation matrix by using the general factor law (R = F’ theta F
where theta = g’g).
To create sample values, we use the mvrnorm function from MASS.
The default is to return population correlation matrices. Sample correlation matrices are generated
if n >0. Raw data are returned if raw = TRUE.
The default values for gload and fload create a data matrix discussed by Jensen and Weng, 1994.
Although written to create hierarchical structures, if the gload matrix is all 0, then a non-hierarchical
structure will be generated.
Value
a matrix of correlations or a data matrix
Author(s)
William Revelle
References
http://personality-project.org/r/r.omega.html
Jensen, A.R., Weng, L.J. (1994) What is a Good g? Intelligence, 18, 231-258.
See Also
omega,schmid,ICLUST,VSS for ways of analyzing these data. Also see sim.structure to simulate
a variety of structural models (e.g., multiple correlated factor models). The simulation uses the
mvrnorm function from the MASS package.
sim.item 307
Examples
gload <- gload<-matrix(c(.9,.8,.7),nrow=3) # a higher order factor matrix
fload <-matrix(c( #a lower order (oblique) factor matrix
.8,0,0,
.7,0,.0,
.6,0,.0,
0,.7,.0,
0,.6,.0,
0,.5,0,
0,0,.6,
0,0,.5,
0,0,.4), ncol=3,byrow=TRUE)
jensen <- sim.hierarchical(gload,fload) #the test set used by omega
round(jensen,2)
#simulate a non-hierarchical structure
fload <- matrix(c(c(c(.9,.8,.7,.6),rep(0,20)),c(c(.9,.8,.7,.6),rep(0,20)),
c(c(.9,.8,.7,.6),rep(0,20)),c(c(c(.9,.8,.7,.6),rep(0,20)),c(.9,.8,.7,.6))),ncol=5)
gload <- matrix(rep(0,5))
five.factor <- sim.hierarchical(gload,fload,500,TRUE) #create sample data set
#do it again with a hierachical structure
gload <- matrix(rep(.7,5) )
five.factor.g <- sim.hierarchical(gload,fload,500,TRUE) #create sample data set
#compare these two with omega
#not run
#om.5 <- omega(five.factor$observed,5)
#om.5g <- omega(five.factor.g$observed,5)
sim.item Generate simulated data structures for circumplex, spherical, or sim-
ple structure
Description
Rotations of factor analysis and principal components analysis solutions typically try to represent
correlation matrices as simple structured. An alternative structure, appealing to some, is a circum-
plex structure where the variables are uniformly spaced on the perimeter of a circle in a two dimen-
sional space. Generating simple structure and circumplex data is straightforward, and is useful for
exploring alternative solutions to affect and personality structure. A generalization to 3 dimensional
(spherical) data is straightforward.
Usage
sim.item(nvar = 72, nsub = 500, circum = FALSE, xloading = 0.6, yloading = 0.6,
gloading = 0, xbias = 0, ybias = 0, categorical = FALSE, low = -3, high = 3,
truncate = FALSE, cutpoint = 0)
sim.circ(nvar = 72, nsub = 500, circum = TRUE, xloading = 0.6, yloading = 0.6,
308 sim.item
gloading = 0, xbias = 0, ybias = 0, categorical = FALSE, low = -3, high = 3,
truncate = FALSE, cutpoint = 0)
sim.dichot(nvar = 72, nsub = 500, circum = FALSE, xloading = 0.6, yloading = 0.6,
gloading = 0, xbias = 0, ybias = 0, low = 0, high = 0)
item.dichot(nvar = 72, nsub = 500, circum = FALSE, xloading = 0.6, yloading = 0.6,
gloading = 0, xbias = 0, ybias = 0, low = 0, high = 0)
sim.spherical(simple=FALSE, nx=7,ny=12 ,nsub = 500, xloading =.55, yloading = .55,
zloading=.55, gloading=0, xbias=0, ybias = 0, zbias=0,categorical=FALSE,
low=-3,high=3,truncate=FALSE,cutpoint=0)
con2cat(old,cuts=c(0,1,2,3),where)
Arguments
nvar Number of variables to simulate
nsub Number of subjects to simulate
circum circum=TRUE is circumplex structure, FALSE is simple structure
simple simple structure or spherical structure in sim.spherical
xloading the average loading on the first dimension
yloading Average loading on the second dimension
zloading the average loading on the third dimension in sim.spherical
gloading Average loading on a general factor (default=0)
xbias To introduce skew, how far off center is the first dimension
ybias To introduce skew on the second dimension
zbias To introduce skew on the third dimension – if using sim.spherical
categorical continuous or categorical variables.
low values less than low are forced to low (or 0 in item.dichot)
high values greater than high are forced to high (or 1 in item.dichot)
truncate Change all values less than cutpoint to cutpoint.
cutpoint What is the cutpoint
nx number of variables for the first factor in sim.spherical
ny number of variables for the second and third factors in sim.spherical
old a matrix or data frame
cuts Values of old to be used as cut points when converting continuous values to
categorical values
where Which columns of old should be converted to categorical variables. If missing,
then all columns are converted.
Details
This simulation was originally developed to compare the effect of skew on the measurement of
affect (see Rafaeli and Revelle, 2005). It has been extended to allow for a general simulation of
affect or personality items with either a simple structure or a circumplex structure. Items can be
sim.item 309
continuous normally distributed, or broken down into n categories (e.g, -2, -1, 0, 1, 2). Items can be
distorted by limiting them to these ranges, even though the items have a mean of (e.g., 1).
The addition of item.dichot allows for testing structures with dichotomous items of different dif-
ficulty (endorsement) levels. Two factor data with either simple structure or circumplex structure
are generated for two sets of items, one giving a score of 1 for all items greater than the low (easy)
value, one giving a 1 for all items greater than the high (hard) value. The default values for low and
high are 0. That is, all items are assumed to have a 50 percent endorsement rate. To examine the
effect of item difficulty, low could be -1, high 1. This will lead to item endorsements of .84 for the
easy and .16 for the hard. Within each set of difficulties, the first 1/4 are assigned to the first factor
factor, the second to the second factor, the third to the first factor (but with negative loadings) and
the fourth to the second factor (but with negative loadings).
It is useful to compare the results of sim.item with sim.hierarchical. sim.item will produce a general
factor that runs through all the items as well as two orthogonal factors. This produces a data set
that is hard to represent with standard rotation techniques. Extracting 3 factors without rotation and
then rotating the 2nd and 3rd factors reproduces the correct solution. But simple oblique rotation of
3 factors, or an omega analysis do not capture the underlying structure. See the last example.
Yet another structure that might be appealing is fully complex data in three dimensions. That
is, rather than having items representing the circumference of a circle, items can be structured to
represent equally spaced three dimensional points on a sphere. sim.spherical produces such data.
Value
A data matrix of (nsub) subjects by (nvar) variables.
Author(s)
William Revelle
References
Variations of a routine used in Rafaeli and Revelle, 2006; Rafaeli, E. & Revelle, W. (2006). A
premature consensus: Are happiness and sadness truly opposite affects? Motivation and Emotion.
http://personality-project.org/revelle/publications/rafaeli.revelle.06.pdf
Acton, G. S. and Revelle, W. (2004) Evaluation of Ten Psychometric Criteria for Circumplex Struc-
ture. Methods of Psychological Research Online, Vol. 9, No. 1 (formerly (http://www.dgps.de/fachgruppen/methoden/mpr-
online/issue22/mpr110_10.pdf) also at http://personality-project.org/revelle/publications/
acton.revelle.mpr110_10.pdf
See Also
See Also the implementation in this to generate numerous simulations. simulation.circ,circ.tests
as well as other simulations ( sim.structural sim.hierarchical)
Examples
round(cor(circ.sim(nvar=8,nsub=200)),2)
plot(fa(circ.sim(16,500),2)$loadings,main="Circumplex Structure") #circumplex structure
#
310 sim.multilevel
#
plot(fa(item.sim(16,500),2)$loadings,main="Simple Structure") #simple structure
#
cluster.plot(fa(item.dichot(16,low=0,high=1),2))
set.seed(42)
data <- mnormt::rmnorm(1000, c(0, 0), matrix(c(1, .5, .5, 1), 2, 2)) #continuous data
new <- con2cat(data,c(-1.5,-.5,.5,1.5)) #discreet data
polychoric(new)
#not run
#x12 <- sim.item(12,gloading=.6)
#f3 <- fa(x12,3,rotate="none")
#f3 #observe the general factor
#oblimin(f3$loadings[,2:3]) #show the 2nd and 3 factors.
#f3 <- fa(x12,3) #now do it with oblimin rotation
#f3 # not what one naively expect.
sim.multilevel Simulate multilevel data with specified within group and between
group correlations
Description
Multilevel data occur when observations are nested within groups. This can produce correlational
structures that are sometimes difficult to understand. This simulation allows for demonstrations that
correlations within groups do not imply, nor are implied by, correlations between group means. The
correlations of aggregated data is sometimes called an ’ecological correlation’. That group level
and individual level correlations are independent makes such inferences problematic.
Usage
sim.multilevel(nvar = 9, ngroups = 4, ncases = 16, rwg, rbg, eta)
Arguments
nvar Number of variables to simulate
ngroups The number of groups to simulate
ncases The number of simulated cases
rwg The within group correlational structure
rbg The between group correlational structure
eta The correlation of the data with the within data
sim.multilevel 311
Details
The basic concepts of the independence of within group and between group correlations is dis-
cussed very clearly by Pedhazur (1997) as well as by Bliese (2009). This function merely simulates
pooled correlations (mixtures of between group and within group correlations) to allow for a better
understanding of the problems inherent in multi-level modeling.
Data (wg) are created with a particular within group structure (rwg). Independent data (bg) are
also created with a between group structure (rbg). Note that although there are ncases rows to this
data matrix, there are only ngroups independent cases. That is, every ngroups case is a repeat. The
resulting data frame (xy) is a weighted sum of the wg and bg. This is the inverse procedure for
estimating estimating rwg and rbg from an observed rxy which is done by the statsBy function.
Value
wg A matrix (ncases * nvar) of simulated within group scores
bg A matrix (ncases * nvar) of simulated between group scores
xy A matrix ncases * (nvar +1) of pooled data
Author(s)
William Revelle
References
P. D. Bliese. Multilevel modeling in R (2.3) a brief introduction to R, the multilevel package and
the nlme package, 2009.
Pedhazur, E.J. (1997) Multiple regression in behavioral research: explanation and prediction. Har-
court Brace.
Revelle, W. An introduction to psychometric theory with applications in R (in prep) Springer. Draft
chapters available at http://personality-project.org/r/book/
See Also
statsBy for the decomposition of multi level data and withinBetween for an example data set.
Examples
#get some parameters to simulate
data(withinBetween)
wb.stats <- statsBy(withinBetween,"Group")
rwg <- wb.stats$rwg
rbg <- wb.stats$rbg
eta <- rep(.5,9)
#simulate them. Try this again to see how it changes
XY <- sim.multilevel(ncases=100,ngroups=10,rwg=rwg,rbg=rbg,eta=eta)
lowerCor(XY$wg) #based upon 89 df
lowerCor(XY$bg) #based upon 9 df --
312 sim.structure
sim.structure Create correlation matrices or data matrices with a particular mea-
surement and structural model
Description
Structural Equation Models decompose correlation or correlation matrices into a measurement (fac-
tor) model and a structural (regression) model. sim.structural creates data sets with known measure-
ment and structural properties. Population or sample correlation matrices with known properties are
generated. Optionally raw data are produced.
It is also possible to specify a measurement model for a set of x variables separately from a set of
y variables. They are then combined into one model with the correlation structure between the two
sets.
Finally, the general case is given a population correlation matrix, generate data that will reproduce
(with sampling variability) that correlation matrix. sim.correlation.
Usage
sim.structure(fx=NULL,Phi=NULL, fy=NULL, f=NULL, n=0, uniq=NULL, raw=TRUE,
items = FALSE, low=-2,high=2,d=NULL,cat=5, mu=0)
sim.structural(fx=NULL, Phi=NULL, fy=NULL, f=NULL, n=0, uniq=NULL, raw=TRUE,
items = FALSE, low=-2,high=2,d=NULL,cat=5, mu=0) #deprecated
sim.correlation(R,n=1000,data=FALSE)
Arguments
fx The measurement model for x
Phi The structure matrix of the latent variables
fy The measurement model for y
fThe measurement model
nNumber of cases to simulate. If n=0, the population matrix is returned.
uniq The uniquenesses if creating a covariance matrix
raw if raw=TRUE, raw data are returned as well for n > 0.
items TRUE if simulating items, FALSE if simulating scales
low Restrict the item difficulties to range from low to high
high Restrict the item difficulties to range from low to high
dA vector of item difficulties, if NULL will range uniformly from low to high
cat Number of categories when creating binary (2) or polytomous items
mu A vector of means, defaults to 0
RThe correlation matrix to reproduce
data if TRUE, return the raw data, otherwise return the sample correlation matrix.
sim.structure 313
Details
Given the measurement model, fx and the structure model Phi, the model is f %*% Phi %*% t(f).
Reliability is f %*% t(f). fφf0and the reliability for each test is the items communality or just the
diag of the model.
If creating a correlation matrix, (uniq=NULL) then the diagonal is set to 1, otherwise the diagonal
is diag(model) + uniq and the resulting structure is a covariance matrix.
Given the model, raw data are generated using the mvnorm function.
A special case of a structural model are one factor models such as parallel tests, tau equivalent tests,
and congeneric tests. These may be created by letting the structure matrix = 1 and then defining a
vector of factor loadings. Alternatively, make.congeneric will do the same.
sim.correlation will create data sampled from a specified correlation matrix for a particular
sample size. If desired, it will just return the sample correlation matrix. With data=TRUE, it will
return the sample data as well.
Value
model The implied population correlation or covariance matrix
reliability The population reliability values
rThe sample correlation or covariance matrix
observed If raw=TRUE, a sample data matrix
Author(s)
William Revelle
References
Revelle, W. (in preparation) An Introduction to Psychometric Theory with applications in R. Springer.
at http://personality-project.org/r/book/
See Also
make.hierarchical for another structural model and make.congeneric for the one factor case.
structure.list and structure.list for making symbolic structures.
Examples
fx <-matrix(c( .9,.8,.6,rep(0,4),.6,.8,-.7),ncol=2)
fy <- matrix(c(.6,.5,.4),ncol=1)
rownames(fx) <- c("V","Q","A","nach","Anx")
rownames(fy)<- c("gpa","Pre","MA")
Phi <-matrix( c(1,0,.7,.0,1,.7,.7,.7,1),ncol=3)
gre.gpa <- sim.structural(fx,Phi,fy)
print(gre.gpa,2)
#correct for attenuation to see structure
round(correct.cor(gre.gpa$model,gre.gpa$reliability),2)
congeneric <- sim.structure(f=c(.9,.8,.7,.6)) # a congeneric model
314 sim.VSS
congeneric
sim.VSS create VSS like data
Description
Simulation is one of most useful techniques in statistics and psychometrics. Here we simulate a
correlation matrix with a simple structure composed of a specified number of factors. Each item is
assumed to have complexity one. See circ.sim and item.sim for alternative simulations.
