Version 1.4.5 is the development release of the psych package. It is available at the repository at http://personality-project.org/r. The released version is 1.4.4. The second digit reflects the year (i.e., 2014), the third set the month (i.e., 1.4.3 was released in March of 2014, the last two digits reflect the day of any modifications).
The psych package includes functions and data sets to do classic and modern psychometrics and to analyze personality and experimental psychological data sets. The psych package has been developed as a supplement to courses in research methods in psychology, personality research, and graduate level psychometric theory. The functions are a supplement to the text (in progress): An introduction to psychometric theory with applications in R.
Additional functions are added sporadically.
This news file reports changes that have been made as the package has been developed.
Suggestions are welcome, but the current list includes (and has included for a long time, so lets be patient):
Add confirmatory clustering to ICLUST
Get cluster scores in ICLUST – analogous to factor scores (requested by Ben Shalet)
Add signal detetection function as an option for dichotmous data. (in progress)
Add the ability to create multiple groups in sim.irt functions
Add labels to factor.plot for more than 2 factors? – this will be hard
Find canonical loadings in set.cor
Add omega factor extension figure option (requested by Sylia Wilson)
add option to do subject density and item density plot to IRT plot.
note that the df in set.cor should reflect the number of cases actually used
add the ability to scale radar plots of raw data from min to max, and add a scale to radar and spider plots (e.g. circular histograms)
add the ability to add labels to lavaan.diagram
clean up lavaan diagram so that the output is more readable
Add the ability to return the true scores for subjects when simulating structures. This will help fitting reliability models but will require not using mvrnorm
To help those who want to simulate a bifactor model, make it explicit
A few tweaks to corr.p to be compatible with corr.test
Improved statsBy so that it can properly handle two grouping variables. statsBy also labels the columns in the within object.
Added pairwise names for factor intercorrelations in print.psych.fa.ci.
Two bugs were introduced into corr.test in version 1.4.3. Both fixed. (reported by louis-charles vannier and Clemens Fell)
Fixed print.psych.fa so that it properly sorts the complexities (bug reported by Kai Horstman)
Fixed fa so that if a non-available rotation is requested, a warning is issued.
Fixed alpha for case of 2 variables.
Fixed the naming of correlations in cor.ci
Fixed the empirical confidence intervals for alpha in alpha.
Added best.scale to empirically combine the best items for a scale to predict particular criteria.
Changed the call to best.items to match fa.lookup.
Added a warning for cor.ci if some of the correlations are NA as would be the resulting scales. Will still find the other correlations but be careful.
Added empirical estimates of fit to the fa and fa.stats functions. These are more robust to misfit for matrices that are not positive definite. (Requested by David Condon and Lorien Elleman).
Added these empirical estimates of fit to vss.
Added the nfactors function to do a call to vss and then do nice graphics
Added fa.organize to hand sort factor output by factor number and by item order.
Added item.lookup to organize items by factor and then by item mean.
Added densityBy to show violin plots. Will do this by groups.
Added cats eyes to error.bars and error.bars.by.
Added normal theory confidence intervals to corr.test. (Suggested by Alexander Weiss)
Added the ability to just resample and not simulate in fa.parallel. (Requested by Ashley Kendall.)
Fixed fa.parallel and fa.parallel.poly for case of very low frequency data. We now resample until we get a matrix with no NA values. (Problem reported by Eric Green).
Added a modification to error bars to print a dark circle at the mean. ( Suggested by Jaroslaw Arlet.)
Add significance and confidence intervals to partial.r (Suggested by Elizabeth Schubach).
cor.plot was incorrectly plotting the numbers by col instead of row. Just a problem for asymmetric matrices. Reported by David Condon.
corr.p was not printing correctly. Reported by Robin Beaumount.
Fixed print.psych.fa to cover several weird cases of no x$R2 or stats.df that were causing HDMD package to fail.
Added the epi data set and epi.dictionary of items.
Added bfi.dictionary to show the item contents of the bfi example.
Added BIC, SABIC, RMSEA, and complexity for each factor model in VSS (vss) (requested by David Condon)
added the alias of fa.congruence and fa.stats as calls to factor.congruence and factor.stats in a slow move to consistency
Added fa.lookup to allow for sorted factor loadings with item contents if using a dictionary for the items.
Modified print.psych.fa so that warnings about how factor score correlations are improper are suppressed (as is the printing of the improper correlations).
defined global=TRUE as default in fa.parallel.poly (reported by Koji E. Kosugi)
changed if(max(R2 > 1) to if(any(max(R2 > 1)) in print.psych.fa (reported by David Condon)
Fixed the se calculation in describe which had been incorrectly changed in version 1.4.1. Thanks to Terry Jorgensen and others for reporting this.
Fixed fa.sort so the that communalities and complexities are sorted as well
Substantial improvements in speed through the use of the parallel package and some coding improvements.
Added parallel processing (multicores) to polychoric, tetrachoric, and mixed.cor. With 2 cores this saves 50%, but this is not a linear function of the number of cores: with 4 cores the savings seems to be 67% and with 8 cores 75%. Requires the parallel package which seems to be supplied with core R. Apparently this benefit will not help those running PCs which don't implement forking.
Added parallel processing using multicores to fa when calculating confidence intervals. Basically changed the interation loop to an lapply and then made that a mclapply.
Added parallel processing using multicores to fa.parallel, cor.ci and alpha (for the empirical bootstrap).
Yet another speed improvement by introducing tableF: a cut down/speeded up version of table. This works only for tabulating two integer vectors, but by reducing error checking, is at least twice as fast. This is used in particular in polychoric and tetrachoric, and therefore in mixed.cor as well.
Changed fa so that confidence intervals are raw rather than fisherz transformed before averaging.
Modified ICLUST to increase speed by about 50% by doing the smcs once instead of every iteration. This does not use multicores so the savings will be for PCs as well as OS X and Unix. 50% of time is now spent doing matrix multiplication – can we speed this up by not rescoring all clusters each time, but rather doing incremental changes?
Minor tweaks to increase speed when creating lists and matrices.
Added omega2latex and modified fa2latex to allow not printing small values.
Changed score.items to scoreItems in a continuing switch to camelCase. score.items will eventually be deprecated unless the entire score. functions are integrated somehow.
