Re: st: Random effects probit

["Guillaume Frechette" <gurst1@hotmail.com>]

Joseph Coveney <jcoveney@bigplanet.com> wrote:
> Wiji Arulampalam wrote that he was having difficulty getting results from 
> -xtprobit-
> to agree with those produced by Limdep.  Weihua Guan identified the problem 
> as
> having to do with the inability of Hermite quadrature to give trustworthy 
> estimates of
> the integral, and this would apply to the results of both -xtprobit- and 
> those of
> Limdep, since they both use similar algorithms.
Wait, that's not exactly right. The results from Limdep are very stable. If 
you look at the coefficient estimates they vary by very little accross 12 
and 20 quad points and in both cases Limdep reports standard errors. As 
pointed out to me in a private email by Stephen Jenkins, Limdep and Stata do 
not use the same algorithm for the optimization. The advantage of the one 
used by Limdep is that you are always garanteed a pos. semi-def. estimate of 
the Hessian, which I believe is not true of the method used by Stata. I 
believe this explains the missing std. errors (I wish somebody at Stata 
would check if that is the case or if I am wrong). This being said, I am 
sure there are advantages to the approach used by Stata, but it is my 
experience that these problems are relatively frequent. Maybe Stata should 
consider adding a DFP or BHHH option to their xtprobit and xtlogit commands.

g

_________________________________________________________________
Send and receive Hotmail on your mobile device: http://mobile.msn.com

*
*   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/