STATISTICAL POWER AND COST IN-TRAINING EVALUATION - SOME NEW CONSIDERATIONS

Two ways to reduce the costs of training evaluation are examined. Firs t, we examine the potential for reducing the costs of training evaluat ion by assigning different numbers of subjects into training and contr ol groups. Given a total N of subjects, statistical power to detect th e effectiveness of a training program can be maximized by assigning th e subjects equally to training and control groups. If we take into acc ount the costs of training evaluation, however, an unequal-group-size design with a larger total N may achieve the same level of statistical power at lower cost. We derive formulas for the optimal ratios of the control group size to the training group size for both ANOVA and ANCO VA designs, incorporating the differential costs of training and contr ol group participation. Second, we examine the possibility that using a less expensive proxy criterion measure in place of the target criter ion measure of interest when evaluating the training effectiveness can pay off. We show that using a proxy criterion increases the sample si ze needed to achieve a given level of statistical power, and then we d escribe procedures for examining the tradeoff between the costs saved by using the less expensive proxy criterion and the costs incurred by the larger sample size.