You are on page 1of 27

An Assessment of Traffic Stops and

Policing Strategies in Nashville

Prepared by the Policing Project at


New York University School of Law


Table of Contents
Executive Summary ....................................................................... 3
Background ................................................................................... 4
About the Policing Project .............................................................. 4
Initial Approach and Recommendations in Nashville ......................... 5
What We Did ................................................................................. 6
Traffic Stops in Nashville ............................................................... 7
Racial Disparity in Traffic Stops ...................................................... 7
Assessing the Efficacy of Traffic Stops ............................................ 9
Officer-Level Differences in Traffic Stop Practices .......................... 10
Social Costs ................................................................................ 11
Recap ......................................................................................... 11
Recommendations ....................................................................... 11
Appendix A: Policing Project Meetings in Nashville ......................... 16
Appendix B: Stanford Computational Policy Lab Report ................... 18

2
Executive Summary
In response to the Gideon’s Army report indicating racial disparities in traffic stops, and the shooting
of Jocques Clemmons, the Nashville Mayor’s Office asked the Policing Project to help develop
strategies to address the disparities and improve community-police relations in Nashville. The Policing
Project is an organization devoted to front-end democratic accountability to assure just and effective
policing.
The Policing Project talked with dozens of Nashville residents about their experiences with policing.
Based on those conversations, we proposed to conduct a thorough assessment of the costs and
benefits of using traffic stops to address crime. And we suggested that the City create a Steering
Committee to guide work around community-police engagement and policing in Nashville.
We conducted the traffic stop data work in collaboration with the Stanford Computational Policy Lab
(SCPL), whose researchers performed the analysis. (The SCPL team’s more detailed report is included
here as Appendix B.) The Metropolitan Nashville Police Department (MNPD) provided the necessary
data, and has from the beginning shown a strong commitment to re-evaluating its traffic stop
strategies and developing alternatives that can achieve public safety with fewer social costs.
As the SCPL report shows, and as we summarize below, there are indeed notable racial disparities in
traffic stops in Nashville. These disparities are higher for traffic stops around non-moving violations,
such as broken taillights or expired tags. Disparity, however, is not necessarily evidence of
discrimination. Any number of neutral factors, including officer deployment patterns or differences in
rates of offending, may explain these and other disparities in the criminal justice system. MNPD
explains these racial disparities in traffic stops on the ground that officers go where the crime is, and
that in Nashville, high-crime neighborhoods tend to have larger minority populations. The SCPL
analysis bears this out. However, even controlling for crime, unexplained racial disparity still remains.
More importantly, the SCPL report shows that traffic stops are not an effective strategy for reducing
crime. In particular, MNPD’s practice of making large numbers of stops in high crime neighborhoods
does not appear to have any effect on crime.
We make a number of recommendations, including that MNPD:
• reduce the number of traffic stops
• acknowledge black residents have been disproportionately affected by MNPD’s stop practices
• monitor racial disparities on an ongoing basis
• redeploy officer resources toward more effective crime-fighting tools
• consider adopting a Neighborhood Policing strategy
• post its department policies online
• conduct a review of certain key policies such as use of force
• conduct a review of training around use of force, traffic stops, and procedural justice
• adopt a body camera policy with attention to transparency regarding the release of body
camera footage

In addition, we suggest that Nashville engage in a public process of strategic planning around public
safety, bringing together the voices of the community and MNPD officials in doing so.

3
Background
In 2016-17, two events took place that focused public attention on policing in Nashville—and
revealed longstanding tensions around police-community relations, particularly in some of Nashville’s
communities of color. In October 2016, Gideon’s Army released a report that pointed to MNPD’s
longstanding practice of making large numbers of stops in high crime neighborhoods—and it pointed
to substantial racial disparities in those stops. Then, in February 2017, an MNPD officer shot and
killed Jocques Clemmons, leading to protests and further concern about policing in Nashville.
Later that year, then-Mayor Megan Barry reached out and asked the Policing Project at NYU Law to
offer suggestions for a plan to address community concerns, and help chart a path to strengthen the
partnership between MNPD and the communities it serves. Although the nature of our assignment
changed somewhat at the mayoral transition, in general we were asked to continue our efforts.
In the succeeding sections, we explain the approach we recommended to Mayors Barry and Briley,
what we learned from our work, and our recommendations for how Nashville and MNPD might
move forward in light of what we have learned.
Before doing so, though, we introduce the Policing Project. This is important because we take a
somewhat unique approach to issues of public safety reforms, which necessarily frames the
recommendations we make.

About the Policing Project


The Policing Project at NYU Law is a not-for-profit center at New York University School of Law
dedicated to assuring just and effective policing through democratic accountability. It is led by
Professor Barry Friedman, who for over a decade was a law professor at Vanderbilt School of Law.
He is the author of Unwarranted: Policing Without Permission, and is leading a national standard-setting
effort on policing for the American Law Institute, Principles of the Law: Policing.
Although there are many organizations that work in the area of policing and public safety, the Policing
Project takes a unique approach, focused on what we call “front-end accountability.” Most of the
attention on policing in this country is on the back end, after something has gone wrong. Remedies
that are proposed run from civil lawsuits to criminal prosecutions of officers, to civil rights
investigations, to civilian review boards, to inspectors general. Any system of accountability needs a
back end, and policing is no different. At the same time, as much media coverage has made clear,
there is ongoing concern about the efficacy of those back-end approaches.
The Policing Project focuses on the front end of policing: the need for democratic voice around how
policing should occur to avoid problems in the first place. Such front-end engagement has historically
been lacking around policing, and we believe changing this could have a transformative effect. To this
end, we advocate for transparency around policing (so the public can make sound choices); we
identify best practices and write model policies on issues ranging from use of force to the use of
policing technologies such as body cameras, predictive policing, or police searches of social media;
and we work with communities and the police on ways to ensure an effective means for the
community to have a voice in how it is policed.

4
More information on our projects can be found on our website, www.policingproject.org, but here we
mention just a few:

• Chicago: We worked closely with the Grassroots Alliance for Police Accountability to assist
them in drafting an ordinance to create a Community Police Commission for the City of
Chicago. We also are working in two Chicago police districts to pilot a comprehensive
Neighborhood Policing Initiative, and to help facilitate more robust community engagement
around policing practices and priorities.
• Body Cameras: We have worked with police departments in New York City, Los Angeles,
and Camden, New Jersey, to gather public views on the policies that should govern how body
cameras are deployed. In Los Angeles, we focused specifically on the question of whether and
when footage ought to be released to the public after a police shooting or other serious use of
force incident. The policy that LAPD ultimately adopted became a model for a state-wide law
that recently went into effect.
• Cleveland: We are working with the federal monitor in Cleveland to help implement the
reforms required under the City’s agreement with the U.S. Department of Justice. We have
helped facilitate community-wide conversations around use of force and community policing,
worked with the department to develop community policing and bias-free training, and
assisted with various other aspects of the monitoring team’s work.
• Racial Disparities in Policing: In partnership with the NYPD and Open Society
Foundations, we recently co-hosted an event that brought together agency heads from across
the City of New York, from education to housing to public health, in order to address the
root causes of racial disparities in policing. The goal of the gathering was to identify the
various steps that each agency could take in order to reduce racial disparities in the outcomes
for which they are responsible—and which may contribute to racial disparities throughout the
criminal justice system.

Initial Approach and Recommendations in Nashville


Because community voice is central to what we do, the Policing Project began its work by making
several trips to Nashville in the summer and fall of 2017. We met with dozens of Nashville residents
representing a variety of stakeholders, including representatives from various communities of color,
faith-based and professional leaders, the legal community, and the agencies concerned with criminal
justice. We also met with MNPD Chief Steve Anderson, members of the department’s command
staff, and representatives from both the Fraternal Order of Police and the Black Police Officer’s
Association. (A complete list of individuals and organizations is included as Appendix A.)
Over the course of several trips it became clear to us that traffic stops were of great concern. In
particular, our visits clarified two things. First, the frustration in minority communities––well beyond
individuals discussed in the Gideon’s Army report––was acute. And second, there were many in the
white community who were largely unaware of this, and were quite concerned at the stories they were
hearing from their fellow residents from communities of color. As one prominent Nashvillian said
quite emphatically, he had been “ignorant” and was “appalled.”
After hearing from all these individuals, we proposed a two-pronged approach: a thorough cost-
benefit analysis of MNPD’s traffic stop practices, and a broader community conversation about
policing, led by a steering committee of Nashville residents. In the fall of 2017, we began working
5
with MNPD and a national team of social scientists on the data portion of our work, while continuing
to work with the Mayor’s office to put together a plan for the rest.
The change in administration—and the surrounding uncertainty about the City’s broader agenda
around police reform—put plans for a broader engagement on hold, but all agreed that we should
continue with the data work. So, we did.
On July 26, 2018, an MNPD officer shot Daniel Hambrick (in the course of a traffic stop). That once
again brought policing issues to the fore. At the invitation of Mayor Briley, we recommenced
conversations with his office about what measures might be taken to promote front-end
accountability in Nashville. Most recently, a referendum on the Community Oversight Board passed
by a large margin. That Board is largely configured to conduct “back end” review of policing, but it
also has front-end authority to issue policy recommendations to the Department and other criminal
justice agencies.
At the conclusion of this Report we make Recommendations as to the steps we believe that Nashville
should pursue to assure effective and equitable policing. Whatever else is true, we believe it is essential
that the community and MNPD work together to “co-produce” public safety for all of Nashville. But
first we turn to the results of the data work.

What We Did
With respect to traffic stops, we sought to answer four questions:

• Are there racial disparities in traffic stops?


• If so, what explains the racial disparities?
• Are traffic stops an effective crime reduction tool?
• What are the social costs of these stops?

