st: Cluster Randomised Trials

[Lists <scmoore.lists@googlemail.com>]

Dear Statalist

We have areas in which interventions are deployed. Areas are in matched 
pairs, matched on certain demographic and other characteristics that 
might determine effect sizes. The intervention is randomised in each 
pair. We also have regions and would like to cluster on region as there 
are large differences in certain demographics across regions. To measure 
the effect of the intervention we survey individuals in each area before 
and after the intervention. Due to the location of the survey we 
anticipate differences in sample sizes for each survey. I’ve been 
looking through Hayes and Moulton’s recently published book Cluster 
Randomised Trials, which is great and very Stata friendly. However, I 
cannot see an easy way to incorporate the levels of information we have 
into the available models. So far I have the option of using t-tests or 
xtmixed on outcome surveys, or even manually computing some metric of 
change. Presumably, though, this is not a particularly unusual problem 
and so wondered whether anyone here on Statalist might be able to point 
me towards suitable resources or even an example? There are other issues 
with the CRT studies that are seemingly avoided in the literature – that 
of contamination. If outcome measures are both sensitive to both the 
intervention’s effect but also tempered by individual characteristics, 
and one has a metric describing the levels of contamination across 
treatment arms it would seem reasonable to adjust accordingly. Though I 
fear this would be a step too far at the moment.

Thanks in advance

Simon

*
*   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/