Date:         Tue, 19 Jun 2007 15:19:41 -0000
Reply-To:     Daniel <daniel.biostatistics@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Daniel <daniel.biostatistics@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Permutation test for group-randomized trial
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset="iso-8859-1"

Good morning all,

I would like to apply the methods of Gail et al. ("On Design
Considerations and Randomization-Based Inference for Community
Intervention Trials"; Stat Med 15; 1069-1092) to my data which come
from a cross-sectional group-randomized trial with matched pairs.

Regarding the design, we had 10 pairs of clinics that were matched and
for each clinic I computed the difference (W_i) of the log odds of my
outcome (dichotomous, measured at the patient level) at times 1 (post-
intervention) and 0 (baseline). For each pair, I then computed the
difference W_1 - W_0, where W_1 corresponds to the experimental
treatment group's difference, and W_0 corresponds to the control
group's difference. I therefore have 10 differences, of which I then
computed the mean (U).

I would like to do a permutation test. More specifically, I would like
to compute, for each of the 2**10 possible permutational scenarios,
the mean of those differences, and then determine the proportion of
those means that are as extreme or more than my original data's U.

My question is: is there a way to do this in SAS, using a procedure's
option (as opposed to programming the algorithm, which I'll be happy
to do if there are no built-in functions)? Because there are only 1024
scenarios, I would like to enumerate all of them as opposed to
selecting just a random sample.

Thank you in advance for your input,

Daniel