Usage
sim.VSS(ncases=1000, nvariables=16, nfactors=4, meanloading=.5,dichot=FALSE,cut=0)
Arguments
ncases number of simulated subjects
nvariables Number of variables
nfactors Number of factors to generate
meanloading with a mean loading
dichot dichot=FALSE give continuous variables, dichot=TRUE gives dichotomous vari-
ables
cut if dichotomous = TRUE, then items with values > cut are assigned 1, otherwise
0.
Value
a ncases x nvariables matrix
Author(s)
William Revelle
See Also
VSS,ICLUST
Examples
## Not run:
simulated <- sim.VSS(1000,20,4,.6)
vss <- VSS(simulated,rotate="varimax")
VSS.plot(vss)
## End(Not run)
simulation.circ 315
simulation.circ Simulations of circumplex and simple structure
Description
Rotations of factor analysis and principal components analysis solutions typically try to represent
correlation matrices as simple structured. An alternative structure, appealing to some, is a cir-
cumplex structure where the variables are uniformly spaced on the perimeter of a circle in a two
dimensional space. Generating these data is straightforward, and is useful for exploring alternative
solutions to affect and personality structure.
Usage
simulation.circ(samplesize=c(100,200,400,800), numberofvariables=c(16,32,48,72))
circ.sim.plot(x.df)
Arguments
samplesize a vector of sample sizes to simulate
numberofvariables
vector of the number of variables to simulate
x.df A data frame resulting from simulation.circ
Details
“A common model for representing psychological data is simple structure (Thurstone, 1947). Ac-
cording to one common interpretation, data are simple structured when items or scales have non-
zero factor loadings on one and only one factor (Revelle & Rocklin, 1979). Despite the common-
place application of simple structure, some psychological models are defined by a lack of simple
structure. Circumplexes (Guttman, 1954) are one kind of model in which simple structure is lack-
ing.
“A number of elementary requirements can be teased out of the idea of circumplex structure. First,
circumplex structure implies minimally that variables are interrelated; random noise does not a
circumplex make. Second, circumplex structure implies that the domain in question is optimally
represented by two and only two dimensions. Third, circumplex structure implies that variables do
not group or clump along the two axes, as in simple structure, but rather that there are always inter-
stitial variables between any orthogonal pair of axes (Saucier, 1992). In the ideal case, this quality
will be reflected in equal spacing of variables along the circumference of the circle (Gurtman, 1994;
Wiggins, Steiger, & Gaelick, 1981). Fourth, circumplex structure implies that variables have a con-
stant radius from the center of the circle, which implies that all variables have equal communality on
the two circumplex dimensions (Fisher, 1997; Gurtman, 1994). Fifth, circumplex structure implies
that all rotations are equally good representations of the domain (Conte & Plutchik, 1981; Larsen
& Diener, 1992)." (Acton and Revelle, 2004)
Acton and Revelle reviewed the effectiveness of 10 tests of circumplex structure and found that
four did a particularly good job of discriminating circumplex structure from simple structure, or
316 simulation.circ
circumplexes from ellipsoidal structures. Unfortunately, their work was done in Pascal and is not
easily available. Here we release R code to do the four most useful tests:
The Gap test of equal spacing
Fisher’s test of equality of axes
A test of indifference to Rotation
A test of equal Variance of squared factor loadings across arbitrary rotations.
Included in this set of functions are simple procedure to generate circumplex structured or simple
structured data, the four test statistics, and a simple simulation showing the effectiveness of the four
procedures.
circ.sim.plot compares the four tests for circumplex, ellipsoid and simple structure data as func-
tion of the number of variables and the sample size. What one can see from this plot is that although
no one test is sufficient to discriminate these alternative structures, the set of four tests does a very
good job of doing so. When testing a particular data set for structure, comparing the results of all
four tests to the simulated data will give a good indication of the structural properties of the data.
Value
A data.frame with simulation results for circumplex, ellipsoid, and simple structure data sets for
each of the four tests.
Note
The simulations default values are for sample sizes of 100, 200, 400, and 800 cases, with 16, 32, 48
and 72 items.
Author(s)
William Revelle
References
Acton, G. S. and Revelle, W. (2004) Evaluation of Ten Psychometric Criteria for Circumplex Struc-
ture. Methods of Psychological Research Online, Vol. 9, No. 1 (formerly at http://www.dgps.de/fachgruppen/methoden/mpr-
online/issue22/mpr110_10.pdf and now at http://personality-project.org/revelle/publications/
acton.revelle.mpr110_10.pdf.
See Also
See also circ.tests,sim.circ,sim.structural,sim.hierarchical
Examples
#not run
demo <- simulation.circ()
boxplot(demo[3:14])
title("4 tests of Circumplex Structure",sub="Circumplex, Ellipsoid, Simple Structure")
circ.sim.plot(demo[3:14]) #compare these results to real data
smc 317
smc Find the Squared Multiple Correlation (SMC) of each variable with
the remaining variables in a matrix
Description
The squared multiple correlation of a variable with the remaining variables in a matrix is sometimes
used as initial estimates of the communality of a variable.
SMCs are also used when estimating reliability using Guttman’s lambda 6 guttman coefficient.
The SMC is just 1 - 1/diag(R.inv) where R.inv is the inverse of R.
Usage
smc(R,covar=FALSE)
Arguments
RA correlation matrix or a dataframe. In the latter case, correlations are found.
covar if covar = TRUE and R is either a covariance matrix or data frame, then return
the smc * variance for each item
Value
a vector of squared multiple correlations. Or, if covar=TRUE, a vector of squared multiple correla-
tions * the item variances
If the matrix is not invertible, then a vector of 1s is returned.
In the case of correlation or covariance matrices with some NAs, those variables with NAs are
dropped and the SMC for the remaining variables are found. The missing SMCs are then estimated
by finding the maximum correlation for that column (with a warning).
Author(s)
William Revelle
See Also
mat.regress,fa
Examples
R <- make.hierarchical()
round(smc(R),2)
318 spider
spider Make "radar" or "spider" plots.
Description
Radar plots and spider plots are just two of the many ways to show multivariate data. radar plots
correlations as vectors ranging in length from 0 (corresponding to r=-1) to 1 (corresponding to an
r=1). The vectors are arranged radially around a circle. Spider plots connect the end points of each
vector. The plots are most appropriate if the variables are organized in some meaningful manner.
Usage
spider(y,x,data,labels=NULL,rescale=FALSE,center=FALSE,connect=TRUE,overlay=FALSE,
scale=1,ncolors=31,fill=FALSE,main=NULL,...)
radar(x,labels=NULL,center=FALSE,connect=FALSE,scale=1,ncolors=31,fill=FALSE,
add=FALSE,linetyp="solid", main="Radar Plot",...)
Arguments
yThe y variables to plot. Each y is plotted against all the x variables
xThe x variables defining each line. Each y is plotted against all the x variables
data A correlation matrix from which the x and y variables are selected
labels Labels (assumed to be colnames of the data matrix) for each x variable
rescale If TRUE, then rescale the data to have mean 0 and sd = 1. This is used if plotting
raw data rather than correlations.
center if TRUE, then lines originate at the center of the plot, otherwise they start at the
mid point.
connect if TRUE, a spider plot is drawn, if FALSE, just a radar plot
scale can be used to magnify the plot, to make small values appear larger.
ncolors if ncolors > 2, then positive correlations are plotted with shades of blue and
negative correlations shades of red. This is particularly useful if fill is TRUE.
ncolors should be an odd number, so that neutral values are coded as white.
fill if TRUE, fill the polygons with colors scaled to size of correlation
overlay If TRUE, plot multiple spiders on one plot, otherwise plot them as separate plots
add If TRUE, add a new spider diagram to the previous one.
linetyp see lty in the par options
main A label or set of labels for the plots
... Additional parameters can be passed to the underlying graphics call
splitHalf 319
Details
Displaying multivariate profiles may be done by a series of lines (see, e.g., matplot), by colors (see,
e.g., cor.plot, or by radar or spider plots.
To show just one variable as a function of several others, use radar. To make multiple plots,
use spider. An additional option when comparing just a few y values is to do overlay plots.
Alternatively, set the plotting options to do several on one page.
Value
Either a spider or radar plot
Author(s)
William Revelle
See Also
cor.plot
Examples
op <- par(mfrow=c(3,2))
spider(y=1,x=2:9,data=Thurstone,connect=FALSE) #a radar plot
spider(y=1,x=2:9,data=Thurstone) #same plot as a spider plot
spider(y=1:3,x=4:9,data=Thurstone,overlay=TRUE)
#make a somewhat oversized plot
spider(y=26:28,x=1:25,data=cor(bfi,use="pairwise"),fill=TRUE,scale=2)
par(op)
splitHalf Alternative estimates of test reliabiity
Description
Eight alternative estimates of test reliability include the six discussed by Guttman (1945), four
discussed by ten Berge and Zergers (1978) (µ0. . . µ3)as well as β(the worst split half, Revelle,
1979), the glb (greatest lowest bound) discussed by Bentler and Woodward (1980), and ωhand ωt
(McDonald, 1999; Zinbarg et al., 2005). Greatest and lowest split-half values are found by brute
force or sampling.
Usage
splitHalf(r,raw=FALSE,brute=FALSE,n.sample=10000,covar=FALSE,check.keys=TRUE,
key=NULL,use="pairwise")
guttman(r,key=NULL)
tenberge(r)
glb(r,key=NULL)
glb.fa(r,key=NULL)
320 splitHalf
Arguments
rA correlation or covariance matrix or raw data matrix.
raw return a vector of split half reliabilities
brute Use brute force to try all combinations of n take n/2.
n.sample if brute is false, how many samples of split halves should be tried?
covar Should the covariances or correlations be used for reliability calculations
check.keys If TRUE, any item with a negative loading on the first factor will be flipped in
sign
key a vector of -1, 0, 1 to select or reverse key items. See if the key vector is less
than the number of variables, then item numbers to be reverse can be specified.
use Should we find the correlations using "pairwise" or "complete" (see ?cor)
Details
Surprisingly, more than a century after Spearman (1904) introduced the concept of reliability to
psychologists, there are still multiple approaches for measuring it. Although very popular, Cron-
bach’s α(1951) underestimates the reliability of a test and over estimates the first factor saturation.
Using splitHalf for tests with 16 or fewer items, all possible splits may be found fairly easily. For
tests with 17 or more items, n.sample splits are randomly found. Thus, for 16 or fewer items, the
upper and lower bounds are precise. For 17 or more items, they are close but will probably slightly
underestimate the highest and overestimate the lowest reliabilities.
The guttman function includes the six estimates discussed by Guttman (1945), four of ten Berge and
Zergers (1978), as well as Revelle’s β(1979) using splitHalf. The companion function, omega
calculates omega hierarchical (ωh) and omega total (ωt).
Guttman’s first estimate λ1assumes that all the variance of an item is error:
λ1= 1 −tr(~
Vx)
Vx
=Vx−tr(~
Vx)
Vx
This is a clear underestimate.
The second bound, λ2, replaces the diagonal with a function of the square root of the sums of
squares of the off diagonal elements. Let C2=~
1(~
V−diag(~
V)2~
10, then
λ2=λ1+qn
n−1C2
Vx
=
Vx−tr(~
Vx) + qn
n−1C2
Vx
Effectively, this is replacing the diagonal with n * the square root of the average squared off diagonal
element.
Guttman’s 3rd lower bound, λ3, also modifies λ1and estimates the true variance of each item as the
average covariance between items and is, of course, the same as Cronbach’s α.
λ3=λ1+
VX−tr(~
VX)
n(n−1)
VX
=nλ1
n−1=n
n−11−tr(~
V)x
Vx=n
n−1
Vx−tr(~
Vx)
Vx
=α
This is just replacing the diagonal elements with the average off diagonal elements. λ2≥λ3with
λ2> λ3if the covariances are not identical.
splitHalf 321
λ3and λ2are both corrections to λ1and this correction may be generalized as an infinite set of
successive improvements. (Ten Berge and Zegers, 1978)
µr=1
Vxpo+ (p1+ (p2+. . . (pr−1+ (pr)1/2)1/2. . .)1/2)1/2, r = 0,1,2, . . .
where
ph=X
i6=j
σ2h
ij , h = 0,1,2,...r−1
and
ph=n
n−1σ2h
ij , h =r
tenberge and Zegers (1978). Clearly µ0=λ3=αand µ1=λ2.µr≥µr−1≥. . . µ1≥µ0,
although the series does not improve much after the first two steps.
Guttman’s fourth lower bound, λ4was originally proposed as any spit half reliability but has been
interpreted as the greatest split half reliability. If ~
Xis split into two parts, ~
Xaand ~
Xb, with corre-
lation rab then
λ4= 21−VXa+VXb
VX=4rab
Vx
=4rab
VXa+VXb+ 2rabVXaVXb
which is just the normal split half reliability, but in this case, of the most similar splits. For 16 or
fewer items, this is found by trying all possible splits. For 17 or more items, this is estimated by
taking n.sample random splits.
λ5, Guttman’s fifth lower bound, replaces the diagonal values with twice the square root of the
maximum (across items) of the sums of squared interitem covariances
λ5=λ1+2p¯
C2
VX
.
Although superior to λ1,λ5underestimates the correction to the diagonal. A better estimate would
be analogous to the correction used in λ3:
λ5+ =λ1+n
n−1
2p¯
C2
VX
.
λ6,Guttman’s final bound considers the amount of variance in each item that can be accounted for
the linear regression of all of the other items (the squared multiple correlation or smc), or more
precisely, the variance of the errors, e2
j, and is
λ6= 1 −Pe2
j
Vx
= 1 −P(1 −r2
smc)
Vx
.
The smc is found from all the items. A modification to Guttman λ6,λ6∗reported by the score.items
function is to find the smc from the entire pool of items given, not just the items on the selected
scale.
Guttman’s λ4is the greatest split half reliability. Although originally found here by combining the
output from three different approaches,this has now been replaced by using splitHalf to find the
322 splitHalf
maximum value by brute force (for 16 or fewer items) or by taking a substantial number of random
splits.
The algorithms that had been tried before included:
a) Do an ICLUST of the reversed correlation matrix. ICLUST normally forms the most distinct
clusters. By reversing the correlations, it will tend to find the most related clusters. Truly a weird
approach but tends to work.
b) Alternatively, a kmeans clustering of the correlations (with the diagonal replaced with 0 to make
pseudo distances) can produce 2 similar clusters.
c) Clusters identified by assigning items to two clusters based upon their order on the first principal
factor. (Highest to cluster 1, next 2 to cluster 2, etc.)
