Epidemiologically inspired approaches to land-use policy evaluation

Environmental policy evaluation is crucial to determining if policy objectives were achieved. In most cases, some of the outcomes can be measured but a proper statistical analysis is difficult to achieve since the data may not represent a random sample (i.e., the data is biased), are not representative of the population or cannot be compared to a control group. This work adapts quasi-experimental statistical methods widely used in epidemiological studies that could be applied to land use policy evaluation in situations of relatively poor data. In order to test and develop this set of methods, we evaluated the effect of a land-use policy known as the rural environmental registry (CAR) on the reduction of deforestation rates in the Brazilian Amazon rainforest. The random variable of interest is the number of deforested hectares in given private properties and the statistic of interest is the difference of the annual deforestation rate between the properties before and after the policy intervention. Since no formal statistical distribution properly fit the data, non-parametrical approaches such as Monte Carlo simulations and Bootstrap were used. Data from the Brazilian states of Mato Grosso and Pará were used, with different time periods and three rural property size classes. Results show that the properties inside the Rural Environmental Registry have reduced their deforestation rates in some property classes and time periods, but this effect has not been systematic across time and space indicating that the policy is only partially effective. We conclude that the proposed statistical methods can be useful in environmental policy evaluation in different contexts due to low demands in terms of data availability and statistical distribution assumptions.