Role of Ethanol Plants in Dakotas Land Use Change: Incorporating Flexible Trends in the Difference-in-Difference Framework with Remotely-Sensed Data
Gaurav Arora, Peter T. Wolter, Hongli Feng, David A. Hennessy
March 2016 [16-WP 564]
Arora, G., P. Wolter, H. Feng, and D.A. Hennessy. 2016. "Role of Ethanol Plants in Dakotas Land Use Change: Incorporating Flexible Trends in the Difference-in-Difference Framework with Remotely-Sensed Data." Working paper 16-WP 564. Center for Agricultural and Rural Development, Iowa State University.
The focus of this study is the Dakotas’ recent land use transitions from grass to corn and soybean cultivation. Recent literature has extensively characterized these land use changes and related concerns. However, formal analyses to understand the factors underlying these conversions are lacking. We study the role of Dakotas’ ethanol plants in these land use changes. We construct a spatially delineated dataset and implement a Difference-in-Difference (DID) model in conjunction with Propensity Score Matching to estimate the impact of a corn-based ethanol plant on nearby corn-acres. We hold the advent of an ethanol plant to be the treatment and estimate the treatment effects for each ethanol plant based on the parallel paths assumption that is standard for the DID methods. We find that effects vary by ethanol plants and so we view as inappropriate the single point estimates for all ethanol plants in a region that are usually provided in the literature. Surprisingly, we find insignificant positive, and significant but negative ethanol plant impacts on local corn-acres. Negative estimates are hard to reconcile with the economic incentives due to ethanol plants. We also find intensified corn production and reduced corn-soy rotations due to the ethanol plants. Furthermore, based on placebo tests and pre-treatment trends in corn acres, we find that the identifying parallel paths assumption of the standard DID model does not hold. We incorporate differentiated trends into the DID framework through more flexible assumptions. To validate the flexible assumptions due to differentiated trends, we implement a spatial placebo and find that estimating identified localized treatment effects in this study is challenging. The estimated treatment effects are identified for only two out of the four ethanol plants in North Dakota. The identified treatment effects on local corn acreage are found to be positive for one plant and negative for the other. In light of economic incentives provided by the establishment of an ethanol plant, the negative treatment effect is puzzling.