These three procedures will produce keys vectors for assigning items to the two splits. The max-
imum split half reliability is found by taking the maximum of these three approaches. This is not
elegant but is fast.
The brute force and the sampling procedures seem to provide more stable and larger estimates.
Yet another procedure, implemented in splitHalf is actually form all possible (for n items <= 16)
or sample 10,000 (or more) split halfs corrected for test length. This function returns the best and
worst splits as item keys that can be used for scoring purposes, if desired.
There are three greatest lower bound functions. One, glb finds the greatest split half reliability,
λ4. This considers the test as set of items and examines how best to partition the items into splits.
The other two, glb.fa and glb.algebraic, are alternative ways of weighting the diagonal of the
matrix.
glb.fa estimates the communalities of the variables from a factor model where the number of
factors is the number with positive eigen values. Then reliability is found by
glb = 1 −Pe2
j
Vx
= 1 −P(1 −h2)
Vx
This estimate will differ slightly from that found by glb.algebraic, written by Andreas Moeltner
which uses calls to csdp in the Rcsdp package. His algorithm, which more closely matches the
description of the glb by Jackson and Woodhouse, seems to have a positive bias (i.e., will over esti-
mate the reliability of some items; they are said to be = 1) for small sample sizes. More exploration
of these two algorithms is underway.
Compared to glb.algebraic,glb.fa seems to have less (positive) bias for smallish sample sizes
(n < 500) but larger for large (> 1000) sample sizes. This interacts with the number of variables
so that equal bias sample size differs as a function of the number of variables. The differences are,
however small. As samples sizes grow, glb.algebraic seems to converge on the population value
while glb.fa has a positive bias.
Value
beta The worst split half reliability. This is an estimate of the general factor satura-
tion.
tenberge$mu1 tenBerge mu 1 is functionally alpha
tenberge$mu2 one of the sequence of estimates mu1 ... mu3
glb glb found from factor analysis
keys scoring keys from each of the alternative methods of forming best splits
statsBy 323
Author(s)
William Revelle
References
Cronbach, L.J. (1951) Coefficient alpha and the internal strucuture of tests. Psychometrika, 16,
297-334.
Guttman, L. (1945). A basis for analyzing test-retest reliability. Psychometrika, 10 (4), 255-282.
Revelle, W. (1979). Hierarchical cluster-analysis and the internal structure of tests. Multivariate
Behavioral Research, 14 (1), 57-74.
Revelle, W. and Zinbarg, R. E. (2009) Coefficients alpha, beta, omega and the glb: comments on
Sijtsma. Psychometrika, 2009.
Ten Berge, J. M. F., & Zegers, F. E. (1978). A series of lower bounds to the reliability of a test.
Psychometrika, 43 (4), 575-579.
Zinbarg, R. E., Revelle, W., Yovel, I., & Li, W. (2005). Cronbach’s α, Revelle’s β, and Mc-
Donald’s ωh): Their relations with each other and two alternative conceptualizations of reliability.
Psychometrika, 70 (1), 123-133.
See Also
alpha,omega,ICLUST,glb.algebraic
Examples
data(attitude)
splitHalf(attitude)
splitHalf(attitude,covar=TRUE) #do it on the covariances
glb(attitude)
glb.fa(attitude)
if(require(Rcsdp)) {glb.algebraic(cor(attitude)) }
guttman(attitude)
#to show the histogram of all possible splits for the ability test
#sp <- splitHalf(ability,raw=TRUE) #this saves the results
#hist(sp$raw,breaks=101,ylab="SplitHalf reliability",main="SplitHalf
# reliabilities of a test with 16 ability items")
sp <- splitHalf(bfi[1:10],key=c(1,9,10))
statsBy Find statistics (including correlations) within and between groups for
basic multilevel analyses
324 statsBy
Description
When examining data at two levels (e.g., the individual and by some set of grouping variables), it is
useful to find basic descriptive statistics (means, sds, ns per group, within group correlations) as well
as between group statistics (over all descriptive statistics, and overall between group correlations).
Of particular use is the ability to decompose a matrix of correlations at the individual level into
correlations within group and correlations between groups.
Usage
statsBy(data, group, cors = FALSE, cor="cor", method="pearson", use="pairwise",
poly=FALSE, na.rm=TRUE)
statsBy.boot(data,group,ntrials=10,cors=FALSE,replace=TRUE,method="pearson")
statsBy.boot.summary(res.list,var="ICC2")
faBy(stats, nfactors = 1, rotate = "oblimin", fm = "minres", free = TRUE, all=FALSE,
min.n = 12,quant=.1, ...)
Arguments
data A matrix or dataframe with rows for subjects, columns for variables. One of
these columns should be the values of a grouping variable.
group The names or numbers of the variable in data to use as the grouping variables.
cors Should the results include the correlation matrix within each group? Default is
FALSE.
cor Type of correlation/covariance to find within groups and between groups. The
default is Pearson correlation. To find within and between covariances, set
cor="cov". Although polychoric, tetrachoric, and mixed correlations can be
found within groups, this does not make sense for the between groups or the
pooled within groups. In this case, correlations for each group will be as speci-
fied, but the between groups and pooled within will be Pearson. See the discus-
sion below.
method What kind of correlations should be found (default is Pearson product moment)
use How to treat missing data. use="pairwise" is the default
poly Find polychoric correlations within groups if requested.
na.rm Should missing values be deleted (na.rm=TRUE) or should we assume the data
clean?
ntrials The number of trials to run when bootstrapping statistics
replace Should the bootstrap be done by permuting the data (replace=FALSE) or sam-
pling with replacement (replace=TRUE)
res.list The results from statsBy.boot may be summarized using boot.stats
var Name of the variable to be summarized from statsBy.boot
stats The output of statsBy
nfactors The number of factors to extract in each subgroup
rotate The factor rotation/transformation
fm The factor method (see fa for details)
statsBy 325
free Allow the factor solution to be freely estimated for each individual (see note).
all Report individual factor analyses for each group as well as the summary table
min.n The minimum number of within subject cases before we factor analyze it.
quant Show the upper and lower quant quantile of the factor loadings in faBy
... Other parameters to pass to the fa function
Details
Multilevel data are endemic in psychological research. In multilevel data, observations are taken on
subjects who are nested within some higher level grouping variable. The data might be experimental
(participants are nested within experimental conditions) or observational (students are nested within
classrooms, students are nested within college majors.) To analyze this type of data, one uses
random effects models or mixed effect models, or more generally, multilevel models. There are
at least two very powerful packages (nlme and multilevel) which allow for complex analysis of
hierarchical (multilevel) data structures. statsBy is a much simpler function to give some of the
basic descriptive statistics for two level models. It is meant to supplement true multilevel modeling.
For a group variable (group) for a data.frame or matrix (data), basic descriptive statistics (mean, sd,
n) as well as within group correlations (cors=TRUE) are found for each group.
The amount of variance associated with the grouping variable compared to the total variance is the
type 1 IntraClass Correlation (ICC1): ICC1=(MSb−MSw)/(MSb+M Sw ∗(npr−1)) where
npr is the average number of cases within each group.
The reliability of the group differences may be found by the ICC2 which reflects how different the
means are with respect to the within group variability. ICC2 = (MSb −MSw)/MSb. Because
the mean square between is sensitive to sample size, this estimate will also reflect sample size.
Perhaps the most useful part of statsBy is that it decomposes the observed correlations between
variables into two parts: the within group and the between group correlation. This follows the
decomposition of an observed correlation into the pooled correlation within groups (rwg) and the
weighted correlation of the means between groups discussed by Pedazur (1997) and by Bliese in
the multilevel package.
rxy =etaxwg ∗etaywg ∗rxywg +etaxbg ∗etaybg ∗rxybg
where rxy is the normal correlation which may be decomposed into a within group and between
group correlations rxywg and rxybg and eta is the correlation of the data with the within group values,
or the group means.
It is important to realize that the within group and between group correlations are independent of
each other. That is to say, inferring from the ’ecological correlation’ (between groups) to the lower
level (within group) correlation is inappropriate. However, these between group correlations are
still very meaningful, if inferences are made at the higher level.
There are actually two ways of finding the within group correlations pooled across groups. We can
find the correlations within every group, weight these by the sample size and then report this pooled
value (pooled). This is found if the cors option is set to TRUE. It is logically equivalent to doing
a sample size weighted meta-analytic correlation. The other way, rwg, considers the covariances,
variances, and thus correlations when each subject’s scores are given as deviation score from the
group mean.
If finding tetrachoric, polychoric, or mixed correlations, these two estimates will differ, for the
pooled value is the weighted polychoric correlation, but the rwg is the Pearson correlation.
326 statsBy
Confidence values and significance of rxywg , pwg, reflect the pooled number of cases within groups,
while rxybg , pbg, the number of groups. These are not corrected for multiple comparisons.
withinBetween is an example data set of the mixture of within and between group correlations.
sim.multilevel will generate simulated data with a multilevel structure.
The statsBy.boot function will randomize the grouping variable ntrials times and find the statsBy
output. This can take a long time and will produce a great deal of output. This output can then
be summarized for relevant variables using the statsBy.boot.summary function specifying the
variable of interest. These two functions are useful in order to find if the mere act of grouping leads
to large between group correlations.
Consider the case of the relationship between various tests of ability when the data are grouped
by level of education (statsBy(sat.act,"education")) or when affect data are analyzed within and
between an affect manipulation (statsBy(flat,group="Film") ). Note in this latter example, that
because subjects were randomly assigned to Film condition for the pretest, that the pretest ICC1s
cluster around 0.
faBy uses the output of statsBy to perform a factor analysis on the correlation matrix within each
group. If the free parameter is FALSE, then each solution is rotated towards the group solution
(as much as possible). The output is a list of each factor solution, as well as a summary matrix of
loadings and interfactor correlations for all groups.
Value
means The means for each group for each variable.
sd The standard deviations for each group for each variable.
nThe number of cases for each group and for each variable.
ICC1 The intraclass correlation reflects the amount of total variance associated with
the grouping variable.
ICC2 The intraclass correlation (2) reflecting how much the groups means differ.
FThe F from a one-way anova of group means.
rwg The pooled within group correlations.
rbg The sample size weighted between group correlations.
etawg The correlation of the data with the within group values.
etabg The correlation of the data with the group means.
pbg The probability of the between group correlation
pwg The probability of the within group correlation
rIn the case that we want the correlations in each group, r is a list of the within
group correlations for every group. Set cors=TRUE
within is just another way of displaying these correlations. within is a matrix which
reports the lower off diagonal correlations as one row for each group.
pooled The sample size weighted correlations. This is just within weighted by the sam-
ple sizes. The cors option must be set to TRUE to get this. See the note.
statsBy 327
Note
If finding polychoric correlations, the two estimates of the pooled within group correlations will
differ, for the pooled value is the weighted polychoric correlation, but the rwg is the Pearson corre-
lation.
The statsBy.boot function will sometimes fail if sampling with replacement because if the group
sizes differ drastically, some groups will be empty. In this case, sample without replacement.
The statsBy.boot function can take a long time. (As I am writing this, I am running 1000 replications
of a problem with 64,000 cases and 84 groups. It is taking about 3 seconds per replication on a
MacBook Pro.)
The faBy function takes the output of statsBy (with the cors=TRUE option) and then factors each
individual subject. By default, the solutions are organized so that the factors "match" the group
solution in terms of their order. It is also possible to attempt to force the solutions to match by order
and also by using the TargetQ rotation function. (free=FALSE)
Author(s)
William Revelle
References
Pedhazur, E.J. (1997) Multiple regression in behavioral research: explanation and prediction. Har-
court Brace.
See Also
describeBy and the functions within the multilevel package.
Examples
#Taken from Pedhazur, 1997
pedhazur <- structure(list(Group = c(1L, 1L, 1L, 1L, 1L, 2L, 2L, 2L, 2L,
2L), X = c(5L, 2L, 4L, 6L, 3L, 8L, 5L, 7L, 9L, 6L), Y = 1:10), .Names = c("Group",
"X", "Y"), class = "data.frame", row.names = c(NA, -10L))
pedhazur
ped.stats <- statsBy(pedhazur,"Group")
ped.stats
#Now do this for the sat.act data set
sat.stats <- statsBy(sat.act,c("education","gender"),cor=TRUE) #group by two grouping variables
print(sat.stats,short=FALSE)
lowerMat(sat.stats$pbg) #get the probability values
#show means by groups
round(sat.stats$mean)
#Do separate factor analyses for each group
#faBy(sat.stats,1)
328 structure.diagram
structure.diagram Draw a structural equation model specified by two measurement mod-
els and a structural model
Description
Graphic presentations of structural equation models are a very useful way to conceptualize sem and
confirmatory factor models. Given a measurement model on x (xmodel) and on y (ymodel) as well
as a path model connecting x and y (phi), draw the graph. If ymodel is not specified, just draw the
measurement model (xmodel + phi). If the Rx or Ry matrices are specified, show the correlations
between the x variables, or y variables.
Perhaps even more usefully, the function returns a model appropriate for running directly in the
sem package written by John Fox. For this option to work directly, it is necessary to specfy that
errrors=TRUE.
Input can be specified as matrices or the output from fa, factanal, or a rotation package such as
GPArotation.
For symbolic graphs, the input matrices can be character strings or mixtures of character strings and
numeric vectors.
As an option, for those without Rgraphviz installed, structure.sem will just create the sem model
and skip the graph. (This functionality is now included in structure.diagram.)
structure.diagram will draw the diagram without using Rgraphviz and is probably the preferred
option. structure.graph will be removed eventually.
lavaan.diagram will draw either cfa or sem results from the lavaan package (> .4.0)
Usage
structure.diagram(fx, Phi=NULL,fy=NULL,labels=NULL,cut=.3,errors=FALSE,simple=TRUE,
regression=FALSE,lr=TRUE,Rx=NULL,Ry=NULL,digits=1,e.size=.1,
main="Structural model", ...)
structure.graph(fx, Phi = NULL,fy = NULL, out.file = NULL, labels = NULL, cut = 0.3,
errors=TRUE, simple=TRUE,regression=FALSE, size = c(8, 6),
node.font = c("Helvetica", 14), edge.font = c("Helvetica", 10),
rank.direction = c("RL", "TB", "LR", "BT"), digits = 1,
title = "Structural model", ...)
structure.sem(fx, Phi = NULL, fy = NULL,out.file = NULL, labels = NULL,
cut = 0.3, errors=TRUE, simple=TRUE,regression=FALSE)
lavaan.diagram(fit,title,...)
Arguments
fx a factor model on the x variables.
