The effect of sample size on the mean efficiency in DEA: Comment

Zhang and Bartels (1998) show formally how DEA efficiency scores are affect ed by sample size. They demonstrate that comparing measures of structural i nefficiency between samples of different sizes leads to biased results. Thi s note argues that this type of sample size bias has much wider implication s than suggested by their example. Models which implicitly restrict the com parison set like some models for non-discretionary variables lead to biased efficiency scores as well. A reanalysis of the Banker and Morey (1986b) da ta shows that the efficiency scores derived there are significantly influen ced by the variation in sample size implicit in their model.