"Roger Newson" <[email protected]>:
> I do not know your supervisor. However, if you want a risk ratio instead of
> an odds ratio, and you want to cluster the standard errors by twin pair,
> then you should use -glm- to fit a model with log link and binomial error.
Thank you Roger for your suggestion.
However, I think that the key word here is not risk but conditional. The
criticism I received against using logistic regression is that I have
dependent (matched) case control data. This would (for reasons I'm not to
sure about) disqualify logistic regression even with the cluster option. And
I would think also your suggestion for the same reasons.
For some reason, ANOVA and t-test for dependent measures would be valid
(with continuous outcomes) but not logistic regression even with the cluster
option - only conditional regression.
Why can't I use logistic regression with the robust option in the same way I
can use ANOVA or t-test for dependent measures? I this a matter of culture
(epidemiologists vs psychologists) or is there some fundamental property
that I have missed?
Michael Ingre
-----------------
PhD-student
Department of Psychology
Stockholm University &
National Institute for
Psychoscial Medicine
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
