Criminology researchers study the explanations of crime at different geographic scales, from counties down to street blocks. Whereas many studies have asked why certain cities or counties have more crime than others, a limitation of this research is that it ignores what is going on in neighborhoods or micro locations within these cities. We propose a unique solution to this problem by utilizing insights from existing literature on which neighborhoods or micro locations tend to have more crime, and incorporate this information to impute crime data from the city to smaller geographic units within the city. This strategy allows researchers to estimate full multilevel models that account for sub-city level factors when comparing across cities. We demonstrate that existing literature failing to account for this can obtain considerably different (and therefore potentially problematic) results.
You can access the article by Dr. John R. Hipp and Seth A. Williams in the Journal of Quantitative Criminology entitled, “Accounting for Meso- or Micro-Level Effects When Estimating Models using City-level Crime Data: Introducing a Novel Imputation Technique”
Abstract: “Objectives: Criminological scholars have long been interested in how macro-level characteristics of cities, counties, or metropolitan areas are related to levels of crime. The standard analytic approach in this literature aggregates constructs of interest, including crime rates, to the macro geographic units and estimates regression models, but this strategy ignores possible sub-city-level processes that occur simultaneously.
Methods: One solution uses multilevel data of crime in meso-level units within a large number of cities; however, such data is very difficult and time intensive to collect. We propose an alternative approach which utilizes insights from existing literature on meso-level processes along with meso-level socio-demographic measures in cities to impute crime data from the city to the smaller geographic units. This strategy allows researchers to estimate full multilevel models that estimate the effects of macro-level processes while controlling for sub-city level factors.
Results: We demonstrate that the strategy works as expected on a sample of 91 cities with meso-level data, and also works well when estimating the multilevel model on a sample of cities different from the imputation model, or even in a different time period.
Conclusions: The results demonstrate that existing studies aggregated to macro units can yield considerably different (and therefore potentially problematic) results when failing to account for meso-level processes.”