Assessment of new methods for incorporating contextual variables into efficiency measures: a Monte Carlo simulation

Abstract

The treatment of the contextual variables (Z) has been one of the most controversial topics in the literature on efficiency measurement. Over the last three decades of research, different methods have been developed to incorporate the effect of such variables in the estimation of efficiency measures. However, it is unclear which alternative provides more accurate estimations. The aim of this work is to assess the performance of two recently developed estimators, namely the nonparametric conditional DEA method (Daraio and Simar in J Prod Anal 24(1):93–121, 2005; J Prod Anal 28:13–32, 2007a) and the StoNEZD (Stochastic Non-Smooth Envelopment of Z-variables Data) approach (Johnson and Kuosmanen in J Prod Anal 36(2):219–230, 2011). To do this, we conduct a Monte Carlo experiment using three different data generation processes to test how each model performs under different circumstances. Our results show that the StoNEZD approach outperforms conditional DEA in all the evaluated scenarios.