I frequently see arguments of the form "ANOVA residuals were non-normal, so we used a nonparametric test instead" in papers I review. (IMO) this does not make sense, since for the sample sizes considered there, the CLT kicks in quite nicely... and sometimes the nonparametric alternatives are misunderstood or nontrivial (think interactions or repeated measures). Is there some canonical reference I could point people to? Do we have a thread on that I haven't found?