Teaching How to Derive Directly Interpretable Coding Schemes for Multiple Regression Analysis

Multiple linear regression is a versatile model for encompassing analysis of variance, analysis of covariance, and aptitude-by-treatment interaction designs. The question of how to teach the coding of levels of a qualitative variable is addressed in this paper. Although a variety of coding schemes will produce invariant omnibus statistical results for a given set of data, one’s interpretation of treatment effects and treatment differences depends on the particular code values chosen. A general procedure is presented that allows the user to generate, on an a priori basis, code values that result in directly interpretable estimates of interest.