PS296 Chapter Notes - Chapter 16: Bonferroni Correction, Multiple Comparisons Problem, John Tukey
Document Summary
First requirement is that the f for the overall analysis of variance must be significant. If f was not significant we simply declare there are no group differences and stop there. This modified t test simply runs an apprpriate t test but replaces the pooled variance estimate in the standard t formula with mserror from the overall analysis of variance. However we don"t make this replacement if the variances of the groups are very different from each other. This works bc ms error is the average of the variance within each group (if there were only 2 groups, the ms error from anova would be the same as the sp2 from two sample t test) Ms error is based on variability within all groups rather than just the two groups we"re comparing at the moment (which would be pooled variance in a t test) thus it is a better estimate of e.
