Model Specification and Spatial Interdependence∗

Researchers now regularly estimate spatial models in applied political science, both to enhance the validity of their “direct” (i.e., non-spatial) covariate effect estimates and to test explicitly spatial theories. While this is a welcome advance over past practices, we worry that much of this first generation of applied spatial research overlooks certain aspects of spatial models. In particular, while different theories imply different spatial-model specifications, statistical tests frequently have power against incorrect alternatives. As a consequence, researchers who fail to discriminate explicitly between the different manifestations of spatial association in their outcomes are likely to erroneously find support for their theoretically preferred spatial process (e.g., diffusion) even where an alternative process instead underlies the association. To help researchers avoid these pitfalls, we elaborate the alternative theoretical processes that give rise to a taxonomy of spatial models, and indicate why, and provide evidence that, these alternative processes are frequently mistaken for one another during conventional hypothesis testing, and suggest a set of strategies for effectively discriminating between various models. We illustrate the utility of this strategy with an application on the relationship between development and democracy.