Re: Longitudinal SEM

[Terrence Jorgensen]

I think you need to worry about empirical underidentification.  The identification of a 2-indicator factor is tenuous, depending on substantial between-factor correlation(s).  I'm not sure these can be disaggregated from the residual correlations one would normal include in longitudinal CFAs (i.e., correlated specifics), so the tenuous empirical identification could be threatened.

http://ww.w.statpower.net/Steiger%20Biblio/Steiger02.PDF

If your model is empirical identified, negative variance estimates are not necessarily a sign of anything other than sampling error.  You can test the feasibility of that null hypothesis:

https://doi.org/10.1177/0049124112442138

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

http://www.uva.nl/profile/t.d.jorgensen