METHODOLOGICAL ISSUES IN MODELING ABSENCE - A COMPARISON OF LEAST-SQUARES AND TOBIT ANALYSES

Predictions of absence made from ordinary least squares (OLS) regression analysis were compared with those made from Tobit analysis (for truncated distributions) in a field survey of 193 male professionals in a large Canadian aerospace organization. In both OLS and Tobit analyses significant predictors of time lost included number of children, comparative absence, job involvement, and life-events stress. Age, comparative absence, job involvement, and life-events stress were significant predictors of absence frequency in the OLS analysis; number of children became an additional predictor in the Tobit analysis. Variance explained increased by 4% for time lost and by 6% for absence frequency when the Tobit model was used. The methodological advantages of using Tobit for truncated distributions and their implications for future research on absence are outlined.