Thank you Joseph and David for the replies. I have a
couple of follow-up questions:
--- Joseph Coveney <[email protected]> wrote:
[snip]
> -xtlogit, re- would seem to be the remaining
> alternative available in Stata,
> unless I'm overlooking something. If there
> is a substantial correlation between the fixed
effects (physician covariates) and the random
> effect, then the parameters are liable not to be
> consistently estimated.
How can I test this?
If doing an <xtlogit, re> command, you could get the intraclass
correlation with loneway as they did at:--- David Airey <[email protected]> wrote:
> But when stuck with a small data set, why not run a
> model designed for
> that data structure, as opposed to running a model
> not designed for the data structure?
This is an interesting point, because in my data I
have 50 clusters and 410 observations, so the number
of patients referred by a physician ranged from 1 to
43 with a median of 5 referrals.
I guess the question that remains is whether or not I
can justifiably use this approach?
One answer I received off list for my query was sometimes the
underlying math of the method doesn't like the small sample, even if
the data fit the model by design. I'd still use the model that fit the
design, since the conclusion has to be qualified by small sample size
or by ignored clustering anyway. Ignore the clustering and see what
differences happen. That's usually my approach, not being one with the
underlying math!
