Dynamic Random Effects Generalised Ordered Probit

[David Clark]

Using Stata 13, I am utilising -regoprob2- in order to try and estimate a dynamic random effects generalised ordered probit model using household panel data, over three waves.

The variable which encaptulates the dynamic element is generated thus:

g lvar = var[_n-1] if id==id[_n-1]

When using the aformentioned variable using -regoprob2-, using the following syntax:

regoprob2 var lvar $IndepVars, i(id)

where $IndepVars is a macro of explanatory variables and id is the panel identifier.

Having done so, Stata produces the following:

Fitting constant-only model:

Fitting full model:

Iteration 0:   log likelihood = -37689.925  
Iteration 1:   log likelihood = -35263.161  
no observations
r(2000);

This I find puzzling, as the same model estimated using -oprobit- and -gologit2- using the option link(probit) estimates without such issues with 34883 observations.

Just to check I have tried the same with the simplist dynamic model:

regoprob2 var lvar, i(id)

and the error is still the same yet, using -oprobit- and -gologit2-, the estimation executes without error.

Can anyone advise me on what may be going wrong here?

If I have been sparse with any information on the issue, please do let me know and I will provide elucidation.