Valuing Neighborhood Amenities with Continuous Spatial Difference-in-Differences

• In Event: Spatial Methods and Geographic Dimensions of Policy Evidence

Abstract

Many studies use difference-in-differences to identify the impact of an event occurring at a location in space on house prices. For example, an industrial plant opens at a particular address and spews pollutants into the air, lowering nearby houses' values. A common research design to evaluate these price effects is a ring DiD, which divides houses into discrete groups based on their distance to a treatment site. For an industrial plant, a researcher might compare the prices among houses within a mile of the plant to the prices for houses 1-2 miles away, before and after the plant opens. Researchers often take the resulting price impact estimate as a welfare measure for homeowners' willingnesses to pay (WTP) for the treatment site by relating to the hedonic model. Such estimates may be useful to inform cost-benefit analyses on, for example, pollution abatement policies.
However, there is a disconnect between most ring DiD applications, which discretize space, and the hedonic model, which hinges on a continuous price function. For instance, the relation that would characterize people’s preference for proximity to a polluting industrial would be estimated by the derivative of price with respect to distance from a plant.
I formalize the econometric framework for this setting in a potential outcomes model where the treatment intensity is a continuous function of distance to the treatment site, thereby revealing that standard difference-in-differences approaches require a treatment effect homogeneity assumption to recover the price derivative. Violations to this assumption are likely because what motivated most researchers to adopt the DiD design in the first place was that the treatment sites were selected. For example, planners might only site plants in neighborhoods where pollution would least affect house prices. In this case, the derivative of ATTs estimated by DiD corresponds to the derivative of the hedonic price function plus an additional selection bias term coming from the fact that treatment effects may be heterogeneous by distance to a treatment site.
As a solution to the identification challenge, I show that bounds on homeowner WTP can be derived under an alternative restriction that effects are concave in distance from the treatment site—a natural assumption in many applications with spatial decay. These bounds can be calculated from simple transformations of DiD estimates.
This project's framework has concrete implications for a large number of empirical settings. I illustrate the econometric and modeling framework by revisiting three applications from the literature: toxic industrial plants (TRI), Low Income Housing Tax Credit developments (LIHTC), and sex offender move ins. In my leading empirical application, for example, I estimate price effects based on millions of house sales surrounding the openings of more than 11,000 industrial plants flagged by the EPA as emitting chemical pollutants. Based on these price impacts, I estimate nearby homeowners would be willing to pay between $13.1 and $19.5 million dollars to avoid a polluting industrial plant.

Author

• Sean Elliott McCulloch, Brown University
  Presenting Author