st: RE: Marginal Effects after Biprobit with Reliable Standard Errors?

["Nick Cox" <n.j.cox@durham.ac.uk>]

I have two banal suggestions. They may or may not apply to the specifics
not given here. 

1. Sometimes the model being fitted is just too complicated. Omitting
some predictors may improve the situation with standard errors. 

2. Sometimes this is Stata's way of telling you that the model is
straining very hard to fit the data. Rethinking the functional form may
be in order. In other words, fixing the poorly determined standard
errors is a lesser issue than getting the right model in the first
place. 

As I said, these are banal! 

Nick 
n.j.cox@durham.ac.uk 

Claudia Berg

I am trying to obtain marginal effects with reliable standard errors
after a
Seemingly Unrelated Biprobit model.  I have tried using the commands
"mfx
compute, predict()" but Stata warns that it is "unsuitable" for biprobit
and
imposes the option "nose".  I know that the option "force" can be used
to
obtain standard errors but with no guarrantee that they are reliable.  I
have refered to the earlier discussion on statalist found at
http://www.stata.com/support/faqs/stat/mfx_nose.html which says that:
"if
diag(vce) shows a large relative difference (say, bigger than 10^-2 for
example) the standard errors given by using force will probably be
wrong..."  I checked the "diagnose(vce)" option for my data and found
that
for my data the relative difference was about 0.029.

Can anyone suggest a way to get reliable standard errors?  If I am
forced to
use "force", how unreliable would the standard errors be?


*
*   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/