Kf. Widaman, COMMON FACTOR-ANALYSIS VERSUS PRINCIPAL COMPONENT ANALYSIS - DIFFERENTIAL BIAS IN REPRESENTING MODEL PARAMETERS, Multivariate behavioral research, 28(3), 1993, pp. 263-311
Citations number
47
Categorie Soggetti
Social Sciences, Mathematical Methods","Psychologym Experimental","Statistic & Probability","Mathematical, Methods, Social Sciences
The aim of the present article was to reconsider several conclusions b
y Velicer and Jackson (1990a) in their review of issues that arise whe
n comparing common factor analysis and principal component analysis. S
pecifically, the three conclusions by Velicer and Jackson that are con
sidered in the present article are: (a) that common factor and princip
al component solutions are similar, (b) that differences between commo
n factor and principal component solutions appear only when too many d
imensions are extracted, and (c) that common factor and principal comp
onent parameters are equally generalizable. In contrast, Snook and Gor
such (1989) argued recently that principal component analysis and comm
on factor analysis led to different, dissimilar estimates of pattern l
oadings, terming the principal component loadings biased and the commo
n factor loadings unbiased. In the present article, after replicating
the Snook and Gorsuch results, an extension demonstrated that the diff
erence between common factor and principal component pattern loadings
is inversely related to the number of indicators per factor, not to th
e total number of observed variables in the analysis, countering claim
s by both Snook and Gorsuch and Velicer and Jackson. Considering the m
ore general case of oblique factors, one concomitant of overrepresenta
tion of pattern loadings is an underrepresentation of intercorrelation
s among dimensions represented by principal component analysis, wherea
s comparable values obtained using factor analysis are accurate. Diffe
rences in parameters deriving from principal component analysis and co
mmon factor analysis were explored in relation to several additional a
spects of population data, such as variation in the level of communali
ty of variables on a given factor and the moving of a variable from on
e battery of measures to another. The results suggest that principal c
omponent analysis should not be used if a researcher wishes to obtain
parameters reflecting latent constructs or factors.