CRITICAL REVIEW 2

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Offi cer-Involved Gun Deaths 217

Public Administration Review,

Vol. 77, Iss. 2, pp. 217–226. © 2017 by

The American Society for Public Administration.

DOI:10.1111/puar.12738.

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies

and Rates of Officer-Involved Gun Deaths

Meghan E. Rubado is assistant

professor of public administration and city

management in the Maxine Goodman

Levin College of Urban Affairs at Cleveland

State University. Her research examines

how institutions shape the decision-

making process and affect local policy

outcomes. Her primary interests are urban

service delivery, local governance, and

environmental policy.

E-mail: [email protected]

Jay T. Jennings is a postdoctoral

research fellow in the Annette Strauss

Institute for Civic Life at the University of

Texas at Austin. His research focuses on the

relationships between institutions and the

capacity of citizens to effectively participate

in politics. His work draws on the fields of

political psychology, public opinion, and

political communication.

E-mail: [email protected]

Jay T. Jennings University of Texas at Austin

Meghan E. Rubado Cleveland State University

Abstract : Killings of civilians by police officers have become a matter of intense public concern in the United States. High-profile deaths, especially those of black citizens, have caused outrage and sparked the Black Lives Matter move- ment with calls for dramatic changes in how police agencies operate. However, little systematic research exists to answer questions about which policies should be ended or put in place to reduce these deaths. The authors leverage a large data set of gun deaths by police officers in the United States, combined with agency-level policy data and community demographic data, to examine whether certain policies are associated with lower or higher rates of officer-involved gun deaths. Findings show that one policy—the requirement that officers file a report when they point their guns at people but do not fire—is associated with significantly lower rates of gun deaths.

Practitioner Points • Police agencies that require officers to file a report when they point their guns but do not shoot at civilians

had significantly lower rates of gun deaths by police officers. Roughly half of agencies required such reports in 2013.

• This requirement may cause officers to use more caution when considering using force by pointing their guns and may demonstrate to officers an agency commitment to best practices that protect civilians from unnecessary use of force.

• The reporting requirement for displaying a gun is not associated with higher rates of gun deaths of police officers, suggesting that the policy does not endanger officers.

• While community policing and racial representation on police forces have been tied to improved outcomes in several areas, we do not find that these factors are associated with lower rates of police-involved gun deaths.

Fatal shootings of civilians by police officers in the past several years have drawn the attention of the news media, protesters, and investigators from all levels of government. Public outcry from individuals and organizations, including the Black Lives Matter movement, has called for major changes in police agencies that result in reductions in these deaths, especially those of black citizens. But which police agency policies are associated with higher incidence of fatal police-involved shootings? What changes could public officials make to reduce these fatalities? These questions have received little in the way of systematic research by scholars. Given the continual loss of civilian lives in these incidents and the recent unrest that has followed them in cities from Cleveland to Minneapolis, Baltimore to Ferguson, these questions warrant attention from scholars, public leaders, and American society in general.

We suggest that certain policy choices by local officials may reduce the rates of these deaths, and we test

our hypotheses on a large sample of U.S. cities and counties. We find that one policy—the requirement that police officers file a report when they point their gun at someone but do not fire—is associated with systematically lower rates of gun deaths of civilians.

Our analysis relies on a unique data set compiled by a nonprofit organization that has attempted to catalog every instance of police-involved gun death since 2000. These are cases in which a police officer fired at a civilian, resulting in death. This database project, called Fatal Encounters, is crowdsourced and fact-checked by the nonprofit. Fatal Encounters and several less organized databases like it have sprung up in recent years as a citizen solution to the lack of reliable government data. The federal government does not mandate collection of these data in a public database. The Federal Bureau of Investigation (FBI) does collect information on police-involved deaths of civilians, but these data are systematically incomplete because of voluntary self-reporting by state and local

218 Public Administration Review • March | April 2017

agencies (Burghart 2014 ; Davis and Lowery 2015 ). We merge the Fatal Encounters data with agency-level policy variables gathered by the U.S. Bureau of Justice to examine whether particular policies or department features are associated with higher or lower rates of gun deaths by police officers. Understanding which agency variables are associated with reduced incidence may provide police agencies with options to reduce their rates of police-involved deaths and help direct policy change by local and state legislative bodies. Additionally, it will provide a foundation for future academic research on an extremely important, although neglected, topic.

Existing scholarly research suggests that the use of force by police is affected by factors at multiple levels, including the individual, agency, and community. For example, racial diversity within communities (Jacobs and O ’ Brien 1998 ) and violent crime rates (Smith 2003 ) have been found to affect the frequency of fatal police-involved shootings. City-level policies such as the presence of a citizen review board have also been associated with police brutality complaints (Smith and Holmes 2003 ). Other factors contributing to the use of force more generally include the racial or ethnic minority status of the citizen(s) involved, education level of the police officer, income of the police officer, and training efforts by the police agency. Our work here focuses on agency policies and characteristics that are associated with relatively low or high rates of gun deaths by police, although we control for important community characteristics as well. Recent deaths of civilians by police have drawn public attention to strained relations between government and citizens, particularly citizens of color. We seek to illuminate potential policy options that may help reduce the rates of these deaths—a normatively positive outcome from the perspective of leaders, police, and citizens.

Much of the past research on the relationship between police use of force and agency- or community-level factors has relied either on case studies of a single police agency (e.g., McElvain and Kposowa 2008 ; Oberfield 2012 ) or on FBI data on police-involved homicides, creating concern regarding the external validity of findings. This article represents a significant contribution, as we leverage a far more complete database of police-involved gun deaths to test which agency policies are associated with fatal police shootings of civilians. The article reviews the literature, lays out hypotheses related to police-involved gun deaths, discusses the data used for the analysis, and reports results from statistical models.

