Stephen Brand
www.statisticsdoc.com
Andrew,
The best way to test for factorial variance and invariance between groups is to use AMOS, or another confirmatory factor analysis package, to carry out a multi-sample confirmatory factor analysis. Briefly, you would compare a model in which the factor loadings and inter-factor correlations were constrained to be equal across the two samples with models in which some or all of the factor loadings are allowed to differ between samples. A key question would be whether the fit of the model improves when factor loadings are allowed to be unequal. If you find that the fit improves with unequal loadings, then examine the magnitude of the difference in loadings to see whether there is some really substantive difference. A word of caution - you mention that each sample has an n = 110 - how many items are you factoring?
There is also a "rough and ready" method for comparing factor structures from the days before high-speed PCs: Tucker's Coefficient of Congruence. This methods does not take the place of mutliple-sample CFA, but it is still a useful tool to have available. The essence of the Coefficient of Congruence is that you correlate the item loadings between Factor1 in sample1 with the loadings for Factor1 in sample2 (treating the items as though they were cases). Do the same for Factor 1 in Sample1 and Factor2 in Sample2, etc. etc. The resulting matrix of congruence coefficients will look like an identity matrix when the factors are perfectly congruent.
HTH,
Stephen Brand
---- Andrew Walsh <
[hidden email]> wrote:
> hello all,
> i have run two separate exploratory factor analyses (principle component
> analysis with a promax rotation) exploring the factor structure of a
> pain intensity and interference questionnaire in a older and younger
> sample (each n=110). on inspection, the factor structure patterns are
> similar, but the loading appear to differ between the two groups. is
> there a way to test for differences in factor loading in spss?
> thanks very much,
> ~andrew
--
For personalized and experienced consulting in statistics and research design, visit www.statisticsdoc.com
Locked
