One-stage and two-stage DEA estimation of the effects of contextual variables

► In two-stage DEA (2-DEA) efficiency estimates are regressed on contextual variables. ► We show that 2-DEA is consistent under more general conditions than earlier assumed. ► We develop a new semi-nonparametric one-stage DEA (1-DEA) estimator. ► 1-DEA performs better than 2-DEA in Monte Carlo simulations.