Using the bootstrap method to detect influential DMUs in data envelopment analysis

This paper proposes a statistical approach to handle the problem of detecting influential observations in deterministic nonparametric Data Envelopment Analysis (DEA) models. We use the bootstrap method to estimate the underlying distribution for efficiency scores in order to avoid making unrealistic assumptions about the true distribution. To measure whether a specific DMU is truly influential, we employ relative entropy to detect the change in the distribution after the DMU in question is removed. A statistical test has been applied to determine the significance level. Two examples from the literature are discussed and comparisons to previous methods are provided.