Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

AW: st: suest after probit or oprobit?


From   "Joachim Henkel" <[email protected]>
To   <[email protected]>
Subject   AW: st: suest after probit or oprobit?
Date   Fri, 6 Feb 2009 09:12:47 +0100

Vince, 

   Thanks very much for solving this riddle. Now, with three oprobit
regressions and correlated error terms, would there be a way to increase
efficiency of the estimation (compared to individual oprobit regressions) by
using the covariance structure of the overall model (without switching to
probit)?

-- Joachim 


-----Ursprüngliche Nachricht-----
Von: [email protected]
[mailto:[email protected]] Im Auftrag von Vince Wiggins,
StataCorp
Gesendet: Freitag, 6. Februar 2009 06:50
An: [email protected]
Betreff: Re: st: suest after probit or oprobit?

Joachim Henkel <[email protected]> why -suest- produces the same standard
errors as maximum likelihood estimators with the -vce(robust)- option,

> I am trying to estimate three equations with oprobit followed by
> suest. The problem is, the standard errors from suest are absolutely
> identical to those of the individual robust regressions without
> suest. The same happens when I use suest after probit (which
> definitely should work with suest).  [...]

That is exactly what we expect.  

Both -suest- and -vce(robust)- use the linearization/huber/white/sandwich
estimator of the VCE.

What -suest- adds are the covariance estimates BETWEEN separately fitted
models.  Among other things, those covariances let us test parameter
estimates
across the fitted models.  -suest- performs this magic by utilizing the
scores
from the separately fitted models and the fact that those scores are
computed
on the same observations.  If we examine the Methods and Formulas for
-suest-,
it is clear that the VCE of any one of the fitted models is affected only by
its own scores and that it is not affected by the scores of the other
models.
So, -suest-'s SEs for any model's coefficients are just the SEs of the
sandwich estimator; that is to say, vce(robust).  This is strictly true only
for maximum likelihood estimators, vce(robust) makes a degree of freedom
adjustment for -regress- which -suest- does not apply.

 
-- Vince 
   [email protected]

 
*
*   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/


*
*   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/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index