1

Good Cop, Bad Cop: An Analysis of Chicago Civilian Allegations of Police Misconduct

November 8, 2016

Kyle Rozema* Max Schanzenbach**

*Post-Doctoral Fellow, Northwestern University, kyle.rozema@law.northwestern.edu. **Seigle Family Professor of Law, Northwestern University, m-schanzenbach@law.northwestern.edu. We thank Rajiv Sinclair and the Invisible Institute’s Citizens Police Data Project for making data on allegations against Chicago police officers available. For helpful comments we thank Sheila Bedi, Bernie Black, Ezra Friedman, Jacob Goldin, Kate Litvak, David Shapiro, Wes Skogan, and participants at the Institute of Policy Research Performance Measurement Colloquium and the Soshnick Colloquium on Law and Economics at Northwestern University Pritzker School of Law. We thank David (Shih-Hsuan) Chuang for his excellent research assistance.

2

Abstract

Preventing police officer misconduct, while giving police officers the incentives and flexibility to

fight crime, is a notoriously difficult challenge. We explore whether civilian complaints can help identify and reduce police officer misconduct. Using data on over 50,000 civilian allegations against police officers in Chicago over a 13-year period, we find that civilian allegations are strongly associated with (1) non-civilian allegations of misconduct, such as allegations from supervisors, and (2) the likelihood of and outcomes in civil rights litigation. We conclude that civilian allegations contain a strong signal of police officer misconduct. In particular, lawsuits against police officers are external and quantifiable measures of officer misconduct and typically arise from serious incidents. We further find that, if an administrative body sustains an allegation against an officer, that officer’s civilian allegation rate decreases and the probability that he or she exits the force increases. We conclude that civilian allegations against police officers in Chicago can be better utilized to prevent misconduct.

INTRODUCTION In the last few years, police shootings and deaths in custody in New York, New York;

Baltimore, Maryland; Chicago, Illinois; and Ferguson, Missouri severely damaged police-community relations, prompted widespread protests, and, in some cities, riots. But less salient police officer misconduct can be widespread within a department. For example, a Justice Department report on the Ferguson, Missouri police department detailed a police force in which racist verbal abuse, excessive force, and arbitrary use of fines and arrests were pervasive.

Preventing police officer misconduct, while giving police officers the incentives and

flexibility to fight crime, is a notoriously difficult challenge. For good reasons, this challenge is now at the top of the public policy agenda. Efforts to improve police officer conduct must confront the fact that police officers are tasked with a very complicated undertaking: they must make arrests and prevent crimes without violating the rights of civilians, under subjective standards such as “reasonable force” and “probable cause,” and they often must make decisions quickly. Many of these decisions are imbued with a great deal of officer discretion. Detection of police officer misconduct by supervisors is difficult because police interactions with civilians are usually not directly observed, and therefore the discovery of much misconduct relies on the allegations of civilians through either an administrative process or litigation. Police officer organizations and some criminal justice scholars have questioned the usefulness of civilian allegations, arguing that civilian allegations mainly signal officer productivity (Worden et al., 2012; Lersch, 2002). Moreover, state and local regulations and union contracts often create a civilian allegation administrative process favorable to police officers (e.g., requiring claimants to swear out an affidavit, tightly regulating the investigation process, and subjecting discipline to lengthy appeal and arbitration processes). Nonetheless, pressure to use civilian allegations more effectively is growing. For example, commentators have pointed out that Chicago police officer Jason van Dyke, recently indicted for the shooting death of a young African-American man named Laquan MacDonald, had a long history of civilian allegations, including twenty allegations in the five years leading up to the shooting (Complaint to DOJ, 2016).

3

This article assesses the potential for civilian-based police monitoring programs to

reduce police officer misconduct. This possibility rests on one of two conditions: (1) that officers who receive excessive allegations are in fact likely committing misconduct and therefore merit some form of intervention; or (2) that investigatory process resulting in a finding of a sustained allegation and any resulting disciplinary action deters future misconduct.

To investigate these conditions, we match recently released data on over 50,000 civilian

allegations of police officer misconduct in Chicago to a panel of police personnel records and data on civil lawsuits and resulting payments. The first part of this article investigates the signal strength contained in civilian allegations in predicting misconduct, as measured against two other measures of misconduct: (1) reports of officer misconduct from supervisors and off-duty behavior, and (2) the involvement in and outcomes1 of civil rights litigation. Civilian allegation rates vary significantly across police units and over time, so we create a risk-adjusted measure of officer allegations that compares allegations between officers within a district in a given time period. We use the risk-adjusted measure of allegations throughout the analysis.

We find that the relationship between risk-adjusted civilian allegations and other

measures of misconduct is strong. First, an additional allegation against an officer arising from within the department is associated with an increase of 0.4 civilian allegations, and an additional allegation against an officer arising from off-duty conduct is associated with an increase of 0.7 civilian allegations. Second, civilian allegations against an officer predict the probability that a lawsuit is brought naming the officer, the probability that there is a payment made to plaintiffs in these lawsuits, and the size of the payment. The risk of litigation and size of payouts increases exponentially across the distribution of allegations, suggesting that efforts could be fruitfully concentrated among a relatively small group of officers. The relationship between litigation and allegations stays roughly flat for the 70% of officers in the bottom of civilian allegations distribution, but increases dramatically for officers farther up the distribution. Officers in the top decile of the risk-adjusted distribution of allegations are 40 to 50 percentage points more likely to be named in a lawsuit than officers below the 70th percentile. Perhaps more importantly, there is a non-linear relationship between risk-adjusted allegations and the size of payments made to civil rights plaintiffs. Damages increase markedly for officers above the 70th percentile of the risk-adjusted civilian allegations and are particularly concentrated at the very top.

These relationships suggest that allegations are not solely related to officer productivity.

It is doubtful that negative behavior toward supervisors and anti-social off-duty behaviors signal productivity. Likewise, lawsuits against police officers and resulting payments are external and quantifiable measures of officer misconduct. Civil rights litigation tends to filter for the most serious incidents of officer misconduct, particularly ones in which significant damages could be awarded. Claims against police officers are generally difficult to win because officers are allowed to make honest mistakes and have many other defenses available to them.

1 Notably, and unlike most research in civil litigation, we are able to observe not only award amounts from a plaintiff victory at trial but also payment amounts resulting from settlements. This is possible because the collective bargaining agreements between Chicago police officers and the City of Chicago provide for indemnification of on-duty police officers, and Chicago’s public spending is a public record.

4

Moreover, only claims likely to give rise to significant damages are brought because of the prevalence of contingency fee arrangements in civil rights litigation. Most civilian allegations of police officer misconduct, such as verbal abuse or wrongful arrests after which the victim is quickly released, will not give rise to significant damages.

The second part of this article investigates the deterrence effect of sustained allegations.

We investigate deterrence in two ways. First, we examine how a sustained allegation affects police officer promotion and exit. We find that, even though civilian allegations have rarely led to terminations, a sustained civilian allegation increases the probability of exit from the police force within a year by roughly 50%. We find no evidence that a sustained civilian allegation affects promotion probabilities. Next, we study deterrence by investigating how officer allegation rates change after a sustained allegation. We find that after an allegation is sustained, the allegation rate for that officer decreases by 40%. The effect of a sustained allegation persists years into the future, which we argue is consistent with specific deterrence story rather than incapacitation resulting from reassignment to low-risk areas or activities.

We conclude that civilian allegations, if properly managed, could play an important role

in reducing police officer misconduct and costly civil liabilities for state and local governments. Chicago has historically made little use of civilian allegations. The city has no effective early detection program designed to intervene when an officer builds up a history of civilian allegations. Moreover, the city sustains 2.4% of civilian allegations, a rate that we believe seems low considering the predictive power of civilian allegations regarding other measures of misconduct.

A recent review of the literature on police officer misconduct concluded that “the

accumulated empirical evidence is not voluminous, and our understanding of the well-springs of police officer misconduct and the organizational mechanisms by which misconduct can be better regulated is not very deep.” (Worden et. al, 2014; see also National Research Council, 2004, p. 290). We make a unique contribution to the literature on role of civilian-based police monitoring programs by demonstrating (1) a strong link between civilian allegations and civil litigation, and (2) a deterrent effect when allegations are sustained.

The remainder of the paper is organized as follows. Part I discusses the civilian

allegations process in Chicago, Part II discusses the various data sources, Part III presents the results, and Part IV concludes. I. THEORETICAL AND INSTITUTIONAL BACKGROUND

The interactions between supervisors and police officers, and between police officers and the community, are governed by a complicated web of laws and institutions. Between police officers and their supervisors are state laws, union contracts, and human resource (or “internal affairs”) departments. Between police officers and the civilians are criminal laws, Constitutional rights, civil remedies for rights violations, and in most large cities an administrative review process, either independent of or within the police department, through which civilians may file allegations against police officers. While civil litigation can result in payments to civilians, the administrative process for which civilians file allegations does not. Civilians pursue the administrative route for entirely non-monetary motives, such as protecting

5

dignitary interests. To understand the possible interactions between civilian allegation processes, civil law claims, and police incentives, we review the basic incentive structure of the Chicago Police Department and the civil and administrative actions that civilians may pursue.

A. Incentives, Police Pay, and Promotion Police officers usually act outside the presence of supervisors while engaging in multiple

interactions with civilians almost every day.2 Monitoring these interactions is, even with advances in technology, a significant challenge. On top of this, state laws, city regulations, and police union contracts at times greatly restrain the power of elected officials to monitor and discipline police officers (Harmon, 2012). Disciplinary actions are usually subject to an appeals process that may end in binding arbitration. Investigation techniques into misconduct are also highly regulated. The Chicago Fraternal Order of Police Contract (2012) specifies, for instance, that no officer can be questioned until 24 hours have elapsed since a shooting and that an officer cannot be disciplined if recordings later contradict his or her statements.3 The contract also provides that records of misconduct and disciplinary actions should be destroyed periodically.