In order to conduct this cost-benefit analysis, the Policing Project assembled a national team of social
scientists. In addition to the four researchers affiliated with Stanford’s Computational Policy Lab
(SCPL), the team included a number of prominent researchers who advised on various aspects of the
project, including Jack Glaser (U.C. Berkeley), Mark Cohen (Vanderbilt), Crystal Yang (Harvard), and
Richard Carson (U.C. San Diego). Between them, the social scientists have extensive experience
working with policing and criminal justice research. Several are renowned experts on cost-benefit
analysis, and in particular on incorporating intangible social benefits and costs.
We also benefited greatly from the assistance of Robert Haas, who was formerly the police
commissioner of Cambridge, Massachusetts, and Massachusetts’ Executive of Public Safety. More
recently, Bob Haas has been working with the Policing Project to pilot the Neighborhood Policing
Initiative in Chicago, and also is working separately with MNPD to pilot a variety of crime reduction
strategies as part of the national Public Safety Partnership.
Together with Bob Haas and our social science partners, we conducted two rounds of focus groups
with MNPD commanders and officers to learn more about the department’s strategy, and to better
understand the patterns we were seeing. We joined officers on ride-alongs both to hear from officers
directly and to learn more about how they record data and conduct stops. We also worked closely

6
with crime analysts at MNPD to identify the data sets that we would work with, and to resolve various
confidentiality concerns.
Throughout the Spring and Summer of 2018, Policing Project team members met weekly with the
SCPL data team to work through the analysis and identify follow-up questions. The report that
follows is the product of this work. Stanford’s Computational Policy Lab prepared a more detailed
technical analysis, which is included as Appendix B.

Traffic Stops in Nashville


The Metro Nashville Police Department (MNPD) has for many years employed a strategy of making
large numbers of traffic stops in high crime areas as a way to address crime and violence. The strategy
is thought to work in one of two ways: (a) deterring crime in the area by establishing a visible officer
presence, or (b) creating opportunities for officers to identify suspects or seize contraband. While on
patrol, officers are instructed to be on the lookout for potential traffic violations and—after making a
stop—to be on the lookout for signs of possible criminal activity including, if appropriate, asking
additional questions or seeking permission to search the car. At the height of this strategy in 2012,
MNPD conducted nearly 450,000 stops. The number of stops has since gradually gone down.
In 2017, MNPD conducted approximately 250,000 traffic stops—approximately 458 stops for every
1,000 driving-age residents. Of these, nearly half (45%) were for non-moving violations, which mostly
consist of equipment or registration violations (e.g. broken tail lights, broken headlights, expired tags).
It is not uncommon for police departments to utilize stops (traffic and pedestrian) as a tool to fight
crime and violence. What the SCPL team found, however, is that Nashville’s per capital stop rate is
considerably higher than in other cities of approximately the same size. Nashville makes more than
twice the number of stops per capita than Raleigh or Charlotte, and more than five times the number
in Austin and Columbus. See SCPL Report at p. 2.

Racial Disparity in Traffic Stops


I. The Data on Disparity

Nashville’s driving-age population is 58% white, 27% black, 9% Hispanic, and 6% other.
Over the course of many years, black drivers have been stopped at a higher rate than white drivers
relative to their percentage of Nashville’s population. As the overall number of stops has gone down
over time, the racial gap has narrowed as well. Still, in 2017, the per capita stop rate was 44% higher
for black drivers than for white drivers. What this means is that while MNPD made approximately
433 stops for every 1,000 white residents of driving age, it made 623 stops for every 1,000 black
residents of driving age.
Racial disparities are notably higher for non-moving violation stops than for moving violations. Thus,
if one disaggregates the 44% figure into moving and non-moving violations, in 2017 the per capita
stop rate for black drivers was 68% higher for non-moving violations—as compared to 24% for
moving violations. For that reason—and because non-moving violation stops are arguably less
important for traffic safety—the data team focused much of its analysis on non-moving violation
stops.

7
Figure 1: Racial Disparity in All Traffic Stops

1,400

Black
1,200

1,000
Stops per 1,000 people

800 White

600

Hispanic
400

200

0
2011 2012 2013 2014 2015 2016 2017

II. Understanding racial disparities

It is important to understand that the fact that black drivers are stopped at a higher rate than white
drivers is not, in and of itself, evidence of racial bias or what is often referred to legally as
“discrimination.” Racial disparities in policing may reflect a variety of factors, including where officers
are deployed, the crimes that they are instructed to prioritize, as well as potential differences in rates
of offending among different demographic groups. Still, racial disparities are concerning and so it is
important to look for causes—first, to try to rule out intentional racial discrimination, but second,
because disparities should be reduced if possible no matter what their causes.
In response to the Gideon’s Army report, MNPD argued that officers go where the crime is—and
that the racial disparities in traffic stops are largely attributable to the fact that Nashville’s high crime
neighborhoods tend to have larger minority populations.
The SCPL team examined this argument and found that:

• Nashville officers do make more non-moving violation stops in high crime neighborhoods,
regardless of their racial composition. That is, stop rates in higher crime, predominantly white
neighborhoods, are comparable to stop rates in higher crime, predominantly non-white
neighborhoods. This is consistent with MNPD’s explanation that officers go where the crime
is.

8
• However, officer deployment patterns explain only part of the overall racial disparity in traffic
stops. On average, within a neighborhood, black drivers are still 37% more likely than white
drivers to be stopped for a non-moving violation.

Figure 2: Distribution of Residents, Stops, and Reported Crimes

Population Stops Crime

The first map shows the population of Nashville by demographic group. The second map shows the distribution of non-moving
violation stops, and the race of the race or ethnicity of the person stopped. The third shows the distribution of reported crimes.

The question, then, is what explains these within-neighborhood disparities? The sorts of factors that
may be in play here certainly include the possibility of either implicit or explicit biases on the part of
officers. There is a large and growing literature about the impact of implicit racial biases in society,
and policing is not immune from the biases that affect us all. But it is important to recognize some
other possible causes, especially when it comes to solutions. For example, lower-income residents may
drive older cars, or may lack the resources to get broken taillights or plate-lights repaired as quickly as
other drivers. They also may be more likely to have expired tags. If lower-income residents tend to be
disproportionately black, this could explain at least some of the remaining disparity. There also may be
differences between the demographic distribution of residents in a particular neighborhood (which is
what was used as the “baseline”), and the makeup of drivers actually on the road at any given time.
More work could, and perhaps should, be done to assess the precise cause of these disparities. If they
were part of a crime-fighting strategy that was successful, it would be very important to do so. But
that raises the question—to what extent are traffic stops, especially for non-moving violations, an
important crime-fighting tool?

Assessing the Efficacy of Traffic Stops


Therefore, we next considered to what extent traffic stops are in fact an effective crime reduction
tool. The theory, as we indicated at the outset, is that stops may act as a deterrent: when officers step
up activity, would-be offenders decide it is too risky to try anything. Stops also may lead to arrests,
taking would-be offenders off the street.
However, the SCPL team found that:

9
• Traffic stops do not appear to have a significant impact on long-term crime trends. As the
number of traffic stops declined between 2012 and 2017, crime rates remained quite flat. See
SCPL Report at Fig. 6, p. 6.
• Traffic stops also do not appear to have any effect on crime in the short-term. This was some
of the SCPL team’s most sophisticated and important analysis. As officers increase the
number of stops in a particular area, crime does not necessarily fall as a result. In some weeks,
officers made an above average number of stops—and crime indeed went down. But
sometimes crime went down without any change in the number of stops. And sometimes
crime would go up despite the stops. On average, we simply did not find a relationship
between stops and crime. See SCPL Report at pp. 6-7.
• Finally, non-moving violation stops rarely lead to an arrest, or to the recovery of drugs or
weapons. For every 1,000 non-moving violation stops, just over 2% (or 21) resulted in an
arrest, or the recovery of drugs or other contraband. An additional 61 stops (6.1%) resulted in
a misdemeanor citation for a non-drug related charge.1 The vast majority of these citations
(89%) were for driving with a revoked or suspended license.2

This suggests that MNPD could safely reduce the number of stops—and in doing so, reduce the
overall racial disparities in stops as well.

Officer-Level Differences in Traffic Stop Practices


We also examined whether particular officers or units make a disproportionate number of stops.
Each of MNPD’s eight precincts has patrol officers who respond to calls for service, and make stops
and engage in other activities between calls. Each precinct also has 2-3 “flex” teams of 6 officers each.
Flex officers typically do not respond to calls for service, and are expected by MNPD to engage in
various proactive activities, including making traffic stops.
During our focus groups, we learned that individual officers, supervisors, and district commanders are
given a great deal of leeway to decide what strategies to pursue, including the degree to which they
ought to rely on traffic stops. Some flex officers, for example, reported making few if any traffic stops
in a given week, while others said they typically made eight or more stops each shift.
The SCPL team found that flex officers conduct about twice as many non-moving violation stops per
officer. Whereas the average patrol officer makes 109 stops each year, the average flex officer makes
217. However, because there are far more patrol officers than flex officers, patrol officers still make
60% of all non-moving violation stops.
Consistent with the focus group discussions, the data team found that a small number of officers
conduct a very large proportion of non-moving violation stops. The ten most active officers—which


1 Under Tennessee law, a misdemeanor citation is considered a non-custodial “arrest.”
2 The large number of citations for driving with a suspended license may reflect in part Tennessee’s longstanding practice
of revoking the drivers’ licenses of individuals who were unable to pay traffic fines or court costs. A federal district court
recently deemed this practice unconstitutional—and ordered the state to reinstate these licenses.
10
includes both flex and patrol officers—made approximately 9,399 stops (about 9% of stops).3 The
most active 125 officers (17% of officers) made 50% of all stops.4

Social Costs
From the outset of our evaluation we have expressed interest in examining the social costs of traffic
stops. Social costs are the costs that are felt by individuals subjected to a particular policing tactic, and
by the communities of which they are apart. When it comes to traffic stops, those costs could include
the lost time of the driver, psychological costs of an unwelcome encounter with the police, dignitary
costs felt by those who perceived the stop as imposed for illegitimate racial reasons, and loss of trust
in particular communities from the negative perceptions of the stops.
In the course of our work, we had many conversations with a diverse group of Nashvillians across
many walks of life. In those conversations, we found ample evidence that frequent stops were having
these effects, particularly among communities of color.
Valuing social costs precisely can be expensive. We believe this is important work, and it could be
done in Nashville. However, there is a foundational rule in cost-benefit analysis: if there are no
identifiable benefits in the first place, costs should be avoided altogether. As indicated, we have not
found any crime-fighting benefits in MNPD’s strategy of proactive traffic stops.

