Hi Stephen,
I have run the model using three different programs. (i) xtprobit; (ii) limdep; (iii) an ado file I wrote for xtprobit. The problem seems to be happening whenever I have a lagged dep variable. xtprobit is not moving from the initial ordinary probit starting values, whereas programs (ii) and (iii) are doing so.
I have also changed the no of quad points in xtprobit but it still wont move. I will play around with the other things as you suggested.
Thanks very much.
best
wiji
=================================
Professor Wiji Arulampalam,
Department of Economics,
University of Warwick,
Coventry,
CV4 7AL,
UK.
Tel: +44 (24) 7652 3471
Sec. Tel: +44 (24) 7652 3202
Fax: +44 (24) 7652 3032
email: [email protected]
http://www.warwick.ac.uk/Economics/arulampalam/
RES2003: http://www.warwick.ac.uk/res2003/
>>> "Stephen P. Jenkins" <[email protected]> 09/16/02 02:47PM >>>
Wiji,
> Thanks very much for your suggestion. I don't think that the problem
> is to do with the number of quadrature points.
... but I still think worth checking this one out -- to eliminate all
possible sources of difference. I looked at Limdep 7.0 manual, pp.
426-31. It says it used DFP algorithm and BHHH for asym cov matrix, and
8 point quadrature. Stata uses 12 points by default (and uses a
different, modified Newton, maximiser -- though I suspect that not the
issue). Try the commands:
xtprobit sue lague if wave>1, i(ind1) /* NB "re" is default */
quadchk /* checks out sensitivity to using 8 or 16 point instead */
> The coefficients being
> different by a very small number does not matter. But the problem I
> have with Stata xtprobit is that it gives me a zero 'rho' with no std
> error. Limdep always gives me an estimate of 'rho' with a well
> defined std error. I have a general program which I am going to
> change now and check the results against stata's results. I will let
> you know the outcome.
Another thing that could be going on is different convergence criteria.
Have a look at -maximize-. That tells you of a number of generic
options that you could add to the -xtprobit- command to do things like
get a more detailed trace of what happens at each step, and to change
default tolerances for convergence, etc. NB 'difficult' option too.
When no std error is reported by a Stata program, it usually means that
it couldn't invert the Hessian -- singular (as far as numerically
relevant) .
It seems that in cases where your rho is very close to zero, the 2
programs are taking different courses of action.
> In most of the runs I get the result that there is no unobserved
> heterogeneity as the coefficient estimates are identical to ordinary
> probit when I use xtprobit. xtprobit also does not give me a std
> error for the rho. But I have a mle for RE probit in stata and I have
> just run that and find that my results are exactly the same as
> Limdep's.
Sorry, I don't understand your last sentence. Are you saying that you
have some results from Stata that are now the same as Limdep's, whereas
they differed before?
best wishes
Stephen
----------------------
Professor Stephen P. Jenkins <[email protected]>
Institute for Social and Economic Research (ISER)
University of Essex, Colchester, CO4 3SQ, UK
Tel: +44 (0)1206 873374. Fax: +44 (0)1206 873151.
http://www.iser.essex.ac.uk
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
