Correction for Johnson et al., Officer characteristics and racial disparities in fatal officer-involved shootings


Correction

PSYCHOLOGICAL AND COGNITIVE SCIENCES
Correction for “Officer characteristics and racial disparities in
fatal officer-involved shootings,” by David J. Johnson, Trevor
Tress, Nicole Burkel, Carley Taylor, and Joseph Cesario, which
was first published July 22, 2019; 10.1073/pnas.1903856116 (Proc.
Natl. Acad. Sci. U.S.A. 116, 15877–15882).
The authors wish to note the following: “Recently, we pub-

lished a report showing that, among civilians fatally shot, officer
race did not predict civilian race and there was no evidence of
anti-Black or anti-Hispanic disparities (1). Specifically, we esti-
mated the probability that a civilian was Black, Hispanic, or
White given that a person was fatally shot and some covariates.
The dataset contains only information about individuals fatally
shot by police, and the race of the individual is predicted by a set
of variables. Thus, we compute Pr(racejshot, X) where X is a set
of variables including officer race.
“Although we were clear about the quantity we estimated and

provide justification for calculating Pr(racejshot, X) in our report
(see also 2, 3), we want to correct a sentence in our significance
statement that has been quoted by others stating ‘White officers
are not more likely to shoot minority civilians than non-White
officers.’ This sentence refers to estimating Pr(shotjrace, X). As
we estimated Pr(racejshot, X), this sentence should read: ‘As the
proportion of White officers in a fatal officer-involved shooting
increased, a person fatally shot was not more likely to be of a
racial minority.’ This is consistent with our framing of the results
in the abstract and main text.
“We appreciate the feedback that led us to clarify this sen-

tence (4). To be clear, this issue does not invalidate the findings
with regards to Pr(racejshot, X) discussed in the report.”

1. D. J. Johnson, T. Tress, N. Burkel, C. Taylor, J. Cesario, Officer characteristics and racial
disparities in fatal officer-involved shootings. Proc. Natl. Acad. Sci. U.S.A. 116,
15877–15882 (2019).

2. D. J. Johnson, J. Cesario, Reply to Knox and Mummolo and Schimmack and Carlsson:
Controlling for crime and population rates. Proc. Natl. Acad. Sci. U.S.A. 117, 1264–
1265 (2020).

3. D. J. Johnson, J. Cesario, Reply to Knox and Mummolo: Critique of Johnson et al. (2019).
https://doi.org/10.31234/osf.io/dmhpu (16 August 2019).

4. D. Knox, J. Mummolo, Making inferences about racial disparities in police violence. Proc.
Natl. Acad. Sci. U.S.A. 117, 1261–1262 (2020).

Published under the PNAS license.

First published April 13, 2020.

www.pnas.org/cgi/doi/10.1073/pnas.2004734117

www.pnas.org PNAS | April 21, 2020 | vol. 117 | no. 16 | 9127

C
O
R
R
EC

TI
O
N

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

D
o
w

n
lo

a
d
e
d
 a

t 
C

a
rn

e
g
ie

 M
e
llo

n
 U

n
iv

e
rs

ity
 o

n
 A

p
ri
l 5

, 
2
0
2
1
 

https://doi.org/10.31234/osf.io/dmhpu
https://www.pnas.org/site/aboutpnas/licenses.xhtml
https://www.pnas.org/cgi/doi/10.1073/pnas.2004734117
https://www.pnas.org


