I have a small, but untidy, dataset. I think I should use non-parametric tests, but I want to check my understanding since I have not often used non-parametric tests.

My dataset is from a small chain of health clubs (10 clubs) that created a program to encourage members to use their benefits more frequently. There are 12 outcome variables representing 12 different benefit types. I have data on member use a year before the program started and for two years while the program ran. The data are recorder as the percentage of members at each club who used a particular benefit type during a given year. So there are 10 clubs and 12 outcome variables; each outcome variables is measured at three points in time.

I should point out that the membership changed somewhat during the three years. For some members I have data for all three years. But some members left during the three years while other members joined during the three years.

Iím think of the ten clubs as the observed units in my study. Since I have data at three points in time for each club, it is a repeated measures. I donít know if the populations are normally distributed, and the group variances are not equal, so I think I should use a non-parametric test. In this case, I think I would use the Friedman test, and then use Wilcoxon signed-rank for post hoc testing.

I plan to drop some outcome variables, but even if I kept only 3 or 4, I will still adjust my alpha level to account for multiple tests.

Is my thinking correct on this? Have I missed anything? Is there another test that would be better suited to my data?

I will appreciate any help.

Thanks,
Jeff