Public Safety Lab
Bringing data science to public safety
WHAT WE DO
The Public Safety Lab uses the tools of data science and social science to support communities’ efforts to improve public safety outcomes.
For example, many relatively minor offenses are easy to detect and interdict. Many more serious offenses are much harder to detect and interdict. We may then overpunish relatively minor offenses, and underpunish more serious offenses. The Public Safety Lab works with communities and law enforcement agencies to achieve more productive allocations of public safety resources.
Communities and agencies interested in working with the Public Safety Lab can contact us at publicsafetylab@nyu.edu.
WHO WE ARE
The Public Safety Lab coordinates the work of researchers across multiple disciplines and universities.
Public Safety Lab Data Science Team
Vidhi Patel
Chief Data Scientist
Public Safety Lab
New York University
Public Safety Lab Doctoral Students
Jail Data initiative
The problem of overincarceration has received considerable attention of late. The focus of this attention, however, has largely been on state and federal prison systems. Often overlooked are local jails. In any given year, approximately 11 million Americans will be detained in a county jail, often because they lack the funds to post bail, or even because they lack the funds to pay their fines. Yet we know little about county jail practices, including which counties are systematically jailing defendants pre-trial and/or post-fine, and about the consequences of these practices.
It is possible that longer periods of pre-trial detention, or detainees’ delayed access to counsel, actually increase recidivism, and/or escalate petty offending into more serious offending. Through the generous support of Arnold Ventures, our data science team is crawling daily county jail rosters and criminal case records in over 1,000 counties. Using these data we will be able both to identify counties that are systematically jailing defendants pre-trial and/or post-fine, and to estimate the impacts of these practices on recidivism. We will make all our code and data publicly available to the policy and research communities.
The Jail Data Initiative is a founding project of Arnold Ventures’ $39 million National Partnership for Pretrial Justice.
Media Coverage: Forbes, Inside Philanthropy, Philanthropy News Digest, The Crime Report
prosecutorial reform initiative
Prosecutors play a vital role in the criminal justice system. After an arrest, prosecutors have numerous choices to make regarding whether and how to pursue prosecution of an offense. In many jurisdictions, prosecutors use this discretion to pursue prosecutions of subfelony violations to the full extent of the criminal law. In these jurisdictions, cases involving subfelony offenses typically consume a large share of law enforcement, prosecutorial, and judicial resources.
Yet it is unclear whether the aggressive prosecution of subfelony offenses actually reduces offender recidivism; the practice may actually result in increased recidivism, and/or the escalation of minor offending into more serious offending. The practice may also exacerbate existing inequalities in criminal justice outcomes. Some prosecutors’ offices, however, are pursuing reforms that implement alternatives to prosecution for subfelony offenses.
Working with Boston’s Suffolk County District Attorney’s Office, the St. Louis County Prosecuting Attorney’s Office, and the New York County District Attorney’s Office, and in partnership with the Justice Tech Lab at Texas A&M University and the Legal Innovation and Technology Lab at Suffolk University, we are investigating the impacts of these reforms on offender recidivism, and designing and experimentally testing a decision assist tool to support prosecutorial decisions that reduce reoffending.
CUrrent Projects
Identifying Sex Trafficking
Sex trafficking is a crime of extraordinary violence. Its frequency is likely severely underreported, given the isolation within which its victims are kept. Working with a multi-university team, we are extracting structured content from a very large corpus of online commercial sex ads and provider reviews crawled over a period of several years, and integrating this content with law enforcement records on sex trafficking cases. We are then looking for features in the ad and review data that are predictive of the presence of sex trafficking, and that may assist law enforcement agencies in identifying newly posted ads and reviews associated with trafficking. We are particularly focusing on urban school district calendars, which typically include days when high school-aged children are not in school, but adults (including teachers) are at work. Initial results suggest that school calendars may be a good predictor of the incidence of sex trafficking, highlighting for agencies the importance of investigating online sex ads posted on days when high school-aged children are not in school, but adults are at work.
Criminal Appeals Equity Project
Existing work has indicated that competitive partisan elections appear to induce more anti-defendant rulings, relative to other judicial retention institutions. This work has led many to advocate against the use of partisan elections for judicial selection and retention. Yet existing work has been constrained to cross-judge estimates, limiting the ability to draw causal inferences. In New York State, intermediate appellate judges must be elected, and re-elected, in contested partisan elections in order to be eligible for gubernatorial appointment, and reappointment, as Appellate Division judges. Using content extracted from the crawled text corpus of the approximately 38,000 slip opinions in criminal appeals heard by New York State's intermediate appellate courts between 2003-2017, and defendant demographic and crime data crawled and extracted from the New York State Department of Corrections' inmate database, we find that impending reappointment induces more rulings against black defendants, but not against nonblack defendants. We find no additional effect of impending reelection on appellate judge votes in criminal appeals.