Added option to sim.congeneric (and others) to do multiple simulations with same theta values. (Actually, it was there all along. Just set.seed(some value) before each run).
A minor tweak to r.test to better report the call in the case of a correlated correlation.
A minor tweak to corr.test so that if all the sample sizes are the same, just report one number.
Reduced the number of executable examples to speed up compilation. The examples are just commented out. Worth running for understanding.
Added the Gleser data set for an example of generalizability theory.
Added the ability to find scores automatically in omega.
Added the ability data set (just the iqitems data set scored for correct/incorrect). This is more convenient for demonstrations.
Added the check option in describe (turning it off increases speed drastically, but at the risk of not detecting logical or factor data).
Added the describeData helper to do a quick pass to list the data types in a data.frame.
Rewrote the summary function to use switch rather than a series of ifs.
Added a parameter to factor.scores to allow the data and a correlation matrix to both be passed for scoring.
Added Structure as an output for principal (requested by Emeka Egbuna)
Added best.items and lookup as two helper functions.
sim.cor.poly – fixed so that it actually works.
Fixed alpha so that it will properly reverse key if given correlations. (Not sure when this stopped working.)
Corrected the sign of at.rest in the TA scoring key for the msq.Rd file
Corrected reverse.code to not add a constant to positive items (reported by Jian Jin)
Perhaps finally fixed pairs.panels so it will not change options for graphics window.
Fixed polychoric so that it will not blow up if the number of response alternatives are not equal. (Reported by Jeanette Lim )
Fixed fa.poly so that the scores are calculated based upon the polychoric correlations rather than the Pearson correlations (based partly on trying to solve another problem reported by Jeanette Lim)
Added a brief wrapper function (corFiml) to call FIML procedures taken from lavaan. Inspired by some code from Joshua Wiley and with the help of Ashley Brown. The lavaan functions necessary for this are not exported from lavaan and hence have been partially moved (with adaptations) to psych.
Completely modified guttman to use the splitHalf function. This provides much more accurate estimates of the greatest lower bound.
Added the splitHalf function to find all possible split halves of scales of up to 16 items, and to sample repeatedly random splits for more than 16 items.
Modified plot.parallel.poly and fa.parallel.poly to allow choice in plotting between both, fa, and pc. (Matches the fa.parallel function).
Added the weight option to fa.poly, polychoric, tetrachoric, and mixed cor. This allows for cases to be given unequal weights. Requested by Fabricio Flalho.
Enhanced the documentation for irt.fa and plot.irt to explain when to use plot.irt vs. plot.poly (Requested by David Condon).
Enhanced cor2latex with the help of Davide Morselli. Now can find correlations directly and can also (if desired) show probability stars.
Added standard errors and 95% confidence boundaries to alpha and score.items (suggested by Doug Lawson)
Minor tweaks to irt2latex and fa2latex.
Corrected the documentation to fa to specify that it allows targetT and targetQ rotation.
Corrected kaiser to return as class "psych" and "fa" so that solutions can be used in fa.diagram. (Reported by Gouri Shankar Mishra)
Corrected alpha so that it reports r.drop in the case of working from a correlation matrix. (Reported by Nicolas Hubner)
A serious speedup to tetrachoric and polychoric was initiated with the help of Jason French. The increase in speed is roughly 1- (nc-1)^2/nc^2 where nc is the number of categories. Thus, for tetrachorics where nc=2, this is a 75% reduction, whereas for polychorics of 6 item responses this is just a 30% reduction.
Added the ability to rotate the numbers in cor.plot by specifying srt. (Thanks to a suggestion on Stack OverFlow by shujaa).
Added the use parameter to the fa function to allow for other than pairwise correlations.
Added an invisible return from the print.psych.fa function to include the variances accounted for and the cumulative variance accounted for. (Requested by Eric Green).
Fixed a bug in print.psych.fa so that it did not always work with the HDMD package.
Modified fa.parallel.poly so that the progress bar is not called during the tetrachorics or polychoric simulations.
print.psych.fa did not work for covariance input when adding the complexity option.
Added Yule2phi and Yule2tetra to convert Yule Q coefficients to phis or tetrachorics. Completely rewrote Yule.inv and Yule2phi.
Added sim.poly.mat to generate polytomous or dichotomous items given a particular item structure.
Modified describeBy to allow specification of digits in the matrix output.
Modified multi.hist to allow one more control over graphic output. (Requested by Lars Carlsen).
added iclust.sort to documentation of iclust.
Added label option to fa2latex (requested by Robert Carlisie)
Added sanitize.latex function to all the latex functions so that they properly escape underscore and ampersand.
Added method option to scatter.hist to allow for other correlation methods
Modified tetrachoric and polychoric so that (0,1) and (1,2) data can be mixed together. tetrachoric and polychoric agree for dichotomous data (as they should)
Added an option to the mixed.cor function to allow for spearman or kendall correlations.
Added Hoffman's complexity index for the factor analyses output
Modified make.keys to allow addressing items by name rather than location.
Modified alpha to allow for specify items to reverse by name or location.
Modified the msq data set Rd file to address by name rather than location
Minor tweaks to fa.plot and cluster.plot to allow more graphic control
modify structure diagram so that the error arrows are to the left for lr printing
Modified fa.plot and cluster.plot to allow positioning and sizing of item labels for two dimensional plots.
Made sure all usage and example lines are less than 90 characters.
Added the ability to calculate factors scores in fa.poly.
Significant improvement in biplot.psych to allow much greater control over plotting.
Added cor.ci to find bootstrapped confidence intervals for raw and synthetic correlations.
Added group reliability calculation for omega.
return (invisibly) the eigen values of the factors and components from scree.
Added unit weighted multiple correlation and unit weighted set correlation to set.cor.
Added test.all to allow for testing whether other packages work (or at least their examples) when psych is loaded. (Can be used for any set of packages.)
Added the ability to simulate as well as resample dichotomous data in fa.parallel.poly (requested by Balal Izanloo).
Drop (with a warning) items that have no variance in alpha (requested by Eric Green). This feature has also been added to score.items.