We hypothesize that agencies with certain policies will have lower rates of gun deaths. These include hiring police forces that better match the racial makeup of communities, requiring paperwork of officers who draw their guns and point them at civilians, assigning regular beats to officers, and requiring more community policing training. Of these policies, we find that only one has a significant relationship with the rate of gun deaths: the requirement that officers document incidents in which they draw their firearms. Agencies that required this documentation had lower rates of officer-involved gun deaths on average. This finding was remarkably robust and quite substantively significant.

We interviewed leaders of several police agencies about this policy to better understand how it works and why some agencies put it in place while others do not. Our research suggests that the paperwork requirement may act as a deterrent to unnecessary use of force because of the added demand on officers’ time, the implicit recommendation from leadership of avoiding unnecessary gun draws, and a general commitment to best practices among agencies that have this policy. Consistent with expectations, we also find that several control variables affected gun death rates by police. For example, agencies situated in communities with higher levels of economic inequality and lower education levels had more police- involved gun deaths per capita.

The Literature: What Predicts the Use of Police Force, Especially Deadly Force? Past research on what influences officers to use force against civilians typically focuses on one of three major sets of explanatory variables or some combination of these: personal or psychological, organizational, and situational (Friedrich 1980 ; Worden 1996 ). Personal or psychological explanations center on individual officers and their backgrounds, personalities, and demographic characteristics. Organizational explanations focus on policies, structures, and cultures of police agencies that may alter individuals’ tendencies to use force. Situational explanations place emphasis on the contextual circumstances of an event in which force might be used, such as the neighborhood, time of day, and race of the civilian(s) involved.

In a two-year study of 80 new police recruits in an unnamed city, Oberfield ( 2012 ) finds that while organizational features, such as training and culture, can have significant effects on officers’ attitudes about using force, these effects are short-lived, and the officers return to their previous views. Meanwhile, personal characteristics were significant and lasting predictors of views about using

force. These included the recruits’ views on the use of force prior to entering the police academy, family income, status as former military, age, and race. At the last point in the survey, for example, officers who had entered the academy thinking that the use of force was effective were significantly more likely to hold that view, as were white officers. At various points in the study, higher income was associated with lower levels of agreement with the view that the use of force is effective. This finding is in line with previous research that supports a relationship between individual characteristics of police officers and their behavior.

Among those characteristics associated with higher incidence of using force are low levels of education and experience of police officers (Terrill and Mastrofski 2002 ; Worden 1996 ), positive views of the use of force (Oberfield 2012 ; Worden 1996 ), and negative impressions of citizens (Worden 1996 ). Research on contextual or situational factors associated with the use of force finds that characteristics of the citizens matter. For example, officers were more likely to use force when citizens were young or poor (Terrill and Mastrofski 2002 ), racial or ethnic minorities (Alpert and Dunham 2004 ; Fyfe 1982 ; Garner and Maxwell 2002 ; Meyer 1980 ; Worden 1996 ), located in disadvantaged neighborhoods

Recent deaths of civilians by police have drawn public

attention to strained rela- tions between government and citizens, particularly citizens of

color.

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Offi cer-Involved Gun Deaths 219

(Terrill and Reisig 2003 ), or viewed as disrespectful or antagonistic (Friedrich 1980 ; Gallo, Collyer, and Gallagher 2008 ). Other research has shown that heterogeneous communities suffer from lower levels of trust and increased levels of intergroup conflict that may elevate police–citizen conflict (Putnam 2007 ; Sampson and Groves 1989 ).

The Role of Race While race has been at the center of many recent conflicts between officers and protesters in Ferguson, Baltimore, and New York, previous research has not found that more racially diverse agencies have different types of arrest behavior (Riksheim and Chermak 1993 ), lower instances of police brutality (Cao and Huang 2000 ), or fewer instances of police-involved deaths (Smith 2003 ). However, research has consistently shown that minorities, particularly African Americans, have significantly higher rates of interaction with law enforcement (Baumgartner et al. 2016 ; Harris 1999 ; Provine 2007 ). Several studies have found that the share of a community ’ s population that is black is associated with higher incidence of police-involved deaths and other outcomes, such as traffic stops that result in a search (Petrocelli, Piquero, and Smith 2003 ; Smith 2003 , 2004 ; Smith and Holmes 2003 ). These findings support what some scholars refer to as the “racial threat” hypothesis, which is drawn from the conflict theory of law and posits that police officers use force as a response to a perceived threat from racial or economic groups that are viewed as threats to the existing social order (Harring et al. 1977 ; Smith 2003 ).

A suggested connection, particularly after the Ferguson protests, is the relationship between the racial composition of the community and the agency serving the community. Theobald and Haider- Markel ( 2009 ) do find a strong connection between the race of the arresting officer and the race of the citizen when considering the legitimacy of the arrest. When a white officer arrests a black citizen, black citizens are significantly less likely to view the arrest as legitimate. In a study of Welsh and English police forces, Hong ( 2017 ) finds that an increase in the proportion of officers who are ethnic minorities is associated with a subsequent reduction in officer misconduct. Black citizens, Hong finds, make fewer misconduct complaints against ethnically diverse police forces.

The Role of Agency Policies Evidence about the association between agency policies and the use of force is less developed, but several scholars have found that police agency decisions can make a difference. In a study of criminal justice reforms after 1950, Walker ( 1993 ) finds that agency policies aimed at controlling the discretion of police officers can significantly reduce instances of officer-involved shootings. A study on police- involved deaths shows that agencies with more stringent recruitment guidelines had lower incidence of use of deadly force (Willits and Nowacki 2014 ). Another shows that officers working in more highly bureaucratized police agencies were more likely to use “reasonable” force, although there was no relationship between bureaucratization levels and use of “unreasonable” force (Worden 1996 ).

