The Wandering Officer

$ +'( )-% %,%%"$ +'( )-% %,%%"

 %$%*$ %$%*$

%($%'%'! $&'' ( $,$

%$%# (

%($%'$() )*)%',$%$%# (



$' $.'$' $.'

$'*$,"

%$&&&%')

%""%,) ($ ) %$",%'!()))&( %*$%*$* %*",$%$%# (



BEN GRUNWALD & JOHN RAPPAPORT

The Wandering Ocer

abstract. “Wandering ocers” are law-enforcement ocers ﬁred by one department, some-

times for serious misconduct, who then ﬁnd work at another agency. Policing experts hold dispar-

ate views about the extent and character of the wandering-ocer phenomenon. Some insist that

wandering ocers are everywhere—possibly increasingly so—and that they’re dangerous. Others,

however, maintain that critics cherry-pick rare and egregious anecdotes that distort broader reali-

ties. In the absence of systematic data, we simply do not know how common wandering ocers

are or how much of a threat they pose, nor can we know whether and how to address the issue

through policy reform.

In this Article, we conduct the ﬁrst systematic investigation of wandering ocers and possibly

the largest quantitative study of police misconduct of any kind. We introduce a novel data set of

all , full-time law-enforcement ocers employed by almost  dierent agencies in the

State of Florida over a thirty-year period. We report three principal ﬁndings. First, in any given

year during our study, an average of just under , ocers who were previously ﬁred—three

percent of all ocers in the State—worked for Florida agencies. Second, ocers who were ﬁred

from their last job seem to face diculty ﬁnding work. When they do, it takes them a long time,

and they tend to move to smaller agencies with fewer resources in areas with slightly larger com-

munities of color. Interestingly, though, this pattern does not hold for ocers who were ﬁred ear-

lier in their careers. Third, wandering ocers are more likely than both ocers hired as rookies

and those hired as veterans who have never been ﬁred to be ﬁred from their next job or to receive

a complaint for a “moral character violation.” Although we cannot determine the precise reasons

for the ﬁrings, these results suggest that wandering ocers may pose serious risks, particularly

given how dicult it is to ﬁre a police ocer. We consider several plausible explanations for why

departments nonetheless hire wandering ocers and suggest potential policy responses to each.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



authors.

Ben Grunwald is Assistant Professor, Duke University School of Law. John Rap-

paport is Assistant Professor and Ludwig and Hilde Wolf Research Scholar, University of Chicago

Law School. The authors thank Guangya Liu at Duke and Morgen Miller, Rafeh Qureshi, and

Bartosz Woda in the Coase-Sandor Institute at the University of Chicago Law School for help

constructing the data set; Dhammika Dharmapala and Richard McAdams for generously sharing

data; Matt Adler, Will Baude, Panka Bencsik, Sam Buell, Mitch Downey, John de Figueredo, Mi-

chael Frakes, Barry Friedman, Brandon Garrett, Roger Goldman, Hiba Haﬁz, Emma Kaufman,

Kate Levine, Darrell Miller, Richard Myers, Michael Pollack, Kyle Rozema, Seth Stoughton, and

Samuel Walker for feedback and suggestions; Dylan Demello, Carly Gibbs, Morgan Gehrls, Viraj

Paul, and Catherine Prater for research assistance; and Terry Baker at the Florida Department of

Law Enforcement for assistance interpreting the data. The paper also beneﬁted from presentations

at Duke University School of Law, Gonzaga University School of Law, the American Society of

Criminology Annual Conference, the Law and Society Association Conference, the Criminal Law

Roundtable at UNC School of Law, the Junior Faculty Forum at Richmond School of Law, the

Stockholm Criminology Symposium, the ETH Zurich Conference on Data Science and Law, the

European University Institute for the Rule of Law, and the Law of the Police Conference. John

Rappaport acknowledges The Darelyn A. & Richard C. Reed Memorial Fund for ﬁnancial support.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



article contents

introduction 

i. the law-enforcement labor market 

A. Hiring 

B. Discipline 

ii. literature review 

A. Correlates of Police Misconduct 

B. Labor Economics 

iii.data 

A. Automated Training Management System (ATMS) 

B. Supplemental Data Sources 

C. Limitations 

iv. describing the wandering officer 

A. The Law-Enforcement Labor Market in Florida 

. Hirings 

. Separations 

B. The Wandering Ocer 

. How Common Are Wandering Ocers? 

. How Easily Do Wandering Ocers Find New Work? 

a. Reemployment Rates 

b. Time to Reemployment 

c. Distance Traveled for Reemployment 

d. Number of Subsequent Jobs 

. Where Do Wandering Ocers Go? 

a. Agency Size 

b. Agency Resources 

c. Racial Composition 

d. Unemployment 

e. Crime 

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



.

 Do Wandering Ocers Engage in More Misconduct? 

a. Firing 

b. Complaints 

c. Explanations 

v. predicting which wandering officers get fired again 

vi.mechanisms and reforms 

A. Poor Information 

B. Unawareness of Risk 

C. Inadequate Alternatives 

D. Countervailing Beneﬁts 

E. Cost Externalization 

conclusion 

appendix 

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



introduction

With all that has been said and written about the tragic death of twelve-year-

old Tamir Rice in Cleveland, one fact attracts less attention than it should: the

ocer who ﬁred the fatal shot had been “allowed . . . to resign” from his previ-

ous job in Independence, Ohio, aer suering a “dangerous loss of composure”

during ﬁrearms training.

According to his supervisors, Tim Loehmann “would

not be able to substantially cope, or make good decisions” in stressful scenarios.

A year or so later, however, the Cleveland Police Department failed to review

Loehmann’s personnel ﬁle before giving him a gun.

Another Ohio department

later hired Loehmann aer he killed Rice and was ﬁred by Cleveland.

This story is not unique. Consider what happened in tiny Tulia, Texas. In a

massive early-morning raid on July , , police arrested a full ﬁh of Tulia’s

black adults.

Aer parading them across the courthouse lawn in their night-

clothes, Tulia authorities charged the arrestees—roughly forty out of ﬁy of

whom were black—with felony drug oenses. The evidence in each case con-

sisted of the testimony of a single undercover narcotics ocer, Tom Coleman.

Coleman claimed he had purchased drugs, mostly powder cocaine, from each of

the defendants—over one hundred buys in total. Most were convicted, their sen-

tences ranging from  to  years. The State crowned Coleman “Lawman of

the Year.”

1. Shaila Dewan & Richard A. Oppel Jr., In Tamir Rice Case, Many Errors by Cleveland Police,

Then a Fatal One, N.Y.

TIMES (Jan. , ) (internal quotation marks omitted), https://

www.nytimes.com////us/in-tamir-rice-shooting-in-cleveland-many-errors-by

-police-then-a-fatal-one.html [https://perma.cc/QYN-YML].

2. Id.

3. Id.

4. Matthew Haag, Cleveland Ocer Who Killed Tamir Rice Is Hired by an Ohio Police Department,

N.Y.

TIMES (Oct. , ), https://www.nytimes.com////us/timothy-loehmann

-tamir-rice-shooting.html [https://perma.cc/V-SDT]. Loehmann ended up withdraw-

ing his application, however, before commencing work. Amir Vera, Ocer Who Shot Tamir

Rice Withdraws Application to Small Police Department in Ohio, CNN (Oct. , ),

https://www.cnn.com////us/tamir-rice-ocer-application/index.html [https://

perma.cc/TST-BHKS]. He continues to contest Cleveland’s decision to terminate him. Jane

Morice, Appeal Filed on Behalf of Cleveland Police Union to Overturn Firing of Timothy Loehmann,

Ex-Cleveland Cop Who Fatally Shot Tamir Rice, C

LEVELAND (Mar. , ), https://

www.cleveland.com/crime///appeal-ﬁled-on-behalf-of-cleveland-police-union-to

-overturn-ﬁring-of-timothy-loehmann-ex-cleveland-cop-who-fatally-shot-tamir-rice.html

[https://perma.cc/UFC-KKW].

5. The following facts are drawn from NATE BLAKESLEE, TULIA: RACE, COCAINE, AND CORRUP-

TION IN A

SMALL TEXAS TOWN (); and Vanita Gupta, Critical Race Lawyering in Tulia,

Te xa s,  F

ORDHAM L. REV.  ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Under pressure from the media and postconviction litigation, Coleman’s

cases later began to crumble. Coleman, it turns out, had never recorded his buys,

nor were there any witnesses; most of the time, there was no corroboration of

any sort. No drugs, money, or weapons had been seized during the raid. Cole-

man’s written reports were vague. He misidentiﬁed suspects, some of whom had

rock-solid alibis. And marijuana and crack, not powder cocaine, were the preva-

lent vices in Tulia’s impoverished black community. By , Coleman’s credi-

bility was shredded. He was indicted for perjury. Seeing the writing on the wall,

the prosecutors eventually joined the trial judge in recommending that the con-

victions be vacated. In August , the governor pardoned the Tulia defend-

ants.

Much of what brought Coleman down stemmed from what the New York

Times called his “wretched work history.”

His ﬁrst job was at a jail in the City

of Pecos,

where he was “lazy and inattentive at work and in constant danger of

being ﬁred.”

He “abruptly quit” and le the state, only to return and ﬁnd work

as a deputy at the nearby Pecos County Sheri’s Oce.

Aer ﬁve years there,

Coleman again “abruptly le town . . . owing thousands of dollars in delinquent

bills.”

Aer a “brief stint as a jailer” in Denton County, Coleman became a

sheri’s deputy in Cochran County.

He lasted about two years there, skipping

town aer the county attorney witnessed him stealing gas from the county

pumps.

He owed thousands of dollars to local businesses.

The Cochran

County Sheri sent an angry letter about Coleman to the State. “Coleman

should not be in law enforcement,” the sheri wrote, “if he is going to do people

the way he did this town.”

At this point, Coleman managed to join the regional task force that sent him

to Tulia. The task force hired Coleman despite a background check revealing that

he “was a discipline problem, that he was ‘too gung ho,’ that he had been accused

of kidnapping his son in a custody battle, . . . and . . . that he had . . . ‘possible

6. Bob Herbert, Kaa in Tulia, N.Y. TIMES (July , ), https://www.nytimes.com//

//opinion/kaa-in-tulia.html [https://perma.cc/DDD-RX].

7. Nate Blakeslee, The Color of Justice, TEX. OBSERVER (June , ), https://

www.texasobserver.org/-the-color-of-justice [https://perma.cc/FX-LNVB].

8. BLAKESLEE, supra note , at .

9. Id.

10. Blakeslee, supra note .

11. Id.

12. Id.; see BLAKESLEE, supra note , at .

13. BLAKESLEE, supra note , at .

14. Blakeslee, supra note .

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



mental problems.’”

During Coleman’s tenure with the task force, Cochran

County indicted him for stealing the gas and notiﬁed Swisher County, where

Tulia sits.

The Swisher County Sheri’s Oce arrested Coleman—during the

undercover operation—but he never faced trial and the charges were dropped.

Even aer leaving Tulia, Coleman continued to bounce around. In the eighteen

months aer departing, Coleman worked for three dierent task forces. He was

ﬁred from the third, in Waxahachie, for sleeping with a sex worker who was an

informant for his then-employer.

Coleman is the archetypal “wandering ocer,” or what those in policing cir-

cles have called a “gypsy cop.” These are police ocers who are ﬁred or who

resign under threat of termination and later ﬁnd work in law enforcement else-

where.

And although Coleman and Loehmann are prime examples, there are

scores of others. Indeed, as the following examples show, wandering ocers ap-

pear all over.

 While William Melendez was working for the Detroit Police Depart-

ment in , local prosecutors alleged that he had leveled false accusa-

tions of drug possession. The federal government later indicted him for

planting evidence, ﬁling bogus reports, and perjury (he was acquitted).

Melendez was forced out of the department in  aer his license

lapsed. Two other Michigan municipalities—Highland Park and then

Inkster—put Melendez back on the street. In , while in Inkster,

Melendez brutally beat a motorist about the head, leading to a . mil-

lion civil settlement and two criminal convictions.

 While working for the St. Louis Police Department in , Eddie Boyd

III pistol-whipped a twelve-year-old girl; a year later, he struck another

child in the face with his gun or handcus before falsifying a report.

15. BLAKESLEE, supra note , at  (quoting Pecos County Chief Deputy Sheri Cli Harris).

16. Id. at -; Gupta, supra note , at .

17. BLAKESLEE, supra note , at .

18. Id. at .

19. See, e.g., TOM BARKER, POLICE ETHICS: CRISIS IN LAW ENFORCEMENT  (d ed. ); Gordon

Dill, South Carolina Police Shortage Means Employment for “Gypsy” Ocers, N

EWS (Feb. ,

, : PM), https://www.wspa.com/news/south-carolina-police-shortage-means

-employment-for-gypsy-ocers/ [https://perma.cc/KTS-VP] (“A gypsy

cop! That’s been termed an ocer that will jump from agency to agency. They have maybe

 agencies under their belt within a  year period.” (quoting Florence McCants, South Car-

olina Criminal Justice Academy)).

20. Jim Schaefer & Gina Kaufman, How Problem Cops Stay on Michigan’s Streets, DET. FREE PRESS

(Sept. , ), https://www.freep.com/story/news/local/michigan////how

-problem-cops-stay-street/ [https://perma.cc/CRU-HUB].

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Shortly aer he resigned his position with St. Louis, Boyd was hired in

St. Ann, Missouri, and later, again, in Ferguson.

 Nicholas Hogan, an ocer with the Tukwila Police Department in

Washington, pepper-sprayed a suspect who was restrained on a gurney

in a hospital in . Hogan was federally indicted for the act and

Tukwila ﬁred him. In , the police department in nearby Snoqualmie

hired him only to ﬁre him later for having an aair with the wife of a

fellow ocer. He was also subsequently incarcerated for the pepper-

spray incident.

 New Orleans Police Department ocer Carey Dykes was “sued for al-

leged brutality, accused of having sex with a prostitute while on duty

and caught sleeping in his patrol car instead of responding to a shoot-

ing.”

An internal aairs investigation found seventeen violations of de-

partment rules. New Orleans ﬁred Dykes in . Later the same year,

Dykes found police work at the Delgado Community College in New

Orleans and then the Orleans Parish Sheri’s Oce.

Additional examples abound, each as shocking as the last.

Yet the scope and

nature of the wandering-ocer phenomenon are dicult to pin down. Some

21. Timothy Williams, Cast-Out Police Ocers Are Oen Hired in Other Cities, N.Y. TIMES

(Sept. , ), https://www.nytimes.com////us/whereabouts-of-cast-out-police

-ocers-other-cities-oen-hire-them.html [https://perma.cc/DP-WPM].

22. Mike Carter, Jail Time for Ex-Tukwila Cop Who Pepper-Sprayed Handcued Man in Hospital,

EATTLE TIMES (Mar. , ), https://www.seattletimes.com/seattle-news/crime/ex

-tukwila-cop-sentenced-to--months-for-pepper-spraying-handcued-man-in-hospital

[https://perma.cc/GWV-ZR].

23. Kimbriell Kelly et al., Forced Out over Sex, Drugs and Other Infractions, Fired Ocers Find Work

in Other Departments, W

ASH. POST (Dec. , ), https://www.washingtonpost.com

/investigations/forced-out-over-sex-drugs-or-child-abuse-ﬁred-ocers-ﬁnd-work-in

-other-departments////e-da-e-bf-dfc_story.html

[https://perma.cc/MCX-EXK].

24. Id.

25. See, e.g., Anthony Cormier & Matthew Doig, Embattled Ocers Land on Their Feet, HERALD-

RIB. (Sarasota, Fla.) (Dec. , ), https://www.heraldtribune.com/news//special

-report-embattled-ocers-land-on-their-feet [https://perma.cc/KFX-CTTG]; Anthony L.

Fisher, Why It’s So Hard to Stop Bad Cops from Getting New Police Jobs, R

EASON (Sept. , ),

https://reason.com/archives////why-its-so-hard-to-stop-bad-cops-from-ge

[https://perma.cc/MWC-HFH]; Jose Gaspar, McFarland’s Hiring of Four Police Ocers

Raises Questions, C

ALIFORNIAN (Nov. , ), http://www.bakersﬁeld.com/news/jose

-gaspar-mcfarland-s-hiring-of-four-police-ocers-raises/article_dfdf-d--ac

-eeedbe.html [https://perma.cc/CFQ-NDCE]; Gary A. Harki, Still in Uniform: Prob-

lem Police Rarely Lose Certiﬁcation in West Virginia, S

UNDAY GAZETTE-MAIL (Charleston, W.

Va.), Dec. , , at A; David Kroman, “Disqualifying Conduct” Rarely an Obstacle for Fired

Police to Get Rehired, C

ROSSCUT (Apr. , ), https://crosscut.com///ﬁred-ocers

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



experts, from their own experience, or from anecdotes like these, insist that wan-

dering ocers are legion

—and possibly increasingly so.

Others deny that

wandering ocers exist

or discern an exaggerated narrative cobbled together

-can-become-hired-ocers [https://perma.cc/WD-MVK]; Matt Lait, Convicted Cop

Hired as Police Chief, L.A.

TIMES (Feb. , ), http://articles.latimes.com//feb/

/local/me-maywood [https://perma.cc/KQA-ZTGV]; Nomaan Merchant et al., Broken

System Lets Problem Ocers Jump from Job to Job, C

HI. TRIB. (Nov. , ), https://

www.chicagotribune.com/news/nationworld/ct-police-ocer-sexual-misconduct

-investigation--story.html [https://perma.cc/BD-HD]; Christopher N. Osher,

Colorado Laws Allow Rogue Ocers to Stay in Law Enforcement, D

ENV. POST (July , ),

https://www.denverpost.com////colorado-laws-allow-rogue-ocers-to-stay-in

-law-enforcement [https://perma.cc/NQM-SMEM]; Push to Keep “Gypsy Cops” with Ques-

tionable Pasts O the Streets, CBS

NEWS (Sept. , , : AM), https://www.cbsnews.com

/news/gypsy-cops-with-questionable-pasts-hired-by-dierent-departments-lack-of

-oversight-police [https://perma.cc/UCN-UPT]; Casey Toner & Jared Rutecki, The Re-

volving Door: Troubled Ocers Get Frequent Career Chances, WBEZ (Jan. , ), http://

interactive.wbez.org/taking-cover/revolving-door [https://perma.cc/QDT-MX]; Steven

Yoder, How to Keep Bad Cops on the Beat, A

M. PROSPECT (July , ), http://prospect.org

/article/how-keep-bad-cops-beat [https://perma.cc/XR-DQT].

26. See, e.g., BLAKESLEE, supra note , at  (“Everybody’s talking about Tom Coleman—well,

there are whole task forces of Tom Colemans out there.” (quoting Barbara Markham, former

narcotics task force ocer)); Roger Goldman & Steven Puro, Revocation of Police Ocer Cer-

tiﬁcation: A Viable Remedy for Police Misconduct,  S

T. LOUIS U. L.J. ,  () (“Even

when [unﬁt ocers] are terminated, these ocers oen go to work for other departments

within the state.”); Martha L. Shockey-Eckles, Police Culture and the Perpetuation of the Ocer

Shue: The Paradox of Life Behind ‘The Blue Wall’,  H

UMANITY & SOC’Y ,  () (“In

urban areas such as St. Louis, the ocer who resigns rather than face licensure revocation

typically ﬁnds employment in a neighboring municipality with relative ease.”); Richard Ab-

shire, Sheri: Cases Show Staers Not Above Law—Kaufman: He Faults Agencies That Let Ocers

Become “Gypsy Cops,” D

ALLAS MORNING NEWS, Sept. , , at B (“Sheri Byrnes said too

many law enforcement agencies have quietly dismissed problem ocers and not prosecuted

them for criminal conduct, enabling so called ‘gypsy cops’ to go from agency to agency, oen

taking trouble with them.”); Dill, supra note  (“It happens every day. It’s happened here. It

happens everywhere.” (quoting Pacolet, South Carolina Police Chief Raymond Webb));

Candice Norwood, Can States Tackle Police Misconduct with Certiﬁcation Systems?, A

TLANTIC

(Apr. , ), https://www.theatlantic.com/politics/archive///police-misconduct

-decertiﬁcation/ [https://perma.cc/FDF-XYJ] (“There are many cases around the

country where ocers leave their departments because of misconduct and then they are re-

hired—sometimes knowingly, sometimes not—by other departments.” (quoting Professor

Roger Goldman)).

27. See, e.g., Schaefer & Kaufman, supra note  (describing former executive director of the Mich-

igan Commission on Law Enforcement Standards as conceding that the wandering-ocer

phenomenon is “a concern, and could be getting worse because of widespread cuts to police

pay and beneﬁts in recent years”).

28. See, e.g., Sarah Childress, How States Are Moving to Police Bad Cops, FRONTLINE (Apr. , ),

https://www.pbs.org/wgbh/frontline/article/how-states-are-moving-to-police-bad-cops

[https://perma.cc/QCH-GBKD] (“Skeptics of certiﬁcation . . . argue that no police chief or

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



from cherry-picked anecdotes that distort broader realities.

When the rhetoric

is swept away, “[i]t is unclear how far-reaching such problems may be.”

policing expert Samuel Walker has remarked: “It is believed to be a problem

nationwide. The phrase ‘gypsy cops’ has come up. There’s not any solid research

on that. We don’t know how common it is.”

The answer matters. If wandering ocers are rampant and dangerous, iden-

tifying and stopping them should be a police-reform priority—especially be-

cause, by their nature, they touch new communities with each move. And

“[p]oor communities,” writes Monica Bell, “are more likely to hire ‘gypsy

cops’ . . . because their resource constraints make it more dicult for them to

discriminate between good and bad ocers.”

The answer also matters because,

for many individuals, policing represents—indeed, embodies—“the law.”

Law-

enforcement ocers interact with tens of millions of American residents each

year,

many of whom have little other contact with the state.

And “[t]he be-

sheri would hire an ocer with a tarnished record . . . .”); Heather Goldin, Bill Seeks Licens-

ing for Massachusetts Police Ocers, S

ENTINEL & ENTERPRISE (July , ), https://

www.sentinelandenterprise.com////bill-seeks-licensing-for-mass-police-ocers

[https://perma.cc/JQ-DSC]; id. (“It’s not possible for an ocer [ﬁred for misconduct]

to get another job in civil service.” (quoting Ray McGrath, Legislative Director, International

Brotherhood of Police Ocers)); Yoder, supra note  (“[California] doesn’t need to cancel

certiﬁcates . . . because its training program and standards for entering the profession are

among the best in the country—rogue ocers are kept out of the force from the get-go.”).

29. See, e.g., Fisher, supra note  (“Police representatives maintain that these anecdotes are cherry

picked. [The International Brotherhood of Police Ocers’ legislative director] says those who

support a national database [of ocer decertiﬁcations] ‘use these wild examples’ of ‘some-

what outlandish’ cases ‘that happened years ago.’”).

30. Williams, supra note .

31. Harki, supra note ; see Schaefer & Kaufman, supra note  (citing former executive director

of the Michigan Commission on Law Enforcement Standards as saying that the Commission

“does not know how many problem cops there are in Michigan, let alone how many jump

from job to job”).

32. Monica C. Bell, Police Reform and the Dismantling of Legal Estrangement,  YALE L.J. ,

 ().

33. Montré D. Carodine, “Street Cred,”  U.C. DAVIS L. REV. ,  () (internal quotation

marks omitted).

34. See Elizabeth Davis et al., Contacts Between Police and the Public, 2015, BUREAU JUST. STAT. 

(Oct. ), https://www.bjs.gov/content/pub/pdf/cpp.pdf [https://perma.cc/PA

-XFU].

35. See TOM R. TYLER & YUEN J. HUO, TRUST IN THE LAW: ENCOURAGING PUBLIC COOPERATION

WITH THE

POLICE AND COURTS  ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



havior of individual police ocers” in these encounters “communicates infor-

mation to members of the public that they use to make judgments about the

nature of legal authority within their society.”

If wandering ocers are just scapegoats, however, they may distract from

other, more pressing problems in policing. Aer all, just because some wander-

ing ocers commit misconduct does not mean that, ex ante, they were any more

likely to do so than their peers. Plenty of ocers who have never been ﬁred end

up breaking the rules. In some other labor settings, experts have recognized that

past experience does not predict future performance. “Malpractice claims against

physicians,” for example, “are simply too stochastic to lend them much credence

as an indicator of physician quality or risk.”

This Article brings much-needed data to the debate. It presents the ﬁrst

large-scale empirical investigation of the wandering-ocer phenomenon and

possibly the largest quantitative study of police misconduct of any kind.

conduct our analysis using a novel data set that begins with employment records

of all , full-time law-enforcement ocers employed in the State of Florida,

covering nearly ﬁve hundred agencies, over a thirty-year period. Crucially, our

data permits us to distinguish between ocers who separated from their agen-

cies voluntarily and those who separated because they were ﬁred. And for nearly

two decades, it also identiﬁes ocers who resigned while under investigation.

Although we cannot know precisely why all of these ocers were pushed out—

36. Id. at . Tyler and Huo’s work “suggest[s] that personal experiences generalize to shape

broader views about the law and legal institutions,” id. at , as well as one’s “status in the

democratic community,” Vesla M. Weaver, The Only Government I Know: How the Criminal

Justice System Degrades Democratic Citizenship, B

OS. REV. (June , ), http://

bostonreview.net/us/vesla-m-weaver-citizenship-custodial-state-incarceration [https://

perma.cc/FR-XFNZ]; see also Bell, supra note , at  (explaining how current policing

regimes “can operate to eectively banish whole communities from the body politic”); Tom

R. Tyler et al., Street Stops and Police Legitimacy: Teachable Moments in Young Urban Men’s Legal

Socialization,  J.

EMPIRICAL LEGAL STUD. ,  () (asserting that “personal experiences

with the police . . . are [a] key determinant of legal socialization,” which is “the developmental

process by which individuals internalize the norms of the law”). Experiences with the police

“are translated into common stories about who is an equal member of a rule-governed society

and who is subjected to arbitrary surveillance and inquiry.” C

HARLES R. EPP ET AL., PULLED

OVER: HOW POLICE STOPS DEFINE RACE AND CITIZENSHIP  ().

37. Michelle M. Mello & Troyen A. Brennan, Deterrence of Medical Errors: Theory and Evidence for

Malpractice Reform,  T

EX. L. REV. ,  () (citing FRANK A. SLOAN ET AL., INSUR-

ING

MEDICAL MALPRACTICE - ()).

38. In a previous study, touted as “perhaps the largest study of police misconduct ever conducted

in the United States,” Robert Kane and Michael White analyzed the careers of roughly ,

New York City ocers (plus a ,-ocer comparison group) across a twenty-year period.

See R

OBERT J. KANE & MICHAEL D. WHITE, JAMMED UP: BAD COPS, POLICE MISCONDUCT, AND

THE

NEW YORK CITY POLICE DEPARTMENT - (). As we explain, our study covers more

ocers, more jurisdictions, and more years than does the Kane and White study.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



something that can be dicult to ascertain, even with access to an agency’s per-

sonnel records—we do have a general indication, such as whether the ﬁring re-

lated to performance, training, or misconduct. The employment records also

contain demographic information about each ocer—such as age, race, sex, and

education—enabling us to describe the wandering ocer in detail.

We report three principal ﬁndings. First, wandering ocers—deﬁned as of-

ﬁcers who have been ﬁred from a position in law enforcement or corrections be-

fore landing a law-enforcement job at another agency—are fairly common in ab-

solute terms. In any given year, roughly , full-time law-enforcement ocers

who had been previously ﬁred were working for other Florida agencies. And for

reasons we explain, we suspect that this is a low-end estimate, both for Florida

and for what it implies about other jurisdictions. At the same time, when viewed

in relative terms, the number appears more modest: no more than  of ocers

employed in a given year in Florida during the study period were wanderers.

Second, assuming that many ﬁred ocers are seeking new law-enforcement

work and are willing to move to another agency, they seem to face diculty ﬁnd-

ing employment. Ocers who were ﬁred from their immediately preceding job

subsequently obtain work in Florida law enforcement at half the rate of ocers

who separate voluntarily, and the discrepancy is growing over time. Fired oc-

ers also take much longer to start another job and typically move to smaller agen-

cies with fewer resources in communities with slightly higher proportions of res-

idents of color. Interestingly, most of these discrepancies disappear for ocers

who were ﬁred earlier in their career rather than from their immediately preced-

ing job. We hypothesize that agencies view these ocers as having redeemed

themselves.

