Probabilistic Measures of Efficiency and the Influence of Contextual Variables in Nonparametric Production Models: An Application to Agricultural Research Evaluation in Brazil

In a research institution it is important to identify which management practices have influence on the production efficiency. In this paper we assess the statistical significance of contextual variables type, size, financial resources acquisition, intensity of partnerships, processes improvements and management change. The analysis is carried out for the Brazilian Agricultural Research Corporation over the period 1999-2006. The statistical analysis uses a balanced dynamic panel data model. We conclude that only financial resources acquisition is statistically significant. The association with the production process is positive. We also found statistically significant the two lag inertial component of the ratio conditional FDH (Free Disposal Hull) to unconditional FDH indicating a two year effort to improve efficiency. Introduction The Brazilian Agricultural Research Corporation (Embrapa) uses a production model to monitor its research production. Embrapa has 37 research centers, spread throughout the country. The model has multiple objectives. Firstly it allows the measurement of outputs and inputs in a systematic way. Proper qualification of inputs and outputs provides a quantitative basis that easier the understanding of the company’s operations. Secondly it provides a sound basis for decision making and strategic planning at the administration level. Thirdly the computation of measures like productivity, economic efficiency and total factor productivity allows the identification of benchmarks and best procedures intended to increase overall performance and reduce differences within the organization. Finally, measures of variability in efficiency through time serve the purpose to assess the performance of the administration. In this context, the Embrapa’s performance evaluation model is a decision support system. This article is concerned with the identification of contextual variables external to the production process that may be affecting or causing efficiency. Typically these variables are in control of the institution. The assessment of their effect is of importance, since they may serve as a tuning device to promote efficiency. The use of technical efficiency as a performance and evaluation measure raises some questions within the organization. An important one is whether or not the process generates unwanted competition among the research centers. A typical criticism is that the evaluation system may inhibit partnerships. The identification of causal factors of efficiency demands appropriate statistical modeling. In Embrapa, Data Envelopment Analysis (DEA) technical efficiencies are computed, since 1996, under constant returns to the scale. Recently, Souza (2006) and Souza et al. (2007) assessed the influence of covariates on the DEA efficiency measurements using analysis of variance, dynamic panel data and maximum likelihood methods. A potential problem arises in this approach: the contextual variables used may affect the production frontier. This