So far we have considered contrasts only in the situation of “between subjects” designs. That is, every element to which we assigned a weight, prediction, or λ was comprised of different sampling units. But it happens frequently in the conduct of judgment studies that we measure our sampling units more than once, as in studies of sensitivity to nonverbal cues in various channels of nonverbal communication, or in studies of learning, or in developmental studies of changes over time. Indeed, sound principles of experimental design lead us frequently to employ within-subject or repeated measures designs in order to increase experimental precision and, therefore, statistical power (Rosenthal & Rosnow, 1984, Chapter 22). Our need for employing contrasts is just as great within repeated-measures designs as in between-subjects designs. Consider the following study: Four subjects each are measured on three occasions one week apart on a test of nonverbal decoding skill. The entries of Table 9.1 are the observations of nonverbal skill scores obtained on three occasions by each of the four subjects.

The preliminary analysis of variance of Table 9.1 tells us that nonverbal skill performance varies from occasion to occasion, but that was not our question. Our primary question was whether there was a steady increase in performance over time, that is, a linear trend with weights of -1, 0, +1.