Retraction

PSYCHOLOGICAL AND COGNITIVE SCIENCES
Retraction for “Officer characteristics and racial disparities in
fatal officer-involved shootings,” by David J. Johnson, Trevor
Tress, Nicole Burkel, Carley Taylor, and Joseph Cesario, which
was first published July 22, 2019; 10.1073/pnas.1903856116 (Proc.
Natl. Acad. Sci. U.S.A. 116, 15877–15882).
The authors wish to note the following: “Our article estimated

the role of officer characteristics in predicting the race of civilians
fatally shot by police. A critique pointed out we had erroneously
made statements about racial differences in the probability of being
shot (1), and we issued a correction to rectify the statement (2).
Despite this correction, our work has continued to be cited as

providing support for the idea that there are no racial biases in
fatal shootings, or policing in general. To be clear, our work does
not speak to these issues and should not be used to support such
statements. We take full responsibility for not being careful
enough with the inferences made in our original report, as this
directly led to the misunderstanding of our research.
While our data and statistical approach were appropriate for

investigating whether officer characteristics are related to the race
of civilians fatally shot by police, they are inadequate to address
racial disparities in the probability of being shot.
Given these issues and the continued use of our work in the

public debate on this topic, we have decided to retract the article.”

1. D. Knox, J. Mummolo, Making inferences about racial disparities in police violence.
Proc. Natl. Acad. Sci. U.S.A. 117, 1261–1262 (2020).

2. D. J. Johnson, T. Tress, N. Burkel, C. Taylor, J. Cesario, Correction for Johnson et al.,
Officer characteristics and racial disparities in fatal officer-involved shootings. Proc.
Natl. Acad. Sci. U.S.A. 117, 9127 (2020).

Published under the PNAS license.

First published July 10, 2020.

www.pnas.org/cgi/doi/10.1073/pnas.2014148117

18130 | PNAS | July 28, 2020 | vol. 117 | no. 30 www.pnas.org

https://www.pnas.org/site/aboutpnas/licenses.xhtml
https://www.pnas.org/cgi/doi/10.1073/pnas.2014148117


P
SY

C
H

O
LO

G
IC

A
L

A
N

D
C

O
G

N
IT

IV
E

SC
IE

N
C

ES

Officer characteristics and racial disparities in fatal
officer-involved shootings
David J. Johnsona,b,1, Trevor Tressb, Nicole Burkelb, Carley Taylorb, and Joseph Cesariob

a Department of Psychology, University of Maryland at College Park, College Park, MD 20742; and b Department of Psychology, Michigan State University,
East Lansing, MI 48824

Edited by Kenneth W. Wachter, University of California, Berkeley, CA, and approved June 24, 2019 (received for review March 5, 2019)

Despite extensive attention to racial disparities in police shoot-
ings, two problems have hindered progress on this issue. First,
databases of fatal officer-involved shootings (FOIS) lack details
about officers, making it difficult to test whether racial dispari-
ties vary by officer characteristics. Second, there are conflicting
views on which benchmark should be used to determine racial
disparities when the outcome is the rate at which members from
racial groups are fatally shot. We address these issues by creating
a database of FOIS that includes detailed officer information. We
test racial disparities using an approach that sidesteps the bench-
mark debate by directly predicting the race of civilians fatally shot
rather than comparing the rate at which racial groups are shot to
some benchmark. We report three main findings: 1) As the pro-
portion of Black or Hispanic officers in a FOIS increases, a person
shot is more likely to be Black or Hispanic than White, a disparity
explained by county demographics; 2) race-specific county-level
violent crime strongly predicts the race of the civilian shot; and
3) although we find no overall evidence of anti-Black or anti-
Hispanic disparities in fatal shootings, when focusing on different
subtypes of shootings (e.g., unarmed shootings or “suicide by
cop”), data are too uncertain to draw firm conclusions. We high-
light the need to enforce federal policies that record both officer
and civilian information in FOIS.

officer-involved shootings | racial disparity | racial bias | police use of force |
benchmarks

Recent high-profile police shootings of Black Americans haveraised questions about racial disparities in fatal officer-
involved shootings (FOIS). These shootings have captured public
concern, leading in part to the Black Lives Matter movement and
a presidential task force on policing (1). Central to this debate
are questions of whether Black civilians are overrepresented in
FOIS and whether racial disparities are due to discrimination by
White officers. However, a lack of data about officers in FOIS
and disagreement on the correct benchmark for determining
racial disparity in FOIS have led to conflicting conclusions about
the degree to which Black civilians are more likely to be fatally
shot than White civilians. We address both issues by creating a
comprehensive database of FOIS that includes officer informa-
tion and by using a method for testing racial disparities that does
not rely on benchmarks.