A perplexing finding in two studies suggests that the required number of hours of field training is positively associated with

homicides by police (Bailey 1996 ; Smith 2004 ). That is, agencies that required more training hours had higher rates of police- involved deaths. Smith ( 2004 ) notes that more systematic study of organizational predictors of police killings, particularly with respect to training, is needed before drawing conclusions about the meaning of this finding. Fyfe ( 1982 ) finds some evidence that restrictive shooting policies by agencies decreased the number of police-involved deaths in New York City. While nearly all agencies require documentation of instances when a firearm is discharged, only 46 percent require documentation of instances when the firearm is displayed but not fired (U.S. Department of Justice 2013 ).

It has become common for police agencies to introduce a variety of procedures and policies to foster a more positive relationship with communities. Community policing policies and training are generally considered to decrease crime and strengthen the relationship between the community and law enforcement agencies (Cordner 2014 ). While it would be expected that these positive relationships would result in fewer fatal encounters between police and the community, there is no empirical evidence to support that hypothesis. Furthermore, it is unclear to what degree community policing tactics are being fully implemented. The 2013 Law Enforcement Management and Administrative Statistics show that a large proportion of agencies train recruits in community policing tactics, but a smaller proportion provide training for nonrookies. Specifically, the share of agencies providing at least half of their recruits with this training is 65 percent, while the share providing the training to at least half of all full-time officers is 49 percent.

Hypotheses We focus our analysis on agency-level policies that we expect may be associated with the rate of gun deaths by police officers. First, given

expectations about the racial match of citizens and police officers and the past finding that black citizens view arrests of black citizens by white officers as less legitimate (Theobald and Haider-Markel 2009 ), we expect that departments with police rosters that are disproportionately white compared with their

citizenry will be associated with more strained community relations and conflict. This increased conflict and reduced legitimacy is expected to produce higher rates of gun deaths by police. Rosters that better match the racial composition of the community served are expected to have lower rates of these fatalities.

Hypothesis 1: Police agencies with a racial composition of police officers that better matches the community served will have lower rates of gun deaths.

By implementing a policy in which officers are required to complete a report any time their firearm is displayed (but not discharged), agencies are more strictly regulating the use of firearms by officers. According to our interviews with leaders of police agencies, such policies mean that when officers point their guns at civilians, even momentarily, they must file reports documenting that their guns were displayed and explaining the circumstances. These policies are often nested within agencies’ “use of force” policies on the

When a white offi cer arrests a black citizen, black citizens are signifi cantly less likely to view

the arrest as legitimate.

220 Public Administration Review • March | April 2017

grounds that pointing a gun at a person is a use of force (personal communications, May–June 2016).

We believe the mechanism connecting firearm display documentation and officer-involved citizen deaths is a two-step process. The first step is the documentation requirement ’ s role in decreasing the frequency with which officers draw their firearms. The second step is increased officer-involved deaths once a firearm has been displayed by an officer. So, these policies are expected to lead to lower rates of gun deaths through their effect of discouraging unnecessary gun draws by officers.

We expect the policy to discourage the drawing of officers’ guns in two key ways. First, by simply instituting the policy and requiring officers to document the display of their firearms, policy makers and agency heads are recommending that officers take more seriously the act of drawing their weapons and pointing them at civilians. Research on the implementation of policies requiring changes in documentation standards has shown that people interpret such policies as recommendations from policy makers (Johnson and Goldstein 2003 ; Madrian and Shea 2001 ). In the case of firearms, while each officer may interpret this implicit recommendation differently, requiring documentation when a firearm is displayed becomes a de facto policy discouraging the unnecessary use of firearms.

Second, in decision making, people give more weight to potential losses than to potential gains (Tversky and Kahneman 1991 ). Humans tend to be loss adverse (Thaler and Sunstein 2008 ). Most people view paperwork as both a loss of time and a negative experience. Viewing the documentation as a punishment or a loss of something valuable (time), we can see that the policy may make officers susceptible to loss aversion. If an officer views the documentation as a loss, then the officer is less likely to draw his or her firearm because this action will lead to a certain loss, even if the potential gains (increased personal safety of the officer) are more valuable.

It is also possible that the inclusion of this policy is most common among agencies that stay current with best practices and trends within the policing profession. In fact, in our interviews with police agency leaders, several noted that their agencies had added this policy in order to maintain or achieve certification by an outside organization or to adopt the recommended policy of a consulting firm. For example, the Rome Police Department in Georgia has required reports for pointing a gun without firing for at least 25 years and maintains this policy and other best practices as part of its commitment to earning certification by the Commission on Accreditation for Law Enforcement Agencies (M. Tyson, private communication, May 31, 2016). In another case, the Flagstaff Police Department each year tailors state- specific template law enforcement policy produced by Lexipol, which includes this policy of mandatory reporting for pointing a firearm (D. Musselman, private communication, May 31, 2016). Agencies adhering to a broad set of best practices, such as the firearm display documentation requirement, may also foster the development of a police force that is more committed to

minimizing the use of force and therefore better able to reduce police-involved deaths.

Hypothesis 2: Agencies with a policy requiring a written report for instances of the display of an officer ’ s firearm will have lower rates of police-involved deaths.

Community policing policies and training are expected to reduce crime and foster improved relationships between agencies and the communities they serve (Cordner 2014 ). We expect this positive relationship to lead to better interactions between officers and civilians, fewer violent conflicts involving police, and, therefore, reduced incidence of police-involved gun deaths. Community policing focuses on building connections between

police agencies and individuals and groups in the communities. The idea is to build partnerships and trust so that the police agencies and communities are working together for better neighborhoods. If the intended goal is realized, officers will have more positive connections in the community, making incidents that result in gun death less

likely. Community policing policies are associated with police beats, in which officers are assigned to regular neighborhoods for long durations. Under these conditions, we expect officers to know the community members better, to feel more connected to them, and therefore to be more likely to hesitate to use deadly force in certain situations.

