The neighborhood context of foreclosures and crime

Abstract

The epidemic of foreclosures across the United States in the mid-2000 decade offers a unique opportunity to examine neighborhood changes that result over extended periods of time. In this analysis, a series of structural and contextual factors are modeled to determine whether concentrated foreclosures correspond to later crime increases that might lead to long-term neighborhood problems. The results first revealed a conflict between Census tract and block group data that lead to a discussion of the Simpson’s Paradox as it relates to the selection of the appropriate geographic scale for further analysis. Using spatial analysis techniques it was determined that the block group (micro-level) results are more accurate for interpretation. With the block group results, a positive and complex relationship between foreclosure and crime is found across various neighborhood contexts. The use of spatial analysis approaches allow for a more in-depth discussion to gain a deeper understanding of the relationship between foreclosures and crime identified in the model results.

Keywords:

• foreclosures
• crime
• neighborhood effects
• geographic scale comparison
• Simpson’s Paradox

Acknowledgments

The author would like to thank Brian Lawton of George Mason University, Elizabeth Groff of Temple University, Mark Shroder of the U.S. Department of Housing and Urban Development, and the blind reviewers for providing comments to refine and improve this paper. The author would also like to thank Michael Bess from the Charlotte-Mecklenburg Police Department for providing insight into trends and findings that led to a better understanding of the geography of the city and county. Finally, the author would like to thank Brandon Behlendorf of the University of Maryland for his contributions to shaping this paper and help in modeling the data.

Notes

1. Foreclosure is the process by which a lender relinquishes an owner’s claim to property from a loan default (Giliberto and Houston, 1989). After a mandatory waiting period the lender repossess the property and the property is vacated, leaving the house exposed to vandalism and theft of valuable building materials.

2. Frame (2010) provides an overview of several of these studies.

3. Residential burglary can be a subset of property crimes, but is a crime type that can inflict severe duress on the victim. This crime can be so concerning to the victim because it violates their private space and can make them feel as though they were personally accosted. This finding has criminologists debating whether it is a property crime of a form of violence. Research has shown that when a household or neighboring households are burglarized it prompts residents to move or taking serious protective actions. Also, residential burglary represents a graduation stage of offences at the nexus of property and violent crime because it is moving into the realm of violence by having the effect outlined above, but still only results in the theft or damage of property. Residential burglary was tested alone for an effect, but was still part of leaving property crimes to represent only those incidents that did not involve breaking into and entering an occupied house.

4. Crime data were supplied by the Charlotte-Mecklenburg Police Department, followed the UCR classifications, and were geocoded to the specific street location of occurrence.

5. There are several pre-foreclosure processes that can prevent vacancy while a house is in the foreclosure process. Four common voluntary foreclosure vacancy mechanisms are: 1) arm’s length sale; 2) short sale; 3) Chapter 7 bankruptcy; and 4) a ‘cash for keys’ program where the lender gives the borrower cash to assist in relocation.

6. Parcel data were provided by the Geography Department at the University of North Carolina, Charlotte.

7. Concentrated disadvantage is measured as a cross-section. It represents a contextual moment in time of what the socioeconomic conditions are in a geographic unit. Given the short time span of the analysis, a cross-sectional is a good representation of contextual conditions in which the socioeconomic factors would not have changed much.

8. Previous studies examining neighborhood effects on crime have noted the problem of multicollinearity, especially the high correlation among factors regarding concentrated disadvantage (Morenoff, Sampson, and Raudenbush 2001). To ensure the models did not suffer from multicollinearity, a factor analysis was run with varimax rotation using maximum likelihood estimation to calculate factor loadings for these two latent factors. Analyses were also conducted using principal components analysis with a varimax rotation and Kasier normalized, but the results were not significantly different.

9. Building grade refers to the classification of a property based on amenities and materials used to construct the house. Classifications such as ‘good’ or ‘better’ refer to houses that have, for example, higher grade siding or heating infrastructure than those of ‘average’ build grade. As well, grade refers to quality of construction in which grades of ‘good’ or ‘better’ refer exceeding building codes or better designs.

10. It is important to caution against strict causal analysis of the results here, since the relationship between crimes and indicators of neighborhood composition may be reciprocal (Kubrin and Weitzer 2003). If so, then there exists the possibility of simultaneity bias primarily because no controls are provided for the effects of crimes on single-family foreclosures or any of the other structural covariates.

11. In order to account for the variation based on areal unit, all the regression models will be run at both the tract level and the block level to determine stability of factors across geographical differences. In addition, a sufficient amount of research on neighborhood effects uses a cross-sectional analysis in a single-year period or a pool of year periods. Rarely are models tested for subsequent years to determine the stability of effects year to year. All of the control variables are held constant between the 2 years, while the foreclosure rates, spatial lags, and 2-year lags of crime rates vary.

12. The above model uses place holders, for example [CONCENTRATED DISADVANTAGE], to denote all the factors of associated with that grouping within the model (refer to Table 1 for a listing of factors and their groups).

13. Although these factors conflict, their bivariate correlations are less than 0.5 and variance inflation factors do not detect multicollinearity as a problem in this model.

14. An in-depth technical discussion of MAUP can be found in Openshaw (1994).

15. For more details about how the Simpson’s Paradox occurs with the results from this analysis, see Wilson (2013).

16. The mean house price was $116,581 with a standard deviation of $53,137 meaning that with flexible loan packages lower income populations had the opportunity to obtain a house than in overinflated markets, which even low-income populations would unlikely concentrate in other neighborhoods.