st: Some questions on probit for panel data

[Andrea Molinari <anmolinari@gmail.com>]

Dear Statalisters,

I estimated a probit with a panel of 10 countries and 14 years with
the -pa- option of -xtprobit- (the command I=B4m using is -xtprobit
depvar explvars, pa corr(ar1) robust force-). I have three questions:

1. When I then ran the post-estimation results (to have the marginal
effects, command -mfx-), I get different P-values and some of the
variables
become non-significant (when they were significant in the original
results). Does anyone know what this might be indicating? I found it
strange that an
coefficient changes its significance when the marginal effects are
calculated, but perhaps I=B4m missing some theoretical argument
here...

2. I also had problems with the convergence in some of the
specifications used, are there any ways of changing the iteration
method when estimating a
probit model?

3. Finally, when it comes to probit-panel, what model (and what
command) would you suggest me to estimate, especially if I want to
capture the
individual effects in the error term?

Any help to either of the three questions would be much appreciated!

Best,
Andrea

--
Andrea Molinari, PhD
PhD in Economics
University of Sussex
*
*   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/