Margins with gsem, random effects, and nominal dependent variable

[Elizabeth Anderson]

I used gsem to estimate a model with a nominal dependent variable and a random slope.

 gsem (ctype <- c.age##c.age i.qpercbaR M1[zcta]), mlogit

I would like to estimate margins, but I am running into this error:

. margins, at(qpercbaR=(1 5)) predict(outcome(2)) post
invalid outcome() option;
depvar 2 not found
r(198);

. margins, at(totscore_cntr=(-1.5(.5)1.5) qpercbaR=(1 5)) predict(outcome(1)) post
invalid outcome() option;
depvar 1 not found
r(198);

This occurs even as the output contains the categories of 1.ctype and 2.ctype.

Is there a way to estimate the margins for each level of the dependent variable?

[Andrew Musau]

Easy enough to provide a reproducible example here. See FAQ Advice #12 on how best to pose questions. You need to specify the full outcome.

webuse sysdsn1, clear
gsem (insure <- c.age##c.age i.male i.nonwhite i.site), mlogit

*WHAT YO HAVE
margins, at(site=(2 3)) predict(outcome(2)) post

*WHAT YO NEED
margins, at(site=(2 3)) predict(outcome(2.insure)) post

Res.:

. 
. *WHAT YO HAVE

. 
. margins, at(site=(2 3)) predict(outcome(2)) post
invalid outcome() option;
depvar 2 not found
r(198);

. 
. 
. 
. *WHAT YO NEED

. 
. margins, at(site=(2 3)) predict(outcome(2.insure)) post

Predictive margins                                         Number of obs = 615
Model VCE: OIM

Expression: Predicted mean (2.insure), predict(outcome(2.insure))
1._at: site = 2
2._at: site = 3

------------------------------------------------------------------------------
             |            Delta-method
             |     Margin   std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
         _at |
          1  |   .5295559   .0326818    16.20   0.000     .4655007    .5936111
          2  |   .3408466    .033945    10.04   0.000     .2743155    .4073776
------------------------------------------------------------------------------