holzinger.swineford | R Documentation |
A classic data set in psychometrics is that from Holzinger and Swineford (1939). A 4 and 5 factor solution to 24 of these variables problem is presented by Harman (1976), and 9 of these are used by the lavaan package. The two data sets were supplied by Keith Widaman.
data(holzinger.swineford)
data(holzinger.raw)
data(holzinger.dictionary)
A data frame with 301 observations on the following 33 variables. Longer descriptions taken from Thompson, (1998).
case
a numeric vector
school
School Pasteur or Grant-White
grade
Grade (7 or 8)
female
male = 1, female = 2
ageyr
age in years
mo
months over year
agemo
Age in months
t01_visperc
Visual perception test from Spearman VPT Part I
t02_cubes
Cubes, Simplification of Brighams Spatial Relations Test
t03_frmbord
Paper formboard-Shapes that can be combined to form a target
t04_lozenges
Lozenges from Thorndike-Shapes flipped over then identify target
t05_geninfo
General Information Verbal Test
t06_paracomp
Paragraph Comprehension Test
t07_sentcomp
Sentence Completion Test
t08_wordclas
Word clasification-Which word not belong in set
t09_wordmean
Word Meaning Test
t10_addition
Speeded addition test
t11_code
Speeded codetest-Transform shapes into alpha with code
t12_countdot
Speeded counting of dots in shap
t13_sccaps
Speeded discrimation of straight and curved caps
t14_wordrecg
Memory of Target Words
t15_numbrecg
Memory of Target Numbers
t16_figrrecg
Memory of Target Shapes
t17_objnumb
Memory of object-Number association targets
t18_numbfig
Memory of number-Object association targets
t19_figword
Memory of figure-Word association target
t20_deduction
Deductive Math Ability
t21_numbpuzz
Math number puzzles
t22_probreas
Math word problem reasoning
t23_series
Completion of a Math Number Series
t24_woody
Woody-McCall mixed math fundamentals test
t25_frmbord2
Revision of t3-Paper form board
t26_flags
Flags-possible substitute for t4 lozenges
The following commentary was provided by Keith Widaman:
“The Holzinger and Swineford (1939) data have been used as a model data set by many investigators. For example, Harman (1976) used the “24 Psychological Variables" example prominently in his authoritative text on multiple factor analysis, and the data presented under this rubric consisted of 24 of the variables from the Grant-White school (N = 145). Meredith (1964a, 1964b) used several variables from the Holzinger and Swineford study in his work on factorial invariance under selection. Joreskog (1971) based his work on multiple-group confirmatory factor analysis using the Holzinger and Swineford data, subsetting the data into four groups.
Rosseel, who developed the ‘lavaan’ package for R, included 9 of the manifest variables from Holzinger and Swineford (1939) as a “resident" data set when one downloads the ‘lavaan’ package. Several background variables are included in this “resident" data set in addition to 9 of the psychological tests (which are named x1 – x9 in the data set). When analyzing these data, I found the distributions of the variables (means, SDs) did not match the sample statistics from the original article. For example, in the “resident" data set in ‘lavaan’, scores on all manifest variables ranged between 0 and 10, sample means varied between 3 and 6, and sample SDs varied between 1.0 and 1.5. In the original data set, scores ranges were rather different across tests, with some variables having scores that ranged between 0 and 20, but other manifest variables having scores ranging from 50 to over 300 – with obvious attendant differences in sample means and SDs.
After a bit of snooping (i.e., data analysis), I discovered that the 9 variables in the “resident" data set in ‘lavaan’ had been rescored through ratio transformations. The ratio transformations involved dividing the raw score for each person on a given test by a particular constant for that test that transformed scores on the test to have the desired range.
I decided to perform transformations of all 26 variables so that two data sets could be available to interested researchers:"
holzinger.raw are the raws scores on all variables from Holzinger & Swineford (1939)
holzinger.swineford are rescaled scores on all variables from Holzinger & Swineford.
holzinger.dictionary is a list of the variable names in short and long form.
... Widaman continues:
“As several persons have noted, Harman (1976) used data only from the Grant-White school (N = 145) for his 24 Psychological Variables data set. In doing so, Harman replaced t03_frmbord and t04_lozenges with t25_frmbord2 and t26_flags, because the latter two tests were experimental tests that were designed to be more appropriate for this age level. This substitution is fine, as long as one analyzes data from only the Grant- White school. If one wishes to perform multiple-group analyses and uses school as a grouping variable (as Meredith, 1964a, 1964b, and Joreskog, 1971, did), then tests 25 and 26 should not be used."