Phi A matrix of directed relationships. Lower diagonal values are drawn. If the
upper diagonal values match the lower diagonal, two headed arrows are drawn.
For a single, directed path, just the value may be specified.
structure.diagram 329
fy a factor model on the y variables (can be empty)
Rx The correlation matrix among the x variables
Ry The correlation matrix among the y variables
out.file name a file to send dot language instructions.
labels variable labels if not specified as colnames for the matrices
cut Draw paths for values > cut
fit The output from a lavaan cfa or sem
errors draw an error term for observerd variables
simple Just draw one path per x or y variable
regression Draw a regression diagram (observed variables cause Y)
lr Direction of diagram is from left to right (lr=TRUE, default) or from bottom to
top (lr=FALSE)
e.size size of the ellipses in structure.diagram
main main title of diagram
size page size of graphic
node.font font type for graph
edge.font font type for graph
rank.direction Which direction should the graph be oriented
digits Number of digits to draw
title Title of graphic
... other options to pass to Rgraphviz
Details
The recommended function is structure.diagram which does not use Rgraphviz but which does not
produce dot code either.
All three function return a matrix of commands suitable for using in the sem package. (Specify
errors=TRUE to get code that will run directly in the sem package.)
The structure.graph output can be directed to an output file for post processing using the dot graphic
language but requires that Rgraphviz is installed.
The figure is organized to show the appropriate paths between:
The correlations between the X variables (if Rx is specified)
The X variables and their latent factors (if fx is specified)
The latent X and the latent Y (if Phi is specified)
The latent Y and the observed Y (if fy is specified)
The correlations between the Y variables (if Ry is specified)
A confirmatory factor model would specify just fx and Phi, a structural model would include fx,
Phi, and fy. The raw correlations could be shown by just including Rx and Ry.
lavaan.diagram may be called from the diagram function which also will call fa.diagram,
omega.diagram or iclust.diagram, depending upon the class of the fit.
Other diagram functions include fa.diagram,omega.diagram. All of these functions use the var-
ious dia functions such as dia.rect,dia.ellipse,dia.arrow,dia.curve,dia.curved.arrow,
and dia.shape.
330 structure.diagram
Value
sem (invisible) a model matrix (partially) ready for input to John Fox’s sem package.
It is of class “mod" for prettier output.
dotfile If out.file is specified, a dot language file suitable for using in a dot graphics
program such as graphviz or Omnigraffle.
A graphic structural diagram in the graphics window
Author(s)
William Revelle
See Also
fa.graph,omega.graph,sim.structural to create artificial data sets with particular structural
properties.
Examples
fx <- matrix(c(.9,.8,.6,rep(0,4),.6,.8,-.7),ncol=2)
fy <- matrix(c(.6,.5,.4),ncol=1)
Phi <- matrix(c(1,0,0,0,1,0,.7,.7,1),ncol=3,byrow=TRUE)
f1 <- structure.diagram(fx,Phi,fy,main="A structural path diagram")
#symbolic input
X2 <- matrix(c("a",0,0,"b","e1",0,0,"e2"),ncol=4)
colnames(X2) <- c("X1","X2","E1","E2")
phi2 <- diag(1,4,4)
phi2[2,1] <- phi2[1,2] <- "r"
f2 <- structure.diagram(X2,Phi=phi2,errors=FALSE,main="A symbolic model")
#symbolic input with error
X2 <- matrix(c("a",0,0,"b"),ncol=2)
colnames(X2) <- c("X1","X2")
phi2 <- diag(1,2,2)
phi2[2,1] <- phi2[1,2] <- "r"
f3 <- structure.diagram(X2,Phi=phi2,main="an alternative representation")
#and yet another one
X6 <- matrix(c("a","b","c",rep(0,6),"d","e","f"),nrow=6)
colnames(X6) <- c("L1","L2")
rownames(X6) <- c("x1","x2","x3","x4","x5","x6")
Y3 <- matrix(c("u","w","z"),ncol=1)
colnames(Y3) <- "Y"
rownames(Y3) <- c("y1","y2","y3")
phi21 <- matrix(c(1,0,"r1",0,1,"r2",0,0,1),ncol=3)
colnames(phi21) <- rownames(phi21) <- c("L1","L2","Y")
f4 <- structure.diagram(X6,phi21,Y3)
# and finally, a regression model
structure.list 331
X7 <- matrix(c("a","b","c","d","e","f"),nrow=6)
f5 <- structure.diagram(X7,regression=TRUE)
#and a really messy regession model
x8 <- c("b1","b2","b3")
r8 <- matrix(c(1,"r12","r13","r12",1,"r23","r13","r23",1),ncol=3)
f6<- structure.diagram(x8,Phi=r8,regression=TRUE)
structure.list Create factor model matrices from an input list
Description
When creating a structural diagram or a structural model, it is convenient to not have to specify
all of the zero loadings in a structural matrix. structure.list converts list input into a design matrix.
phi.list does the same for a correlation matrix. Factors with NULL values are filled with 0s.
Usage
structure.list(nvars, f.list,f=NULL, f.labels = NULL, item.labels = NULL)
phi.list(nf,f.list, f.labels = NULL)
Arguments
nvars Number of variables in the design matrix
f.list A list of items included in each factor (for structure.list, or the factors that cor-
relate with the specified factor for phi.list
fprefix for parameters – needed in case of creating an X set and a Y set
f.labels Names for the factors
item.labels Item labels
nf Number of factors in the phi matrix
Details
This is almost self explanatory. See the examples.
Value
factor.matrix a matrix of factor loadings to model
See Also
structure.graph for drawing it, or sim.structure for creating this data structure.
332 superMatrix
Examples
fx <- structure.list(9,list(F1=c(1,2,3),F2=c(4,5,6),F3=c(7,8,9)))
fy <- structure.list(3,list(Y=c(1,2,3)),"Y")
phi <- phi.list(4,list(F1=c(4),F2=c(1,4),F3=c(2),F4=c(1,2,3)))
fx
phi
fy
superMatrix Form a super matrix from two sub matrices.
Description
Given the matrices nXm, and jYk, form the super matrix of dimensions (n+j) and (m+k) with
with elements x and y along the super diagonal. Useful when considering structural equations.
The measurement models x and y can be combined into a larger measurement model of all of the
variables. If either x or y is a list of matrices, then recursively form a super matrix of all of those
elements.
Usage
superMatrix(x,y)
super.matrix(x, y) #Deprecated
Arguments
xA n x m matrix or a list of such matrices
yA j x k matrix or a list of such matrices
Details
Several functions, e.g., sim.structural,structure.graph,make.keys use matrices that can be
thought of as formed from a set of submatrices. In particular, when using make.keys in order to
score a set of items (scoreItems or scoreOverlap) or to form specified clusters (cluster.cor), it
is convenient to define different sets of scoring keys for different sets of items and to combine these
scoring keys into one super key.
Value
A (n+j) x (m +k) matrix with appropriate row and column names
Author(s)
William Revelle
table2matrix 333
See Also
sim.structural,structure.graph,make.keys
Examples
mx <- matrix(c(.9,.8,.7,rep(0,4),.8,.7,.6),ncol=2)
my <- matrix(c(.6,.5,.4))
colnames(mx) <- paste("X",1:dim(mx)[2],sep="")
rownames(mx) <- paste("Xv",1:dim(mx)[1],sep="")
colnames(my) <- "Y"
rownames(my) <- paste("Yv",1:3,sep="")
mxy <- superMatrix(mx,my)
#show the use of a list to do this as well
key1 <- make.keys(6,list(first=c(1,-2,3),second=4:6,all=1:6)) #make a scoring key
key2 <- make.keys(4,list(EA=c(1,2),TA=c(3,4)))
superMatrix(list(key1,key2))
table2matrix Convert a table with counts to a matrix or data.frame representing
those counts.
Description
Some historical sets are reported as summary tables of counts in a limited number of bins. Trans-
forming these tables to data.frames representing the original values is useful for pedagogical pur-
poses. (E.g., transforming the original Galton table of height x cubits in order to demonstrate
regression.) The column and row names must be able to be converted to numeric values.
Usage
table2matrix(x, labs = NULL)
table2df(x, count=NULL,labs = NULL)
Arguments
xA two dimensional table of counts with row and column names that can be
converted to numeric values.
count if present, then duplicate each row count times
labs Labels for the rows and columns. These will be used for the names of the two
columns of the resulting matrix
Details
The original Galton (1888) of heights by cubits (arm length) is in tabular form. To show this as
a correlation or as a scatter plot, it is useful to convert the table to a matrix or data frame of two
columns.
This function may also be used to convert an item response pattern table into a data table. e.g., the
Bock data set bock.
334 test.psych
Value
A matrix (or data.frame) of sum(x) rows and two columns.
Author(s)
William Revelle
See Also
cubits and bock data sets
Examples
data(cubits)
cubit <- table2matrix(cubits,labs=c("height","cubit"))
describe(cubit)
ellipses(cubit,n=1)
data(bock)
responses <- table2df(bock.table[,2:6],count=bock.table[,7],labs= paste("lsat6.",1:5,sep=""))
describe(responses)
test.psych Testing of functions in the psych package
Description
Test to make sure the psych functions run on basic test data sets
Usage
test.psych(first=1,last=5,short=TRUE,all=FALSE,fapc=FALSE)
Arguments
first first=1: start with dataset first
last last=5: test for datasets until last
short short=TRUE - don’t return any analyses
all To get around a failure on certain Solaris 32 bit systems, all=FALSE is the de-
fault
fapc if fapc=TRUE, then do a whole series of tests of factor and principal component
extraction and rotations.
test.psych 335
Details
When modifying the psych package, it is useful to make sure that adding some code does not break
something else. The test.psych function tests the major functions on various standard data sets. It
also shows off a number of the capabilities of the psych package.
Uses 5 standard data sets:
USArrests Violent Crime Rates by US State (4 variables)
attitude The Chatterjee-Price Attitude Data
Harman23.cor$cov Harman Example 2.3 8 physical measurements
Harman74.cor$cov Harman Example 7.4 24 mental measurements
ability.cov$cov 8 Ability and Intelligence Tests
It also uses the bfi and ability data sets from psych
Value
out if short=FALSE, then list of the output from all functions tested
Warning
Warning messages will be thrown by fa.parallel and sometimes by fa for random datasets.
Note
Although test.psych may be used as a quick demo of the various functions in the psych packge, in
general, it is better to try the specific functions themselves. The main purpose of test.psych is to
make sure functions throw error messages or correct for weird conditions.
The datasets tested are part of the standard R data sets and represent some of the basic problems
encountered.
When version 1.1.10 was released, it caused errors when compiling and testing on some Solaris 32
bit systems. The all option was added to avoid this problem (since I can’t replicate the problem on
Macs or PCs). all=TRUE adds one more test, for a non-positive definite matrix.
Author(s)
William Revelle
Examples
#test <- test.psych()
#not run
#test.psych(all=TRUE)
# f3 <- fa(bfi[1:15],3,n.iter=5)
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="Varimax")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="varimax")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="bifactor")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="varimin")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="bentlerT")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="geominT")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="equamax")
336 tetrachoric
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="Promax")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="cluster")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="biquartimin")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="equamax")
# f3 <- fa(bfi[1:15],3,n.iter=5,rotate="Promax")
#
# fpoly <- fa(bfi[1:10],2,n.iter=5,cor="poly")
# f1 <- fa(ability,n.iter=4)
# f1p <- fa(ability,n.iter=4,cor="tet")
tetrachoric Tetrachoric, polychoric, biserial and polyserial correlations from var-
ious types of input
Description
The tetrachoric correlation is the inferred Pearson Correlation from a two x two table with the
assumption of bivariate normality. The polychoric correlation generalizes this to the n x m table.
Particularly important when doing Item Response Theory or converting comorbidity statistics using
normal theory to correlations. Input may be a 2 x 2 table of cell frequencies, a vector of cell
frequencies, or a data.frame or matrix of dichotomous data (for tetrachoric) or of numeric data (for
polychoric). The biserial correlation is between a continuous y variable and a dichotmous x variable,
which is assumed to have resulted from a dichotomized normal variable. Biserial is a special case
of the polyserial correlation, which is the inferred latent correlation between a continuous variable
(X) and a ordered categorical variable (e.g., an item response). Input for these later two are data
frames or matrices. Requires the mnormt package.
Usage
tetrachoric(x,y=NULL,correct=.5,smooth=TRUE,global=TRUE,weight=NULL,na.rm=TRUE,
delete=TRUE)
polychoric(x,smooth=TRUE,global=TRUE,polycor=FALSE,ML=FALSE, std.err=FALSE,
weight=NULL,correct=.5,progress=TRUE,na.rm=TRUE, delete=TRUE)
biserial(x,y)
polyserial(x,y)
polydi(p,d,taup,taud,global=TRUE,ML = FALSE, std.err = FALSE,
weight=NULL,progress=TRUE,na.rm=TRUE,delete=TRUE,correct=.5)
#deprecated use polychoric instead
poly.mat(x, short = TRUE, std.err = FALSE, ML = FALSE)
Arguments
xThe input may be in one of four forms:
a) a data frame or matrix of dichotmous data (e.g., the lsat6 from the bock data
set) or discrete numerical (i.e., not too many levels, e.g., the big 5 data set, bfi)
for polychoric, or continuous for the case of biserial and polyserial.
tetrachoric 337
b) a 2 x 2 table of cell counts or cell frequencies (for tetrachoric) or an n x m
table of cell counts (for both tetrachoric and polychoric).
c) a vector with elements corresponding to the four cell frequencies (for tetra-
choric)
d) a vector with elements of the two marginal frequencies (row and column) and
the comorbidity (for tetrachoric)
yA (matrix or dataframe) of discrete scores. In the case of tetrachoric, these
should be dichotomous, for polychoric not too many levels, for biserial they
should be discrete (e.g., item responses) with not too many (<10?) categories.
correct Correction value to use to correct for continuity in the case of zero entry cell for
tetrachoric, polychoric, polybi, and mixed.cor. See the examples for the effect
of correcting versus not correcting for continuity.
smooth if TRUE and if the tetrachoric/polychoric matrix is not positive definite, then
apply a simple smoothing algorithm using cor.smooth
global When finding pairwise correlations, should we use the global values of the tau
parameter (which is somewhat faster), or the local values (global=FALSE)? The
local option is equivalent to the polycor solution, or to doing one correlation
at a time. global=TRUE borrows information for one item pair from the other
pairs using those item’s frequencies. This will make a difference in the presence
of lots of missing data. With very small sample sizes with global=FALSE and
correct=TRUE, the function will fail (for as yet underdetermined reasons.
polycor A no longer used option, kept to stop other packages from breaking.
weight A vector of length of the number of observations that specifies the weights to
apply to each case. The NULL case is equivalent of weights of 1 for all cases.
short short=TRUE, just show the correlations, short=FALSE give the full hetcor out-
put from John Fox’s hetcor function if installed and if doing polychoric Depre-
cated
std.err std.err=FALSE does not report the standard errors (faster) deprecated
progress Show the progress bar (if not doing multicores)
ML ML=FALSE do a quick two step procedure, ML=TRUE, do longer maximum
likelihood — very slow! Deprecated
na.rm Should missing data be deleted
delete Cases with no variance are deleted with a warning before proceeding.
pThe polytomous input to polydi
dThe dichotomous input to polydi
taup The tau values for the polytomous variables – if global=TRUE
taud The tau values for the dichotomous variables – if globabl = TRUE
Details
Tetrachoric correlations infer a latent Pearson correlation from a two x two table of frequencies with
the assumption of bivariate normality. The estimation procedure is two stage ML. Cell frequencies
for each pair of items are found. In the case of tetrachorics, cells with zero counts are replaced with
.5 as a correction for continuity (correct=TRUE).
