ABSTRACT
In paired-observation applications of it, the differences between scores in the first group or occasion and the matching scores in the second are ranked in absolute value, and then the signs of the differences are reattached. One can develop unbiased estimates analogous of the variances and covariance rather than simply substituting the sample quantities that parallel the terms. A test of monotonic relation between score and time, within subjects, that is more direct than the Friedman-Kendall procedure, and probably more powerful, is possible. The suggestion early in the chapter concerning the pretest-posttest between-groups design shows that d analysis can be applied to mixed designs as well, and that suggestion can be elaborated to more complex designs. In traditional ANOVA, interest would focus instead on the main effect of construction. A "main effect" reflects a mean difference between conditions that is averaged over an orthogonal dimension of the design.
