Geospatial Synthetic Controls for Agricultural Impact Evaluation
Abstract
Identifying the effects of interventions is a difficult problem—especially when random assignment of treatment and continued observation of control units are uncommon and extremely rare. This paper adapts cutting edge methods from the econometrics literature to allow us to infer the effect of a treatment on a remote-sensible outcome like agricultural productivity in an observational setting. Given some treated areas, it samples thousands of points outside of the treated plots and constructs a synthetic control that best matches the pre-treatment trend of the outcome of interest for each treated area. The synthetic control is constructed using observed covariates and unobserved factors driving the variation of treated units in the pre-treatment period. This allows us to generate unbiased estimates of average and individual treatment effects as well as confidence intervals for those estimates. The attached figure shows a treatment effect that would be confounded by unobserved covariates but is accurately estimated by synthetic controls. Furthermore, the method allows us to observe the spatial distribution of points that were heavily drawn upon for any synthetic control unit and the distribution of factors that went into the creation of that control. Rather than relying on a black-box machine learning method, we show how the method constructs counterfactual outcomes for the treated plots. We test the method on real data with a simulated effect and show how we applied the method to a land-titling intervention in Benin. The paper has two major contributions: First, it allows a researcher to estimate the treatment effect of an intervention in which only treated units were demarcated, and with satellite data this can be done ex-post. Second, it allows a researcher to observe what areas and characteristics might have been good choices when selecting a control group.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMGC43B..08S
- Keywords:
-
- 0402 Agricultural systems;
- BIOGEOSCIENCESDE: 1615 Biogeochemical cycles;
- processes;
- and modeling;
- GLOBAL CHANGEDE: 1640 Remote sensing;
- GLOBAL CHANGEDE: 6309 Decision making under uncertainty;
- POLICY SCIENCES