Identifying spatial interactions in the presence of spatial error autocorrelation: An application to land use spillovers

Abstract

Spatial spillovers—interaction effects among neighboring agents in space—are a common characteristic of a variety of processes that are of interest to environmental and resource economists. Empirical identification of these interactions is challenging, however, due to the endogenous nature of the interactions and the inevitable unobserved spatial correlation that, if uncontrolled, can result in spurious estimates of the interaction parameters. Traditional spatial econometric models rely on maintained assumptions that impose separate structures for the spatial error and interaction processes and thus are insufficient for solving this identification problem. To identify spatial land use spillovers in a hedonic model of residential housing values, we pursue an alternative approach by exploiting a natural experiment in the data. We use exogenous physical land features that impose a direct constraint on residential development on some, but not all, of the land that falls within our study region and use this to construct a “partial population identifier.” We find that this estimation strategy solves the endogeneity problem and reduces spatial error autocorrelation, but does not fully eliminate it. Estimation of the model using a more restricted sample in combination with the partial population identification strategy is successful in eliminating the remaining spatial error autocorrelation. We conclude that less restrictive approaches to controlling for unobserved spatial correlation, such as the natural experiment pursued here, may provide a superior alternative to identifying spatial spillovers.