Re: st: xtlogit with fe for rare events

[Austin Nichols <austinnichols@gmail.com>]

Marc Peters <marcpeters32c@gmail.com>:
That is appropriate--if you think there is unobserved heterogeneity
that calls for -xtlogit, fe- or -clogit- then you need to drop the
cases with no variation in outcomes. You have a small sample (of cases
with variation in outcomes)--better to accept that fact. You can
cluster SE in a pooled -logit- as well, but you will get biased point
estimates from -logit- in the case that the assumptions for -xtlogit,
fe- are met.

On Fri, Jun 29, 2012 at 9:07 AM, Marc Peters <marcpeters32c@gmail.com> wrote:
> Dear all,
>
> I am running logistic regressions with CSTS data using Statas xtlogit
> with fixed effects. My dependent variable is quite rare and 2/3 of all
> the groups do not have any positive outcomes during the 20 years I
> have in my dataset. As a consequence all these groups are excluded and
> I end up with a relatively small sample. I wonder if you have any
> suggestions on how to improve the model? Are there any way to
> circumvent this problem, but still cluster the standard errors?
*
*   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/