Third, wandering ocers are far more likely—than both rookies and veter-

ans who have never been ﬁred—to be ﬁred from their next job. They are also

more likely to receive complaints at the state licensing board for “moral character

violations,”

including complaints for violent or sexual misconduct and for in-

tegrity-related misdeeds. We cannot fully rule out the possibility that these ele-

vated risks are due to the characteristics of the agencies that hire wandering of-

ﬁcers or to enhanced monitoring or discipline of these ocers. Perhaps some

agencies even hire wandering ocers on a de facto “probationary” basis, intend-

ing simply to terminate them if problems arise. For reasons we discuss, however,

we are doubtful that these are the principal explanations for our results.

We also explore whether certain ocer or agency characteristics—such as

ocer age or education or agency hiring and training requirements—predict

39. As we describe in more detail below, “moral character violations” can include committing any

felony or certain enumerated misdemeanors (regardless of criminal prosecution) or commit-

ting other speciﬁed acts such as using excessive force or making false statements in a court

proceeding. See F

LA. ADMIN. CODE ANN. r. B-.() ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



which wandering ocers are most likely to fail. The idea is that agencies might

manage risk by screening for certain ocer characteristics or adopting more

stringent hiring or training requirements in the event that, for whatever reason,

they choose to hire a wandering ocer. Unfortunately, we ﬁnd little reason for

optimism on this front, although agencies may have information about ocer

or agency characteristics that is more predictive than what we can observe here.

These ﬁndings present a puzzle: if wandering ocers are so risky, why do

agencies hire them? Our data do not permit us to isolate a single causal mecha-

nism, and the reality is that several are probably at play. First, agencies may hire

wandering ocers because they fail to identify them as such, either due to inad-

equate background checks or candidates’ deliberate concealment of their disci-

plinary history.

The favored solution seems to be improving the existing na-

tional decertiﬁcation database. This database records decisions by state agencies

to “decertify” ocers, which prevent them from working elsewhere in the same

state. Coverage, however, is spotty. Likewise, in , the President’s Task Force

st Century Policing recommended expanding the existing database into a

comprehensive national register to address the problem of ocers who are ﬁred

and decertiﬁed in one state and then move to another state and land a job in law

enforcement there.

We discuss the importance, but also the substantial limita-

tions, of the national decertiﬁcation database as a tool to stop wandering ocers,

and we suggest potential improvements.

Second, agencies might know they are hiring a wandering ocer but be un-

aware that such ocers are, in general, risky hires. Some agencies might, for

example, think an ocer who has been ﬁred will be more conscientious than oth-

ers, if ﬁring acts as a deterrent sanction. As one ocial put it, “You think it’s a

second chance so they’ll try hard, which is what they’re telling you.”

Moreover,

most agencies probably hire too few wandering ocers to notice that, as our data

40. See, e.g., Goldman & Puro, supra note , at  (describing two ocers hired by the West

Palm Beach Police Department despite serious, undiscovered problems at their previous po-

lice departments); Ian Cohen, Questionable Hires, Low Morale Plague Palm Beach Police, P

ALM

BEACH DAILY NEWS (Apr. , ), https://www.palmbeachpost.com/news/

/exclusive-questionable-hires-low-morale-plague-palm-beach-police [https://perma.cc

/MKP-USD] (reporting that the Palm Beach Police Department “ignored or missed red

ﬂags in the applications of its ocers, some of whom had applied [to] and were rejected with

cause from multiple agencies before being accepted by Palm Beach”).

41. See Final Report, PRESIDENT’S TASK FORCE ON ST CENTURY POLICING - (), https://

cops.usdoj.gov/pdf/taskforce/taskforce_ﬁnalreport.pdf [https://perma.cc/SPA-DQT]

(quoting National Decertiﬁcation Index—FAQs, I

NT’L ASS’N DIRECTORS L. ENFORCEMENT

STANDARDS & TRAINING, https://www.iadlest.org/Portals//Files/NDI/FAQ/ndi_faq.html

[https://perma.cc/EAH-QAUQ]).

42. Cormier & Doig, supra note  (quoting Lieutenant David Hubbard of the Eustis, Florida

Police Department).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



suggest, their actual pattern of behavior seems to cut the other way. In light of

our new evidence, law-enforcement agencies should be cautious about hiring

wandering ocers. And agencies that do hire them might invest in enhanced

monitoring and support or, alternatively, adopt recidivist penalties designed to

deter misconduct in this high-risk population.

Third, agencies may know that wandering ocers are risky hires but lack

any better alternatives—that is, the wandering ocers they hire may be less risky

than the alternative candidates. Consistent with our ﬁnding that wandering of-

ﬁcers tend to move to agencies with fewer resources, cash-strapped agencies—

and particularly those in undesirable locations—may be unable to oer compen-

sation competitive enough to attract candidates of higher quality than the wan-

dering ocers they hire. In that case, and assuming ocers must be hired at all,

the solution may be to improve the pool of candidates by raising salaries or re-

ducing barriers to entry. Certainly, if law-enforcement agencies were sophisti-

cated, for-proﬁt entities that internalized the costs of bad hiring decisions, this

story would be compelling. As we discuss in a moment, however, there are rea-

sons to think that agencies do not internalize these costs.

Ideally, to test this third hypothesis, we would compare each wandering of-

ﬁcer who was hired with the “marginal ocer”—the ocer the agency would

have hired had it decided not to hire the wanderer. Unfortunately, we are unable

to identify the actual marginal ocer with our data. We make progress on this

problem by isolating, for each wandering ocer, a plausible candidate cohort—

a group of ocers who were hired around the same time by nearby agencies with

similar budgetary resources. We ﬁnd that wandering ocers are still riskier than

this narrower comparator group, providing some evidence that this third hy-

pothesis is, at best, a partial explanation.

Fourth, agencies may know that wandering ocers are risky but hire them

because of the unique beneﬁts they are perceived to bring. Some agencies, for

example, may actually seek out “cowboy” veteran ocers to work the toughest

beats. Given the “band of brothers” ethos that pervades American policing, some

law-enforcement leaders, too, may feel a “warm glow” upon hiring ocers who

have been cast out by other agencies.

This may explain the seemingly cavalier

attitudes police chiefs sometimes express toward hiring ocers who have been

ﬁred before. “We believe in redemption,” explained one police chief.

“This

43. See, e.g., Barbara E. Armacost, Organizational Culture and Police Misconduct,  GEO. WASH. L.

REV. ,  () (describing how, “[i]n the face of outside criticism, cops tend to circle

the wagons, adopting a ‘code of silence,’ protecting each other, and defending each other’s

actions”).

44. Schaefer & Kaufman, supra note  (quoting Police Chief Chester Logan of Highland Park,

Michigan).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



stu is supposed to follow you forever?” another wondered. “For the rest of your

career? Of course I’m going to give somebody a second chance.”

Relatedly,

agencies might hire wandering ocers who are riskier than the alternatives if the

cost is lower. Given that they typically have little if any discretion over salaries,

it is unlikely that police administrators hire wandering ocers because they are

willing to accept lower salaries than similarly experienced candidates who have

never been ﬁred.

But wandering ocers may be cheaper than fresh recruits, as

most Florida agencies pay police-academy tuition when they onboard rookie

hires. Compared to rookies, wandering ocers are able to hit the streets more

quickly, too.

Finally, agencies may externalize, and thus discount, the costs of hiring wan-

dering ocers. Although agencies nearly always indemnify ocers against ﬁ-

nancial liability, the ocers themselves enjoy qualiﬁed immunity, which, in

practice, protects the agencies as well.

Direct municipal liability for negligent

hiring, moreover, is rare.

And even when municipalities do end up paying for

harms wandering ocers have caused, whether agencies internalize those costs

depends on the institutional and budgetary niceties of municipal governance.

If cost externalization contributes to the hiring of wandering ocers, and we

suspect it does, the appropriate response is to improve existing mechanisms of

accountability. This is in some sense the central challenge of all civil-rights lia-

bility regimes, however; many have tried and failed to accomplish it. Barring

successful accountability reforms, and if future research corroborates our ﬁnd-

ings, states could consider following Connecticut and banning local agencies

from hiring wandering ocers altogether.

The remainder of the Article proceeds as follows. Part I describes the law-

enforcement labor market. Part II reviews the pertinent literature. Part III de-

scribes our data in detail. Part IV presents our ﬁndings about the wandering of-

45. Cormier & Doig, supra note  (quoting Police Chief Roberto Fulgueira of Sweetwater, Flor-

ida).

46. See infra Section VI.D.

47. See, e.g., Devon W. Carbado, Blue-on-Black Violence: A Provisional Model of Some of the Causes,

 G

EO. L.J. , - ().

48. See MICHAEL AVERY ET AL., POLICE MISCONDUCT LAW AND LITIGATION  : (d ed. )

(describing just how dicult it is for plaintis to make out a claim for municipal liability

based on bad hiring).

49. In some jurisdictions, for example, payments come from the general treasury and the agency

is never held accountable. See Joanna C. Schwartz, How Governments Pay: Lawsuits, Budgets,

and Police Reform,  UCLA

L. REV. ,  ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



ﬁcer. Part V examines whether we can predict which wandering ocers are like-

liest to fail. Part VI considers potential causal mechanisms for the wandering-

ocer phenomenon and corresponding reforms.

i. the law-enforcement labor market

Every year, over ﬁeen thousand individual law-enforcement agencies,

spread across ﬁy states, hire thousands of ocers.

Because of this segmenta-

tion, it is challenging to oer a comprehensive description of the law-enforce-

ment labor market or the features of the system that inﬂuence whether a local

agency hires a wandering ocer. In this Part, however, we sketch out general

patterns in the labor market across the states and oer details on Florida—the

site of the current study—as an illustrative example.

A. Hiring

The vast majority of law-enforcement ocers work for county or municipal

agencies; a small number work directly for the state.

In nearly every state, to

become a law-enforcement ocer at any level, an applicant must ﬁrst obtain cer-

tiﬁcation—essentially an occupational license—from a state-level licensing en-

tity.

In most states, this body is called the Peace Ocer Standards and Training

(POST) Board.

Certiﬁcation procedures vary widely from state to state.

In Florida, the certifying entity is the Criminal Justice Standards and Train-

ing Commission (CJSTC), which is part of the Florida Department of Law En-

forcement (FDLE).

To obtain certiﬁcation in the state, candidates must clear a

50. BRIAN A. REAVES, U.S. DEP’T OF JUSTICE, BUREAU OF JUSTICE STATISTICS, LOCAL POLICE DE-

PARTMENTS

, : PERSONNEL, POLICIES, AND PRACTICES  tbl. ().

51. Id.

52. See MATTHEW J. HICKMAN, POST AGENCY CERTIFICATION PRACTICES, , at  () (re-

porting that Hawaii does not have a POST board and the District of Columbia’s POST board

does not certify ocers). On July , , Hawaii enacted legislation to create a POST board,

which was scheduled to ﬁnalize its standards and certiﬁcation process by July , . See 

Haw. Sess. Laws .

53. See Roger Goldman & Steven Puro, Decertiﬁcation of Police: An Alternative to Traditional Rem-

edies for Police Misconduct,  H

ASTINGS CONST. L.Q. , - ().

54. See Roger Goldman, Importance of State Law in Police Reform,  ST. LOUIS U. L.J. , 

(); Goldman & Puro, supra note , at -.

55. FLA. STAT.  ., .() ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



basic abilities test, graduate the police academy, and then pass a written certiﬁ-

cation examination.

They must also meet certain minimum qualiﬁcations re-

garding age, citizenship, and education.

Once certiﬁed, ocers must undergo

continuing training and education to maintain their certiﬁcation.

State law generally regulates the process by which agencies hire ocers. In

Florida, local agencies must conduct a background investigation and gather doc-

umentation to prove compliance with the statewide minimum qualiﬁcations.

State law speciﬁes that background investigations “should include information

setting forth the facts and reasons for any of the applicant’s previous separations

from private or public employment or appointment, as the applicant under-

stands them.”

Implementing regulations require that local agencies “ver-

ify . . . [p]rior criminal justice employments of the applicant and the facts and

reasons for any prior separations of employment”

by “[o]btain[ing] previous

employment data from prior employers.”

Local agencies are expected to con-

tact CJSTC to conﬁrm prior employment and discipline.

As part of the background investigation in Florida, the hiring agency must

conﬁrm that the candidate has “good moral character.”

Under Florida regula-

tions, moral-character violations can include committing any felony or certain

misdemeanors (regardless of criminal prosecution), using excessive force, mis-

using an ocial position to secure a privilege or beneﬁt, participating in sexual

conduct while on duty, engaging in sexual harassment, making false statements

during the job application process, subverting training and testing processes,

and making false statements in a court proceeding.

Regulations also provide

that CJSTC is available to assist local agencies in examining moral character:

56. Id.  .()-(); id.  . (certiﬁcation examination); id.  . (police academy);

id.  . (basic abilities test).

57. Id.  ..

58. Id.  .(), ..

59. Id.  .()-(); FLA. ADMIN. CODE ANN. r. B-.() () (documentation); id. r.

B-. (background investigation).

60. FLA. STAT.  .().

61. FLA. ADMIN. CODE ANN. r. B-.()(a).

62. Id. r. B-.()(a).

63. See Employment Background Investigative Report, No. CJSTC-, https://www.fdle

.state.ﬂ.us/CJSTC/Documents/Rules-Forms/WordDoc/CJSTC------TR.aspx

[https://perma.cc/LGJ-VXE], cited in F

LA. ADMIN. CODE ANN. r. B-.()(a)().

64. See FLA. STAT.  .(); FLA. ADMIN. CODE ANN. r. B-.()(d); see also id. r. B-

.().

65. FLA. ADMIN. CODE ANN. r. B-.().

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



upon request, the CJSTC “shall evaluate the qualiﬁcation of an applicant to de-

termine compliance with ‘good moral character’ pursuant to this rule section.”

The CJSTC’s assistance focuses on the applicant’s criminal history, especially

out-of-state or federal court records.

Within the constraints set forth by state law, local agencies have fairly broad

discretion over hiring. Such discretion is not absolute, however, as it is oen

subject to civil service requirements and sometimes to provisions of a collective-

bargaining agreement with a police ocers’ union.

To facilitate the hiring pro-

cess, most agencies designate certain hiring prerequisites, such as a minimum

age or education level, and a set of screening exams, such as a physical ﬁtness or

driving test or a polygraph examination.

Only candidates who satisfy the pre-

requisites and pass the exams are eligible to be hired. Local agencies may aug-

ment, but not diminish, state-law hiring prerequisites.

The same is true for

continuing education and training.

B. Discipline

Each local agency also administers its own disciplinary process for ocers

who commit crimes or violate agency policy. As with hiring, the agency’s author-

ity over discipline may be circumscribed by civil-service laws or provisions of a

collective-bargaining agreement. Collective-bargaining agreements frequently

provide for arbitration of disciplinary decisions, including termination. Arbitra-

tors commonly order agencies to reinstate terminated ocers.

66. Id. r. B-.().

67. Email from Terry Baker, Training & Research Manager, Fla. Dep’t of Law Enf’t, to John Rap-

paport (Aug. , , : AM CDT) (on ﬁle with John Rappaport).

68. See SAMUEL WALKER & CHARLES M. KATZ, THE POLICE IN AMERICA -,  (th ed. ).

69. See id. at -, -.

70. FLA. STAT.  . (). Florida law, for example, requires ocers to be at least nineteen

years old, id.  .(),

but a local agency is free to raise the minimum age to twenty-one.

71. Id.  ..

72. See Mark Iris, Unbinding Binding Arbitration of Police Discipline: The Public Policy Exception, 

A. J. CRIM. L.  (); Stephen Rushin, Police Disciplinary Appeals,  U. PA. L. REV. 

(); Kimbriell Kelly et al., Fired/Rehired: Police Chiefs Are Oen Forced to Put Ocers

Fired for Misconduct Back on the Streets, W

ASH. POST (Aug. , ), https://

www.washingtonpost.com/graphics//investigations/police-ﬁred-rehired [https://

perma.cc/EQZ-AB].

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



In forty-ﬁve states, the government entity responsible for certifying ocers

also has the power to decertify upon certain conditions.

Ocers who have been

decertiﬁed are prohibited from working in law enforcement anywhere in the

state. As with certiﬁcation, the criteria for decertiﬁcation vary widely among the

states. All states with decertiﬁcation authority, for example, can decertify for fel-

ony convictions, but only  can decertify for failure to meet training or qual-

iﬁcation requirements,  for general misconduct,  for termination for

cause, and  for any misdemeanor conviction.

Like most of its counterparts,

the CJSTC in Florida has the authority to decertify Florida ocers.

Decertiﬁ-

cation can happen when an ocer has committed a felony or a misdemeanor

involving dishonesty (again, regardless of criminal prosecution) or fails to main-

tain good moral character.

Note that this standard, detailed above, covers only

fairly egregious types of misconduct.

The CJSTC learns about potentially disqualifying activity through several

channels, but two are particularly important. First, local agencies are required to

notify the CJSTC whenever an ocer separates from employment, “setting forth

in detail the facts and reasons for such separation.”

Second, local agencies must

conduct an internal investigation when they have cause to suspect that an ocer

has committed a disqualifying crime or moral character violation.

If the

agency’s suspicion is substantiated, the agency must notify the CJSTC.

Florida

is one of the more active states in decertifying ocers even though the substan-

tive scope of its decertiﬁcation authority is not the broadest.

73. See HICKMAN, supra note , at  (reporting that forty-four states allow decertiﬁcation of of-

ﬁcers). Aer Hickman wrote, in October , the responsible agency in New York promul-

gated regulations permitting it to decertify ocers. See N.Y.

COMP. CODES R. & REGS. tit. ,

 . (). New Jersey does not have an agency that decertiﬁes ocers. Certain criminal

convictions, however, can trigger “forfeiture of oce” by court order, which can, in some

cases, entail permanent disqualiﬁcation from holding any public oce. See N.J.

STAT. ANN.

 C:- (West ); State v. Hupka,  A.d , - (N.J. ).

74. HICKMAN, supra note , at . These ﬁgures do not cover New York, which adopted decertiﬁ-

cation regulations aer the report was written. See supra note .

75. FLA. STAT.  .().

76. Id.  .()-(); FLA. ADMIN. CODE ANN. r. B-. ().

77. FLA. STAT.  .(); FLA. ADMIN. CODE ANN. r. B-.().

78. FLA. STAT.  .(); FLA. ADMIN. CODE ANN. r. B-.().

79. FLA. STAT.  .(); FLA. ADMIN. CODE ANN. r. B-.()(b).

80. While Florida is sometimes identiﬁed as the second-highest state (aer Georgia) by number

of decertiﬁcations, see H

ICKMAN, supra note , at , that observation misses a few key points.

First, Florida is the third-largest state by population and therefore has more ocers than most

other states. Second, most ocers decertiﬁed in Florida are corrections ocers, not the law-

enforcement ocers on whom we focus here. In , for example, Florida decertiﬁed  law-

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Given all of these regulations, how do wandering ocers still manage to ﬁnd

work? For starters, local agencies do not always conduct thorough background

investigations before hiring.

Even when they do, past employers are not always

forthcoming and sometimes conceal the real reasons for an ocer’s separation.

Anecdotal evidence suggests that ocers who commit misconduct are oen al-

lowed to resign, with a guaranteed positive work reference, in exchange for for-

going legal action.

Similarly, local agencies do not always notify their state

POST boards about ocer misconduct. Even setting aside cases in which local

agencies disregard mandatory disclosure obligations,

reporting to POST is

wholly voluntary in most states.

Agencies are reportedly reluctant to disclose

negative employment information—either to other local agencies or state POST

boards—for fear of being sued for defamation.

Even more important, as men-

enforcement ocers at a rate of . decertiﬁcations per , ocers, which made it the nine-

teenth-most-frequent decertiﬁer per ocer in the country. See Loren T. Atherley & Matthew

J. Hickman, Ocer Decertiﬁcation and the National Decertiﬁcation Index,  P

OLICE Q. , -

 tbl. (). Still, Florida does decertify ocers more frequently than other big states such

as California, Texas, Pennsylvania, North Carolina, Illinois, and Ohio. For examples of states

with apparently broader decertiﬁcation authority than Florida, see S.D.

CODIFIED LAWS  -

-() (), which permits decertiﬁcation for ocers who “have been discharged from

employment for cause” or “have engaged in conduct unbecoming of a law enforcement of-

ﬁcer”; and W

IS. STAT. ANN.  .()(cm) (West ), which authorizes the POST board

to “[d]ecertify law enforcement . . . ocers who terminate employment or are terminated.”

81. See, e.g., Goldman & Puro, supra note , at ; Cohen, supra note ; Dewan & Oppel, supra

note ; Williams, supra note .

82. See, e.g., Goldman, supra note , at ; Cara E. Rabe-Hemp & Jeremy Braithwaite, An Ex-

ploration of Recidivism and the Ocer Shue in Police Sexual Violence,  P

OLICE Q. , 

(); Williams, supra note . In Florida, speciﬁcally, see Anthony Cormier & Matthew

Doig, Police Agencies Undermine System, H

ERALD-TRIB. (Sarasota, Fla.) (Dec. , , :

AM), http://www.heraldtribune.com/news//special-report-police-agencies

-undermine-system [https://perma.cc/QFA-QEC].

83. See Cormier & Doig, supra note .

84. See HICKMAN, supra note , at .

85. See, e.g., Goldman & Puro, supra note , at ; Steven Puro et al., Police Decertiﬁcation:

Changing Patterns Among the States, 1985-1995,  P

OLICING: INT’L J. POLICE STRATEGIES &

MGMT. , - (); see also J. Hoult Verkeke, Legal Regulation of Employment Reference

Practices,  U.

CHI. L. REV. ,  () (“Providing such negative information creates a

risk of defamation liability while oering few clear beneﬁts to the referring employer. Indeed,

the available empirical evidence suggests that former employers are less likely to reveal em-

ployee misconduct than any other information about the employee.”). Florida law attempts

to ameliorate this and several of the other problems mentioned. See, e.g., F

LA. STAT.

 .() () (requiring background checks); id.  .()(b)-()(a) (requiring

prior employers to disclose disciplinary history and reasons for separation); id.  .()

(providing immunity for disclosure of employment information to a subsequent hiring

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



tioned earlier, many states deﬁne the scope of POST-reportable conduct nar-

rowly—twenty states, for example, require a criminal conviction before an ocer

can be decertiﬁed.

In other words, not all “police misconduct” must be re-

ported even in mandatory-reporting states. In addition, local agencies some-

times learn about prior misconduct and hire the ocers anyway.

Ocer mobility across state lines introduces yet another layer of complexity.

A signiﬁcant problem with state-by-state certiﬁcation is that an ocer decerti-

ﬁed in one state can move across state lines and obtain certiﬁcation, and then

employment, in another. In an eort to address this problem, the International

Association of Directors of Law Enforcement Standards and Training con-

structed a national database called the National Decertiﬁcation Index (NDI).

State POST boards are encouraged to enter their decertiﬁcation decisions into

the database. When a decertiﬁed ocer attempts to ﬁnd employment in another

state, that state’s POST board—or, in some cases, the local hiring agency—can

query the database and review the prior decertiﬁcation record.

Unfortunately, the NDI is far from watertight. As mentioned, ﬁve states plus

the District of Columbia—which collectively employ a signiﬁcant share of all

law-enforcement ocers nationwide—have no decertiﬁcation authority.

Among the majority of states that do decertify ocers, reporting to the NDI is

voluntary.

In , only thirty states contributed to the database; by , that

number rose to thirty-eight.

On the back end, only  local agencies have per-

mission to query the NDI directly when hiring. The rest must rely on their state

POST boards, only twenty-eight of which say they “always” or “frequently”

agency); id.  .() (providing immunity for disclosure to CJSTC). It is unclear how

eective these provisions are and, in any event, many states have no analogs.

86. Fisher, supra note ; Merchant et al., supra note .

87. See HICKMAN, supra note , at  (stating that “four POSTs reported that law enforcement

agencies in their state have hired individuals as ocers who had been decertiﬁed in another

state”). Consider Tom Coleman as well.

88. See Raymond A. Franklin et al., 2009 Survey of POST Agencies Regarding Certiﬁcation Practices,

AT’L CRIM. JUST. REFERENCE SERV. - (), https://www.ncjrs.gov/pdles/nij

/.pdf [https://perma.cc/BJT-RK].

89. See Goldman, supra note , at ; sources cited supra note .

90. Roger L. Goldman, State Revocation of Law Enforcement Ocers’ Licenses and Federal Criminal

Prosecution: An Opportunity for Cooperative Federalism,  S

T. LOUIS U. PUB. L. REV. , 

(); Merchant et al., supra note .

91. HICKMAN, supra note , at .

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



query the NDI.

In , Florida reported that it “occasionally” queries the na-

tional database.

The conﬂuence of all these legal and institutional forces is thought to channel

wandering ocers toward small, understaed, and resource-strapped agen-

cies.

Budget constraints impede thorough background checks. They also make

wandering ocers, who may be prepared to settle for modest salaries and more

limited opportunities for professional advancement, more appealing—especially

where agencies must otherwise foot the bill to put rookie hires through the police

academy. And experienced wandering ocers who are already trained and certi-

ﬁed can hit the streets immediately.

92. Id. at .

93. Franklin et al., supra note , at .

94. For versions of this narrative, see, for example, Atherley & Hickman, supra note , at 

(“[D]ismissal isn’t always the ﬁnal word on the matter. Ocers may be rehired by another

jurisdiction, in which case the new jurisdiction inherits another jurisdiction’s problem. This

can be a conscious decision by the hiring agency, especially in small jurisdictions where ﬁnan-

cial resources are limited and lateral ocers are simply scarce.” (citation omitted)); Bell, supra

note , at  (“[T]he prevalence of very small departments in close proximity to each other

increases the likelihood that an ocer ﬁred from one jurisdiction for serious reasons could

ﬁnd work as an ocer in another. Poor communities are more likely to hire ‘gypsy cops,’

ocers with spotty work histories who have been ﬁred elsewhere, because their resource con-

straints make it more dicult for them to discriminate between good and bad ocers.”);

Goldman, supra note , at , , which describes the pressures faced by small departments

that lead them to hire previously terminated ocers; Goldman & Puro, supra note , at 

(“Although it might seem unusual for a police department to hire an ocer with a past record

of misconduct, the second department is usually located in a poor community that cannot

aord to pay high salaries to its police.”); Shockey-Eckles, supra note , at  (“These mu-

nicipalities, although well known for high crime rates and excessive violence, typically oer

low pay and few beneﬁts to their ocers. Hence, they are the very communities willing to

hire gypsy cops when other departments with more resources are unwilling to do so.”); Chil-

dress, supra note , which notes that “some departments still hire [wandering] ocers, par-

ticularly those that are smaller and strapped for cash”; Cormier & Doig, supra note  (“Vet-

erans in trouble oen ﬁnd second chances by heading down the career ladder, to smaller police

forces in need of experience.”); Dill, supra note , which describes eorts to remedy a state

shortage of police ocers “while trying to avoid problem ocers who bounce from depart-

ment to department”; Toner & Rutecki, supra note , which reports that “poorer communi-

ties” in the Chicago suburbs “are also places where ocers with troubled histories and records

of multiple shootings are oen employed”; Williams, supra note  (“[S]maller departments

and those that lack sucient funding or are understaed are most likely to hire applicants

with problematic pasts if they have completed state-mandated training, which allows depart-

ments to avoid the cost of sending them to the police academy.”); and Yoder, supra note 

(“Even if a background check turns up past rogue behavior, a small department may go ahead

anyway. Such agencies usually are in poor communities that can’t aord high salaries.”).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



ii. literature review

At least two academic literatures provide helpful background on the wander-

ing-ocer phenomenon. First, a number of studies, mostly in criminology, have

examined the correlates of police misconduct. Second, a large literature in labor

economics describes the dynamics of labor markets, largely for professions and

industries other than policing. We summarize each literature in turn.

A. Correlates of Police Misconduct

Empirical research directly examining law-enforcement hiring and separa-

tion has been fairly limited. Perhaps the most pertinent study concerns the New

York City Police Department (NYPD), the nation’s largest law-enforcement

agency. Criminologists Robert Kane and Michael White examined all involun-

tary separations in the NYPD between  and . They identiﬁed , of-

ﬁcers who were separated for so-called “career-ending misconduct” during that

period—roughly  of the , individuals the NYPD had employed.

They

then compared these ocers with randomly selected members of their respective

police academy classes.

Using multivariate analyses, Kane and White identiﬁed dierences between

the study and comparison groups that served as both risk and protective factors

for misconduct. In particular, black ocers were signiﬁcantly more likely than

white ocers to be terminated for misconduct.

Prior criminal history, docu-

mented problems in prior jobs, civilian complaints, and assignment to busy pa-

trols also signiﬁcantly predicted misconduct.

Ocers with associate’s or bach-

elor’s degrees, in contrast, were less likely to be ﬁred for misconduct, as were

ocers who were older when hired or who were married.

