Type I error rate comparisons of post hoc procedures for I x J chi-square tables

The authors used Monte Carlo methods to assess the per-contrast and experimentwise Type I error rates of two post hoc tests of cellwise residuals and four post hoc tests of pairwise contrasts in 3 x 4 chi-square contingency tables. The six post hoc procedures were evaluated under three sample sizes and under the null hypotheses of independence and homogeneity. Results of the study indicate that the cellwise adjusted residual method provided adequate experimentwise Type I error rate control when appropriate adjustments to the alpha level were made, and the Gardner pairwise post hoc procedure provided several advantages over the other pairwise procedures. This was true for both the independence and homogeneity models.