Can a measurement error perspective improve estimation in neighborhood effects research? A hierarchical Bayesian methodology

Abstract

AbstractObjectiveNeighborhood effects research often employs aggregate data at small geographic areas to understand neighborhood processes. This article investigates whether empirical applications of neighborhood effects research benefit from a measurement error perspective.MethodsThe article situates neighborhood effects research in a measurement error framework and then details a Bayesian methodology capable of addressing measurement concerns. We compare the proposed model to conventional linear models on crime data from Detroit, Michigan, as well as two simulated examples that closely mirror the sampling process.ResultsThe Detroit data example shows that the proposed model makes substantial differences to parameters of interest and reduces the mean squared error. The simulations confirm the benefit of the proposed model, regularly recovering parameters and conveying uncertainty where conventional linear models fail.ConclusionA measurement error perspective can improve estimation for data aggregated at small geographic areas.