Post-hoc Tests in GLM, GRM, and ANOVA

Post-hoc comparisons can be performed on the GLM, GRM, and ANOVA More Results - Post-hoc tab.

Post-hoc comparisons. Post-hoc comparisons are usually performed when an analysis (involving categorical predictor variables) yields unexpected results (patterns of means), and one wants to be sure that those unexpected results are reliable (see Contrast analysis and post-hoc tests for additional details). STATISTICA GLM will perform all post-hoc tests on observed weighted marginal means (see also the Means tab for details). In a full factorial design, when the highest order interaction is chosen in the Effect drop-down box (see GLM, GRM, and ANOVA More Results - Post-hoc tab, these are simply the observed means within each cell of the design; if there are no continuous predictors (covariates) in the design, these are also the least squares means. The different post-hoc test procedures available in STATISTICA GLM are described below.

Post-hoc tests for random effects. The error terms for all post-hoc tests will always be computed from the sums of squares residuals. Those error terms may not be appropriate, and when random effects are involved, you should interpret the results post-hoc tests with caution. Note that the Post-hoc tab provides options for specifying a user-defined Error term (see Error term for post-hoc tests in GLM, GRM, and ANOVA).

Post-hoc tests for effects involving repeated measures. Post-hoc comparisons for effects involving repeated measures pose a particular problem regarding the choice of the appropriate estimate of sigma, i.e., the residual or error variation in the data. Specifically, it is not clear how to choose the appropriate error term for comparisons of means across the levels of between-group factors and within the levels of the repeated measures factors, and for the means across the levels of one or more repeated measures factors within the levels of the between group factors. A detailed discussion of this issue -- the post-hoc comparisons of means in effects involving between-group and repeated measures factors -- is provided in Winer, Brown, and Michels (1991, pp. 526-531), and Milliken and Johnson (1992, pp. 322-350). The GLM module provides you with full control over the choice of error terms; see Error term for post-hoc tests in GLM, GRM, and ANOVA. See also Post-hoc tests for repeated measures effects in Comparison with other general linear model programs.