Teaching Applied Researchers to Create Their Own Statistical Models

Abstract

The purpose of the following remarks is to give you something of the flavor of a novel approach to the teaching of statistical model building and manipulation. Historically, it evolved out of an applied environment in which many of the classical models appeared to be inadequate or at least deficient in one or more respects. Students in applied areas who have been exposed to the approach respond enthusiastically to it, and, in general, the more "traditional" work they have had, the greater their enthusiasm. The response of teachers has been mixed. Many of the critics make remarks similar to those criticisms that are directed at the "new math." It is certainly accurate to state that students of this approach get very little practice in arithmetic for even the most elementary models. In fact, the primary text [6] is almost totally devoid of computing formulae.

With respect to mathematical and statistical foundations, we rely very heavily on the theory of the classical fixed-x linear model, and the text bears some superficial resemblance to a typical text on linear models. However, a great deal of the material covered in a typical linear models text will be found in ours only indirectly, if at all. Conversely, the concepts we identify and the skills we try to develop are only indirectly inferable from the typical text.