Re: st:Command for clustering without sampling

[Austin Nichols <austinnichols@gmail.com>]

Nanlesta Pilgrim <nanlesta@gmail.com>:
In general, (3) is a good approach, since clustering will get you
better (more accurate) SEs if you have a reasonable number of clinics
(you obviously should not cluster by state with only 3 states). You
should only "control" for state and clinic (i.e. include fixed
effects) in a logit if you have very large numbers of obs per clinic.
You might also check out
http://www.stata.com/help.cgi?xtmelogit
and read
http://www.stata.com/statalist/archive/2007-10/msg00926.html
http://www.stata.com/statalist/archive/2007-10/msg00935.html
http://www.stata.com/statalist/archive/2007-10/msg01008.html
etc. for background.

On Tue, Aug 31, 2010 at 10:56 AM, Nanlesta Pilgrim <nanlesta@gmail.com> wrote:
> Hello All!
>
> I am trying to figure out which stata command I should use given the
> dataset.  Basically, we collected information in clinics across three
> states.  It was not the manner in which we would have preferred but we
> did not sample clinics nor state due to the nature of the study. In a
> sense, there really was no "cluster sampling," however, participants
> attending the same clinics will have similar experiences than those
> attending other clinics.  Participants were also not sampled given the
> type of participant we were trying to recruit. Our main interest is
> how experiences at baseline clinic visit influence behavior at
> follow-up.   In running the analyses, I'm debating over what command
> to use to control for "clustering,"  given participants were
> interviewed within clinics within states.  I've contemplated (1)gee,
> (2)svy, (3)logit with clustering option, and (4)logit controlling for
> state and clinic.  Any guidance would be very helpful!
>
> Thanks,
> Nandy

*
*   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/