Answers are grounded in this paper and cite their sources.
A two-step framework for the interpretation of significant two-treatment interactions is proposed. First, a contrast between the means of one treatment is estimated separately at each level of the second treatment. A partial interaction test tests the hypothesis that the variabiliy among these contrasts is zero. The second step consists of computing a difference between the separately estimated contrasts. An interaction contrast test tests the hypothesis that this difference is zero. The familywise Type I error rate can be controlled at alpha by employing Gabriel's simultaneous test procedure for partial interaction tests and Scheffe's method for interaction contrast tests. Recently, an issue has arisen concerning a posteriori tests after detection of a significant interaction in the analysis of variance. Marascuilo and Levin (1970) and Levin and Marascuilo (1972) have criticized the use of simple effects tests in which the means within a single row or column of the data matrix are compared. Their basic criticism is that the null hypotheses tested by simple effects tests are not coherent with the null hypothesis tested by the a priori omnibus interaction test. To maintain
No takes yet. Share an insight, caveat, or question.
Robert J. Boik (1979) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: