st: Using cmp with many models

["Tysinger, Bryan" <btysinge@healthpolicy.usc.edu>]

Hi there!  My goal is to use cmp to estimate the covariance of the error terms for about 20 models.  I've into some issues.

The first issue I'm dealing with is a variable with two levels of selection.  I have a model for work status, a model for having a pension, and a model for pension wealth.  In my data, pension wealth is only observed if a person has a pension and pension status is only observed if a person is working.  Consequently, I have something like this in mind:

#delimit ;
cmp (pension_level = `vars1') (any_pension = `vars2') (working = `vars3') 
, indicators( any_pension*$cmp_cont working*$cmp_probit $cmp_probit )  
  ghkdraws(10) tech(dfp) ;
#delimit cr

I can't get this to estimate.  I get the following error message:

Fitting constants-only model for LR test of overall model fit.
         dPhi_dpE_dSig():  3200  conformability error [11]
      cmp_lnL_censored():     -  function returned error [80]
               cmp_lnL():     -  function returned error [433]
                 <istmt>:     -  function returned error
Mata run-time error
r(3200);

Is this possible to estimate using cmp?  What should I try to get this to work?

Ultimately, I have about 17 other "seemingly unrelated" models I would like to include in this estimation (health outcomes and other economic outcomes).  I think cmp is the appropriate routine to use, but I haven't had much luck getting more than a half dozen models to estimate simultaneously and converge to a solution, even when I skip the models with selection issues.  Any tips?

Thanks in advance,
Bryan Tysinger


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