Hypothesis 3: Requiring community policing training will lead to lower rates of police-involved deaths.

Hypothesis 4 : Assigning officers to regular beats will lead to lower rates of police-involved deaths.

Data and Method This article incorporates three data sources: the Fatal Encounters Project (FE), the U.S. Bureau of Justice Statistics’ Law Enforcement Management and Administrative Statistics (LEMAS), and the U.S. Census Bureau ’ s American Community Survey. Data from the three sources were combined to create a single data set in which the unit of analysis is the law enforcement agency. The Fatal Encounters Project is a nonprofit organization that has been tracking police- involved deaths, with data going from 2000 through the end of 2015. Through a combination of paid researchers, volunteers, and Internet crowdsourcing, FE has collected records on more than 10,000 police-involved deaths in that 15-year time period. All instances are sourced through media reports, Freedom of Information Act requests, or agency reporting. 1 While this data set cannot claim to be complete or without some errors, it is the most complete data set of police-involved deaths in the United States. While the FBI does collect data on police-involved deaths, the reporting by agencies is voluntary and some states do not report data at all (U.S Department of Justice 2015).

For a variety of reasons, the data presented in this article do not represent the total number of police-involved deaths in the FE database. First, only gun-related deaths are included in our data set, as the other types of deaths are less clearly defined as intentional

Community policing focuses on building connections between police agencies and individuals and groups in the communities.

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Offi cer-Involved Gun Deaths 221

uses of deadly force. For instance, many people in the data set died as a result of being Tasered. It is difficult to assume that the officer chose to use deadly force in those instances because typically a Taser does not kill someone. Even murkier were cases of vehicular deaths and asphyxiation while in police custody. Next, cases in which multiple law enforcement agencies were involved in the death had to be eliminated. Because the unit of analysis is the agency, these deaths were not included because of an attribution problem. As we were primarily interested in local agencies, state and federal agency- related deaths were also eliminated. In a small number of instances in which the agency responsible for the death in the FE data could not be matched with an agency in the LEMAS, those agencies were not included in our data set. Finally, for the analysis presented in this article, only agencies serving communities with populations over 25,000 have been included. 2 This leaves us with a total of 5,141 police-involved deaths across 1,107 law enforcement agencies from 2000 to 2015. Our data set is not time serial; rather, it is cross- sectional with a count of police-involved deaths over the 15-year time period, standardized by population. The data include police agencies for communities of various sizes, including the nation ’ s largest cities and counties, as well as small cities and other municipalities.

The dependent variable in our analysis is the number of officer- involved deaths for every 100,000 people in the community served by the police department. This variable ranged from 0 to nearly 13, with a mean of 1.75. While many large agencies had particularly high death rates, other big cities and counties had comparably low rates (see table 1 ). Meanwhile, several of the agencies with the highest death rates were relatively small and medium-sized cities (see table 2 ). Figure 1 shows the highest death rates for the largest agencies in our data set, those serving populations over 500,000.

The data used to construct our agency-level independent variables come from the LEMAS study of 2013. That study,

funded by the U.S. Department of Justice and produced by the Inter-University Consortium for Political and Social Research, provides organizational data for more than 3,000 law enforcement agencies, including all those with 100 or more sworn officers and a representative sample of smaller agencies. The data set allowed for construction of variables on agency policy regarding required documentation for display of firearm, training in community policing, and assignment of officers to a regular beat. It also provided data on racial composition of the agency, which were used to create a variable for the degree to which the agencies match their communities.

To test hypothesis 1, our model includes a measure of white representation in the agency compared with white representation in the community. This variable simply takes the percentage of the community that is white and subtracts the percentage of full-time officers who are white. So, when the number is positive, it indicates that whites are underrepresented in the agency, and when it is negative, whites are overrepresented in the agency. Agencies with negative values, then, have police forces that are less racially diverse than the communities they serve. To test hypothesis 2, the model includes a variable indicating the departmental policy on a required report when the officer displays a firearm. As can be seen from the summary statistics (found in appendix A in the online version of this article), about half of the agencies in our data set require a written report if an officer displays a firearm. For the purposes of testing hypotheses 3 and 4, the model also includes variables for the proportion of officers who receive in-service community policing training and whether the agency assigns officers to specific beats or areas of the community.

We merged the LEMAS police agency data with local-level demographic data on the community served, including population counts, race, education levels, and income from the U.S. Census Bureau ’ s American Community Survey using place and county data sets of five-year estimates for 2013. These data were obtained using the National Historical Geographic Information System (Minnesota Population Center 2016 ). 3 Control variables created from these data include the percentage of the population that is black, percentage of the population with a higher education degree, a Gini index of income inequality, and an index of racial homogeneity. Past work suggests there may be a positive relationship between the percentage of the community that is black and the incidence of police use of force because of a perceived threat to the existing social order. We also expect poorer, more racially diverse, and more economically unequal localities to experience higher rates of fatalities by police. Given that police shootings of civilians occur more frequently in larger cities (Willits and Nowacki 2014 ) and that many studies have restricted analysis to large cities (e.g., Smith 2003 ; Smith and Holmes 2003 ), we include a control variable for log population. We expect larger cities by population to have higher rates of gun deaths per capita. We also include a control variable for the percentage of the population that lives in an urban area. All else being equal, we expect less urban communities to be associated with lower rates of gun deaths by police. Finally, the percentage of the community that is non- Hispanic white also was used for creation of the agency-level variable that measures the racial composition match between the agency and the community.