Recap
To summarize, in 2017, black drivers in Nashville were 68% more likely to be stopped for a non-
moving violation than were white drivers. A substantial portion, but certainly not all of, this disparity
stems from the fact that MNPD officers spend more of their time in high crime neighborhoods—and
make more stops in these neighborhoods as well.
Yet, non-moving violation stops do not appear to have a discernible effect on either long-term or
short-term crime rates. And they result in a relatively small number of arrests.
This suggests that if MNPD’s primary concern is crime reduction, it could reduce the number of
equipment and registration stops, and direct officer resources to more productive strategies that could
potentially lead to greater reductions in crime, while strengthening the relationship between MNPD
and the communities it serves.
We note that this is an important result, not only for Nashville, but for other communities as well.
Police in many places rely heavily on stops as a crime-fighting tactic. Although the value of these
stops may vary from locale to locale, our work suggests the use of such stops should be explored
carefully, especially given that racial disparities frequently result from such stops.

Recommendations
Although considerable effort went into the traffic stop study, in the course of our work we had a
chance to talk with many Nashvillians, from many walks of life, about policing and public safety. This
includes members of the black and white communities, other communities of color and immigrant

3Of these, 8 were flex officers and 2 were patrol officers.
4Again, 46.5% of these were flex officers, and 53.5% were patrol officers. 19 of the officers served as both flex and patrol
over the course of the year.
11
communities, key leadership in the criminal justice system, and members of MNPD, both line officers
and command staff.
We believe it appropriate to share a set of recommendations, not only about traffic stops, but other
aspects of policing and public safety that were raised in these conversations, and about which we have
expertise. These other recommendations relate to the work we were asked to do around both traffic
stops and the issue of community-police relations.
I. Traffic Stops
Traffic stops impose obvious costs—even if we have not quantified them precisely—and seem to
produce few crime reduction benefits. The question then arises, what should be done to rectify the
racial disparities caused by these stops?
The first answer is that Nashville ought to recognize in some official way the burden that these racial
disparities have imposed on communities of color in Nashville, particularly the African American
community. This sort of recognition has proven important and effective throughout the country in
opening up dialogue about next steps.
The second answer is to reduce the number of stops, as well as tracking, remaining conscious of, and
working to eliminate as much as possible any disparity. That is true for reasons of racial justice, but
also for reasons of overall public safety.
For many years, MNPD has used traffic stops to pursue two goals at once: to promote traffic safety,
and to address crime. Given that there do not appear to be any crime reduction benefits to stops, we
encourage MNPD to focus traffic enforcement efforts in areas where traffic safety is of particular
concern—and to direct crime-reduction resources toward more successful crime-fighting strategies.
As noted above, relatively few officers perform a very high number of stops, which may facilitate
bringing down the number of stops. Based on conversations with MNPD, it appears that many of
these officers are some of the department’s most dedicated and high-performing officers—whose
efforts may simply need to be directed toward other strategies.
In making this recommendation we want to emphasize that this change will not happen overnight and
not without substantial effort and direction. As the high number of traffic stops shows, this has been
one of MNPD’s core strategies for fighting crime for some time now. MNDP will need to develop a
set of alternative strategies, and its officers will need to be trained accordingly.
We note, however, that MNPD may face some difficulty in adopting these new strategies and likely
will require outside assistance. Because traffic stops have been a core MNPD strategy for so long,
there is not necessarily the expertise or knowledge base within the department to transition to new
strategies. Throughout our conversations with MNPD, officials have been willing to consider
alternatives, but are not certain what those alternatives are. In addition, Nashville uses a command
structure that gives a great deal of discretion to precinct commanders to pursue their own
enforcement strategies. Although discretion may have its benefits, for example in terms of fostering
individual initiative, the amount of such discretion in Nashville is uncommon in our experience and
may hinder system-wide change.
As of the time of this report, we are engaged in discussion with MNPD leadership, including the
precinct commanders, about what a change in deployment might look like, and alternative strategies
12
they might pursue. In particular, Robert Haas, who has a great deal of experience with policing
models, and also is working with MNPD through the U.S. Department of Justice violence reduction
efforts, has been consulting on our behalf with MNPD. Those conversations are ongoing.
II. Neighborhood Policing
The most promising strategy is likely to move toward a model of neighborhood policing in
communities suffering from high crime. Here we explain what that might look like, as well as
challenges of implementation. MNPD has expressed interest in piloting the model in one or more
precincts.
Neighborhood policing is based on the philosophy that officers ought to be familiar with and engaged
with people in the neighborhood they police, and work collaboratively with those residents to achieve
public safety in a just and effective way.
The theoretical basis for this sort of policing is that violence-torn or crime-ridden neighborhoods in
particular cannot be made safe unless the police and the community work together to do so. All over
the country we hear the same thing: there is crime, or there are homicides, and the community does
not “cooperate” with the police. Yet, what city police chiefs have come to recognize is that relations
in those communities are sundered by heavy enforcement efforts: stop-and frisk, aggressive
enforcement of low-level offenses, the imposition of fines and fees, and high levels of incarceration.
People in those communities are reluctant to interact with the police.
The alternative is a form of policing, which we specify in greater detail below, in which the police
actively work to partner with communities to address problems of crime and blight.
In talking about neighborhood policing, we want to distinguish it from community policing, about
which much is said in the public sphere. Beginning in the late 1970s, and continuing to the present,
many have recommended community policing as an alternative to aggressive enforcement and
random patrols. The difficulty, as our extensive research shows, is that the phrase community policing
came to mean so many different things to so many different people, that it lost all coherent meaning.
In many departments, it involved little more than assigning a couple of officers in various locales to
attend community meetings, while the rest of the department went on with the “real” business of
policing. Another difficulty with community policing is that a typical meeting held by police involves
the police talking at a group of people, largely those who already have good relationships with the
police. There is very little engagement of a meaningful sort with the community at large and in
particular with members of heavily-policed communities.
The evolving concept of neighborhood policing seeks to build on the sentiments that drove a push
toward community policing, but to take it seriously as a holistic form of policing that should extend to
every aspect of a department’s operations. Although various communities are experimenting with
aspects of neighborhood policing, to date the most comprehensive form of it is in New York City.
The NYPD has deployed an intensive neighborhood policing model with great success. Under the
NYPD’s plan, officers are expected to stay within their assigned sectors or beats throughout the day.
In order to enable them to get to know residents, officers are given substantial time “off the radio” to
engage with their communities. Any overflow in calls for service is handled by a small number of
rapid response cars. Meanwhile, neighborhood coordinating officers work closely with community
groups to identify community concerns and develop response strategies.

13
Each city of course needs its own tailored form of Neighborhood Policing. We are in the process of
implementing such a model in pilot districts in Chicago. To deal with the huge break in community
trust in that city, we are incorporating a very heavy community engagement component, in which
community members are given real voice in how they are policed. In a city like Nashville, this
approach may make more sense in some neighborhoods or communities than in others. But we
believe that in terms of reducing crime and developing healthy police-community relations, it ought to
be considered seriously.
III. Strategic Vision
Implementing Neighborhood Policing, or anything like it, requires addressing a host of issues, from
resources to the style of policing best suited to a community. We will address some of these below,
but it brings us to an umbrella issue that needs to be addressed.
It is no secret to anyone that Nashville is a fast-growing metropolitan area with a host of concerns
from gentrification to displacement of residents to a booming city center. This sort of transformation
has both good and ill effects. And one key area of focus necessarily must be public safety. Indeed,
perhaps public safety should be the initial focus.
Yet, it is our sense that although MNPD has taken a number of steps to respond to these trends,
there has not been an opportunity to engage in strategic, holistic thinking around public safety in a
changing Nashville.
We recommend that the City consider initiating this conversation. It is an essential step in developing
leadership in the department, approaches to public safety, and addressing resource concerns.
This sort of strategic planning could be done in any of a number of ways, but what is certain is that it
should bring a variety of city stakeholders into dialogue with MNPD about what the future of public
safety in Nashville should look like.
IV. Front End Accountability
We believe a strong system of front-end accountability around policing leads to safer communities,
better relations between the police and communities, greater legitimacy of policing, and better
outcomes. There are a number of steps we think MNPD and Nashville should take in this regard. We
have discussed several of these with MNPD and there is a willingness to consider or pursue them.
First, we believe MNPD’s policy manual should be put on the web so that anyone can see its policies.
Many departments throughout the country do this, and we have given assistance to others. There are
undoubtedly some policies—such as how active-shooter situations are handled—that should not be
public. But that is not true of most of the policies that govern policing.
Second, we think it would be useful to conduct a policy review of some of the critical MNPD policies,
especially around Use of Force and aspects of stops and searches such as Consent Searches. The goal
is to make sure MNPD is adhering to best practices in these areas.
Similarly, we believe it would be valuable to review some of MNPD’s training around things like Use
of Force. We have not done so and express no views whatsoever, but given community concern on
these issues such a review would make sense. It also might be profitable to examine other areas of
training like Procedural Justice or community engagement.

