A place-based method of evaluation and spatial units of analysis were used to measure the extent to which allocating police resources to high-rick areas, derived from risk terrain modeling (RTM), affects the frequency and spatial distribution of new crime events. This quasi-experimental project had two primary goals: 1) to replicate and validate RTM in multiple jurisdictions and across many different crime types; and 2) to evaluate intervention strategies targeted at high-risk micro-level environments across 5 cities: Chicago, IL; Colorado Springs, CO; Glendale, AZ; Kansas City, MO; and Newark, NJ.
In completing the risk terrain models (RTM), the RTMDx Utility, developed by the Rutgers Center on Public Security, was used. The Utility applied a precise set of statistical tests to evaluate the relative importance of spatial factors in influencing crime outcomes. The Utility begins by building an elastic net penalized regression model assuming a Poisson distribution of events. It does this using cross-validation. The Utility then further simplifies the model in subsequent steps via a bidirectional step wise regression process (Poisson and negative binomial) and measures the Bayesian Information Criteria (BIC) score. The best model with the lowest BIC score between Poisson and negative binomial distributions is selected.
RTMDx outputs are tabular and cartographic; for each significant risk factor, tabular outputs include a relative risk value (RRV), which is the exponentiated factor coefficient (i.e., relative weight), and the optimal operationalization and distal extent of spatial influence. A risk terrain map is also produced to show highest risk places throughout the study area.
Following the RTM analysis in each city, each Police Department developed an intervention strategy that targeted the spatial influences of select significant risk factors. The Police Department also worked with the research team in the selection of target areas for the intervention. In evaluating the intervention, statistical comparisons were made to equivalent control areas locally within each city. Control areas were matched to treatment areas through Propensity Score Matching.
A census of all elements of each city was used for this project. No sampling was done for this project.
Longitudinal: Cohort / Event-based
City-level crime incidents
Chicago Police Department (Chicago, IL)
Glendale Police Department (Glendale, AZ)
Newark Police Department (Newark, NJ)
Colorado Springs Police Department (Colorado Springs, CO)
Arlington Police Department (Arlington, TX)
Kansas City Police Department (Kansas City, MO)
geographic information system (GIS) data
Variables used in analysis were X and Y coordinates of risk factor locations including foreclosures, parks, bars, and liquor stores as well as coordinate locations of city level crime occurrences, including car theft, shootings, aggravated violence, and drug related crime.