Table 1 Top 10 Agencies by Population of Community Served

State Population Death Rate (percent)

Los Angeles County Sheriff California 9,893,481 1.10 New York City Police New York 8,268,999 2.26 Cook County Sheriff Illinois 5,212,372 0.04 Harris County Sheriff Texas 4,182,285 1.29 Los Angeles Police California 3,827,261 5.77 San Diego County Sheriff California 3,138,265 0.80 Orange County Sheriff California 3,051,771 0.16 Chicago Police Illinois 2,706,101 4.62 Miami-Dade County Police Florida 2,549,075 3.41 Dallas County Sheriff Texas 2,412,481 0.04

Table 2 Top 10 Agencies with Highest Police-Involved Gun Death Rates

State Death Rate (percent) Population

New Orleans Police Louisiana 12.60 357,013 Redding Police California 12.17 90,370 Inglewood Police California 11.76 110,585 Anderson Police S. Carolina 11.24 26,699 Midland Police Texas 11.22 115,903 Sierra Vista Police Arizona 11.07 45,187 Jackson Police Mississippi 10.92 173,997 Tucson Police Arizona 10.89 523,278 Salt Lake City Police Utah 10.63 188,141 Suisun City Police California 10.58 28,350

222 Public Administration Review • March | April 2017

We include a control variable for city departments. Compared with other types of police agencies, city departments tend to cover all of the territory within their boundaries. Because county agencies are typically layered over city departments, they may patrol only the territory that is not covered by a city police force. Therefore, we expect city departments to be more likely to patrol the most densely populated parts of a region, to have more interactions with civilians, and, therefore, to have higher incidence of police-involved shootings.

The model also includes a variable for the percentage of full-time officers in the agency who are female. While research has generally found that female officers are less likely to be accused of excessive force (Spillar, Harrington, and Wood 2000 ) or use deadly force (McElvain and Kposowa 2008 ), other research has shown that agencies with higher percentages of female officers have higher rates of police-involved deaths (Smith 2003 ). Summary statistics for all model variables may be found in appendix A in the online version of this article, while a table describing each variable is provided in appendix B online.

As our dependent variable is a death rate with a fairly large and continuous range, we use ordinary least squares (OLS) regression to

analyze the data. Because state laws and rules are likely to affect the outcome variable, we include state fixed effects. It should also be noted that our dependent variable is constructed from data ranging from 2000 to 2015, while our agency and community independent variables are from 2013 data. Therefore, it is possible that certain agency policies were enacted between 2000 and 2013, and these cases would obfuscate the impact of the policy in our findings. Because of this, nonsignificant results cannot be interpreted as a clear finding of policy ineffectiveness.

Results Results are shown in table 3 . Of the four policies that we expected to be associated with higher or lower gun death rates, one turned out to be a significant predictor of gun death rates: the policy requiring a written report for drawing and pointing a firearm. Consistent with hypothesis 2, the firearm reporting requirement had a negative relationship with death rates. The policy of requiring officers to provide written documentation of instances in which they display their firearm was associated with a 0.322 decrease in the number of fatalities per 100,000 population. Given that the mean for the dependent variable is 1.75, this effect is substantively significant. Of large city departments that do not currently have the firearm display report policy, if the 10 with the highest officer- involved death rates instead had this policy, our findings suggest that over a 15-year period, at least 40 fewer people would have died from officer-involved shootings. Figure 2 shows the potential lives saved by enacting this policy for the large city departments with the 10 highest death rates.

Contrary to our expectations, communities with police forces that had larger shares of white officers than the general population did not have systematically higher rates of gun deaths. Similarly, agencies that required community policing training for nonrookie officers and those that assign officers to regular beats did not have lower rates of gun deaths. Thus, hypothesis 1, hypothesis 3, and hypothesis 4 are not supported by the data.

As expected, the binary control for city departments had a significant positive effect, as did the logged population of the community. Larger communities as well as those situated within city boundaries had systematically higher rates of gun deaths by police.

Figure 1 Highest Gun Death Rates among Large Agencies (population over 500,000)

Table 3 OLS Model Results, with State Fixed Effects: Officer-Involved Death Rates in Communities with Populations over 25,000

Coefficient Standard Error

Departmental Variables

City department 2.619* 0.227 Firearm display report −0.322* 0.120 Community policing training −0.058 0.049 Offi cer beat assign 0.118 0.148 % whites overrepresented −0.008 0.006 % of department female 0.003 0.011 Community Variables Inequality Gini 0.087* 0.015 % of district urban −0.019* 0.005 Education −0.033* 0.005 Racial homogeneity −0.002 0.007 % of district black 0.007 0.006 Population (log) 0.524* 0.087

Constant −5.686* 1.404

N = 1,102 | Adjusted R 2 = .351 | * p < .05

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Offi cer-Involved Gun Deaths 223

Additionally, the model shows that higher levels of inequality were associated with significantly higher rates of police-involved deaths. Educational attainment and the percentage of the population living in urban areas were associated with lower rates of gun deaths. We expected that communities with more educated populations would have lower death rates. However, the finding for urban population runs contrary to expectations. It suggests that, holding constant whether the department is a city or county agency, higher shares of population living in urban areas were associated with lower rates of death by police.

While inequality was associated with higher rates of fatalities, racial characteristics of the community served had no effect. Previous research suggested that police use of force would increase with higher shares of black residents in the community due to officers’ perceptions of a threat to the existing social order. However, the percentage of the community that is black had no effect on the

death rate. Additionally, racial homogeneity, which measures the degree to which a community is dominated by a single race, had no effect. Also producing null results was the control variable for the percentage of the police force that is female.

While most of the significant findings in our model are associated with inherent traits of an agency (e.g., city or county) or attributes of a community (e.g., inequality and education), the firearm display report policy is easy to implement and associated with significantly lower levels of fatal officer involved shootings.