338 tetrachoric
The data typically will be a raw data matrix of responses to a questionnaire scored either true/false
(tetrachoric) or with a limited number of responses (polychoric). In both cases, the marginal fre-
quencies are converted to normal theory thresholds and the resulting table for each item pair is
converted to the (inferred) latent Pearson correlation that would produce the observed cell frequen-
cies with the observed marginals. (See draw.tetra and draw.cor for illustrations.)
This is a very computationally intensive function which can be speeded up considerably by using
multiple cores and using the parallel package. The number of cores to use when doing polychoric or
tetrachoric may be specified using the options command. The greatest step in speed is going from
1 core to 2. This is about a 50% savings. Going to 4 cores seems to have about at 66% savings, and
8 a 75% savings. The number of parallel processes defaults to 2 but can be modified by using the
options command: options("mc.cores"=4) will set the number of cores to 4.
The tetrachoric correlation is used in a variety of contexts, one important one being in Item Re-
sponse Theory (IRT) analyses of test scores, a second in the conversion of comorbity statistics to
correlation coefficients. It is in this second context that examples of the sensitivity of the coefficient
to the cell frequencies becomes apparent:
Consider the test data set from Kirk (1973) who reports the effectiveness of a ML algorithm for the
tetrachoric correlation (see examples).
Examples include the lsat6 and lsat7 data sets in the bock data.
The polychoric function forms matrices of polychoric correlations by an local function (polyc) and
will also report the tau values for each alternatives. Earlier versions used John Fox’s polychor
function which has now been replaced by the polyc function.
polychoric replaces poly.mat and is recommended. poly.mat is an alternative wrapper to the polycor
function.
biserial and polyserial correlations are the inferred latent correlations equivalent to the observed
point-biserial and point-polyserial correlations (which are themselves just Pearson correlations).
The polyserial function is meant to work with matrix or dataframe input and treats missing data by
finding the pairwise Pearson r corrected by the overall (all observed cases) probability of response
frequency. This is particularly useful for SAPA procedures (http://sapa-project.org) with
large amounts of missing data and no complete cases.
Ability tests and personality test matrices will typically have a cleaner structure when using tetra-
choric or polychoric correlations than when using the normal Pearson correlation. However, if either
alpha or omega is used to find the reliability, this will be an overestimate of the squared correlation
of a latent variable the observed variable.
A biserial correlation (not to be confused with the point-biserial correlation which is just a Pearson
correlation) is the latent correlation between x and y where y is continuous and x is dichotomous
but assumed to represent an (unobserved) continuous normal variable. Let p = probability of x
level 1, and q = 1 - p. Let zp = the normal ordinate of the z score associated with p. Then,
rbi =rs ∗p(pq)/zp.
The ’ad hoc’ polyserial correlation, rps is just r=r∗sqrt(n−1)/n)σy/ P(zpi)where zpi are
the ordinates of the normal curve at the normal equivalent of the cut point boundaries between the
item responses. (Olsson, 1982)
All of these were inspired by (and adapted from) John Fox’s polychor package which should be
used for precise ML estimates of the correlations. See, in particular, the hetcor function in the
polychor package. The results from polychoric match the polychor answers to at least 5 decimals
when using correct=FALSE, and global = FALSE.
tetrachoric 339
Particularly for tetrachoric correlations from sets of data with missing data, the matrix will some-
times not be positive definite. Various smoothing alternatives are possible, the one done here is
to do an eigen value decomposition of the correlation matrix, set all negative eigen values to 10 *
.Machine$double.eps, normalize the positive eigen values to sum to the number of variables, and
then reconstitute the correlation matrix. A warning is issued when this is done.
For very small data sets, the correction for continuity for the polychoric correlations can lead to
difficulties, particularly if using the global=FALSE option, or if doing just one correlation at a time.
Setting a smaller correction value (i.e., correct =.1) seems to help.
For combinations of continous, categorical, and dichotomous variables, see mixed.cor.
If using data with a variable number of response alternatives, it is necessary to use the global=FALSE
option in polychoric.
Value
rho The (matrix) of tetrachoric/polychoric/biserial correlations
tau The normal equivalent of the cutpoints
fixed If any correlations were adjusted for continuity, the total number of adjustments
will be reported.
Note
For tetrachoric, in the degenerate case of a cell entry with zero observations, a correction for con-
tinuity is applied and .5 is added to the cell entry. A warning is issued. If correct=FALSE the
correction is not applied. This correction is, by default, on. It can be adjusted by specifying a
smaller value. See the examples.
For correct=FALSE, the results agree perfectly with John Fox’s polycor function.
Switched to using sadmvn from the mnormt package to speed up by 50%.
Author(s)
William Revelle
References
A. Gunther and M. Hofler. Different results on tetrachorical correlations in mplus and stata-stata
announces modified procedure. Int J Methods Psychiatr Res, 15(3):157-66, 2006.
David Kirk (1973) On the numerical approximation of the bivariate normal (tetrachoric) correlation
coefficient. Psychometrika, 38, 259-268.
U.Olsson, F.Drasgow, and N.Dorans (1982). The polyserial correlation coefficient. Psychometrika,
47:337-347.
See Also
mixed.cor to find the correlations between mixtures of continuous, polytomous, and dichtomous
variables. See also the polychor function in the polycor package. irt.fa uses the tetrachoric
function to do item analysis with the fa factor analysis function. draw.tetra shows the logic
behind a tetrachoric correlation (for teaching purpuses.)
340 tetrachoric
Examples
#if(require(mnormt)) {
data(bock)
tetrachoric(lsat6)
polychoric(lsat6) #values should be the same
tetrachoric(matrix(c(44268,193,14,0),2,2)) #MPLUS reports.24
#Do not apply continuity correction -- compare with previous analysis!
tetrachoric(matrix(c(44268,193,14,0),2,2),correct=0)
#the default is to add correct=.5 to 0 cells
tetrachoric(matrix(c(61661,1610,85,20),2,2)) #Mplus reports .35
tetrachoric(matrix(c(62503,105,768,0),2,2)) #Mplus reports -.10
tetrachoric(matrix(c(24875,265,47,0),2,2)) #Mplus reports 0
polychoric(matrix(c(61661,1610,85,20),2,2)) #Mplus reports .35
polychoric(matrix(c(62503,105,768,0),2,2)) #Mplus reports -.10
polychoric(matrix(c(24875,265,47,0),2,2)) #Mplus reports 0
#Do not apply continuity correction- compare with previous analysis
tetrachoric(matrix(c(24875,265,47,0),2,2), correct=0)
polychoric(matrix(c(24875,265,47,0),2,2), correct=0) #the same result
#these next examples are impossible!
tetrachoric(c(0.02275000, 0.0227501320, 0.500000000))
tetrachoric(c(0.0227501320, 0.0227501320, 0.500000000))
#give a vector of two marginals and the comorbidity
tetrachoric(c(.2, .15, .1))
tetrachoric(c(.2, .1001, .1))
#} else {
# message("Sorry, you must have mnormt installed")}
# 4 plots comparing biserial to point biserial and latent Pearson correlation
set.seed(42)
x.4 <- sim.congeneric(loads =c(.9,.6,.3,0),N=1000,short=FALSE)
y <- x.4$latent[,1]
for(i in 1:4) {
x <- x.4$observed[,i]
r <- round(cor(x,y),1)
ylow <- y[x<= 0]
yhigh <- y[x > 0]
yc <- c(ylow,yhigh)
rpb <- round(cor((x>=0),y),2)
rbis <- round(biserial(y,(x>=0)),2)
ellipses(x,y,ylim=c(-3,3),xlim=c(-4,3),pch=21 - (x>0),
main =paste("r = ",r,"rpb = ",rpb,"rbis =",rbis))
dlow <- density(ylow)
dhigh <- density(yhigh)
points(dlow$y*5-4,dlow$x,typ="l",lty="dashed")
lines(dhigh$y*5-4,dhigh$x,typ="l")
thurstone 341
}
thurstone Thurstone Case V scaling
Description
Thurstone Case V scaling allows for a scaling of objects compared to other objects. As one of the
cases considered by Thurstone, Case V makes the assumption of equal variances and uncorrelated
distributions.
Usage
thurstone(x, ranks = FALSE, digits = 2)
Arguments
xA square matrix or data frame of preferences, or a rectangular data frame or
matrix rank order choices.
ranks TRUE if rank orders are presented
digits number of digits in the goodness of fit
Details
Louis L. Thurstone was a pioneer in psychometric theory and measurement of attitudes, interests,
and abilities. Among his many contributions was a systematic analysis of the process of comparative
judgment (thurstone, 1927). He considered the case of asking subjects to successively compare
pairs of objects. If the same subject does this repeatedly, or if subjects act as random replicates of
each other, their judgments can be thought of as sampled from a normal distribution of underlying
(latent) scale scores for each object, Thurstone proposed that the comparison between the value of
two objects could be represented as representing the differences of the average value for each object
compared to the standard deviation of the differences between objects. The basic model is that each
item has a normal distribution of response strength and that choice represents the stronger of the two
response strengths. A justification for the normality assumption is that each decision represents the
sum of many independent inputs and thus, through the central limit theorem, is normally distributed.
Thurstone considered five different sets of assumptions about the equality and independence of the
variances for each item (Thurston, 1927). Torgerson expanded this analysis slightly by consid-
ering three classes of data collection (with individuals, between individuals and mixes of within
and between) crossed with three sets of assumptions (equal covariance of decision process, equal
correlations and small differences in variance, equal variances).
The data may be either a square matrix of dataframe of preferences (as proportions with the proba-
bility of the column variable being chosen over the row variable) or a matrix or dataframe of rank
orders ( 1 being prefered to 2, etc.)
342 tr
Value
GF Goodness of fit 1 = 1 - sum(squared residuals/squared original) for lower off
diagonal.
Goodness of fit 2 = 1 - sum(squared residuals/squared original) for full matrix.
residual square matrix of residuals (of class dist)
data The original choice data
...
Author(s)
William Revelle
References
Thurstone, L. L. (1927) A law of comparative judgments. Psychological Review, 34, 273-286.
Revelle, W. An introduction to psychometric theory with applications in R. (in preparation), Springer.
http://personality-project.org/r/book
Examples
data(vegetables)
thurstone(veg)
tr Find the trace of a square matrix
Description
Hardly worth coding, if it didn’t appear in so many formulae in psychometrics, the trace of a
(square) matrix is just the sum of the diagonal elements.
Usage
tr(m)
Arguments
mA square matrix
Details
The tr function is used in various matrix operations and is the sum of the diagonal elements of a
matrix.
Tucker 343
Value
The sum of the diagonal elements of a square matrix.
i.e. tr(m) <- sum(diag(m)).
Examples
m <- matrix(1:16,ncol=4)
m
tr(m)
Tucker 9 Cognitive variables discussed by Tucker and Lewis (1973)
Description
Tucker and Lewis (1973) introduced a reliability coefficient for ML factor analysis. Their example
data set was previously reported by Tucker (1958) and taken from Thurstone and Thurstone (1941).
The correlation matrix is a 9 x 9 for 710 subjects and has two correlated factors of ability: Word
Fluency and Verbal.
Usage
data(Tucker)
Format
A data frame with 9 observations on the following 9 variables.
t42 Prefixes
t54 Suffixes
t45 Chicago Reading Test: Vocabulary
t46 Chicago Reading Test: Sentences
t23 First and last letters
t24 First letters
t27 Four letter words
t10 Completion
t51 Same or Opposite
Details
The correlation matrix from Tucker (1958) was used in Tucker and Lewis (1973) for the Tucker-
Lewis Index of factoring reliability.
344 vegetables
Source
Tucker, Ledyard (1958) An inter-battery method of factor analysis, Psychometrika, 23, 111-136.
References
L.~Tucker and C.~Lewis. (1973) A reliability coefficient for maximum likelihood factor analysis.
Psychometrika, 38(1):1–10.
F.~J. Floyd and K.~F. Widaman. (1995) Factor analysis in the development and refinement of
clinical assessment instruments., Psychological Assessment, 7(3):286 – 299.
Examples
data(Tucker)
fa(Tucker,2,n.obs=710)
omega(Tucker,2)
vegetables Paired comparison of preferences for 9 vegetables
Description
A classic data set for demonstrating Thurstonian scaling is the preference matrix of 9 vegetables
from Guilford (1954). Used by Guiford, Nunnally, and Nunally and Bernstein, this data set allows
for examples of basic scaling techniques.
Usage
data(vegetables)
Format
A data frame with 9 choices on the following 9 vegetables. The values reflect the perecentage of
times where the column entry was preferred over the row entry.
Turn Turnips
Cab Cabbage
Beet Beets
Asp Asparagus
Car Carrots
Spin Spinach
S.Beans String Beans
Peas Peas
Corn Corn
VSS 345
Details
Louis L. Thurstone was a pioneer in psychometric theory and measurement of attitudes, interests,
and abilities. Among his many contributions was a systematic analysis of the process of comparative
judgment (thurstone, 1927). He considered the case of asking subjects to successively compare
pairs of objects. If the same subject does this repeatedly, or if subjects act as random replicates of
each other, their judgments can be thought of as sampled from a normal distribution of underlying
(latent) scale scores for each object, Thurstone proposed that the comparison between the value of
two objects could be represented as representing the differences of the average value for each object
compared to the standard deviation of the differences between objects. The basic model is that each
item has a normal distribution of response strength and that choice represents the stronger of the two
response strengths. A justification for the normality assumption is that each decision represents the
sum of many independent inputs and thus, through the central limit theorem, is normally distributed.
Thurstone considered five different sets of assumptions about the equality and independence of the
variances for each item (Thurston, 1927). Torgerson expanded this analysis slightly by consid-
ering three classes of data collection (with individuals, between individuals and mixes of within
and between) crossed with three sets of assumptions (equal covariance of decision process, equal
correlations and small differences in variance, equal variances).