Until recently, the only nationwide data on FOIS was compiled
yearly in the Federal Bureau of Investigation (FBI) Uniform
Crime Report. On a voluntary basis, departments report the
number of justifiable homicides by on-duty law-enforcement offi-
cers. Not only are these shootings underreported (by ∼50%;
ref. 2), such reports do not provide information about the
officers or circumstances surrounding these shootings. Begin-
ning in 2015, news companies such as The Washington Post
and The Guardian began to collect information about FOIS
to address the issues with the FBI data. Through reporting
and Freedom of Information Act requests to law-enforcement
agencies, such organizations have created more complete FOIS
databases. These databases provide information about shoot-
ings not available in federal databases: where they took place,

what police departments were involved, and demographic infor-
mation about civilians. However, even these databases fail to
provide information about involved officers, which prevents ask-
ing whether certain types of officers are more likely to show racial
disparities.∗

When officers fire their weapons at civilians, there are three
possible outcomes: 1) They miss the civilian, 2) they result in a
nonfatal hit, or 3) they result in a fatal hit. Not only do officers
miss civilians the majority of times they fire [estimates of hit rates
range from 20 to 40% (5, 6)], many shootings do not result in
fatalities. Thus, it is important to be clear at the outset that our
analyses speak to racial disparities in the subset of shootings that
result in fatalities, and not officers’ decisions to use lethal force
more generally.

Why should we expect officer characteristics to relate to the
race of a person fatally shot? Decades of research from criminal
justice have investigated whether officer characteristics relate to
the degree of force used by police. Whereas officer race does
not typically predict how much force an officer uses (7–11), male
and inexperienced officers use more force (7, 8, 10), perhaps
due to their use of more aggressive tactics (e.g., initiating more
stops; ref. 11). One issue with this research is that it focuses
on whether officer characteristics increase the degree of force
used, not whether force is used disproportionately by civilian
race. Some researchers have proposed that racial disparities in
FOIS might be driven by discrimination by White officers (12),
but research on this issue is uncommon due to a lack of officer

Significance

There is widespread concern about racial disparities in fatal
officer-involved shootings and that these disparities reflect
discrimination by White officers. Existing databases of fatal
shootings lack information about officers, and past analytic
approaches have made it difficult to assess the contributions
of factors like crime. We create a comprehensive database of
officers involved in fatal shootings during 2015 and predict
victim race from civilian, officer, and county characteristics.
We find no evidence of anti-Black or anti-Hispanic dispari-
ties across shootings, and White officers are not more likely
to shoot minority civilians than non-White officers. Instead,
race-specific crime strongly predicts civilian race. This suggests
that increasing diversity among officers by itself is unlikely to
reduce racial disparity in police shootings.

Author contributions: D.J.J. and J.C. designed research; D.J.J., T.T., N.B., and C.T.
performed research; D.J.J. analyzed data; and D.J.J. and J.C. wrote the paper.y

The authors declare no conflict of interest.y

This article is a PNAS Direct Submission.y

Published under the PNAS license.y
1 To whom correspondence should be addressed. Email: djjohnson@smcm.edu.y

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1903856116/-/DCSupplemental.y

Published online July 22, 2019.

