Geographies of urban crime: An intraurban study of crime in Nashville, Tennessee; Portland, Oregon; and Tucson, Arizona

Abstract

Understanding the context of crime is key to developing informed policy that will reduce crime in communities. In exploring criminal contexts, this dissertation tests criminal opportunity theory, which integrates social disorganization and routine activity theories. Methodologically, the dissertation presents unique ways of modeling space in crime studies. Analyses are undertaken in three cities, Nashville, TN; Portland, OR; and Tucson, AZ, chosen for their similar crime rates and varied demographic and social characteristics. This dissertation includes three papers submitted for publication. Crime data were collected for nine crimes over the period 1998-2002. Census data, used to create an array of socioeconomic measures, and land use data were also used in the analyses, presented at the census block group level. The first paper attempts to determine whether certain structural associations with violence are generalizable across urban areas. The idea is tested by first developing an Ordinary Least Squares model of crime for all three cities, then replicating the results for each city individually. The models provide support for a general relationship between violence and several structural measures, but suggest that the exploration into geographic variation of crime and its covariates both within urban areas and across urban areas should be undertaken. The second paper explores an alternative to crime rates: location quotients of crime. A comparison of location quotients and rates is provided. The location quotients are then used in a regression modeling framework to determine what influences the crime profile of a place. The results demonstrate the efficacy of simple techniques and how location quotients can be incorporated into statistical models of crime. The models provide modest support for the opportunity framework. The final paper explores possible spatial variation in crime and its covariates through a local analysis of crime using Geographically Weighted Regression (GWR). Those results are compared to the results of a 'base' global OLS model. Parameter estimate reaps confirm the results of the OLS model for the most part and also allow visual inspection of areas where specific measures have a strong influence in the model. This research highlights the importance of considering local context when modeling urban violence.