Modified upperLower so that the rownames come from the lower diagonal matrix, and the column names from the upper diagonal matrix.
fixed statsBy so that it correctly reports the pooled values (reported by John Rauthman)
corrected a problem with fa that was leading to errors when handling very non-positive matrices
fixed a bug in Yule.inv that was leading to impossible values being reported
fixed the way that biplot.psych returns the op$mfrow parameter
applied the same fix to pairs.panels to perhaps fix the recurring problem
Fixed fa.parallel to not break if just principal components output is requested
Identified and fixed bug in mixed.cor for case of improper data.
Identified and fixed? bug in mixed.cor so it will treat randomly mixed data
Corrected corr.test and cor.p to properly pass the adjust option (reported by David Weisman and Russell Pierce)
Corrected error.bars, error.bars.by and errorCircles to use n-1 for df in the call to qt. Modified the documentation to point out that we are using the t-distribution. (Thanks to Trevor Dodds)
Fixed fa.parallel so that it correctly prints out the legend for the case of fa="pc" or fa="fa" (reported by Andrew Hsiao and others)
Corrected the way confidence intervals were found in fa.poly
Correctly name the scored variables in case of impute = 'none' (reported by David Condon
Complains, rather than blows up when doing irt.fa and plot.poly for Heywood cases.
Added column names to factor scores.
Corrected the way that data are randomized in the fa.parallel.poly function so that each variable retains (roughly) the original difficulty. Also did this for fa.parallel. Reported by Balal Izanloo.
Fixed (?) progressBar so that it does not freeze the R.Gui for Mac.
Cleaned up help files so that lines don't over run the page in the pdf version. (Suggested by Jennie Miller). Still seems to be a problem.
Fixed factor.stats so that it does not complain when finding scores for principal
Fixed fa2latex so that the apa option correctly puts in the last cr
Added ECV to omega to give an estimate of unidimensionality
Add a parameter to set.cor and mat.regress to allow for square data matrices (suggested by Thomas Richardson).
Changed the default in principal to find scores (scores=TRUE). This makes it compatible with fa.
Fixed principal so that it labels the components as rotated or transformed (as documented and previously implemented. Somehow this feature went away.
Fixed fa so that specifying SMC=FALSE will replace the diagonals with 1s instead of 0s. (Seems to have been an error introduced when the SMC= a vector option was introduced)
Minor bug fix to sim.omega so that it will not choke on badly structured sem commands.
Added fa2irt to convert factor analysis output to Item Response Theory type output. (Basically doing what irt.fa already does, but from a separate factor analysis.
Added fa.extend to make factor extension easier to do.
Added an overall goodness of fit tests for fa.extension applied to the extension variables.
Cleaned up factor.stats to make it do fm="minchi" more readily.
Added two trivial helper functions r2d and d2r to convert from correlations to effect sizes and back.
added the von Neumann test of successive differences (mssd and rmssd). This is useful for studying within subject changes in mood.
Changed the help files for test.psych, irt.fa, iqitems to dontrun the examples (cutting down time to compile and test the functions)
fix the output in the example for fa. Fixed by noticing that:
in fa, fm="pa" did not return labels for the factors.
in statsBy, fixed a problem if the by returned NULL values for some group values
Finally fixed bug in describeBy for the case of NULL categories for the by variable. (Reported by Nipa Phojanamongkolkij).
fixed bug in smc for case of missing values of diag(smc)
added an error message to cor.smooth for the case of NA eigen values
Changed the random seed in overview.rnw for the example of omegaSem to 17 (from 42). The previous seed was generating a bad omega solution which in turn caused sem to fail.
fixed the call function in ICLUST to properly report the parameters being called.
Added the rangeCorrection function to correct for range restriction.
Added a new factor method (minchi) to the fa function. This weights residual correlations by the pairwise sample size. The minimum weighted residual is then found. This will give OLS (minres) solutions for the case of equal sample sizes, but slightly better (weighted) fits for unequal sample sizes. Most appropriate for the Massively Missing Completely at Random structure of SAPA data.
Added a measure of the Kaiser-Meyer-Olkin index of factoring adequacy
Updated the overview vignette
Added the ability to do target rotations (TargetQ) in omega.
Improved documentation of fa to explain how oblique SS are found.
A better set of items is used for the iq example (iqitems).
add CVE estimate to fa (see Reise 2010) (had already done this for omega).
Changed ICLUST.sort so that it can sort the loadings of principal (requested by Gudmundur Arnkelsson and then print them with the number of digits requested in the print command.
Modified score.irt so that subjects who miss all items or pass all items are given an estimate based upon the (product) of the difficulty of the items they miss (pass) and then adjusted based upon half the quantile difference from 0 (if they miss all items) and 100 (if they pass all items).
Modified sim.omega to allow for specifying a general factor. This allows for tests of not just the bias in the case of no general factor, but also the ability to detect a general factor. Also modified it to include calls to omegaSem.
modified iclust.diagram so that cluster names can be specified rather than all start with C1 ... Cn (requested by Michael Kubovy)
Slightly improved the documentation for r.test so it is clear which correlation is which.
Modified iclust so that the fit statistic is based upon the off-diagonal elements unless otherwise specified. Use diagonal =TRUE to get fits matching previous analyses.
Modified the print.psych.iclust function to print out the Root Mean Square Residual correlation. (It was previously reported in iclust output, but not printed.)
fixed a bug in score.items, such that if the number of subjects is equal to the number of items, no scores were returned. Reported by Jeromy Anglim.
Modified factor.stats so that rare condition of an exact fit in omega does not lead to an error.
Fixed plot.irt so that item labels are correct
Modified omegaSem to pass n.obs to sem
Fixed df2latex so that it will on text variables as well as numeric variables
Improved statsBy to find within and between group correlations.
Added a data set, withinBetween, to graphically show the issue of multilevel correlations.
Added a simulation function to generate multilevel data (sim.multilevel)
Added a function (irt.responses) to plot responses as a function of the latent score for multiple choice alternatives.
Modified progressBar so that it only shows dots when using the console, but not when using a text file for output (i.e., with Sweave).
fixed plot.poly and plot.irt so that multiple scales for one factor can be plotted.
Slight modification to cor.plot to allow for better control of multiple plots on the same page.
Added the ability to put values into the cor.plot output (numbers=TRUE)
Added a function (lowerUpper) to combine two symmetric matrices and output one as the above diagonal entries, and the other as the below diagonal entries of a square matrix.
