# Predicting Local Crime Clusters Using (Multinomial) Logistic Regression

## Article excerpt

(ProQuest: ... denotes formulae omitted.)

Introduction

The use of spatial statistical methods in criminology is becoming increasingly popular with the recognition that spatial relationships exist in crime data that complicate traditional statistical estimation. In addition, spatial statistical techniques are becoming applied more because of their ability to ask new questions to better understand the criminal event. One of these statistical techniques is explicitly spatial and considers local relationships: local crime clusters using local Moran's I. This technique and others like it are "local" because, rather than calculating a statistic tor the entire study area (a "global" statistic), it calculates a statistic tor each unit under analysis (census tracts, tor example).

Although instructive on their own, simply with their identification, a better understanding of the determinants of these local crime clusters is of interest. This article discusses a two-stage research methodology that incorporates local crime clusters and their prediction. The two-stage approach first identifies the local crime clusters in the study area and then a multinomial logistic regression to identity the predictors tor each local crime cluster type. This two-stage approach shows that the output irom a local Moran's I and multinomial logistic regression is able to identity iactors specific to each local crime cluster type beyond what would be tound in more common regression analyses.

Data and Methods

The data tor the tollowing analyses come irom Vancouver, British Columbia, Canada. Vancouver, a city on the west coast of Canada, just north of the U.S. border with Washington State, is Canada's third largest census metropolitan area (CMA), with a population that is slightly more than 2.3 million. Although Vancouver is the third largest CMA in Canada, it historically has had the highest crime rate tor CMAs in the country. In 2013, tor example, Vancouver's total crime rate (excluding traffic) was 6,897 per 100,000 residents, more than double the crime rate in the largest Canadian CMA, Toronto, at 2,941 per 100,000 (Boyce, Cotter, and Perreault, 2014).

Data and Spatial Units of Analysis

The spatial unit of analysis tor Vancouver is the dissemination area (DA). The DA, approximately the size of a census block group in the U.S. census, contains approximately 400-700 people and is composed of one or more blocks. In Vancouver, 990 DAs are used tor our analyses.

Crime data tor Vancouver consist of calls tor service made to the Vancouver Police Department tor the year 2001. The calls-tor-service database is the set of calls made to the Vancouver Police Department directly, calls allocated to them through the 911 emergency system, and calls made by police while on duty. These data consist of automotive theit (theit of vehicle and theit irom vehicle), burglary (commercial and residential), and violent crime (assault, fighting, holdups, homicide, robbery, sexual assault, and stabbing). Each call tor service includes the complaint code/ description, listed previously, and the location in the torm of an address of the call tor subsequent mapping and spatial analysis. These data were geocoded to the street network with a 94-percent success rate, exceeding the minimum acceptable hit rate of 85 percent set by Ratcliffe (2004).

The census data used in the inierential analyses represent the appropriate years of crime data tor Vancouver (2001) at the dissemination area level. These explanatory variables were chosen, in both cases, to represent social disorganization theory and routine activity theory (Cohen and Felson, 1979; Shaw and McKay, 1942). For the Vancouver analyses, the iollowing variables were employed: population change, percent; males ages 15 to 24, percent; single-parent tamilies, percent; recent immigrants, percent; ethnic diversity, measured using the Blau Index; unemployment, percent; post-secondary completion, percent; average income in thousands of dollars; population density; average dwelling value in thousands of dollars; rentals, percent; and housing in major repair, percent. …

Search by...
Show...

### Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.