PSYC 301 Lecture Notes - Lecture 18: Family-Wise Error Rate, Orthogonality
Document Summary
Dunn/bonferroni correction: fw= familywise error rate, c= comparisons. Holm correction (also called sequential bonferroni: better type i control but less power, most conservative other than dunn/bon, sort p values from smallest to largest, compare to . 05 divided by # of tests yet to be conducted. Benjamini-hochberg procedure: less type i control but more power, sort p"s from largest to smallest, determine new alpha by multiplying it by its rank order divided by # of tests. If 4 tests: . 05* (4/4) = 0. 05 . 05. 05* (3/4) = 0. 0375: etc, compare observed p to its new alpha. If one is significant as you go through them sequentially, then all the others are significant as well. Contrast analyses can be done either: post-hoc, when overall anova is significant, can use post-hoc to test both pairwise and complex h0s, a priori. When you have a very specific hypotheses, you can skip the overall anova and just do the contrast analysis.