In addition to such restrictions, police officer compensation is quite rigid. Compensation

is often determined under collective bargaining agreements or civil service statutes that specify pay. In Chicago, salaries are entirely determined by rank and years of service. For example, the 2012 collective bargaining agreement with the Fraternal Order of Police provides three (increasing) salary grades corresponding to police officer, detective, and sergeant, with pay differentials within those grades based on years of service (Chicago Fraternal Order of Police Contract, 2012). Even if rigid pay were not required by law or contract, fixed police compensation may be the best payment structure because of well-known problems of incentive pay in multitask environments such as policing (Holmstrom and Milgrom, 1991). Much of the personnel economics literature focuses on agency problems in the employment relationship and the potential for incentive pay to improve outcomes,4 but this is, of course, not a feasible option for police officers (e.g., pay per arrest).

Although discipline and pay are highly regulated, promotion—if appropriately used—may serve as strong incentive for good performance. Although police officers do not have formal incentive pay, police departments may create incentives via promotions in rank, a concept well-developed in the literature on tournaments (Lazear and Rosen, 1981; Green and Stokey, 1983; Nalebuff and Stiglitz, 1983). Tournament theory describes how workers who are compensated similarly within each level of the firm can be motivated by the possibility of promotion.

The institutional setting suggests that police officers should be incentivized by

promotion possibilities. Figure 1 plots the average salary for police officers, police officers assigned as a detective, and sergeants over time. The step-ups for each officer rank in Figure 1 represent new contracts. In the time period we examine, the police force consisted of an average

2 Walker and Katz (2012) estimate that there are 1,100 use-of-force incidents per day in the United States, and 43 million significant “police-citizen” encounters every year. 3 Unless the tape was not available to investigators at the time. 4 See Lazear and Oyer (2012) for a thorough review.

6

of 10,248 police officers, 1,001 detectives, and 1,210 sergeants. From 2002 to 2014, the average salary for police officers, detectives, and sergeants was $70,851, $82,066, and $90,740 respectively. In Chicago, promotion from police officer to the detective rank comes with a roughly 17% pay rise, and promotion to sergeant from police officer with an almost 35% pay rise. Other than seniority pay, the primary way in which an officer can raise his or her salary is through promotion.

[Insert Figure 1 about here]

There are two paths to promotion in the Chicago Police Department. The first path relies

primarily on standardized test scores and is called the “General Hiring Process” (Police Department Hiring Plan for Sworn Titles, 2011, p. 16). Candidates who score above a pre-determined threshold are considered in order of their score. Subjective factors that are based on prior performance evaluations and discipline history are then taken into account.

The other promotion mechanism is so-called “merit promotion.” Candidates for merit

promotion are first nominated by supervisors. The “Merit Board” then conducts an initial assessment and forwards selected officers to the Superintendent of Police. A number of subjective factors are considered in this process, including “disciplinary history as permitted by the [Collective Bargaining Agreement]” (Police Department Hiring Plan, Exhibit A, p. 3). About 30% of promotions to sergeant are based on “merit” and 20% to lieutenant are based on “merit.” (Chicago Police Department Hiring Plan, p. 22.)

In light of Chicago’s promotion policies and fixed compensation within rank, personnel tools such as termination, suspensions, reductions in rank, or diminishment of promotion possibilities are necessary to properly align police officer incentives. Standard economic theory predicts that officers will be too aggressive if only positive performance metrics (such as arrest records) are used for promotion. Even without such incentives, police with a taste for misconduct will indulge in it if misconduct is not costly to them. Penalizing officers for misconduct, however, requires effective monitoring of police officer conduct. Without cameras accurately documenting all actions of police officers and a complete review of the footage, civilian allegations will be a necessary feature of monitoring.5 The possibility of using civilian allegations to identify and prevent police officer misconduct is not a new idea. Almost all large police departments have standardized administrative procedures under which civilians may file allegations against police officers. Investigations of these allegations take place under a variety of procedures that vary greatly from city to city (Finn, 2001). Some cities have independent bodies that investigate civilian allegations against police officers. Other cities rely on internal affairs departments for the investigations. The investigation process in many cities can be very slow: in Chicago the average time to process and complete an investigation of a civilian allegation is about one year, after which there may be a lengthy appeals process for any resulting discipline; in New York City the process is only slightly faster (Kane and White, 2013).

5 Even the use of monitoring equipment in Chicago has run into problems because of equipment failures. Roughly eighty percent of audio fails to record, which the department has acknowledged is often caused intentionally by officers disabling equipment (Chicago Tribune, 2015).

7

Some departments use civilian allegations as part of so-called “Early Intervention” or

“Early Warning” programs (Shultz, 2015; Walker and Katz, 2012; Walker et al., 2001). A recent expert review of Chicago policing recommended the implementation of a meaningful early warning system (Police Accountability Task Force, 2016). However, even though evidence suggests that such programs reduce civilian allegations (Walker et al., 2001), some have argued that they decrease officer productivity and punish the most effective officers who have more civilian interactions (Worden et al., 2013). Other evidence suggests that the most active officers, in terms of arrests and stops, are the ones most likely to draw civilian allegations (Brandl et al. 2001; Lersch 2002; Terrill and McCluskey 2002). It is not clear to us that arrests and ticketing are themselves appropriate measures of officer productivity.

We address concern over differences in risk exposure and officer productivity by (1)

conditioning on district-time fixed effects and (2) linking civilian allegations to non-civilian allegations and to civil litigation. As explained in greater detail below, there are institutional arguments for why civil rights litigation is a good external metric for serious misconduct. Allegations arising from other sources, such as from supervisors, likewise provide some external validation.

We also consider whether allegations or sustained allegations result in changes in officer behavior and other long-term career consequences. Discipline resulting from sustained allegations could have both a general and specific deterrent effect on officer misconduct. A credible threat of discipline provides general deterrence ex ante, but general deterrence resists easy measurement. Instead, we assess whether an individual officer who is disciplined is deterred from future bad acts after a sustained allegation. This is usually referred to as “specific deterrence.” Sustained allegations can affect the officer in several ways. The first may be the psychological costs of a negative record. This effect might be evident in the fact that the collective bargaining agreements attempt to limit the time period during which a sustained civilian allegation may remain on the officer’s record. Second, any discipline resulting from the sustained allegation, if it includes suspension, directly reduces an officer’s pay and seniority. Moreover, sustained allegations, disciplines, or merely having a history of civilian allegations could reduce an officer’s promotion prospects (either formally or informally), significantly reducing lifetime income. The resulting diminished promotion prospects and other psychic costs of disciplinary actions could induce departures from the police force, improving overall quality of police officers.

The potential for civilian allegations and resulting discipline to improve police officer conduct is far from universally accepted. Union contracts and state laws restrict the investigatory powers of agencies charged with investigating police officer misconduct,6 and police officers often complain about frivolous civilian allegations. We are not aware of any academic study that has validated civilian allegations as a metric against alternate metrics such as civil litigation or explored in great depth the consequences of disciplinary actions. In the empirical analysis, we take this important first step using Chicago as a case study. To

6 State law constraints often come in the form of so-called “Police Officer Bill of Rights” legislation that limits how internal investigations can be carried on and what methods may be used in investigating officers. Illinois’s Uniform Peace Officer’s Disciplinary Act (50 ILCS 725) is one such example.

8

understand the institutional setting in our empirical analysis, the next two sections discuss the Chicago civilian allegation process and how civil rights litigation proceeds. B. Chicago’s Civilian Allegation Process There are three agencies in Chicago that are tasked with investigating and resolving allegations of police officer misconduct: the Independent Police Review Authority (IPRA), the Police Board, and the Internal Affairs Division (IAD). The IPRA is an independent, civilian-run agency, while the Police Board and IAD are part of the Chicago Police Department. Figure 2 details the rather confusing structure of IPRA process for civilian allegations in Chicago.

[Insert Figure 2 about here]

Prior to 2007, the agency in charge of investigating and clearing police officer

misconduct charges was the Office of Professional Standards, which was a part of the Chicago Police Department and under the control of the Superintendent of Police. The IPRA was created in 2007 by a city ordinance passed in response to a series of high-profile police brutality cases, in particular the notorious Burge torture cases.7 The IPRA has initial responsibility for intake and processing of civilian allegations.8 The head of the agency is appointed by the mayor. Although there were some structural changes when the IPRA replaced the OPS, the allegation process remained largely the same across the two agencies. In fact, one complaint against the IPRA is that it merely took the Office of Professional Standards outside of the Police Department without really changing how civilian allegations were investigated or the disciplinethat could be imposed.9

Civilian allegations of police officer misconduct are submitted to the IPRA. Once received, the IPRA makes a determination on whether it or the IAD has jurisdiction over the subject matter which gave rise to the allegation. Key civilian allegations that are within the IPRA’s jurisdiction are excessive force, wrongful arrest, and verbal abuse. The IPRA also conducts investigations in two other key areas. First, the IPRA has jurisdiction anytime there is a weapon discharged in a manner that could strike someone or anytime someone suffers death or injury in police custody.10 Second, the IPRA has jurisdiction, and must initiate an investigation, any time the city pays a settlement or award over a claim of a civil rights violation. By contrast, noncivilian allegations of criminal misconduct, including bribery, driving under the influence, and allegations of domestic abuse, are under the jurisdiction of the IAD. The IAD can delegate jurisdiction for civilian allegations to the officer’s district-level supervisor if the conduct is deemed minor. (Chicago has twenty-five geographically defined police

7 Jon Burge, a Chicago police detective, allegedly coerced dozens of confessions over a twenty-year period, resulting in dozens of overturned convictions and tens of millions in civil damages (Finney, 2015). 8 The Chicago Police Board adjudicates allegations against officers brought by the Superintendent of Police if discipline sought is dismissal or more than 30 days suspension. Also, the Board hears appeals from the IPRA disciplinary process if the Chief Administrator of the IPRA and the Superintendent of Police cannot agree on a disciplinary action. 9 For a discussion of the continuity between the Office of Professional Standards and the IPRA, see Safer et al. (2014). 10 Safer (2014 at 14-15). The IPRA retains control over cases with multiple allegations if any one allegation would be in its jurisdiction.

