Factor Analysis Question

[Paul Mills]

In factor analysis the factors are the result of analysis. So STATA figures out that Variables A, C and E for example are related, and assigns factor weights.

Is it possible to do an analysis where I can tell STATA which variables I want to "force" to include in a factor and have it do the analysis (give me the eigenvalues, factor weights, allow me to do rotations). I have theory that I want to test, and it seems like it would be convenient if I could construct factors rather than derive them. I hear you can do this in SAS, but wanted to see if anyone has any ideas on doing such an analysis and whether STATA can support it.

Thanks.

[Nick Cox]

Stata's sem suite includes confirmatory factor analysis as from Stata 12.

[Dick Campbell]

SEM will do this, but in a confirmatory factor analysis (CFA) the notion of rotations is moot. You can specify if you want a set of factors to be orthogonal or not, but if you allow the factors to correlate CFA estimates the correlations and "factor loadings" (actually regression coefficients linking latent variables (factors) and observables) in such a way as to maximize the fit of the estimated associations among variables as implied by the model to the observed data.