Assessing the Significance of Covariates in Output Oriented Data Envelopment Analysis with Two Stage Regression Models

We propose alternative regression models to assess the effects of covariates in output oriented DEA scores. We use probability choice models combined with specifications related to the gamma and to the truncated normal families of distributions. These specifications imply different two stage regression models and alternative quasi maximum likelihood estimators. We apply these methods to assess the significance of technical effects – type of unit, processes improvement and technology impact – affecting DEA efficiency scores computed to agricultural research production in Brazil, measured through Embrapa. We favor models taking into account the whole sample of efficiency scores and not only the inefficient units. In our application we conclude that type of unit and processes improvement are significant effects, with the latter effect negatively associated to the efficiency of the units classified into the type of unit considered more efficient. Key-words Data envelopment analysis; Fractional regression models; Second stage DEA regressions; Bootstrap; Contextual variables; Agricultural research.