Thanks Austin and Stas!
That is good to know. I prefer your first practical approach for simulation, and the xtlogit for modeling the real data.
-Dave
> If you have 1000 obs per cluster, you might prefer to include fixed
> effects (50 or so dummies) in regular logit--the bias from using
> dummies is quite small when the number of obs per cluster is large.
> Also, if you have 50-60 clusters (20-30 per T/C group rather than
> 20-30 in both groups) you can use the cluster-robust VCE. The bias
> (in SEs) is quite small with more than 50 balanced clusters. This
> should be fairly easy to simulate, since -logit- is fast. A better
> simulation would compare to -xtlogit- of course, but 10000 iterations
> of something that takes a day to converge is not feasible... though if
> you supply the true parameter vector with option from(b), it might
> converge a bit faster.
*
* 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/
