PSYC2010 Lecture Notes - Lecture 6: Repeated Measures Design, Statistical Model, Variance
Document Summary
Follow up tests need to examine differences among means or sets of means with a priori and posthoc comparisons. Pairwise tests could simply do a series of t-tests but would end up with large familywise error rate. Multiple t-tests if we performed 3 t-tests we need to control familywise error rate, could evaluate each at a more conservative alpha level (performing du(cid:374)(cid:374)"s o(cid:396) bo(cid:374)fe(cid:396)(cid:396)o(cid:374)i t-test) Linear contrasts used to compare groups of means (effect of treatments combined vs placebo), a way of organising multiple t-test to evaluate more complicated hypotheses (non-pairwise comparisons) aka sum of weighted averages. Assigning weights the sum of the weights/coefficients for a comparison must equal zero. Non-orthogonal contrasts can do them but need to be aware that the contrasts contain overlapping information and you may have less power in your set. Post-hoc comparisons comparisons formulated after data examined (not planned), a range of different tests available, typically fix familywise error rates for all possible comparisons.