Improving discriminating power in data envelopment models based on deviation variables framework

Abstract Lack of discriminating power in efficiency values remain a major contention in the literature of data envelopment analysis (DEA). To overcome this problem, a well-known procedure for ranking efficient units; that is, the super-efficiency model was proposed. The method enables an extreme efficient DMU to achieve an efficiency value greater than one by excluding the DMU under evaluation from the reference set of the DEA model. However, infeasibility problems may persist while applying the super-efficiency DEA model under the constant returns-to-scale (CRS), and this problem tends to be compounded under the variable returns-to-scale (VRS). In order to address this drawback sufficiently, we extend the deviation variable form of classical VRS technique and propose a procedure for ranking efficient units based on the deviation variables values framework in both forms – CRS and VRS. With our proposed method, scholars who wish to prescribe theories based on a set of contextual factors need not remove large number of DMUs that are infeasible, thus avoiding problems in generalizability of their findings. We illustrate the performance and validate the efficacy of our proposed method against alternative methods with two established numerical examples.