We evaluate three bootstrap techniques for the statistical analysis of a non parametric production model for which a DEA measure of efficiency is potentially affected by a set of exogenous factors. The application of interest relates to the assessment of the significance of the contextual variables income generation, processes improvement, intensity of partnerships, type and size on the technical efficiencies of Embrapa's research centers. It is concluded that the bootstrap of the maximum likelihood estimator provides the best fit from the point of view of Pearson correlation between observed and predicted values and is the most informative in regard to the significance of the variables considered. With the exception of size all contextual variables are statistically significant. Income generation, processes improvement and intensity of partnerships are all positively associated with technical efficiency.
DEA; bootstrap; two-stage estimation