This vegetable data set is used by Guilford and by Nunnally to demonstrate Thurstonian scaling.
Source
Guilford, J.P. (1954) Psychometric Methods. McGraw-Hill, New York.
References
Nunnally, J. C. (1967). Psychometric theory., McGraw-Hill, New York.
Revelle, W. An introduction to psychometric theory with applications in R. (in preparation), Springer.
http://personality-project.org/r/book
See Also
thurstone
Examples
data(vegetables)
thurstone(veg)
VSS Apply the Very Simple Structure, MAP, and other criteria to determine
the appropriate number of factors.
346 VSS
Description
There are multiple ways to determine the appropriate number of factors in exploratory factor anal-
ysis. Routines for the Very Simple Structure (VSS) criterion allow one to compare solutions of
varying complexity and for different number of factors. Graphic output indicates the "optimal"
number of factors for different levels of complexity. The Velicer MAP criterion is another good
choice. nfactors finds and plots several of these alternative estimates.
Usage
vss(x, n = 8, rotate = "varimax", diagonal = FALSE, fm = "minres",
n.obs=NULL,plot=TRUE,title="Very Simple Structure",use="pairwise",cor="cor",...)
VSS(x, n = 8, rotate = "varimax", diagonal = FALSE, fm = "minres",
n.obs=NULL,plot=TRUE,title="Very Simple Structure",use="pairwise",cor="cor",...)
nfactors(x,n=20,rotate="varimax",diagonal=FALSE,fm="minres",n.obs=NULL,
title="Number of Factors",pch=16,use="pairwise", cor="cor",...)
Arguments
xa correlation matrix or a data matrix
nNumber of factors to extract – should be more than hypothesized!
rotate what rotation to use c("none", "varimax", "oblimin","promax")
diagonal Should we fit the diagonal as well
fm factoring method – fm="pa" Principal Axis Factor Analysis, fm = "minres" min-
imum residual (OLS) factoring fm="mle" Maximum Likelihood FA, fm="pc"
Principal Components"
n.obs Number of observations if doing a factor analysis of correlation matrix. This
value is ignored by VSS but is necessary for the ML factor analysis package.
plot plot=TRUE Automatically call VSS.plot with the VSS output, otherwise don’t
plot
title a title to be passed on to VSS.plot
pch the plot character for the nfactors plots
use If doing covariances or Pearson R, should we use "pairwise" or "complete cases"
cor What kind of correlation to find, defaults to Pearson but see fa for the choices
... parameters to pass to the factor analysis program The most important of these is
if using a correlation matrix is covmat= xx
Details
Determining the most interpretable number of factors from a factor analysis is perhaps one of the
greatest challenges in factor analysis. There are many solutions to this problem, none of which
is uniformly the best. "Solving the number of factors problem is easy, I do it everyday before
breakfast. But knowing the right solution is harder" (Kaiser, 195x).
Techniques most commonly used include
1) Extracting factors until the chi square of the residual matrix is not significant.
VSS 347
2) Extracting factors until the change in chi square from factor n to factor n+1 is not significant.
3) Extracting factors until the eigen values of the real data are less than the corresponding eigen
values of a random data set of the same size (parallel analysis) fa.parallel.
4) Plotting the magnitude of the successive eigen values and applying the scree test (a sudden drop
in eigen values analogous to the change in slope seen when scrambling up the talus slope of a
mountain and approaching the rock face.
5) Extracting principal components until the eigen value < 1.
6) Extracting factors as long as they are interpetable.
7) Using the Very Simple Structure Criterion (VSS).
8) Using Wayne Velicer’s Minimum Average Partial (MAP) criterion.
Each of the procedures has its advantages and disadvantages. Using either the chi square test or the
change in square test is, of course, sensitive to the number of subjects and leads to the nonsensical
condition that if one wants to find many factors, one simply runs more subjects. Parallel analysis is
partially sensitive to sample size in that for large samples the eigen values of random factors will
be very small. The scree test is quite appealling but can lead to differences of interpretation as to
when the scree "breaks". The eigen value of 1 rule, although the default for many programs, seems
to be a rough way of dividing the number of variables by 3. Extracting interpretable factors means
that the number of factors reflects the investigators creativity more than the data. VSS, while very
simple to understand, will not work very well if the data are very factorially complex. (Simulations
suggests it will work fine if the complexities of some of the items are no more than 2).
Most users of factor analysis tend to interpret factor output by focusing their attention on the largest
loadings for every variable and ignoring the smaller ones. Very Simple Structure operationalizes
this tendency by comparing the original correlation matrix to that reproduced by a simplified version
(S) of the original factor matrix (F). R = SS’ + U2. S is composed of just the c greatest (in absolute
value) loadings for each variable. C (or complexity) is a parameter of the model and may vary from
1 to the number of factors.
The VSS criterion compares the fit of the simplified model to the original correlations: VSS = 1
-sumsquares(r*)/sumsquares(r) where R* is the residual matrix R* = R - SS’ and r* and r are the
elements of R* and R respectively.
VSS for a given complexity will tend to peak at the optimal (most interpretable) number of factors
(Revelle and Rocklin, 1979).
Although originally written in Fortran for main frame computers, VSS has been adapted to micro
computers (e.g., Macintosh OS 6-9) using Pascal. We now release R code for calculating VSS.
Note that if using a correlation matrix (e.g., my.matrix) and doing a factor analysis, the parame-
ters n.obs should be specified for the factor analysis: e.g., the call is VSS(my.matrix,n.obs=500).
Otherwise it defaults to 1000.
Wayne Velicer’s MAP criterion has been added as an additional test for the optimal number of
components to extract. Note that VSS and MAP will not always agree as to the optimal number.
The nfactors function will do a VSS, find MAP, and report a number of other criteria (e.g., BIC,
complexity, chi square, ...)
A variety of rotation options are available. These include varimax, promax, and oblimin. Others
can be added. Suggestions are welcome.
348 VSS
Value
A data.frame with entries: map: Velicer’s MAP values (lower values are better)
dof: degrees of freedom (if using FA)
chisq: chi square (from the factor analysis output (if using FA)
prob: probability of residual matrix > 0 (if using FA)
sqresid: squared residual correlations
RMSEA: the RMSEA for each number of factors
BIC: the BIC for each number of factors
eChiSq: the empirically found chi square
eRMS: Empirically found mean residual
eCRMS: Empirically found mean residual corrected for df
eBIC: The empirically found BIC based upon the eChiSq
fit: factor fit of the complete model
cfit.1: VSS fit of complexity 1
cfit.2: VSS fit of complexity 2
...
cfit.8: VSS fit of complexity 8
cresidiual.1: sum squared residual correlations for complexity 1
...: sum squared residual correlations for complexity 2 ..8
Author(s)
William Revelle
References
http://personality-project.org/r/vss.html, Revelle, W. An introduction to psychometric
theory with applications in R (in prep) Springer. Draft chapters available at http://personality-project.
org/r/book/
Revelle, W. and Rocklin, T. 1979, Very Simple Structure: an Alternative Procedure for Estimat-
ing the Optimal Number of Interpretable Factors, Multivariate Behavioral Research, 14, 403-414.
http://personality-project.org/revelle/publications/vss.pdf
Velicer, W. (1976) Determining the number of components from the matrix of partial correlations.
Psychometrika, 41, 321-327.
See Also
VSS.plot,ICLUST,omega,fa.parallel
Examples
#test.data <- Harman74.cor$cov
#my.vss <- VSS(test.data,title="VSS of 24 mental tests")
#print(my.vss[,1:12],digits =2)
#VSS.plot(my.vss, title="VSS of 24 mental tests")
#now, some simulated data with two factors
VSS.parallel 349
#VSS(sim.circ(nvar=24),fm="minres" ,title="VSS of 24 circumplex variables")
VSS(sim.item(nvar=24),fm="minres" ,title="VSS of 24 simple structure variables")
VSS.parallel Compare real and random VSS solutions
Description
Another useful test for the number of factors is when the eigen values of a random matrix are greater
than the eigen values of a a real matrix. Here we show VSS solutions to random data. A better test
is probably fa.parallel.
Usage
VSS.parallel(ncases, nvariables,scree=FALSE,rotate="none")
Arguments
ncases Number of simulated cases
nvariables number of simulated variables
scree Show a scree plot for random data – see omega
rotate rotate="none" or rotate="varimax"
Value
VSS like output to be plotted by VSS.plot
Author(s)
William Revelle
References
Very Simple Structure (VSS)
See Also
fa.parallel,VSS.plot,ICLUST,omega
Examples
#VSS.plot(VSS.parallel(200,24))
350 VSS.plot
VSS.plot Plot VSS fits
Description
The Very Simple Structure criterion ( VSS) for estimating the optimal number of factors is plotted
as a function of the increasing complexity and increasing number of factors.
Usage
VSS.plot(x, title = "Very Simple Structure", line = FALSE)
Arguments
xoutput from VSS
title any title
line connect different complexities
Details
Item-factor models differ in their "complexity". Complexity 1 means that all except the greatest
(absolute) loading for an item are ignored. Basically a cluster model (e.g., ICLUST). Complexity 2
implies all except the greatest two, etc.
Different complexities can suggest different number of optimal number of factors to extract. For
personality items, complexity 1 and 2 are probably the most meaningful.
The Very Simple Structure criterion will tend to peak at the number of factors that are most inter-
pretable for a given level of complexity. Note that some problems, the most interpretable number
of factors will differ as a function of complexity. For instance, when doing the Harman 24 psycho-
logical variable problems, an unrotated solution of complexity one suggests one factor (g), while a
complexity two solution suggests that a four factor solution is most appropriate. This latter probably
reflects a bi-factor structure.
For examples of VSS.plot output, see http://personality-project.org/r/r.vss.html
Value
A plot window showing the VSS criterion varying as the number of factors and the complexity of
the items.
Author(s)
Maintainer: William Revelle <revelle@northwestern.edu>
References
http://personality-project.org/r/r.vss.html
VSS.scree 351
See Also
VSS,ICLUST,omega
Examples
test.data <- Harman74.cor$cov
my.vss <- VSS(test.data) #suggests that 4 factor complexity two solution is optimal
VSS.plot(my.vss,title="VSS of Holzinger-Harmon problem") #see the graphics window
VSS.scree Plot the successive eigen values for a scree test
Description
Cattell’s scree test is one of most simple ways of testing the number of components or factors in a
correlation matrix. Here we plot the eigen values of a correlation matrix as well as the eigen values
of a factor analysis.
Usage
scree(rx,factors=TRUE,pc=TRUE,main="Scree plot",hline=NULL,add=FALSE)
VSS.scree(rx, main = "scree plot")
Arguments
rx a correlation matrix or a data matrix. If data, then correlations are found using
pairwise deletions.
factors If true, draw the scree for factors
pc If true, draw the scree for components
hline if null, draw a horizontal line at 1, otherwise draw it at hline (make negative to
not draw it)
main Title
add Should multiple plots be drawn?
Details
Among the many ways to choose the optimal number of factors is the scree test. A better function to
show the scree as well as compare it to randomly parallel solutions is found found in fa.parallel
Author(s)
William Revelle
352 winsor
References
http://personality-project.org/r/vss.html
See Also
fa.parallel VSS.plot,ICLUST,omega
Examples
scree(attitude)
#VSS.scree(cor(attitude)
winsor Find the Winsorized scores, means, sds or variances for a vector, ma-
trix, or data.frame
Description
Among the robust estimates of central tendency are trimmed means and Winsorized means. This
function finds the Winsorized scores. The top and bottom trim values are given values of the
trimmed and 1- trimmed quantiles. Then means, sds, and variances are found.
Usage
winsor(x, trim = 0.2, na.rm = TRUE)
winsor.mean(x, trim = 0.2, na.rm = TRUE)
winsor.means(x, trim = 0.2, na.rm = TRUE)
winsor.sd(x, trim = 0.2, na.rm = TRUE)
winsor.var(x, trim = 0.2, na.rm = TRUE)
Arguments
xA data vector, matrix or data frame
trim Percentage of data to move from the top and bottom of the distributions
na.rm Missing data are removed
Details
Among the many robust estimates of central tendency, some recommend the Winsorized mean.
Rather than just dropping the top and bottom trim percent, these extreme values are replaced with
values at the trim and 1- trim quantiles.
withinBetween 353
Value
A scalar or vector of winsorized scores or winsorized means, sds, or variances (depending upon the
call).
Author(s)
William Revelle with modifications suggested by Joe Paxton and a further correction added (Jan-
uary, 2009) to preserve the original order for the winsor case.
References
Wilcox, Rand R. (2005) Introduction to robust estimation and hypothesis testing. Elsevier/Academic
Press. Amsterdam ; Boston.
See Also
interp.median
Examples
data(sat.act)
winsor.means(sat.act) #compare with the means of the winsorized scores
y <- winsor(sat.act)
describe(y)
xy <- data.frame(sat.act,y)
#pairs.panels(xy) #to see the effect of winsorizing
x <- matrix(1:100,ncol=5)
winsor(x)
winsor.means(x)
y <- 1:11
winsor(y,trim=.5)
withinBetween An example of the distinction between within group and between group
correlations
Description
A demonstration that a correlation may be decomposed to a within group correlation and a be-
tween group correlations and these two correlations are independent. Between group correlations
are sometimes called ecological correlations, the decomposition into within and between group cor-
relations is a basic concept in multilevel modeling. This data set shows the composite correlations
between 9 variables, representing 16 cases with four groups.
Usage
data(withinBetween)
354 withinBetween
Format
A data frame with 16 observations on the following 10 variables.
Group An example grouping factor.
V1 A column of 16 observations
V2 A column of 16 observations
V3 A column of 16 observations
V4 A column of 16 observations
V5 A column of 16 observations
V6 A column of 16 observations
V7 A column of 16 observations
V8 A column of 16 observations
V9 A column of 16 observations
Details
Correlations between individuals who belong to different natural groups (based upon e.g., ethnicity,
age, gender, college major,or country) reflect an unknown mixture of the pooled correlation within
each group as well as the correlation of the means of these groups. These two correlations are inde-
pendent and do not allow inferences from one level (the group) to the other level (the individual).
This data set shows this independence. The within group correlations between 9 variables are set to
be 1, 0, and -1 while those between groups are also set to be 1, 0, -1. These two sets of correlations
are crossed such that V1, V4, and V7 have within group correlations of 1, as do V2, V5 and V8,
and V3, V6 and V9. V1 has a within group correlation of 0 with V2, V5, and V8, and a -1 within
group correlation with V3, V6 and V9. V1, V2, and V3 share a between group correlation of 1, as
do V4, V5 and V6, and V7, V8 and V9. The first group has a 0 between group correlation with the
second and a -1 with the third group.
statsBy can decompose the observed correlation in the between and within correlations. sim.multilevel
can produce similar data.
