This paper considers the appropriateness of the F test when the dependent measure is discretely distributed and takes on only a small number of possible values ( m = 3, 5 or 7). Effects of α level, number and size of treatment groups, and skewness and kurtosis of the parent population are described. Results are consistent with those previously reported for continuously distributed measures; the F test is quite robust with respect to Type I error rate.
No takes yet. Share an insight, caveat, or question.
Bevan et al. (1974) studied this question.