Ulric B. and Evelyn L. Bray Social Sciences Seminar

Abstract: Learning about the impact of localized events taking place over time is a multi-faceted problem, as it requires taking into account the influence of multiple dimensions, including: geographical location, timing, and attributes of events. In this paper, I argue that traditional regression approaches---which assume the existence of a linear, or otherwise known, relationship between predictors and outcomes---are inappropriate for learning about the impact of spatial and temporal proximity to events. Instead, I propose using regression trees, an approach that allows addressing the problem in a non-parametric and efficient manner. I illustrate the usefulness of the proposed procedure by studying the impact of mass shootings on opinions about gun control and of distance to border crossings on support for immigration reform.