Source
The data were created for this example
References
P. D. Bliese. Multilevel modeling in R (2.3) a brief introduction to R, the multilevel package and
the nlme package, 2009.
Pedhazur, E.J. (1997) Multiple regression in behavioral research: explanation and prediction. Har-
court Brace.
Revelle, W. An introduction to psychometric theory with applications in R (in prep) Springer. Draft
chapters available at http://personality-project.org/r/book/
See Also
statsBy,describeBy, and sim.multilevel
Yule 355
Examples
data(withinBetween)
pairs.panels(withinBetween,bg=c("red","blue","white","black")[withinBetween[,1]],
pch=21,ellipses=FALSE)
stats <- statsBy(withinBetween,'Group')
print(stats,short=FALSE)
Yule From a two by two table, find the Yule coefficients of association, con-
vert to phi, or tetrachoric, recreate table the table to create the Yule
coefficient.
Description
One of the many measures of association is the Yule coefficient. Given a two x two table of counts
a b R1
c d R2
C1 C2 n
Yule Q is (ad - bc)/(ad+bc).
Conceptually, this is the number of pairs in agreement (ad) - the number in disagreement (bc)
over the total number of paired observations. Warren (2008) has shown that Yule’s Q is one of
the “coefficients that have zero value under statistical independence, maximum value unity, and
minimum value minus unity independent of the marginal distributions" (p 787).
ad/bc is the odds ratio and Q = (OR-1)/(OR+1)
Yule’s coefficient of colligation is Y = (sqrt(OR) - 1)/(sqrt(OR)+1) Yule.inv finds the cell entries
for a particular Q and the marginals (a+b,c+d,a+c, b+d). This is useful for converting old tables of
correlations into more conventional phi or tetrachoric correlations tetrachoric
Yule2phi and Yule2tetra convert the Yule Q with set marginals to the correponding phi or tetrachoric
correlation.
Bonett and Price show that the Q and Y coefficients are both part of a general family of coefficients
raising the OR to a power (c). If c=1, then this is Yule’s Q. If .5, then Yule’s Y, if c = .75, then this
is Digby’s H. They propose that c = .5 - (.5 * min(cell probabilty)^2 is a more general coefficient.
YuleBonett implements this for the 2 x 2 case, YuleCor for the data matrix case.
Usage
YuleBonett(x,c=1,bonett=FALSE,alpha=.05) #find the generalized Yule cofficients
YuleCor(x,c=1,bonett=FALSE,alpha=.05) #do this for a matrix
Yule(x,Y=FALSE) #find Yule given a two by two table of frequencies
#find the frequencies that produce a Yule Q given the Q and marginals
Yule.inv(Q,m,n=NULL)
#find the phi coefficient that matches the Yule Q given the marginals
356 Yule
Yule2phi(Q,m,n=NULL)
Yule2tetra(Q,m,n=NULL,correct=TRUE)
#Find the tetrachoric correlation given the Yule Q and the marginals
#(deprecated) Find the tetrachoric correlation given the Yule Q and the marginals
Yule2poly(Q,m,n=NULL,correct=TRUE)
Arguments
xA vector of four elements or a two by two matrix, or, in the case of YuleBonett
or YuleCor, this can also be a data matrix
c1 returns Yule Q, .5, Yule’s Y, .75 Digby’s H
bonett If FALSE, then find Q, Y, or H, if TRUE, then find the generalized Bonett coffi-
cient
alpha The two tailed probability for confidence intervals
YY=TRUE return Yule’s Y coefficient of colligation
QEither a single Yule coefficient or a matrix of Yule coefficients
mThe vector c(R1,C2) or a two x two matrix of marginals or a four element vector
of marginals. The preferred form is c(R1,C1)
nThe number of subjects (if the marginals are given as frequencies
correct When finding a tetrachoric correlation, should small cell sizes be corrected for
continuity. See {link{tetrachoric} for a discussion.
Details
Yule developed two measures of association for two by two tables. Both are functions of the odds
ratio
Value
QThe Yule Q coefficient
RA two by two matrix of counts
result If given matrix input, then a matrix of phis or tetrachorics
rho From YuleBonett and YuleCor
ci The upper and lower confidence intervals in matrix form (From YuleBonett and
YuleCor).
Note
Yule.inv is currently done by using the optimize function, but presumably could be redone by solv-
ing a quadratic equation.
Author(s)
William Revelle
Yule 357
References
Yule, G. Uday (1912) On the methods of measuring association between two attributes. Journal of
the Royal Statistical Society, LXXV, 579-652
Bonett, D.G. and Price, R.M, (2007) Statistical Inference for Generalized Yule Coefficients in 2 x 2
Contingency Tables. Sociological Methods and Research, 35, 429-446.
Warrens, Matthijs (2008), On Association Coefficients for 2x2 Tables and Properties That Do Not
Depend on the Marginal Distributions. Psychometrika, 73, 777-789.
See Also
See Also as phi,tetrachoric,Yule2poly.matrix,Yule2phi.matrix
Examples
Nach <- matrix(c(40,10,20,50),ncol=2,byrow=TRUE)
Yule(Nach)
Yule.inv(.81818,c(50,60),n=120)
Yule2phi(.81818,c(50,60),n=120)
Yule2tetra(.81818,c(50,60),n=120)
phi(Nach) #much less
#or express as percents and do not specify n
Nach <- matrix(c(40,10,20,50),ncol=2,byrow=TRUE)
Nach/120
Yule(Nach)
Yule.inv(.81818,c(.41667,.5))
Yule2phi(.81818,c(.41667,.5))
Yule2tetra(.81818,c(.41667,.5))
phi(Nach) #much less
YuleCor(ability[,1:4],,TRUE)
YuleBonett(Nach,1) #Yule Q
YuleBonett(Nach,.5) #Yule Y
YuleBonett(Nach,.75) #Digby H
YuleBonett(Nach,,TRUE) #Yule* is a generalized Yule
Index
∗Topic cluster
00.psych,5
cluster.fit,40
cluster.loadings,41
cluster.plot,43
iclust,165
ICLUST.cluster,170
iclust.diagram,171
ICLUST.graph,173
ICLUST.rgraph,176
∗Topic datagen
sim,296
sim.congeneric,304
sim.hierarchical,305
sim.item,307
sim.structure,312
sim.VSS,314
simulation.circ,315
∗Topic datasets
ability,14
affect,16
Bechtoldt,21
bfi,26
blot,33
bock,34
burt,35
cities,39
cubits,76
cushny,77
Dwyer,92
epi,95
epi.bfi,98
galton,148
Gleser,153
Gorsuch,154
Harman,155
Harman.5,156
Harman.8,157
Harman.political,159
heights,162
income,180
iqitems,182
msq,209
neo,218
peas,239
sat.act,270
Schmid,273
Tucker,343
vegetables,344
withinBetween,353
∗Topic hplot
bi.bars,29
biplot.psych,30
cluster.plot,43
cor.plot,52
densityBy,78
diagram,87
draw.tetra,89
ellipses,93
error.bars,99
error.bars.by,102
error.crosses,105
errorCircles,107
fa.diagram,119
iclust.diagram,171
ICLUST.graph,173
ICLUST.rgraph,176
multi.hist,217
pairs.panels,234
scatter.hist,272
spider,318
structure.diagram,328
VSS.scree,351
∗Topic models
00.psych,5
alpha,17
bestScales,24
circ.tests,37
358
INDEX 359
cor.ci,50
cor.smooth,55
cor2dist,58
corFiml,59
corr.test,60
correct.cor,62
count.pairwise,72
cta,73
describe,80
describeBy,83
dummy.code,91
eigen.loadings,92
fa,109
fa.multi,125
factor.congruence,133
factor.fit,136
factor.model,137
factor.residuals,138
factor.rotate,139
factor.scores,141
factor.stats,143
factor2cluster,145
fisherz,147
ICLUST.sort,178
irt.1p,184
irt.fa,186
irt.item.diff.rasch,190
irt.responses,191
kaiser,193
KMO,194
make.keys,198
mardia,200
mat.sort,202
mediate,204
mixed.cor,207
mssd,215
omega,220
outlier,230
p.rep,231
paired.r,233
phi,241
phi.demo,242
phi2tetra,243
polychor.matrix,248
predict.psych,249
principal,251
Promax,256
r.test,262
rangeCorrection,264
read.clipboard,265
rescale,267
residuals.psych,268
scaling.fits,271
schmid,274
score.alpha,276
score.irt,278
score.multiple.choice,280
scoreItems,282
scoreOverlap,287
SD,291
setCor,292
sim.anova,301
sim.hierarchical,305
sim.multilevel,310
sim.VSS,314
statsBy,323
structure.list,331
table2matrix,333
thurstone,341
VSS,345
VSS.parallel,349
VSS.plot,350
Yule,355
∗Topic multivariate
00.psych,5
alpha,17
bestScales,24
biplot.psych,30
block.random,32
circ.tests,37
cluster.fit,40
cluster.loadings,41
cluster.plot,43
cluster2keys,45
cohen.kappa,46
comorbidity,49
cor.ci,50
cor.plot,52
cor.smooth,55
cor.wt,57
cor2dist,58
corFiml,59
corr.test,60
correct.cor,62
cortest.bartlett,64
cortest.mat,65
360 INDEX
cosinor,67
count.pairwise,72
densityBy,78
describe,80
diagram,87
draw.tetra,89
dummy.code,91
eigen.loadings,92
ellipses,93
error.bars,99
error.bars.by,102
error.crosses,105
errorCircles,107
fa,109
fa.diagram,119
fa.extension,122
fa.multi,125
fa.parallel,128
fa.sort,132
factor.congruence,133
factor.model,137
factor.residuals,138
factor.rotate,139
factor.scores,141
factor.stats,143
factor2cluster,145
fisherz,147
geometric.mean,149
glb.algebraic,150
harmonic.mean,160
headTail,161
ICC,163
iclust,165
ICLUST.cluster,170
iclust.diagram,171
ICLUST.graph,173
ICLUST.rgraph,176
ICLUST.sort,178
irt.1p,184
irt.fa,186
irt.item.diff.rasch,190
irt.responses,191
kaiser,193
KMO,194
logistic,195
lowerUpper,197
make.keys,198
mardia,200
mat.sort,202
matrix.addition,203
mediate,204
mixed.cor,207
mssd,215
multi.hist,217
omega,220
omega.graph,228
outlier,230
paired.r,233
pairs.panels,234
parcels,237
partial.r,238
phi,241
phi.demo,242
plot.psych,244
polar,247
polychor.matrix,248
predict.psych,249
principal,251
print.psych,255
Promax,256
psych.misc,259
r.test,262
rangeCorrection,264
read.clipboard,265
rescale,267
residuals.psych,268
reverse.code,269
scatter.hist,272
schmid,274
score.alpha,276
score.irt,278
score.multiple.choice,280
scoreItems,282
scoreOverlap,287
scrub,290
setCor,292
sim,296
sim.anova,301
sim.congeneric,304
sim.hierarchical,305
sim.item,307
sim.multilevel,310
sim.structure,312
sim.VSS,314
simulation.circ,315
smc,317
INDEX 361
spider,318
splitHalf,319
statsBy,323
structure.diagram,328
structure.list,331
superMatrix,332
test.psych,334
tetrachoric,336
tr,342
VSS,345
VSS.plot,350
VSS.scree,351
Yule,355
∗Topic package
00.psych,5
∗Topic univar
describe,80
describeBy,83
interp.median,181
p.rep,231
rescale,267
winsor,352
∗Topic utilities
df2latex,84
%+% (matrix.addition),203
00.psych,5
00.psych-package (00.psych),5
ability,14,183
affect,16,213,214
all.income (income),180
alpha,7,8,10,17,163,277,286,289,294,
323
anova.psych,118
anova.psych (print.psych),255
Bechtoldt,21,258
bestItems,25
bestItems (bestScales),24
bestScales,24,24,25
bfi,8,13,26,130
bfi.dictionary,25
bi.bars,28,29,218
bifactor,22,120,155,156
bifactor (Promax),256
biplot.psych,30
biquartimin,22,120
biquartimin (Promax),256
biserial,207
biserial (tetrachoric),336
block.random,32
blot,33
bock,34,333,334,338
burt,35,56,114,155,156
Chen (Schmid),273
circ.sim,314
circ.sim (sim.item),307
circ.sim.plot,316
circ.sim.plot (simulation.circ),315
circ.simulation,38
circ.simulation (simulation.circ),315
circ.tests,8,37,248,309,316
circadian.cor,6,12,69
circadian.cor (cosinor),67
circadian.F,69
circadian.F (cosinor),67
circadian.linear.cor,6,12,69
circadian.linear.cor (cosinor),67
circadian.mean,6,12,69
circadian.mean (cosinor),67
circadian.phase (cosinor),67
circadian.reliability,69
circadian.reliability (cosinor),67
circadian.sd (cosinor),67
circadian.stats,69
circadian.stats (cosinor),67
circular.cor,69
circular.cor (cosinor),67
circular.mean,69
circular.mean (cosinor),67
cities,8,13,39
city.location (cities),39
cluster.cor,6,8,10,19,21,25,41,42,45,
50–52,62,63,146,198,199,256,
282,286,288,294,295,332
cluster.cor (scoreOverlap),287
cluster.fit,40,138,169,171,179
cluster.loadings,10,41,63,256,277,286
cluster.plot,7,43,246,248
cluster2keys,45
cohen.kappa,46
comorbidity,12,49
con2cat (sim.item),307
congeneric.sim (sim.congeneric),304
cor,61,208
cor.ci,50,53,54,62,236,285
cor.plot,9,12,51,52,198,202,236,319
362 INDEX
cor.plot.upperLowerCi,51,52,54,236
cor.smooth,36,55,114
cor.smoother (cor.smooth),55
cor.test,62,234
cor.wt,57,83,236
cor2dist,58
cor2latex (df2latex),84
corCi (cor.ci),50
corFiml,59
corPlot (cor.plot),52
corr.p,61,239
corr.p (corr.test),60
corr.test,7,9,52,54,60,234,236,241,
261,263
correct.cor,10,62,277,286
cortest,66