911 Response Analytics
Calls to 911 in large urban jurisdictions are typically first assigned to a responding service (e.g., police, medical, fire), and are then assigned priority codes determining the speed and nature of the service's response. These human-assigned call and priority codes may not be optimal. For example, calls involving mental health and substance abuse issues may be assigned to policing agencies when they might more appropriately be assigned to medical professionals. Calls involving behavioral health issues assigned to policing agencies may or may not be assigned codes providing for response by officers trained in crisis intervention. Call taker biases may affect the assignment of call and priority codes. Working with a large urban policing agency, we are exploiting the as-if random assignment of 911 calls to call takers to evaluate the causal effects of call taker discretion on the assignment of call and priority codes, and subsequently on call outcomes, and to design call response protocols that may reduce the risks of escalated encounters.
911 Response Field Experiment
Many policing agencies seek to improve their relationships with the communities they police, yet struggle to find effective means to do so. Working with a large urban policing agency, we are developing a randomized controlled trial of a platform that pushes an SMS-based survey to recent 911 callers, asking them to provide feedback on how they were treated by responding officers. Responding patrol officers may then access their average survey ratings, as well as those of the other officers in their unit, through a web-based interface. The platform may both incentivize responding officers to treat 911 callers with greater professionalism, and increase caller satisfaction with the police response to their call. More generally, the platform may provide an effective means for agencies to receive real time feedback from the communities they police.
Detecting Domestic Violence
Domestic violence is both distressingly common and significantly underreported to law enforcement authorities. Its widespread presence is costly to its victims, to their children, and to society more generally. Yet there may be strategies to assist both law enforcement and social service agencies in detecting and addressing the presence of unreported domestic violence. For example, we know that the timing of reported domestic violence responds to the timing of the distribution of social benefits. In partnership with NYU Wagner’s Policies For Action Health Hub, and leveraging the randomly assigned timing of the distribution of social benefits to individual households, we are working with New York State’s Medicaid data to identify female and child injuries that may be due to unreported domestic violence. We are then looking for features that may help to explain both the incidence of domestic violence, and its relative underreporting.
Policing for Revenue
In recent years numerous observers have raised concerns about “policing for profit,” or the deployment of law enforcement resources to raise funds for cash-strapped jurisdictions. However, identifying the causal effect of fiscal incentives on law enforcement behavior has remained elusive. We leverage a discontinuity in the rules allocating fine revenue from traffic citations issued by a large highway patrol agency, finding that the frequency and severity of traffic accidents increase sharply just above the threshold reducing the share of fine revenue captured by the agency. We also find that cited drivers in towns just below this threshold are given fewer days to pay their fines and are less likely to pay their fines on time, leading to higher risks of late fees and license suspension. These findings suggest that fiscal concerns can in fact impact public safety decisions.
Diversifying Policing
Recent events have directed renewed attention to the question of whether increasing the proportion of nonwhite officers in law enforcement agencies would lead to different public safety outcomes. Increasing the proportion of nonwhite police officers may lead to an increased acceptance of policing in nonwhite communities, fewer instances of the use of force, and fewer citizen complaints, leading in turn to increases in the reporting of crime and in the willingness of civilians to cooperate with police investigations. Nonwhite officers may also exert greater effort, or be better equipped, to clear crimes in largely nonwhite communities. Working with both survey and law enforcement administrative data, we are investigating the impacts of police force diversity on public safety outcomes.
Public Safety Lab
Workshop on Data-Driven Criminal Justice Reform
Friday October 19th 2018 9 AM-4:30 PM
Saturday October 20th 2018 9 AM-1 PM
Lester Pollack Colloquium Room
Furman Hall, New York University School of Law
Data science and social science may offer opportunities to support communities’ efforts to pursue criminal justice reform. For example, many relatively minor offenses are easy to detect and interdict. Many more serious offenses are much harder to detect and interdict. As a consequence we may overpunish relatively minor offenses, and underpunish more serious offenses. Data science and social science strategies may allow us to achieve more socially productive allocations of criminal justice resources.
This workshop will convene university researchers along with representatives from public agencies and nonprofit organizations to engage in a broad-ranging conversation on data-driven criminal justice reform.
Space is limited and registration is required; please rsvp here if you wish to attend. Rsvps will close when capacity is reached.
Friday, October 19
8:30-9:00: Light breakfast available
9:00-9:15: Opening remarks
9:15-10:45: Do we overpolice minor offenses?