Kane and White

concluded that “police departments should continue to invest heavily in pre-em-

ployment screening processes that exclude people who have demonstrated rec-

ords of criminal involvement and employee disciplinary problems” and embrace

95. KANE & WHITE, supra note , at .

96. See id. at -.

97. More precisely, black ocers were more likely to be ﬁred for two out of three types of mis-

conduct. Initially, the same was also true for Hispanic and Asian ocers, but over time, their

separation rates converged with that for white ocers. See id. at .

98. See id. at -, .

99. See id. at -.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



“racial/ethnic diversity and post-secondary educational requirements.”

100

In-

formative as it is, Kane and White’s study was set entirely within a single law-

enforcement agency and does not speak to the lateral movement of ocers

among agencies, our primary interest in this Article.

Kane and White’s ﬁndings are largely consonant with the broader literature

examining ocer-level correlates of police misconduct. Some additional re-

search has also found, for example, that past misconduct predicts future prob-

lems.

101

Unlike Kane and White’s study, much of the research focuses speciﬁcally

on ocer use of force. Studies ﬁnd that younger ocers tend to use force more

oen.

102

So do less experienced ocers,

103

although that may be precisely be-

cause they are younger.

104

Research on female ocers is mixed. Studies have

found that female ocers use less force than male ocers in arrest situations

105

and are less likely to shoot suspects

106

but use similar levels of force in general

citizen encounters.

107

100. Id. at .

101. See, e.g., Samuel Carton et al., Identifying Police Ocers at Risk of Adverse Events,  PROC.

ND ACM SIGKDD INT’L CONF. ON KNOWLEDGE DISCOVERY & DATA MINING , 

(“[O]cers who are routinely found to have been engaged in an adverse event are likely to

engage in another such event in the future.”); James P. McElvain & Augustine J. Kposowa,

Police Ocer Characteristics and the Likelihood of Using Deadly Force,  C

RIM. JUST. & BEHAV.

,  () (“[P]revious history of shootings was a very strong predictor of future shoot-

ings.”); Kyle Rozema & Max Schanzenbach, Good Cop, Bad Cop: Using Civilian Allegations to

Predict Police Misconduct,  A

M. ECON. J.: ECON. POL’Y ,  () (ﬁnding that past civil-

ian allegations predict future misconduct).

102. See, e.g., JOEL H. GARNER & CHRISTOPHER D. MAXWELL, UNDERSTANDING THE PREVALENCE

AND

SEVERITY OF FORCE USED BY AND AGAINST THE POLICE (); Steven G. Brandl et al.,

Who Are the Complaint-Prone Ocers? An Examination of the Relationship Between Police Ocers’

Attributes, Arrest Activity, Assignment, and Citizens’ Complaints About Excessive Force,  J.

CRIM.

JUST. ,  (); Christopher Chapman, Use of Force in Minority Communities Is Related

to Police Education, Age, Experience, and Ethnicity,  P

OLICE PRAC. & RES. ,  ().

103. See, e.g., Eugene A. Paoline, III & William Terrill, Police Education, Experience, and the Use of

Force,  C

RIM. JUST. & BEHAV. ,  (); William Terrill & Stephen D. Mastrofski,

Situational and Ocer-Based Determinants of Police Coercion,  J

UST. Q. , - ().

104. See Chapman, supra note , at  (ﬁnding that, controlling for age, less experienced ocers

use less force).

105. See GARNER & MAXWELL, supra note ; Amie M. Schuck & Cara Rabe-Hemp, Women Police:

The Use of Force by and Against Female Ocers,  W

OMEN & CRIM. JUST.  ().

106. See McElvain & Kposowa, supra note , at .

107. See Eugene A. Paoline, III & William Terrill, Women Police Ocers and the Use of Coercion, 

OMEN & CRIM. JUST. , - () (“[B]oth males and females choose not to invoke

their coercive authority rather similarly (i.e.,  of the female encounters resulted in no co-

ercion versus  for males).”).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



Education, too, has received sustained attention. Studies have found that of-

ﬁcers with more education use less force

108

and are subject to fewer disciplinary

allegations and founded complaints.

109

In tension with Kane and White’s rec-

ommendation, however, agency-level studies have not found that minimum ed-

ucation requirements reduce misconduct or use of force.

110

One possible explana-

tion is that educational requirements shrink the pool of eligible candidates,

excluding otherwise-promising individuals.

111

Empirical scholars have also closely examined the relationship between cer-

tain hiring requirements and police misconduct. Perhaps the largest body of re-

search examines the capacity of psychological exams to predict ocer perfor-

mance and, in particular, to identify candidates likely to have disciplinary

problems. Many studies ﬁnd that personality proﬁles predict performance,

112

108. See, e.g., McElvain & Kposowa, supra note , at ; Jason Rydberg & William Terrill, The

Eect of Higher Education on Police Behavior,  P

OLICE Q. ,  (); Terrill & Mastrofski,

supra note , at , . But see Brandl et al., supra note , at  (“None of the analyses

conducted here would lead one to believe that ocers’ race or level of education played a role

in the receipt of excessive use of force complaints.”).

109. See, e.g., Victor E. Kappeler et al., Police Ocer Higher Education, Citizen Complaints and De-

partmental Rule Violations,  A

M. J. POLICE ,  () (“Although ocers with college de-

grees had fewer citizen-initiated complaints and fewer founded complaints for rudeness, they

did not have signiﬁcantly fewer department-generated complaints for violations of agency

rules and procedures.”); Kim Michelle Lersch & Linda L. Kunzman, Misconduct Allegations

and Higher Education in a Southern Sheri’s Department,  A

M. J. CRIM. JUST. ,  ().

But see Donald M. Truxillo et al., College Education and Police Job Performance: A Ten-Year Study,

 P

UB. PERSONNEL MGMT. ,  () (reporting that police ocers’ education levels

had “an inconsistent relationship with measures of disciplinary action”).

110. See, e.g., David Eitle et al., The Eect of Organizational and Environmental Factors on Police Mis-

conduct,  P

OLICE Q. ,  () (ﬁnding that “neither ﬁeld training nor educational

standards had a statistically discernible association with” police misconduct); Dale W. Willits

& Jerey S. Nowacki, Police Organisation and Deadly Force: An Examination of Variation Across

Large and Small Cities,  P

OLICING & SOC’Y ,  () (reporting that college requirements

and training hours do not have a statistically signiﬁcant relationship to deadly force inci-

dents).

111. See, e.g., Lisa Kay Decker & Robert G. Huckabee, Raising the Age and Education Requirements

for Police Ocers: Will Too Many Women and Minority Candidates Be Excluded?,  P

OLICING:

INT’L J. POLICE STRATEGIES & MGMT. ,  () (“Not surprisingly, raising the educa-

tional requirements for sworn police applicants to require a four-year college degree would

eliminate a large number of the traditionally successful police applicants.”).

112. See, e.g., Ryan M. Roberts et al., Predicting Postprobationary Job Performance of Police Ocers

Using CPI and MMPI-2-RF Test Data Obtained During Preemployment Psychological Screening,

PERSONALITY ASSESSMENT  () (ﬁnding correlations between prehire psychological

test results in California and Minnesota and supervisor ratings for police ocers); Anthony

M. Tarescavage et al., Criterion Validity and Practical Utility of the Minnesota Multiphasic Person-

ality Inventory–2–Restructured Form (MMPI–2–RF) in Assessments of Police Ocer Candidates,

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



though a fair number reach the contrary conclusion.

113

Numerous studies have

also identiﬁed particular psychological proﬁles that predict whether someone

will be a “problem ocer.”

114

As with educational requirements, however, there

is little evidence that psychological exams improve agency-level outcomes such

as civilian complaints or deaths.

115

The empirical evidence on the eects of training is particularly conﬂicted.

Various studies have found that training is negatively,

116

positively,

117

or simply

not

118

associated with adverse outcomes such as the use of force, civilian deaths,

civilian complaints, and general misconduct. One plausible explanation is that

the quality of training may matter more than the quantity. Researchers in one

 J. PERSONALITY ASSESSMENT ,  () (ﬁnding correlations between prehire psycho-

logical test results in Minnesota and supervisor ratings for police ocers).

113. See, e.g., Suzanne Daniels & Emily King, The Predictive Validity of MMPI-2 Content Scales for

Small-Town Police Ocer Performance,  J.

POLICE & CRIM. PSYCHOL. ,  (); Beth A.

Sanders, Using Personality Traits to Predict Police Ocer Performance,  P

OLICING: INT’L J. PO-

LICE

STRATEGIES & MGMT. ,  (). These studies oen measure ocer performance

using ratings given by supervisors, an imperfect outcome measure. See Steven Falkenberg et

al., An Examination of the Constructs Underlying Police Performance Appraisals,  J.

CRIM. JUST.

,  (). Additional studies using more objective outcomes—such as internal investi-

gations, involuntary termination, turnover, and disciplinary complaints—are similarly mixed.

Compare, e.g., Martin Sellbom et al., Identifying MMPI-2 Predictors of Police Ocer Integrity and

Misconduct,  C

RIM. JUST. & BEHAV. ,  (), with, e.g., Jose M. Cortina et al., The

“Big Five” Personality Factors in the IPI and MMPI: Predictors of Police Performance,  P

ERSON-

NEL

PSYCHOL. ,  ().

114. See, e.g., Michael J. Cuttler & Paul M. Muchinsky, Prediction of Law Enforcement Training Per-

formance and Dysfunctional Job Performance with General Mental Ability, Personality, and Life His-

tory Variables,  C

RIM. JUST. & BEHAV. ,  (); Charles D. Sarchione et al., Prediction of

Dysfunctional Job Behaviors Among Law Enforcement Ocers,  J.

APPLIED PSYCHOL. , 

().

115. See, e.g., Liqun Cao et al., A Test of Lundman’s Organizational Product Thesis with Data on Citizen

Complaints,  P

OLICING: INT’L J. POLICE STRATEGIES & MGMT. ,  (); Brad W.

Smith, Structural and Organizational Predictors of Homicide by Police,  P

OLICING: INT’L J. PO-

LICE

STRATEGIES & MGMT. ,  ().

116. See, e.g., EMILY G. OWENS ET AL., PROMOTING OFFICER INTEGRITY THROUGH EARLY ENGAGE-

MENTS AND

PROCEDURAL JUSTICE IN THE SEATTLE POLICE DEPARTMENT (); Cao et al., su-

pra note , at .

117. See, e.g., William C. Bailey, Less-Than-Lethal Weapons and Police-Citizen Killings in U.S. Urban

Areas,  C

RIME & DELINQ. ,  (); Hoon Lee et al., An Examination of Police Use of

Force Utilizing Police Training and Neighborhood Contextual Factors: A Multilevel Analysis,  P

O-

LICING

: INT’L J. OF POLICE STRATEGIES & MGMT.  ().

118. See, e.g., Lee et al., supra note ; Willits & Nowacki, supra note .

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



study, for example, found that roughly a quarter of the variation in ocers’ com-

plaint rates was attributable to the identity of their ﬁeld training ocers, sug-

gesting that qualitatively bad training may hurt more than it helps.

119

B. Labor Economics

Although it does not focus on policing, speciﬁcally, a large body of research

in economics examines the dynamics of labor markets. A number of papers, for

example, study the costs to employers of employee turnover or “churn.” These

costs include the expense of temporarily covering the departing employee’s du-

ties, such as through overtime for other workers; replacement costs, such as

screening new applicants; training costs, including on-the-job training and uni-

forms; lost productivity for the departing employee, who may spend his last days

writing exit memos or laboring with reduced morale; damaged morale for other

workers; and lost institutional knowledge.

120

The costs of churn appear to vary

by region and industry

121

and may, in some settings, be partially or wholly oset

by the beneﬁts of hiring new workers.

122

Nevertheless, one recent literature re-

view concludes that, on average, the cost of replacing an employee is roughly

one-ﬁh of the employee’s salary (excluding the very highest-paid jobs).

123

High turnover costs may, therefore, discourage law-enforcement agencies from

hiring and ﬁring wandering ocers.

A related literature explores the employee-side costs of job separation and

unemployment. One consistent ﬁnding concerns the “unemployment scar”: dis-

placed and nonemployed workers suer long-term earnings losses.

124

Such

119. See Ryan M. Getty et al., How Far from the Tree Does the Apple Fall? Field Training Ocers, Their

Trainees, and Allegations of Misconduct,  C

RIME & DELINQ. ,  ().

120. See Heather Boushey & Sarah Jane Glynn, There Are Signiﬁcant Business Costs to Replacing Em-

ployees, C

TR. FOR AM. PROGRESS  (Nov. , ), https://cdn.americanprogress.org

/wp-content/uploads////CostofTurnover.pdf [https://perma.cc/DR

-YRCK].

121. See id. at -.

122. See Zeynep Ton & Robert S. Huckman, Managing the Impact of Employee Turnover on Perfor-

mance: The Role of Process Conformance,  O

RG. SCI. ,  () (mentioning “improvement

of matches between employees and ﬁrms over time” and increased eort by replacement em-

ployees).

123. Boushey & Glynn, supra note , at .

124. See, e.g., Louis S. Jacobson et al., Earnings Losses of Displaced Workers,  AM. ECON. REV. ,

 () (ﬁnding that high-tenure workers who separate from distressed ﬁrms suer

longer-term earnings losses averaging  per year). See generally William J. Carrington &

Bruce Fallick, Why Do Earnings Fall with Job Displacement?, 

INDUS. REL.  () (re-

viewing the literature).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



workers, for example, may leave the work force or move to ﬁrms that pay lower

wages.

125

Unsurprisingly, earnings losses are greater for workers whom ﬁrms

exercise discretion to ﬁre than for workers displaced by, say, a plant closing—the

former reveals important information about worker quality.

126

Fired workers

also experience longer unemployment spells.

127

Based on these ﬁndings from the

labor economics literature, we expect that wandering ocers will take longer to

land new jobs than other ocers and that they will tend to move to less desirable

agencies.

The concept of “wandering workers” is not itself novel. Journalists have, for

example, penned numerous stories about clergy or teachers who, following mis-

conduct, leave one parish or school and ﬁnd work in another.

128

As far as we can

tell, however, the labor-economics literature has not focused on most of the core

questions that concern us here: which ﬁrms hire displaced workers and how do

those workers fare upon reemployment? The closest study of which we are aware

examines the market for ﬁnancial advisers.

129

The authors ﬁnd that seven per-

cent of working ﬁnancial advisers have misconduct records, and roughly one-

third of these are repeat oenders.

130

Approximately half of these advisers lose

their jobs aer misconduct, yet forty-four percent of them are rehired by other

ﬁrms within a year—they are, in our terminology, “wandering ﬁnancial advis-

ers.”

131

This is true even though advisers with prior misconduct are ﬁve times as

likely as the average adviser to commit misconduct in the future.

132

Still, advisers

125. See, e.g., Fatih Guvenen et al., Heterogeneous Scarring Eects of Full-Year Nonemployment, 

M. ECON. REV.: PAPERS & PROCEEDINGS ,  () (ﬁnding heterogeneous scarring ef-

fects from one year’s nonemployment, resulting primarily from a higher incidence of future

nonemployment rather than lower earnings conditional on working); Kristiina Huttunen et

al., How Destructive Is Creative Destruction? Eects of Job Loss on Job Mobility, Withdrawal and

Income,  J.

EURO. ECON. ASS’N ,  () (ﬁnding that displacement increases the prob-

ability of leaving the labor force by  and, for workers who remain, moderately depresses

earnings due to movement between ﬁrms).

126. See Robert Gibbons & Lawrence F. Katz, Layos and Lemons,  J. LAB. ECON. ,  ().

127. Id.

128. See, e.g., Tara Isabella Burton, Scathing Report Reveals 300 Pennsylvania Catholic Priests Sexually

Abused over 1,000 Children, V

OX (Oct. , ), https://www.vox.com////

/catholic-sex-abuse-priest-crisis-pennsylvania-report [https://perma.cc/GEW-HZG];

Martha Irvine & Robert Tanner, Sexual Misconduct Plagues US Schools, W

ASH. POST

(Oct. , ), http://www.washingtonpost.com/wp-dyn/content/article///

/AR.html [https://perma.cc/MHD-P].

129. Mark Egan et al., The Market for Financial Adviser Misconduct,  J. POL. ECON.  ().

130. Id. at , .

131. Id. at .

132. Id. at .

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



who commit misconduct do experience an elevated likelihood of industry exit,

longer gaps between employment stints, and, for those who ﬁnd new jobs, lower

compensation at smaller, less desirable ﬁrms.

133

Observing that some ﬁrms em-

ploy substantially more wandering advisers than others and that misconduct is

more common in wealthy, elderly, and less educated counties, the authors hy-

pothesize that “misconduct may be targeted at customers who are potentially less

ﬁnancially sophisticated.”

134

To our knowledge, there is no prior quantitative empirical work on wander-

ing police ocers. In the following parts, we report the results of the ﬁrst sys-

tematic exploration of this phenomenon.

Where possible, we examine how the

correlates of misconduct described above interact with wandering-ocer status

and other outcomes of interest.

iii. data

Our primary source of data is the Automated Training Management System

(ATMS), a database constructed and maintained by the FDLE. We supplement

ATMS with a range of other data sources, including an annual survey of Florida

law-enforcement agencies, the United States Census, the Uniform Crime Re-

ports, and others.

A. Automated Training Management System (ATMS)

ATMS compiles information on employment, termination, complaints, dis-

cipline, and demographics for every law-enforcement and corrections ocer

hired in the State of Florida. Our ATMS extract runs up through June .

We rely on three ATMS data sets in particular. First, the “employment” data

set is structured at the ocer-employment level. This means that the same of-

ﬁcer can appear in multiple rows, one for each job he has held. The raw employ-

ment data set has , observations. We drop a substantial number of them.

First, given our focus on wandering law-enforcement ocers, we drop all em-

ployment positions categorized by ATMS as “corrections,” “corrections proba-

tion,” “concurrent” (both law enforcement and corrections), “civilian,” “instruc-

tor,” and “auxiliary.”

135

Doing so removes approximately , rows from the

data. Second, we drop all rows for part-time ocers, focusing only on full-time

133. Id. at -. The ﬁnding concerning the gap between employment stints is driven by advisers

who are not rehired; conditional on being rehired, advisers recently disciplined for miscon-

duct ﬁnd work marginally faster than other advisers. See id. at -.

134. Id. at ; see id. at -.

135. When we include concurrent ocers, the core results of the Article do not change.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



employment. Third, we drop all employment associated with a very small num-

ber of ocers whose full-time employment dates across two or more positions

overlap by more than  days. Fourth, we drop all employment stints beginning

aer  because we have only six months of data for . Fih, we exclude all

employment stints that ended before  due to concerns about whether ATMS

is comprehensive in earlier years.

136

Sixth, we drop all stints that end with oc-

ers “transferring within agency” to other full-time law-enforcement positions

because these are not true separations for our purposes.

137

Before dropping these

stints, however, we assign their start dates to the subsequent employment stints

(to which the ocers transferred) to account for the full length of employment.

Aer applying each of these sample restrictions, the employment data set

contains the , full-time law-enforcement job stints that spanned at least

one day between  and ; they correspond to , unique ocers. Most

of our analyses, however, focus on the , of those stints that began in or

aer  and that correspond to , unique ocers in  unique agencies.

Among those agencies,  are police departments,  are sheris’ oces, and

 are other agencies, which include state-level, university, school, port, and

other kinds of law-enforcement agencies. Table  presents descriptive statistics

at the agency-year level. The bottom row indicates that, on average, agencies

were closed for operation in roughly ﬁeen percent of agency-years; we thus

drop those agency-years from the rest of the descriptive statistics reported in the

table.

138

On average, agencies in the data set employed a mean of  and a me-

dian of  full-time ocers on at least one day in each year. They hired a mean

of . and a median of  full-time ocers per year, and experienced a mean of 

separations and a median of .

136. Our coverage concerns stem from a substantial increase in the volume of hirings in the years

before , which suggests that the database may not have been capturing all employment

stints during that period. To be clear, although we exclude stints that ended before  from

our analytic data set, we still use those stints to determine ocers’ ﬁring histories. To illus-

trate, suppose a law-enforcement ocer was ﬁred from a job in  and then found work

again in . For purposes of our analysis, this ocer would have only one row in our data

set—for the employment stint beginning in —which would indicate that the ocer had

been previously ﬁred.

137. In contrast, when an ocer transfers within agency from a full-time law-enforcement posi-

tion to a part-time or non-law-enforcement position, we keep the employment stint and treat

it as a voluntary separation because the subsequent stint does not meet the eligibility criteria

for our sample. In some cases, multiple full-time law-enforcement ocers “transfer within

agency” on the same day to full-time law-enforcement positions in a dierent agency, likely

because the initial agency was absorbed by another agency. We assume that these are not true

separations. We therefore assign the start date and other relevant information from the ﬁrst

stint to the second and drop the initial stint from the data.

138. We assume an agency is closed in a given year if it has no full-time law-enforcement ocers

employed on any day in that year.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE 1.

agency-year-level descriptive statistics, 1988-2016

Mean Median SD Min. Max.

Officers Employed 98.0 27 241.7 1 3,272

Hirings 8.4 4 16.30 0 353

Separations 5.0 2 8.90 0 237

Voluntary Separations 4.2 2 7.20 0 125

Firings 0.78 0 1.65 0 28

Firings for Misconduct 0.54 0 1.18 0 20

Complaints 0.61 0 1.94 0 85

Agency Closed 0.15 0 0.36 0 1

The employment database provides a range of information on each job, in-

cluding agency name, start and end date, and cause of separation. That last var-

iable is our principal variable of interest. A separation can be either voluntary or

involuntary; it is simply the end of an employment stint, regardless of the rea-

son. We refer to involuntary separations as “terminations,” “involuntary termi-

nations,” or, more colloquially, “ﬁrings.” In total, the variable measuring cause

of separation has thirty-seven dierent code values. Unfortunately, some of

those values have not been used consistently over time. In consultation with the

FDLE, we have grouped the codes to produce two cause-of-separation measures

that correspond to salient separation categories—voluntary and involuntary sep-

arations, or quitting and getting ﬁred—and minimize inconsistencies in data col-

lection over time.

139

To reﬁne the broad category of involuntary separations, we develop two

measures of ﬁring. Our ﬁrst, and narrower, measure captures terminations for

“moral character violations” or violations of a local agency’s policy. For conven-

ience, we refer to these terminations as ﬁrings for “misconduct.”

140

Agency pol-

icy violations may include things like insubordination or failing to follow orders.

139. For the frequency distributions of these codes among law-enforcement positions beginning

between  and  and their inclusion in our cause-of-separation measures, see infra Ta -

ble A.

140. There is no legal authority or academic consensus on the deﬁnition of “police misconduct.”

See K

ANE & WHITE, supra note , at -.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Theoretically, they can also include more minor oenses, such as not having

one’s uniform pressed—agency policy manuals are hey tomes. Generally speak-

ing, however, the types of agency policy violations that warrant termination are

serious or represent the culmination of a pattern of misconduct. It is, aer all,

notoriously dicult to ﬁre a police ocer.

141

The agencies we studied experi-

enced an average of . ﬁrings for misconduct per year, accounting for roughly

. of all separations.

A second, and broader, measure captures all instances in which ocers are

ﬁred for cause. This measure includes ﬁrings for misconduct, but it also includes

terminations for training and performance problems. It does not, however,

count involuntary separations due to downsizing or the dissolution of an agency,

which together account for no more than . of all separations in the data. As

Table  shows, the agencies in our data set saw an average of almost . ﬁrings

for cause per year during the study period, accounting for roughly . of all

separations. We deﬁne the remaining . of the separations as voluntary sep-

arations.

One signiﬁcant methodological problem in any study of police employment

and misconduct is that ocers who are under investigation are frequently al-

lowed to resign before being terminated involuntarily.

142

Fortunately, at least af-

ter , ATMS tracks ocers who resign “in lieu of separation” or “while being

investigated” for misconduct.

143

We include these separations in our ﬁring

measures.

The vast majority of employment stints are easy to deﬁne: they are cleanly

marked by a start and an end date. But there are some edge cases that complicate

that seemingly simple line. Perhaps most important, as mentioned earlier, in

141. See, e.g., Kelly et al., supra note ; Tess Owen, Why It’s Hard to Fire Cops, VICE NEWS (Oct.

, ), https://www.vice.com/en_us/article/bjyxw/why-is-it-so-hard-for-cops-to-get

-ﬁred [https://perma.cc/YD-RPSL]; Mike Riggs, Why Firing a Bad Cop Is Damn Near Im-

possible, R

EASON (Oct. , ), https://reason.com////how-special-rights-for-law

-enforcement-m [https://perma.cc/SKQ-RWC].

142. See, e.g., Jeremy Gorner, 2 Chicago Cops Resign Aer Facing Firing for O-Duty Trac Dispute

That Led to Gunﬁre, C

HI. TRIB. (Mar. , , : PM), https://www.chicagotribune.com

/news/breaking/ct-met-chicago-cop-trac-dispute-police-board--story.html

[https://perma.cc/VC-WU]; see also Cara E. Trombadore, Police Ocer Sexual Miscon-

duct: An Urgent Call to Action in a Context Disproportionately Threatening Women of Color, 

ARV. J. RACIAL & ETHNIC JUST. ,  () (“Thus, an ocer accused of sexual miscon-

duct can resign before an investigation is completed, and then be hired by another department

where he may continue the behavior.”).

143. We suspect that FDLE was already counting these cases under another, more generic ﬁring

code before  because we do not observe any bump in the total number of ﬁrings and

misconduct-related terminations in that year.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



some cases a labor arbitrator will reverse a ﬁring decision, forcing the agency to

reinstate the ﬁred ocer—oen months or years later. Unfortunately, there is no

code in the ATMS database that indicates when a ﬁred ocer is reinstated aer

arbitration. Nevertheless, in a small number of cases—roughly  of all ﬁrings

of full-time law-enforcement ocers from  to —we do observe ﬁred

ocers beginning their next employment stint in the same agency that ﬁred

them. Based on communications with FDLE sta, we suspect that many of these

cases represent arbitral reversals.

144

Arbitral reversals pose not only a data challenge but also a conceptual one.

For reinstated ocers, is the initial term of employment—or what we call the

“preﬁring stint”—and the period aer reinstatement—the “postﬁring stint”—

one employment with a pause in the middle? Or is it two separate employments?

Should the initial ﬁring count as a ﬁring when the department is ultimately

forced to reinstate the ocer? The answers likely vary depending on the research

question we seek to answer. Accordingly, for each of our analyses, we indicate

how we handle ocers who are ﬁred and then rehired by the same agency.

We use our ﬁring measures to construct professional-history variables,

which indicate whether an ocer was ﬁred from his last job or from an earlier

job. In constructing these variables, we include ﬁrings not only from law-en-

forcement positions but also from employment stints in corrections positions.

That is, although our subject is law-enforcement ocers, speciﬁcally—and we

have dropped corrections ocers from the employment data set—we use the

corrections-related data to check whether the law-enforcement ocers we study

previously worked in corrections positions from which they separated involun-

tarily.

In addition to the employment data set, we use an ATMS data set containing

state-level “moral character” complaints against ocers. Most of these com-

plaints were initiated and investigated by the local agencies that employed the

ocers named. Under Florida law, if an agency has cause to believe that an of-

ﬁcer has committed a moral character violation, it must investigate. If the agency

sustains the allegation, it must submit its ﬁndings to the FDLE, which will then

initiate a state-level complaint. The consistency with which local agencies inves-

tigate and report complaints to FDLE likely varies.

145

The FDLE also has the

power to initiate complaints on its own, which are included in the data set. We

use the ATMS complaint data to compute the number of complaints ﬁled against

ocers during each of their employment stints. We do this using the date on

144. See Email from Terry Baker, supra note .

145. See Anthony Cormier & Matthew Doig, Special Report: How Florida’s Problem Ocers

Remain on the Job, H

ERALD-TRIB. (Sarasota, Fla.) (Dec. , , : AM), https://

www.heraldtribune.com/article/LK//News//SH [https://perma.cc

/AW-TJK].

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



which complaints were initiated, which will typically succeed the date on which

the alleged misconduct occurred. That said, based on communications with the

FDLE, we suspect that, in most cases, these two dates are close in time. These

moral-character complaints are rare—agencies experienced an average of

roughly . complaints per year. Each complaint in the data set is also tagged

with “oense codes” that indicate the substantive nature of the misconduct al-

leged, and some complaints are associated with multiple oense codes. We use

these codes to identify the subset of complaints that include any allegation of

violent or sexual misconduct or misconduct that implicates the ocer’s integrity.

Finally, ATMS contains an ocer-level database that provides demographic

information including race, gender, age, and education. Ocer race is desig-

nated as white, black, Hispanic, Asian, or other, and we use these terms through-

out. We merge these demographic data with the employment data.