14
We have consistently heard disagreement about the functioning of the Office of Professional
Accountability, including wildly different estimates of the “sustain” rate of complaints. We would
point out this is a complicated issue. Even if sustain rates are low, this could be for very different
reasons: officers could be behaving very well, or OPA may not be sufficiently diligent. We have no
basis for an opinion one way or another, but public faith in the back-end system of police discipline is
essential. It could well be that the creation of the COB will address this sufficiently, but one item to
consider is an audit of OPA and a report to the community so that there is a common set of facts
from which to start.
Nashville is transitioning to body cameras, an area in which we have considerable expertise. If there is
one lesson we have learned, it is that the substantial money spent on BWCs is squandered without
sound policy in place that deals with, among other things, release of the video to the public, or to
individuals who wish to file a complaint.

15
Appendix A: Partial List of Individuals with Whom
Policing Project Has Met or Spoken 5

Educational & Religious Institution Affiliated


Dr. Emilie Townes (Vanderbilt Divinity School)
Amy Steele (Vanderbilt Divinity School)
Herbert Marbury (Vanderbilt Divinity School)
Candice Ninn (Vanderbilt Divinity School)
A. Dexter Samuels (Meharry Medical College)
Aerial Ellis (Lipscomb University)
Sekou Franklin (Middle Tennessee State University; Community Oversight Now)
Brodrick Thomas (Trevecca Nazarene)
David Tucker (American Baptist College)
Pastor Darrell Drumwright (Temple Church)
Bishop Joseph Walker (Mt. Zion Baptist Church)
Pastor Breonus Mitchell (Greater Grace Temple Community Church; Mount Gilead Missionary
Baptist Church)
Rev. Martin Espinoza (Ray of Hope Community Church)
Harold Love (International Ministerial Fellowship, Lee Chapel AME Church)

Advocacy Organization Affiliated


Heidi Weinberg (ACLU)
Ludye Wallace (NAACP)
Sharon Roberson (YWCA)
Hanna Cornfield (YWCA)
Jessica Guzman (YWCA)
Bishop Campbell (Gentlemen and not Gangsters)
Gerald Brown (Nashville Dismas House)
Marsha Edwards (Martha O'Bryan Center)
Walter Searcy (NOAH)
Martin Hodge (NOAH)
Joe Engle (NOAH)
Kyle Mothershead (NOAH)
Rev. W. Antoni Sinkfield (NOAH)
Rev. Ed Thompson (NOAH)
Eric Brown (Forward Nashville)
Fallon Wilson (Black in Tech Nashville)
Jurnell Cockhren (Black in Tech Nashville)
Eric Brown (Children’s Defense Fund Nashville Team)
Rashed Fakhruddin (Islamic Center of Nashville)
Kasar Abdulla (Tennessee Immigrant and Refugee Rights Coalition)|
Rasheedat Fetuga (Gideon’s Army; Community Oversight Now)
Theeda Murphy (Community Oversight Now)
Sheila Clemmons Lee and Mark T. Lee (Justice for Jocques Coalition)

5 This list of individuals has been reconstructed from Policing Project staff notes, taken during Nashville based

meetings, and may not be comprehensive.


16
Ethan Link (Laborers’ International Union of North America (LIUNA)

Government Affiliated
Reggie Miller (Black Police Officers Association)
James Smallwood (Fraternal Order of Police)
Jimmy Gafford (Fraternal Order of Police)
Bob Nash (Fraternal Order of Police)
Brenda Wynn (Davidson County Clerk)
Jocelyn Stevenson (Tennessee Bar Association)
Judge Sheila Calloway (Davidson County Juvenile Court Judge)
Mel Fowler Green (HRC)
Dr. Phyllis Hildreth (HRC)
Dawn Deaner (MPDO)
Martesha Johnson (MPDO)
Glen Funk (DA)
Mary Carolyn Roberts (City Council)
Bob Mendes (City Council)
Scott Davis (City Council)
Hershell Warren (Mayor's Office, Senior Advisor)

Miscellaneous Affiliations
Charles Bone (Bone McAllester Norton PLLC)
Wallace Dietz (Bassy Berry & Sims)
Byron Traguer (Trauger & Tuke)
Steven A. Riley (Riley Warnock & Jacobson)
Jarrett Strickland (UBS Financial)
Ben Rechter (President of Rogers Group Investments, Inc.)
Amy Adam Strunk (Tennessee Titans)
Fina Tuggle (Tennessee Titans)
Burke NiHill (Tennessee Titans)
Demetria Kalodimos (WSMV)
Itzel Gonzalez Patino
Narnelle Cochran
Avi Poster (Community Organizer)
Paul Galloway (Executive Director of The American Muslim Advocacy Center)

17
Appendix B: Report Prepared by the Stanford
Computational Policy Lab

18
Stanford Computational Policy Lab Page 1 of 9

An Analysis of the Metropolitan Nashville Police


Department’s Traffic Stop Practices
Alex Chohlas-Wood* , Sharad Goel† , Amy Shoemaker‡ and Ravi Shro↵§
Stanford Computational Policy Lab
November 19, 2018

EXECUTIVE SUMMARY L ike all police departments, the Metropolitan Nashville


Police Department (MNPD) uses a wide range of en-
forcement tools to ensure public safety. Traffic stops are
For the last several years, Nashville has made con- one such tool. These interactions typically involve an offi-
siderably more traffic stops per capita than the na- cer pulling over a motorist, issuing a warning or citation,
tional average, with stops disproportionately involv- and—more rarely—conducting a search for contraband or
ing black drivers. Here we examine the Metropolitan making a custodial arrest. The prevalence and nature of
Nashville Police Department’s (MNPD) traffic stop traffic stops vary widely across American cities, but they
practices in 2017, drawing on an extensive dataset are generally the most common way police departments
of records provided by the department. Black drivers initiate contact with the public [6].
were stopped 44% more often per driving-age res-
ident when compared to white drivers; this gap is In the past several years, the MNPD made more traf-
particularly pronounced among stops for non-moving fic stops per capita than many similarly sized Ameri-
violations (68%), such as broken tail lights and ex- can cities—in some cases, over ten times as many (Fig-
pired registration tags. These disparities stem, in ure 1). Local community groups have also raised con-
part, from a strategy that concentrates traffic stops cerns that the MNPD’s traffic stop practices dispropor-
in high-crime areas. In particular, after controlling tionately impact black drivers. In 2016, Gideon’s Army
for location, disparities among non-moving violation published a report, “Driving While Black,” documenting
stops drop from 68% to 37%. This policy of concen- racial disparities in MNPD traffic stops between 2011 and
trating stops in high-crime areas may be predicated 2015 [4]. Notably, there were more stops of black drivers
on the belief that traffic stops are an e↵ective tactic per year than the number of black driving-age residents
for reducing burglaries, robberies, and other crimi- in Nashville. The MNPD, in response, argued that such
nal activity. We find, however, no immediate or long- disparities resulted from higher deployment to areas with
term impact of traffic stops on serious crime. We fur- greater incidence of crime and requests for police services.
ther find that only 1.6% of stops result in a custodial Our goals in this report are three-fold. First, we aim to
arrest—often for license violations or drugs. These quantify racial disparities in the MNPD’s current traffic
findings suggest that the MNPD could reduce traf- stop practices. In particular, we focus on stops in 2017, a
fic stops without an associated rise in serious crime, year in which the MNPD’s traffic stop rates had dropped
while bringing Nashville’s traffic stop rates more in by almost 50% from their peak during the years covered
line with similar cities around the country. In par- by the Gideon’s Army report. Second, we seek to assess
ticular, the MNPD could substantially reduce racial the extent to which any observed racial disparities may
disparities by curtailing stops for non-moving viola- be driven by concerns for public safety. Finally, and most
tions. Notably, a small proportion of active MNPD importantly, we strive to provide concrete, data-driven
officers conduct the majority of non-moving viola- insights to improve both the equity and efficacy of the
tion stops, potentially facilitating any e↵ort to re- MNPD’s policing strategies. Our analysis builds on a long
duce such stops. line of empirical research examining traffic stops [2, 3, 8,
13–20, 22].
To conduct our analysis, we used several datasets pro-
vided to us by the MNPD, including traffic stop records
Commissioned by the Metropolitan Government of and crime reports. We also incorporated information from
Nashville and Davidson County, Office of the Mayor. the U.S. Census to construct population benchmarks for
Nashville neighborhoods. Though we focus on 2017, our
dataset covers traffic stops occurring between 2011 and
2017, permitting comparisons with historical trends.
Last year, the MNPD conducted approximately 246,000
*
Deputy Director at the Stanford Computational Policy Lab; † Assistant
traffic stops, or roughly one stop for every two driving-
Professor at Stanford University, Department of Management Science & age residents. We start by comparing stop rates for black
Engineering, and, by courtesy, Computer Science, Sociology, and Stanford motorists and non-Hispanic white motorists. We focus
Law School; ‡ Data Scientist at the Stanford Computational Policy Lab; on these two groups, which comprise about 85% of
§
Assistant Professor at New York University, Department of Applied Nashville’s population, in part for ease of exposition and
Statistics, Social Science, and Humanities
Stanford Computational Policy Lab Page 2 of 9

800

Annual traffic stops per 1,000 residents


in part to mitigate statistical difficulties with analyzing Nashville, 2012
groups that comprise a smaller share of the local pop-
ulation.[1] We find that the stop rate for black drivers 600
in Nashville in 2017 was 44% higher than the stop rate
for white drivers, where stop rates are computed relative
to the driving-age population. Further, certain types of 400 Nashville, 2017
stops exhibited far greater disparities than others. Among
moving violations (e.g., speeding or reckless driving), the Nashville, 50% N.M.V. reduction compared to 2017
stop rate for black drivers was 24% higher than white Nashville, 90% N.M.V. reduction compared to 2017
200
drivers; in contrast, among non-moving violations (e.g.,
broken tail lights or expired registration tags), the stop PPCS National Average
rate for black drivers was 68% higher than for white
drivers. Moreover, stops for non-moving violations were 0

om so

ity

ha h
Ba tte

Fr ore

C sco

ti

th

in

Sa ttle

se
relatively common, comprising 45% of all traffic stops in

Fo nve
ew nna

an

bu
g

st
or
C
Pa

Jo
ei

r lo

a
m

ci

Au

um
r le

W
al

Se
e
a

ci
lti

an

n
D
El

R
Nashville in 2017.

