Non-parametric sample size requirements: per group, or overall?


Hopefully this is a reasonably straightforward question that belongs in this forum but I was wondering whether the sample size requirements (i.e. not too small to be <3-6) for non-parametric methods such as the Mann-Whitney and Wilcoxon Signed Rank test refer to per group (such as each arm in a clinical trial) or to the overall sample size?

Many thanks for your time if you have read this. I look forward to your insights.

sounds like these numbers are per arm. Once you get below 3 per arm it would be getting to a point where you could not generate p<0.05 for any possible outcome of the experiment, or nearly so. Usually when the sample sizes are small you would use the 'exact' wilcoxon test. There are probably some implications for the normality approximations used in these tests when not 'exact'.