Added a function (topBottom) to act like headTail but to do it without ellipses.
In a gradual switch to camelCase, changed the name of headtail to headTail, and describe.by to describeBy. Warning messages are issued about the change.
Added a function, statsBy, to find summary statistics (means, sample sizes, standard deviations) by a grouping variable. Similar to describeBy but somewhat easier to use. Also will report (as an option) the pooled within group correlations.
Added a function to find sample size weighted correlations, cor.wt. Useful for SAPA types of analysis.
Modified superMatrix to take list input so that more than two matrices can be combined at once (suggested by David Condon).
Added errorCircles, a greatly improved version of error.crosses.
Added the option to specify starting communalities in the fa function (requested by David Gosar).
cor.plot was actually switching row and colnames (but since it is mainly used on symmetric matrices, this was not a problem). Fixed.
principal failed when handling missing data in data.frame. Fixed. (reported by Neil Stewart)
(Note that 1.2.3 was not released but was replaced with 1.2.4)
fixed ICLUST so that it converts covar matrices to correlations.
Modified pairs.panels so that the method of correlation can be specified (suggested by Carsten Dormann)
added error values to the structure.diagram/lavaan diagram as an option (suggested by José Luis Cañadas )
Changed the oblique.scores option in fa to be FALSE (use the structure the matrix) following a request by Niels Waller.
Fixed the size of the boxes in dia.rect so that omega.diagram and fa.diagram provide reasonable size boxes for long names (requested by Erich Studerus)
modified output of corr.test so that it announces whether or not it is adjusting the probability values for multiple tests.
added the ability to export dot.plot commands without requiring Rgraphviz. (requested by Erich Studerus). Although just implemented for fa.graph, omega.diagram functions can be exported using fa.graph applied to the schmid$sl object from omega.
Added 3 options to the skew and kurtosis functions, and thus to the describe and describe.by functions as well. (Suggested by Bruce Dudek).
Added the ability to plot general factors (ala omega) in the fa.diagram function.
Added the function lowerCor to find correlations (with pairwise deletion as a default) and print out the lowerMat of the results.
A documentation error in r.test was fixed (reported by Nicholas S. Holtzman). The example of Steiger case B was incorrectly ordered and thus led to incorrect results. Specifying by name (rather than location) is now recommended.
In fa, the rotation was not returned, fixed. (reported by Niels Waller)
predict.psych would not work for a single case (reported by Jonathan Williams). Fixed. Also made the documentation a bit clearer to explain the way the factor scores are standardized.
By adding a global-local parameter to mixed.cor it now seems to work on those data sets with badly distributed polytomous data. (Reported by Eric Smith and David Condon.)
Modified scrub to be a bit easier to use (suggested by Josh Wilt)
Added a check in mixed.cor so that if the data are incorrect, announce it first, rather than waiting until late in the process.
Added a progress indicator to particularly slow functions (e.g. tetrachoric, polychoric, biserial). This will show up when using these functions in other functions (such as irt.fa or mixed.cor).
Modified cluster.cor so that in the case of correlations that are NA, it will not find smcs of the raw correlation matrix. In addition, it will impute scale level correlations based upon the average between scale correlation (previously it was reporting an NA). Requested by David Condon.
Added the helper function lowerMat to print the lower submatrix. Added this as an option to the print.psych function.
Modified mixed.cor so that it will automatically (if desired) find the Pearson, polychoric, and tetrachoric correlation matrices and then reorganize the matrix back to the original form.
Minor modifications to tetrachoric and polychoric to give slightly more helpful error messages in the case of missing data and corrections for continuity.
Minor modifications to summary to give cleaner summary stats for fa and irt.fa analyses.
Added a residual and plot.residuals function to plot either qq plots of the residuals, or a cor.plot of the residuals.
Fixed a few references that were working but doing partial matching in print.psych, test.psych, pairs.panels (e.g, in print.psych digit=digits was changed to digits=digits, test.psych nf was changed to nfactors, col was changed to col.smooth in pairs.panels). This had been flagged by linux checking as a note about "partial matching" for a long time. Finally figured out what it meant.
corrected biplot.psych to properly take the xlim and ylim parameters (reported by Andreas M. Brandmaier)
Correction to tetrachoric (and related functions) to return NA in the case of no cases for x or y. (Reported by David Condon)
Fixed fa so that the score.oblique option actually works. (Reported by Jessica Jaynes).
Modified scrub to be more general.
Added factor structure as an object in the fa function and modified the print.psych.fa function to note that it is the pattern matrix being reported.
Modified the ICLUST output so that the eigenvalues reflect the independent contribution of each cluster.
Added a comment and an option to ICLUST.sort about the order in which clusters are sorted (suggested by Gudmundur Arnkelsson).
Added the rug option to pairs.panels. This will, by default draw a rug underneath the histograms.
Added a covar option to principal to find principal components from covariance matrices as well as correlation matrices.
Cleaned up the objects returned from fa.parallel.poly to more closely match those from fa.parallel.
Added the ability to find and report residual correlations in set.cor. Minor formatting corrections to set.cor and mat.regress.
Fixed a bug in fa such that the oblique.scores option did not work.
fa.parallel.poly was ignoring the fm parameter and was not printing correctly.
Added various rotation options to principal to match those in fa.
Added "components" as an option to factor score to score principal components using just the component loadings. This is most appropriate when not rotating the components.
Added the Harman.5 (socio-demographic) data set to allow comparisons with SAS.
The way component scores were found had changed from "regression" to "tenBerge" without being documented. Changed the documentation and added the method of scoring as an option. Problem reported by Alexander Weiss.
This is basically a rerelease of version 1.1.10 trying to solve a problem with 32 bit machines and one test that fails. Originally call 1.11.01 but that one still failed the tests.
Added ... as an option in multi.hist (following a suggestion by Rui Barradas)
Added the all=FALSE option to test.psych. This turns off one test to avoid a problem in the testing on Solaris 32 bit machines.
factor.scores and factor.stats were failing in tests for Linux operating systems, but not Macs or PCs. Added some protection against complex eigen values.
fixed problem in irt.scores reported by David Condon
fixed irt.fa so it will properly plot for polytomous items
Started to modify all calls to sd and mean to use apply to respond to their deprecation for data.frames and matrices
Changed the release numbering system to reflect year and month of release.
