Memphis Murder Revisited: Do Housing Voucher Households Cause Crime?
The August 2008 Atlantic Monthly article “American Murder Mystery,” by Hanna Rosin, brought considerable popular attention to housing mobility programs when it implicated the Housing Choice Voucher Program (HCV) as the cause of rising crime rates in U.S. suburbs. Through anecdotes from local law enforcement officials and residents, interviews with criminologists and housing policy experts, and published research, Rosin made the case that increased crime in suburban Memphis neighborhoods and other suburbs throughout the United States was caused by the relocation of residents through the HCV program. Rosin offers anecdotal evidence to make her case, but little scientific evidence exists in the article, or in the larger body of social science scholarship, to support her claim of a causal relationship between the relocation and subsequent concentration of HCV participants and an increased incidence of neighborhood crime.
On September 19, 2011, HUD’s Office of Policy Development and Research (PD&R) used a portion of its Quarterly Briefing to discuss new evidence of the relationship between HCV participants and neighborhood crime rates that begins to dispel the claims in Rosin’s article. Ingrid Gould Ellen from the Robert F. Wagner Graduate School of Public Service at New York University presented the findings of the HUD-sponsored study, “Memphis Murder Revisited: Do Housing Vouchers Cause Crime?” which directly examined the relationship between the concentration of HCV program participants and neighborhood crime.
Given the claims in Rosin’s article, neighborhood crime would be expected to increase as the number of residents using housing vouchers in the neighborhood increased. Ellen, along with Michael Lens and Katherine O’Regan, used data from HUD’s HCV program along with crime data from 10 U.S. cities aggregated by census tract to develop a regression model that formed the empirical basis of the study. Multivariate regression is a common statistical modeling tool used widely in social science research to understand the relationship between a dependent variable—in this case, crime—and a number of independent variables.
Although the initial model result showed a positive correlation between an increase in housing choice voucher residents and neighborhood crime rates, it does not support the causal effect suggested by Rosin’s article when variables used to approximate time and other factors are included. This finding means that, at a given point in time (for example, the year 2001), there may appear to be a positive relationship between the two variables of interest, but when you control for the effect of general crime trends—by including crime data from 2000 and citywide crime statistics as independent variables in the model— the impact of HCV households on neighborhood crime rates is significantly reduced. Moreover, the research suggests evidence of reverse causality: HCV residents are relocating to neighborhoods where crime is already on the rise, but they themselves are not the cause.
The presentation by Ellen spurred a lively discussion among the panel about policy implications and the need for more research and evidence to further inform decisionmaking. Given the evidence presented by Ellen, there is reason to believe that neighborhoods with rising crime provide lower barriers to entry for HCV participants and do not lead to desirable neighborhood outcomes. Supporting this claim, Ron Wilson of PD&R noted that most HCV participants relocate to neighborhoods near their previous residence. Among the very preliminary policy ramifications discussed by the panel, Marina Myhre of PD&R suggested that local housing authorities need to work with HCV participants to expand housing opportunities in neighborhoods where crime is low and adequate services exist for low-income residents. This could potentially be achieved through better information systems—for instance, enabling local PHAs with geographic information systems grounded in rich local datasets—and coordinated outreach to landlords regarding HCV program requirements to build the supply of quality rental opportunities in low crime neighborhoods.
Much of the discussion among the panel and the audience focused on the need for more research to better understand the relationships Ellen and her colleagues explored. For example, aggregating crime and housing voucher data at smaller geographic areas would allow researchers to examine the effect of voucher holders at a smaller spatial unit; given that the relationship between housing voucher residents and crime is inherently spatial, this is particularly important and would likely lead to more robust findings. Examining the relationship within suburbs — Ellen’s research included only urban areas — would be useful in understanding the impact HCV participants have on neighborhood crime across different urban structures. Furthermore, research that examines subpopulations within the HCV program would be beneficial in understanding to what degree, if any, subgroups of HCV participants affect neighborhood crime.
Although the evidence presented by Ellen at the Quarterly Briefing is preliminary, it does begin to refute the most inflammatory elements of Rosin’s article and discredit some of the common perceptions of HCV residents. The study also calls attention to the broader commitment at HUD to ground policy decisions in evidence-based empirical research, which further advances HUD’s mission.