9

districts, and there are also other units of the police that are not geographically constrained including administrative units and crime-specific task forces.)

When a civilian allegation is made, the IPRA assigns an investigative team. The lead

investigator may take witness testimony and interview the officer against whom the civilian allegation is made. The investigator must ultimately write a report recommending any of three different outcomes: “sustained,” “not sustained,” or “exonerated.” The evidentiary standard is “clear and convincing,” which is higher than the traditional civil lawsuit “preponderance” standard, but less than the criminal law standard of “beyond a reasonable doubt.” A sustained civilian allegation indicates that the investigator found evidence supporting the facts of the reported misconduct, and that those facts indeed gave rise to a violation of the department’s code of conduct. An exoneration means that the officer has been determined with reasonable certitude not to have engaged in the conduct alleged.

A civilian allegation that is “not sustained” indicates that there was some, but not

sufficient, evidence of misconduct. This can occur for two reasons. First, the facts did not establish misconduct by a preponderance of the evidence. Second, many allegations are ultimately closed because of an unwillingness on the part of a witness or complainant to cooperate with the investigation or file an affidavit. Illinois state law requires an affidavit to be filed for a citizen allegation to proceed (Uniform Peace Officers’ Disciplinary Act, 50 ILCS 725/1 et seq.).

The requirement of a sworn affidavit from the accuser has come under criticism for

chilling the allegation process. As part of the affidavit process, the accuser is warned of potential criminal liability for perjury if he or she swears to something they know not to be true.11 Recent policy proposals have suggested relaxing the affidavit requirement (Police Accountability Task Force 2016). Of course, the justification behind the affidavit requirement is that it filters out false allegations and such a requirement could improve the signal contained in the allegation process. Below we test for whether sworn or unsworn civilian allegations have different predictive power. We find no difference between them. If an allegation is “sustained” by the IPRA, the IPRA only recommends a punishment to the Superintendent of police. There are no clear guidelines for what discipline is appropriate, a fact highlighted in the external evaluations of the IPRA (Safer et al., 2014; Police Accountability Task Force, 2016). If the IPRA and the Superintendent cannot agree on a punishment, the Chicago Police Board resolves the dispute. If termination is recommended, the recommendation must proceed to Police Board review. Appeals from any punishments relating to sustained civilian allegations of police officer misconduct can proceed to arbitrators or through internal administrative review as well.

11 Indeed, Illinois has a state law, similar to that in many other states and often referred to as a “police officer’s bill of rights” that permits investigations into allegations made by civilians to proceed only if there is a sworn affidavit. The law further directs that “[a]ny complaint, having been supported by a sworn affidavit, and having been found, in total or in part, to contain knowingly false material information, shall be presented to the appropriate State's Attorney for a determination of prosecution.” 50 ILCS 725/3.8.

10

Officers are rarely disciplined as a result of civilian allegations. In our data, 3.1% of civilian allegations are “sustained.”12

C. Civil Litigation against Police Officers An individual whose constitutional rights were violated by a Chicago police officer can

bring a civil action against the officer under federal or Illinois law. The most popular cause of action is the “Section 1983” claim brought under 42 U.S.C. 1983. This federal statute, originally passed in 1871 as part of the Ku Klux Klan Act, allows a private right of action against a state or local official who “under color of law” subjects anyone to “deprivation of any rights, privileges, or immunity secured by the Constitution and laws.” Sometimes called a “constitutional tort”, a Section 1983 suit is commonly used by those subject to excessive force, unlawful search and seizure, or wrongful arrest by a police officer. Most states, including Illinois, allow state tort actions against police officers in some contexts as well, but Section 1983 is the most pursued vehicle for civil recovery against alleged violations of civil rights by police officers. Moreover, Section 1983 claims may be brought in state courts, which have concurrent jurisdiction with federal courts.13 However, Section 1983 actions are easily removable to federal court, and consequently most cases proceed in federal court.14

It is difficult to win a Section 1983 action against a police officer for several reasons. First,

police officers enjoy “qualified immunity,” which means that in order to be found liable, police actions must have been both objectively unreasonable and violate a clearly established statute or right.15 “[Q]ualified immunity is an immunity from suit rather than a mere defense to liability.”16 Thus, Section 1983 cases are dismissed with some regularity because a judge can dismiss a Section 1983 case on a threshold determination of whether qualified immunity applies.

Second, the burden of proof is on the plaintiff. Thus, to win an excessive force claim, for

instance, the plaintiff must show the officer objectively used unreasonable force under the circumstances and it would have been clear to a reasonable police officer that the alleged conduct violated a statute or constitutional right.17 Because there is no clear legal rule that defines reasonable force, the amount of force required to be “unreasonable” is a fact-intensive

12 A separate study of all 409 police-involved shootings in Chicago between 2007 and 2015 found that only two resulted in a sustained allegation, and both involved shootings by off-duty officers (Complaint to DOJ 2016). 13 Indeed, some claims, such as malicious prosecution claims, must be brought in state court. See Parish v. City of Chicago, 594 F.3d 551 (2009). 14 A few Illinois state cases have proceeded in state court. In conversations with practitioners, we were told that one reason to file a state action may be to change the jury pool. Federal cases would draw the jury pool from several counties, while cases proceeding against Chicago police officers in state court would draw only from Cook County, the county in which Chicago is located. 15 Plumhoff v. Rickard, 134 S. Ct. 2012 (2014). 16 Pearson v. Callahan, 555 U.S. 223, 231 (2009). 17 Saucier v. Katz, 533 U.S. 194 (2001).

11

inquiry. The Court’s reasonableness standard incorporates the fact that “officers are often forced to make split-second judgments” in “tense, uncertain, and rapidly evolving” situations.18

Moreover, damages are not necessarily available merely because a Constitutional tort

was committed, and not all police misconduct implicates Constitutional rights. For example, in Corpus v. Bennett,19 a police officer struck a suspect in the head after “words were exchanged” during a booking. The suspect sued and the jury found that, although rights were violated, there were only “nominal” damages.20 In addition, verbal abuse, by itself, will not ordinarily give rise to a Section 1983 claim because such abuse does not violate a Constitutional right, even if the abuse would constitute assault under state law. For example, the Ninth Circuit held in Freeman v. Arpaio21 that verbal abuse directed at religious and ethnic background does not give rise to a cognizable constitutional violation.

Although attorney’s fees are available for prevailing parties in Section 1983 actions

under federal law, such fees are often difficult to collect and the definition of “prevailing” is not always straightforward. Without a significant damages award, courts have been generally reluctant to find that the plaintiff prevailed (Schwartz, 2007). As one observer has written, “plaintiffs with low damages or those seeking injunctive relief have no remedy under section 1983, because no lawyer will take their cases to court” (Reingold, 2008, p. 47). In sum, Section 1983 claims face a variety of hurdles and are brought by attorneys only when there is a prospect of a significant recovery.

Because we examine the effect of allegations on litigation, it is important to consider whether civilian allegations can be mechanically related to litigation in a way that might raise identification concerns. One potential source of concern is the fact that the IPRA is directed to investigate cases in which a payment was made. We removed these cases from the data, but in any case their inclusion does not materially change the results.

Another issue is the extent to which allegations can influence litigation. For example,

identification concerns would arise if lawyers target officers with large numbers of allegations or if allegations are admissible as evidence. The number and extent of allegations is unlikely to influence the course of litigation because unsustained allegations are not generally discoverable and, if discovered, are not generally admissible. First, prior to the public disclosure of the allegations data in late 2015, the public did not have access to police disciplinary records. Thus, lawyers could not target police officers who had lengthy histories of allegations or disciplines when deciding to file a suit—that information would not be known at filing. This implies that, in our sample period, the decision of whether to file a lawsuit was could not be based on the number of allegations an officer has received.

18 Graham v. Connor, 490 U.S 386 (1989), at 936-97. 19 430 F.3d 912 (2005). 20 The jury awarded $75,000 in damages, later reduced to $1 on appeal because the jury concluded that there was no significant physical harm. 21 125 F.3d 732, (1997).

12

Second, after filing a lawsuit, rules of evidence limit the discovery and admissibility of allegations. Chicago’s attorneys vigorously fought against discovery of allegation histories and such discovery was often granted on only a limited basis and mainly when the allegations bore directly on the conduct at issue (St. Clair et al., 2016). Even if discovered, allegations are unlikely to be admissible. Under the Federal Rules of Evidence, evidence of prior bad acts is not generally admissible to show a defendant's propensity to commit an act. Thus, the mere number of allegations or sustained allegations is not admissible. A specific prior bad act may be admissible, however, to show “motive, opportunity, intent, plan, knowledge, identity, or absence of mistake.” How this is determined is subject to a complicated four-part test. In Okai v. Verfuth,22 the Seventh Circuit Court of Appeals applied this test to a 1983 claim of excessive force, and concurred with the district court’s refusal to admit prior sustained administrative allegations against the officers at issue, even though it was for similar conduct. The exceptions to this rule are governed a complicated rule of evidence and related case law, which all but requires the allegation to have been (1) based on very similar facts to the case at hand, and (2) to have been sustained, which allegations rarely are.23 In short, there is no reason to think that a feedback loop exists between allegations and litigation.