Added a parcels function to form item parcels of size 2 or size 3.
Added a jitter option to the factor.plot function.
Added 3 more factor scoring options to the factor.scores function and changed the default scoring option in fa to be the tenBerge
Added the kaiser function to do kaiser normalization.
Added the df2latex function to make LaTeX tables.
Added plot.circular to draw radar and spider plots
Added a short function to create dummy codes (dummy.code)
Added a score.irt function to find IRT based scores.
Added a table of information by attribute level to be (silently) produced in the plot(irt.fa) function.
Added bifactor and biquartimin as rotation options to fa (based on the Jennrich and Bentler 2011 papers).
Implemented an improvement to print.psych function to use switch. (Following a suggestion by Joshua Wiley ). This does not change the functionality, but makes the code easier to debug and to change.
Added an option to alpha to automatically flip items if they seem to be negatively correlated with total score. (Suggested by Jeremy Miles).
Serious modification to cor.plot to allow for colorRamp data. Makes much prettier correlation plots. (Suggested by David Condon and Joshua Wilt).
Modified sim and sim.simplex to create State Trait Auto Regressive simplex structures (suggested by Deepika Anand).
Added a correct option to irt.fa (passes correct to tetrachoric).
Added the ability for omega to be rerun on the same correlation matrix found by previous omega runs. This speeds up analyses of large matrices but particularly that of tetrachoric/polychoric matrices.
Modified headtail so that it will not choke on mixed numeric and string data
Added a smoothing option to the tetachoric correlation function to smooth the resulting matrix if it is not positive definite.
Added cor.smooth to do a principal components based smoothing for correlations.
Added global=TRUE option to polychoric (and tetrachoric) correlations. If global=FALSE, each correlation is found using the pairwise taus. This seems to match John Fox's polycor function. If global=TRUE, the taus are the datawise values. This will differ from polycor in the case of a great deal of missing data. Somewhat faster and perhaps more appropriate given that we are trying to find a global solution.
Added a correction for multiple tests to the p values in the
corr.test function and added a new function to just report probabilities
Fixed a very subtle bug in score.items for the case of impute="none" where some subjects have no non-missing data. This affected the estimate of the average number of items (num.ob.item) per scale. Reported by Ben Schalet.
Fixed bug in factor.stats that gave much too narrow confidence intervals for RMSEA (Thanks to Rick Zinbarg and Ashley Kendall for pointing out the problem).
revised factor extension to properly handle oblique solutions
fa now reports the call correctly – since introduction of iterations it had reported the wrong call
Fixed mixed.cor so that it can handle several special cases.
Perhaps finally fixed pairs.panels so it leaves us with the original parameter settings
Fixed a problem in print.psych such that mat.regress output would not print. (Reported by Ben Schalet.)
Fixed plot.poly so that it correctly plots the ICC results, particularly for negatively keyed items.
Fixed fa so that the covar option works. (It had been killed when iterations were introduced.)
Modified fa, fa.irt, factor.stats, factor.scores so that they handle bad data more gracefully.
Improvements to set.cor to report the canonical correlations and their chi squares. Also added a plot function to set.cor so that it will plot the eigen values of the canonical correllations.
Added the cushny data set to show the original drug data from student.
Resolved why the denomiator df in set.cor does not precisely match the Cohen example. This seems to be because Systat is doing an unbiased estimator and is thus multiplying by (u+v)/v.
Added a polychoric and tetrachoric option to fa.parallel
Added a keys option to plot.psych (plot.poly and plot.irt) to allow for plotting information functions for selected items while retaining the factor loadings from the complete set. (Requested by Ben Schalet.)
Added a summary of the graphic output for irt.fa that includes the average information for each item as well as where the item is most informative. (Suggested by David Condon.)
Integrated the interactions between fa, fa.poly, irt.fa, fa.diagram, plot, etc. to make commands more consistent.
added a factor analysis for tetrachoric or polychoric correlations (fa.poly) that will work without doing an irt analysis using irt.fa.
Added bootstrapped confidence intervals for loadings and correlations in factor analysis (fa), for reliability estimates in omega (omega) and fa.poly.(Originally requested by Steve Miller and Renee Engeln-Maddox for polychoric correlations, extended to the omega case.)
Modified fa.diagram to include graphic output for fa.extension.
Added Cohen's set correlation to the mat.regress function. (Requested by Amanda Uliaszek.)
Changed the order of the parameters in the mat.regress function to be more consistent with the y~x notation of most regression functions.
Added an oblique.scores option to principal. This means that if oblique rotations are done (not the default) then the scores from principal will be oblique as well.
Factor extension (fa.extension) now can extend an omega as well as a normal factor analysis.
Modified fa so that the factor scores are found for the oblique factors rather than the orthogonal factors. This now produces factor scores that have roughly the same correlations as do the factors. Factor score statistics remain the same. This leads to a different set of factor scores than factanal finds, for it reports the factor scores for the orthogonal factors.
Modified response.frequencies to incorporate a very nice suggestion by Joshua Wiley to identify unique item responses that are not necessarily integers.
Modified fa.extension to correctly find extended oblique loadings.
check added to omega for bizarre case of 1 factor extracted through PC. (This should not be done, for it makes no sense, but nonetheless, it was leading to omega_h > omega_t.) A warning for this condition is now issued and omega_h = omega_t. Reported by Alison Lewis.
Fixed News so that it works as a NEWS.rd file
Added an example to msq
Added a function to perform factor extension (fa.extension)
Added the Dwyer and Gorsuch data sets as examples of fa.extension.
Added lavaan.diagram to the set of diagram functions. Modified to work with lavaan 0.4-7
Added scatter.hist to draw a scatter plot and associated histograms and densities.
Modified score.items so that it reports the number of missing responses for each person for each scale. It had previously reported the total number of missing responses but did not break this down by scale. Requested by Ryne Sherman. Changed the documentation to reflect that missing is calculated.
Modified error.bars.by so that it can not draw lines (lines=FALSE) if desired. (Requested by Ryne Sherman.)
