Testing the accuracy of DEA for measuring efficiency in education under endogeneity

DEA is the most common and well known technique to evaluate efficiency in education because it easily allows dealing with a complex multi-output multi-input production framework. After an interesting debate in this journal with Ruggiero (2003), Bifulco and Bretschneider (2001, 2003) concluded that DEA does not perform well to measure efficiency in the presence of endogeneity and measurement error because it leads to quite biased efficiency estimates. However, their experiment was conducted considering a one-output framework, constant returns to scale and a simple Cobb-Douglas specification. Following Bifulco and Bretschneiders’ original idea, in this paper we update the adequacy of DEA in the presence of endogeneity and measurement error in a scenario closer to the educational context. To do this we perform a Monte Carlo experimentation in a flexible multi-output multi input translog context. Our results point out that DEA obtains quite accurate measures of technical efficiency when the sample size is large enough even in the presence of endogeneity.