O
in

rt

ol
C

C
ah

N
kl

Sa
These di↵erences in stop rates are striking. It bears em-

O
phasis, though, that such di↵erences may result from a
variety of complex factors, and are not necessarily the Figure 1: Per capita traffic stop rates in Nashville
product of racial bias [1, 5, 9, 13, 19]. In particular, we compared with the national average and activity in
find that the observed disparities are in part attributable other American cities between approximately 2011–
to deployment patterns, particularly the MNPD’s concen- 2016.[4] This figure is intended for approximate compar-
tration of stops in high-crime neighborhoods, which, in ison, not to suggest optimal levels of policing. Traffic
Nashville, tend to have disproportionately large minority stop rates for comparison cities were calculated using
populations. data compiled by the Stanford Open Policing Project
One reason—and arguably the primary rationale—for (OPP). All OPP cities with populations between approx-
carrying out large numbers of traffic stops in high- imately 500,000 and 1 million were included for compar-
crime areas is a belief that this enforcement strategy ison. Cincinnati, New Orleans, and Raleigh have popula-
has broader benefits for public safety. One might posit tions under 500,000, but were added for additional con-
that traffic stops deter future crime or lead to apprehend- text. Green reference lines display historical stop rates
ing those responsible for past incidents. Though plausi- for Nashville, blue lines display stop rates for hypotheti-
ble, we find little evidence of such a connection between cal reductions in non-moving violation stops, and the or-
traffic stops and serious crime levels in Nashville. Over ange line displays the 2015 Police-Public Contact Survey
the 2011–2017 time period, crime levels for Part I of- (PPCS) national average [6].
fenses[2] remained steady despite substantial reductions
in stop rates over the same period. Further, week-to-week more than 95%[3] with no associated increase in crime.
changes in area-specific stop rates were uncorrelated with Further, the MNPD itself has nearly halved its use of
changes in local crime levels. traffic stops over the last several years, while crime rates
Traffic stops might also benefit public safety by facil- have held steady.
itating the arrest of those individuals charged for past Such a reduction may be facilitated by the fact that
crimes but who may have been difficult to otherwise track a relatively small set of officers carry out the bulk of
down. We find, however, that only 1.6% of traffic stops non-moving violation stops, allowing the MNPD to work
lead to a custodial arrest, often for license violations or directly with that group to redirect enforcement activ-
drug possession. An additional 5.8% of traffic stops end ity. For example, 50% of these stops were conducted by
in a misdemeanor citation (resulting in a non-custodial 125 individuals, or 17% of all officers who conducted at
arrest), typically for driving without a valid license. least one traffic stop in our observation period. It is un-
These findings suggest that the MNPD could cur- clear why stops are concentrated among such a relatively
tail traffic stops without increasing serious crime. Given small group. We note, however, that officers in many
the substantial disparities in non-moving violations, one jurisdictions are given considerable discretion to enforce
might first focus on reducing these stops. In particular, traffic laws as they see fit, which may in turn result in
we note that a 90% reduction in non-moving violation the observed pattern.
stops would bring Nashville more in line with per capita
traffic stop rates in similar cities across the U.S. (Fig-
ure 1), and we estimate this change would reduce stop Background
rate disparities between black and white drivers from 44% Police departments may conduct traffic stops for many
to 28%. This reduction in proactive policing would be siz- reasons, including traffic safety, crime reduction, and pub-
able, though not unprecedented. For example, the New lic engagement and education. Traffic stops and traffic
York Police Department reduced pedestrian stops from [3] https://www.nyclu.org/en/stop-and-frisk-data
nearly 700,000 in 2011 to 11,000 in 2017, a reduction of [4] Several cities in this chart do not have data over the entire

[1] In 2011–2016 period. In addition, some cities only share data on stops
2017, the driving-age population in Nashville was 58% that ended with a citation. As a result, strict comparisons should be
white, 27% black, 9% Hispanic, and 6% Asian and other groups. avoided; this chart is intended to demonstrate the notable di↵erence
[2] Part I o↵enses are murder, rape, robbery, assault, burglary, between Nashville traffic stop rates and other proxies for what could
larceny, and motor vehicle theft. be considered typical behavior.
Stanford Computational Policy Lab Page 3 of 9

1,400
safety have a clear connection, given that certain driving
behaviors (e.g., speeding or DUI) directly threaten the Black
1,200
safety of motorists and pedestrians. Conducting traffic
stops may therefore increase compliance with laws de-
signed to minimize the risk of serious or fatal traffic col- 1,000

Stops per 1,000 people


lisions. Some departments also consider traffic stops to
be an e↵ective tool in fighting crime. Under this premise, 800 White
a traffic stop may directly impede the commission of a
crime in progress; less directly, the presence of officers
600
may discourage criminal activity in the areas being pa-
trolled. Traffic stops may also impact crime levels through Hispanic
the discovery of people with outstanding arrest warrants, 400
or by recovering weapons or other contraband. Further-
more, officers may also conduct stops to make contact 200
with members of the public and remind them of traffic
laws, inform them about policing programs, or provide 0
educational materials. Finally, we note that some jurisdic- 2011 2012 2013 2014 2015 2016 2017
tions rely on minor infractions like traffic stops to gener- Figure 2: While stop rates (for all types of traffic stops)
ate revenue [7], a controversial practice that has recently of both black and white drivers have been decreasing
come under scrutiny. Regardless of these broader policy since 2012, the stop rate for black drivers has remained
aims, individual officers may simply be enforcing traffic or consistently higher than the stop rate for white or His-
criminal codes without explicit attention to longer-term panic drivers.
objectives.
Government practices which disproportionately burden
(or benefit) one racial group in comparison to another ized responses, described below); the zone and reporting
are often undesirable, but such practices may be justi- area of the stop (two MNPD-specific geographies); the
fied by legitimate policy considerations. In the case of race of the stopped driver; information about the offi-
traffic stops, it is theoretically possible that such activity cer who conducted the stop; whether weapons or other
has a net benefit for drivers themselves, by deterring un- contraband were found, a custodial arrest was made, or
safe behavior on the road, or by acting as an educational a misdemeanor citation was issued; and narrative details
and community relations strategy for police officers to about the incident.
engage with the public. In the specific case of stops for Almost all traffic stops in 2017 were categorized with
non-moving violations, arguably the primary objective is one of four stop reasons. Moving violations were the most
crime suppression and detection, as the benefits for traffic common, constituting 51% of all traffic stops. These vi-
safety are likely attenuated. Despite such potential ben- olations include illegal driving behavior such as speeding,
efits, research has shown that police stops also impose talking on a cellphone while driving, or reckless driving.
a substantial burden on residents. Police stop practices The next most common categories were equipment vi-
may create stress for stopped individuals, result in fines olations (27%), registration violations (9%), and safety
and fees which are difficult for some residents to pay, violations (9%), comprising 45% in aggregate. A man-
and threaten police-community relations [10, 21]. As po- ual review of the narrative details for 100 records marked
lice rely on residents to report crime and cooperate with as safety violation stops found that they most often in-
investigators, any erosion of trust between residents and volved equipment violations (like broken headlights or tail
law enforcement is a particular concern. lights).[5] Throughout this report, we refer to stops for
these latter three reasons—equipment, safety, and reg-
istration violations—as non-moving violation stops. The
Data remaining 4% of stops are marked with other stop rea-
Our analysis primarily used three datasets provided by sons, including investigatory stops, seatbelt violations,
the MNPD, restricted to 2017 unless otherwise noted. and child restraint violations. We note that regardless
Traffic stop records were used in every part of the study. of the type of stop, officers may issue a verbal or written
We used arrest and crime incident records to gauge the warning instead of a citation. In Nashville, warnings are a
efficacy of traffic stop enforcement. We also used shape- frequent occurrence—in 2017, roughly three out of every
files of MNPD geographies, along with publicly available four traffic stops ended in a warning alone.
data from the U.S. Census, when calculating per capita
We use the MNPD’s incident-record dataset to investi-
stop rates by race and location.
gate the relationship between reported crime and the en-
Traffic stop records were provided by the MNPD for forcement of traffic violations. The MNPD receives over
the period 2011–2017, during which MNPD conducted 80,000 incident reports annually, with over 100,000 re-
2.57 million traffic stops. However, as noted previously, ported crimes, for a total of approximately 725,000 re-
traffic stops in Nashville have seen a marked decline since ported crimes between 2011 and 2017.[6] These records
their peak in 2012: the MNPD conducted almost 450,000
traffic stops that year, but fewer than 250,000 stops in [5] We note that the narrative details of all other types of stops
2017. The traffic stop dataset includes many relevant at- were more closely aligned with their marked reasons.
tributes, including the date and time of the stop; the [6] These figures exclude non-crime incidents, which MNPD
reason for the stop (chosen from among several standard- marks as “matter of record.”
Stanford Computational Policy Lab Page 4 of 9

(a) Population. (b) Non-moving violation stops. (c) Reported crime.

