LARGE SAMPLE AND OTHER MULTIPLE COMPARISONS AMONG MEANS

Abstract

If the number of multiple comparisons for K large samples is limited to the (k/2) differences between pairs of means plus ‘a few’ for possible post hoc multiple comparisons, one can frequently obtain shorter confidence intervals using a t statistic than through the use of Marascuilo's procedure (1966) for large samples. The savings in the length of the confidence interval increase with the number of samples and with the confidence level. The length of the t intervals is compared with the length of Marascuilo's intervals under various conditions in Tables 1–6. Modifications are given for use in the case of moderate and small samples. A numerical example (Marascuilo's Example 1 (1966)) is presented, and two non‐parametric analogues of the procedure are given.