ABSTRACT
The primary goal of the models approach to data analysis is to develop a model that is an adequate representation of the data. Up to now, we have approached this task using as explanatory or predictor variables only those variables that denote group membership. In most situations, such group-membership variables account for a relatively small proportion of the total variance in the dependent variable. The between-group sum of squares typically is less than half of the total sum of squares and frequently is much smaller than the within-group sum of squares. This should not be surprising. Although the more extreme early behaviorists may have hoped that they could explain nearly all the variance in behavior by their experimental manipulations, most researchers in the behavioral sciences today expect that preexisting differences among subjects are at least as important a predictor of their scores on the dependent variable as any treatment variable. This chapter considers how best to make use of information you might obtain about the individual differences among subjects that are present at the beginning of your study.