B. Supplemental Data Sources

We supplement the ATMS database with several other data sources to lever-

age additional information on the agencies that employ wandering ocers. To

collect information on agency hiring and training requirements, we obtained

from the FDLE all data from the Criminal Justice Agency Proﬁle (CJAP), an an-

nual survey of all law-enforcement agencies in Florida that has run from  to

.

146

We extract from CJAP information on hiring and training requirements

for all municipal police agencies and sheris’ oces in the state. We do not cap-

ture data for these variables for other law-enforcement agencies, such as those

in schools, universities, or ports. In total,  unique police departments and

sheris’ oces appear in at least one year of the survey. Some agencies are not

present every year, either because they formed aer  or dissolved before

, or because they did not respond to the survey. We create an agency-year

panel data set in which every agency that appears at least once has a row for each

of the twenty years of our study—, rows in total.

With respect to hiring prerequisites, CJAP gathers information on minimum

age, minimum education, prior criminal-justice experience, and tobacco use.

CJAP also collects information on whether each agency requires a driving his-

tory, an in-person interview, a physical ﬁtness test, a polygraph examination, a

psychological examination, a written exam, a swimming test, a vision test, or a

voice-stress analysis. It also records the length of any probationary employment

146. We are grateful to Guangya Liu for her heroic eorts to extract and process the relevant vari-

ables from CJAP.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



period.

147

To compute a rough estimate of each agency’s overall hiring strin-

gency in a given year, we create a composite measure combining these require-

ments.

148

On average, agencies have a hiring-stringency score of . on our scale.

The average score increased from  to , from . to ..

CJAP also collects information on ongoing training requirements. These in-

clude training on chemical agents, defensive tactics, driving, ﬁrearms, and ﬁrst

aid. In addition, CJAP records the length of the training period required of new

ocers under a ﬁeld training ocer (FTO).

149

As with hiring, we combine the

training requirements to create a composite measure of the stringency of an

agency’s training regimen.

150

On average, agencies have a hiring-stringency

score of .. Agencies’ average composite-training score increased from  to

, from . to ..

We also supplement the data with additional sources of information, which

we describe in greater detail below. First, using Google Maps, we geocoded the

longitude and latitude coordinates of law-enforcement agencies to measure the

distance that wandering ocers travel from one job to the next. Second, we ob-

tained agency-level annual crime data from the FBI’s Uniform Crime Reports.

Third, we gathered county- and city-level information on the racial and ethnic

composition of resident populations and unemployment rates from the United

States Census and Bureau of Labor Statistics. Finally, we obtained information

on county- and city-level law-enforcement expenditures from the Florida De-

partment of Financial Services.

C. Limitations

Although our data is rich and reasonably comprehensive, there are important

limitations worth noting. One substantive limitation is that the separation codes

147. Descriptive statistics for these variables are available in Table A.

148. We construct our composite score by giving an agency one point for every hiring requirement.

To maximize the length of our observation period, we do not include hiring variables that

were excluded from one or more years of the survey. This leads us to drop driving history,

voice-stress analysis, and the swimming test. Three variables—age, education, and probation-

ary period—are not binary. We therefore specify a threshold at which to assign an agency a

point for these requirements. Based on the statistics reported in Table A, we assign agencies

a point for requiring ocers to be older than nineteen, to have an associate’s degree or college

credit, and to undergo more than twelve months of probationary employment.

149. Descriptive statistics for these variables are available in Table A.

150. To construct the composite-training score, we give agencies one point for requiring training

in chemical agents, defensive tactics, driving, ﬁrearms, or ﬁrst aid every six months or year.

We give them an additional point for requiring more than twelve months of training under

an FTO.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



we use to construct our cause-of-separation measures (listed in Table A) reveal

only the general reason for each separation, such as a failure to complete training

requirements, budgetary constraints, misconduct, or a voluntary departure.

Within the all-important category of ﬁrings for misconduct, we are not able to

identify the speciﬁc nature of the misconduct, such as excessive force, embezzle-

ment, substance abuse, and so on. That said, even if we had more speciﬁc infor-

mation on the ocial reason for termination, it would remain dicult in many

cases to determine reliably the actual, underlying conduct at issue. As just one

example, NYPD ocer Martin Tisdale shot and killed a woman during a strug-

gle over his ﬁrearm, ﬂed the scene, and then disposed of the ﬁrearm. The ocial

reason for his termination was recorded as “failure to safeguard a weapon.”

151

Other limitations relate to the geographic scope of our data, all of which are

drawn from the State of Florida. We cannot, for example, observe ocers who

were ﬁred out of state and then obtained law-enforcement work in Florida. That

our data is limited to one state also raises questions about the external validity of

our results—that is, the extent to which our conclusions generalize to other lo-

cations. It is certainly possible that law-enforcement labor markets vary from

state to state in ways that implicate our research questions. Indeed, we note be-

low some reasons to believe that wandering ocers may be relatively more prev-

alent in some other states,

152

except for Connecticut, which bans their hiring al-

together.

153

At the same time, we are unaware of any reason to think that Florida

is idiosyncratic in pertinent respects.

A national study, it bears noting, would not be practicable at this time. Alt-

hough we have not conducted an exhaustive survey, we are unaware of any other

state that collects and makes available the type of data contained in Florida’s

ATMS. There are, moreover, methodological advantages to working within a

single state. For one thing, any state-level covariates, such as state law or other

state characteristics, are held constant, simplifying statistical analysis and reduc-

ing the risk of omitted variable bias. In addition, that a single entity (a state

agency) collects all of the critical data signiﬁcantly mitigates concerns about data

consistency.

iv. describing the wandering officer

Thousands of law-enforcement ocers begin and end jobs in Florida each

year. Some are hired for the ﬁrst time; some for the third or fourth. Some sepa-

151. KANE & WHITE, supra note , at -.

152. See infra Section IV.B..

153. See infra Section VI.E.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



rate voluntarily to retire or change careers; others are ﬁred for grievous miscon-

duct. Section IV.A presents an overview of the law-enforcement labor market in

Florida, describing hiring patterns and then separations. Section IV.B details the

prevalence, characteristics, movement patterns, and behavior of wandering of-

ﬁcers.

A. The Law-Enforcement Labor Market in

Florida

1. H

irings

The law-enforcement labor market in Florida is large. As the black line in

Figure  reveals, Florida agencies hired between , and , full-time oc-

ers in most years between  and .

154

In general, trends in hiring appear

to track conditions in the wider American economy. The number of ocers hired

each year dropped dramatically during the economic recession in  and .

It then began rising again until it peaked in the late s and fell during the

Great Recession, from  to . Since then, the number has been steadily

rising.

154. To give a complete picture of the ATMS database over time, all of the data we report in this

subsection include both the pre- and postﬁring terms for ocers who were ﬁred and then

rehired by the same agency for their next job. As noted, the postﬁring term likely represents

an employment stint resulting from an arbitrator’s decision to reinstate the ocer. See supra

Section III.A.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



FIGURE 1.

total number of law-enforcement officers hired and separated, 1988-2016

Men held the

vast majority of full-time law-enforcement jobs in Florida from

 to —roughly . Most jobs also went to white ocers——with

 to black ocers and  to Hispanic ocers. Educational information is

available for about  of the jobs in the employment database. Of those, 

were held by ocers with only a high school education. Another  were held

by ocers with an associate’s degree and  by ocers with a bachelor’s de-

gree. Just  were held by ocers with a master’s degree.

155

2. Separations

urning to separations, Figure  shows that, in most years from  to ,

about , to , ocers separated from their jobs, whether voluntarily or

involuntarily. Unsurprisingly, the number of separations tracks the number of

hires, rising in most years except during recessions, when fewer jobs are availa-

ble.

155. Although our data identify Asian ocers and ocers with doctoral degrees, we do not report

them here because they are too scarce. We also decline to report ﬁgures where ocer charac-

teristics are unknown or where ocer race is coded as “Other.”

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



In Figure , we disaggregate separations by cause.

156

From  to , an

average of  of separations were involuntary—meaning that the ocer was

ﬁred—while the remaining  were voluntary. The proportion of involuntary

separations fell in the s, from roughly  to , but has remained rela-

tively stable since then. The same basic pattern characterizes ﬁrings for miscon-

duct, speciﬁcally.

FIGURE 2.

proportion of separations by cause of separation, 1988-2016

Table  b

reaks down by demographic characteristics the proportions of sep-

arations that were ﬁrings and ﬁrings for misconduct, respectively. The most

striking observation is that black ocers were most likely to be ﬁred, both in

general and for misconduct:  of all separations involving black ocers were

ﬁrings and  were ﬁrings for misconduct. Those rates were substantially

higher than the rates for Hispanic ocers—who were ﬁred  of the time and

ﬁred for misconduct  of the time—and white ocers—who were ﬁred 

of the time and ﬁred for misconduct  of the time. To be clear, these ﬁgures do

156. To give a complete picture of separations in the ATMS database, throughout this subsection

we count all involuntary separations, including those in the pre- and postﬁring terms for of-

ﬁcers who were ﬁred and then rehired by the same agency for their next job.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



not necessarily imply any problem with the relative performance of black oc-

ers. Black ocers, for example, may be ﬁred more oen due to discrimination in

the disciplinary process

157

or because they are disproportionately assigned duties

that present elevated opportunities for misconduct.

158

Educa

tional background is also correlated with involuntary termination, at

least within the  of employment stints for which education data are availa-

ble. Ocers with only a high school education are most likely to be ﬁred, and

the rate of ﬁring generally decreases with higher educational attainment. Ocers

without a four-year college degree are also most likely to be ﬁred for misconduct.

TABLE 2.

cause of separation by demographic groups, 1988-2016

Demographics n Fired

Fired for

Misconduct

Race

White 69,103

11.7%

8.6%

Black 8,228 21.6% 14.8%

Hispanic 8,323

17.8% 11.7%

Gender

Male 75,946

13.3%

9.7%

Female 11,155 13.4% 7.6%

Education

High School 4,028 16.9% 8.8%

Associate’s 12,728 12.1% 8.5%

Bachelor’s 22,503 9.5% 6.3%

Master’s 3,637 7.2% 4.7%

Overall 87,116 13.3% 9.5%

157. See, e.g., KANE & WHITE, supra note , at , - (discussing possible heightened scrutiny

of black ocers due to “tokenism”); Kate Levine, Discipline and Policing,  D

UKE L.J. ,

- ().

158. See KANE & WHITE, supra note , at ; see also James J. Fyfe, Police Use of Deadly Force: Re-

search and Reform,  J

UST. Q. ,  () (“Disparities in on-duty shooting rates were

attributable largely to racial dierences in rank and assignment.”); William A. Geller & Kevin

J. Karales, Shootings of and by Chicago Police: Uncommon Crisis—Part I: Shootings by Chicago

Police,  J.

CRIM. L. & CRIMINOLOGY ,  () (similar).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



B. The Wandering Ocer

Despite the salience of wandering ocers, we have remarkably little system-

atic data about them. In this Section, we attempt to answer four basic questions

that frequently arise in public discourse about police misconduct. First, how

common are wandering ocers? Second, how easy is it for them to land new

jobs? Third, where do they go? And fourth, are they really a problem, in the sense

that they cause more harms than other ocers to the communities they police?

We take each of these questions in turn.

Before we begin, a brief deﬁnitional note. There is no legal or even informal

consensus deﬁnition to tell us who, exactly, counts as a wandering ocer. We

therefore adopt an expansive deﬁnition: a wandering ocer is someone who was

ﬁred from at least one full-time law-enforcement or corrections position in the

State of Florida and later lands another full-time law-enforcement job in the

state. At times, we also break wandering ocers into two smaller groups—those

who were ﬁred from their last job and those who were ﬁred from a job earlier in

their employment history—because the results for these groups dier in im-

portant respects.

1. How Common Are Wandering Ocers?

In absolute numbers, wandering ocers are fairly common. From  to

, an average of roughly , full-time law-enforcement ocers who had

previously been ﬁred, and just under  ocers who had been ﬁred for mis-

conduct, were employed by new agencies in any given year.

159

Figure  depicts

the number of wandering ocers employed over time. As the dark and light gray

lines show, the number of wandering ocers employed throughout the state in

any given year has been relatively stable over time, with a slight decrease in re-

cent years. The black line also shows that the number of all ocers—divided by

ten to depict all three lines on the same axis—has been increasing steadily over

time. In , just under , full-time ocers had been previously ﬁred and

just over  had been ﬁred for misconduct. To be clear, these counts include

only wandering ocers, not all ocers who worked in law enforcement aer hav-

ing been ﬁred. If we include ocers who were rehired by the same agency that

ﬁred them, the counts are even higher: from  to , on average, roughly

, and  ocers were employed who had previously been ﬁred or ﬁred for

misconduct, respectively.

159. In this Section we exclude from our ocer counts the post-ﬁring stints of ocers who were

ﬁred and then rehired by the same agency (before being hired at any other agency). While

these ocers have previously been ﬁred, they are not wandering ocers because they have

not, as of yet, moved to another agency.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



FIGURE 3.

number of employed officers by professional history, 1988-2016

When viewed in relative terms, the prevalence of wandering ocers seems

more limited. Almost  of all ocers employed in any given year were wander-

ing ocers previously ﬁred and just  were wandering ocers previously ﬁred

for misconduct. As Figure  shows, the relative proportion of wandering ocers

has fallen gradually over time. This is partly because—as we observed Figure —

the total number of ocers employed has increased. By , just over  of all

ocers employed were wandering ocers who had ever been ﬁred and .

were wandering ocers who had been ﬁred for misconduct.

FIGURE 4.

percent of all employed officers who are wandering officers, 1988-2016

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



Whether there are many wandering ocers or few, therefore, may be in the

eye of the beholder. In absolute terms, Florida law-enforcement agencies employ

many wanderers: in recent years, roughly , wandering ocers who had pre-

viously been ﬁred and about  who had been ﬁred for misconduct. In total,

these ocers likely interact with hundreds of thousands of civilians each year.

160

et when viewed in relative terms, we see that the proportion of wandering of-

ﬁcers is small and decreasing gradually over time.

Still, we suspect that our ﬁgures underestimate the actual number of wan-

dering ocers for at least three reasons. First, because our data is limited to Flor-

ida agencies, we are unable to identify wandering ocers who were previously

ﬁred from law-enforcement agencies in other states. Second, a small number of

ocers may successfully obscure their employment history, even within the

Florida market. Some who have been ﬁred may simply lie and get away with it.

Others may have negotiated an apparently “voluntary” separation in exchange

for separating without a legal ﬁght.

161

Third, other states may have more wan-

dering ocers than Florida does. Florida has a robust statewide data system

tracking ocer employment and requires hiring agencies to investigate appli-

cants’ employment history. Although it may be possible for an applicant to con-

ceal a prior ﬁring, it is probably not easy. Florida also decertiﬁes more ocers

than many other states, and a decertiﬁed ocer cannot subsequently gain em-

ployment with any Florida agency. Because our estimate of the volume of wan-

dering ocers is likely a lowball—both for Florida, speciﬁcally, and for other

states—it is hard to conclude that wandering ocers are a negligible phenome-

non.

How Easily Do Wandering Ocers Find New Work?

A second core question is how much diculty ocers have ﬁnding work af-

ter being ﬁred.

162

In this Section, we probe this question along several dimen-

sions, including how oen ﬁred ocers land a new job, how long it takes them

to do so (assuming they are looking for work), how far they have to travel, and

how many dierent jobs they tend to hold. We count as reemployment only full-

160. Drawing upon ocial Florida data and prior academic research, Jordan Blair Woods recently

estimated that law-enforcement ocers in Florida have conducted between . million and

. million trac stops annually during the past decade. See Jordan Blair Woods, Policing,

Danger Narratives, and Routine Trac Stops,  M

ICH. L. REV. ,  (). If wandering

ocers conducted a proportional number of these stops—three percent—they would have

conducted between , and , stops each year.

161. See supra note  and accompanying text.

162. See, e.g., Shockey-Eckles, supra note , at  (asserting that ﬁred ocers who resign rather

than face license revocation ﬁnd work “with relative ease”).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



time law-enforcement jobs in public agencies in Florida. We exclude jobs that

end with a ﬁring followed immediately by reemployment with the same agency,

a pattern that, as mentioned earlier, likely reﬂects reinstatement by an arbitrator

rather than the ocer’s eorts and success on the job market.

a. Reemployment Rates

First, do ocers who have been ﬁred obtain law-enforcement work less oen

than ocers who have not been? Interestingly, the answer depends on when

during their careers they were ﬁred. To explore this issue, we exclude all employ-

ment stints ending aer  to allow all ocers in the data at least three years

to obtain a new position.

As the bottom row of Panel A in Table  reports, from  to ,  of

ocers who separated and had never been ﬁred obtained a new full-time law-

enforcement position in Florida. In contrast, just  of ocers who were ﬁred

from their last job secured a new position, which represents a statistically signif-

icant dierence.

163

Yet ocers who were ﬁred further back in their employment

history obtained a new job at a rate similar to ocers who had never been

ﬁred.

164

The rest of Table  shows that the same basic results hold across ocer

characteristics and when, in Panel B, we group ocers by their history of ﬁrings

for misconduct, speciﬁcally.

165

163. p < .. All of the statistical tests in Section IV.B. are two-sided t-tests clustered at the

person and agency level.

164. While the dierence is substantively small, it is statistically signiﬁcant at the . level. Ideally,

we would also calculate rehiring rates while excluding ocers who have decided to retire, to

concentrate on the subset of separated ocers who were most plausibly looking for law-en-

forcement work. We do not have data on retirements for ﬁred ocers, however. As an imper-

fect alternative, we exclude ocers who were over ﬁy years of age at the time of separation;

in a separate analysis, we exclude ocers with at least twenty-ﬁve years of full-time service at

the time of separation. The basic pattern of results for both analyses is substantively similar

to the results we report in the text.

166. Sample sizes are shown in parentheses.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE 3.

rehiring rates by demographic groups and professional history, 1988-2013

166

166. Sample sizes are shown in parentheses.

Panel A: Fired Panel B: Fired for Misconduct

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Race

White

37%

(51,538)

18%

(6,702)

41%

(2,498)

37%

(53,908)

15%

(5,177)

42%

(1,653)

Black

40%

(5,073)

14%

(1,451)

48%

(414)

40%

(5,626)

12%

(1,074)

44%

(238)

Hispanic

49%

(5,164)

19%

(1,123)

51%

(304)

47%

(5,607)

17%

(795)

52%

(189)

Gender

Male

39%

(54,841)

17%

(8,283)

44%

(2,928)

39%

(57,692)

15%

(6,431)

45%

(1,929)

Female

35%

(7,926)

16%

(1,151)

32%

(311)

34%

(8,506)

13%

(717)

27%

(165)

Education

High

School

56%

(1,664)

19%

(284)

60%

(60)

54%

(1,813)

19%

(160)

60%

(35)

Associate’s

36%

(9,359)

20%

(1,249)

40%

(453)

36%

(9,816)

17%

(940)

42%

(305)

Bachelor’s

41%

(16,996)

21%

(1,657)

45%

(596)

41%

(17,732)

18%

(1,174)

45%

(343)

Master’s

52%

(2,739)

31%

(187)

55%

(126)

52%

(2,849)

24%

(136)

58%

(67)

All

39%

(62,780)

17%

(9,436)

43%

(3,239)

38%

(66,212)

15%

(7,149)

44%

(2,094)

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



e next examine reemployment rates over time. Figure  shows the propor-

tion of ocers who obtained a new job within three years, conditional on their

professional history of ﬁrings.

167

We limit our rehiring measure to three years to

address a potential censoring problem: ocers who separated in the last few

years had less time to secure a new job. If we did not limit the rehiring measure

in this way, censoring might severely deﬂate the rehiring rate in the last few years

relative to years prior.

As Figure  shows, the lower reemployment rates we observe for ﬁred ocers

date back at least to the late s. Moreover, the reemployment rate for ocers

ﬁred from their last job fell in the early s, from  to roughly . That

number fell again around , probably because—as we showed above—law-

enforcement hiring generally fell throughout the state during the Great Reces-

sion. Since then, the reemployment rate for ﬁred ocers appears to have in-

creased slightly and may continue to do so if the  drop was driven primarily

by the economic downturn.

FIGURE 5.

proportion of separations in which officer obtains subsequent employment

within three years, by professional history of firing, 1988-2013

167. See Figure A for similar results on ﬁrings for misconduct, speciﬁcally.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



We cannot be sure why reemployment rates are so dierent for ocers who

were ﬁred from their most recent job versus ocers who were ﬁred from a job

further back in their employment history. One possible explanation is that the

subset of ocers who landed an intervening job were initially ﬁred for conduct

that, on average, was less egregious than the ocers who were ﬁred from their

last job. Another possibility is that wandering ocers who voluntarily separated

from their last job seek employment in law enforcement at higher rates than

those who were ﬁred from their last job. A third potential explanation is that

law-enforcement agencies believe at least some previously ﬁred ocers are “re-

deemed” if they have held at least one other job in the intervening period without

having been ﬁred. We interrogate this “redemption” story in further detail be-

low.

b. Time to Reemployment

To assess how much ﬁred ocers may struggle to obtain new employment,

we also consider how long it takes them to secure their next job. Folk wisdom

says not long. Our data, however, show something dierent.

As the bottom row of Table  reports, ocers who were ﬁred or ﬁred for

misconduct from their most recent job who later obtain another job take sub-

stantially longer to do so, on average, than ocers who have never been ﬁred or

ﬁred for misconduct—more than  days longer.

168

In Table A, we report the

median time to reemployment rather than the mean. The same basic pattern is

present, but the dierence is even starker.

169

These ﬁndings are driven, at least

partially, by the fact that ocers who separate voluntarily oen do not leave until

they have another job lined up.

170

Once again, as the rest of Table  shows, ﬁr-

ings further back in time are substantially less important and there is relatively

little variation in these patterns across demographic groups.

168. p < ..

169. The median time for ocers ﬁred ( days) or ﬁred for misconduct ( days) from their

previous job is over  times longer than for ocers who have never been ﬁred ( days) or

ﬁred for misconduct ( days).

170. We can see, for example, that many ocers who separate voluntarily begin a new employment

stint the day aer ﬁnishing the last one.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE 4.

mean time to new employment by demographic groups and professional

history, 1988-2013

171

Panel A: Fired Panel B: Fired for Misconduct

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Race

White

400

(19,280)

685

(1,175)

443

(1,024)

402

(19,977)

782

(800)

454

(702)

Black

317

(2,049)

703

(210)

209

(198)

325

(2,224)

729

(129)

213

(104)

Hispanic

316

(2,519)

787

(212)

402

(154)

322

(2,649)

910

(137)

473

(99)

Gender

Male

378

(21,451)

694

(1,441)

407

(1,288)

380

(22,327)

785

(986)

428

(867)

Female

430

(2,750)

767

(180)

355

(98)

435

(2,892)

886

(92)

402

(44)

Education

High

School

416

(937)

859

(55)

665

(36)

423

(977)

1,038

(30)

789

(21)

Associate’s

370

(3,401)

674

(247)

483

(179)

372

(3,538)

785

(162)

535

(127)

Bachelor’s

375

(6,982)

686

(347)

295

(270)

378

(7,236)

760

(210)

286

(153)

Master’s

440

(1,436)

774

(58)

586

(69)

439

(1,491)

897

(33)

854

(39)

All

384

(24,205)

703

(1,621)

403

(1,386)

386

(25,223)

794

(1,078)

426

(911)

171. Sample sizes are shown in parentheses.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



c. Distance Traveled for Reemployment

We also probe how much wandering ocers may struggle to ﬁnd new work

by examining how far they travel to obtain their next job. The conventional wis-

dom is that wandering ocers obtain jobs at agencies that are relatively far away

and so have less (oen informal) information about their past conduct. Surpris-

ingly, we ﬁnd little evidence of such dierential movement.

172

Among ocers who have never been ﬁred, those who separate and ﬁnd new

employment move to an agency that is, on average, forty-two miles away from

their last job. Similarly, ocers who were ﬁred from their last job or some job

further back in their employment history move to agencies that are, on average,

forty-four and forty-ﬁve miles away, respectively. The dierence in median dis-

tances was just slightly larger.

173

We also ﬁnd little dierence in the movement

patterns of ocers who were ﬁred for misconduct, speciﬁcally, either from their

most recent job or earlier.

d. Number of Subsequent Jobs

Finally, we examine how many full-time jobs wandering ocers hold over

the course of their careers and, perhaps more important, aer their ﬁrst ﬁring.

As a baseline for comparison, Figure  shows the proportion of ocers who,

over the course of an entire career, worked a particular number of full-time

jobs.

174

One thing is clear: most ocers move around very little. The vast ma-

jority hold no more than one full-time job, and virtually all of them—roughly

—hold no more than two.

172. To obtain geographic coordinates, we geocoded the names of each agency using an R package

called “ggmaps” that automatically runs queries on Google Maps. In total, we were able to

obtain geographic coordinates for  of the agencies in our sample. Most of the agencies we

could not geocode are state-level agencies with ambiguous (and potentially multiple) geo-

graphic locations. In total, we are missing geographic distance information for  of all em-

ployment stints from which an ocer separated and later obtained another job.

173. Ocers who have never been ﬁred move a median of fourteen miles. Ocers who were ﬁred

from their last job move a median of twenty miles, while ocers ﬁred further back in their

employment history move a median of seventeen miles.

174. We do not count the postﬁring stint for ocers who were ﬁred and then rehired by the same

agency.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Of course, not all ocers who work more than one job are wandering oc-

ers; many were never ﬁred at all. The top panel of Figure  shows the proportion

of ocers who, aer having been ﬁred, worked a particular number of subsequent

full-time jobs; the bottom panel presents the same information for ocers who

were ﬁred for misconduct. Both are consistent with what we already know: the

vast majority of ﬁred ocers—around —never secure another job. Moreo-

ver, very few—fewer than —obtain more than one additional full-time job.

Virtually none obtain more than three. Wandering ocers “jump[ing] from

agency to agency” who “may have  agencies under their belt within a  year

period” therefore appear to be exceedingly rare, if not apocryphal, at least as far

as our data can detect.

175

FIGURE 6.

number of full-time positions held by unique officers, 1988-2016

175. Dill, supra note .

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



FIGURE 7.

number of full-time positions held by unique officers after being fired for

the first time, 1988-2016

Taken tog

ether, our data suggest that ocers who were ﬁred from their most

recent job may face signiﬁcant challenges in securing new law-enforcement work

in Florida. They are half as likely as other ocers to obtain a new position and

it takes them twice as long to do so. Moreover, the vast majority of ocers who

were ﬁred hold a very small number of full-time positions throughout their ca-

reer—virtually all of them have fewer than three. Interestingly, we also ﬁnd evi-

dence that ﬁrings from earlier in an ocer’s career appear to pose a much smaller

obstacle to ﬁnding a new job.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



3. Where Do Wandering Ocers Go?

If ﬁred ocers struggle to ﬁnd work, they might seek employment at smaller

agencies, which may have fewer resources, or at agencies that serve socioeco-

nomically disadvantaged communities of color with higher crime. Various ac-

counts advance these hypotheses.

176

To test them empirically, we compare, for

each wandering ocer, characteristics of the previous agency in the year of sep-

aration to characteristics of the hiring agency in the year of hiring. We exclude all

years prior to  because, for most of our agency-level variables, we lack data

before then. As in the previous Section, we also exclude jobs that end with a

ﬁring followed by reemployment with the same agency, which likely reﬂects re-

instatement by an arbitrator rather than the discretionary decisions of ocers

and agencies on the market.

Our data bear out some, but not all, of the hypotheses about the movement

patterns of wandering ocers. In particular, we ﬁnd that wandering ocers tend

to migrate to agencies with fewer resources in communities with slightly higher

proportions of residents of color. Ocers who were just ﬁred tend to move to

smaller agencies as well. We ﬁnd no evidence, however, that wandering ocers

move to areas with more unemployment or crime.

a. Agency Size

Ocers who were ﬁred from their last job move to smaller agencies on aver-

age. We calculate agency size using the ATMS employment data set by counting

the number of unique full-time law-enforcement ocers employed in each

agency each year. From  to , ocers who had never been ﬁred and who

landed a new job moved, on average, from agencies with roughly  ocers to

agencies with roughly  ocers—representing a relative increase of nearly .

In contrast, ocers who were ﬁred or ﬁred for misconduct from their previous

job and who obtained a new position moved from agencies with  and  of-

ﬁcers to agencies with  and  ocers, respectively—relative decreases of

 and . These dierences between the increase experienced by ocers

who had never been ﬁred, on the one hand, and the decreases experienced by

wandering ocers, on the other, are both large and statistically signiﬁcant.