cortest (cortest.mat),65
cortest.bartlett,12,64,67,114,118
cortest.jennrich,65,66
cortest.mat,12,62,65,65,66,263
cortest.normal,65,66
cosinor,6,12,67
count.pairwise,10,60,72,186,187
cov.wt,57,58
cta,73,74
cta.15,74
cubits,8,13,76,149,162,240,334
cushny,77
d2r (fisherz),147
densityBy,78,82,84
describe,5,7,9,79,80,83,84,99,106,201,
216,223
describe.by,9,82,102,201,292
describe.by (describeBy),83
describeBy,7,79,83,106,108,216,327,354
describeData,81,82
describeData (describe),80
df2latex,12,25,84
dia.arrow,172,329
dia.arrow (diagram),87
dia.cone (diagram),87
dia.curve,172,329
dia.curve (diagram),87
dia.curved.arrow,329
dia.curved.arrow (diagram),87
dia.ellipse,172,329
dia.ellipse (diagram),87
dia.ellipse1 (diagram),87
dia.rect,172,329
dia.rect (diagram),87
dia.self (diagram),87
dia.shape,329
dia.shape (diagram),87
dia.triangle (diagram),87
diagram,11,87,119,168,169,329
diff,216
draw.cor,338
draw.cor (draw.tetra),89
draw.tetra,89,338,339
dummy.code,12,91,91
Dwyer,92,124
eigen.loadings,10,92
ellipses,77,93,162
epi,95
epi.bfi,13,98
equamax (Promax),256
error.bars,5,7,9,79,82,84,99,103,104,
106
error.bars.by,9,79,82,84,101,102,106,
108
error.bars.tab,100
error.crosses,9,80,82,101,104,105,108
errorCircles,106,107
fa,6,7,9,10,25,26,30,31,43,53,54,56,
60,64,66,85,87,109,112–114,
121–124,127–129,132,133,135,
139,141–143,145,146,158,
186–188,191,193–195,202,214,
222,244,246,248,250–252,254,
258,268,288,294,305,317,324,
328,339
fa.congruence (factor.congruence),133
fa.diagram,7,9,11,88,119,119,133,228,
329
fa.extend,123
fa.extend (fa.extension),122
fa.extension,9,10,113,118,122,123,214,
268
fa.graph,9,11,13,44,119,276,330
fa.graph (fa.diagram),119
fa.lookup,24,25,118
fa.lookup (bestScales),24
fa.multi,118,125
fa.organize,118
fa.organize (fa.sort),132
INDEX 363
fa.parallel,7,9,128,130,254,347–349,
351,352
fa.parallel.poly,9,130
fa.plot,31,246
fa.plot (cluster.plot),43
fa.poly,30,31,90,113,114
fa.rgraph,119
fa.rgraph (fa.diagram),119
fa.sort,9,25,118,132
fa.stats (factor.stats),143
fa2irt (irt.fa),186
fa2latex (df2latex),84
faBy,326,327
faBy (statsBy),323
fac (fa),109
factanal,111–113,115,144,253
factor.congruence,11,123,133,254
factor.fit,11,40,41,136,137,138
factor.minres,7,9,145,258
factor.minres (fa),109
factor.model,11,137
factor.pa,7–10,145,244,258,266
factor.pa (fa),109
factor.plot (cluster.plot),43
factor.residuals,11,138
factor.rotate,11,139,258
factor.scores,9,31,110,112,116,141,141
factor.stats,142,143
factor.wls,9
factor.wls (fa),109
factor2cluster,6,8,11,41,42,45,118,
145,146,179,254,287–289,294,
295
fisherz,12,147
fisherz2r,12
fisherz2r (fisherz),147
flat (affect),16
galton,8,13,77,148,162,240
geometric.mean,9,149
glb (splitHalf),319
glb.algebraic,10,13,150,322,323
glb.fa,152,322
Gleser,153
Gorsuch,154
guttman,6,7,9–11,21,150,152,223,284,
286,317
guttman (splitHalf),319
Harman,23,36,155,158
Harman.5,156
Harman.8,156,157
Harman.Burt,36
Harman.political,158,159,195
Harman74.cor,156,158
harmonic.mean,9,150,160
head,161
headTail,161
headtail,9
headtail (headTail),161
heights,8,13,76,77,149,162,162,240
het.diagram,88,119,121
het.diagram (fa.diagram),119
histBy (multi.hist),217
histo.density (multi.hist),217
Holzinger,155,258
Holzinger (Bechtoldt),21
Holzinger.9,155
ICC,6,10,12,46,163
ICC2latex (df2latex),84
ICLUST,6–8,18,21,40–45,115,118,121,
136–139,144,146,170–174,177,
178,222,226,244,246,248,256,
266,275,276,283,288,294,295,
306,314,323,348–352
ICLUST (iclust),165
iclust,9,26,87,165,238,289
ICLUST.cluster,138,169,170,179
ICLUST.diagram,11,88
ICLUST.diagram (iclust.diagram),171
iclust.diagram,166,171,329
ICLUST.graph,7,10,11,44,121,138,166,
168,169,171,173,177–179,226
iclust.graph (ICLUST.graph),173
ICLUST.rgraph,10,13,168,169,171–173,
176,230
ICLUST.sort,42,178
iclust.sort,169
iclust.sort (ICLUST.sort),178
income,180
interp.boxplot (interp.median),181
interp.median,9,82,181,353
interp.q (interp.median),181
interp.qplot.by (interp.median),181
interp.quantiles (interp.median),181
interp.quart (interp.median),181
interp.quartiles (interp.median),181
364 INDEX
interp.values (interp.median),181
iqitems,8,13,14,182
irt.0p (irt.1p),184
irt.1p,184
irt.2p (irt.1p),184
irt.discrim,185
irt.discrim (irt.item.diff.rasch),190
irt.fa,6,8,9,14,15,27,28,34,35,56,85,
90,113,114,117,183–185,186,
187,188,190,191,208,225,244,
246,268,278,279,286,305,339
irt.item.diff.rasch,13,185,190
irt.person.rasch,13,191
irt.person.rasch (irt.1p),184
irt.responses,188,191,279,286
irt.select,187
irt.select (irt.fa),186
irt.stats.like,246,279
irt.stats.like (score.irt),278
irt.tau,279
irt.tau (score.irt),278
irt2latex (df2latex),84
item.dichot (sim.item),307
item.lookup,25
item.lookup (bestScales),24
item.sim,243,305,314
item.sim (sim.item),307
kaiser,10,116,193,193
keysort (parcels),237
KMO,114,118,194
kurtosi,9,82,292
kurtosi (mardia),200
lavaan.diagram,88,329
lavaan.diagram (structure.diagram),328
layout,90
logistic,185,195,299
logit (logistic),195
lookup,25
lookup (bestScales),24
lowerCor,52,62,241,260,261
lowerCor (psych.misc),259
lowerMat,52,62,260
lowerMat (psych.misc),259
lowerUpper,62,197,261
lsat6 (bock),34
lsat7 (bock),34
mahalanobis,231
make.congeneric,313
make.congeneric (sim.congeneric),304
make.hierarchical,226,230,313
make.hierarchical (sim.hierarchical),
305
make.keys,8,10,27,45,52,198,214,279,
282,286,287,332,333
MAP,6,7,10,130
MAP (VSS),345
maps (affect),16
mardia,200
mat.regress,6,8,11,239,288,289,317
mat.regress (setCor),292
mat.sort,54,202
matReg,294
matReg (setCor),292
matrix.addition,203
mean,150
median,182
mediate,204,294
minkowski (ellipses),93
misc (psych.misc),259
mixed.cor,10,113,207,339
moderate.diagram (mediate),204
msq,13,16,209,289
mssd,215
multi.hist,9,217,273
mvrnorm,306
nearPD,56
neo,218
nfactors,117,132,346
nfactors (VSS),345
omega,6,7,10,13,15,18,19,21,22,53,87,
118,120,122,123,127,128,138,
155,163,165,167,169,171,183,
187,220,222,224,225,228–230,
244,246,268,276,277,282–284,
286,306,309,320,323,348,349,
351,352
omega.diagram,11,88,329
omega.diagram (omega.graph),228
omega.graph,6,7,10,11,121,221,226,228,
276,330
omega2latex (df2latex),84
omegaFromSem,224,225
omegaFromSem (omega),220
INDEX 365
omegah (omega),220
omegaSem,10,224,225
omegaSem (omega),220
optim,112
options,208,338
outlier,230
p.adjust,61,62
p.rep,6,12,231
p.rep.r,234
paired.r,12,233,263
pairs,236
pairs.panels,5,7,9,31,81,82,93,94,234,
235,272,273
pairwiseDescribe (count.pairwise),72
panel.cor (pairs.panels),234
panel.ellipse (pairs.panels),234
panel.hist (pairs.panels),234
panel.lm (pairs.panels),234
panel.smoother (pairs.panels),234
parcels,237
partial.r,6,10,61,238
pca (principal),251
peas,8,13,77,149,239
phi,8,12,50,241,355,357
phi.demo,11,12,242,249
phi.list (structure.list),331
phi2poly,12,13,249
phi2poly (phi2tetra),243
phi2poly.matrix,12,244
phi2poly.matrix (polychor.matrix),248
phi2tetra,242,243
plot.irt,186
plot.irt (plot.psych),244
plot.poly,186
plot.poly (plot.psych),244
plot.poly.parallel (fa.parallel),128
plot.psych,11,44,188,244
plot.residuals,246
plot.residuals (plot.psych),244
polar,8,12,247
poly.mat,7,10,13
poly.mat (tetrachoric),336
polychor.matrix,13,248
polychoric,6,8,10,27,31,55,56,113,114,
186–188,191,207,209,225,242,
261,305
polychoric (tetrachoric),336
polydi (tetrachoric),336
polyserial,10,11,207
polyserial (tetrachoric),336
predict,10
predict.psych,116,118,249,254
principal,7–9,26,30,31,53,64,87,92,
111,115,117,124,133,135,139,
144–146,222,244,246,250,251,
258,268,288,294,295
princomp,252
print.psych,114,133,169,255
progressBar (psych.misc),259
Promax,6,225,256
promax,258
psych,9,80
psych (00.psych),5
psych-package (00.psych),5
psych.misc,259
r.con,6,12,263
r.con (fisherz),147
r.test,6,8,12,62,234,261,262
r2d (fisherz),147
r2t (fisherz),147
radar,318,319
radar (spider),318
rangeCorrection,264,264
read.clipboard,5,7,9,81,82,265,287
read.clipboard.csv,9
read.clipboard.lower,9,198
read.clipboard.upper,9
read.https (read.clipboard),265
reflect (psych.misc),259
Reise (Bechtoldt),21
rescale,9,267,267,291
resid.psych (residuals.psych),268
residuals,115
residuals.psych,268
response.frequencies,282
response.frequencies (scoreItems),282
reverse.code,269,291
rmssd,216
rmssd (mssd),215
sat.act,8,13,270
scale,267
scaling.fits,12,271
scatter.hist,90,236,272
Schmid,273
schmid,6,10,13,221,222,226,274,306
366 INDEX
schmid.leiman (Schmid),273
score.alpha,276
score.irt,185,188,191,192,209,278,278,
286
score.items,6–8,10,19,27,45,63,198,
208,209,237,238,252,256,276,
277,279,281,282,321
score.items (scoreItems),282
score.multiple.choice,6,8,10,192,280,
286
scoreItems,6,20,21,25,51,52,112,118,
168,198,199,214,222,282,284,
285,287–289,332
scoreOverlap,52,198,199,214,286,287,
288,332
scree (VSS.scree),351
scrub,12,290
SD,291
set.cor,10,234,294
set.cor (setCor),292
setCor,206,292
setCor.diagram,206
shannon,261
shannon (psych.misc),259
sim,11,296,296,303
sim.anova,8,11,296,300,301
sim.circ,6,8,11,38,296,300,316
sim.circ (sim.item),307
sim.congeneric,6,11,296,300,304
sim.correlation,312,313
sim.correlation (sim.structure),312
sim.dichot,8,296,300
sim.dichot (sim.item),307
sim.general,299
sim.hierarchical,6,11,296,300,305,309,
316
sim.irt,11,185,188,296
sim.item,6,8,11,296,300,307
sim.minor,11,130,132,296,299,300
sim.multilevel,300,310,326,354
sim.npl,296
sim.npn,296
sim.omega,296,299
sim.parallel,296,300
sim.poly,296
sim.poly.ideal,296,299
sim.poly.ideal.npl,296
sim.poly.ideal.npn,296
sim.poly.npl,296
sim.poly.npn,296
sim.rasch,185,296
sim.simplex,296,298
sim.spherical,11,309
sim.spherical (sim.item),307
sim.structural,6,8,11,296,309,316,330,
332,333
sim.structural (sim.structure),312
sim.structure,298,299,306,312,331
sim.VSS,11,296,314
simulation.circ,309,315,315
skew,9,82,292
skew (mardia),200
smc,10,130,317
spider,12,318,319
splitHalf,319,320–322
statsBy,9,57,58,79,83,84,106–108,216,
236,311,323,325,326,354
statsBy.boot,326
statsBy.boot.summary,326
structure.diagram,11,88,121,228,299,
328
structure.graph,11,13,331–333
structure.graph (structure.diagram),328
structure.list,313,331
structure.sem (structure.diagram),328
summary,80
summary.psych (print.psych),255
super.matrix (superMatrix),332
superMatrix,199,332
table2df,9,34,77,162
table2df (table2matrix),333
table2matrix,76,77,162,333
tableF,261
tableF (psych.misc),259
tail,161
target.rot,6,113,221,252
target.rot (Promax),256
TargetQ (Promax),256
tenberge (splitHalf),319
test.all,261
test.all (psych.misc),259
test.psych,13,334
tetrachor,12
tetrachor (tetrachoric),336
tetrachoric,6,8,10,11,15,31,35,55,56,
90,113,114,183,185–188,191,
INDEX 367
207,209,242,244,259,261,336,
355,357
Thurstone,13
Thurstone (Bechtoldt),21
thurstone,12,272,341,345
topBottom (headTail),161
tr,12,342
Tucker,13,343
varimin (Promax),256
veg (vegetables),344
vegetables,8,13,272,344
vgQ.bimin (Promax),256
vgQ.targetQ (Promax),256
vgQ.varimin (Promax),256
violinBy,82,84
violinBy (densityBy),78
VSS,6,7,10,40,41,115,117,130,132,
136–139,144,165,168,169,171,
179,226,244,246,247,254,276,
306,314,345,350,351
vss,7
vss (VSS),345
VSS.parallel,7,10,132,349
VSS.plot,7,10,132,174,178,246,348,349,
350,352
VSS.scree,7,10,254,351
VSS.sim (sim.VSS),314
VSS.simulate,243
VSS.simulate (sim.VSS),314
West (Schmid),273
winsor,352
withinBetween,311,326,353
wkappa,12
wkappa (cohen.kappa),46
Yule,8,12,50,241,242,355
Yule.inv,12,242
Yule2phi,12,242,248,249
Yule2phi (Yule),355
Yule2phi.matrix,244,357
Yule2phi.matrix (polychor.matrix),248
Yule2poly,249
Yule2poly (Yule),355
Yule2poly.matrix,357
Yule2poly.matrix (polychor.matrix),248
Yule2tetra,12,248
Yule2tetra (Yule),355
YuleBonett (Yule),355
YuleCor (Yule),355

Navigation menu