Partitioning Variance in Multiple Regression Analyses as a Tool For Developing Learning Models

Multiple regression will continue to be an extremely valuable analytical tool in educational research for many years to come even though many investigatiors (e.g., Hanushek, Levin, Michelson, Werts) recognize that such primitive single-equation models must in the long run be superseded by more sophisticated structural models. The necessity for more complicated models that make some kind of causal sense is best appreciated by econometricians who have now achieved considerable success in formulating and verifying structural models of economic activities. The most massive example is a comprehensive model of the U. S. economy (Duesenberry, et al.). We seem to be quite a long way from that happy state of affairs in education. At least my own struggles to understand educational data have not led me to any very convincing causal connections; in fact they have led me to believe that education is, at best, an order of magnitude more complex than economics and that we have much floundering searching to do before we can confidently write down identities which relate endogenous to exogenous variables as econometricians do when they state, for example, that