O. Vassend et A. Skrondal, The problem of structural indeterminacy in multidimensional symptom reportinstruments. The case of SCL-90-R, BEHAV RES T, 37(7), 1999, pp. 685-701
The factor structure of SCL-90-R items and scales was analyzed using both e
xploratory factor analysis (EFA) and confirmatory factor analysis (CFA). Re
sults of CFA studies at the item-level failed to support the original nine-
factor model, as well as several alternative models and EFA suggested very
different dimensionality, depending on which criteria were used. Analyses a
t the scale-level (i.e. the nine original symptom dimensions) suggested tha
t a one- or two-factor model was satisfactory according to descriptive good
ness of fit criteria. However, using the likelihood ratio test, specificati
on of four factors was necessary to avoid rejection. According to the likel
ihood ratio test in a multi-group analysis, a lack of factorial invariance
across gender was indicated. Moreover, the factorial structure of the instr
ument was clearly different across levels of negative affectivity (NA); the
dimensionality was substantially higher in the low-NA group as compared to
the high-NA group. It is concluded that we are confronted with a profound
structural indeterminacy problem and that factor analytic methods and model
acceptance criteria alone are insufficient to solve this problem. The inde
terminacy problem can be accounted for, at least in part, by the complex lo
gical-semantical structure of SCL-90-R items and scales and the role of the
NA trait as a structure generating factor. (C) 1999 Elsevier Science Ltd.
All rights reserved.