Assessing the productivity consequences of agri-environmental practices when adoption is endogenous

We address the general problem of selection bias, an issue endemic to policy analysis when adoption is voluntary, with an empirical application to environmental policies for agriculture. Many voluntary practices for mitigating the environmental impacts of agriculture provide external benefits while lowering productivity. Policy analysis of the productivity consequences is complicated by the fact that decision makers can choose their own policy levers, an action that ruins any notion of random assignment. We introduce an identification strategy to correct this kind of endogeneity, combining classic methods from stochastic frontier analysis and selection models. Applying it to micro-level data from Finnish grain farms, we find that more efficient producers are more likely to enroll in subsidized practices. And, because those practices tend to reduce yield, frontier analysis without the endogeneity correction greatly understates the productivity loss. In other words, naïvely basing the frontier estimator on the subset of less-productive farms leads to downward bias in the frontier estimates. In fact, average inefficiency more than doubles after the correction in this case. An outlier investigation suggests that the lowest decile of farms are responsible for most of the selection bias in the uncorrected model.