Assessing the Link Between Foreclosure and Crime Rates: A Multi-level Analysis of Neighborhoods Across 29 Large United States Cities, 2007-2009 (ICPSR 34570)

Time Period(s) View help for Time Period(s)

Date of Collection View help for Date of Collection

Data Collection Notes View help for Data Collection Notes

Study Purpose View help for Study Purpose

The primary purpose of this project was to evaluate the possible link between foreclosure and crime in America. The project addressed three specific questions: (1) Are levels of foreclosure significantly associated with crime rates across neighborhoods after controlling for other factors?; (2) Is any observed effect of foreclosure on neighborhood crime rates contingent on (i.e., moderated by) other neighborhood conditions, including pre-existing structural disadvantage, pre-existing vacancy rates, or racial and ethnic context?; and (3) Does the effect of foreclosure rates on neighborhood crime levels vary across cities in systematic ways?

Study Design View help for Study Design

The research questions for this study were addressed by integrating census tract-level data on crime rates gathered from local police agencies with foreclosure data from RealtyTrac, and a wide variety of social, economic, and demographic control variables from multiple sources. The general strategy of the researchers was to regress 2009 neighborhood crime rates on foreclosure rates measured for 2007-2008, while accounting for 2007 crime rates and other control variables. The bulk of the latter were drawn from the sole source of data on contemporary social, economic, and demographic context for American neighborhoods - the American Community Survey (ACS) pooled (2005-2009) census tract file - which was treated as reflective of conditions present at approximately the mid-point of the period covered in these data (i.e., 2007) as this was when foreclosure rates exhibited particularly notable spikes in most areas of the country.

To facilitate a meaningful assessment of cross-city variability in neighborhood crime patterns, researchers specifically designed their effort to yield a sample of approximately 50 cities. Researchers chose to focus the project on relatively large cities, defined as those with 100,000 or more persons based on estimates drawn from the 2005-2007 ACS; there were approximately 270 U.S. cities with populations of 100,000 during the middle of the decade. Since 40 percent of United States cities with populations of 100,000 or more were located in just three states (California, Texas, and Florida), researchers defined the sampling frame in two stages. The first targeted large cities from the 50 most populous metropolitan areas and the other targeted cities from other metropolitan areas.

Sample View help for Sample

The study focused on relatively large cities, defined as those with 100,000 or more persons based on estimates drawn from the 2005-2007 American Community Survey (ACS). To facilitate broader regional and state coverage, the sampling frame was defined in two stages. Both stages focused on the selection of large cities (those with 100,000 or more persons), but the first targeted cities from the 50 most populous metropolitan areas and the other targeted cities from other metropolitan areas.

In the first stage of sampling, researchers selected at least one large city (populations greater than 100,000) from each of the largest 50 metropolitan areas. For metropolitan areas with more than one such city, researchers chose one randomly. In cases where data were not provided by a selected city or were provided in a form that could not be meaningfully integrated with data from other cities (e.g., counts of crime within locally defined beats or neighborhoods that have important local value, but are not comparable to "neighborhood" definitions that could be applied across multiple cities), researchers randomly selected a replacement city. Overall, researchers requested data from 80 cities within the largest 50 metropolitan areas, obtaining data in some fashion from 58 of them, and data that could be meaningfully integrated with other cities (i.e., data that could be aggregated to census tracts) from 50 cities. Of the 80 cities included in the initial sampling frame, 62 had participated in the National Neighborhood Crime Survey (NNCS) [ICPSR 27501 (National Neighborhood Crime Study (NNCS), 2000], which encompassed neighborhood data on crime and other conditions at the beginning of the 2000s. Researchers wished to expand the breadth of the sample to include cities outside the largest United States metropolitan areas, while also facilitating more extensive linkages between these data and the NNCS. Thus, researchers also requested in a second stage of sampling data from the other 29 cities represented in the NNCS. In practice, these additional large cities were chosen randomly within regions from the universe of all cities with 100,000 or more persons located outside the largest 50 metropolitan areas. Researchers received data in some form or another from 20 of these cities, and data that could be meaningfully integrated with data from other jurisdictions from 17. In total, researchers requested neighborhood data in writing from 109 cities with 100,000 or more persons; researchers received data in some form or another from 78 cities, and data that could be integrated fully for 67 cities.

The sample used for the analyses was further constrained by the specific design employed and the availability of other data elements. Specifically, researchers excluded Knoxville, Tennessee, Columbus, Ohio, and Seattle, Washington because the police agencies in these cities did not provide the requisite crime data for both 2007 and 2009 (an important element of the design). According to the 2005-2009 ACS census tract file, the 64 cities included in the sample contained 7,842 census tracts that fall wholly or partly within them (based on 2009 place definitions). Researchers excluded from the analysis 295 census tracts with less than 50 persons or 50 housing units and 132 tracts for which researchers were unable to obtain data on foreclosure and other data elements. After these data exclusions, the maximum analysis sample consisted of 7,415 census tracts within 64 cities and 29 states. These 64 cities serve as the pooled sample for the analysis of foreclosure and burglary. One city - Minneapolis, MN - did not provide parallel crime counts for robbery, so that portion of the analysis was based on 7,294 census tracts within 63 cities.

Time Method View help for Time Method

Universe View help for Universe

Census tracts within 64 large cities in the United States

Unit(s) of Observation View help for Unit(s) of Observation

Data Source View help for Data Source

Data Type(s) View help for Data Type(s)

Mode of Data Collection View help for Mode of Data Collection

Description of Variables View help for Description of Variables

The data file (99 variables, n=8,198) contains the following elements:

1. Geographic codes such as FIPS, city, and state
2. Crime rates including counts of burglaries and robberies known to police for the years 2006-2009
3. Foreclosure data including raw foreclosure counts and rates for the years 2006-2009
4. Economic data including vacancy data and mortgage data
5. Population, economic, and demographic data such as total population, poverty percentage, and percentage of female-headed households
6. City level measures such as total population, median family income, and percentage of vacant homes
7. Unemployment rates
8. Police force counts

Response Rates View help for Response Rates

Not applicable.

Original Release Date View help for Original Release Date

2016-09-29

Version History View help for Version History