* Although some news organizations have gathered officer information, they have either
not released it (3) or gathered information only about large departments (4).

www.pnas.org/cgi/doi/10.1073/pnas.1903856116 PNAS | August 6, 2019 | vol. 116 | no. 32 | 15877–15882

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
July 10, 2020

https://www.pnas.org/site/aboutpnas/licenses.xhtml
mailto:djjohnson@smcm.edu
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/cgi/doi/10.1073/pnas.1903856116
http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1903856116&domain=pdf


data. The only national examination of this question found that
White officers were no more likely to fatally shoot Black or His-
panic civilians than non-White officers (13). However, their key
analyses were based on only a small subset (19–23%) of all fatal
shootings. Beyond race, researchers have not tested whether offi-
cer sex or experience impact racial disparities in fatal shootings.
To address this gap, we created a comprehensive database of all
FOIS in 2015 with information about officer race, sex, and years
of experience. However, even with this officer information on
hand, there is still a challenge of exactly how to quantify racial
disparities in FOIS.

How to Calculate Racial Disparities in FOIS
A persistent point of debate in studying police use of force con-
cerns how to calculate racial disparities. Racial disparities in fatal
shootings have traditionally been tested by asking whether offi-
cers fatally shoot a racial group more than some benchmark, such
as that group’s population proportion in the United States. Dis-
parity is assumed when the rate of fatal shootings deviates from
this benchmark. For example, 26% of civilians killed by police
shootings in 2015 were Black (3, 14), even though Black civilians
comprise only 12% of the US population. According to this 12%
benchmark, more Black civilians are fatally shot than we would
expect, indicating disparity. News organizations and researchers
using this method (12, 15–19) find robust evidence of anti-Black
disparity in fatal shootings.

However, using population as a benchmark makes the strong
assumption that White and Black civilians have equal exposure
to situations that result in FOIS. If there are racial differences
in exposure to these situations, calculations of racial disparity
based on population benchmarks will be misleading (20, 21).
Researchers have attempted to avoid this issue by using race-
specific violent crime as a benchmark, as the majority of FOIS
involve armed civilians (22). When violent crime is used as a
benchmark, anti-Black disparities in FOIS disappear or even
reverse (20, 23–25).

In essence, benchmarking approaches test whether mem-
bers from certain racial groups are shot more than we would
expect relative to some benchmark. The issue is that conclu-
sions regarding racial disparities depend more on the benchmark
used (population or violent crime) than the data (the number
of people fatally shot). Rather than trying to identify which
benchmark is best, another way to test for racial disparities in
FOIS is to directly predict the race of a person fatally shot.
Specifically, we used multinomial regression with civilian race
as the outcome and various factors—officer, civilian, and county
characteristics—as predictors. In this way, we approached racial
disparity from a different angle and asked: “What factors predict
the race of a person fatally shot by police?”

This approach has several benefits. By focusing on individual
shootings, we can test how much officer and civilian character-
istics predict racial disparities in FOIS. A benchmark approach
necessarily blends data on individual shootings with the broader
population, stripping away the context in which FOIS take place.
Second, this approach can test the degree to which common
benchmarks like violent crime predict the race of a person shot.
This is more informative than tying FOIS deaths to a single
benchmark, which provides no information about the predic-
tive validity of that factor. Third, this approach estimates racial
disparity in FOIS, controlling for civilian, officer, and other
contextual variables simultaneously. Whatever remains when
controlling for all relevant variables provides an upper bound for
racial disparity in FOIS. Finally, this approach can test whether
racial disparities vary by the type of shooting.

Racial Disparities by Type of Shooting
By creating a more detailed database of FOIS and focusing on
individual shootings, we are able to address how the type of

shooting might impact racial disparities in FOIS. For example,
anti-Black or anti-Hispanic disparities in fatal shootings may
depend on whether the civilian was armed or suicidal.

Examination of National Violent Death Reporting System
data shows racial differences across types of fatal shootings.
Black civilians fatally shot by police (relative to White civilians)
are more likely to be unarmed and less likely to pose an immedi-
ate threat to officers (26). In contrast, White civilians (relative to
Black civilians) are nearly three times more likely to be fatally
shot by police when the incident is related to mental-health
concerns and are seven times more likely to commit “suicide
by cop” (26). These are incidents where a civilian threatens a
police officer for the purpose of ending their life (27) and reflect
higher rates of suicide overall among Whites relative to Black
and Hispanic civilians (28).

