Accounting for Micro- or Meso-level Effects When Building Models using City-level Crime Data: Introducing an Imputation Technique

• In Event: Spatial Criminology: Consequences for Failing to Consider the Spatial Scale when Modeling the Location of Crime

Abstract

Although the need to account for varying geographic levels for micro- and meso-level analyses is well-known, less attention has been paid to this issue for macro-level studies. Instead, macro-level studies of crime levels across cities typically ignore possible neighborhood-level processes when estimating models. Ideally, we would have data at all geographic levels when estimating models. But such data are extremely difficult to come by, raising the question of whether there are reasonable alternative strategies? We propose one that uses information on neighborhoods (or smaller units) in a city to impute crime in those smaller units. The model is then estimated on this synthetic data. We demonstrate this approach using National Neighborhood Crime Study (NNCS) data (in which the “true” model is known), and show that it performs reasonably well. Importantly, we show that results from macro studies ignoring these smaller scale patterns can in some cases be quite biased.

Authors

• John Hipp, University of California, Irvine

• Seth A. Williams, University of California, Irvine