Discussion Our results demonstrate that while controlling for a host of important variables, police agencies mandating written reports by officers who display their firearms have systematically lower rates of police-involved gun deaths. In 2013, this policy was in place in nearly half of the police agencies in our data set. Our interviews with leaders of several randomly selected police agencies indicated that some of the agencies requiring documentation have had the policy for decades, while others put it in place more recently to match changing standards and best practices in law enforcement. Requiring documentation in these circumstances is recommended by Lexipol, a provider of police policy manuals and updates that are individualized by state (D. Musselman, private communication, May 31, 2016). It is also required for certain types of police agency certification, such as that offered by the Commission on Accreditation for Law Enforcement Agencies (M. Tyson, private communication, May 31, 2016). The policy created no controversy among the ranks in the agencies we contacted, and it was not opposed by unions or other police associations in those cities (P. Daley, M. Tyson, and D. Musselman, private communications, May 31, 2016).

We hypothesized that this rule would discourage officers from drawing their guns, at least in some cases, in order to avoid onerous paperwork and to follow the recommendation of agency leadership. Fewer draws of the weapon mean a reduction in opportunities to use deadly force with a firearm, thereby decreasing incidence of police-involved gun deaths. Our analysis supports this association, although additional research will be required to establish causality.

Another reason for optimism about the possibility of using this policy to reduce civilian deaths is that the success of the policy could build over time as officers build habits of caution in displaying their

weapons. Research on habit formation and habit breaking shows that when a habit is well established, informational incentives are not nearly as effective as economic incentives (Verplanken and Wood 2006 ). Furthermore, research has shown time-restrained decisions, much like the ones officers are frequently asked to make, are particularly susceptible to the effects of habits or routines and are

difficult to break (Betsch et al. 2004 ). If officers are prone to frequently display their weapons, even in instances in which it is not necessary, this policy may help them break the habit.

A natural follow-up question to the benefit of the display documentation policy is whether it may endanger police officers by

Figure 2 Model Estimates for Lives Saved by Enacting Firearm Display Report Policy

Note: Departments with the highest offi cer-involved death rates of city departments serving areas with populations over 500,000 that also did not have the fi rearm display documentation policy, according to the 2013 LEMAS.

Police agencies mandating written reports by offi cers who display their fi rearms have sys-

tematically lower rates of police- involved gun deaths.

224 Public Administration Review • March | April 2017

causing them to hesitate in drawing their guns in life-threatening situations. To address this question we ran a separate OLS regression using the same control variables as the model earlier but switching the dependent variable to the rate of gun deaths of on-duty police officers themselves. We found that the documentation policy had no effect on the rate of gun deaths of police officers, which, when compared with deaths of civilians, are extremely rare events (see appendix C online for model results).

Contrary to our expectations, community policing training and assignment of regular beats had no effects on rates of gun deaths. We hypothesized that the beat system would create personal connections among beat officers and the communities they patrol, making officers less likely to use deadly force. We reason that the beat system may make officers more familiar with members of the community who are repeat offenders or wanted by police. For example, if an officer on a beat knows that someone in the neighborhood is wanted on a warrant, that officer (compared with a randomly assigned patrol officer) is more likely to recognize, confront, and pursue the wanted person. This was the case in a recent use of deadly force in Raleigh, North Carolina (State of North Carolina 2016)—a city that uses community policing and patrol beats. It is possible that in the use of deadly force, these sorts of interactions may counteract the other proven benefits of community policing, including increased positive connections between officers and their assigned neighborhoods.

Conclusion We note that building the dependent variable as a population- standardized aggregate total of police-involved gun deaths over a 15-year span creates some limitations. Our agency policy and community population data are for a single point in time (2013), while the dependent variable covers 2000–2015. Therefore, we cannot use these data to make claims about whether changes in agency policy or community demographics during this time frame had meaningful effects on the rate of police-involved gun deaths. Instead, we compare agencies with one another using a snapshot of their policies and demographics from 2013. By combining multiple years into one agency-level dependent variable, we were able to diminish the potential randomness of our death rate variable. Particularly in smaller communities, one or two police- involved deaths in a year could create the appearance of a pattern that may not be truly present. By aggregating over multiple years, any observed patterns are less likely to be the result of potential randomness and more likely to be related to substantial factors. However, this does create other problems, including the inability to rule out reverse causality. As we do not analyze when the deaths occurred or when the agencies began implementing policies, it is possible that as a result of high rates of police-involved deaths, some agencies put into place policies such as mandatory documentation of gun display. However, our interviews with police agency leaders suggest this is not the case. Additionally, if high death rates were a significant contributor to the decision to require this policy, we would expect that association to be positive, while we find the association to be substantively and significantly negative. To the extent that agencies may select this policy to reduce high civilian

death rates, this would suggest our result is a conservative estimate of the impact of the policy on death rates.

Further research will be necessary to fully confirm the causal relationships of our findings. Perhaps if the U.S. Justice Department (2016) follows through with its promise of collecting accurate and comprehensive use of force statistics, a more detailed time-serial project can add further evidence of causality. Such a database would include reliable information about incident-level factors, including the race of the officer and the race of the victim. This specificity would allow for more direct testing of several potential predictors of deadly force than our study allowed. Additionally, experiments could be conducted in a controlled setting to examine the causal link. Previous laboratory research has been effective at showing

the effect of racial bias in the reactions officers have to citizens (Correll et al. 2007 ; Greenwald, Oakes, and Hoffman 2003 ). Further research should test the effect of the documentation requirement for firearm display in potentially dangerous situations that demand quick reactions.