Racial differences in the frequency of certain types of FOIS
matter because they may mask racial disparities in other types
of fatal shootings. Even if a person fatally shot during a criminal
encounter is more likely to be Black than White, this disparity
will be difficult to detect if White civilians commit suicide by
police intervention more frequently and such cases represent a
large proportion of the overall FOIS. As past work has not distin-
guished between shootings where the civilian is or is not suicidal,
it is unclear how much these disparities cancel each other out.

Results
Given the lack of national data on officers in FOIS, we first
briefly described the officers involved in fatal shootings dur-
ing 2015. Civilian and county characteristics are provided in SI
Appendix. In a majority of FOIS (56%), a single officer fired their
weapon. In 39% of cases, two to four officers fired their weapons.
Cases with five or more officers were rare (5%). Compared with
officers nationwide (73% White, 12% Black, 12% Hispanic, 88%
male; ref. 29), 79% of officers were White, 12% Hispanic, 6%
Black, and 3% from other racial groups. Officers were over-
whelmingly male (96%). The average officer had almost 10 y of
experience (officers often retire after 20 y; ref. 30).

Officer and Civilian Characteristics. To test whether officer charac-
teristics predict the race of a person fatally shot, we regressed
victim race against all officer and civilian predictors. Predictors
and coefficients for this model are reported in Table 1. For all
effects, we report odds ratios (OR) comparing Black or Hispanic
individuals to Whites and 95% CIs (in brackets). In terms of
officer race, as the percentage of Black officers who shot in a
FOIS increased, a person fatally shot was more likely to be Black

Table 1. Predicting Race from Officer and Civilian Factors

Black Hispanic

Variable OR 95% CI OR 95% CI

Intercept 0.25 0.14, 0.44 0.29 0.18, 0.47
Civilian age 0.54 0.45, 0.66 0.57 0.47, 0.70
Civilian armed 0.57 0.26, 1.23 0.98 0.35, 2.75
Civilian mental health issue 0.70 0.44, 1.14 0.59 0.29, 1.20
Civilian suicidal 0.28 0.12, 0.64 1.15 0.45, 2.91
Civilian attacking 1.74 0.54, 5.56 1.04 0.30, 3.63
Officer number 1.04 0.89, 1.22 1.08 0.91, 1.28
Officer % Black 1.23 1.03, 1.48 1.15 0.92, 1.44
Officer % Hispanic 1.29 1.07, 1.56 1.84 1.54, 2.20
Officer % women 1.13 0.95, 1.35 1.04 0.84, 1.28
Average officer experience 1.12 0.93, 1.33 1.00 0.82, 1.23

OR above (below) 1.00 indicate a positive (negative) relationship
between the predictor and the odds that a person fatally shot is Black or
Hispanic. Whites served as the referent group. n = 917. χ2 (20) = 71.73;
P < 0.001; R2 = 0.24.

15878 | www.pnas.org/cgi/doi/10.1073/pnas.1903856116 Johnson et al.

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
July 10, 2020 

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/cgi/doi/10.1073/pnas.1903856116