Figure 3: The distribution of Nashville’s residential driving-age population (3a) and locations of non-moving violation
stops (3b), colored by race (white, black, Hispanic, and other). Non-moving violation stops are concentrated in
neighborhoods where reported crimes (3c) are the most dense, which, in Nashville, also have disproportionately large
minority populations.

contain a date and time; a reporting area, marking the lo- per 1,000 white driving-age residents), and this disparity
cation of the alleged crime; and the Federal Bureau of In- dropped to 44% by 2017 (623 vs. 433 stops per 1,000).
vestigation’s National Incident-Based Reporting System These stop rate disparities are particularly pronounced
categorization. In the case of drug-related incidents, we for non-moving violation stops, though they have also
also have drug type and quantity. been declining over time. Among stops for non-moving
Finally, we combine MNPD shapefiles with public U.S. violations, the disparity dropped from 82% in 2012 (578
Census records to generate population benchmarks for vs. 317 stops per 1,000) to 68% (309 vs. 184 stops per
each MNPD geographic unit. The MNPD uses three 1,000) in 2017.
geographic divisions of increasing resolution: precincts Such disparities may arise from a variety of factors,
(8), zones (65), and reporting areas (2,003). We trans- including a deployment strategy that concentrates offi-
lated American Community Survey (ACS) estimates[7] to cers in high-crime areas. We next examine this possibility
MNPD geographies by distributing population from each in several di↵erent ways. Given the substantial dispari-
block group proportionally according to the area of each ties associated with stops for non-moving violations, we
MNPD geography that overlaps. To calculate per capita focus this analysis on that subset, though we note that
stop rates, we then compare stop counts in each geogra- qualitatively similar patterns hold for the full set of stops.
phy with the driving-age residential population recorded First, we visually investigate the geographic distribu-
by the Census in that area.[8] tion of residents and non-moving violation stops, disag-
gregated by race. As shown in Figures 3a and 3b, non-
moving violation stops occur largely in predominantly
Racial disparities in stop rates black neighborhoods. In particular, there are relatively
Since 2012, the per capita traffic stop rate has de- few such stops in the predominantly white neighborhoods
creased substantially for both black and white drivers. on the southwestern side of Nashville. Figure 3c further
However, the stop rate for black drivers has been consis- shows that the geographic distribution of non-moving vi-
tently higher than for white drivers across all years (Fig- olation stops is quite similar to the geographic distribu-
ure 2).[9] In 2012, the stop rate disparity was 61% (1,275 tion of reported crimes throughout the city. These maps
stops per 1,000 black driving-age residents vs. 792 stops thus provide some indication that the racial disparities in
[7] Due to data availability, we use ACS block-group estimates for
non-moving violation stops are at least partly attributable
to such stops being made in high-crime areas—which, in
2013–2016. When analyzing 2011 and 2012 traffic stop data, we
benchmark to 2013 estimates; we similarly use 2016 ACS estimates Nashville, tend to be predominantly black.
as a benchmark for the 2017 traffic stop data. To more rigorously quantify this pattern, we next com-
[8] To our knowledge, driving-age population estimates by race
pare non-moving violation stop rates in predominately
are not available at the block-group level. We accordingly estimate
these figures as follows: for each block group, we compute the white and predominately non-white zones, controlling for
fraction of driving-age residents, and scale the population of each reported Part I crime. As shown in Figure 4, we see stop
race group by that fraction. Citywide estimates are computed by rates and crime rates are positively correlated, meaning
aggregating these block-group level estimates. We note that these
driving-age benchmarks are only a proxy for the number of drivers, that officers are making more stops in zones with higher
and do not account for daytime populations, or the amount of crime rates. Also, among zones with similar crime rates,
time drivers spend on the road. In rare cases, we exclude extreme
instances of areas with high daytime populations as outliers.
[9] Throughout this period, we find lower stop rates for Hispanic Survey (PPCS), which is based on a nationally representative sam-
drivers, consistent with a national analysis of police stops by Pier- ple of approximately 50,000 people who report having been recently
son et al. [13], and with results from the Police-Public Contact stopped by the police [6, 11].
Stanford Computational Policy Lab Page 5 of 9

1,000 1,400
Zone demographics

Black NMV stops per 1,000 people by zone


Majority white 1,200
Majority non−white
NMV stops per 1,000 people

750
1,000

800
500
600

400
250

200

0 0
0 50 100 150 200 0 200 400 600 800 1,000 1,200 1,400
Part I crimes per 1,000 people White NMV stops per 1,000 people by zone

Figure 4: Per capita stops for non-moving violations Figure 5: Black versus white per capita stops for non-
(NMV) vs. per capita Part I crimes for the year 2017, moving violations (NMV). Each circle represents a police
by police zone. Each circle represents a police zone, col- zone, sized by number of stops (black and white) made
ored by whether the zone population is majority white in each zone in 2017. More points lie above the reference
(open circles, dashed line), or majority non-white (shaded line than below, indicating that within-location stop rates
circles, solid line). Zones with similar levels of reported are higher for black drivers than for white drivers.
crime have similar stop rates, regardless of the zones’
racial compositions.
Instead of looking at patterns across zones, we can also
stop rates in predominately white zones are similar to look at patterns within zones. Figure 5 shows that in
stop rates in predominately non-white zones. It thus ap- the majority of zones, the per capita non-moving viola-
pears that stops are concentrated in neighborhoods where tion stop rate for black drivers is higher than for white
crimes are most frequently reported, regardless of the de- drivers. This visual pattern is corroborated with a statis-
mographic composition of the zone. tical model that estimates zone-level disparities:
We add quantitative detail to this result by fitting the
following Poisson regression model: sr,g ⇠ Poisson pr,g · e↵r + g
,
⇣ ⌘
sg = Poisson pg · eµ+↵ log(cg )+ rg
, where sr,g is the stop count of drivers of race r in zone g,
and pr,g is the driving-age population of race r in zone g.
We include coefficients for each race group, denoted by
where sg is the stop count in zone g, pg is the number ↵r , and for each zone, denoted by g . Comparing the co-
of driving-age residents in zone g, cg is the number of efficients ↵white and ↵black , we find that after controlling
crimes per capita in zone g, and rg is the racial composi- for location at the zone-level, the non-moving violation
tion (proportion non-white) of zone g. Under this model, stop rate for black drivers is 37% higher (95% CI: (18%,
a positive value of would indicate that zones with pre- 59%)) than for white drivers.[12][13]
dominately minority populations were being stopped at In summary, our analysis of stop rate disparities sug-
higher rates than predominately white zones with simi- gests three high-level trends. First, though racial dispari-
lar crime rates. We find, however, that is not statisti- ties have been declining over the last several years, black
cally significantly di↵erent from 0 ( ˆ = 0.4, 95% CI: drivers are still stopped more often than white drivers,
(-1.1, 0.4)).[10] That is, we do not find statistically sig- and this gap is particularly large for the subset of stops
nificant evidence that predominately white and predomi- for non-moving violations. Second, this pattern is in part
nately black zones are di↵erentially policed after adjusting driven by the concentration of stops in high-crime neigh-
for reported crime.[11] borhoods, with such activity uncorrelated with zone-level
[10] Confidence intervals for Poisson regression in this study use a [12] Comparing the coefficients ↵
white and ↵hispanic , we find that
dispersion parameter that allows variance to scale proportional to after controlling for location at the zone-level, the non-moving vio-
the mean, accounting for overdispersion. lation stop rate for Hispanics drivers is 40% lower (95% CI: (55%,
[11] We also fit this model restricting to zones with similar crime
22%)).
profiles. Specifically, for each predominately non-white zone, we [13] Using moving violation stops instead of non-moving violation
selected its nearest-neighbor, matching on reported Part I crime stops, we found that black-white stop rate disparities for mov-
rate, using the MatchIt package in R. Under this matched subset, ing violations exhibit a small—but not statistically significant—
ˆ = 0.5 with CI (-1.5, 0.4), in line with the model fit on all zones. reduction, from 24% to 18% (95% CI: (0%, 41%)).
Stanford Computational Policy Lab Page 6 of 9

900
expect to see crime rates change as stop enforcement
800 changes. We examine this potential relationship on two
All stops per 1,000 people time scales: first, over a longer, multi-year time frame;
700
and second, over many shorter, week-long time frames.
Occurrences per 1,000 people