Added to error.bars.Rd to describe how to supply means and standard errrors.
Made various data files available as separate rather than collated. (e.g., Thurstone, Bechtoldt,Holzinger) in the bifactor data set (they remain there as well). (Requested by Michael Friendly.)
Changed VSS.scree and scree to provide a few more options.
Cleaned up circ.tests to be more compatible with rest of psych.
Clean up output from schmid to more match other functions.
Used tools::compactPDF on the vignettes to make them smaller. Also switched to using png instead of pdf for some of the graphics in the vignettes. This reduced the size drastically.
fixed error.bars.by so that it draws the alpha level instead of 2 alpha level error bars.
fixed fa for special case of 3 x 3 matrix with correlations of exactly .5 which would lead to a non-conforming array problem
fixed partial.r to not round output. For rounded output, the print.psych function defaults to 2 digits (reported by David Freedman)
Finally fixed the output of print.psych.fa so that it labels sorted factors independent of cut (Reported by Reinhold Hatzinger )
Eliminated references to polychor in Yule2poly,
fixed read.clipboard.upper so that it will read labels
fixed mat.regress so that it can work with a single predictor
Added mixed.cor to do mixtures of Pearson, polychoric and tetrachoric correlations
Added legends to error.bars.by bar graph
Modified score.items so it will work on correlation matrices as well as raw data.
Improved the plot.irt and plot.poly functions to plot irt output. Called by plot.psych or generic plot call.
Added a trivial little function, bi.bars to do paired histograms
Modified irt.fa to allow for multiple factors
Modified scrub to be a general recoding function
Modified error.bars.by to allow for ... parameters to be passed into axis
Modified diagram so it will dispatch a number of diagram functions (fa, prinicpal, omega, iclust, lavaan).
Added a biplot.psych function to do biplots of fa or principal results.
Added an unbiased estimate of kurtosis in the kurtosi function.
Added a predict function to fa and principal (predict.fa and predict.principal, both called from predict.psych)
Added lavaan.diagram to draw the output from a cfa or sem from the lavaan package.
Added a function to calculate Mardia's test for multivariate normality
Added lengends and improved the examples for pairs.panels.by
Removed the "short" option in score.items
Added the ability to score scales with greatly missing data (for SAPA). Modified the impute option to include = "none"
Added the term KR20 and KR21 to documentation of alpha
Modified pairs.panels so that with the cor option it can draw regressions and give correlations.
Modified the documentation for error.crosses to give a better example.
Added the ability to read rownumbers (variable names?) in first column of read.clipboard.lower and read.clipboard.upper
Tweaked the zlim default for cor.plot to be c(-1,1) rather than c(0,1)
Minor adjustments to Yule and phi.demo to call tetrachoric instead of polycor, thus making the package work with fewer dependencies.
Cleaned up the fa summary output
Fixed dia.curved.arrow to draw one headed curved arrows (as documented)
Fixed a problem with fa.print option sort – Although giving the correct communalities and uniquenesses in the unsorted option, when sorted, it was not sorting h2 or u2.
Added omega total from sem to the omegaSem function.
error.bars.by was ignoring the user supplied ylim. (reported by Helena Chui) Fixed.
Was drawing incorrect sl solution in omega.graph (reported by Mark Difford).fixed.
omega.Rd and schmid.Rd incorrectly referred to fm=ml as fm=mle. Fixed.
Added iclust and vss as an alternative names to ICLUST and vss to be consistent with most naming conventions.
Added the simple utility scrub to do data cleaning.
Added biserial and polyserial to do biserial/polyserial correlations (not to be confused with point biserial)
Modified score.multiple.choice so that it handles missing data without using imputation
Changed the bfi data set to include more subjects and to correctly define the Openness scale.
Modified score.multiple.choice and score.items to report response frequencies for each item (frequencies based upon number of non-missing responses).
Added the burt data set of 11 emotional variables (from Burt, 1915)
Added draw.tetra to draw tetrachoric correlations for fixed tau values (teaching demo)
Added plot.irt and plot.poly to the plot.psych function. These have options to draw the item characteristic function, the item information function, and the test information.
Added irt.fa to do factor analysis of dichotomous or discrete items by first forming a tetrachoric or polychoric correlation matrix. Then convert the results to IRT form.
Added tetrachoric and polychoric functions to find tetrachoric/polychoric correlations as well as item discrimination parameters.
Added omegaSem and omegaFromSem to calculate omega from a confirmatory factor model using John Fox's sem package.
Added (with permission) Bond's Logical Operations Test (BLOT) data set.
sim.irt was not properly handling vectors of item discrimination or guessing parameters. Fixed.
Minor problem in fa.stats that would lead to an error message if chisquare was NaN.
Minor bug in reverse.code for case of 1 variable (why one would want to use 1 the function for one variable is unclear.)
tetrachoric would not apply the correction for continuity when given a table input but would when given normal data. Fixed.
Added a few parameters to fa.parallel to a) set the y label and b) to show or not show a legend. Useful for multipanel displays.
Minor improvements to fa.diagram to allow for input of factor loadings and intercorrelations from other functions.
Added sim.omega to simulate multifactorial hierarchical structures as a test (and demonstration) of omega.
Minor changes to the parameter call list for sim.minor to be compatible with sim.omega.
Fixed a problem in ICLUST where it would sometimes not stop at the right number of clusters for n.clus = 1
Fixed problem in omega.diagram where it would sometimes use the incorrect line type for hierarchical (non-Schmid-Leiman) diagrams.
Fixed bug in schmid where it would not rotate exactly orthogonal factors and find exactly omega = 0.
Added a warning message that oblique rotation does not work, and does a cluster based rotation instead in the case of exactly orthogonal factors. This is just a problem for artificial data sets.
History (Both of these are probably due to a problem in GPArotation which has since been fixed.)
Rearranged the order of parameters for fa so that it is easier to call.
Added various sem stats to fa output (e.g. RMSEA, BIC, rms, crms). These are also provided for omega output.
Add fit statistics to omega for g solution only.
Added a measure of general factor adequacy in Omega. This is just the percent of common variance that is general factor variance for each item. High values and low variances suggest a more meaningful estimate than lower average values or higher variances.
Various clean up of output of ICLUST and iclust.diagram. Added color option for positive and negative loadings.