P
SY

C
H

O
LO

G
IC

A
L

A
N

D
C

O
G

N
IT

IV
E

SC
IE

N
C

ES

Table 2. Predicting Race from Officer, Civilian, and County
Factors

Black Hispanic

Variable OR 95% CI OR 95% CI

Intercept 0.14 0.08, 0.25 0.18 0.12, 0.27
Civilian age 0.58 0.46, 0.72 0.55 0.44, 0.70
Civilian armed 0.54 0.24, 1.24 1.13 0.44, 2.91
Civilian mental health issue 0.52 0.30, 0.89 0.41 0.19, 0.87
Civilian suicidal 0.28 0.13, 0.61 1.24 0.43, 3.63
Civilian attacking 2.22 0.63, 7.81 1.02 0.33, 3.15
Officer number 1.04 0.86, 1.26 1.11 0.90, 1.37
Officer % Black 1.06 0.87, 1.30 1.02 0.81, 1.29
Officer % Hispanic 1.23 1.00, 1.51 1.26 1.05, 1.51
Officer % women 1.04 0.87, 1.25 0.94 0.75, 1.19
Average officer experience 1.04 0.85, 1.26 1.01 0.80, 1.28
County population size 1.18 0.96, 1.46 1.13 0.91, 1.39
County median income 1.42 1.10, 1.82 1.16 0.89, 1.52
County income inequality 1.15 0.88, 1.50 1.07 0.77, 1.49
County % rural 1.24 0.89, 1.72 1.25 0.78, 1.98
County % White homicide 1.31 0.28, 6.13 0.61 0.30, 1.27
County % Black homicide 4.52 1.09, 18.8 0.96 0.48, 1.89
County % Hispanic homicide 1.32 0.36, 4.77 2.12 1.07, 4.20

Odds ratios (OR) above (below) 1.00 indicate a positive (negative) rela-
tionship between the predictor and the odds that a person fatally shot is
Black or Hispanic. Whites served as the referent group. n = 917. χ2(34) =
183.57; P < 0.001; R2 = 0.52.

(OR = 1.23 [1.03, 1.48]) than White. As the percentage of His-
panic officers who shot in a FOIS increased, a person fatally
shot was more likely to be Hispanic (OR = 1.84 [1.54, 2.20])
or Black (OR = 1.29 [1.07, 1.56]) than White. The number of
officers, percentage of female officers, and average experience
of officers did not predict civilian race. Older civilians were 1.85
times less likely (OR = 0.54 [0.45, 0.66]) to be Black than White
and 1.75 times less likely (OR = 0.57 [0.47, 0.70]) to be Hispanic
than White. Suicidal civilians were 3.57 times less likely (OR =
0.28 [0.12, 0.64]) to be Black than White. In sum, as the percent-
age of Black or Hispanic officers increased, the likelihood that
a civilian fatally shot was Black or Hispanic (respectively) also
increased.

Greater anti-Black and anti-Hispanic disparity among fatal
shootings where more Black or Hispanic officers were involved
might not be due to bias on the part of Black or Hispanic offi-
cers, but instead to simple overlap between officer and county
demographics. To test this, we reran the model including county
demographics. Model coefficients are reported in Table 2. When
county variables were included, the relationship between offi-
cer and civilian race was attenuated or eliminated. Black officers
were not more likely to fatally shoot Black civilians (OR = 1.06
vs. 1.23), and Hispanic officers were less likely to fatally shoot
Black (OR = 1.23 vs. 1.29) and Hispanic (OR = 1.32 vs. 1.84)
civilians, although the latter disparity was still significant. This
suggests that the association between officer race and Black and
Hispanic disparities in FOIS largely occur because officers and
civilians are drawn from the same population. Additional analy-
ses (SI Appendix) provided converging evidence for this account;
counties with more Hispanic civilians also had more Black or
Hispanic officers (r = 0.82 and 0.87, respectively).

County Characteristics. We also tested whether county variables
predict the race of a person fatally shot. An advantage of
conducting our analyses at the level of individual shootings
is the ability to test the degree to which race-specific violent
crime and population proportions predict the race of a per-
son fatally shot. We could not test this question in the model

with all county-level predictors because of the strong correlation
between violent crime and population size for all races (r > 0.85;
SI Appendix). We therefore examined the effects of each variable
independently.