The research presented here relied on a combination of data sets never before used to study the patterns of fatal officer-involved shootings. While it has not confirmed the long-standing assumptions regarding the importance of race or community policing when it comes to the use of deadly force, it has provided an evaluation of a simple yet powerful policy. It is beyond the scope of this article to be able to provide definitive evidence as to the causal relationship between the firearm display report policy and fatal officer-involved shootings. This study provides evidence of a strong relationship between adoption of this policy and lower rates of gun deaths by police. As appendix C (online) shows, this policy does not correlate with increased levels of officers dying by gunshot in the line of duty. Therefore, our research suggests this policy change has the potential of saving lives without putting officers at further risk.

Notes 1 . For more on the methodology of Fatal Encounters, see http://www.

fatalencounters.org/methodology/ (accessed April 1, 2016). 2 . Because our dependent variable was constructed using population size, if we

included very small communities, one or two officer-involved deaths would appear as extremely high death rates. We also did not want to make the cut point too high, so as to greatly limit the number of agencies. We tested the sensitivity of our findings for population cut points +/–10,000 people from our 25,000 cut point, and our results held across all iterations.

3 . Town and township police departments were excluded from the analysis because the Census Bureau organizes data on these localities separately from cities and other municipalities. There were 89 such departments in the LEMAS data set that had populations of 25,000 or more, which was the cutoff for inclusion in our analysis.

References Alpert , Geoffrey , and Roger Dunham . 2004 . Understanding Police Use of Force . New

York : Cambridge University Press . Bailey , William C. 1996 . Less-than-Lethal Weapons and Police-Citizen Killings in

U.S. Urban Areas . Crime & Delinquency 42 ( 4 ): 535 – 52 .

Contrary to our expectations, community policing training

and assignment of regular beats had no eff ects on rates of gun

deaths.

Preventing the Use of Deadly Force: The Relationship between Police Agency Policies and Rates of Offi cer-Involved Gun Deaths 225

Baumgartner , Frank R. , Derek A. Epp , Bayard Love , and Kelsey Shoub . 2016 . Targeting Young Men of Color for Search and Arrest during Traffic Stops: Evidence from North Carolina, 2002–2013. Politics, Groups, and Identities. Published electronically on April 13. doi.org/10.1080/21565503.20 16.1160413.

Betsch , Tilmann , Susanne Haberstroh , Beate Molter , and Andreas Glöckner . 2004 . Oops, I Did It Again—Relapse Errors in Routinized Decision Making . Organizational Behavior and Human Decision Processes 93 ( 1 ): 62 – 74 .

Burghart , D. Brian . 2014 . What I ’ ve Learned from Two Years Collecting Data on Police Killings. Gawker, August 22. http://gawker.com/what-ive-learned-from- two-years-collecting-data-on-poli-1625472836 [accessed January 14, 2017].

Cao , Liqun , and Bu Huang . 2000 . Determinants of Citizen Complaints against Police Abuse of Power . Journal of Criminal Justice 28 ( 3 ): 203 – 13 .

Cordner , Gary . 2014 . Community Policing . In The Oxford Handbook of Police and Policing , edited by Michael D. Reisig , and Robert J. Kane , 148 – 71 . New York : Oxford University Press .

Correll , Joshua , Bernadette Park , Charles M. Judd , Bernd Wittenbrink , Melody S. Sadler , and Tracie Keesee . 2007 . Across the Thin Blue Line: Police Officers and Racial Bias in the Decision to Shoot . Journal of Personality and Social Psychology 92 ( 6 ): 1006 – 23 .

Davis , Aaron C. , and Wesley Lowery . 2015 . FBI Director Calls Lack of Data on Police Shootings “Ridiculous,” “Embarrassing.” Washington Post , October 7. https://www.washingtonpost.com/national/fbi-director-calls-lack-of-data-on- police-shootings-ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16–11e5- b31c-d80d62b53e28_story.html?tid=a_inl [accessed January 14, 2017].

Friedrich , Robert J. 1980 . Police Use of Force . Annals of the American Academy of Political and Social Science 452 : 82 – 97 .

Fyfe , James J. 1982 . Blind Justice: Police Shootings in Memphis . Journal of Criminal Law and Criminology 73 ( 2 ): 707 – 22 .

Gallo , Frank J. , Charles E. Collyer , and Patricia L. Gallagher . 2008 . Prevalence of Force by Police in Rhode Island Jurisdictions: Implications for Use-of-Force Training and Reporting . Criminal Justice Review 33 ( 4 ): 480 – 501 .

Garner , Joel H. , and Christopher D. Maxwell . 2002 . Understanding the Use of Force by and against Police in Six Jurisdictions . Washington, DC : U.S. Department of Justice .

Greenwald , Anthony G. , Mark A. Oakes , and Hunter G. Hoffman . 2003 . Targets of Discrimination: Effects of Race on Responses to Weapons Holders . Journal of Experimental Social Psychology 39 ( 4 ): 399 – 405 .

Harring , Sid , Tony Platt , Richard Speiglman , and Paul Takagi . 1977 . The Management of Police Killings . Crime and Social Justice 8 : 34 – 43 .

Harris , David A. 1999 . Driving While Black: Racial Profiling on Our Nation ’ s Highways . New York : American Civil Liberties Union .

Hong , Sounman . 2017 . Does Increasing Ethnic Representativeness Reduce Police Misconduct? Public Administration Review 77 ( 2 ): 195 – 205 .

Jacobs , David , and Robert M. O ’ Brien . 1998 . The Determinants of Deadly Force: An Empirical Assessment of Police Violence . American Journal of Sociology 103 ( 4 ): 837 – 62 .

Johnson , Eric J. , and Daniel G. Goldstein . 2003 . Do Defaults Save Lives? Science 302 : 1338 – 39 .