A significant share of law enforcement effort is directed at policing relatively minor offenses, such as infractions, violations, and misdemeanors. Law enforcement personnel are often asked to respond to situations that might better be handled by medical or social service professionals. How are these resource allocation decisions made? Are they effective? Are they equitable? What consequences do they have? Are there better ways to make these decisions? What kinds of data would we need to answer these questions?
Wesley Bell, Democratic Candidate for St. Louis County Prosecutor
Jenn Rolnick Borchetta, Deputy Director, Impact Litigation, The Bronx Defenders
Captain Clay Farmer, North County Police Cooperative, St. Louis County
Ravi Shroff, Assistant Professor of Applied Statistics and Urban Informatics, New York University
11:00-12:30: Do we underpolice more serious offenses?
Many more serious offenses, such as gun violence, gun trafficking, drug trafficking, and sex trafficking, are difficult to detect and/or seemingly intractable. Recent developments in data science may offer tools to help law enforcement agencies address these more serious offenses. How do these predictive policing tools work? Are they effective? Are they equitable? What consequences do they have? What kinds of data would we need to answer these questions?
Anna Harvey, Professor of Politics and Director, Public Safety Lab, New York University
Justin McCrary, Paul J. Evanson Professor of Law, Columbia Law School
Daniel Neill, Associate Professor of Computer Science and Public Service, New York University
Chief Lori Pollock, Crime Control Strategies, New York Police Department
12:30-1:30: Lunch
1:30-3:00: Is pretrial decision making optimal?
A number of important decisions are made during the period between arrest and the disposition of a defendant’s case. For example, prosecutors make decisions over whether to pursue prosecution, the nature and count of charges brought, bail and detention recommendations, and plea offers. Judges make bail decisions, rule on pretrial motions, and approve or disapprove of plea bargains. How are these decisions made? What consequences do they have? What kinds of data-driven strategies might assist prosecutorial decision making? Do bail algorithms offer the promise of more effective bail decisions? What kinds of data would we need to answer these questions?
Sharad Goel, Assistant Professor of Management Science and Engineering, Executive Director, Stanford Computational Policy Lab, Stanford University
Scott Levy, Special Counsel to the Criminal Defense Practice, The Bronx Defenders
Jens Ludwig, McCormick Foundation Professor of Social Service Administration, Law, and Public Policy; Director, University of Chicago Crime Lab
Maria McKee, Principal Analyst, Office of San Francisco District Attorney
3:15-4:30: Access to counsel
Jurisdictions vary widely in their practices with respect to providing counsel to the indigent. What factors affect access to counsel? How does access to counsel affect outcomes? How does the quality of counsel affect outcomes? Are there data-driven strategies to promote more effective access to counsel? What kinds of data would we need in order to answer these questions?
Amanda Agan, Assistant Professor of Economics and Affiliated Professor of Criminal Justice, Rutgers University
Andrew Davies, Director of Research, New York State Office of Indigent Legal Services
Miguel de Figueiredo, Associate Professor of Law and Terry J. Tondro Research Scholar, University of Connecticut
Saturday, October 20
8:30-9:00: Light breakfast available
9:00-10:15 Incarceration and reentry
Incarceration is expensive, both in terms of its direct financial costs and in terms of the indirect costs it imposes on defendants post-release. Is incarceration an effective strategy to achieve deterrence? Is it equitable? What consequences does it have? Are there more cost-effective alternatives that can achieve both greater deterrence and greater reentry success? What kinds of data would we need to answer these questions?
Monica Deza, Assistant Professor of Economics, Hunter College
Elizabeth Glazer, Director, Mayor’s Office of Criminal Justice, New York City
Jennifer Skeem, Mack Distinguished Professor of Social Welfare and Public Policy, University of California, Berkeley
10:30-11:45 Transparency and accountability
Recent media attention directed at police shootings, instances of officer misconduct, and civilian complaints has raised questions about the availability and accuracy of law enforcement data. What kinds of data should law enforcement agencies report to the communities they police? What consequences, both positive and negative, might follow from reporting these data? How would we know?
Cynthia Conti-Cook, Staff Attorney, Criminal Special Litigation Unit, Legal Aid Society, New York City
Jonathan Mummolo, Assistant Professor of Politics and Public Affairs, Princeton University
Anita Ravishankar, Fellow, The Lab @ DC
12:00-1:00: Working Lunch: Developing strategies for data-driven criminal justice reform
In this final session we will review the most promising strategies identified to promote data-driven criminal justice reform, discuss potential partnerships, and develop a roadmap for future research.
Chris Dawes, Associate Professor of Politics, New York University
Sharad Goel, Assistant Professor of Management Science and Engineering, Executive Director, Stanford Computational Policy Lab, Stanford University
Jonathan Mummolo, Assistant Professor of Politics and Public Affairs, Princeton University
Daniel Neill, Associate Professor of Computer Science and Public Service, New York University
