A comparison of an actuarial and a linear model for predicting managerial behavior

The objective of the current research was to compare an actuarial and a linear model for predicting criteria related to managerial success. Two subject samples were involved, both of which contained managers and potential managers who were current or past employees of a large petrochemical company. Each sample contained 2,899 individuals, who had been tested in the company's ongoing managerial assessment program, and each sample was predominantly white and male, although females and minorities were present in both samples. One sample was a validation sample; the other served as a cross-validation sample. In the first step of the actuarial analysis, twelve homogeneous subgroups of employees were identified through the hierarchical and convergent clustering of the validation sample subjects on thirteen scores available from the company's managerial assessment battery. In the cross-validation sample the twelve subgroups were replicated through a minimum distance comparison of each subject and the twelve validation sample subgroup centroids. Cross-validation subjects were assigned to the subgroup they most closely resembled. In the second step of the analysis, the twelve subgroups were cross-tabulated against various descriptive and predictive criteria. In both samples subgroup membership was found significantly associated with ethnic group, age, education, occupation, manpower classification, employment status, and two factor analytically derived job performance scores. Descriptions of the subgroups were developed in terms of the thirteen assessment scores and the various descriptive criteria. In terms of the predictive criteria, despite the significant association, it was found that subgroup membership could not be used to predict employment status better than the base rate of the high frequency criterion category. However, knowledge of subgroup membership could be used to influence the base rate of the criterion. The job performance variables were observed to have differential affinity for the subgroups in both samples, and, thus, knowledge of subgroup membership could be used to predict job performance at better than the base rate levels. In the analysis of the linear model, the thirteen assessment scores were used as independent variables in predicting employment status and the job performance scores. Multiple group discriminant analysis was employed to predict employment status. Statistically significant results were observed; however, in both samples the model could not develop better than base rate predictions of the criterion and could not be used to influence the base rate of the high frequency category. Multiple regression analysis was employed to predict the job performance scores. In both samples significant multiple R's and better than base rate predictions of job performance were observed. In comparing the models the actuarial model was slightly superior to the linear model in predicting employment status since the model could be used to influence the employment status base rate and the linear model could not. In terms of predicting job performance, the models were found equal in the validation sample. On cross-validation the linear model was observed to be significantly more accurate than the actuarial model. However, this superiority was traced to an artifact of the coarse grouping of job performance, which was done to facilitate the presentation of the data. Therefore, the models were ultimately found equal in accuracy in predicting job performance.