The Limits of Housing Investment as a Neighborhood Revitalization Tool: Crime in New York City

Abstract

Problem, research strategy, and findings: Local residents often oppose place-based affordable housing on the grounds that such housing will increase crime and decrease property values. New York City has actually used affordable housing investment as a neighborhood revitalization tool, leading to a positive impact on neighborhood property values. Households in distressed neighborhoods consistently cite crime as a problem, but we know little about the impact of housing investments on crime. Using a unique set of point-specific data on affordable housing and crime locations between 2002 and 2008 in New York City, I estimate a set of regression models to identify the effect that affordable housing investments have on crime on the block where they are situated. I find little evidence that affordable housing investments either reduce or increase crime on New York City blocks, suggesting there are limits to the revitalization effects of these subsidies and that crime fears about subsidized housing are unwarranted.

Takeaway for practice: Cities with tight rental markets such as New York should continue to invest in affordable housing construction. However, these cities need to find ways to expand housing options in higher-income, less-distressed neighborhoods, or they risk exacerbating concentrated poverty and further subjecting low-income households to unsafe living environments.

Keywords:

• affordable housing
• crime
• New York City
• housing subsidies
• neighborhood revitalization

Acknowledgments

I would like to thank the NYU Furman Center for data and research support, the New York City Police Department for providing crime data, and the U.S. Department of Housing and Urban Development and NYU for generous funding. I also thank Ingrid Gould Ellen, Katherine O’Regan, Johanna Lacoe, Lance Freeman, Sewin Chan, Patrick Sharkey, John MacDonald, Ed Olsen, Elyzabeth Gaumer, and anonymous referees for important comments on this paper. I also thank seminar participants at the Association of Public Policy and Management annual conference and USC Lusk Center Seminar for insightful feedback. Finally, I thank Alan Biller for research assistance. All errors are my own.

Notes

1. Negative binomial models are the ideal choice (rather than a more common model such as ordinary least squares [OLS]), given the distribution of the crime variable, which is strongly skewed toward zero. For more on the empirical strategy, see the online supplemental data (Technical Appendix).

2. Although the coefficients are not significant, the model fit (which is normally presented using R 2, but there is not an ideal version for negative binomial models) is good. All of the models have likelihood ratios well above the critical chi-square values, allowing us to reject the null hypothesis that the model is not explaining the variance in crime.