st: GSEM Combined, Nonrecusive Structural Model

[Matt Clement <matthewtclement2@yahoo.com>]

Hello,


Using 'gsem' in Stata/MP (13.0), I tried to estimate a combined, nonrecursive structural model with two generalized response variables: 'y1' (gaussian with identity-link) and 'y2' (poisson with log-lnk).


Here is the relevant command line:

gsem (y2 -> y1, family(gaussian) link(identity)) (y1 -> y2,
family(poisson) link(log)) (x2 -> y2, family(poisson) link(log)) (x1
-> y2, family(poisson) link(log)) (x1 -> y1, family(gaussian)
link(identity)), vce(robust) cov( e.y1*e.y2) nocapslatent

The error I get is this:

"invalid path specification; a nonrecursive system among the variables 'y2' and 'y1' is not allowed"

In the model builder, when I try to create a covariance link between 'y2' and 'y1', I get the following error:


"For generalized responses,
only the errors of Gaussian/Identity responses covary."

(Note: When I change the family/link of 'y2' to Gaussian/Identity, there is no problem estimating the model; but 'y2' is Poisson/log.)

What am I doing wrong? Any suggestions on how to estimate a combined, nonrecursive structural model in Stata? I appreciate your responses.

Sincerely,
Matt Clement
PhD Candidate
Department of Sociology
University of Oregon

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/