177

Ocers who were ﬁred at some point before their most recent separation

actually move in the opposite direction. These ocers went from agencies with,

176. See sources cited supra note .

177. p < .. All of the statistical tests in Section IV.B. are two-sided t-tests clustered at the

person and agency level.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



on average,  ocers to agencies with  ocers—a relative increase of

roughly . This increase is statistically signiﬁcantly dierent from the change

in agency size experienced by ocers who had never been ﬁred.

178

Note, how-

ever, that ocers who were ﬁred at some point continue to work at smaller agen-

cies, on average, than ocers who had never been ﬁred.

b. Agency Resources

Wandering ocers appear to migrate toward agencies with fewer resources.

To examine this issue, we collected data on law-enforcement expenditures by

every county and municipality in Florida from the Florida Department of Finan-

cial Services from  to . The data set, therefore, contains information

only on municipal police departments and sheris’ oces and not state-level,

university, school, or port agencies.

179

Our expenditure ﬁgures include “personal

services” and “operating costs”—which cover salaries and beneﬁts—but exclude

“capital outlays.”

180

Ocers who had never been ﬁred and landed a new job moved, on average,

from agencies with operating budgets of . million to agencies with budgets

of . million—a relative increase of . In contrast, ocers who were ﬁred

from their last job moved, on average, from agencies with . million budgets

to agencies with . million budgets—which represents a  decline that is

statistically signiﬁcantly dierent from the change experienced by ocers who

had never been ﬁred.

181

Similarly, ocers who were ﬁred for misconduct from

their last job moved, on average, from agencies with . million budgets to

agencies with . million budgets—a . decline that is statistically signiﬁ-

cantly dierent from the change experienced by ocers who had never been

178. p < .. When we do the same comparison for ﬁrings for misconduct, however, the results

are not statistically signiﬁcantly dierent.

179. We are missing expenditure data for roughly  of all employment stints, but nearly all of

those missing observations—almost —involve an ocer moving to or from a law-en-

forcement agency that is neither a sheri’s oce nor a police department. Another limitation

of our data is that they are reported by municipalities and counties, not by the law-enforce-

ment agencies themselves. Some municipalities contract with other municipalities or counties

for policing services. Our estimates of agency budgets may exclude funds provided to an

agency by another municipality.

180. We exclude capital outlays because they are spiky over time and because, while they poten-

tially support an agency for many years, we observe only the year in which the money was

spent and not the years in which the beneﬁts of the purchase actually accrued.

181. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



ﬁred for misconduct.

182

As before, the experience of ocers who were ﬁred fur-

ther back in their employment history more closely resembles that of ocers

who had never been ﬁred. Ocers who were ﬁred further back in their career

moved, on average, from agencies with . million-dollar budgets to agencies

with . million-dollar budgets—a relative increase of  that is not statis-

tically signiﬁcantly dierent from the change experienced by ocers who had

never been ﬁred.

The same basic patterns emerge when we examine agencies’ budgetary dol-

lars per ocer, though the dierences are smaller in magnitude.

183

Ocers who

had never been ﬁred and obtained a new job moved, on average, from agencies

with , per ocer to agencies with , per ocer—a relative increase

of . In contrast, ocers ﬁred from their last job moved, on average, from

agencies with , per ocer to agencies with , per ocer—a relative

decline of , which is statistically signiﬁcantly dierent from the change ex-

perienced by ocers who had never been ﬁred.

184

Similarly, ocers ﬁred for

misconduct from their most recent job moved, on average, from agencies with

, per ocer to agencies with , per ocer—a relative decline of 

that is statistically signiﬁcantly dierent from the change experienced by ocers

who had never been ﬁred.

185

Yet again, the behavior of ocers who were ﬁred

further back in their employment history more closely resembles the behavior of

ocers who had never been ﬁred. Ocers who were ﬁred at some point further

back in their employment history moved, on average, from agencies with

, per ocer to agencies with , per ocer, which is not statistically

signiﬁcantly dierent from the increase experienced by ocers who had never

been ﬁred.

c. Racial Composition

Our data suggest that wandering ocers tend to move to areas with slightly

higher proportions of residents of color. We use municipal measures of race and

ethnicity from the Census for municipal agencies, and county-level measures for

182. p < ..

183. For this measure, we divide the agency’s budget by the total number of ocers employed in

that same year. Because some law-enforcement agencies also employ correctional ocers, we

include these ocers in this estimate of agency size. We also ﬁnd the same basic patterns when

we examine budgetary dollars per resident rather than per ocer.

184. p < ..

185. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



all other agencies.

186

Although county-level estimates are available annually, mu-

nicipal estimates are not. We therefore use decennial measures from , ,

and . We apply the data from each of those Census years to all subsequent

years until the next decennial census.

187

From  to , ocers who had never been ﬁred and found new work

moved, on average, from agencies in areas with black populations averaging

. to agencies in areas with black populations averaging .—a relative de-

crease of . By contrast, ocers who were ﬁred from their last job and found

new work moved, on average, from agencies in areas with black populations of

. to agencies in areas with black populations of .—a relative increase

of , which is a statistically signiﬁcant dierence from the change experienced

by ocers who had never been ﬁred.

188

Similarly, ocers who were ﬁred for

misconduct moved from agencies in areas with black populations of . to

agencies in areas with black populations of .—a relative increase of ,

which is statistically signiﬁcantly dierent from the change experienced by oc-

ers who had never been ﬁred for misconduct.

189

There is no statistically signiﬁ-

cant dierence between the movement patterns of ocers who were ﬁred further

back in their career—whether or not for misconduct—and ocers who had never

been ﬁred. The same basic patterns emerge if we examine population data for

Hispanic residents.

190

As we have shown so far, our data support certain aspects of the conventional

wisdom about wandering ocers—many move to smaller agencies, with fewer

resources, in communities with slightly more residents of color. But not every

186. Our race data is missing for roughly  of all employment stints because it is mostly limited

to municipal police departments and sheris’ oces. Indeed, roughly  of all the missing

observations come from employment stints in which an ocer moves from or to an agency

that is neither a municipal police department nor a sheri’s oce.

187. In other words, we assign  data to all years between  and ; we assign  data

to all years between  and ; and we assign  data to all years aer .

188. p < ..

189. p < ..

190. Ocers who had never been ﬁred moved, on average, from agencies in areas with  His-

panic populations to agencies in areas with . Hispanic populations—a relative decrease

of . Ocers who were ﬁred from their last job moved in the other direction, from agencies

in areas with . Hispanic populations to agencies in areas with . Hispanic popula-

tions—a relative increase of , which is statistically signiﬁcantly dierent from the change

experienced by ocers who had never been ﬁred (p < .). Similarly, ocers ﬁred for mis-

conduct for their last job moved from agencies in areas with . Hispanic populations to

agencies in areas with . Hispanic populations—a relative increase of , which is not

statistically signiﬁcantly dierent from the change experienced by ocers who had never been

ﬁred for misconduct. And once again, ocers who were ﬁred further back in their employ-

ment history behaved similarly to ocers who had never been ﬁred.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



element of the conventional wisdom is borne out by the data. We turn to those

other elements next.

d. Unemployment

Contrary to the conventional wisdom, we ﬁnd little evidence that wandering

ocers move to agencies in communities with higher unemployment, which we

use as a rough proxy for socioeconomic well-being.

191

This result might appear

at odds with our earlier ﬁnding that wandering ocers move to agencies in areas

with larger communities of color, but the magnitude of the change in racial com-

position was quite small.

We use two measures of unemployment rates, which produce similar

(though not identical) results. Our ﬁrst measure is annual county-level unem-

ployment rates from  to , which are available from the Bureau of Justice

Statistics Current Population Survey.

192

During that period, ocers who had

never been ﬁred and secured a new position moved, on average, from agencies

in counties with an unemployment rate of . to agencies in counties with a

rate of .. Ocers who were ﬁred from their most recent job and secured sub-

sequent employment moved from agencies in counties with an unemployment

rate of ., on average, to agencies in counties with a rate of .—a change

that is statistically signiﬁcantly dierent from the change experienced by ocers

who had never been ﬁred.

193

Ocers who were ﬁred for misconduct experienced

a similar increase, moving from agencies in counties with an unemployment rate

of ., on average, to agencies in counties with an average rate of nearly .—

a change that is also statistically signiﬁcantly dierent from that of ocers who

had never been ﬁred.

194

While these changes are statistically signiﬁcant, they are

191. See, e.g., John P. Crank, The Inﬂuence of Environmental and Organizational Factors on Police Style

in Urban and Rural Environments,  J.

RES. CRIME & DELINQ. ,  () (using unem-

ployment rate as one of two measures of community economic conditions in examining their

relationship with policing style); Vickie L. Shavers, Measurement of Socioeconomic Status in

Health Disparities Research,  J.

NAT’L MED. ASS’N , - () (listing unemploy-

ment rate as a “commonly used” measure of community socioeconomic status in public-health

research); see also Bell, supra note , at  (asserting that “[p]oor communities are more

likely to hire ‘gypsy cops’”).

192. Our unemployment data is missing for roughly  of all employment stints because it is

limited to municipal police departments and sheris’ oces. Indeed, roughly  of all the

missing cases come from employment stints in which an ocer moves from or to an agency

that is neither a municipal police department nor a sheri’s oce.

193. p < ..

194. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



substantively very small.

195

When we look at the data for ﬁrings further back in

an ocer’s career, once again we ﬁnd little dierence between ocers who have

and have not been ﬁred. If anything, ocers who were ﬁred earlier in their career

appear to go to agencies in counties with slightly lower unemployment.

Our county-level estimates of unemployment, however, may mask variation

within counties, especially as municipal agencies, which serve municipalities,

make up the majority of all agencies in our data set. To address this problem, we

construct a second measure of unemployment that assigns municipal-level un-

employment data to municipal agencies. Unfortunately, municipal-level data is

available only back to . We therefore assign all agencies the relevant county-

level or municipal-level unemployment rate from  for all years.

196

This ap-

proach is not ideal, as it requires the strong assumption that any changes in the

unemployment rate over time are constant across all localities. In other words,

in unmasking spatial variation in unemployment rates within counties, we are

forced to mask temporal variation. Still, we think the analysis useful as a check

on our results above.

Based on this second measure, ocers who had never been ﬁred and landed

another job moved, on average, from agencies in localities with an unemploy-

ment rate of . to agencies in localities with an unemployment rate of ..

Ocers who were ﬁred from their last job moved, on average, from agencies in

localities with an unemployment rate of . to agencies in localities with a rate

of ., while ocers who were ﬁred for misconduct from their last job moved

from agencies in localities with a rate of . to those with a rate of .. Nei-

ther of these latter changes is statistically signiﬁcantly dierent from the change

experienced by ocers who had never been ﬁred. Finally, ocers who were ﬁred

or ﬁred for misconduct further back in their career moved from agencies in lo-

calities with roughly . unemployment to agencies in localities with the same

rate.

Thus, our ﬁrst and second measures of unemployment point to similar an-

swers. The ﬁrst measure suggests that wandering ocers migrate toward agen-

cies in communities with very slightly more unemployment than other ocers—

from about . to .—while our second measure ﬁnds no evidence at all of

any dierence in migration patterns for wandering ocers.

195. Indeed, a . or . change in the county-level unemployment rate is well within the or-

dinary range of annual ﬂuctuations for the counties in our data set.

196. Our measure of unemployment data is missing for  of all employment stints from 

to . Roughly  of the missing cases come from employment stints in which an ocer

moves from or to an agency that is neither a municipal police department nor a sheri’s oce.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



e. Crime

We also ﬁnd little evidence that wandering ocers are more likely than other

ocers to migrate toward areas with more crime. We measure crime using data

from the FBI’s Uniform Crime Reports that capture “oenses known to the po-

lice” from  to .

197

As it turns out, both wandering and non-wandering

ocers tended to move to agencies with less crime, and, if anything, wandering

ocers tended to experience larger decreases in crime from one job to the next—

although the dierence in the size of these drops was not statistically signiﬁcant.

From  to , ocers who had never been ﬁred and landed a new job

moved, on average, from an agency with , violent crimes per , resi-

dents annually to an agency with , such crimes—a relative decrease of .

Ocers who were ﬁred from their last job moved, on average, from an agency

with , violent crimes per , residents to an agency with , such

crimes—a relative decrease of , which is not statistically signiﬁcantly dier-

ent from the decrease experienced by ocers who had never been ﬁred. Ocers

who were ﬁred for misconduct from their last job experienced an even greater

decrease, moving from agencies with , violent crimes per , residents

to agencies with , such crimes—a relative decrease of , which is also not

statistically signiﬁcantly dierent from the change experienced by ocers who

had never been ﬁred for misconduct.

198

Once again, ocers who were ﬁred from

a job earlier in their employment history behaved similarly to ocers who had

never been ﬁred.

199

Taken together, these results suggest that wandering ocers

do not move to agencies in communities with more crime.

197. Our crime data begin in  because a large number of agencies did not report crime data in

. Even between  and , we are missing crime data for roughly  of our employ-

ment stints. But nearly all of those missing observations——involve an ocer moving to

or from a law-enforcement agency that is neither a sheri’s oce nor a police department.

Both the crime and population data are at the agency level.

198. We observe a similar pattern using property crime. Ocers who had never been ﬁred and

found a new job moved from an agency with , property crimes per ,, on average,

to an agency with , such crimes, a relative decrease of . Ocers who were ﬁred from

their most recent job moved from agencies with , property crimes per , to agen-

cies with , such crimes—a relative decrease of . Similarly, ocers who were ﬁred for

misconduct from their last job and found new work moved from agencies with an average of

, property crimes per , to agencies with ,—a relative decrease of .

199. Ocers who were ﬁred further back in their career moved, on average, from an agency with

, violent crimes per , to an agency with , such crimes—a relative decrease of

, which is similar to the decrease experienced by ocers who had never been ﬁred. And

ocers who were ﬁred for misconduct at some point before their most recent job moved from

agencies with an average of , violent crimes per , to agencies with ,—a relative

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



4. Do Wandering Ocers Engage in More Misconduct?

Many worry that when wandering ocers move, they “tak[e] trouble with

them.”

200

In the absence of systematic data, however, analysts have not known

whether this fear is justiﬁed. In this Section, we examine employment and dis-

ciplinary data to assess whether wandering ocers seem to pose heightened risks

to the communities they are hired to serve. We ﬁnd that they do. We then con-

sider potential explanations for these ﬁndings.

Because one goal of our analysis is to enable critical evaluation of the choice

to hire wandering ocers, we focus on employment stints in which agencies ac-

tually exercise their discretion to bring an ocer onto the force. Throughout this

Section, therefore, we drop the post-ﬁring employment stint for ocers who are

ﬁred and then rehired by the same agency. As noted, many of these stints likely

represent cases in which an arbitrator forced the agency to reinstate the ocer

against its will.

a. Firing

We consider ﬁrst whether wandering ocers are ﬁred more oen than other

ocers. We begin by examining ﬁring rates across groups of ocers with dier-

ent professional histories.

201

As Panel A in Table  shows, from  to ,

202

ocers who had never been ﬁred and who secured a new position were subse-

quently ﬁred . of the time and ﬁred for misconduct . of the time.

203

contrast, ocers who were ﬁred from their most recent position and landed a

new job were ﬁred and ﬁred for misconduct, respectively, . and . of the

time. This is more than twice as oen, and the dierence is statistically signiﬁ-

cant.

204

Ocers who voluntarily separated from their last position but who had

been ﬁred at some point earlier in their career were subsequently ﬁred and ﬁred

for misconduct . and . of the time, respectively. This is roughly 

more oen than ocers who had never been ﬁred, and the dierence is again

decrease of , which is not statistically signiﬁcantly dierent from the change experienced

by ocers who had never been ﬁred for misconduct.

200. Abshire, supra note , at B.

201. We do not break out these ﬁring rates by demographic categories due to small sample sizes.

202. We exclude stints that began aer  to ensure we have at least three years of follow-up for

every observation. The results are substantively similar if we start our analysis in  rather

than , as we did in the previous Section due to data limitations. See Table A.

203. The denominator includes both employment stints that have ended and those that have not

yet ended.

204. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



statistically signiﬁcant.

205

Because wandering and non-wandering ocers may

remain in their jobs for dierent lengths of time, we also report ﬁring rates

within three-year windows (as well as one- and ﬁve-year windows in Table A),

and the results are substantively similar. The same basic pattern holds in Panel

B when we examine ocers’ history of ﬁrings for misconduct, speciﬁcally.

206

TABLE 5.

subsequent firing by professional history, 1988-2013

207

205. p < ..

206. Ocers who had never been ﬁred for misconduct and who secured a new position were sub-

sequently ﬁred  of the time and ﬁred for misconduct just . of the time. In contrast,

ocers who were ﬁred for misconduct from their most recent position were ﬁred  of the

time and ﬁred for misconduct . of the time—more than twice as oen as ocers who had

never been ﬁred (p < .). Ocers who voluntarily separated from their last position but

had been ﬁred for misconduct at some point earlier in their career were subsequently ﬁred

 of the time and ﬁred for misconduct . of the time—roughly  more oen than

never-ﬁred ocers (p < .). Once again, the same pattern emerges when we examine ﬁr-

ings within three-year windows (and one- and ﬁve-year windows in Table A).

207. As noted earlier, although our focus is on law-enforcement ocers, our variables measuring

whether an ocer was previously ﬁred include any past ﬁrings from a position in corrections

as well. A very small number of law-enforcement rookies—ocers who are working for the

ﬁrst time in a law-enforcement position—were previously ﬁred from corrections positions.

We count these ocers as rookies when we estimate ﬁring rates in Table , but excluding them

has very little eect on the result.

Fired

Fired for

Misconduct

n Ever

Years

Ever

Years

Panel A:

Firings

Never 29,888 8.7% 4.5% 6.6% 3.0%

Fired, last job 1,969 18.4% 11.2% 13.8% 7.9%

Fired, earlier job 1,631 14.7% 7.5% 11.1% 5.6%

Panel B:

Firings for

Misconduct

Never 31,182 9.0% 4.7% 6.7% 3.1%

Fired, last job 1,297 20.0% 12.2% 15.3% 9.0%

Fired, earlier job 1,009 15.0% 8.1% 11.2% 6.2%

Panel C:

Rookie

54,476 10.5% 5.8% 7.2% 3.2%

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



Earlier, we found that ocers who have been ﬁred are far more likely to se-

cure a new job if they voluntarily separate from at least one intervening position.

We hypothesized that hiring agencies might view these ocers as rehabilitated.

The results just reported, however, imply a story of only partial redemption: of-

ﬁcers who were ﬁred from a job further back in their career are less likely to be

ﬁred than ocers who were ﬁred from their most recent position. Perhaps this

is unsurprising, as these ocers have been double-screened: two dierent agen-

cies have made the decision to hire them since they were ﬁred. Still, these ocers

are substantially more likely to be ﬁred than ocers who have never been ﬁred

before.

Of course, when hiring, law-enforcement agencies do not simply choose

among veteran ocers. They might also decide to hire a rookie who has never

had a full-time job in law enforcement before. Ocers hired as rookies therefore

oer another potentially helpful performance benchmark. As Table  reports in

Panel C, rookies are ﬁred . of the time and ﬁred for misconduct . of the

time. This makes them substantially less risky than wandering ocers—both

those who were ﬁred from their most recent job and those who were ﬁred from

an earlier one.

208

Ocers hired as rookies tend to be slightly riskier than veterans

who have never been ﬁred, but the dierences are not consistently statistically

signiﬁcant across our speciﬁcations.

So far, we have compared the ﬁring rates of wandering ocers to those of all

veterans who have never been ﬁred and all ocers hired as rookies. Yet

employment markets may vary across both space and time. To account for such

potential local variations, we next narrow our analysis both geographically and

temporally. The idea is to capture, as best we can, the subset of ocers who may

have applied for each job, or at least to approximate the type and quality of

ocers who were likely to have been in the candidate pool.

Our approach is to match each wandering ocer with all nonwandering

ocers who were hired within  miles and up to  days aer the wandering

ocer’s hiring date.

209

The results are similar when we change the time window

to  days or  year and the geographic limit to  or  miles. We exclude all

208. All but one of the dierences between ocers hired as rookies and wandering ocers are

statistically signiﬁcant. The sole exception is when we compare ﬁrings within one year for

ocers hired as rookies and ocers who had been ﬁred prior to their most recent job.

209. We do not allow ocers from the same agency to serve as comparators for each other. Each

wandering ocer is matched, on average, to roughly one hundred comparators. Because some

wandering ocers are matched with more comparator ocers than others, we weight each

comparator ocer by the inverse of the number of comparator ocers assigned to the same

wandering ocer. In other words, for a wandering ocer with twenty-ﬁve comparators, we

weight each of the comparators /.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



observations for the  law-enforcement agencies for which we were unable to

obtain geographic coordinates.

210

As a result, the data set we use for this analysis

diers somewhat from the data set used previously.

In Table , Panel A presents ﬁring rates based on professional history. As a

basis for comparison, the ﬁrst row reports ﬁring rates for all ocers who had

never previously been ﬁred. Overall, these ocers were ﬁred  of the time.

The next row reports ﬁring rates for wandering ocers who had been ﬁred from

their immediately preceding job. Similar to our results in Table , these ocers

ended up being ﬁred . of the time—almost  more oen. More

important, the following row reports the ﬁring rate for the never-ﬁred ocers

who were chosen as matched comparators based on location and timing. These

ocers were ﬁred . of the time—slightly more than all ocers who had

never been ﬁred, but still far less than the wandering ocers.

211

Matching

comparator ocers to wandering ocers based on geography and timing thus

appears to have slightly increased the ﬁring rate of the comparison group, but

the comparator ocers are still ﬁred at rates much lower than the wandering

ocers are.

The remaining rows in Panel A conduct the same analysis for wandering

ocers who were ﬁred from a job further back in their employment history. The

results are substantively similar.

212

In the remaining columns of Panel A, the

same basic pattern also appears for ﬁrings within a three-year window and

subsequent ﬁrings for misconduct. Panel B shows similar results when we deﬁne

wandering ocers as ocers who had been ﬁred for misconduct, speciﬁcally.

210. The vast majority of these agencies are state-level agencies for which assigning a speciﬁc lo-

cation is not straightforward.

211. The statistical signiﬁcance tests we report here are two-sided t-tests based on cluster-robust

standard errors clustered at the person level. The dierence in ﬁring rate between these wan-

dering ocers and matched comparators is statistically signiﬁcant at the . level.

212. The dierence in ﬁring rate between these wandering ocers and matched comparators is

again statistically signiﬁcant at the . level.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE 6.

subsequent firing by professional history for matched comparators based

on timing and geography, 1988-2013

Fired Fired for Misconduct

Ever 3 Years Ever 3 Years

Panel A:

Firings

Never

10.0% 5.4% 7.1% 3.3%

Fired,

last job

Wanderer 18.6% 11.1% 14.0% 8.0%

Comparator 10.8% 6.0% 7.9% 3.8%

Fired,

earlier job

Wanderer 14.6% 7.4% 11.1% 5.7%

Comparator 10.2% 5.6% 7.1% 3.3%

Panel B:

Firings for

Misconduct

Never

10.1% 5.5% 7.2% 3.3%

Fired,

last job

Wanderer 20.2% 12.1% 15.7% 9.2%

Comparator 10.9% 6.0% 8.1% 3.9%

Fired,

earlier job

Wanderer 15.3% 8.3% 11.5% 6.5%

Comparator 10.5% 5.7% 7.4% 3.5%

We can narrow our comparator pool even further if we assume that

applicants similarly evaluate the desirability of working at speciﬁc agencies and

that agencies similarly evaluate the desirability of job applicants. If so, at least

within local labor markets, more desirable candidates are likely to be hired by

more desirable agencies.

213

Under this logic, we can further narrow our analysis

to the pool of candidates who were likely vying for the same job by matching

wandering ocers not only based on timing and geography but also based on

whether they were hired by a similarly desirable agency.

Designing an objective measure of agency desirability poses two challenges.

The ﬁrst is that we need to know what agency characteristics ocers value.

Unfortunately, there is little relevant empirical evidence in the policing literature.

Moreover, dierent ocers may value agency characteristics dierently. As just

one example, some ocers may prefer to work at agencies with high levels of

213. See David Card et al., Workplace Heterogeneity and the Rise of West German Wage Inequality, 

Q.J.

ECON. , - () (ﬁnding that more educated workers in Germany tend to be

hired by higher-paying employers and that this correlation increased over time from the s

to the late s).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



crime—to be close to the action—while others may prefer agencies in safer areas.

The second challenge is that we need observable measures of agency

characteristics—measures that ideally vary over time, given the longitudinal

nature of our study.

Given these constraints, we proxy for agency desirability using annual

expenditures per ocer. The assumption is that job candidates prefer to work at

agencies with more money to spend on each ocer—for salaries, beneﬁts, perks,

and other organizational resources.

214

Because the expenditure data start in

, we drop all employment stints that began before that year. We then use

agency expenditure data to compute, within each year,

215

the percentile rank of

each agency in terms of expenditures per ocer. We group the agencies into

quintiles—with the bottom twenty percent in the ﬁrst quintile, the next twenty

percent in the second quintile, and so on, until the ﬁh quintile, which includes

the top twenty percent in each year. We then use these quintile scores to select

comparator ocers who were hired within ninety days and ﬁy miles of the

wandering ocer by an agency within the same desirability quintile.

216

As we

show in Table A, our basic results do not change when we match on agency

desirability.

This geographic, temporal, and agency-desirability matching process helps

us get closer to capturing a picture of the “marginal ocer”—the ocer the

agency would have hired had it not hired the wandering ocer. Ideally, we

would compare wandering ocers and marginal ocers more directly. Without

data on job applicants,

217

however, it is impossible to identify the actual marginal

ocers, and we cannot probe this question any more closely.

The foregoing analysis has revealed consistent evidence that wandering

ocers are ﬁred at signiﬁcantly higher rates than both veterans who have never

214. To obtain a comprehensive measure of the number of ocers employed by an agency, we

count both law-enforcement ocers and correctional ocers.

215. Because some agencies’ expenditures vary substantially from year to year, we also calculated,

for each agency, average expenditures per ocer over the entire period,  to . We then

reran our analysis using this time-invariant measure of agency desirability. The results are

similar.

216. Because matching on desirability quintiles substantially reduces the pool of potential compar-

ator ocers,  out of the , wandering ocers who were hired from  to  who

had previously been ﬁred were not matched with any comparators. Also,  of the , of-

ﬁcers hired during this period who had previously been ﬁred for misconduct did not receive

a match. We drop these unmatched wandering ocers from this analysis. The remaining

wandering ocers were matched with an average of about twenty-one comparators.

217. We did not try to collect applicant data because it would require obtaining sensitive personnel

records from each of the hundreds of law-enforcement agencies in Florida, likely an impossi-

ble task.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



been ﬁred and rookies. One lingering question is whether this relationship—

between wandering-ocer status and ﬁring rates—is mediated by other

observable ocer characteristics. Might it be, for example, that wandering

ocers tend to be young and that youth predicts a higher ﬁring rate? If so, what

appears to be a “wandering-ocer eect” might really be a “youth eect.”

To explore this possibility, Table  reports a series of linear probability

models, at the employment-stint level, on our variable measuring ﬁrings within

three years.

218

Model  essentially replicates the results in Table  because it

contains only the professional-history variables: whether an ocer was ﬁred

from his last job, was ﬁred from a job further back in his employment history,

or was hired as a rookie. Veterans who have never been ﬁred are the comparison

group. The model conﬁrms that ocers who were ﬁred from their last job and

ocers ﬁred further back in their employment history are subsequently ﬁred .

and . percentage points more oen, respectively, than ocers who have never

been ﬁred.

In Model , we add some ocer-level demographic variables: age and

gender.

219

The addition of these variables has little eect on the coecients for

either type of wandering ocer, meaning that the predictive power of being a

wandering ocer is not merely driven by these demographic characteristics.

Model  adds educational attainment to the model, but this step is,

unfortunately, more complicated because the variable is frequently missing. The

coecient for ocers who were ﬁred from their last job falls slightly, while the

coecient for ocers who were ﬁred from an earlier job falls dramatically, by

roughly two-thirds. It is possible, however, that dropped observations due to

missing data, rather than the predictive power of educational attainment, are

driving these dierences. To test this possibility, in Model  we replicate Model

 but drop all observations for which we lack education data. The results are

similar to Model . This shows that the change in coecients we observed in

Model  is likely due to the loss of observations from missing data rather than to

the eect of the education variables.

220

In the end, therefore, we ﬁnd little

evidence that observable ocer-level variables mediate the heightened ﬁring rate

that wandering ocers experience.

218. We report the results of linear probability models (or linear regressions) because the magni-

tudes of their coecients are easy to interpret. The results are substantively similar, however,

when we use logistic regression, which better ﬁts the binary structure of the dependent vari-

able. The results are also substantively similar when we ﬁt the model on our variable measur-

ing ﬁrings for misconduct within three years.