Another possible complication occurs if people pursue allegations and civil lawsuits

over the same incident. Most lawyers would not permit a client to use the allegation process because it requires a sworn statement that may then become admissible in court. Moreover, damages are not available in the allegation process, so claims giving rise to damages would not benefit, but could only be harmed, by the allegation process. We will present some robustness checks below that account for the possibility that allegations and lawsuits proceed simultaneously for the same incident.

III. DATA AND DESCRIPTIVE STATISTICS To conduct our analysis, we link personnel, allegation, and litigation data from five

sources: (1) a panel of roughly 50,000 civilian allegations of police officer misconduct and the outcomes of the investigations, as well as 28,000 internal allegations and 5,000 “other allegations” and their outcomes; (2) a panel of the universe of federal and state lawsuits in which Chicago police officers are named; (3) a panel of the universe of lawsuit payments made on behalf of these officers by the City of Chicago (officers are indemnified by the City); (4) a panel of personnel data on the district or unit to which police officers were assigned on any given day while they were on the force; and (5) panel data on each Chicago police officers rank every six months. Except for the analyses using data on civil ligation, which is available from 2009 to 2014, our analysis will rely on data from 2002 through 2014.

22 275 F.3d 606, (7th Cir. 2001). 23 In Okai, the court held that “actual evidence, in the form of sustained complaints or potential witness testimony....” should make admissibility of disciplinary records much more likely. Okai at 611 (emphasis in original).

13

A. Personnel Data We obtained personnel records on all Chicago police officers by filing a Freedom of

Information Act (FOIA) request with City of Chicago Department of Human Resources. The bi-annual personnel data include each officer's name, position title (rank), and original hire date. These records are available from 2002 to 2014. We use the officers first and last name and the original hire date to link records with the other datasets, and our time units are in six-month windows (January to June; July to December). We limit our analysis to officers at four levels of rank: officer, officer in special unit, detective, and sergeant. This is because higher-level officers, such as captains and other supervisors, are often named in lawsuits as a matter of course, which means there is little information content in a lawsuit against a supervisor. On average from 2002 to 2014, there were roughly 12,000 police officers, detectives, and sergeants in Chicago.

We first match the personnel records to data on the district to which each police officer

was assigned at the start of each period. There are 25 police districts in Chicago defined geographically, as well as a couple dozen “special” or city-wide units. We will discuss in the next section that an important part of our identification strategy rests in the ability to control for “risk factors”, including the police district an officer works in any given period. This is because (1) crime rates vary across police districts in certain areas of Chicago and vary within police districts over time, and (2) civilian-police interactions, and the frequency and intensity of those interactions, vary over time, across geographic areas, and even within a geographic area over time. Officers in each district-time period therefore face different risks of having an allegation made against them.

B. Civilian Allegations Data Chicago Police Department records of officer misconduct allegations were not available

to the public until December 2015. In 2014, a court decision Kalven v. Chicago24 opened to the public a partial register of allegations made against police officers (from 2001 to 2006 and from 2011 to 2015), which became available to the public in December 2015. A second round of data was released in October 2016 based on a freedom of information request from the Chicago Tribune.25 We use allegations made against Chicago police officers from 2002 to 2014 from the second round of the data release, which we obtained from the Citizens Police Data Project at the Invisible Institute, an organization that serves as a clearing house for the data.26 Most allegation records contain (1) the identity of the officer in question; (2) the outcome of the investigation; (3) the date in which the incident for the civilian allegation occurred; (4) the type of allegation (whether verbal abuse, wrongful arrest, or excessive force); and (5) the date in which the outcome of the allegation occurred. We exclude the over 3,000 allegations that arose because of civil litigation, which is indicated as a type of allegation.

The data include the type of conduct that gave rise to the allegation, ranging from

allegations about off-duty behavior to the use of excessive force. Table 1 provides a summary of the type and number of allegations by IPRA coding and which ones we identify as “civilian.”

24 7 N.E.3d 741 (2014). 25 For a description, see Richards et al. (2016). 26 http://invisible.institute/police-data/

14

We identify “civilian allegations” as those that relate to interactions between civilians and on-duty police officers, such as “First Amendment/Wrongful Arrest” and “Excessive Force.” We separate the remaining non-civilian allegations into “internal allegations,” which are allegations that arise from reports from supervisors or fellow officers, and “other allegations,” most of which are generated by off-duty behavior. Internal allegations include insubordination, failure to report, failure of a drug or alcohol test, and so forth. The remaining “other” allegations include behavior that is often criminal in nature, such as drunk driving and domestic violence, but also includes repeated traffic violations. Table 1 also reports the rate at which each type of allegation is sustained. Internal and other allegations are sustained at a much higher rate than civilian allegations, perhaps because they involve the evidence and testimony of officers or are the result of objective indicators such as a drug test. Figure 3 plots the number of civilian, internal, and other allegations against police officers over time.

[Insert Figure 3 about here]

Using geospatial data on Chicago’s police districts available from the City of Chicago,

Figure 4 plots the mean per-officer average number of annual civilian allegations by police district over the same period. With the city-wide average rate of 0.29 annual civilian allegations per officer over the sample period, rates vary substantially across districts, from as low as in 0.17 allegations per-officer in District 19 to as high as 0.62 annual allegations per-officer in District 11. This substantial variation motivates a risk-adjusted approach to measuring excess civilian allegations. The risk-adjusted approach is also necessary if the number of allegations have evolved over time. Even within a given district, allegations in any six-month period vary substantially. In fact, the standard deviation of the civilian allegation rate at the district-period level is larger than the within-district standard deviation rate (0.22 compared to 0.15).

[Insert Table 1 and Figure 4 about here]

C. Data on Civil Litigation Using Bloomberg Law, we obtained the universe of federal and state civil actions from

2009 to 2014 where Chicago or a city department was listed as a party. We separately obtained all positive payments from the City of Chicago from 2009 to 2014.

For each civil action, the Bloomberg data contains the filing date of the lawsuit and the

names of each defendant in the suit, which we use in combination with officer names to link up to the personnel data. Bloomberg also codes the data by cause of action. We identified 5,810 instances between 2009 and 2014 in which a Chicago police officer was listed as a party to a federal Section 1983 action or a state action based on a deprivation of constitutional rights.27 27 To match officers from the personnel data to officers named in litigation in the Bloomberg data, we followed four-step process. First, we exclude officers from the personnel data who do not have a unique name. Second, we used a fuzzy match record linkage process to link all officers in the personnel data to all named defendants in the Bloomberg data. This created matches with a probabilistic assessment of the similarity of the character strings of the officer names in both datasets. Third, we required a probabilistic match rate to be at least 0.9. This resulted in 19,770 potential officer matches. In a final step, a research assistant manually assessed the officer names and scored the potential name matched on a standard 4 step scale (not a match, maybe a match, very likely a match, and definitely a match). We retained

15

[Insert Figure 5 about here] We obtained the universe of actual payment amounts to civil rights plaintiffs made by

the City of Chicago from 2009 to 2014. The collective bargaining agreements between police officers and the City of Chicago provide for indemnification of on-duty police officers in civil suits, and all payments made by the city in litigation matters are public.28 Moreover, settlements above $100,000 require approval by the Chicago City Council (Iris 2014).

The payment data does not list the officers that were named in the litigation or the filing

date of the lawsuit. However, the data contains the civil action number, which we link to the civil litigation data from Bloomberg, thus identifying the officer defendants in the litigation.29 We identified 2,099 occasions that an officer was part of a lawsuit that resulted in a payment. Figure 5 plots the number of instances where an officer was named in a lawsuit, was named in a lawsuit that had a positive payment, and was named in a lawsuit in which over $100,000 was paid out, requiring the approval of the City Council.

Chicago’s total payments in police officer misconduct cases have averaged nearly $50 million annually in the period we observe payouts (2009 to 2014). Figure 6 shows the distribution of payments that were matched to the personnel records, where the x-axis is in natural log scale because the long right tail of payment amounts (further discussed below). Conditional on a payment, the median payment is $30,000 and the average is $238,900. Settlements are clustered below $100,000, likely due to the requirement that the City Council approve settlements in excess of $100,000.

[Insert Figure 6 about here] E. Descriptive Statistics Table 2 provides descriptive statistics. In a six-month period, the average number of

civilian allegations per officer is 0.156. Internal allegations are about half as frequent as civilian allegations, but the rate in which they are sustained is over four times higher. Other allegations are about 25% as frequent as internal allegations but are sustained at a similar rate. Per six-month period, 1.9% of officers are named in a lawsuit, and 1.3% of officers are named in a lawsuit that pays out, and payouts average $2,195 per officer. Each year, 1.1% of police officers are promoted and 3.5% leave the police force.

matches that the research assistant identified to be “very likely a match” and “definitely a match”, but the results are not largely sensitive to the precise definition we use for the match quality. 28 Chicago discloses the data at http://www.cityofchicago.org/city/en/depts/dol.html. 29 We created crosswalks for both federal and state cases. Federal and state cases can be distinguished by the civil action number. For instance, all federal cases follow the format of XX-CV-XXXX, and state cases formats include XX-M1-XXXXX and XX-L-XXXX. To construct the federal crosswalk, we first acquired all the federal actions coded as Section 1983 claims or prisoner claims in the Northern District of Illinois from Bloomberg Law, and all state court dockets for cases in the Illinois Circuit Court for Cook County in which the City of Chicago or Chicago Police were named as a party. Section 1983 and prisoner claims receive numeric codes of 440 and 550 respectively from the Federal Judicial Center.

