gsem: a two-equation multilevel model with two random effects

[Secil Danakol]

Dear all,

I am running a two-equation multilevel model with random intercepts at the nation and region levels. Furthermore, I use STATA's gsem command. Regions are nested within nations. The dependent variables are continuous, and they share the same right-hand-side variables.

My code is as follows and M1[nation] and M2[nation>region] refer to random intercepts at the nation and region levels.

 gsem (dv1 <- var1 var2 var3 var4 M1[nation] M2[nation>region])
(dv2 <- var1 var2 var3 var4 M1[nation] M2[nation>region])

My question is that whether the use of identical random effects variables M1[nation] and M2[nation>region] is correct in both equations? Or should I use M3[nation] and M4[nation>region] in the second equation? When I use STATA's mixed command, we simply add the same variables in the form of | nation: | region: Therefore, I suspect gsem should follow a similar logic but could not be sure. Will you please help me?

I hope my question is clear, and highly appreciate any suggestions.

Many thanks in advance

Best wishes,
Secil

[Phil Bromiley]

I suspect the question is whether you want to allow different random effects for the two different dependent variables. Since the issue is largely a question of controlling for omitted panel level variables, if both equations are likely to have the same omitted panel level variables (assumed to be uncorrelated with the included variables) than the same random effects are probably okay. I might be tempted to try it both ways – if you get the same results you have a more robust set of results.