This article presents a new framework for the impact evaluation studies of programs that aim to increase efficiency and productivity of firms. Likelihood ratio statistics support a spatial autoregressive stochastic production model that estimates the technical efficiencies of matched subsamples of treated and control agents determined using Propensity Score Matching. The model is used to analyze the efficiency of farmers who participate to an agricultural extension program in rural areas of the North West of Tanzania. This study reveals a 9.2% gain in technical efficiency from the program participation. There is evidence of a benefit of 5.7% gain in output value of non-participants as a spatial spillover effect