219. We divide age by ten so that the coecients are not rounded to zero.

220. The fact that the results change dramatically thus suggests that the education variable is not

missing randomly.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE 7.

regression models on firing within three years, 1988-2013

221

Model 1 Model 2 Model 3 Model 4

Intercept

0.045**

[0.002]

0.053**

[0.006]

0.044**

[0.007]

0.025**

[0.005]

Fired from

Last Job

0.067**

[0.007]

0.067**

[0.007]

0.063**

[0.010]

0.064**

[0.010]

Fired from

Earlier Job

0.029**

[0.007]

0.030**

[0.007]

0.011

[0.009]

0.011

[0.009]

Rookie

0.012**

[0.002]

0.011**

[0.002]

0.010**

[0.002]

0.011**

[0.002]

Age

-0.002

[0.001]

0.004**

[0.001]

0.004**

[0.001]

Male

-0.002

[0.003]

-0.007*

[0.003]

-0.005

[0.003]

Associate’s

-0.009*

[0.005]

Education Bachelor’s

-0.024**

[0.004]

Master’s

-0.034**

[0.005]

n 87,964 87,948 46,974 46,974

b. Complaints

As an alternative measure of ocer performance, we also consider com-

plaints ﬁled with the state licensing board—the CJSTC—alleging “moral char-

acter violations” as deﬁned by Florida law. As detailed above,

222

these complaints

typically begin as civilian or internal aairs allegations investigated by a local

221. An asterisk (*) denotes an estimate that is statistically signiﬁcant at the . level; two aster-

isks (**) denote an estimate that is statistically signiﬁcant at the . level. While our thresh-

old of statistical signiﬁcance throughout the paper is ., we also note estimates that are

statistically signiﬁcant at the . level with a dagger (†). Cluster-robust standard errors clus-

tered at the person- and agency-level are reported in brackets.

222. See supra Section III.A.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



agency. If the local agency sustains the allegation and the oense implicates the

ocer’s “moral character,” the agency must submit its ﬁndings to the FDLE,

which opens a “complaint” and begins an independent disciplinary process. The

FDLE also has the power to initiate complaints on its own, which are included

in the data set. Once again, our results suggest that wandering ocers may pose

signiﬁcant risks.

Table  shows the rate at which ocers received complaints conditional on

their professional history. Because the FDLE appears to have begun consistently

recording complaints in , our analyses of complaints include only employ-

ment stints beginning between  and . As in our analysis of ﬁrings, we

exclude job stints in which an ocer is employed by an agency that had ﬁred the

ocer in his immediately preceding job. Panel A shows that ocers who had

never been ﬁred and who secured a new position received an average of .

complaints during their next job. In contrast, ocers who were ﬁred from their

last job received an average of . complaints, almost ninety percent more, a

dierence that is statistically signiﬁcant.

223

Ocers who voluntarily separated

from their previous job but had been ﬁred at some point earlier in their career

received an average of . complaints, which is also statistically signiﬁcantly

more than ocers who had never been ﬁred.

224

223. p < ..

224. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE 8.

number of complaints by professional history and type, 1993-2013

All

Violent/

Sexual

Integrity

n Ever

Years

Ever

Years

Ever

Years

Panel A:

Firings

Never 24,711 0.07 0.02 0.02 0.01 0.02 0.01

Fired,

last job

1,394 0.13 0.07 0.04 0.02 0.06 0.03

Fired,

earlier jobs

1,295 0.12 0.05 0.03 0.01 0.05 0.02

Panel B:

Firings for

Misconduct

Never 25,686 0.07 0.02 0.02 0.01 0.02 0.01

Fired,

last job

934 0.16 0.08 0.04 0.03 0.07 0.03

Fired,

earlier jobs

780 0.14 0.05 0.03 0.01 0.06 0.02

Panel C:

Rookie

44,584 0.08 0.03 0.02 0.01 0.03 0.01

One question is whether the additional complaints that wandering ocers

receive concern the types of misconduct the public ﬁnds most troubling. While

we are constrained by the relatively small number of complaints in the data, we

are able to break complaints into broad categories based on the character of the

misconduct alleged. Almost a quarter of the complaints are for violent or sexual

conduct (including implied violence), the most common allegations of which are

“excess force,” “assault,” “battery – domestic violence,” and “sex oense.”

225

An-

other third are integrity-related complaints, the most common allegations of

which are “false statements,” “perjury,” “misuse of public position,” and

225. In our primary speciﬁcation, we exclude from this category complaints concerning prostitu-

tion, sex on duty, intimidation, harassing communication, sexual harassment, resisting an of-

ﬁcer, unprofessional relationships, and the manufacture, possession, or transportation of ob-

scene materials. The results, however, are substantively similar when we include these

complaints. Details of the coding scheme are available from the authors upon request.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



“fraud.”

226

We also create a category for drug-related allegations, which account

for just  of the complaints. This is too few to support any reliable results,

but also indicates that our general results, reported above, are not driven by

drug-related oenses.

As Table  shows, ocers who had never been ﬁred and who secured a new

position received an average of . complaints for violent or sexual conduct

during their next job. In contrast, ocers who were ﬁred from their last job re-

ceived an average of ., roughly twice as many, a dierence that is statistically

signiﬁcant.

227

Ocers who voluntarily separated from their previous job but had

been ﬁred at some point earlier in their career received an average of . com-

plaints for violent or sexual conduct, which is statistically signiﬁcantly dierent

from ocers who had never been ﬁred.

228

We ﬁnd a similar pattern of results for

integrity-related complaints.

Because employment stints vary in length, Table  also reports the average

number of complaints against ocers within a three-year window, and the re-

sults are similar (as are the results within one- and ﬁve-year windows, reported

in Table A).

229

Furthermore, as Panel B of Table  shows, all of these same pat-

terns hold if we examine ocers who have been ﬁred for misconduct instead.

Here, too, ocers hired as rookies provide another useful performance

benchmark for wandering ocers. As shown in Panel C of Table , ocers hired

as rookies receive a roughly similar number of complaints as veterans who have

never been ﬁred

230

—and fewer than wandering ocers.

231

As before, we next match each wandering ocer with the nonwandering

ocers who were hired within  miles and fewer than  days aer the

226. We exclude traditional the oenses and oenses involving stolen property, which do not

necessarily implicate truthfulness or the abuse of an ocial position.

227. p < ..

228. p < ..

229. The one exception is that, for violent and sexual complaints, the dierence between ocers

who have never been ﬁred and ocers who voluntarily separated from their last job but who

were ﬁred earlier in their career is not statistically signiﬁcant within a one-, three-, or ﬁve-

year window.

230. When we count all complaints, all violent or sexual complaints, or all integrity-related com-

plaints incurred during the employment stint, ocers hired as rookies receive statistically sig-

niﬁcantly more complaints than veterans who have never been ﬁred, at a threshold of p < .,

though the dierences are substantively small. In the time-limited comparisons, most of the

dierences are not statistically signiﬁcant.

231. The dierences are always statistically signiﬁcantly dierent for ocers who were ﬁred from

their last job (p < .). They are also statistically signiﬁcant for ocers who were ﬁred fur-

ther back in their employment history when we count all complaints and integrity-related

complaints (p < .), but not when we count only violent or sexual complaints.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



wandering ocer’s hiring date.

232

The results, reported in Table , follow the

now-familiar pattern: wandering ocers receive, on average, . complaints—

about  more than all ocers with no prior history of ﬁring (. complaints)

and the comparator ocers (. complaints).

233

The same pattern of results

holds for wandering ocers who were ﬁred from a job earlier in their

employment history, for complaints received within a three-year window, and

for wandering ocers who were ﬁred for misconduct, speciﬁcally. The same

patterns also hold when we count only violent or sexual complaints or integrity-

related complaints.

234

And, as reported in Table A, the same basic pattern holds

when we match comparator ocers not only on timing and geography but also

on agency desirability.

TABLE 9.

number of complaints by professional history for matched comparators

based on timing and geography, 1993-2013

All Violent/Sexual Integrity

Ever

Years

Ever

Years

Ever

Years

Panel A:

Firings

Never

0.08 0.03 0.02 0.01 0.03 0.01

Fired,

last job

Wanderer 0.14 0.07 0.04 0.02 0.06 0.03

Comparator 0.08 0.03 0.02 0.01 0.03 0.01

Fired,

earlier job

Wanderer 0.12 0.05 0.03 0.01 0.06 0.02

Comparator 0.07 0.03 0.02 0.01 0.03 0.01

Panel B: Firings

for Misconduct

Never

0.08 0.03 0.02 0.01 0.03 0.01

Fired,

last job

Wanderer 0.16 0.08 0.05 0.03 0.07 0.03

Comparator 0.08 0.03 0.02 0.01 0.03 0.01

Fired,

earlier job

Wanderer 0.14 0.06 0.03 0.01 0.07 0.02

Comparator 0.07 0.03 0.02 0.01 0.03 0.01

232. The results are similar when we expand the time window to  days or one year and when

we expand the geographic limit to  or  miles.

233. The statistical signiﬁcance tests we report here are two-sided t-tests based on cluster-robust

standard errors clustered at the person level. The dierence in complaints received by these

wandering ocers and their matched comparators is statistically signiﬁcant at the . level.

234. The dierences are not always statistically signiﬁcant, however, for violent and sexual com-

plaints within a three-year window.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



Finally, as with our ﬁring measures, we consider whether the higher number

of complaints against wandering ocers is mediated by other observable ocer

characteristics. Table  reports a series of linear-regression models that mirror

those described in Table .

235

The basic results are the same: we ﬁnd no evidence

that adding ocer age, gender, or education to the models decreases the size of

the coecient on the professional-history variables. Dierences in coecients

among the models appear to be due to the loss of observations from missing data

rather than to the eect of the independent variables. We therefore ﬁnd no

evidence that observable ocer-level characteristics mediate the heightened risk

associated with hiring wandering ocers.

TABLE 10.

regression models on number of complaints within three years, 1993-2013

236

Model 1 Model 2 Model 3 Model 4

Intercept

0.024**

[0.001]

0.033**

[0.003]

0.025**

[0.004]

0.019**

[0.004]

Fired from

Last Job

0.044**

[0.008]

0.043**

[0.008]

0.038**

[0.010]

0.039**

[0.010]

Fired from

Earlier Job

0.023**

[0.007]

0.024**

[0.006]

0.01

[0.009]

0.01

[0.009]

Rookie

0.002

[0.001]

-0.001

[0.002]

0.001

[0.002]

Age

-0.005**

[0.001]

-0.001

[0.001]

-0.001†

[0.001]

Male

0.007**

[0.002]

0.004†

[0.002]

0.005*

[0.002]

235. We report the results of linear models because the magnitudes of their coecients are easy to

interpret. The results are substantively similar, however, when we use negative binomial re-

gression.

236. An asterisk (*) denotes an estimate that is statistically signiﬁcant at the . level; two aster-

isks (**) denote an estimate that is statistically signiﬁcant at the . level. While our thresh-

old of statistical signiﬁcance throughout the paper is ., we also note estimates here that are

statistically signiﬁcant at the . level with a dagger (†). Cluster-robust standard errors clus-

tered at the person- and agency-level are reported in brackets.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Education

Associate’s

-0.001

[0.003]

Bachelor’s

-0.008**

[0.002]

Master’s

-0.011**

[0.003]

71,984 71,971 40,436 40,436

c. Explanations

Taken together, our results suggest that wandering ocers are signiﬁcantly

more likely than comparable ocers—either ocers hired as rookies or veterans

who have never been ﬁred from a Florida law-enforcement agency—to be ﬁred

and to incur complaints of serious misconduct. What explains these ﬁndings? At

least four hypotheses strike us as plausible.

First, agencies may scrutinize wandering ocers more rigorously than other

ocers by monitoring them more closely or applying a lower threshold for ini-

tiating the disciplinary process. To put the point most forcefully, agencies might

even hire wandering ocers on a de facto “probationary” basis, intending simply

to ﬁre them if problems arise. Although this is possible, it strikes us as unlikely

that it could fully explain the sizable gaps in the rates at which ocers are ﬁred

and incur complaints.

As an initial matter, employee turnover, or “churn,” is typically expensive: a

recent review of the labor economics literature found that the median cost of

replacing a worker is roughly one-ﬁh of the worker’s salary.

237

This makes less

plausible the notion that agencies would hire wandering ocers on a probation-

ary rationale, particularly as the agencies that hire wandering ocers tend to be

more poorly resourced. At the same time, the literature emphasizes that the costs

of turnover vary by industry, region, and other factors, and they may be lower

for Florida law enforcement than this general evidence suggests.

238

237. See Boushey & Glynn, supra note , at . We have not found any research on the cost of

turnover in law enforcement speciﬁcally. For potential analogs, see G

ARY BARNES ET AL., THE

COST OF TEACHER TURNOVER IN FIVE SCHOOL DISTRICTS: A PILOT STUDY - () (estimat-

ing the cost of teacher turnover in ﬁve school districts in  at , to ,); and

Michelle I. Graef & Erick L. Hill, Costing Child Protective Services Sta Turnover,  C

HILD

WELFARE ,  () (estimating the cost of turnover for a child protective services

worker in a midwestern state in  at ,).

238. Where labor requires “knowledge exploitation” (i.e., implementation, execution) rather than

“knowledge exploration” (i.e., discovery, innovation), however—as seems generally true of

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



To probe this ﬁrst hypothesis more closely, we consider a series of additional

empirical checks. None of these checks is conclusive on its own but, collectively,

they suggest that heightened scrutiny cannot fully explain our results. To begin

with, we found earlier that wandering ocers receive not only more complaints

in general, but also more complaints speciﬁcally about violent or sexual miscon-

duct. While certainly possible, it seems less likely that heightened scrutiny could

drive those results, as agencies presumably take violent and sexual conduct more

seriously than other misconduct regardless of which ocers are accused.

Our complaint data oer another way to test the heightened-scrutiny hy-

pothesis. The FDLE, a statewide entity independent from the local agencies that

employ law-enforcement ocers, itself initiates roughly one-third of the com-

plaints in our data set. These include complaints opened in response to an of-

ﬁcer’s arrest, a news report, a veriﬁed citizen complaint tendered to the FDLE

directly, or a problem FDLE sta discovered while auditing local-agency rec-

ords.

239

If heightened scrutiny by employing agencies explained why wandering

ocers receive more complaints, we would expect the gap in complaint rates to

disappear in this subset of the data (because, again, the employing agencies do

not trigger these complaints). The data does not bear this out. We reran our

analysis of complaints on the FDLE-initiated subset. As Table A shows, the

results are substantively similar to when we use all complaints: across each of the

speciﬁcations, wandering ocers were roughly twice as likely as ocers who had

never been ﬁred to receive a complaint initiated by the FDLE.

240

Next, we examine ﬁring and complaint rates across agencies of varying size.

What little has been written on the topic suggests that smaller agencies are less

policing—the average net eects of turnover are expected to be negative. See Ton & Huckman,

supra note , at -, .

239. The ATMS database contains seven codes that deﬁne the source of a complaint: adavit of

separation, arrest hit notiﬁcation, FDLE sta documentation, internal investigation, newspa-

per, other, and veriﬁable complaint. Based on conversations with the FDLE, we deﬁne a com-

plaint as initiated by the FDLE (and not the ocer’s agency) if its code is anything other than

adavit of separation or internal investigation. As noted, see supra Section III.A, we have data

only on the date a complaint was opened and not the date on which the alleged misconduct

occurred. It is possible that, in some cases, there is a nontrivial gap between the date a com-

plaint was opened and the date of the alleged misconduct.

240. The dierence between never-ﬁred ocers and ocers who were ﬁred from their last job is

statistically signiﬁcant across all comparisons at p < .. The dierence between never-ﬁred

ocers and ocers who were ﬁred further back in their employment history is signiﬁcant

across all comparisons at p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



likely than their larger counterparts to have the resources and organizational ap-

paratus to closely monitor and remediate particular ocers.

241

Thus, if wander-

ing ocers are ﬁred more frequently or receive more complaints than other of-

ﬁcers because they face heightened scrutiny, we would expect these gaps to

narrow or disappear in smaller agencies.

242

We therefore reran our analyses of

ﬁring and complaints aer breaking the data into two groups of agencies, by

size: agencies that employed fewer than sixty-two ocers—the smallest quartile

of employment stints—and agencies that employed sixty-two ocers or more.

We then reran the analysis again using even smaller thresholds: thirty-one oc-

ers and ﬁeen ocers. Across all of these comparisons, our results were substan-

tively similar.

243

In both large and small agencies, ocers who were ﬁred from

their last job were roughly twice as likely to be ﬁred,

244

be ﬁred for miscon-

duct,

245

or receive a complaint

246

than ocers who had never been ﬁred before.

241. See, e.g., STEVEN G. BRANDL, POLICE IN AMERICA  () (“[L]arger police departments tend

to have more rules and policies than smaller ones.”); P

RESIDENT’S TASK FORCE ON ST CEN-

TURY

POLICING, supra note , at - (reporting that “small forces oen lack the resources

for training and equipment accessible to larger departments” and encouraging consolidation);

David N. Falcone et al., The Small-Town Police Department,  P

OLICING: INT’L J. POLICE STRAT-

EGIES

& MGMT. ,  () (“Given the low number of [full-time employees] for small-

town police departments, and the near absence of organizational dierentiation, all ocers,

regardless of rank, must carry out general patrol functions.”);

Kevin Johnson, Lack of Training,

Standards Mean Big Problems for Small Police Departments, USA

TODAY (June , , :

PM), https://www.usatoday.com/story/news/nation////small-police-departments

-standards-training/ [https://perma.cc/YFL-LEJG] (“[Q]uestions about leader-

ship, training and basic competence track an array of unmet public safety needs that threaten

small-town policing operations in communities across the country.”); cf. Casey Toner & Jared

Rutecki, 113 Suburban Cop Shootings, Zero Discipline, WBEZ (Jan. , ), https://interac-

tive.wbez.org/taking-cover/zero-discipline [https://perma.cc/CV-WLTT] (“The investi-

gation found little evidence that suburban police agencies do any self-reﬂection or post-mor-

tem reviews aimed at retraining ocers involved in deadly shootings, a practice oen

employed at larger departments.”).

242. Alternatively, it is possible that police chiefs in smaller agencies are better able to monitor

wandering ocers closely. Our results run contrary to this prediction, too.

243. See infra Table A. To reduce the size of the table, we present only the results for agencies that

are bigger and smaller than ﬁeen ocers at the time the ocer was hired and for ﬁrings (but

not ﬁrings for misconduct). The results are substantively similar when we break up agencies

based on the thirty-one-ocer and sixty-two-ocer thresholds and when we examine ﬁrings

for misconduct.

244. p < ..

245. p < ..

246. p < ..

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



While ocers who were ﬁred further back in their employment history also ex-

perienced heightened rates of ﬁring and complaints—both in small and large

agencies—the dierences were only sometimes statistically signiﬁcant.

Finally, we probe the heightened-scrutiny hypothesis by examining the tim-

ing of ﬁrings and complaints. If wandering ocers are, like probationary em-

ployees, scrutinized more closely than others, this heightened monitoring is

probably not indeﬁnite. Indeed, monitoring can be costly and its expected re-

turns may diminish with time. It seems likely, then, that wandering ocers

blend in with the rest of the workforce at some point. We therefore calculate the

average rate at which ocers are ﬁred and receive complaints during their fourth

through seventh years of employment.

247

As Table A shows, wandering oc-

ers continue to experience higher rates of ﬁrings

248

and complaints

249

even dur-

ing those later years, though the gap between wandering and nonwandering of-

ﬁcers is somewhat smaller during this period than during years one through

three.

A second potential explanation for the elevated rates of ﬁring and complaints

that wandering ocers receive is that, even if wandering ocers are not treated

dierently within agencies, they may, on average, take jobs at agencies that are

better at detecting misconduct or more likely to initiate the disciplinary process.

We are doubtful this hypothesis is correct. If anything, agencies that are better

at detecting misconduct or more likely to initiate the disciplinary process are,

because of their high quality, probably less likely to hire wandering ocers in the

ﬁrst place. While we cannot directly observe the stringency of an agency’s disci-

plinary process, our composite measures of agencies’ hiring and training re-

quirements provide useful proxies. In our data, ocers who were ﬁred from

their last job and land a new position tend to move to agencies with less stringent

247. We also examine the data in one-year increments. The basic results are substantively similar

but noisy due to the limited number of ﬁrings and complaints within each period. In calcu-

lating the ﬁring and complaint rates in years four through seven, we count in the denominator

all ocers who had le the agency in years one through three. If instead we drop these ocers

from the analysis, the dierence in ﬁring rates between wandering and nonwandering ocers

is even larger.

248. The ﬁring rate for ocers ﬁred from their last job is statistically signiﬁcantly dierent from

that of ocers who were never ﬁred at a threshold of .. For ocers ﬁred earlier in their

employment history, the dierence is statistically signiﬁcant at the . level for all compari-

sons except when we measure both professional history and subsequent ﬁrings using ﬁrings

for misconduct.

249. The number of complaints for ocers ﬁred from their last job is only marginally statistically

signiﬁcantly dierent from the number of complaints for ocers who have never been ﬁred,

at the . level. For ocers ﬁred further back in their employment history, the dierence is

statistically signiﬁcant at the . level.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



hiring and training requirements than ocers who have never been ﬁred.

250

This

suggests that, as we expected, wandering ocers are not moving to agencies

with more stringent disciplinary processes than other agencies.

Third, wandering ocers may take jobs at agencies where, due to dierences

in culture, leadership, or beat assignment, they are exposed to conditions con-

ducive to further misconduct. Although we did not ﬁnd any evidence that wan-

dering ocers move to agencies in higher-crime jurisdictions, many move to

smaller agencies with relatively fewer resources. For this reason, we think this

third hypothesis warrants further investigation.

We test the hypothesis in two ways.

251

First, we assess whether observable

agency-level variables reduce the predictive power of ocers’ professional-his-

tory variables. Table  presents a series of linear probability models where the

dependent variable is whether an ocer was ﬁred within three years.

252

Because

some of our agency-level variables begin in , we restrict our analysis to em-

ployment stints that began between  and . Model  contains only the

professional-history variables and shows that, during that period, ocers who

were ﬁred from their last job and ocers ﬁred further back in their employment

history were subsequently ﬁred . and . percentage points more oen, re-

spectively, than veterans who had never been ﬁred, the comparison group.

In Model , we add a number of agency-level variables (as of the date the

ocer’s employment began): total number of ocers employed by the agency,

agency expenditures per full-time ocer, violent crimes per , residents,

250. Our hiring and training scores begin in . We are also missing a substantial number of

observations within our sample. As a result, we lack hiring-requirement and training-require-

ment data for  and  of observations from  to , respectively. Most of these

missing cases— and  of the missing hiring and training observations—concern moves

from or to agencies that are neither municipal police departments nor sheris’ oces. Ocers

who have never been ﬁred, who separated from their previous job voluntarily, and who land

a new job move from agencies that, on average, have hiring and training scores of . and .

to agencies with scores of . and .—almost no change at all. Similarly, ocers who were

ﬁred earlier in their employment history but voluntarily separated from their last job and

found new work moved from agencies with hiring and training scores of . and . to agen-

cies with scores of . and .. In contrast, ocers who were ﬁred from their last job and

obtain new work tend to move to agencies with lower hiring and training scores; indeed, they

move from agencies with average scores of . and . to agencies with scores of . and .

(p < .).

251. As in the rest of our analyses of professional history and ﬁrings, we exclude job stints in which

an ocer is employed by an agency that had ﬁred him in his immediately preceding job.

252. As before, we report the results of linear probability models (or linear regressions) because

the magnitudes of their coecients are easy to interpret. The results are substantively similar,

however, when we use logistic regression, which better ﬁts the binary structure of the depend-

ent variable.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



property crimes per , residents, unemployment, proportion of the pop-

ulation that is black, and proportion of the population that is Hispanic. Adding

these variables decreases the coecient both for ocers who were ﬁred from

their last job and for ocers who were ﬁred further back in their employment

history by  to . Because we are missing data for at least one of the agency-

level variables in roughly  of employment stints from  to , it is once

again possible that missing data—rather than the introduction of the agency-

level variables—explain the change in results from Model  to Model . We assess

this possibility in Model  by replicating Model  while dropping any observa-

tions for which we are missing some agency-level data. The coecients both for

ocers ﬁred from their last job and for ocers ﬁred from an earlier job are

roughly similar to those in Model , suggesting that the addition of the agency-

level variables is responsible for much of the modest decrease in the coecient

from Model  to . Taken together, these models suggest that the agency-level

variables may reduce the predictive power of the professional-history variables

but only very slightly.

Second, we test whether unobservable agency-level characteristics reduce the

predictive power of the professional-history variables. More speciﬁcally, we in-

troduce agency ﬁxed eects to assess whether the heightened rates at which wan-

dering ocers are ﬁred stem from their moving to agencies with higher rates of

ﬁring and complaints in general. As Model  in Table  shows, adding these

ﬁxed eects has little impact on the size of the coecients for either type of wan-

dering ocer. This provides further evidence that this third hypothesis has lim-

ited explanatory power.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE 11.

regression models on firing within three years with agency characteristics,

1997-2013

253

Model 1 Model 2 Model 3 Model 4

Intercept

0.039**

[0.002]

0.052**

[0.006]

0.041**

[0.002]

Fired from Last Job

0.066**

[0.009]

0.055**

[0.009]

0.060**

[0.009]

0.056**

[0.009]

Fired from Earlier Job

0.029**

[0.008]

0.026**

[0.009]

0.029**

[0.009]

0.024**

[0.009]

Rookie

0.015**

[0.003]

0.018**

[0.002]

0.015**

[0.003]

0.016**

[0.002]

Number of Officers

(In Hundreds)

-0.002**

[0.000]

0.002

[0.002]

Expenditure Per Officer

(In Thousands)

[0.000]

Violent-Crime Rate

(Per 100 Residents)

0.005**

[0.001]

0.002

[0.003]

Property-Crime Rate

(Per 100 Residents)

-0.003†

[0.002]

County-Level

Unemployment Rate

0.024

[0.059]

-0.063

[0.058]

Proportion Black

-0.008

[0.014]

-0.053

[0.061]

Proportion Hispanic

-0.015

[0.010]

-0.042

[0.060]

Agency Fixed Effects No No No Yes

n 59,861 49,687 49,687 49,687

The fourth and ﬁnal hypothesis is the most straightforward: holding all else

constant, wandering ocers may simply behave worse than ocers who have

253. Asterisks (*) denote estimates that are statistically signiﬁcant at the . level; two asterisks

(**) denote estimates statistically signiﬁcant at the . level. While our statistical signiﬁ-

cance threshold is ., we note estimates statistically signiﬁcant at the . level with daggers

(†). Cluster-robust standard errors clustered at the person- and agency-level are in brackets.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



never been ﬁred. On top of that, they tend to be hired by smaller agencies with

fewer resources, which may be unable to provide them with the policies, train-

ing, and supervision that could help them stay in line. Because the other hypoth-

eses largely fail, and because this explanation is consistent with anecdotal ac-

counts, we suspect that this is, in the end, the most plausible explanation of the

higher rates of ﬁring and complaints against wandering ocers.

v. predicting which wandering officers get fired again

Before we turn to potential legal reforms for the problems just identiﬁed, we

examine one ﬁnal empirical question: whether certain agencies, or certain wan-

dering ocers, pose greater or lesser risks than others. More speciﬁcally, should

certain agencies be especially wary of hiring wandering ocers? Do agencies’

hiring or training requirements aect the risk that wandering ocers pose? Are

there observable characteristics that mark some wandering ocers as more or

less risky than others?

To examine these questions, we ﬁt a series of linear probability models on a

data set containing an observation for every time, from  to , an agency

hired a wandering ocer.

254

The dependent variable indicates whether the of-

ﬁcer was ﬁred within three years. We include a range of independent variables

measuring characteristics of the ocer, as well as the law-enforcement agencies

he moved from and to, in order to estimate whether any of these characteristics

is associated with a lower probability that the ocer’s separation was involun-

tary. As in the earlier analyses of ﬁrings, we exclude job stints in which an ocer

is employed by an agency that had ﬁred the ocer in his immediately preceding

job.

Table  presents the results.

255

Model  includes one independent variable:

whether the ocer was ﬁred from his most recent job. It shows that such ocers

are . percentage points more likely to be ﬁred than other wandering ocers.

Model  adds various ocer-level independent variables, including the number

of jobs worked, gender, age (at start of employment), and a dummy for prior

complaints. The only statistically signiﬁcant variable is whether the ocer was

ﬁred from his most recent job.

254. We report the results of linear probability models (or linear regressions) because the magni-

tudes of their coecients are easy to interpret. The results are substantively similar, however,

when we use logistic regression, which better ﬁts the binary structure of the dependent vari-

able.

