Categorical Variables in Multiple Regression: Some Cautions.

Various methods of coding categorical variables for use as predictors in multiple regression analyses have been presented in the literature. The limitations of two oft-discussed methods -- dummy coding and nonsense coding -- are detailed for several frequently-used regression designs. Two examples involving potentially inappropriate interpretations of the results of analyses involving dummy coding are presented. Researchers are cautioned that the parameter estimate or estimates and test of significance associated with a predictor variable or set of predictor variables in an equation which involves dummy- and/or nonsense-coded predictors represent an effect of interest only in a limited set of circumstances.