ABSTRACT
In this chapter we consider the case where more than two groups of participants are being compared on several dependent variables simultaneously. We first briefly show how the MANOVA can be done within the regression model by dummy-coding group membership for a small sample problem and using it as a nominal predictor. In doing this, we build on the multivariate regression analysis of two-group MANOVA that was presented in the last chapter. (Note that Section 5.2 can be skipped if you prefer a traditional presentation of MANOVA). Then we consider traditional multivariate analysis of variance, or MANOVA, introducing the most familiar multivariate test statistic Wilks’ Λ. Two fairly similar post hoc procedures for examining group differences for the dependent variables are discussed next. Each procedure employs univariate ANOVAs for each outcome and applies the Tukey procedure for pairwise comparisons. The procedures differ in that one provides for more strict type I error control and better confidence interval coverage while the other seeks to strike a balance between type I error and power. This latter approach is most suitable for designs having a small number of outcomes and groups (i.e., 2 or 3).
