Research That Matters (January 17 - 20, 2008)

Purpose: For more than 20 years, the relationship between disorder and crime has been the focus of a contentious debate in social policy. Wilson and Kelling's (1982) seminal Broken Windows essay asserts that disorder and crime are causally linked, and that policing “the little things” is instrumental in the prevention of serious crime. However, critics of the theory raise concerns about its racial implications and its potential to undermine police legitimacy. Moreover, these critics question the theory's suggested causal links, citing difficulties both in the measurement of disorder, and in modeling its relationship with serious crime. This analysis examines the relationship between physical disorder and violent crime in 1990s New York City.

Methods: This study advances the debate on Broken Windows by carefully examining the ways disorder is measured and perceived, and identifying and controlling for the correlations among disorder, social structure and crime. Combining data on neighborhood disorder from the New York City Housing and Vacancy Survey (HVS) with police measures of violent crime, I define two measures of disorder, one based on a principal components factor analysis of conditions observed by HVS surveyors, and one based on survey responses by neighborhood residents.

I use a series of Ordinary Least Squares (OLS), fixed effects, and first difference models to assess the extent to which physical disorder in New York neighborhoods is linked to crime in those same neighborhoods, in subsequent years. I estimate a set of nonlinear models to determine whether the relationship changes based on neighborhoods' levels of disorder, and whether there is a threshold at which the more disorderly neighborhoods experience a stronger link between disorder and crime.

Results: My analysis reveals that the relationship between disorder and violent crime is considerably more complex than that described in the original Broken Windows essay. Like previous literature, I find a strong cross-sectional association between the surveyor's measure of physical disorder and violent crime, indicating that disorderly neighborhoods also tend to be high in crime, and are likely to have substantial needs for both social services and police intervention. However, my analysis also indicates that when neighborhoods' time-invariant characteristics are considered, there is no remaining relationship between this measure and violent crime. Estimates of models with year fixed effects, but without neighborhood fixed effects, and estimates of models based on the residents' reports of disorder, indicate a significant relationship between disorder and crime, but one that is increasing but concave.

Implications: These findings suggest that addressing conditions of disorder, as assessed by outside reporters, without other efforts to improve the neighborhood, is unlikely to significantly improve crime conditions. The strong tie between residents' perceptions of disorder and later crime suggest that residents may more accurately perceive threatening conditions in their neighborhoods than do outside reporters, but further research is necessary to determine whether residents' perceptions are objective predictors of violent crime, or if they, in turn, are predicted by a neighborhood's criminal history.