A computer simulation to examine the performance of fixed effect F and kruskal-wallis ANOVA tests on educational data

Abstract EN

The present study empirically examined the effect of nonnormality defined by a combination of multimodality, tailedness.

and asymmetry, variance heterogeneity, and sample size on the type I error and power properties of the parametric single-factor ANOVA F and Kruskal-Wallis test.

Simulated data were generated representing multimodal, light, and moderately heavy tailed and skewed distributions, and three sample sizes.

Based on type I error and power results, the parametric ANOVA F test is recommended for all unimodal and bimodal, light and moderately heavy tailed distributions combined with all variance patterns and sample sizes.

However the Kruskal-Wallis test is recommended for trimodal, and moderately heavy tailed distributions, extreme variance heterogeneity, and large sample sizes.