The Use of Regression Diagnostics to Improve Model Fit: A Case of Role Strain and Job Stress

Abstract

This paper illustrates the importance of using regression diagnostics to improve model fit when using standard multiple regression statistical packages such as SASPC. This study examined the relationship between employee perceptions of their work environment and perceived job stress. The analysis was theory driven rather than exploratory in nature, and was performed using SASPC multiple regression procedures. Variables were coded to reduce possible collinearity. Various regression diagnostics were examined to detect the presence of outliers, influential observations, residual correlation, and collinearity (e.g. VIFs, DFFTTS, the C criterion, HAT (leverage) values, and the Durbin-Watson test). These values, coupled with the various regression procedures yielded a final, best nine-variable model of R^2 = .48, significantly larger than the initial value of R^2 = .27. Future research in this area could be strengthened through 1) an examination of the path analytic and LISREL models in the literature that attempt to model indirect effects, 2) possible incorporation of select, higher-order terms from these studies, and 3) utilization of the regression diagnostic procedures outlined in this paper.