----- Original Message -----
From: "Thomas Schmeling" <[email protected]>
To: <[email protected]>
Sent: Sunday, August 17, 2003 4:21 PM
Subject: st: LR test and robust SE under logit
> I am estimating a logit model using robust standard errors (via clustering
> on a subject ID variable). I want to compare a restricted model with a
> full model. STATA tells me that LR test cannot be used with robust SEs,
> although I can get the LR test results by using the 'force' option.
>
> Can anyone point me to a discussion of why LRtest is invalid under robust
> SE? Does anyone have advice on how to proceed where I need to use the
> Robust SE to estimate the model correctly and want to compare the full and
> restricted models?
>
> Tom
>
> --------------------------
> Dept. of Political Science
> Rhode Island College
> 600 Mount Pleasant Ave. Providence, RI 02908
> Phone: (401) 456-8722
>
Bryan Sayer <[email protected]>, on July 24th, replied to a similar question:
"Liklihood ratio tests are not appropriate with robust variance estimates
because we do not know what the sampling distribution of the liklihood is
under the assumptions made with robust. There is an excellent FAQ on the
topic at the Stata web site."
I believe the appropriate FAQ is:
"Why should I not do a likelihood-ratio test after an ML estimation (e.g.,
logit, probit) with clustering or pweights?"
(http://www.stata.com/support/faqs/stat/lrtest.html)
Hope this helps,
Scott
*
* 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/