16

[Insert Table 2 about here]

IV. EMPIRICAL ANALYSIS The analysis proceeds in four parts. Section A investigates the extent to which civilian

allegations of police misconduct are related to internal and other allegations of officer misconduct. Section B investigates the extent to which civilian allegations of police misconduct predict involvement in civil litigation, and Section C provides robustness checks of this relationship and explores related issues. Section D investigates whether an allegation against an officer that is “sustained” by the reviewing agency decreases promotion prospects and promotes exit from the force. Section E investigates whether a sustained allegation affects the propensity of an officer to draw future civilian allegations.

Throughout, we use a risk adjusted measure of civilian allegations. To construct a risk-

adjusted measure of civilian allegations, we estimate the average residual propensity to draw an allegation not explained by the interaction of time period and district by first estimating Equation 1:

𝐴𝐴𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝐷𝐷𝑖𝑖𝑖𝑖 + 𝜖𝜖𝑖𝑖𝑖𝑖 (1)

where 𝐷𝐷𝑖𝑖𝑖𝑖 is a vector of indicators for the police district in which officer i was assigned to

in period t, which nets out the average propensity of any officer in the period and district to have an allegation filed against them. In other words, this approach accounts for an officer’s assignment history to districts by measuring the officer’s history of allegations relative to the history of similarly situated officers. After we estimate Equation 1, the average of each officer’s lifetime propensity to have an allegation filed against them is the officer-level average residual 𝜖𝜖�̅�𝑖 = 1

𝑇𝑇𝑖𝑖∑ 𝜖𝜖𝑖𝑖𝑖𝑖𝑇𝑇𝑖𝑖=1 , where 𝑇𝑇𝑖𝑖 is the number of periods officer i was in the police force during our

sample period. Figure 7 provides a histogram of the risk-adjusted measure of civilian allegations, 𝜖𝜖�̅�𝑖. The long-right tail of 𝜖𝜖�̅�𝑖 suggests that a small group of officers have a propensity to accumulate excess allegations. Because our independent variable of interest is itself an estimate, we bootstrap standard errors throughout.

[Insert Figure 7 about here]

A. Civilian Allegations and Other Allegations We study the relationship between (risk-adjusted) civilian allegations and (risk-

adjusted) internal allegations of misconduct from within the police force. These misconduct allegations originate from different sources (civilians, supervisors, and arrests), and thus may provide validation of civilian allegations as a measure of misconduct.

We estimate Equation 1 for internal allegations and separately for other allegations, and

then calculate the average residual propensity to draw these allegations not explained by the interaction of time period and location, which accounts for potential differences in management

17

and inner workings of districts over time. The risk-adjusted measure of civilian allegations is then regressed on the risk-adjusted measures of internal and other allegations. Column 1 of Table 3 provides the results. Each additional internal (other) risk-adjusted allegation is associated with 0.416 (0.674) additional risk-adjusted civilian allegations.

[Insert Table 3 about here]

We next test for differences in sworn and unsworn civilian allegations. Prior to 2007,

civilian allegations without an affidavit are not separately identified in the data. At the creation of the IPRA in 2007, however, the data record “no affidavit” as a reason for a finding of “not sustained.” Roughly 55% of allegations are dismissed (“not sustained”) for failure to swear an affidavit.30

To test whether no-affidavit allegations are systematically different from affidavit-based

allegations, we separately analyze allegations with and without affidavits in Columns 2 and 3. The results suggest that internal and other allegations have about the same relationship with sworn and unsworn civilian allegations. For internal allegations, the coefficient from Column 1 is split roughly equally between Columns 2 and 3. The split in “other allegations” actually suggests that no-affidavit allegations are more strongly correlated with “other allegations,” but this result is fairly noisy. In short, we find no evidence that civilian allegations made with an affidavit are any different from those made without an affidavit in terms of their relationship with allegations arising from within the department and from off-duty behavior.

B. Civilian Allegations and Civil Rights Litigation In this section, we explore whether civilian allegations against police officers can serve

as a valid metric to predict serious police officer misconduct as measured by litigation brought against Chicago police officers. From 2009 to 2014, roughly 17% of officers are named as defendants in civil rights litigation, and 11% of officers are involved in a case for which a payment was made. The average payment, conditional on a payment being made, is $238,900. The average annual payout per police officer was about $4,400. The analysis here uses involvement in litigation as the outcome, and the variable of interest is an officer’s residual propensity to draw an allegation 𝜖𝜖�̅�𝑖. To this end, we estimate the following cross-sectional regression:

𝐿𝐿𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽𝜖𝜖�̅�𝑖 + 𝑇𝑇𝑖𝑖 + 𝐴𝐴𝑖𝑖 + 𝜀𝜀𝑖𝑖 2)

where 𝐿𝐿𝑖𝑖 is one of four measures of litigation: (1) an indicator variable for whether officer i was named in a civil rights lawsuit over the sample period, (2) an indicator variable for whether officer i was named in a civil rights lawsuit that involved a payment (either through an award or a settlement), (3) an indicator variable for whether officer i was named in a civil rights lawsuit that involved a payment that the City Council had to approve (at least $100,000), or (4) the natural log of the total amount of payments in the case in which officer i was involved (set to 0—ln($1)—for those who have no payments at all). Each model includes appointment year

30 For purposes of comparability, we use 2007 to 2014 data for Column 1. The results are largely unchanged using years 2002 to 2014.

18

fixed effects, 𝐴𝐴𝑖𝑖, and a vector of fixed effects for the number of periods the officer was presenting, 𝑇𝑇𝑖𝑖, which controls for the higher risk of involvement in litigation due to more time as a police officer.31

Note that the litigation outcomes are based on data from 2009 to 2014, but the risk-adjusted allegation measures are based on data from 2002 to 2014. In the robustness checks section, we take different cuts at the data and investigate the relationship between contemporaneous allegations and litigation (by restricting the allegations to the same time period as the litigation data). In addition, we will investigate the relationship of litigation that occurs after allegations (by restricting the allegations to before the time period as the litigation data). Below we provide motivation for the different specifications, but simply note that the analysis here uses data over a 13-year time period to construct a risk-adjusted measure of civilian allegations.

Table 4 provides the results for binary outcomes pertaining to involvement in litigation. The outcome in Column 1 is whether an officer has been named in a lawsuit. The outcome in Column 2 is whether a payment was made in a lawsuit in which the officer was named, and Column 3 restricts to payments that required City Council approval.

The point estimates indicate the percentage point increase in the probability of

involvement in litigation associated with a one-unit increase in the risk-adjusted measure of civilian allegations. To interpret the magnitude of the estimates, one can turn to the distribution of allegation residuals (see Figure 7). The distance between the allegation residual for the median officer (with -0.08 allegation residual) and the allegation residual in the 95th percentile of the distribution (0.42) is about 0.50.32 The point estimates suggests that, compared to the median officer, an officer in the 95th percentile of risk-adjusted civilian allegations has a 117%, 158%, and 133% higher probability of being named in a lawsuit, being named in a lawsuit that pays out, and being named in a lawsuit that pays out more than $100,000, respectively.33

[Insert Table 3 about here]

Table 4 reports OLS results using the logged amount of payments as the dependent

variable as well as quantile regressions of payment amounts in levels. The point estimate on the allegation residual in the logged payment regression is 3.63, implying that a one-unit increase in the allegation residual increases payout by 363%. In terms of the distribution of allegations, an officer in the 95th percentile of risk-adjusted civilian allegation has on average about 158% 31 The results are not sensitive to various specifications. There are qualitatively similar results if no controls are used, if frequency weights are used to account for the fact that officers are observed for different numbers of time periods (because of new hires, retirements, and promotions), a squared allegation residual term, and if the sample is limited to officers with different thresholds for years of experience. 32 The mean officer is close to the median officer (from the 95th percentile a distance of 0.42). The interpretation of the results for the 99th percentile officer would be roughly 50% higher from the median (the distance between the median and the 99th percentile is 0.782). 33 These are calculated by multiplying the point estimates by the distance between the median officer and the officer in the 95th percentile of the measure of risk-adjusted civilian allegations (0.5), and then dividing by the mean of the litigation measure (as reported in the bottom row of the table).

19

higher payouts than the median officer. Columns 2 to 4 report results in levels using quantile regressions. Because only 11 percent of officers are involved in a lawsuit with a payout, and given the presence of outlier damage awards, we report quantile regression estimates at the 90th, 95th, and 99th percentiles. The estimated effect of the allegations residual at the conditional 90th, 95th, and 99th percentiles are increasing, where a shift from the median officer of risk-adjusted civilian allegations to the 95th percentile increases damages by roughly $20,000, $60,000, and $180,000 over the six-year time period.

[Insert Table 4 about here]

There could be significant non-linearities as one moves across the distribution of officer

allegation residuals. To test for this possibility, and also to aid in interpreting the parametric results in Table 4, Figure 8 displays an alternative specification in which we repeat the basic regression in Equation 2 but assign officers to deciles in the residual distribution instead of regressing on the residual itself. Officers in the lowest decile are the reference group. Panel A of Figure 8 shows that the probability an officer is named in a lawsuit (solid line) is flat across the first seven deciles, but increases at the 70th percentile. Officers in the top decile of allegations residuals are 45 percentage points more likely to be named in a lawsuit than officers in the first decile. The results for the probability of a payout (the dashed line) roughly track those of probability of a lawsuit. Panel B of Figure 8 is for the size of the payouts. The solid line tracks log payouts across the allegation residual deciles, and the dashed line tracks the change in the conditional 95th percentile across the allegation residual deciles. The payout levels are flat across the deciles but begin to increase at the 70th percentile, spiking at the 90th percentile. Officers in the top ten percent of the allegation residual distribution have 250% higher payouts than officers in the bottom 70 percent, and the conditional 95th percentile of payments is roughly $100,000 higher for that group.