“As have others, Gorsuch (1983) mentioned that analyses based on the raw data reported by Holzinger and Swineford (1939) will not produce statistics (means, SDs, correlations) that match precisely the values reported by Holzinger and Swineford or Harman (1976). Following Gorsuch, I have assumed that the raw data are correct. Applying factor analytic techniques to the raw data from the Grant-White school and to the summary data reported by Harman (1976) will produce slightly different results, but results that differ in only minor, unimportant details."
These data are interesting not just for the historical completeness of having the original data, but also as an example of suppressor variables. Age and grade are positively correlated, and scores are higher in the 8th grade than in the 7th grade. But age (particularly in months) is negatively correlated with many of the cognitive tasks, and when grade and age are both entered into regression, this negative correlation is enhanced. That is, although increasing grade increases cognitive performance, younger children in both grades do better than the older children.
As discussed by Widaman, the descriptive values reported in Harman (1967) (p 124) do not quite match the descriptive statistics in holzinger.raw
. Further note that the correlation matrix and factor loadings are trivially different from the Harman.24 factor loadings in the GPA rotation package.
The purpose behind presenting both the raw and transformed data is to show that the fit statistics from factor analysis are identical for these two data sets.
The variables v1 ... v9 in the lavaan package correspond to tests 1, 2, 4, 6, 7, 9, 10, 12 and 13.
Keith Widaman (2019, personal communication). Original data from Holzinger and Swineford (1939).
Gorsuch, R. L. (1983). Factor analysis (2nd ed.). Hillsdale, NJ: Erlbaum.
Harman, Harry Horace (1967), Modern factor analysis. Chicago, University of Chicago Press.
Holzinger, K. J., & Swineford, F. (1939). A study in factor analysis: The stability of a bi-factor solution. Supplementary Educational Monographs, no. 48. Chicago: University of Chicago, Department of Education.
Joreskog, K. G. (1971). Simultaneous factor analysis in several populations. Psychometrika, 36, 409-426.
Meredith, W. (1964a). Notes on factorial invariance. Psychometrika, 29, 177-185.
Meredith, W. (1964b). Rotation to achieve factorial invariance. Psychometrika, 29, 177-206.
Meredith, W. (1977). On weighted Procrustes and hyperplane fitting in factor analytic rotation. Psychometrika, 42, 491-522.
Thompson, Bruce. Five Methodology Errors in Educational Research:The Pantheon of Statistical Significance and Other Faux Pas. Paper presented at the Annual Meeting of the American Educational Research Association(San Diego, CA, April 13-17,1998)
psych::Holzinger
data(holzinger.raw)
psych::describe(holzinger.raw)
data(holzinger.dictionary)
holzinger.dictionary #to see the longer names for these data (taken from Thompson)
#Compare these to the lavaan correlation matrix
psych::lowerCor(holzinger.swineford[ 7+ c(1, 2, 4, 6, 7, 9, 10, 12, 13)])
psych::lmCor(t01_visperc + t05_geninfo + t08_wordclas ~ grade + agemo,data = holzinger.raw)
psych::lmCor( t06_paracomp ~ grade + agemo, data=holzinger.swineford)
psych::mediate(t06_paracomp ~ grade + (agemo),data = holzinger.raw,std=TRUE)
#show the omega structure of the 24 variables
om4 <- psych::omega(holzinger.swineford[8:31],4)
psych::omega.diagram(om4,sl=FALSE,main="26 variables from Holzinger-Swineford")
#these data also show an interesting suppression effect
psych::lowerCor(holzinger.swineford[c(3,7,12:14)])
psych::lmCor( t06_paracomp ~ grade + agemo, data=holzinger.swineford)
#or show as a mediation effect
mod <- psych::mediate(t06_paracomp ~ grade + (agemo),data = holzinger.raw,std=TRUE,n.iter=50)
summary(mod)
#now, show a plot of these effets
plot(t07_sentcomp ~ agemo, col=c("red","blue")[holzinger.swineford$grade -6],
pch=26-holzinger.swineford$grade,data=holzinger.swineford,
ylab="Sentence Comprehension",xlab="Age in Months",
main="Sentence Comprehension varies by age and grade")
#we use lmCor to figure out the lines
#note that we need to not plot the default graph
by(holzinger.swineford,holzinger.swineford$grade -6,function(x) abline(
psych::lmCor(t07_sentcomp ~ agemo, data=x, std=FALSE, plot=FALSE),
lty=c("dashed","solid")[x$grade-6]))
text(190,3.3,"grade = 8")
text(190,2,"grade = 7")
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.