Estimating the causal effects of the French agro-environmental schemes on farmers' practices by difference in difference matching

Summary

We present the first disaggregated estimation of the causal effects of a nationwide AES program on environmentally-relevant practices for a nationally representative sample. We study the AES program that has been implemented in France between 2000 and 2006 as part of the EU policy for rural development. The causal effect of the AES program is the average difference between practices of participants in the presence of the program and what their practices would have been had the program not been implemented. This counterfactual situation cannot be observed and intuitive approximations using non participants' practices or participants' practices before receiving the AES are plagued by large sources of bias (selection bias and time trend bias). To correct for these sources of bias, we use Difference-in-Difference (DID) matching. The main idea is to compare each participant to a non participant "twin" with the same observed characteristics. We combine matching with DID to correct for any remaining bias that may be due to unobserved variables. Our results suggest that the size of windfall effects of the AES programs depends on the specific prerequisites for each AES. For example, the AES supporting conversion to organic farming, which combined large payments with the strong requirement of not being an organic farmer when applying for the AES, have had negligible windfall effects. We cannot conclude on the extent of windfall effects of payments for extensive herding. We provide a first attempt to integrate the results of causal analysis in a cost-benefit evaluation of the AES. We argue that evaluation should be prepared at the same time the program is designed. This would improve the collection of data and enable using the first results of the evaluation to design the next program (by suppressing or altering the less efficient AES). Finally, this could allow for the implementation of experimental methods to evaluate policies that do not comply with the assumptions of DID-matching, like payments for extensive herding.