C. Robustness Checks We investigate a number of robustness checks across all four of our litigation outcomes (named in lawsuit, payout, city council approved, and log payout). Table 6 incorporates into the regressions internal and other allegations and separates allegations with an affidavit from those without. The coefficients on civilian allegations are little changed overall and again indicate that allegations with no affidavit are at least as powerful a predictor of litigation as those with an affidavit. Moreover, civilian allegations are by far the best predictor of litigation.

[Insert Tables 6 and 7 about here]

Table 7 checks for whether allegations could be mechanically related to litigation in

some way (see Section III for a discussion). Again note that the data record allegations brought because of civil litigation, which we have already excluded. Panel A takes the conservative assumption that every lawsuit also generated an allegation the year or the year before it was filed. Thus, we subtract one allegation from allegations around each time an officer was named in a lawsuit, and then re-estimate the allegation residuals. We repeat this process for the timing of a payout made in a lawsuit. Under these specifications, the estimated coefficients on civilian

20

allegations are reduced by 15% to 20% compared to the results in Tables 4 and 5, but nonetheless indicate large and statistically significant effects.

Panels B and C of Table 7 report results from estimates of the allegation residual using

different time periods. Panel B examines whether allegations predict contemporaneous litigation, and so uses allegation data from 2009 to 2014, the same time period we observe payouts in civil litigation. Next, Panel C of Table 7 examines whether pre-2009 allegations predict litigation outcomes in 2009-2014. The use of separate time periods breaks the simultaneity between allegations and litigation. In particular, we estimate the risk-adjusted measure of civilian allegations relying on data from 2002 to 2008, and then regress these estimates on the litigation measures from 2009 to 2014. We include only officers whose careers straddle these two time periods. The results in Panels B and C are smaller (by about 20 to 25%) than those in Tables 4 and 5, but indicate large and statistically significant relationships. This is consistent with greater measurement error in the allegation residual when a shorter time period is used to estimate the residual.

D. Consequences of Allegations and Sustained Allegations In this section, we explore the effect of civilian allegations on police officer promotion

and exit by estimating Equation 3.

𝑌𝑌𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽𝐴𝐴𝑖𝑖𝑖𝑖 + 𝛾𝛾𝑆𝑆𝑖𝑖𝑖𝑖 + 𝑋𝑋𝑖𝑖𝑖𝑖 + 𝐻𝐻𝑖𝑖 + 𝜖𝜖𝑖𝑖𝑖𝑖 3) where 𝑌𝑌𝑖𝑖𝑖𝑖 is an indicator for (1) whether the officer was promoted within a year, or (2)

whether the officer exited the police force within a year.34 𝐴𝐴𝑖𝑖𝑖𝑖 is the number of allegations against officer i in period t, and 𝑆𝑆𝑖𝑖𝑖𝑖 is an indicator for whether officer i received a sustained allegation in time t. We separately include internal and other allegations, as well as sustained internal and other allegations. 𝑋𝑋𝑖𝑖𝑖𝑖 are fixed effects for the district officer i was assigned in time t, and 𝐻𝐻𝑖𝑖 are fixed effects for officer appointment year.

Table 8 provides the results. Columns 1 and 2 are for promotion within a year, and

Columns 3 and 4 are for exiting the force within a year. Columns 1 and 3 include civilian allegations, and Columns 2 and 4 add internal and other allegations. The models here are estimated using the full panel, so the inclusion of district and time fixed effects gives the number of a civilian allegations the risk-adjusted interpretation as in the previous sections.

Civilian allegations have no immediate effect on promotion probabilities, even if they

are sustained. However, internal and other allegations, whether sustained or not, reduce the probability of promotion, with sustained internal or other allegations having a large additional effect. In Columns 3 and 4, a civilian or internal allegation actually decreases the probability of exit,35 while a sustained civilian, internal, or other allegation dramatically increases the

34 Because promotion and exit are rare events, we look at a one-year window instead of the six-month window provided in the personnel data. 35 This decrease is consistent with reduced opportunities for officers who receive allegations in transitioning to other work, such as a suburban police department. We do not know if the allegations

21

probability of exit. To be concrete, a sustained civilian allegation increases the probability of exit within a year by over 50%. This is true even though civilian allegations have rarely led to terminations. Viewing the results on promotion and exit together, the large increase in the probability of exit from a sustained allegation could explain why there is no detectable relationship between sustained allegations and promotion: if an officer exits the force, the sustained allegation cannot influence his or her promotion prospects, and perhaps diminished promotion prospects prompt officer exit.

[Insert Table 8 about here]

E. Deterrence Does the civilian allegation process deter police officer misconduct? We provide one

test for the answer to this question by asking whether a sustained allegation reduces the future civilian allegations. To do so, we estimate Equation 4:

𝐴𝐴𝑖𝑖𝑖𝑖 = 𝛼𝛼 + 𝛽𝛽𝑆𝑆𝑖𝑖𝑖𝑖 + 𝛾𝛾𝐼𝐼𝑖𝑖𝑖𝑖 + 𝛾𝛾𝑂𝑂𝑖𝑖𝑖𝑖 + 𝑋𝑋𝑖𝑖𝑖𝑖 + 𝜉𝜉𝑖𝑖 + 𝜖𝜖𝑖𝑖𝑖𝑖 4)

where 𝐴𝐴𝑖𝑖𝑖𝑖 is the number of allegations in period t. The variables of interest are whether as of time t an officer has a sustained civilian allegation on his record, 𝑆𝑆𝑖𝑖𝑖𝑖, an internal allegation on his record, 𝐼𝐼𝑖𝑖𝑖𝑖, or an other sustained allegation on his record, 𝑂𝑂𝑖𝑖𝑖𝑖. 𝑋𝑋𝑖𝑖𝑖𝑖 are fixed effects for the district that officer i was assigned to in time t. Here we include officer fixed effects, 𝜉𝜉𝑖𝑖, which nets out differences in officer propensity to have an allegation filed against him or her. We therefore draw on within-officer variation in allegations.

Because the event 𝑆𝑆𝑖𝑖𝑖𝑖 requires at least one allegation in a previous period, the estimated

effect of a sustained allegation could be overstated. One could choose to not count the allegation that was ultimately sustained, but then one allegation is being removed for any party with a sustained allegation, biasing the effects of sustaining an allegation downward. We therefore report two estimates: the first set of estimates includes all allegations, and the second set of estimates excludes allegations that were eventually sustained. The former provides an upper bound on the effect of a sustained allegation, and the latter provides a lower bound estimate of the effect of a sustained allegation.

Panel A of Table 9 presents the results including all allegations. Having had a sustained

civilian allegation reduces the number of allegations filed against an officer each period by 0.17 in Column 1. Columns 2 and 3 report a much smaller effect for a sustained internal or other allegation. Column 4 reports results including all three types of sustained allegations and again finds that a sustained civilian allegation reduces the number of civilian allegations per period by about 0.17. Panel B repeats the regressions in Panel A for the sample that excludes allegations that were later sustained. The coefficient drops to 0.09 but remains precisely estimated. To interpret the magnitude of the effect of a sustained allegation, note that the mean allegation rate per period for all officers is 0.15 and the allegation rate for officers with a themselves make other departments hesitant or reduce officer opportunity or merely are correlated with other factors observable to prospective employers.

22

sustained allegation is 0.28. Thus, having a sustained allegation produces a substantial decrease in the allegation rate.

[Insert Table 7 about here]

The estimates in Table 9 capture the total average effect of a sustained allegation for all

periods observed. Figure 9 investigates the extent to which a sustained allegation deters future bad behavior over time, as measured by the number of civilian allegations. The figure shows the results of an event study regression, where event time 0 is the period in which an allegation was sustained. The reference category is event time 0 so that the test is whether the propensity to draw allegations in periods before the sustained allegation is higher. The line reports the estimated coefficients on event time indicators along with standard error bars (p<0.05). The specification also includes officer fixed effects and district-time period fixed effects. The rate of civilian allegations falls markedly and immediately after an allegation is sustained, and remains at the new low level for all remaining periods.

There is some decrease in the allegation rate in the year prior to a sustained allegation.

Because investigations take on average 10 months from filing of the complaint to disposition, and allegations that are ultimately sustained take over twice as long, we would expect officers and their supervisors to react during the course of an investigation as well (particularly one that is headed toward a finding of misconduct).

We cannot state with certainty whether the reduction in officer allegation rates after a

sustained allegation is a result of incapacitation or specific deterrence. However, we believe that Figure 9 is most consistent with specific deterrence because there is no reversion to pre-sustained allegation levels of allegations after a sustained allegation. Incapacitation would rely on reassignment and supervisor monitoring, the effect of which would decay over time as officer disciplinary records are expunged every five years for promotion purposes and officer assignment relies heavily on seniority (FOP Contract 2011).

[Insert Figure 9 about here]

VI. CONCLUSION In this article, we have explored whether civilian allegations against police officers can

be better utilized to prevent police officer misconduct. To assess the potential for civilian-based police monitoring programs to identify and reduce misconduct, we matched recently released data on civilian allegations against Chicago police officers to a panel of police personnel records and data on federal civil lawsuits and resulting payments. Lawsuits against police officers and resulting payments are external and quantifiable measures of officer misconduct, arising in the most serious cases.

