Multiple Comparisons with Dichotomous Data

Statistical tests for the overall hypothesis of no difference among multiple groups of a dichotomous response experiment include the likelihood ratio, Pearson's X 2, Goodman's Y 2, and the analysis of variance (ANOVA) F. An empirical study is performed to evaluate these tests and the Bonferroni and maximal contrast methods of multiple comparisons for a four group experiment. The F test is recommended as an overall test and for maximal contrast comparisons, but its nominal significance level can be optimistic for certain Bonferroni comparisons with small samples. The X 2 test is generally more conservative than the F test.