600
We begin by comparing the citywide per capita traffic
stop rate with per capita crime rates over the last several
500 years, shown in Figure 6. The crime rates for both Part I
crimes and violent crimes are roughly steady over the en-
Non−moving violation stops per 1,000 people
400 tire time frame. However, the rate of traffic stops begins
to decrease quite substantially in 2014. Between 2014
300
and 2017, overall traffic stop rates, as well as stop rates
200 for non-moving violations, dropped by more than 40%.
Consequently, at least on this time scale, traffic stops do
100 Part I crimes per 1,000 people not appear to reduce more serious crime.
Violent crimes per 1,000 people In theory, it is possible that other long-term trends—
0
2011 2012 2013 2014 2015 2016 2017 like an improving economy—mask any crime-prevention
benefit from traffic stops. That is, crime might have been
Figure 6: This time series of annual stops and crimes per even lower had traffic stops not declined. To address this
capita suggests the absence of a long-term connection be- concern, we now examine how crime responds to stops on
tween traffic stops and crime levels. MNPD substantially shorter time scales and at higher geographic resolution,
reduced traffic stops over the second half of the seven where such confounding is less likely. In particular, we
year period without any substantial rise in crime. consider stops and crime occurring over the course of a
week in individual reporting areas (RPAs), the MNPD’s
most granular unit of geography.
demographics after controlling for crime. Finally, such an The MNPD generally holds weekly CompStat meetings
enforcement pattern does not account for all the observed on Fridays to make deployment decisions for the follow-
disparities. In particular, black drivers are stopped more ing week, creating and communicating these directives
often than white drivers even within most zones. It is over the next 1–2 days based on current crime trends.
unclear what may be driving this remaining disparity. At Accordingly, we consider weeks starting on Sunday and
least in theory, it may arise from di↵erences in violation ending on the following Saturday. After controlling for in-
rates (e.g., if black drivers are disproportionately more formation available at CompStat meetings, we consider
likely to have broken tail lights), di↵erences in enforce- deployment to be as-if randomly assigned. In practice, it is
ment (e.g., implicit bias), heterogeneity in population or possible that officer assignments are changed mid-week in
crime within zone, or some combination of these factors. response to a serious crime outbreak; further, we cannot
fully account for all information available to commanders
Stop efficacy at the CompStat meetings. Nevertheless, we believe this
assumption is a reasonable, though admittedly imperfect,
As described above, the observed racial disparities in starting point for such an analysis.
stop rates appear to result in part from the concentration
of non-moving violation stops in high-crime areas—in line We first visually examine the short-term relationship
with the MNPD’s explanation. However, unless there are between stop levels and crime levels. In Figure 7, each
discernible benefits of such a policing strategy, we would point represents a week in an RPA in 2017, and the axes
still characterize these disparities as problematic. Here we represent departures from each RPA’s median level of
examine one potential benefit—and ostensibly the pri- crime or median number of traffic stops.[14] As the flat
mary rationale—for such policing practices: that traffic red trend line indicates, we find that weekly crime lev-
stops are an e↵ective means for reducing more serious els within an RPA have almost no relationship with that
crime. week’s traffic stop levels. For example, an RPA could have
a week with the median number of stops for that RPA,
We analyze the efficacy of these stops by measuring two
another week with ten fewer stops than the median, and
di↵erent outcomes: crime levels, and rates of custodial
another with ten more stops than the median. Despite
arrest, misdemeanor citation, and contraband recovery.
these variations in stop enforcement, we would still ex-
Traffic stops may influence crime levels through direct or
pect crime to occur at the median level for that RPA in all
indirect mechanisms. For example, traffic stops could di-
three weeks. This lack of correlation persists when exam-
rectly impede crime by catching criminals (e.g., burglars)
ining more specific crime types, such as violent crimes or
driving to or from from the scene of a crime. On the other
burglaries, when considering non-moving violation stops
hand, traffic stops may also indirectly discourage crime in
specifically, and when including the e↵ect of the previ-
a neighborhood through the active and visible presence
ous week’s crime levels or traffic stop enforcement (as
of an attentive officer in the area. Some traffic stops will
discussed below).
also end with a custodial arrest, a misdemeanor citation,
or the recovery of contraband or weapons, potentially
preventing future criminal activity or apprehending those
[14] Outliers that were far from the median, representing roughly
involved in past crimes.
0.05% of all points, were removed from the analysis. Points are
E↵ects on crime. If changes in traffic stop enforce- downsampled and jittered for the purposes of visualization, but the
ment are connected to changes in crime, one would trend line is constructed from every unjittered point in the domain.
Stanford Computational Policy Lab Page 7 of 9

Per 1,000 stops


Difference in part 1 crimes from RPA median

10 Custodial arrest charge All stops NMV stops


Suspended/revoked licenses 3.7 5.0
Minor marijuana possession 0.7 0.8
5 Other drug crimes 2.2 2.4
DUI 4.6 2.0
FTA/parole violation/warrant 1.9 2.2
Driving violation 0.8 0.7
0
Public misconduct 0.7 0.7
Another crime (burglary, assault) 0.6 0.7
−5 Misdemeanor citation charge
Suspended/revoked licenses 47.1 53.9
Minor marijuana possession 3.3 3.7
−10 Other drug crimes 2.0 2.0
−20 −10 0 10 20 FTA/parole violation/warrant 3.8 4.5
Difference in all traffic stops from RPA median Driving violation 0.3 0.1
Public misconduct 0.3 0.2
(a) Part I crimes vs. all traffic stops. Plate alteration 0.6 1.0
Another crime (burglary, assault) 0.2 0.2
Table 1: Custodial arrest and misdemeanor citation rates
Difference in part 1 crimes from RPA median

10
for traffic stops.[17] For example, 5 out of every 1,000 non-
moving violation stops resulted in a custodial arrest for
5 a suspended or revoked license. Note that 1 out of ev-
ery 1,000 stops and 0.8 out of every 1,000 non-moving
0
violation stops also included a weapons charge.

−5
Arrests, citations, and contraband. Stops may addi-
tionally have an impact on future crime via the custodial
arrest of individuals or the recovery of contraband, includ-
−10 ing illegal weapons. For example, during a non-moving vi-
−20 −10 0 10 20 olation stop, an officer may detain a suspect—who might
Difference in non−moving violation stops from RPA median
otherwise be difficult to locate—with an open warrant
(b) Part I crimes vs. non-moving violation stops. for a string of recent robberies. It is possible that these
custodial arrests prevent future crimes. It is also plausi-
Figure 7: Part I crimes versus both all traffic stops, and ble that contraband recovery, like the recovery of drugs,
also non-moving violation stops specifically, for MNPD thwarts the sale and consumption of illegal materials. Fi-
reporting areas (RPAs) in 2017. Each point corresponds nally, weapon recovery by the MNPD may make it harder
to a specific week in one RPA, where crime and stop for individuals to follow through with violent impulses.
levels are both measured by that week’s di↵erence from Overall, however, both custodial arrests and contraband
the RPA’s 2016 median. Changes in crime levels are ef- recoveries were infrequent occurrences. As noted in Ta-
fectively uncorrelated to changes in traffic stop levels, as ble 1, arrest rates were highest for suspended or revoked
indicated by the flat slope of the red trend line. licenses, or for drug crimes.[18] Custodial arrests which
might be suspected to have a direct impact on future
To more quantitatively examine the short-term relation- crime (e.g., those arrests which are not solely for hold-
ship between non-moving violation stops and crime lev- ing an invalid license, for minor marijuana possession,
els, we fit a Poisson regression model. Specifically, given for public misconduct, or for driving violations) occur in
a crime count yg,t in RPA g in week t, we aim to esti- 0.7% of non-moving violation stops. A larger percentage
mate the relationship with normalized[15] stop counts sg,t every one standard deviation increase in stop activity. The point es-
in the same RPA and week. We include the RPA’s popu- timate is statistically significant when using robust standard errors;
however, the estimated e↵ect is not statistically significant under
lation pg as a baseline, normalized counts of the previous an alternative over-dispersed Poisson model. It is also possible that
week’s crimes and stops, coefficients g for each geog- the result is driven by an unmeasured confounding variable that
raphy, and ✓m[t] for the month in which week t occurs. correlates both with stop activity and crime rates.
[17] When a custodial arrest leads to multiple charges, we count
Accordingly, we fit the following regression model: only the most severe charge per incident, using the following hi-
erarchy: serious crime (assault, burglary, theft, sex o↵ense, child
·yg,t ·sg,t crimes), drug crimes (non-marijuana charges, or possession of at
yg,t ⇠ Poisson(pg · e↵·sg,t + 1+ 1 + g +✓m[t]
). least 0.5 oz of marijuana), DUI, minor marijuana possession (less
than 0.5 oz), FTA/parole violation/warrant (also includes probation
The fitted model suggests that stops do not decrease violations and FTB), public misconduct (public intoxication, disor-
crime (↵ˆ = 1.03, 95% CI: (1.01, 1.04)), confirming our derly conduct, vandalism, trespassing), driving violations, plate al-
terations, license charges (suspended/revoked license, driving with
intuition from the graphical representation in Figure 7.[16] no license).
[18] Only 51% of non-moving violation stops that led to a custo-
[15] Stop and crime counts are normalized for each RPA by sub-
dial arrest matched a corresponding arrest record. Values reported
tracting the mean count for that RPA and dividing by the standard in Table 1 are over the subset of these matched arrests. The cover-
deviation of that count. age for all stops that led to custodial arrest was 56%. The coverages
[16] The fitted model results in a small positive coefficient on stop for all stops and for non-moving violation stops that led to misde-
levels, indicating—counterintuitively—that crime increases 3% for meanor citations were 89% and 91%, respectively.
Stanford Computational Policy Lab Page 8 of 9

(6.6%) of non-moving violation stops led to misdemeanor 100%


citations. However, the majority of these citations were
for license-related charges: 82%[19] of non-moving vio- 90%
lation stops that led to a misdemeanor citation included
only a license-related charge, and no other charge. An ad- 80%
ditional 0.7% of non-moving violation stops resulted in
70%
the recovery of other contraband (typically drugs), but
did not include a custodial arrest. Altogether, 2.2% of

