Confirmatory factor analysis with high-frequency longitudinal data

[Clyde Schechter]

Does anybody know of a Stata program that can do confirmatory factor analysis in longitudinal data of the large T small N type? If it matters, I'm positing a factor structure with only 2 or 3 latent factors, but about a dozen indicators for each.

If not a Stata program, a reference on the overall approach? I've been looking at confirmatory factor analysis for a data set of this type, and so far have found only methods suitable for large N small T. (Well, I found one approach for large T small N but even the author warned that it probably would be computationally infeasible when the number of indicators and time points are both large, as here.)

[Joseph Coveney]

Don't know of any Stata program off-hand (I'm not certain whether Mplus would help here, either), but depending upon what question you're trying to answer maybe you could condense or sample the dataset for something more workable.

For example, if you're interested in trends over time for the latent factor means, then how about forming a simple sumscore for each set of indicator variables (maybe scale them first if needed) as a surrogate for the respective latent factor, and then track changes in the two or three sumscores longitudinally.

Or if you're interested in whether the factor loadings are stable, then maybe take a random subset of the timepoints for a more manageably sized analysis dataset--it could be enough to give an indication of whether the factor structure undergoes radical perturbations or not.

[Clyde Schechter]

Thanks, Joseph. I'm interested in both of those questions. If there's no more direct solution, I think I will go the way(s) you suggest.