Madrian , Brigitte C. , and Dennis F. Shea . 2001 . The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior . Quarterly Journal of Economics 116 ( 4 ): 1149 – 87 .

McElvain , James P. , and Augustine J. Kposowa . 2008 . Police Officer Characteristics and the Likelihood of Using Deadly Force . Criminal Justice and Behavior 35 ( 4 ): 505 – 21 .

Meyer , Marshall . 1980 . Police Shootings at Minorities: The Case of Los Angeles . Annals of the American Academy of Political and Social Science 452 : 98 – 110 .

Minnesota Population Center . 2016 . National Historical Geographic Information System, Version 11.0 [database]. http://doi.org/10.18128/D050.V11.0 [accessed January 14, 2017].

Oberfield , Zachary W. 2012 . Socialization and Self-Selection: How Police Officers Develop Their Views about Using Force . Administration & Society 44 ( 6 ): 702 – 30 .

Petrocelli , Matthew , Alex R. Piquero , and Michael R. Smith . 2003 . Conflict Theory and Racial Profiling: An Empirical Analysis of Police Traffic Stop Data . Journal of Criminal Justice 31 ( 1 ): 1 – 11 .

Provine , Doris Marie . 2007 . Unequal under Law: Race in the War on Drugs . Chicago : University of Chicago Press .

Putnam , Robert D. 2007 . E Pluribus Unum: Diversity and Community in the 21st Century: The 2006 Johan Skytte Prize Lecture . Scandinavian Political Studies 30 ( 2 ): 137 – 74 .

Riksheim , Eric C. , and Steven M. Chermak . 1993 . Causes of Police Behavior Revisited . Journal of Criminal Justice 21 ( 4 ): 353 – 82 .

Sampson , Robert J. , and W. Byron Groves . 1989 . Community Structure and Crime: Testing Social-Disorganization Theory . American Journal of Sociology 94 ( 4 ): 774 – 802 .

Smith , Brad W. 2003 . The Impact of Police Officer Diversity on Police-Caused Homicides . Policy Studies Journal 31 ( 2 ): 147 – 62 .

––––. 2004 . Structural and Organizational Predictors of Homicide by Police . Policing: An International Journal of Police Strategies & Management 27 ( 4 ): 539 – 57 .

Smith , Brad W. , and Malcolm D. Holmes . 2003 . Community Accountability, Minority Threat, and Police Brutality: An Examination of Civil Rights Criminal Complaints . Criminology 41 ( 4 ): 1035 – 63 .

Spillar , Katherine , Penny Harrington , and Michelle Wood . 2000 . Gender Differences in the Cost of Police Brutality and Misconduct: A Content Analysis of LAPD Civil Liability Cases: 1990–1999 . Washington, DC : National Center for Women and Policing .

State of North Carolina, General Court of Justice . 2016 . Release on Akiel Denkins’ Shooting Investigation. http://dig.abclocal.go.com/wtvd/docs/Akiel%20 Denkins%20release%2004%2013%202016.pdf [accessed January 14, 2017].

Terrill , William , and Stephen D. Mastrofski . 2002 . Situational and Officer-based Determinants of Police Coercion . Justice Quarterly 19 ( 2 ): 215 – 48 .

Terrill , William , and Michael D. Reisig . 2003 . Neighborhood Context and Police Use of Force . Journal of Research in Crime & Delinquency 40 ( 3 ): 291 – 321 .

Thaler , Richard , and Cass R. Sunstein . 2008 . Nudge: Improving Decisions about Health, Wealth, and Happiness . New Haven, CT : Yale University Press .

Theobald , Nick A. , and Donald P. Haider-Markel . 2009 . Race, Bureaucracy, and Symbolic Representation: Interactions between Citizens and Police . Journal of Public Administration Research and Theory 19 ( 2 ): 409 – 26 .

Tversky , Amos , and Daniel Kahneman . 1991 . Loss Aversion in Riskless Choice: A Reference-Dependent Model . Quarterly Journal of Economics 106 ( 4 ): 1039 – 61 .

U.S. Department of Justice, Office of Justice Programs, Bureau of Justice Statistics . 2013 . Law Enforcement Management and Administrative Statistics (LEMAS) . ICPSR36164-v2. Ann Arbor, MI : Inter-University Consortium for Political and Social Research [distributor], 2015-09-22. http://doi.org/10.3886/ICPSR36164. v2 [accessed January 14, 2017].

———. 2015 . Assessment of Coverage in the Arrest-Related Deaths Program. October. http://www.bjs.gov/content/pub/pdf/acardp.pdf [accessed January 14, 2017].

———. 2016 . Justice Department Outlines Plan to Enable Nationwide Collection of Use of Force Data. October 13. https://www.justice.gov/opa/pr/justice- department-outlines-plan-enable-nationwide-collection-use-force-data [accessed January 14, 2017].

Verplanken , Bas , and Wendy Wood . 2006 . Interventions to Break and Create Consumer Habits . Journal of Public Policy & Marketing 25 ( 1 ): 90 – 103 .

Walker , Samuel . 1993 . Taming the System: The Control of Discretion in Criminal Justice, 1950–1990 . New York : Oxford University Press .

226 Public Administration Review • March | April 2017

Supporting Information Supplementary appendices may be found in the online version of this article at http://onlinelibrary.wiley.com/doi/10.1111/puar.12738/full.

Willits , Dale W. , and Jeffrey S. Nowacki . 2014 . Police Organisation and Deadly Force: An Examination of Variation across Large and Small Cities . Policing & Society 24 ( 1 ): 63 – 80 .

Worden , Robert . 1996 . The Causes of Police Brutality: Theory and Evidence on Police Use of Force . In Police Violence , edited by William A. Geller , and Hans Toch , 23 – 51 . New Haven, CT : Yale University Press .