Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/280567 
Year of Publication: 
2017
Series/Report no.: 
AEI Economics Working Paper No. 2017-14
Publisher: 
American Enterprise Institute (AEI), Washington, DC
Abstract: 
The highly influential broken windows theory is based on the notion that homelessness and other forms of disorder signal to potential criminals that the police are not monitoring a particular area, leading to more crime. In this paper, we take two approaches to estimating the contemporaneous impact of visible homelessness on crime, using data on the exact times and locations of over 75,000 reports of homeless individuals and 3.3 million crime reports in New York City. First, we use a difference-in-differences approach to compare changes in crime in close proximity to a homeless report to areas centered around nearby homeless reports that occurred in the near past or future but outside of the experimental period. Compared to the same time period the day before and after, we find that within two hours of the homeless report, violent crime falls by 27 percent within 100 feet and increases by 32 percent between 100 and 300 feet of the report. These findings are consistent with homeless individuals deterring crime through a witness effect in very close proximities but increasing crime in less immediate vicinities by signaling a lack of police presence. Meanwhile, misdemeanor assault rises in both regions, but especially within 100 feet, consistent with homeless individuals themselves being victims or perpetrators. Property crime is unaffected. In our second approach, we consider a natural experiment based on nighttime rainfall. Nighttime rainfall reduces homeless encampment reports the following day but does not affect most other non-homeless related city reports. Based on our instrumental variables estimates, the elasticity of citywide misdemeanor assaults with respect to homelessness is 0.08 and marginally statistically significant, and the elasticity of violent crime is 0.09 but not statistically significant. Elasticities for other crime types are smaller and not statistically different from zero.
Document Type: 
Working Paper

Files in This Item:
File
Size





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.