Changed ICC to work on complete data (i.e., using na.omit) as the default (following a question) by Ross Culloch
Changed pairs.panels to allow for color choice of histogram (requested by Elaine Kuo)
Changed describe to handle numeric but annotated vectors (suggested by Soeren Vogel)
Clarified documentation of cohen.kappa and allowed for using explicitly non-numerical categories (as suggested by Peter Ehlers and Scot McNary)
Increased the usefulness of error.bars to allow for input of x values, y values, and confidence interval. Also added the ability to use 1 standard deviation.
Minor change to print.psych.fa to round before using formatting.
Removed the cut <- min(cut,max(abs(load))/2) following a request by Reinhold Hatzinger to be consisent with factanal
Added an option to omega to allow for analysis of data sets from just the factor loadings and correlation matrices.
Add a scree function to replace VSS.scree
Fixed a bug in pairs.panels for the case of not smoothing but with data ellipses (reported by Hirayuki Sato)
Fixed a serious bug in omega for the case of non-positive manifold items (e.g., personality items). The sign of the group factors was not correctly being set which would lead to incorrect estimates of the multiple R2.
Fixed bug in ICLUST.diagram where it would draw on top of other figures in rare cases.
Fix bug in ICLUST for the path coefficients – in the case of negative loadings, some of these were much too small.
Correctly color the paths in omega.diagram.
Fixed print.psych.fa to correctly showing item numbers for sorted output.
Corrected tetrachor so it would not inappropriately claim that some cells had no data (this did not affect the correlations, just gave an inappropriate warning).
Added within option for error.bars and error.bars.by to allow for within subject error estimates.
Reformatted this file (NEWS) so that the command (e.g.,) news(query=Version > "1.0.85",package="psych") provides nice information.
Added more detail to the overview vignette to discuss how to find omega.
Added the block.random function to block randomize for experimental designs.
Corrected schmid (and thus, omega) so that it finds orthogonal factors before rotating them obliquely. This corrected an error introduced in version 1.0.86 when the default rotation for fa was switched to be oblimin.
correct=FALSE option in tetrachor actually works
error.bars (and error.bars.by) would not properly draw < 3 variables
Added logit, logistic and logistic.grm functions to help teach about IRT.
Added the ability to factor analyze covariance matrices (with resulting loadings of covariances rather than correlations) (suggested by Andreas Moeltner)
Added two new functions to estimate the greatest lower bound (glb.algebraic, written by Andreas Moeltner) and glb.fa.
Added read.clipboard.fwf to allow reading of fixed width files from clipboard. (Originally this required adding a patch to the read.fwf function but that was subsequentally fixed.)
Minor adjustment to ICLUST.cluster so that it will produce fewer clusters (if desired) than it thinks are appropriate. That is, if n.clus is less than the appropriate solution, a warning is issued, but n.clus clusters are found.
Changed the default rotation in fa to be oblimin. (Subsequently discovered that this introduced a bug into schmid and therefore omega.)
Added the glb.communality function to calculate the greatest lower bound by using factor communality estimates (based upon correspondece with Andreas Moeltner, although his algorithm (glb.algebraic, which is now included) which uses the Rcsdp provides slightly different estimates).
added names to communalities in principal (requested by Gumundur Arnkelsson).
Minor change to headtail
Added the sim.rasch function to simulate 1 parameter IRT Rasch models.
Added the data sets bock and income. bock is the source of the lsat data,income is US family income from 2008.
Seriously modified the estimation of Cohen's Kappa in the wkappa function and added a new function: cohen.kappa to calculate kappa from raw data or from similarity matrices. Also calculates the variances following Fleiss, Cohen, and Blashfield, and then find the confidence limits.
Modified the sim.minor function to simulate correlation matrices with nfact major factors and nvar/2 minor factors with larger number of variables.
Fixed print.psych.fa to give correct communalities for sorted factors (had been not sorting the communalities, although it had the factor loadings).
Minor change to comorbidity to clean up output (using print.psych)
Fixed guttman to correctly estimate lambda 4 (error pointed out by Andreas Moeltner).
Modified the output from fa.parallel to print summary statistics. Also added an option to use smcs as communality estimates.
Added the sim.minor function to generate major and minor factor structures.
Modified fa to use first derivatives based upon the model rather than just empirically derived. This results in an impressive speed improvement.
Continued to improve the dia.x functions. Labeled the entire set of functions diagram (for easier search).
Improved documentation for the dia functions to show the power of structure diagrams
Improved structure.diagram to allow graphs to go left to right (default) or bottom to top (which means causal arrows go left to right)
Added the affect data set which includes data from two experimental studies of affect.
Added some features from pairs to allow group coloring in pairs.panels
Modified factor.parallel to allow for other factoring models (including minres)
Modified fa to give eigenvalues of the correlation matrix, and of the common factor space.
Modified pairs.panels so it can plot even if the correlations are NA
Added a legend and improved the output of cor.plot.
Having learned how to use strwidth, have cleaned up the dia.rect, fa.diagram, structure.diagram, and iclust.diagram functions.
Changed the way fa.parallel handles correlation matrices if the number of subjects is not specified. Rather than issuing a warning and proceeding as if n.obs = nvars, it now assumes n.obs =100.
Improved documentation for fa to explain minimum correlation of factor score estimates.
Fixed a problem with printing of the principal (components) results
Fixed bug in the print.psych.vss function that returned the wrong value (although identifying the correct number) for the Velicer MAP criterion. (Reported by Maike Luhmann).
Added Suggests Rgraphviz to DESCRIPTION file to get around CRAN test for 2.10 (as I should have done according to documentation for packages). Removed all tests of functions which require RGraphviz for 2.9.2 because Rgraphviz does not work on SnowLeopard.
Improved the output for mat.regress to allow print() and summary().
Corrected problem in describe.by to properly identify the matrix output
Corrected bug to fa for case of oblique and negatively correlated factors (reported by Erich Studerus).
Minor improvements to the dia.x functions
Fixed the formatting of the news file to follow the appropriate format
Removed the old supplementary manual from the docs folder to make for a cleaner installation
Improved the documentation for the dia.X functions and made some of the calls simpler.
