Performing posthoc analyses using factors from factor analysis

[Laura Palm]

Hi all

This may be a basic question and I think I know of the solution in Mplus, but wondered if there was a way to 'save' the factors that are generated from a factor analysis in order to perform posthoc analysis. For example, we're looking at how the factors are associated with other exposures in subgroups of the sample using regressions.

Thanks,
Laura

[Joseph Coveney]

. . . a way to 'save' the factors that are generated from a factor analysis in order to perform posthoc analysis. For example, . . .how the factors are associated with other exposures in subgroups of the sample using regressions.

The factor loadings (and uniqueness etc.) are ereturned in matrices, e(L) (and e(Psi) etc.); you can save those matrices as variables in the dataset or another frame using svmat.

help factor

and scroll down to the section Stored results. and

help svmat

for more details.

[Erik Ruzek]

To follow-up on Joseph's excellent advice, you may also be interested in getting factor scores. To do that, check the help file for factor postestimation and in particular the predict option.

Rather than predicting an inherently unknown and uncertain variable, you can (as you are likely aware) instead build the measurement model into a full structural equation model (sem in Stata) with regression paths into and out of the latent variables. This has the advantage of preserving the most useful information from the measurement model (including the uncertainty in any individual's given value), potentially helping you to have more precision in estimating the regression paths.