st: Alternative models for logistic regression when there is high within-panel correlation

[Nicola Man <nman1011@gmail.com>]

Hi,

I have an issue that is possibly similar to the one in the message
below, only it is now logistic regression that I have a problem with.

http://www.stata.com/statalist/archive/2009-06/msg01083.html

I am getting very high ICC >95% for paired dichotomous outcome (by
subject) data, and got ridiculously high odds ratios (particularly
when I use large number of integration points, but it actually looked
acceptable when I used quad(4)).  Would using a GEE model in xtlogit
be a way to go?

A second problem then is that I have a higher level clustering
variable (by school which has an ICC ~ 4-15%), so presumably I would
then need to use multi-level modelling.  Would you see that using
xtmelogit (or gllamm) could still be a problem?  If it is, what is the
way around this?

Thanks for any help or advice!

Cheers,
Nicola



-- 
Cheers,
Nicola

If urgent, please also email to:
n.man@unsw.edu.au
v2B2@hotmail.com

*
*   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/