PSYCH 100A Lecture Notes - Lecture 14: Type I And Type Ii Errors, Null Hypothesis, John Tukey
Document Summary
Controls the overall experimentwise type i error rate, but compared to a priori t-test, is way less powerful (because critical values are more extreme) Type i error rate highly inflated 1- (1-alpha)^# of comparisons. Used only after a statistically significant f. You know at least one comparisons between the groups are different. The more groups you are comparing, the more type i of falsely rejecting the null hypothesis increases. When the investigator has one complex comparison that they are interested in, a planned comparison can be used instead of the omnibus anova null hypothesis followed by post-hoc tests. These are the most powerful tests--shotgun versus rifle. They are t-tests but comparing a combination of groups. If there are 5 groups, there are 5x4/2 = 10 possible pairwise comparisons. Complex comparisons have a common form and are usually written as. Weights are always 1/the number of groups that you are comparing.
