Using the PISE Criterion to Measure the Effects of Imbalance in the Analysis of Covariance
Describes a new statistical technique for comparing unbalanced experimental designs which will be modeled by the univariate analysis of covariance. The author proposes minimizing a design criterion variable called PISE (percent inflation of the standard error of a contrast). The research was motivated by the need to design an experiment to measure the effectiveness of a potential new Army recruitment policy. The policy would provide greater management flexibility in paying cash bonuses to eligible "high-quality" young men who agree to enlist in the U.S. Army. Results are provided for both the standard Gauss-Markov model (constant error variance) and the model with heteroscedasticity. Also discussed is the problem of attributing the increased variance caused by imbalance in a design to particular covariates. When implemented, the proposed PISE criterion will generate a design which has greater sensitivity to treatment effect differences.