I am afraid I don't know the answer to your questions as I am a very new
STATA user. However, I note that you talk about the sandwich / Huber /
White variance estimator. I am working with a national dataset and a
binary outcome. Many of the small areas that I am investigating have
only one respondent.
I am using xtlogit to investigate area effects and I am wondering
whether the sandwich estimator improves the estimates of the effects and
whether I should / could use this with xtlogit?
I do not believe that robust or vce(robust) is an option compatible
with xtlogit using the default random-effects specification:
. xtlogit bin mvalue kstock, robust
robust invalid
r(198);
I don't have my full documentation handy, but I do not think sandwich
is available here.
Not my area of expertise, but if you have geographic areas with small
number of respondents per area, shouldn't you be using survey methods
such as svy: logit ? Unless you have a timeseries from a single
respondent xtlogit is surely not the right tool for the job. But if
you have some information about the sampling of individuals from
regions, it would seem that the svy: commands would be appropriate.