An officer’s history of civilian allegations predicts civil rights litigation and resulting

payments, and an allegation against an officer that is “sustained” by the reviewing agency deters future misconduct by that officer. The results suggest that civilian allegations, if properly managed, could be better utilized to prevent police officer misconduct and could therefore play

23

an important role in reducing resulting costly civil liabilities for state and local governments. Moreover, we find that allegations not supported by an affidavit are as predictive of police misconduct as those with an affidavit.

We conclude that civilian allegations, if properly managed, could play an important role

in reducing police officer misconduct and costly civil liabilities for state and local governments. Many cities are presently considering reforms to police discipline procedures, in particular how civilian allegations are utilized. If our results are generalizable to other cities, our findings suggest that such efforts may produce significant decreases in officer misconduct.

24

References Agreement between Chicago Fraternal Order of Police of Chicago Lodge no. 7 and the

City of Chicago 2012-2017, available at http://www.chicagofop.org/contract/. Brandl, S., M. Stroshine, and J. Frank (2001). Who are the Complaint Prone Officers? An

Examination of the Relationship between Police Officers’ Attributes, Arrest Activity, Assignment, and Citizens’ Complaints about Excessive Force. Journal of Criminal Justice 29, 521–529.

Chicago Police Department Hiring Plan for Sworn Titles (2011), available at https://www.cityofchicago.org/city/en/depts/dhr/supp_info/hiring_plans.html.

Finn, P. (2001). Citizen Review of Police: Approaches and Implementation. Institute of

Justice, U.S. Department of Justice Finney, L. (2015). The Four-Decade Fight for Chicago Police Reform. Chicago Tribune

November 27, 2015 Green, J. and N. Stokey (1983). A Comparison of Tournaments and Contracts. Journal of

Political Economy 91 (3), 349–364. Harmon, R. (2012). The Problem of Policing. Michigan Law Review 110 (5), 761–818. Holmstrom, B. and P. Milgrom (1991). Multitask Principal-Agent Analyses: Incentive

Contracts, Asset Ownership, and Job Design. Journal of Law, Economics, and Organization 7, 24–52.

Harris, C. (2009). Exploring the Relationship between Experience and Problem

Behaviors: A Longitudinal Analysis of Officers from a Large Cohort. Police Quarterly, 12:192–213.

Harris, C. (2010). Problem Officers? Analyzing Problem Behavior Patterns from a Large

Cohort. Journal of Criminal Justice, 38:216–225. Harris, C. and R. Worden (2012). The Effect of Sanctions on Police Misconduct. Crime &

Delinquency. Kane, R. and M. White (2013). Jammed Up: Bad Cops, Police Misconduct, and the New York

City Police Department. New York University Press. Koenker, R. and K. Hallock (2001). Quantile Regression. Journal of Economic Perspectives

15(4), 143-156. Lazear, Edward P., and Paul Oyer (2012). Personnel Economics. Handbook of

Organizational Economics, Chapter 3.

25

Lazear, E. and S. Rosen (1981). Rank-Order Tournaments as Optimum Labor Contracts. Journal of Political Economy 89 (5), 841–864.

Lersch, K. (1998). Predicting Citizen Race in Allegations of Misconduct Against the

Police. Journal of Criminal Justice 26: 87–97. Lersch, K. (2002). Are Citizen Complaints Just Another Measure of Officer Productivity?

An Analysis of Citizen Complaints and Officer Activity Measures. Police Practice and Research 3, 135–147.

Nalebuff, B. and J. Stiglitz (1983). Prizes and Incentives: Towards a General Theory of

Compensation and Competition. Bell Journal of Economics 14 (1), 21–43. Oyer, P. and S. Schaefer (2011). Personnel Economics: Hiring and Incentives. Volume 4,

Part B, Chapter 20 of Handbook of Labor Economics, pp. 1769–1823. Elsevier. Police Accountability Task Force (2016). Recommendations for Reform: Restoring Trust

between the Chicago Police and the Communities They Serve, available at https://chicagopatf.org/wp-content/uploads/2016/04/PATF_Final_Report_4_13_16-1.pdf.

Reingold, P. (2008). Requiem for Section 1983. Duke Journal of Constitutional Law and

Public Policy 3 (1), 1–47. Safer, R., K. Khemani, and J. O’Keefe (2014). Preventing and Disciplining Police

Misconduct: An Independent Review and Recommendations Concern Chicago’s Police Disciplinary System.

Schmadeke, S., D. Heinzmann, and D. Hinkel (2015). Chicago Police Video Hindered by

Errors, Omissions. Chicago Tribune December 23, 2015 Schwartz, M. (2006). Fundamentals of Section 1983 in Ross, Mary M, and Edwin P. Voss.

Sword and Shield: A Practical Approach to Section 1983 Litigation. Chicago, Ill: Section of State and Local Government Law, American Bar Association, 2006.

Sen, S. (2010). Enforcing Police Accountability through Civilian Oversight, SAGE

Publications. Shultz, A. (2015). Early Warning Systems: What’s New? What’s Working? CNA Analysis

and Solutions. St. Clair, S., J. Coen and J. Smith Richards (2016), City Hall often Fights to Keep Files

Secret in Police Abuse Suits. Chicago Tribune, June 24, 2016. Sword and Shield: A Practical Approach to Section 1983 Litigation. Terrill, W., and J. McCluskey (2002). Citizen Complaints and Problem Officers:

Examining Officer Behavior. Journal of Criminal Justice 30(2): 143-155.

26

Walker, S., P. Alpert, and D. Kenney (2001). Early Warning Systems: Responding to the

Problem Police Officer. U.S. Department of Justice, Office of Justice Programs, National Institute of Justice.

Walker, S., and C. Katz (2012). Police in America. McGraw-Hill. Worden, R., C. Harris, and S. McLean (2014). Risk Assessment and Risk Management in

Policing. Policing: An International Journal of Police Strategies & Management 37(2): 239-258. Worden, R., M. Kim, C. Harris, M. Pratte, S. Dorn, and S. Hyland (2012). Intervention

with Problem Officers: An Outcome Evaluation of an EIS Intervention. Criminal Justice and Behavior 40 (4), 409–437.

Worrall, J. (2002). If You Build It, They Will Come: Consequences of Improved Citizen

Complaint Review Procedures. Crime & Delinquency 48(3): 355-379. Van Brunt, A., S. Bedi, L. Bowman, and C. Futterman (2016), Complaint Submitted to the

United States Department of Justice Documenting the Role of the Independent Police Review Authority in Perpetuating a Code of Silence and Culture of Violence in the Chicago Police Department.

Figures

Figure 1: Mean Salary of Chicago Police Officers (PO), Detectives, and Sergeants Over Time

6070

8090

100

110

Ave

rage

Sal

ary

($10

00)

2003

h1

2004

h1

2005

h1

2006

h1

2007

h1

2008

h1

2009

h1

2010

h1

2011

h1

2012

h1

2013

h1

2014

h1

PO Detective Sergeant

27

Figure 2: Chicago Police Officer Misconduct Investigation and Appeals Process

No Punishment Punishment

Arbitrator

Retain

No Punishment Punishment

Court

Transfers

Civilian Files Allegation IPRA

IPRA

Police DepartmentAppeal

28

Figure 3: Allegations Over Time

050

010

0015

0020

0025

00N

umbe

r of

Alle

gatio

ns

2003

h1

2004

h1

2005

h1

2006

h1

2007

h1

2008

h1

2009

h1

2010

h1

2011

h1

2012

h1

2013

h1

2014

h1

Date

Civilian Internal OtherType of Allegation

29

Figure 4: Mean Number of Annual Allegations Per Officer By Chicago Police District

Annual AllegationsPer Officer(.6,.7](.5,.6](.4,.5](.3,.4](.2,.3][.1,.2]

30

Figure 5: Litigation Over Time

010

020

030

0N

umbe

r of

Pol

ice

Offi

cers

2009

h1

2010

h1

2011

h1

2012

h1

2013

h1

2014

h1

Date

Named in LawsuitLawsuit PayoutLawsuit Payout City Council Approved

31

Figure 6: Size of Payout Amounts

050

100

150

200

Num

ber

of P

aym

ents

Mad

eon

Beh

alf o

f an

Offi

cer

1K 10K 100K 1m >10mPayment Amount ($)

32

Figure 7: Histogram of Allegation Residual

0.0

5.1

.15

.2P

ropo

rtio

n of

Offi

cers

-.6 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 >.6Allegation Residual

33

Figure 8: Nonparametric Relationship between Allegations and Litigation

A. Involvement in Litigation Estimates-1

00

1020

3040

50P

ropo

tion

of O

ffice

rs (

%)

10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100Percentile of Civilian Allegation Residual

Named in Lawsuit Payout in Lawsuit

B. Payment Amount Estimates

-.5

0.5

11.

