Re: st: ttest or xtmelogit?

[Jeph Herrin <junk@spandrel.net>]

You may want to look at a program I wrote called -clchi2-,
which is part of the package -cltest- on SSC. This performs
a chi2 test of independence, adjusted for clustering. It
implements an adjustment suggested by Donner & Klar (don't
recall the original paper, but it's in Cluster Randomization
Trials in Health Services Research, 2000). It needs only 3
clusters per comparison group to estimate the intra-cluster
correlation (though it doesn't account for the uncertainty
in basing the ICC on such a small number of clusters).

hth,
Jeph




David Airey wrote:
> .
>
> I have a typical pilot data set. Small. I have 12 mice, 6 from one 
> group, 6 from another. For each mouse I have about 50 yes/no scores. 50 
> was enough to get precision on a given mouse. I'm interested in the 
> group difference. In the past I used xtgee with mouse as the i(mouse) 
> option, i.e., mouse as the cluster, with family(binomial) and 
> link(logit). But previously, I did this with double the number of mice, 
> so I felt I had enough clusters. Here, I am feeling uncomfortable about 
> the number of clusters (12). The same goes for xtmelogit, new in Stata 
> 10. Given the number of mice, is it better to simply transform the 
> summary statistics for animals (mean of the yes/no vector by animal), 
> like with an arcsine or logit, and use a ttest?
>
> -Dave
> *
> *   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/
*
*   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/