A statistical test for detecting influential observations in data envelopment analysis

This paper deals with the problem of detecting influential observations in deterministic nonparametric DEA models. The technique we present is intended to classify for a further analysis those sample observations considerably affecting the measured efficiency for the remaining units. Then, the analyst will have to check whether these observations are contaminated by data errors or not. This approach also allows to determine when efficiency changes due to the presence of a given unit in the sample are statistically significant. Thus, ours is a statistical alternative to approach the problem of detecting influential observations in deterministic nonparametric DEA models.