Multiivariate probit using panel data and -cmp

[Adama lo]

Dear all,
I hope you are fine.
Is it possible to estimate the random effect multivariate probit model with cmp.
I saw the univariate case in the cmp help, but I don't understand the use of the suffix # to estimate random effect.
Why, in this example, was "south" chosen instead of another variable?

webuse union
cmp (union = age grade not_smsa south##c.year || idcode, ind($cmp_probit) qui

Thank you for your help!

[David Roodman]

Yes.
The # suffix is not for random effects. In this command line, the random effect is at the idcode level.
Note that by default in multi-equation models, both the random effects and the observation-level errors are correlated across equations. This creates a lot of correlation parameters, which it can be hard in practice to identify. See the discussion under the covariance() option in the help file.

[Adama lo]

Thank you for your feedback and orientation,
the random effect model doesn't seem to work when the number of dependent variables is high (6). Here's the error message I get

Fitting full model.
Random effects/coefficients modeled with sparse-grid quadrature.
Precision equivalent to that of one-dimensional quadrature with 12 integration points.
cmp_model::cmp_init(): 3900 unable to allocate real <tmp>[30013,6]
<istmt>: - function returned error
Mata run-time error
r(3900);

end of do-file

r(3900);

On the other hand, "Mixed process regression" works well.
Does this model take into account the Panel's structure?

[David Roodman]

That's an out of memory error. You can try reducing the number of quadrature points from 12 using the intpoint() option.

[Adama lo]

Ok Thank you very much