Spatial Modeling in Neighborhood and Crime Study: Is the Pain Worth the Gain?

• In Event: Spatial Patterns of Crime

Abstract

In the current study, we theoretically discuss why addressing spatial dependence is important and empirically demonstrate its methodological advantages in the context of neighborhood and crime study. We first conduct a Monte Carlo simulation and shows that a model without spatial dependency significantly underestimates the uncertainty of the coefficient estimates and standard errors. Moreover, we empirically demonstrated this using a real-world data in the context of neighborhood and crime study using the sample of block groups in New York City. First, we found that as the uncertainty in measuring neighbors increases, the bias in the coefficient estimates increases. However, importantly, we also observed that even with high rate of uncertainty in the spatial matrix, the bias is smaller than the non-spatial models. Likewise, as the uncertainty in defining neighbors increases, models tend to underestimate the standard errors. However, even with higher uncertainty in the spatial weight matrix, underestimation seems smaller than that of the non-spatial model. Our primary findings suggest that despite of uncertainty in defining neighbors, a kind of spatial modeling approach outperforms the non-spatial models in the context of neighborhood and crime study.

Authors

• Young-An Kim, Florida State University

• Jongwoo Jeong, Washington University in St. Louis