255. We report cluster-robust standard errors clustered at the agency and ocer level.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



In Model , we add a variable reﬂecting the ocer’s educational attainment

to see whether education is associated with the likelihood that a wandering of-

ﬁcer is ﬁred. Because education data is missing for many ocers, we lose a sig-

niﬁcant proportion of our observations. We ﬁnd no statistically signiﬁcant evi-

dence that ocer education makes a dierence here, but we caution that our

estimates are noisy, at least partially due to the decreased sample size.

Finally, in Model , we examine whether any agency characteristics are cor-

related with whether a wandering ocer will be ﬁred again. We add several char-

acteristics of the hiring agency for the year the ocer was hired, including the

proportion of the agency’s jurisdiction that is black, the proportion that is His-

panic, the violent and property crime rates per , residents, county-level

unemployment rates, and agency expenditures per ocer. We also test whether

the agency’s hiring or training requirements are correlated with whether the

agency will ﬁre a wandering ocer it hires by adding two composite scores, de-

scribed in Section III.B, that measure the stringency of the hiring agency’s re-

quirements. The idea is that stricter hiring requirements might direct agencies

toward the least risky wandering ocers and that rigorous training require-

ments might keep wandering ocers on the right path. Finally, the model also

contains two variables capturing the number of full-time ocers employed by

the agencies the ocer moved to and from divided by  (to ease interpretation

of the results). Because some of these agency-level variables are available only

starting in , we exclude all employment stints that began before that year.

We also lose some additional observations due to missingness on some of the

agency-level variables.

In this model, the variable indicating whether an ocer was ﬁred from his

most recent job (rather than a job further back in his employment history) de-

creases in size and is no longer statistically signiﬁcant. More important, the only

agency-level variables that are statistically signiﬁcant relate to agency size; they

show that ocers who move from and to agencies that are smaller tend to be

ﬁred less frequently. Otherwise, we ﬁnd little evidence of any statistical associa-

tion between the agency variables and ﬁring. When we examine ﬁring rates

based on wandering ocers’ history of ﬁrings for misconduct, speciﬁcally, we

ﬁnd a similar pattern of results.

256

Because Model  is restricted to hirings from

 to , however, the number of observations is substantially reduced,

which may partly explain the null results.

256. See Table A.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE 12.

predicting whether wandering officers are fired again, 1988–2013

257

Model 1 Model 2 Model 3 Model 4

Intercept

0.074**

[0.007]

0.103**

[0.023]

0.075†

[0.042]

0.194**

[0.049]

Fired from

Last Job

0.037**

[0.010]

0.035**

[0.013]

0.043*

[0.020]

0.025

[0.020]

Job Number

0.009

[0.017]

-0.002

[0.030]

-0.008

[0.029]

0.005

[0.017]

-0.016

[0.029]

-0.003

[0.030]

0.015

[0.019]

-0.002

[0.032]

-0.024

[0.030]

Age

(At Start)

-0.001

[0.001]

Male

-0.028

[0.018]

-0.031

[0.025]

-0.091**

[0.031]

Any Past

Complaints

0.006

[0.014]

-0.007

[0.019]

-0.011

[0.018]

Education

Associate’s

0.005

[0.028]

Bachelor’s

0.014

[0.027]

Master’s/

Doctorate

-0.011

[0.032]

257. An asterisk (*) denotes an estimate that is statistically signiﬁcant at the . level; two aster-

isks (**) denote an estimate that is statistically signiﬁcant at the . level. While our thresh-

old of statistical signiﬁcance throughout the paper is ., we also note estimates that are

statistically signiﬁcant at the . level with a dagger (†). Cluster-robust standard errors clus-

tered at the person-level are reported in brackets.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Hiring

Agency

Proportion Black

-0.025

[0.050]

Proportion

Hispanic

-0.062

[0.041]

Violent-Crime

Rate

(Per 100,000)

[0.000]

Property-Crime

Rate

(Per 100,000)

[0.000]

County-Level

Unemployment

Rate

0.031

[0.313]

Expenditures Per

Officer (2008)

[0.000]

Hiring Re-

quirement Score

0.003

[0.005]

Training Re-

quirement Score

-0.007

[0.006]

No. of Officers

-0.003*

[0.001]

Separating

Agency

No. of Officers

-0.003**

[0.001]

3,533 3,533 1,390 1,451

Taken together, what should we make of these results? Assuming a law-en-

forcement agency will hire a wandering ocer, are there ways to reduce the

risks? Our models provide little optimism on this front. Most of the models sug-

gest that, among wandering ocers who are hired again, those who have sepa-

rated voluntarily from at least one job since their last ﬁring are less likely to be

ﬁred again. The models also suggest that ocers coming from or going to

smaller agencies may be slightly less likely to be ﬁred. Beyond that, however, the

regression models provide little consistent evidence about how to predict which

wandering ocers will be ﬁred again.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



vi. mechanisms and reforms

We have found that an average of over one thousand wandering ocers were

working in full-time Florida law-enforcement jobs in any given year between

 and . Many of these ocers found work by migrating to smaller, less

desirable agencies—measured by resources per ocer—than the agencies that

ﬁred them. They were then ﬁred and subjected to complaints alleging “moral

character violations” at higher rates than ocers who had never been ﬁred be-

fore, even when we restrict the latter group to ocers who were hired around

the same time and place by similarly desirable agencies. Although we cannot be

sure why wandering ocers experienced these adverse outcomes, aer a review

of several hypotheses, we think the most straightforward explanation is also the

most plausible one: wandering ocers simply behave worse than ocers who

have never been ﬁred. These results appear generally consistent with the perti-

nent labor economics literature and, in particular, with the closest study on

point, which concerns the labor mobility and behavior of ﬁnancial advisers who

commit misconduct.

258

If wandering ocers are so risky, one might reasonably ask, why do police

chiefs keep hiring them? Our data do not permit us to isolate a single mecha-

nism, and we doubt that there is only one cause in any event. In this ﬁnal Part,

we consider ﬁve plausible explanations along with the policy implications of

each. In Section A, we examine the possibility that agencies may not know they

are hiring wandering ocers. In Section B, we consider whether agencies may

not appreciate the risks that wandering ocers pose. In Section C, drawing upon

our empirical analysis in Section IV.B., we discuss whether wandering ocers

are simply the best available candidates—that is, if they are hired when no

stronger candidate is available. In Section D, we explore the beneﬁts that might

make wandering ocers attractive notwithstanding their risks. Finally, in Sec-

tion E, we discuss the possibility that agencies are not concerned about the risks

of hiring wandering ocers because, generally speaking, they do not bear the

costs of any resulting harms.

A. Poor Information

In some instances, an agency may hire a wandering ocer simply because it

does not know about the ocer’s past. Indeed, agencies do not always complete

adequate background checks, possibly due to unprofessional organizational cul-

ture or resource constraints. The Cleveland Police Department, for instance,

“never reviewed [the] personnel ﬁle” for Tim Loehmann, who shot and killed

258. See supra notes - and accompanying text.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Tamir Rice.

259

In addition, some candidates may deliberately conceal their pro-

fessional history. Much of the small academic literature on wandering ocers

emphasizes this explanation for the ability of wandering ocers to ﬁnd work.

260

The favored solution here is to build a robust national decertiﬁcation data-

base. Such a database would record state agency decisions to decertify ocers

and make them available for local agencies in other states to see. A small group

of academics has been pushing this idea for decades and, in , the President’s

Task Force on st Century Policing lent its imprimatur.

261

To be sure, a national

database does already exist—the NDI described in Part I. But its coverage is poor.

Some states “have decided it’s a drain on resources to contribute to the index

[and] the state-by-state discrepancies signiﬁcantly limit the database’s eective-

ness.”

262

As one reference point, Georgia decertiﬁed  ocers in  but sub-

mitted nothing to the database.

263

Our data and ﬁndings highlight both the importance and limits of a national

decertiﬁcation database as a tool to stop wandering ocers. On the one hand,

our ﬁnding that wandering ocers are more likely than other ocers to be ﬁred,

including for misconduct, and more likely to be subject to serious misconduct

complaints, underscores the importance of some kind of national—and manda-

tory—tracking system. Such a tool could help agencies avoid hiring wandering

ocers who saunter in from other states.

On the other hand, as the Loehmann example demonstrates—along with all

of our own ﬁndings—the focus on interstate movement seems to skip a step.

Wandering ocers frequently move around within a single state. Even before

we get to any national database, background checks should be standardized and

259. Dewan & Oppel, supra note ; see CIVIL RIGHTS DIV., U.S. DEP’T OF JUSTICE & U.S. ATTORNEY’S

OFFICE FOR THE N. DIST. OF OHIO, INVESTIGATION OF THE CLEVELAND DIVISION OF POLICE

() (detailing systemic and structural deﬁciencies in the department). On the importance

of organizational culture more generally, see Armacost, supra note .

260. See, e.g., Atherley & Hickman, supra note , at - (describing impediments to the ﬂow

of information necessary to vet candidates thoroughly); Bell, supra note , at  (explaining

that “resource constraints make it more dicult” for agencies “to discriminate between good

and bad ocers”); Goldman & Puro, supra note , at  (“[T]he second department may

be unaware of the previous misconduct, either because the ﬁrst department would not dis-

close the ocer’s previous misconduct, or because the second department does not conduct a

thorough background check.”).

261. See, e.g., Atherley & Hickman, supra note , at -; Goldman & Puro, supra note , at -

; Goldman & Puro, supra note , at -; see also P

RESIDENT’S TASK FORCE ON ST CEN-

TURY

POLICING, supra note , at .

262. Fisher, supra note .

263. Id.

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



mandatory (as they are in some states, including Florida). There is some sug-

gestion, mentioned earlier, that background checks are taxing on very small or

resource-strapped departments. If so, states or the federal government should

consider subsidizing or otherwise assisting local agencies in conducting the nec-

essary investigation of ocer candidates.

In addition, even if the database includes every decertiﬁcation decision na-

tionwide, it is useful only if states regularly decertify problem ocers. But they

do not.

264

Recall that ﬁve states plus the District of Columbia, employing a sig-

niﬁcant share of all law-enforcement ocers nationally, have no decertiﬁcation

option at all. Another twenty states require a criminal conviction before an ocer

can be decertiﬁed. And according to one study, in , over half of all police

decertiﬁcations reported to the NDI came from Florida and Georgia, while Lou-

isiana, Mississippi, and Wyoming did not decertify a single ocer

265

—in fact,

Louisiana has not decertiﬁed anyone in at least a decade.

266

Even in states that

do decertify a relatively high number of ocers, moreover, decertiﬁcation is still

rare. Only a small subset of misconduct will render an ocer eligible for decer-

tiﬁcation.

267

It is worth pointing out that, to our knowledge, none of the employ-

ment stints in our study—including those of wandering ocers—was held by an

ocer who had been decertiﬁed in Florida before the stint began.

268

The federal government—through an exercise of Congress’s spending

power—could encourage the states to decertify ocers under speciﬁed condi-

tions. Short of that, the most productive reform might be to expand the substan-

tive coverage of the national database. Rather than merely cover ocer decertiﬁ-

cations, the database ought to record at least all misconduct-related terminations,

and probably all involuntary terminations or all separations of any kind. This

would resemble the National Practitioner Databank (NPDB) that tracks medical

264. See, e.g., Puro et al., supra note , at  (“The decertiﬁcation authority is available [in most

states] but may not be implemented in many states owing to factors such as a lack of political

will by the state legislature, insucient stang at the POST, or lack of participation by local

police chiefs or sheris.”).

265. Fisher, supra note .

266. Kelly et al., supra note .

267. See Fisher, supra note  (“I think some people are under the misimpression that if a cop gets

ﬁred for anything really bad, they’re going to get decertiﬁed, and that is not the case. . . . It’s

a very narrow range of behavior that will cause them to lose their certiﬁcation.” (quoting Sue

Rahr, former police chief and member of the President’s Task Force on st Century Polic-

ing)).

268. It is possible that some had been decertiﬁed and later recertiﬁed. It is also possible that some

had been decertiﬁed in another state before moving to Florida.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



malpractice and fraud.

269

In fact, the NPDB goes further, requiring reporting of

not only employment separations but also malpractice payments and certain dis-

ciplinary actions.

270

In the s, Congress actually considered two bills that

sought to establish a federal clearinghouse of law-enforcement “employment

termination data.”

271

Both died in committee.

272

Some states have recently taken

steps in this direction for their in-state databases. A  Illinois law, for exam-

ple, requires local agencies to notify the state POST board of any “ﬁnal determi-

nation of willful violation of department or agency policy, ocial misconduct,

or violation of law” in connection with an ocer’s termination or resignation.

273

If other states enact similar reforms, an eective national database will be that

much closer to reality.

B. Unawareness of Risk

Some agencies may know they are hiring wandering ocers but may not

know that wandering ocers are, in general, risky hires.

274

Aer all, plenty of

wandering ocers do not reoend,

275

and plenty of ocers who have not been

ﬁred before commit misconduct. In fact, police administrators sometimes make

the plausible argument that wandering ocers should be more conscientious

than others because they have been reprimanded before—they know, in other

269. National Practitioner Data Bank, U.S. DEP’T HEALTH & HUM. SERVS.,

https://www.npdb.hrsa.gov/index.jsp [https://perma.cc/GCA-WM]; see Goldman &

Puro, supra note , at .

270. See Ilene N. Moore et al., Rethinking Peer Review: Detecting and Addressing Medical Malpractice

Claims Risk,  V

AND. L. REV. ,  ().

271. Law Enforcement and Correctional Ocers Employment Registration Act of , H.R.

, th Cong.  () (); Law Enforcement and Correctional Ocers Employment

Registration Act of , S. , th Cong.  () ().

272. Fisher, supra note .

273. Illinois Police Training Act,  ILL. COMP. STAT.  / .(a) (). But see Toner & Rutecki,

supra note  (reporting in  that only twenty-six ocers’ names had been submitted and

that ten of the twenty-six had gone on to ﬁnd additional police work).

274. See, e.g., Cormier & Doig, supra note  (describing a city manager who knowingly hired of-

ﬁcers who had been ﬁred for perjury, excessive force, and making false statements—but not

decertiﬁed—reasoning that, “[i]f the [certifying] commission says they’re good to go, they’re

good to go”); cf. Egan et al., supra note , at  (hypothesizing that ﬁnancial advisers with

records of misconduct may ﬁnd work because unsophisticated customers do not know “how

to interpret” their records).

275. See, e.g., Shockey-Eckles, supra note , at - (relating the story of a wandering ocer

who “salvaged his career,” was named ocer of the year, and was promoted to lieutenant in

his new agency).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



words, that they are not invincible.

276

Very few agencies, we suspect, hire enough

wandering ocers to notice that, in the aggregate, the pattern of their behavior

actually seems to cut the other way.

Police administrators are now on notice. Even when well intentioned—as a

second chance for a hard-working cop—hiring a wandering ocer is risky busi-

ness. Wandering ocers, we have shown, are ﬁred and subjected to moral-char-

acter complaints more oen than other ocers. Notably, they are riskier, by our

measures, than even ocers hired as rookies. Our ﬁndings are consistent with

prior studies, using dierent data and methods, that examined police miscon-

duct without focusing on wandering ocers speciﬁcally. White and Kane, for

example, found that New York City ocers with “red ﬂags” early in their careers

were at greatest risk of dismissal later on.

277

Machine-learning researchers who

developed an early intervention system to predict adverse events for Charlotte-

Mecklenburg ocers found that “[t]he most predictive features of the model

were those relevant to the prior [internal aairs] history of the ocer.”

278

And

Kyle Rozema and Max Schanzenbach found that Chicago ocers who receive

more civilian complaints are more likely to be sued later for civil-rights viola-

tions.

279

In light of this evidence, law-enforcement agencies should be wary of know-

ingly hiring wandering ocers and, when they do make such hires, should be

realistic about the risk they assume. Agencies should consider enhanced moni-

toring and support of wandering ocers as a potential way to manage this risk.

They might also promulgate recidivist penalties for ocers—common in the

substantive criminal law—designed to deter misconduct by this high-risk pop-

ulation. (Some may have already.) And if given the choice between two wander-

ing-ocer candidates, one of whom was just ﬁred and one of whom was ﬁred

earlier in his career, our evidence suggests that agencies ought to opt for the lat-

ter, all else equal.

C. Inadequate Alternatives

Many have assumed that hiring a wandering ocer is an obvious manage-

ment blunder. It certainly appears so aer the fact, when the public learns that

276. See Cormier & Doig, supra note .

277. Michael D. White & Robert J. Kane, Pathways to Career-Ending Police Misconduct: An Exami-

nation of Patterns, Timing, and Organizational Responses to Ocer Malfeasance in the NYPD, 

RIM. JUST. & BEHAV. ,  ().

278. Carton et al., supra note , at .

279. Rozema & Schanzenbach, supra note , at -.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



an ocer who hurt someone had previously been ﬁred by another agency. But

these cases alone, tragic as they are, actually tell us little about the wisdom of the

decision from an ex ante perspective. Even if agencies suspect that, on average,

wandering ocers carry certain risks, they might still be making optimal hiring

decisions under the circumstances. The wandering ocers who are hired might

be less risky than the available alternative candidates.

280

That wandering ocers

tend to migrate to agencies with fewer resources is evidence consistent with this

account. Certainly, if law-enforcement agencies were epistemically sophisti-

cated, proﬁt-driven private entities that internalized the costs of bad hiring de-

cisions, this story would be compelling. As we discuss in Section E below, how-

ever, there are reasons to think most agencies are not like this.

Fortunately, we were able to do more than speculate in this way. We were

able, as detailed in Part IV above, to identify, for each wandering ocer, a plau-

sible candidate cohort—a group of ocers who were hired around the same time

by nearby, and similarly desirable, agencies. This group roughly approximates

the type of ocers who may have been candidates for the job the wandering

ocer ultimately obtained. As we showed, although this cohort is riskier than

the general population of ocers who have never been ﬁred, the wandering of-

ﬁcers are riskier still. It would be preferable, of course, to compare the wandering

ocers to the actual ocers who competed for the same jobs, but that infor-

mation is unavailable. In its absence, our cohort analysis provides some evidence

that agencies are not always making the optimal hiring decision.

Nevertheless, it remains possible that a shallow applicant pool does explain

some wandering-ocer hiring. If so, and assuming additional hires are actually

necessary, the solution may be to improve the pool of candidates by raising com-

pensation or improving outreach and recruitment.

281

This, of course, is likely far

easier said than done. Still, to the extent that ocials allocating budgets have not

280. See, e.g., Atherley & Hickman, supra note , at  (explaining that small agencies sometimes

consciously hire wandering ocers when “ﬁnancial resources are limited and lateral ocers

are simply scarce”); Goldman & Puro, supra note , at  (“Although it might seem unusual

for a police department to hire an ocer with a past record of misconduct, the second depart-

ment is usually located in a poor community that cannot aord to pay high salaries to its

police.”); Cohen, supra note  (“[I]f it’s not an attractive department, you’d have . . . very

few applicants, and it’s a matter of getting the best of the worst.” (quoting Vincent del Cas-

tillo)); Kyle Hopkins, The Village Where Every Cop Has Been Convicted of Domestic Violence,

ROPUBLICA (July , , : AM), https://www.propublica.org/article/stebbins-alaska

-cops-criminal-records-domestic-violence [https://perma.cc/TC-YFY] (describing small

Alaskan communities in which all applicants for law-enforcement jobs have criminal records).

281. Cf., e.g., Johnson, supra note  (“L[i]ke the mayor, the chief is concerned about the town’s

ability to draw candidates to small-town policing ‘when you can make more at McDonalds.’”

(quoting Roger Dowell, Police Chief, Damascus, Va.)).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



perceived a problem, our ﬁndings might encourage them to reconsider. Locali-

ties that are forced to hire wandering ocers might appeal to intergovernmental

sources of funding, such as the state, for assistance in subsidizing salary en-

hancements.

Another way to improve the candidate pool may be to reduce barriers to en-

try into the profession. This could include minimizing ﬁnancial barriers, such as

the cost of the requisite training or education. Another possibility—although not

one without risks—is to relax the stringency of hiring requirements. Because re-

search ﬁnds that more educated ocers, for example, tend to commit less mis-

conduct, many assume that agencies should raise their education require-

ments.

282

But higher education requirements may screen out some otherwise-

excellent candidates—candidates who might be a superior choice to a better-ed-

ucated wandering ocer. More research is warranted on the eects of hiring re-

quirements on the makeup and performance of the police forces they yield.

D. Countervailing Beneﬁts

Some agencies may understand that wandering ocers are risky—riskier,

even, than alternative candidates—but hire them because of the beneﬁts, both

cultural and ﬁnancial, they’re perceived to bring. The chief, for example, may be

looking for a “cowboy” ocer to work the toughest beat—someone who’s savvy

and unafraid to do what’s necessary to “clean up the streets.”

283

Some agency

leaders may even believe they are doing a service to the profession by helping a

cast-out comrade ﬁnd his way. Unlike a new recruit, a wandering ocer has

earned his spot in policing’s “band of brothers”;

284

that he has been ﬁred may

signal only that he was unfairly maligned or fell victim to “politics.”

285

The

chance to right this perceived wrong may generate for some chiefs a “warm

282. See, e.g., KANE & WHITE, supra note , at .

283. See, e.g., Toner & Rutecki, supra note  (“Certain ocers are more active than other oc-

ers. . . . You have ocers that are out there simply looking for the bad elements. They are

looking for the criminals. They are looking for the drugs. They are looking for the guns out

there, which is what they should be doing.” (quoting Robert Collins Jr., Police Chief, Dolton,

Ill.)).

284. See WILLIAM SHAKESPEARE, KING HENRY V act , sc.  (MIT ed. ) (“We few, we happy

few, we band of brothers; / For he to-day that sheds his blood with me / Shall be my

brother . . . .”).

285. Kelly et al., supra note  (“Former New Orleans police superintendent Ronal Serpas said that

sheris and other chiefs oen justify rehiring ocers by dismissing their problems as ‘politi-

cal.’”).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



glow” that actually makes a wandering ocer more, not less, attractive than a

nonwandering alternative.

286

Alternatively, agencies might hire wandering ocers who are riskier than the

marginal hires if the costs of doing so are lower. Because collective-bargaining

agreements in many law-enforcement agencies across the country, and in Florida

speciﬁcally, establish scheduled pay raises based primarily on years of service,

police administrators typically have little discretion over salaries.

287

It is there-

fore unlikely that an administrator would hire a wandering ocer simply be-

cause that ocer is willing to accept a lower salary than a similarly experienced

marginal candidate who has never been ﬁred. But wandering ocers might be

cheaper than rookies. In many jurisdictions, hiring agencies must pay to send

new recruits to the police academy and fund their salaries during training as

286. See, e.g., Cormier & Doig, supra note  (“‘This stu is supposed to follow you forever? . . .

Of course I’m going to give somebody a second chance.’” (quoting Roberto Fulgueira, Police

Chief of the Sweetwater Police Department)); Schaefer & Kaufman, supra note  (describing

police chief who considers his agency “a place where cops can earn a second chance”); cf. Maya

Lau & Matt Stiles, L.A. County Sheri Alex Villanueva Reinstates Four More Fired Deputies,

L.A.

TIMES (Apr. , ), https://www.latimes.com/local/lanow/la-me-sheri-more

-reinstatements--story.html [https://perma.cc/LC-MNV] (describing a sheri

who “argued that previous sheris were too harsh in punishing deputies and he wants to be

fairer” and who campaigned “on a promise to correct the wrongs of the past, including . . .

addressing the cases of deputies who’d been unfairly disciplined through a ‘truth and recon-

ciliation’ panel”). On the concept of “warm glow,” see James Andreoni, Giving with Impure

Altruism: Applications to Charity and Ricardian Equivalence,  J.

POL. ECON.  ().

287. See, e.g., Seth W. Stoughton, The Incidental Regulation of Policing,  MINN. L. REV. , 

() (describing the New York Police Department’s wage schedule, in which “pay de-

pend[s] on the length of employment, not ocer performance”); Agreement Between the City

of Boca Raton and Fraternal Order of Police Lodge #35, C

ITY OF BOCA RATON  (), http://

bocaraton.granicus.com/DocumentViewer.php?ﬁle=bocaraton_da

ebbe.pdf [https://perma.cc/PRH-QDBP]; Agreement Between the City of St. Peters-

burg and Sun Coast Police Benevolent Association for Police Ocers and Technicians, C

ITY OF ST.

PETERSBURG ,  (), https://www.stpete.org/city_departments/human_resources

/docs/PBAContract.pdf [https://perma.cc/NCN-PG]; City of Orlando Collective Bar-

gaining Agreement with Fraternal Order of Police, Orlando Lodge #25, C

ITY OF ORLANDO

- (), https://orlando.novusagenda.com/AgendaPublic/AttachmentViewer.ashx

?AttachmentID=&ItemID= [https://perma.cc/SP-KWX].

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



well.

288

Training also creates delay; wandering ocers join the agency ready to

deploy.

289

While not easy, it may be possible to diminish the inﬂuence of these per-

ceived beneﬁts on agencies’ hiring decisions. With respect to the cultural factors,

scholars and even some police leaders are increasingly challenging the notion

that high crime—even violent crime—demands a militaristic response.

290

The

idea is that Guardian Ocers, not Warrior Cops, may best accomplish the aspi-

rational aim of law enforcement: “protecting civilians from unnecessary indig-

nity and harm.”

291

“[R]ethinking the professional self-image of policing and

changing some of the core values that inform ocers’ actions and decisions” will

not be easy,

292

but instructive examples do exist.

293

By foregrounding the service

288. See, e.g., Goldman & Puro, supra note , at  (explaining that, when they hire wandering

ocers, “[d]epartments need not pay for the costs of a training academy or the salary of the

trainee while he is in training”); Williams, supra note  (“[S]maller departments and those

that lack sucient funding or are understaed are most likely to hire applicants with prob-

lematic pasts if they have completed state-mandated training, which allows departments to

avoid the cost of sending them to the police academy.”). In Florida, for example, local agencies

are authorized, though not legally required, to pay police-academy tuition. See F

LA. STAT.

 . (). Most do: according to CJAP data, in both  and ,  of the re-

sponding police departments and sheris’ oces reimbursed tuition costs. As a point of ref-

erence, tuition at one program in  ran roughly , for Florida residents and ,

for out-of-state residents. Law Enforcement Ocer, G

A. STONE TECH. COLL., https://

gstc-ecsd-ﬂ.schoolloop.com/pf/cms/view_page?d=x&group_id=&vdid

=igyerxdmb [https://perma.cc/VTZ-HWT]. Given that so many agencies pay these

costs, it may be extremely dicult for agencies that do not to hire new recruits.

289. See, e.g., Williams, supra note  (explaining that wandering ocers “can start work almost

immediately”); Yoder, supra note  (“Given the opportunity to hire a licensed ocer who can

start immediately and for whom the hiring agency doesn’t need to pay training costs, [a small

department] may decide to ignore their history.”).

290. See, e.g., PRESIDENT’S TASK FORCE ON ST CENTURY POLICING, supra note , at ; Seth W.

Stoughton, Principled Policing: Warrior Cops and Guardian Ocers,  W

AKE FOREST L. REV.

, - (); Kate Mather, LAPD Urges Ocers to Be Community Guardians, Not War-

riors on Crime, L.A.

TIMES (Aug. , , : AM), https://www.latimes.com/local/crime

/la-me-warrior-guardians--story.html [https://perma.cc/ABW-FZ]; Nick

Morgan, From Warriors to Guardians, M

AIL TRIB. (Medford, Or.) (Mar. , ),

http://www.mailtribune.com/article//NEWS/ [https://perma.cc

/ZNF-Z].

291. Stoughton, supra note , at .

292. Id. at .

293. See, e.g., Sue Rahr & Stephen K. Rice, From Warriors to Guardians: Recommitting American

Police Culture to Democratic Ideals, N

EW PERSPECTIVES IN POLICING BULL., Apr. , at -,

https://www.ncjrs.gov/pdles/nij/.pdf [https://perma.cc/DXL-NV] (describ-

ing training curriculum reforms in Washington designed to cultivate a guardian mentality by

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



aspects of the job, these same reforms might help temper the “band of brothers”

mentality that may lead chiefs to pity, or at least identify with, wandering oc-

ers. Litigation targeting legal rules that reinforce a self-protective, insular mind-

set might, over time, gradually erode that mentality as well.

294

The perceived ﬁnancial beneﬁts—that wandering ocers are cheaper to hire

than new recruits—may be more straightforward to neutralize. If agencies are

hiring ocers with troubled histories to avoid the startup costs of educating and

training new recruits, then they should not be asked to pay these costs. If the

locality itself cannot allot more money to the police department’s budget, state

and federal authorities may need to intervene. The idea is not a fanciful one—

the federal government, for example, already supports local policing with tens

of billions of dollars. Rachel Harmon has argued that much of this money goes

to programs that are ineective or that may do more harm than good.