Added the ability to draw "diagrams" to replace the functions that required Rgraphviz. These are path diagrams with rectangles for observed variables and ellipses for latent variables. A set of graphic functions for drawing diagrams includes dia.ellipse, dia.rect, dia.arrow, dia.curve and dia.self. This allows the function omega and ICLUST to draw structural diagrams without using Rgraphviz.
The documentation for these new functions is a bit less than desirable and the dia primitives will be modified slightly for the next release.
added the Schmid data set which includes the original Schmid-Leiman correlation matrix as well as data from Chen, West, and Sousa (2006).
modified geometric.mean and harmonic.mean to allow for missing data.
Cleaned up describe.by so that it can handle matrix output of multiple grouping variables and then print correctly.
Modified factor.stats to allow for statistics even if the matrix is very singular (i.e., N var >> n.obs).
Corrected bug to ICLUST introduced in version 74 that can, in some conditions, produce negative betas.
Fixed an error in printing communalities for rotated factor solutions reported by Tamaki Hattori. Also resolved a problem of extra lines appearing in error.bars. Minor revision to polar.
Finally resolved the problem of not passing the CRAN tests for Mac (even though developed on a Mac, the test package assumes no other packages).
Reinstituted reporting communalities and uniquenesses for principal components and for factor analysis. This feature had been dropped sometime in the past and reported missing by Ista Zahn.
Added two new data sets: Harman.Holzinger and Harman.Burt. The former is used by Peter Bentler as an example of alternative methods for estimating reliability, the latter is a nice example of what happens if the correlation matrix is singular. smc estimates are inappropriate in this case and need to be checked.
Added a weighting option to ICLUST. Weighted beta finds the average between cluster correlation and takes the ratio of the n ^2 * average between cluster correlation to the total variance of the pooled clusters as a measure of the general factor. That is, 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. (Unfortunately, in doing this, a bug was introduced that could produce negative betas. Fixed in Version 78).
Modified the fa function to include maximimum likelihood factor analysis.
Added gls fit option to fa
Added mat.sort to sort correlation matrices by factor loadings.
Revised the overview vignette to make more readable.
Fixed (I hope) the documentation error that has been preventing being loaded onto CRAN for Mac.
Added factor stats to omega
Minor fix to print.psych.omega to correct printing error introduced in .71
Combined principal axis, unweighted least squares (minres) and weighted least squares into one combined function (fa). Eventually the three other functions (factor.pa,factor.wls, and factor.minres) will be phased out. Added more rotation and transformation options (basically by making calls to the GPArotation package).
factor.stats now reports the factor weights matrix for factor scores.
A new factor analysis function factor.wls will give weighted least squares (wls).
Omega will now give an estimate for 1 factor and a message about how there are alternatives for 2 factors.
Fixed a bug in print.psych.omega so that it will now report the item numbers correctly when sorting.
Fixed call in phi.demo so that it checks if the polycor package is installed. This was causing problems in the Mac version on CRAN.
Added factor.stats function to report a variety of statistics for factors or components. In particular, added measures of factorial adequacy. factor.stats is now called by the factor.pa, factor.minres, and principal, so it is easier to adjust all three of these functions.
Added a new rotation option to target.rot – if keys are not specified, it will rotate to a cluster solution.
minor tweak to sim.hierarchical to make it consistent with the Jensen and Weng paper
added to omega to calculate omege_infinity
corrected bug in principal for case of singular matrices (reported by Gudmundur Arnkelsson)
corrected bug in print.psych so that it now sorts by factor loadings correctly
added cor.plot to graphically display correlation structures
modified cosinor to do regression as well as optimization for fits
minor correction to ICLUST to catch non-invertible matrices
Added method="minres" to do minimal residual (ULS) factor analysis to the factor.pa function.
Added the target.rot function to do targeted rotation.
Added the sim.anova function to simulate 3 way ANOVA or linear models.
Minor cleanups of code in other functions
Have included the vignettes in the package
Serious modifications to mat.regress to allow for raw data and covariances to be used.
Added corrections for item overlap to use estimated communalities of items to alpha, cluster.cor, etc.
Added estimates of G6+ (Modified Guttman 6) reliabililty to alpha, score.items
Introduced two package vignettes (overview and psych_for_sem)
minor modification to ICLUST with respect to labeling
addition to factor.congruence to allow for comparing omega solutions with factor solutions
modified describe to allow for descriptions of categorical variables.
Added the function alpha to find Cronbach's alpha and Guttman's Lambda 6 for total scales as well as with single item deleted.
Minor enhancements to the structure.graph and structure.sem functions. Further refinements to the psych_for_sem.pdf tutorial.
Added structure.graph and structure.sem to create graphical output and pre-sem commands for sem.
Modified omega.graph to return sem commands for running the sem function from John Fox.
Added Promax to report factor intercorrelations for promax rotations.
bug fix in VSS and omega to allow for promax rotations.
correction to winsor so that it does not sort the vectors of the input data.frame.
bug fix in principal (for problem of oblique rotations)
continued improvements in the print.psych and summary.psych functions
cleaned up sim.congeneric to reduce the amount of output (now handled through print.psych)
added super.matrix function (to combine two matrices – useful for constructing keys and for doing examples in structure.graph)
Improvements to structure graph.
added the structure.graph function
added circadian functions
renamed some functions to make the index order easier to follow.
Introduced the guttman function to estimate the 6 Guttman (1945) estimates (as well as a few others).
Added print.psych and summary.psych generic functions to clean up the output of many of the functions.
Introduced several new functions:
headtail (combines head and tail functions to produce the first N and last M rows of a matrix or dataframe)
error.bars.by plot means with error bars by a grouping variable
omega revised to allow for negatively keyed items. Detects items that should be reversed.
A relatively stable release.
Passed the tests for R 2.6.2
score.multiple.choice provides scores and item statistics for multiple choice items.
circ.sim modified to be more general item simulations
polar converts two dimensional factor loadings to polar coordinates.
A relatively stable release.
A relatively stable release.
Passed tests for R 2.5
The first release to CRAN. Prior versions had been tested at NU.
Included pairs.panels, describe, read.clipboard, iclust, vss, principal.
The first package version on the personality-project.org repository.
included some of the functions previously in the "useful R" source file.
worked with R 2.1.0