Hidden heterogeneity in measuring production factors: Implications for two-stage efficiency analysis

Data envelopment analysis (DEA) is one of the two primary estimators of technical efficiency and is widely applied in policy evaluations within agricultural, environmental, and other domains. In the two-stage efficiency analysis, the DEA efficiency scores are estimated in the first stage, followed by an assessment of the influence of selected policy variables on these scores in the second stage. This paper demonstrates that two-stage efficiency DEA analyses are not robust to variations in the measurement of fundamental input variables, even when the correlation between alternative input measures exceeds 0.9. This lack of robustness is reflected in substantial heterogeneity in both statistical significance and the signs of parameters that capture the effects of environmental variables on efficiency. Consequently, by selecting seemingly interchangeable inputs, it is possible to obtain results that align with prior expectations, raising serious concerns about the reliability of DEA-based policy analyses. We argue that, given the nature of the problem, robustness cannot be achieved through methodological refinements of the DEA itself. Rather, the only viable strategy is to explicitly assess the robustness of the results with respect to alternative input specifications.