Percent of stops
60%
non-moving violation stops resulted either in a custodial
arrest or the recovery of contraband. 50% 17% of officers make 50% of NMV stops
Quantifying the benefits of such stop outcomes is be-
yond the scope of this report. We note, however, that it is 40%
possible that other police activity may be a more e↵ective
use of time. For example, 16% of investigatory stops— 30%
which require that officers have reasonable and articula-
ble suspicion of criminal activity—resulted in a custodial 20%
arrest or contraband recovery, a rate almost eight times
higher than the corresponding rate for non-moving vi- 10% 1% of officers make 9% of NMV stops
olation stops. This di↵erence suggests the MNPD may
be able to more e↵ectively achieve the arrests and con- 0%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
traband recoveries from non-moving violation stops with Percent of officers
other enforcement e↵orts.
Figure 8: The distribution of the number of non-moving
Officer-level di↵erences in stop activity violation stops across MNPD officers in 2017, illustrating
As one might expect, there are significant di↵erences that a small number of officers conduct the majority of
in stop rates across officer assignments. For example, of- such stops.
ficers assigned to flex units—whose duties allow for more
proactive policing—conduct about twice as many non- possible that the MNPD gives officers wide leeway to en-
moving violation stops per officer (217 stops per officer gage in proactive policing, which in turn may result in
in 2017) as patrol units (109 per officer). Such di↵erences the observed heterogeneity. It is also possible that these
ostensibly reflect the discretion that flex officers have in officer-level di↵erences are part of an intentional polic-
carrying out proactive policing duties. Similarly, officers ing strategy, though we are unaware of any such policy
working evening shifts make more such stops than those directives. Regardless of the underlying reason, the rela-
working during the day, likely in part because certain non- tively small number of officers involved makes it easier for
moving violations—like broken lights—are more visible at the department to understand and appropriately address
night. their behavior as necessary.
More surprisingly, however, we find that a relatively
small number of officers conduct the vast majority of
non-moving violation stops. For example, as shown in Discussion
Figure 8, the 10 most active flex and patrol officers made Based on an extensive analysis of the MNPD’s 2017
9,399 stops, or approximately 9% of all non-moving viola- traffic stop data, we find that black drivers were stopped
tion stops over the year; further, half of all non-moving vi- substantially more often than white drivers; these dis-
olation stops were conducted by 17% of active officers— parities were particularly pronounced among stops for
125 officers in total.[20] We find similar patterns when we non-moving violations, such as broken tail lights and ex-
disaggregate by assignment. For example, among patrol pired registration tags. The racial disparities in these non-
officers working the night shift, 15% made 50% of stops. moving violation stops are in part attributable to the con-
It is unclear why such a small group of officers carries centration of stops in high-crime areas, which in Nashville
out the majority of stops. As in many jurisdictions, it is often coincide with predominantly black neighborhoods.
The defensibility of such a policing strategy, however,
[19] This number considers as a baseline only the 91% of non- rests on its e↵ectiveness in ensuring public safety. In this
moving violation stops that matched an arrest record, since for case, we found that traffic stops—including stops for non-
the remaining 9% we do not have data on charges. Implicit in
this computation is an assumption that the remaining 9% have moving violations—had no discernible e↵ect on serious
similar charge distributions as the 91%. We can set a lower bound crime rates, and only infrequently resulted in the recov-
on this estimate by assuming that none of the 9% were license- ery of contraband or a custodial arrest.
only charges, and an upper bound by assuming that all of the 9%
were license-only charges. With this, we conclude that the number These results suggest that the MNPD could safely re-
of non-moving violation misdemeanor citations that were charged duce overall stop rates. In particular, curtailing stops for
with only a license-related charge lies between 74% and 84%.
[20] For this analysis, we consider “active” officers to be flex and non-moving violations could reduce racial disparities, par-
patrol officers who conducted at least one stop during 2017, to tially addressing community concerns about its policing
avoid counting those assigned to administrative duties. These gen- practices. However, in order to bring Nashville’s stop rates
eral patterns hold when when we use a more stringent definition to the level of similar American cities, the MNPD would
of “active”. For example, among flex and patrol officers who car-
ried out at least 10 non-moving violation stops in 2017, 19% were have to significantly reduce the number of such stops it
responsible for half of stops. carries out (Figure 1). A reduction of even 50% in non-
Stanford Computational Policy Lab Page 9 of 9

moving violation stops would still leave the city’s overall special thanks to Maria Ponomarenko for providing exten-
stop rate twice as high (or higher) than other peer cities. sive and invaluable perspective throughout our analysis.
A more substantial 90% reduction in such stops would put We also thank Mark Cohen, Barry Friedman, Robert C.
Nashville on par with peer cities with the highest stop Haas, Farhang Heydari, and Crystal Yang for their helpful
rates. These reductions would have significant impact feedback. Finally, we thank the Metropolitan Nashville
on the day-to-day lives of Nashville residents. Assuming Police Department, including the MNPD Crime Analy-
the MNPD reduced non-moving violation stops by 90%, sis Unit, for their transparency and cooperation on this
and changed nothing else, roughly 100,000 stops—52,000 project.
stops of white drivers, 40,000 stops of black drivers, 6,000 References
stops of Hispanic drivers, and 2,000 stops of drivers of 1. Geo↵rey Alpert, Michael Smith, and Roger Dunham. Toward a better
other races—would be avoided each year. The disparity benchmark: Assessing the utility of not-at-fault traffic crash data in
between overall black and white stop rates would also racial profiling research. Justice Research and Policy, 6(1):43–69, 2004.
2. Kate Antonovics and Brian Knight. A new look at racial profiling:
drop substantially, from 44% to 28%. Evidence from the Boston Police Department. The Review of
The remaining disparities largely result from di↵erences Economics and Statistics, 91(1):163–177, 2009.
3. Shamena Anwar and Hanming Fang. An alternative test of racial
in stop rates for moving violations. In particular, black prejudice in motor vehicle searches: Theory and evidence. The
drivers were stopped for moving violations (which com- American Economic Review, 2006.
prised half of all traffic stops) 24% more often than white 4. Gideon’s Army. Driving while black: A report on racial profiling in
Metro Nashville Police Department traffic stops, 2016. URL:
drivers last year. We expect that reducing such stops will https://drivingwhileblacknashville.files.wordpress.com/
not adversely impact crime levels, though they could have 2016/10/driving-while-black-gideons-army.pdf.
other unintended consequences. For example, one con- 5. Ian Ayres. Outcome tests of racial disparities in police practices.
Justice Research and Policy, 4(1-2):131–142, 2002.
cern is the possible e↵ect of traffic stop reductions on 6. Elizabeth Davis, Anthony Whyde, and Lynn Langton. Contacts
traffic safety. This may be an issue in Nashville, where between police and the public, 2015. Technical report, U.S.
traffic accidents per capita increased by roughly 60% be- Department of Justice, Bureau of Justice Statistics, 2015.
7. Kasey Henricks and Daina Cheyenne Harvey. Not one but many:
tween 2011 and 2017. As such, reductions in moving vi- Monetary punishment and the Fergusons of America. In Sociological
olations require balancing the potential impacts on traf- Forum, volume 32, pages 930–951. Wiley Online Library, 2017.
fic safety with broader community concerns. In contrast, 8. Rebecca Hetey, Benoı̂t Monin, Amrita Maitreyi, and Jennifer
Eberhardt. Data for change: A statistical analysis of police stops,
most non-moving violation stops are for minor traffic in- searches, handcuffings, and arrests in Oakland, Calif., 2013-2014.
fractions, like a broken tail light, a broken license plate Technical report, Stanford University, SPARQ: Social Psychological
light, or an expired registration. We thus expect one could Answers to Real-World Questions, 2016.
9. James Lange, Mark Johnson, and Robert Voas. Testing the racial
safely reduce non-moving violation stops by continuing profiling hypothesis for seemingly disparate traffic stops on the New
to enforce the most serious such o↵enses (e.g., broken Jersey Turnpike. Justice Quarterly, 22(2):193–223, 2005.
headlights) while eliminating stops with a less immediate 10. Richard J Lundman and Robert L Kaufman. Driving while black:
E↵ects of race, ethnicity, and gender on citizen self-reports of traffic
connection to traffic safety. Finally, we note that reduc- stops and police actions. Criminology, 41(1):195–220, 2003.
tions in traffic stops may also reduce opportunities for 11. Michael A Medina Jr. Accounting for membership: an examination of
officers to engage the public, although there are arguably the e↵ects race, sex, and age has on police conduct and force during
traffic stops. PhD thesis, Rutgers University–Camden Graduate
other more appropriate channels for community contact. School, 2016.
Our analysis illustrates the power of a data-driven ap- 12. Emma Pierson, Sam Corbett-Davies, and Sharad Goel. Fast threshold
tests for detecting discrimination. In International Conference on
proach to public policy. Looking forward, more extensive Artificial Intelligence and Statistics, pages 96–105, 2018.
data could yield further insights. For example, we found 13. Emma Pierson, Camelia Simoiu, Jan Overgoor, Sam Corbett-Davies,
inconsistencies in how police searches were classified in Vignesh Ramachandran, Cheryl Phillips, and Sharad Goel. A
large-scale analysis of racial disparities in police stops across the united
the data we analyzed, making it difficult to carry out sta- states. arXiv preprint arXiv:1706.05678, 2017.
tistical tests for racial bias in search decisions [12, 13, 19]. 14. Greg Ridgeway. Assessing the e↵ect of race bias in post-traffic stop
Additionally, inconsistent incident identification numbers outcomes using propensity scores. Journal of Quantitative
Criminology, 22(1):1–29, 2006.
made it difficult to fully link traffic stops to arrest records. 15. Greg Ridgeway and John MacDonald. Doubly robust internal
Finally, many of the categories that the MNPD uses for benchmarking and false discovery rates for detecting racial bias in
traffic stops are relatively coarse. For instance, equip- police stops. Journal of the American Statistical Association,
104(486):661–668, 2009.
ment violations can include both plate light violations 16. Je↵ Rojek, Richard Rosenfeld, and Scott Decker. The influence of
and headlight violations, despite their potentially di↵er- driver’s race on traffic stops in Missouri. Police Quarterly,
ent impacts on traffic safety. Finer classification would 7(1):126–147, 2004.
17. Matt Ryan. Frisky business: race, gender and police activity during
improve the department’s capacity to monitor changes in traffic stops. European Journal of Law and Economics, 41(1):65–83,
enforcement over time, and would be useful information 2016.
to help the MNPD safely curtail traffic stops. We hope 18. Stephanie Seguino and Nancy Brooks. Driving while black and brown
in Vermont, 2017.
our analysis, and these suggestions for future data col- 19. Camelia Simoiu, Sam Corbett-Davies, and Sharad Goel. The problem
lection, help both the MNPD and the broader Nashville of infra-marginality in outcome tests for discrimination. The Annals of
community design more e↵ective and equitable policing Applied Statistics, 11(3):1193–1216, 2017.
20. Michael Smith and Matthew Petrocelli. Racial profiling? A multivariate
strategies. analysis of police traffic stop data. Police Quarterly, 4(1):4–27, 2001.
21. Tom R Tyler, Je↵rey Fagan, and Amanda Geller. Street stops and
police legitimacy: Teachable moments in young urban men’s legal
Acknowledgements socialization. Journal of Empirical Legal Studies, 11(4):751–785, 2014.
This work was conducted in collaboration with the 22. Patricia Warren, Donald Tomaskovic-Devey, William Smith, Matthew
Zingra↵, and Marcinda Mason. Driving while black: Bias processes and
Policing Project at New York University School of Law. A racial disparity in police stops. Criminology, 44(3):709–738, 2006.

You might also like