On the Assumptions Concerning the Assumptions of a t Test

Summary A computer simulation which generated various sample distributions was performed. The equal interval assumption and degree of skewness were systematically manipulated to examine the sensitivity of the t and Mann-Whitney tests to varying degrees of violations of these t test assumptions. Results show that of the 348 sets (2436 t tests), 94 (27%) contained shifts from significant to nonsignificant t test values under mild violations of the equal-interval and normality assumption and would have led to decision errors. Of these, 55% would have been Type I and 45% would have been Type II. The implications of these results are discussed relative to Types I and II errors, to the use of parametric and nonparametric statistics, and to the likelihood of encountering such problems in sampled distributions.