Paying Attention to the Default Reference Category in Several SPSS Statistics Procedures: An Example of Coding Reversal

Abstract

This paper reviews two common approaches to handling categorical predictors in regression analysis and how they are implemented in several Statistical Package for the Social Sciences (SPSS) procedures. Through this review, the aim is to revisit a warning from the SPSS literature regarding dummy coding for categorical predictors with just two classes. In this problem, the original coding of a two-category dummy variable is reversed by the software program without any alert and such a change is likely to result in an incorrect interpretation of the regression coefficient estimate of the corresponding categorical predictor. Further, the paper discusses the generalizability of this conclusion to several other statistics programs: Statistical Analysis System (SAS), JMP, and STATA.