295

If that

is right, the suggestion here is simply to repurpose some of these funds to pay

for training rather than tanks.

296

To be clear, we do not mean to suggest that

agencies are correct to think that a wandering ocer is cheaper than a rookie in

the long run. Even if a wandering ocer is less costly upfront, what he gives

with one hand he may take away with the other, later, in the form of attrition

and potential civil liability. All told, the best approach may be to ensure that

agencies are internalizing these countervailing costs, a point to which we now

turn.

E. Cost Externalization

Finally, agencies may hire wandering ocers because they externalize, and

therefore discount, the costs of doing so—in other words, they know it’s risky

but they don’t care.

297

Agencies’ principal ﬁnancial exposure comes from law-

suits against ocers for the harms their misconduct inﬂicts. Although these ac-

tions technically run against the defendant ocers as individuals, ocers are

emphasizing the nobility of policing, procedural justice, crisis intervention, tactical social in-

teraction, and respect).

294. See Aziz Z. Huq & Richard H. McAdams, Litigating the Blue Wall of Silence: How to Challenge

the Police Privilege to Delay Investigation,  U.

CHI. LEGAL F. .

295. See Rachel A. Harmon, Federal Programs and the Real Costs of Policing,  N.Y.U. L. REV. ,

, - ().

296. See id. at - (describing federal grant programs that ﬁnance the acquisition of militaristic

equipment).

297. See, e.g., Dill, supra note  (“You have some agencies that take the approach, we need warm

bodies, so they will hire that [wandering] individual . . . .” (quoting Union County, South

Carolina Sheri David Taylor)).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



virtually always indemniﬁed by the employing locality.

298

Yet, even setting aside

the fact that many civilians wronged by the police will never sue, the locality

itself bears relatively limited exposure due to robust qualiﬁed immunity protec-

tions that immunize “all but the plainly incompetent or those who knowingly

violate the law.”

299

Moreover, even when the locality does incur liability on behalf of an ocer,

there are additional impediments to the generation of eective behavioral incen-

tives for those who run the agency. Roughly half of the agencies covered by one

recent study contribute nothing to the satisfaction of lawsuits brought against

them; central government funds are used to pay the bills.

300

Not all of the agen-

cies that do contribute, moreover, actually experience ﬁnancial pressures; some,

for example, pay from funds that were earmarked for litigation costs alone.

301

All told, because of these complexities in the way localities ﬁnance liability costs,

the majority of agencies suer no ﬁnancial consequences when liability costs in-

crease—they do not “feel the burn,” so to speak.

302

And though political pres-

sures may increase with liability, most agencies do not track or analyze infor-

mation about police litigation in a way that could facilitate learning and

improvement.

303

Municipalities are also exposed to direct liability for faulty hiring deci-

sions.

304

But the opening for plaintis is narrow, limited to situations in which

an ocer’s misconduct was a “plainly obvious consequence of the decision to

hire” the ocer.

305

That said, courts have upheld claims “where there is a close

connection between information a municipality did or should have learned about

an employee in the hiring process and the constitutional violation that ultimately

occurred.”

306

In one recent case, an ocer who had previously been ﬁred from

298. See Joanna C. Schwartz, Police Indemniﬁcation,  N.Y.U. L. REV. ,  ().

299. Malley v. Briggs,  U.S. ,  (); see, e.g., Carbado, supra note , at -. But cf.

Joanna C. Schwartz, How Qualiﬁed Immunity Fails,  Y

ALE L.J. , - () (ﬁnding that

qualiﬁed immunity rarely ends civil rights cases, although allowing that it may discourage

people from ever ﬁling suit).

300. Schwartz, supra note , at .

301. Id. at .

302. See id. at .

303. See generally Joanna C. Schwartz, Myths and Mechanics of Deterrence: The Role of Lawsuits in

Law Enforcement Decisionmaking,  UCLA

L. REV.  () (ﬁnding that ocials do not

typically learn from civil-rights lawsuits to make informed decisions about how to modify

agency policy).

304. Bd. of the Cty. Comm’rs v. Brown,  U.S.  ().

305. Id. at .

306. AVERY ET AL., supra note , at ; see id. at  n. (collecting cases).

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



four separate law-enforcement agencies shot and killed the former mayor of a

small town in South Carolina. Shortly aer the incident, the town’s police chief

noted that, despite the ocer’s rocky past, he had “proven himself” at his current

job: “He’s done a good job, so I guess he got a second chance.”

307

The decedent’s

family sued, alleging, among other theories, negligent hiring by the town. A jury

awarded the family almost one hundred million dollars, including sixty million

in punitive damages against the town.

308

Given our evidence about the risks wandering ocers pose, the jury in a case

like this—or a court deciding a dispositive motion—is justiﬁed in deeming an

agency’s decision to hire a wandering ocer (or failure to conduct an adequate

background investigation) to be probative of fault. The more serious the ocer’s

past misconduct, the more probative the evidence is. Nevertheless, wandering-

ocer status remains but one risk factor among many and rarely should be dis-

positive.

309

Increased judicial willingness to entertain negligent-hiring suits concerning

wandering ocers might, on the margins, aect the frequency with which wan-

dering ocers are hired. Yet the impediments just discussed—to translating ﬁ-

nancial liability into constructive behavioral incentives—are every bit as power-

ful here as in the case of individual ocer liability. To the extent these

impediments contribute to the hiring of wandering ocers, the appropriate re-

sponse would be to improve the system’s mechanisms of accountability so that

agencies internalize the costs of their hiring decisions. But this is hardly a novel

suggestion; indeed, it is clear by now there is no straightforward solution. Po-

lice-liability insurers might pitch in by converting the “large but improbable po-

tential liabilities” of negligent hiring into more salient premium dollars.

310

raising premiums for agencies that hire wandering ocers, for example, insurers

would force local governments to pay for the incremental increase in risk that

307. Harve Jacobs, Cop Accused of Shooting Ex-Mayor Had Previous Encounter, Chief Says, WCSC

(June , , : AM), http://www.livenews.com/story//sled-investigate-ocer

-involved-shooting-in-cottageville [https://perma.cc/UYD-VLHL].

308. See Heath Hamacher, A Matter of Force: $97.5M Jury Award Trains a Spotlight on the Issue of Law

Enforcement Hiring, S.C.

LAW. WKLY. (Oct. , ), https://sclawyersweekly.com/news

////a-matter-of-force--m-jury-award-trains-a-spotlight-on-the-issue-of-law

-enforcement-hiring [https://perma.cc/RZG-U]. The parties later settled for  mil-

lion. South Carolina Mayor’s Death Settlement Reduced to $10M, I

NS. J. (Mar. , ), https://

www.insurancejournal.com/news/southeast////.htm [https://perma.cc

/QVN-MD].

309. See supra Part II (reviewing research on other ocer-level correlates of misconduct).

310. John Rappaport, How Private Insurers Regulate Public Police,  HARV. L. REV. , 

().

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



wandering ocers present regardless whether that risk ultimately material-

ized.

311

Here, too, however, according special treatment to wandering-ocer

status may make most sense if other ocer-level risk factors are also considered.

Of course, many costs of wrongful policing are nonpecuniary in nature. Po-

lice misconduct humiliates and degrades its subjects, creates racial disparities in

criminal-justice outcomes, causes negative health consequences, and breeds cyn-

icism toward the police—which, in turn, can “stymie or hinder public safety ef-

forts and, instead, keep crime rates higher in the same communities where fair

and just policing practices are most needed.”

312

Agencies externalize many of

these nonpecuniary costs as well.

313

Improving transparency ought to allow the

public to monitor the police more eectively and thus to exert pressure on polit-

ical actors to account for these neglected costs of policing.

314

In the meantime,

federal pattern-or-practice lawsuits might help to achieve organizational change

where ﬁnancial penalties do not. The federal government could consider the hir-

ing of wandering ocers, for example, when determining whether and how to

target particular agencies under  U.S.C.  .

315

Barring successful general-accountability reforms, and if future research cor-

roborates our ﬁndings, states could consider the “nuclear option” that Connect-

icut invoked in . Connecticut law now prohibits any local agency from hiring

311. See id.; see also id. at  (describing “feature rating” of insurance policies, the practice of

“charging more to riskier customers . . . based on the presence of traits correlated with riski-

ness”).

312. Marie Ouellet et al., Network Exposure and Excessive Use of Force: Investigating the Social Trans-

mission of Police Misconduct,  C

RIMINOLOGY & PUB. POL’Y  (); see Harmon, supra note

, at -. As Harmon points out, lawful policing can impose many of these costs as well.

Harmon, supra note , at .

313. See, e.g., Harmon, supra note , at  (explaining that, “[f]or local governments to func-

tion as a check on the nonbudgetary costs of policing, the public must be able to monitor and

attribute responsibility for the harm the police do, and political actors must be able to inﬂu-

ence police conduct,” and highlighting ways that federal policing programs “undermine these

preconditions for local accountability”).

314. See, e.g., Barry Friedman & Maria Ponomarenko, Democratic Policing,  N.Y.U. L. REV. ,

 () (“The people must be able to see what their agents are doing so they can evaluate

those actions and exercise control as necessary.”).

315. Some of the important sources on   include Rachel A. Harmon, Promoting Civil Rights

Through Proactive Policing Reform,  S

TAN. L. REV.  (); Debra Livingston, Police Reform

and the Department of Justice: An Essay on Accountability,  B

UFF. CRIM. L. REV.  ();

Stephen Rushin, Federal Enforcement of Police Reform,  F

ORDHAM L. REV.  (); and

Samuel Walker, The New Paradigm of Police Accountability: The U.S. Justice Department Pattern

or Practice Suits in Context,  S

T. LOUIS U. PUB. L. REV.  (). For more information, see

also C

IVIL RIGHTS DIV., U.S. DEP’T OF JUSTICE, THE CIVIL RIGHTS DIVISION’S PATTERN AND

PRACTICE POLICE REFORM WORK: -PRESENT ().

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



an ocer who was previously employed by another Connecticut agency and who

“() was dismissed for malfeasance or other serious misconduct calling into

question such person’s ﬁtness to serve as a police ocer; or () resigned or re-

tired from such ocer’s position while under investigation for such malfeasance

or other serious misconduct.”

316

We are not prepared, on the basis of our evi-

dence alone, to oer a full-throated defense of the nuclear option, but it is cer-

tainly a plausible backstop, and it could be useful to study Connecticut’s experi-

ence to develop a better sense of the costs and beneﬁts of going down this road.

conclusion

Not all those who wander are lost, but in policing, many are. In any given

year over the last three decades, an average of roughly , full-time law-en-

forcement ocers in Florida walk the streets having been ﬁred in the past, and

almost  having been ﬁred for misconduct, not counting the many who were

ﬁred and reinstated in arbitration. These ocers, we have shown, are subse-

quently ﬁred and subjected to “moral character” complaints at elevated rates rel-

ative to both ocers hired as rookies and veterans with clean professional histo-

ries. And we likely underestimate the prevalence of the phenomenon nationwide.

We have, moreover, only a partial understanding of the extent of the problem

wandering ocers pose. Beyond their own misbehavior, wandering ocers may

undermine eorts to improve police culture, as they carry their baggage to new

locales. Worse yet, wandering ocers may “infect” other ocers upon arrival,

317

causing misconduct to metastasize to the farthest reaches of the law-enforcement

community. Future research should investigate these possibilities.

316. CONN. GEN. STAT.  -c(a) ().

317. Cf. Ouellet et al., supra note  (suggesting that peers inﬂuence whether an individual ocer

will engage in misconduct); Edika G. Quispe-Torreblanca & Neil Stewart, Causal Peer Eects

in Police Misconduct,  N

ATURE HUM. BEHAV.  () (estimating peer eects in police mis-

conduct); Daria Roithmayr, The Dynamics of Excessive Force,  U.

CHI. LEGAL F.  (argu-

ing that use of excessive force should be thought of as contagious); Thibaut Horel et

al., The Contagiousness of Police Violence (unpublished manuscript), https://www.law

.uchicago.edu/ﬁles/-/chicago_contagiousness_of_violence.pdf [https://perma.cc

/FX-XJ] (studying whether police shootings are “contagious”).

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



appendix

FIGURE A1.

proportion of separations in which officer obtains subsequent employment

within three years, by professional history of firing for misconduct, 1988-

2013

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE A1.

frequency distribution of raw separation codes, 1988-2016

Separation Code

Frequency

Proportion

Fired

Fired for

Misconduct

Administrative - Unfavorable

(Historical Use Only)

928 0.009 Yes Yes

Misconduct

(Historical Use Only)

584 0.006 Yes Yes

No Cause for Decertification

(Historical Use Only)

90 0.001 Yes Yes

Resigned - Would Not Rehire

(Historical Use Only)

7 0 Yes Yes

Resigned/Retired in Lieu of Separation for

Violating Agency/Training Center Policy

270 0.003 Yes Yes

Resigned/Retired in Lieu of Separation for

Violating Moral Character Standards

216 0.002 Yes Yes

Resigned/Retired While Being Investigated

for Violating Moral Character Standards

1,027 0.01 Yes Yes

Resigned/Retired While Being Investigated

for Violating Agency Policy

1,039 0.01 Yes Yes

Terminated for Violating Agency Policy

(No Moral Character Violation)

1,143 0.011 Yes Yes

Terminated for Violating Ch. 943.13(4),

FS or Moral Character Standards

1,048 0.01 Yes Yes

Under Investigation

(Historical Use Only)

304 0.003 Yes Yes

Failure to Complete Basic

Recruit Training

476 0.005 Yes

Failure to Complete Elder

Abuse Training

114 0.001 Yes

Failure to Meet Mandatory

Retraining Requirement

438 0.004 Yes

Failure to Pass State

Certification Examination

70 0.001 Yes

Failure to Perform Assigned

Tasks Satisfactorily

283 0.003 Yes

Failure to Qualify with Firearm 225 0.002 Yes

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



Failure to Satisfactorily Complete Agency

Field-Training Program

1,083 0.011 Yes

Involuntary Separation

(Historical Use Only)

2 0 Yes

Other - Excessive Absence, Fail Report for

Duty, Sleep on Duty, Etc.

141 0.001 Yes

Staff Termination

(Historical Use Only)

106 0.001 Yes

Administrative Separation

(Not Involving Misconduct)

1,087 0.011

Budgetary Constraints 86 0.001

Deceased 707 0.007

Extended Leave of Absence 17 0

Extended Leave of Absence or Suspension

(Historical Use Only)

135 0.001

Leave of Absence

(Historical Use Only)

10 0

Military Leave of Absence 138 0.001

Not Separated 39,344 0.39

Processed Fingerprints Not Received

Within One Year

364 0.004

Resigned/Retired

(Historical Use Only)

19 0

Retired

(Not Involving Misconduct)

7,845 0.078

Special Elected or Appointed Position 28 0

Suspension 13 0

Temporary Employment Authorization

(Period Exceeded)

48 0

Transfer Within Agency

(No Break in Service)

5,658 0.056

Voluntary Separation

(Not Involving Misconduct)

35,675 0.354

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE A2.

descriptive statistics of cjap hiring requirements, 1997-2016

Mean SD Min Max n

Age

18 0.01 0.12 0 1 6,539

19 0.64 0.48 0 1 6,539

20 0.02 0.13 0 1 6,539

21 0.33 0.47 0 1 6,539

൒22

0 0.07 0 1 6,539

Education

High School/

GED

0.9 0.3 0 1 6,581

Associate’s/

Some College

0.09 0.29 0 1 6,581

Bachelor’s 0.01 0.08 0 1 6,581

Criminal-Justice

Experience

0.07 0.26 0 1 6,575

Tobacco

Requirement

0.26 0.44 0 1 6,575

Driving Test

0.79 0.41 0 1 5,252

Interview

0.92 0.27 0 1 6,585

Physical-Ability

Test

0.49 0.5 0 1 6,570

Polygraph Test

0.5 0.5 0 1 6,564

Psychological

Test

0.75 0.43 0 1 6,562

Selection Exam

0.54 0.5 0 1 6,575

Swimming Test

0.09 0.29 0 1 6,254

Vision Test

0.61 0.49 0 1 6,570

Voice Test

0.23 0.42 0 1 5,906

Probation Period

(In Months)

11.5 2.72 0 24 5,811

Composite

Requirement Score

4.61 1.96 0 9 6,469

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE A3.

descriptive statistics of cjap training variables, 1997-2016

Mean SD Min Max n

Chemical

Agents

Not Required 0.43 0.49 0 1 6,207

Every 24-48 Months 0.21 0.41 0 1 6,207

Every 6-12 Months 0.36 0.48 0 1 6,207

Self-

Defense

Not Required 0.39 0.49 0 1 6,530

Every 24-48 Months 0.17 0.38 0 1 6,530

Every 6-12 Months 0.44 0.50 0 1 6,530

Driving

Not Required 0.59 0.49 0 1 6,521

Every 24-48 Months 0.20 0.40 0 1 6,521

Every 6-12 Months 0.20 0.40 0 1 6,521

Firearm

Not Required 0.01 0.09 0 1 6,550

Every 12-48 Months 0.89 0.31 0 1 6,550

Every 6 Months 0.10 0.30 0 1 6,550

In-Service

Not Required 0.85 0.36 0 1 5,192

Every 24-48 Months 0.02 0.14 0 1 5,192

Every 6-12 Months 0.13 0.34 0 1 5,192

Medical

Not Required 0.31 0.46 0 1 6,527

Every 24-48 Months 0.49 0.50 0 1 6,527

Every 6-12 Months 0.21 0.40 0 1 6,527

FTO

11.96 5.91 0 52 6,486

Continued

0.68 0.47 0 1 5,887

Composite

Training Score

2.91 1.44 0 6 6,094

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE A4.

median time to new employment by demographic groups and professional

history, 1988-2013

Panel A: Fired Panel B: Fired for Misconduct

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Never

Fired

Fired,

Last Job

Fired,

Earlier

Job

Race

White

(19,280)

318

(1,175)

(1,024)

(19,977)

427

(800)

(702)

Black

(2,049)

401

(210)

(198)

(2,224)

457

(129)

(104)

Hispanic

(2,519)

470

(212)

(154)

(2,649)

553

(137)

(99)

Gender

Male

(21,451)

335

(1,441)

(1,288)

(22,327)

442

(986)

(867)

Female

(2,750)

406

(180)

(98)

(2,892)

692

(92)

(44)

Education

High School

(937)

509

(55)

158

(36)

(977)

664

(30)

269

(21)

Associate’s

(3,401)

322

(247)

(179)

(3,538)

498

(162)

(127)

Bachelor’s

(6,982)

327

(347)

(270)

(7,236)

404

(210)

(153)

Master’s

(1,436)

376

(58)

(69)

(1,491)

642

(33)

(39)

All

(24,205)

350

(1,621)

(1,386)

(25,223)

450

(1,078)

(911)

TABLE A5.

subsequent firing by professional history, 1996-2013

Fired Fired for Misconduct

n Ever 3 Years Ever 3 Years

Panel A: Firings

Never 21,693 7.6% 4.0% 5.3% 2.3%

Fired, last job 1,200 16.8% 10.8% 11.3% 6.7%

Fired, earlier job 1,143 13.9% 7.0% 10.1% 5.0%

Panel B: Firings

for Misconduct

Never 22,603 7.8% 4.1% 5.4% 2.4%

Fired, last job 761 18.1% 11.3% 12.6% 7.2%

Fired, earlier job 672 14.4% 7.4% 10.1% 5.4%

Panel C: Rookie

39,007 9.4% 5.4% 6.0% 2.5%

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE A6.

subsequent firing by professional history with one-, three-, and five-year

time windows, 1988-2013

318

Fired Fired for Misconduct

n 1-Year 3-Year 5-Year 1-Year 3-Year 5-Year

Panel A:

Firings

Never 29,888 2.7% 4.5% 6.0% 1.5% 3.0% 4.3%

Fired,

last job

1,969 7.3% 11.2% 13.7% 4.8% 7.9% 9.8%

Fired,

earlier job

1,631 4.6% 7.5% 9.9% 3.2% 5.6% 7.5%

Panel B:

Firings

for Mis-

conduct

Never 31,182 2.8% 4.7% 6.1% 1.6% 3.1% 4.4%

Fired,

last job

1,297 7.7% 12.2% 14.8% 5.4% 9.0% 11.2%

Fired,

earlier job

1,009 5.2% 8.1% 10.7% 3.6% 6.2% 8.1%

Panel C:

Rookie

54,476 3.8% 5.8% 7.3% 1.6% 3.2% 4.5%

TABLE A7.

subsequent firing by professional history for matched comparators based

on timing, geography, and agency desirability, 1996-2013

Fired Fired for Misconduct

Ever 3 Years Ever 3 Years

Panel A: Firings

Never

10% 5.4% 7.1 3.3

Fired, last job

Wanderer 16.7% 10.3% 11.3% 6.2%

Comparator 9.03% 5.3% 6.2% 2.9%

Fired, earlier job

Wanderer 13.8% 6.7% 10.1% 5.0%

Comparator 9.8% 5.4% 6.4% 2.7%

Panel B: Firings

for Misconduct

Never

10.1% 5.5% 7.2% 3.3%

Fired, last job

Wanderer 17.5% 10.8% 12.2% 6.7%

Comparator 9% 5.3% 6.0% 2.9%

Fired, earlier job

Wanderer 14.5% 7.2% 10.2% 5.4%

Comparator 9.6% 5.3% 6% 2.6%

318. For the -year time window, we exclude all job stints beginning aer  to ensure that we

have at least  years of follow-up for each observation. The sample sizes are thus slightly

smaller for these estimates.

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



TABLE A8.

number of complaints by professional history with one-, three-, and five-

year time windows, 1993-2013

n 1-Year 3-Year 5-Year

All Complaints

Panel A: Firings

Never 24,711 0.01 0.02 0.04

Fired, last job 1,394 0.02 0.07 0.09

Fired, earlier job 1,295 0.02 0.05 0.06

Panel B: Firings

for Misconduct

Never 25,686 0.01 0.02 0.04

Fired, last job 934 0.03 0.08 0.10

Fired, earlier job 780 0.02 0.05 0.07

Panel C: Rookie Panel C: Rookie 44,584 0.01 0.03 0.04

Violent and

Sexual Com

laints

Panel A: Firings

Never 24,711 0.00 0.01 0.01

Fired, last job 1,394 0.01 0.02 0.03

Fired, earlier job 1,295 0.01 0.01 0.02

Panel B: Firings

for Misconduct

Never 25,686 0.00 0.01 0.01

Fired, last job 934 0.01 0.03 0.03

Fired, earlier job 780 0.01 0.01 0.02

Panel C: Rookie Panel C: Rookie 44,584 0.00 0.01 0.01

Integrity Complaints

Panel A: Firings

Never 24,711 0.00 0.01 0.01

Fired, last job 1,394 0.01 0.03 0.03

Fired, earlier job 1,295 0.01 0.02 0.03

Panel B: Firings

for Misconduct

Never 25,686 0.00 0.01 0.01

Fired, last job 934 0.01 0.03 0.04

Fired, earlier job 780 0.01 0.02 0.03

Panel C: Rookie Panel C: Rookie 44,584 0.00 0.01 0.01

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE A9.

number of complaints by professional history for matched comparators

based on timing, geography, and agency desirability, 1996-2013

All

Violent/

Sexual

Integrity

Ever

Years

Ever

Years

Ever

Years

Panel A:

Firings

Never

0.08 0.02 0.02 0.01 0.03 0.01

Fired,

last job

Wanderer 0.13 0.06 0.04 0.02 0.05 0.02

Comparator 0.08 0.03 0.02 0.01 0.03 0.01

Fired,

earlier job

Wanderer 0.13 0.06 0.03 0.01 0.05 0.02

Comparator 0.07 0.03 0.01 0.01 0.03 0.01

Panel B:

Firings for

Misconduct

Never

0.08 0.02 0.02 0.01 0.03 0.01

Fired,

last job

Wanderer 0.16 0.07 0.04 0.02 0.06 0.03

Comparator 0.08 0.04 0.02 0.01 0.03 0.01

Fired,

earlier job

Wanderer 0.14 0.06 0.03 0.01 0.06 0.02

Comparator 0.07 0.02 0.01 0.01 0.03 0.01

TABLE A10.

predicting if wandering officers are fired for misconduct again, 1988-2013

Model 1 Model 2 Model 3 Model 4

Intercept

0.062**

[0.008]

0.060†

[0.030]

0.011

[0.056]

0.114*

[0.054]

Fired from

Last Job

0.029*

[0.011]

0.036*

[0.015]

0.063*

[0.026]

0.021

[0.022]

Job

Number

[0.023]

-0.021

[0.049]

-0.03

[0.037]

0.007

[0.023]

-0.009

[0.049]

-0.002

[0.038]

≥ 4

0.023

[0.026]

0.016

[0.053]

-0.012

[0.039]

Age

(At Start)

-0.001

[0.001]

-0.001

[0.001]

Male

0.007

[0.021]

-0.005

[0.034]

-0.031

[0.036]

Any Past

Complaints

-0.002

[0.013]

-0.011

[0.020]

-0.006

[0.017]

Electronic copy available at: https://ssrn.com/abstract=3589544

the wandering officer



Education

Associate’s

0.033

[0.024]

Bachelor’s

0.051*

[0.024]

Master’s/Doctorate

0.047

[0.034]

Hiring

Agency

Proportion Black

-0.086

[0.062]

Proportion Hispanic

-0.043

[0.039]

Violent-Crime Rate

(Per 100,000)

[0.000]

Property-Crime Rate

(Per 100,000)

[0.000]

County-Level

Unemployment Rate

0.129

[0.365]

Expenditures Per Officer

(2008)

[0.000]

Hiring Requirement

Score

-0.001

[0.004]

Training Requirement

Score

-0.001

[0.007]

Number of Officers

-0.002

[0.001]

Separating

Agency

Number of Officers

-0.001

[0.001]

2,272 2,272 848 882

TABLE A11.

number of fdle-initiated complaints by professional history, 1993-2013

n Ever 3 Years

Panel A: Firings

Never 24,711 0.02 0.01

Fired, last job 1,394 0.04 0.02

Fired, earlier job 1,295 0.05 0.02

Panel B: Firings

for Misconduct

Never 25,686 0.03 0.01

Fired, last job 934 0.05 0.02

Fired, earlier job 780 0.06 0.02

Panel C: Rookies

44,584 0.03 0.01

Electronic copy available at: https://ssrn.com/abstract=3589544

the yale law journal : 



TABLE A12.

subsequent firings and number of complaints by professional history and

agency size, 1988-2013

319

Firings

Firings for

Misconduct

Complaints

Size

n Ever 3 Yr. Ever 3 Yr. n Ever 3 Yr.

≤15

Officers

Never 2,295 10% 6% 7% 4% 1,822 0.07 0.04

Fired, last job 413 21% 12% 15% 9% 314 0.12 0.09

Fired, earlier job 268 19% 10% 15% 9% 210 0.12 0.06

Rookie 2,001 12% 7% 9% 5% 1,554 0.07 0.03

>15

Officers

Never 27,593 9% 4% 7% 3% 22,889 0.07 0.02

Fired, last job 1,556 18% 11% 13% 8% 1,080 0.14 0.06

Fired, earlier job 1,363 14% 7% 10% 5% 1,085 0.12 0.04

Rookie 52,475 10% 6% 7% 3% 43,030 0.08 0.03

TABLE A13.

subsequent firings and number of complaints by professional history and

timing, 1993-2009

320

Firing

Firing for

Misconduct

Complaints

Years

1-3

Years

4-7

Years

1-3

Years

4-7

Years

1-3

Years

4-7

Panel A:

Firings

Never 26,346 4.6% 2.3% 3.2% 2.0% 21,169 0.024 0.024

Fired,

last job

1,811 11.2% 4% 8% 3.4% 1,236 0.070 0.038

Fired,

earlier job

1,487 7.5% 4% 5. 9% 2.8% 1,151 0.050 0.042

Panel B:

Firings

for Mis-

conduct

Never 27,516 4.8% 2.4% 3.3% 2.0% 22,020 0.025 0.024

Fired,

last job

1,198 12.1% 4.4% 9.1% 3.7% 835 0.079 0.044

Fired,

earlier job

930 8.3% 4% 6.6% 2.6% 701 0.056 0.054

Panel C:

Rookie

47,371 5.8% 2.4% 3.3% 2.1% 37,479 0.026 0.033

319. The ﬁring data is based on employment stints beginning between  and , and the

complaint data is based on stints beginning between  and .

320. We exclude all employment stints that began aer  to ensure that we have a full seven-

year window for each employment stint through which to observe ﬁrings and complaints.

Electronic copy available at: https://ssrn.com/abstract=3589544