Abstract
The current study builds on prior research in an analysis of the relationship between monthly violent and property crime rates in New York City census tracks and the New York City Police Department’s highly contentious stop, question, and frisk (SQF) policy. We find that higher doses of SQF are associated with small crime reductions generally and specific crime reductions for stops of blacks, Hispanics, and whites. But the way the policy was implemented precludes strong causal conclusions. Now that a federal court has intervened and SQF is undergoing change, the court monitor, New York Police Department, and city officials should partner with researchers in experimental evaluations to determine the optimal mix and dosage of enforcement strategies that safeguard the rights and liberties of citizens while enhancing public safety.
Keywords:
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 The figures are from the New York Civil Liberties Union (http://www.nyclu.org/content/stop-and-frisk-data) taken, in turn, from the NYPD’s SQF website (http://www.nyc.gov/html/nypd/html/analysis_and_planning/stop_question_and_frisk_report.shtml).
2 In their 2007 study, Smith & Purtell also assessed SQF effects on crime in impact zones.
3 Hereafter, Non-Hispanic blacks and whites are referred to as “blacks” and “whites.”
4 The crime data were provided by the NYPD. Rape is excluded because sexual violence is not well measured in police data. The SQF data are from the NYPD’s SQF database (http://www.nyc.gov/html/nypd/html/analysis_and_planning/stop_question_and_frisk_report.shtml). For measurement reliability, analyses are restricted to census tracts with 100 or more inhabitants.
5 This decision was based on discussions with an NYPD crime analyst.
6 The Census defines linguistically isolated households as those in which “no member 14 years old and over (1) speaks only English or (2) speaks a non-English language and speaks English ‘very well’” (http://www.census.gov/hhes/socdemo/language/about/faqs.html).
7 Orthogonal rotation forces the correlations among the factors to zero. To assess the sensitivity of the results to this assumption, we also performed a factor analysis with oblique rotation, which allows the factors to be correlated. The results are virtually identical to those reported.
8 Log transformation and the addition of 1.0, or some other positive constant, to a measure before taking the log can induce unwanted changes to the original distribution. We checked the resulting distributions of the log-transformed measures in our analysis and found that the transformations had the intended effect: the influence of extreme values was reduced and the zeros were retained in the transformed measures (ln(1) = 0), without otherwise distorting the distributions of the original measures. Results available on request from the lead author.
9 The correlation (r) between the total crime rate and the SQF rate in census tracts is .679.
10 An “instrument” is a variable that, when controlled in a regression equation, reduces suspected endogeneity of one or more of the predictors. The logic behind using past values of the outcome and predictors as instruments is that they cannot be influenced by current or more recent values of the outcome, thereby increasing confidence that significant effects of the predictors represent causal influences on the outcome. As discussed below, however, strong causal inferences cannot be drawn from observational data of the kind used in this study.
11 Because a constant was added to the original measures before taking the log, these impact estimates should be treated as approximations.
12 The mean monthly stop rate is 6.900 stops per 1,000 tract residents. The mean monthly violent and property crime rates are .484 and .908 crimes per 1,000 tract residents, respectively. The mean number of residents per tract is 3,884 (see Table ). Therefore, the mean number of stops per month is 26.800 (6.900 × 3.884), the mean number of violent crimes is 1.880 (.484 × 3.884), and the mean number of property crimes is 3.527 (.908 × 3.884). A 10% increase in the mean number of stops per month would yield 2.680 additional stops. Applying the summed coefficients reported in the text and Table , an increase of 2.680 stops would result in .039 (1.880 × .021) fewer violent crimes and .131 (3.527 × .037) fewer property crimes per month.
13 We thank an anonymous reviewer for both suggestions.
Additional information
Notes on contributors
Richard Rosenfeld
Richard Rosenfeld is the Thomas Jefferson Professor of Criminology and Criminal Justice at the University of Missouri—St. Louis. His current research focuses on the factors associated with change over time in crime rates.
Robert Fornango
Robert Fornango received his PhD in criminology and criminal justice from the University of Missouri—St. Louis. He directs the Informatics Research program at the Health Services Advisory Group in Phoenix, Arizona.