52

2.5

OLS

(ln

($))

-25

025

5075

100

125

Qua

ntile

Reg

ress

ion

($10

00)

10-20 20-30 30-40 40-50 50-60 60-70 70-80 80-90 90-100Percentile of Civilian Allegation Residual

Quantile Regression (p95) OLS34

Figure 9: Event Study for Sustained Allegation

-.15

-.1

-.05

0.0

5.1

.15

Est

imat

ed C

oeffi

cien

ton

Eve

nt T

ime

Dum

my

-9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9<-10 >10Event Time to Sustained Allegation (6 month periods)

35

Tables

Table 1: Police Officer Misconduct Allegations

Number of SustainedType of Misconduct Allegations Rate (%)

Civilian Allegations

First Amendment and Illegal Arrest 23840 0.4

Arrest/Lock-up Procedures 19944 3.2

Search-Related 4624 9.2

Verbal Abuse 2793 2.2

Total 51201 2.4

Internal Allegations

Operation/Personnel Violations 27782 10.4

Supervisory Responsibilities 571 16.1

Total 28353 10.5

“Other” Allegations

Conduct Unbecoming (Off-duty) 1490 28.2

Bribery / Official Corruption 268 6.7

Traffic 1636 4.8

Alcohol Abuse 218 72.0

Drug/Substance Abuse 125 36.0

Criminal Misconduct 864 13.0

Total 4601 18.1

Note: The table reports allegations made between 2002 and 2014.

36

Table 2: Descriptive Statistics

Variable Mean StDev Min Max

Civilian Allegations 0.147 0.489 0 16

Sustained Civilian Allegations 0.004 0.066 0 3

Internal Allegations 0.079 0.304 0 10

Sustained Internal Allegations 0.009 0.097 0 5

Other Allegations 0.014 0.123 0 4

Sustained Other Allegations 0.003 0.053 0 2

Promotion within a year 0.011 0.119 0 1

Exit within a year 0.035 0.196 0 1

Named in Lawsuit 0.019 0.136 0 1

Named in Lawsuit with Payment 0.013 0.111 0 1

Payout Amount ($) 2195 80280 0 7500000

(Conditional on a Payout) 174608 694843 1 7500000

Note: Observation is an officer-6 month period. The table reportsallegations, promotion, exit, and whether an officer was named in alawsuit between 2002 and 2014. The table reports whether an officerwas named in a lawsuit with a payment and the size of the paymentbetween 2009 and 2014.

37

Table 3: Internal and No Affidavit Allegations Predicting Civilian Allegations

Civilian Allegation Residual

w/o withAll Affidavit Affidavit(1) (2) (3)

Residual of 0.416∗∗∗ 0.214∗∗∗ 0.202∗∗∗

Internal Allegations (0.039) (0.029) (0.016)

Residual of 0.674∗∗∗ 0.393∗∗∗ 0.281∗∗∗

Other Allegations (0.072) (0.062) (0.052)

N 14,658 14,658 14,658R2 0.157 0.121 0.123

Note: Bootstrapped standard errors in parentheses andare clustered at the officer level. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗

p<0.01. Each model controls for the number of time pe-riods the officer was observed and the officer appoint-ment year fixed effects. Because the current affidavitrequirement became active in 2007, the table reportsallegations made between 2007 and 2014. 55% of civil-ian allegations did not have an affidavit.

38

Table 4: Effect of Residual of Allegations on Involvement in Civil Litigation

Involvement in Lawsuit

Named in Positive Payout

Lawsuit All >$100k(1) (2) (3)

Residual of 0.656∗∗∗ 0.349∗∗∗ 0.053∗∗∗

Civilian Allegations (0.019) (0.017) (0.010)

N 13,756 13,756 13,756R2 0.158 0.093 0.015Dep Var Mean 0.28 0.11 0.02

Note: Bootstrapped standard errors in parentheses andare clustered at the officer level. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗

p<0.01. Each model controls for the number of time pe-riods the officer was observed and the officer appoint-ment year fixed effects. The table reports allegationsand lawsuits made between 2009 and 2014. Payouts inexcess of $100,000 must be approved by the City Coun-cil.

39

Table 5: Effect of Residual of Allegations on Involvement in Civil Litigation

Amount of Payout

QuantileRegression ($1,000)

OLS p90 p95 p99Residual (1) (2) (3) (4)

Residual of 3.631∗∗∗ 44.3∗∗∗ 128.2∗∗∗ 364.0∗∗∗

Civilian Allegations (0.172) (4.1) (10.2) (69.6)

N 13,756 13,756 13,756 13,756R 2 0.090 0.018 0.034 0.060Dep Var Mean 1.15 21.82 21.82 21.82

Note: Bootstrapped standard errors in parentheses andare clustered at the officer level. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗

p<0.01. Each model controls for the number of time pe-riods the officer was observed. The OLS model controlsfor officer appointment year fixed effects. Amount ofpayout for OLS is in natural log, where $1 was added toeach observation before taking the natural log. PseudoR2 reported for quantile regressions.

40

Table 6: Robustness Check and Decomposition

Named Payout Approved ln(Payout)(1) (2) (3) (4)

Residual of Civilian 0.659∗∗∗ 0.256∗∗∗ 0.053∗∗∗ 2.813∗∗∗

Allegations with Affidavit (0.037) (0.036) (0.016) (0.332)

Residual of Civilian 0.591∗∗∗ 0.501∗∗∗ 0.059∗∗∗ 4.998∗∗∗

Allegations without Affidavit (0.053) (0.047) (0.019) (0.452)

Residual of 0.153∗∗∗ 0.018 -0.020∗ 0.149Internal Allegations (0.042) (0.028) (0.011) (0.325)

Residual of 0.091 -0.188∗∗ -0.003 -1.913∗∗∗

Other Allegations (0.123) (0.076) (0.029) (0.694)

N 13,756 13,756 13,756 13,756R 2 0.159 0.095 0.016 0.092Dep Var Mean 0.28 0.11 0.02 1.15

Note: Bootstrapped standard errors in parentheses and are clustered at theofficer level. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗ p<0.01. Each model controls for thenumber of time periods the officer was observed and the officer appointmentyear fixed effects. For the amount of payment, $1 was added to each observationbefore taking the natural log.

41

Table 7: Robustness Checks – Timing

Named Payout Approved ln(Payout)(1) (2) (3) (4)

a. Remove Allegation

Residual of Civilian Allegations 0.374∗∗∗ 0.260∗∗∗ 0.026∗∗∗ 2.650∗∗∗

(0.025) (0.018) (0.010) (0.184)

N 13,756 13,756 13,756 13,756R2 0.084 0.068 0.010 0.065Dep Var Mean 0.17 0.11 0.02 1.15

B. Litigation Contemporaneous with Allegations

Residual of 0.443∗∗∗ 0.302∗∗∗ 0.034∗∗∗ 3.093∗∗∗

Civilian Allegation (0.025) (0.017) (0.010) (0.188)

N 13,756 13,756 13,756 13,756R2 0.104 0.079 0.011 0.076Dep Var Mean 0.17 0.11 0.02 1.15

C. Post-Allegation Litigation

Residual of 0.462∗∗∗ 0.226∗∗∗ 0.040∗∗∗ 2.380∗∗∗

Civilian Allegation (0.018) (0.015) (0.007) (0.128)

N 12,102 12,102 12,102 12,102R2 0.114 0.068 0.013 0.067Dep Var Mean 0.31 0.12 0.02 1.28

Note: Bootstrapped standard errors in parentheses and are clustered at the officerlevel. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗ p<0.01. Each model controls for the number of timeperiods the officer was observed and the officer appointment year fixed effects.For the amount of payment, $1 was added to each observation before taking thenatural log.

42

Table 8: Civilian Allegations Predicting Promotion and Exit (one-year intervals)

Promotion Exit

(1) (2) (3) (4)

Civilian Allegations 0.037 0.052 -0.714∗∗∗ -0.709∗∗∗

(0.050) (0.050) (0.045) (0.045)

Sustained Civilian Allegations 0.293 0.318 2.172∗∗∗ 2.007∗∗∗

(0.399) (0.399) (0.674) (0.666)

Internal Allegations -0.226∗∗∗ -0.420∗∗∗

(0.059) (0.094)

Sustained Internal Allegations -0.469∗∗∗ 2.948∗∗∗

(0.174) (0.506)

Other Allegations -0.249∗ 1.087∗∗∗

(0.144) (0.323)

Sustained Other Allegations -0.773∗∗∗ 12.557∗∗∗

(0.268) (1.439)

N 230,465 230,465 230,465 230,465R2 0.013 0.013 0.069 0.071Dep Var Mean 1.07 1.07 3.52 3.52

Note: Standard errors in parentheses and are clustered at the officer level.∗ p<0.1, ∗∗ p<0.05, ∗∗∗ p<0.01. Each model controls for officer appoint-ment year fixed effects, and the officer’s current period police district.Models restricted to police officers. Coefficients are in percentage points(multiplied by 100).

43

Table 9: Deterrence: The Effect of Sustained Allegations on Future Allegations

Allegations in Period t

(1) (2) (3) (4)

A. All Allegations

Sustained Civilian -0.171∗∗∗ -0.168∗∗∗

Allegation in the Past (0.015) (0.015)

Sustained Internal -0.041∗∗∗ -0.033∗∗∗

Allegation in the Past (0.010) (0.010)

Sustained Other -0.035∗∗ -0.025∗

Allegation in the Past (0.015) (0.015)

R2 0.245 0.243 0.243 0.245Dep Var Mean 0.15 0.15 0.15 0.15

B. Allegations Not Sustained

Sustained Civilian -0.088∗∗∗ -0.085∗∗∗

Allegation in the Past (0.014) (0.014)

Sustained Internal -0.038∗∗∗ -0.033∗∗∗

Allegation in the Past (0.010) (0.009)

Sustained Other -0.031∗∗ -0.025∗

Allegation in the Past (0.015) (0.015)

R2 0.244 0.243 0.243 0.244Dep Var Mean 0.14 0.14 0.14 0.14

N 276,298 276,327 276,298 276,280

Note: Standard errors in parentheses and are clustered at theofficer level. ∗ p<0.1, ∗∗ p<0.05, ∗∗∗ p<0.01. Each model includesofficer, current police district, time period, and current rank fixedeffects.

44