If crime matters for police shootings, as race-specific crime
rates increase for a given group (i.e., Black or Hispanic civilians),
the odds of a person fatally shot belonging to that group should
increase as well. Conversely, as the rate at which Whites commit
violent crime increases, the odds of a person fatally shot being
Black or Hispanic should decrease (because Whites serve as the
comparison group in our models). Finally, crime-rate changes
for the noncomparison minority group (Hispanics for Blacks and
Blacks for Hispanics) should not predict the race of a person
fatally shot.

We found strong support for these predictions, as the race
of a person fatally shot closely followed race-specific homicide
rates. As illustrated in Fig. 1, as the proportion of violent crime
committed by Black civilians increased, a person fatally shot was
more likely to be Black (OR = 3.66 [2.97, 4.51]). As the propor-
tion of violent crime committed by Hispanic civilians increased,
a person fatally shot was more likely to be Hispanic (OR = 3.34
[2.45, 4.56]). Conversely, as White crime rates increased, a per-
son fatally shot was less likely to be Black (OR = 0.28 [0.22, 0.37])
or Hispanic (OR = 0.29 [0.20, 0.41]). Finally, Hispanic crime
rates were unrelated to the odds of a person fatally shot being
Black (OR = 0.88 [0.66, 1.17]), and Black crime rates were unre-
lated to the odds of a person fatally shot being Hispanic (OR =
0.95 [0.73, 1.23]).

Race-specific violent crime was a very strong predictor of civil-
ian race, explaining 44% of the variance in the race of a person
fatally shot. This reveals that the race of a person who is fatally
shot closely tracks same-race violent crime, at least as indexed by
Centers for Disease Control and Prevention homicide data. We
largely replicated this pattern with population data (lower half of
Fig. 1). Race-specific population rates accounted for 43% of the
variance in civilian race, showing that the race of a person who is
fatally shot also closely tracks population size.

Hispanic % Population

Black % Population

White % Population

Hispanic % Crime

Black % Crime

White % Crime

6 5 4 3 2 1 2 3 4 5 6
Odds of Civilian Being White vs. Black/Hispanic

Hispanic

Black

Fig. 1. Odds ratios predicting the race of civilians fatally shot by police from
county-level race-specific violent crime (estimated by race-specific homicide
data) and population size. Values to the left (right) of the dotted line indi-
cate that the civilian was more likely to be White (Black/Hispanic). Civilian
race was regressed on each variable individually due to multicollinearity.
Lines represent 95% CI. n = 917.

Johnson et al. PNAS | August 6, 2019 | vol. 116 | no. 32 | 15879

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
,  July 10 2020

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental


Do Racial Disparities in FOIS Vary by Type of Shooting? We con-
ducted a set of regression models to test whether a person fatally
shot was more likely to be Black (or Hispanic) than White in
certain types of FOIS. In this set of models, we strategically
centered predictors to identify racial disparities across shoot-
ing types. When all predictors are centered or effects coded,
the test of the intercept indicates racial disparities in the aver-
age shooting. This provides an estimate of racial disparities
across all shootings. When categorical predictors are dummy-
coded so that zero represents the absence of the factor, model
intercepts reflect whether anti-Black and anti-Hispanic racial
disparity was observed for this type of shooting (e.g., unarmed
shootings). When continuous predictors are centered a SD below
the mean, model intercepts reflect whether anti-Black and anti-
Hispanic racial disparity was observed for this type of shooting
(e.g., shootings of young civilians). We tested racial disparities
across all types of shootings as defined by civilian and officer
factors.

Table 3 reports tests of racial disparities by type of shooting.
Model 0 tests whether there is evidence of racial disparity in
the typical shooting (all variables are centered or effects coded).
Controlling for predictors at the civilian, officer, and county lev-
els, a person fatally shot by police was 6.67 times less likely
(OR = 0.15 [0.09, 0.27]) to be Black than White and 3.33
times less likely (OR = 0.30 [0.21, 0.47]) to be Hispanic than
White. Thus, in the typical shooting, we did not find evidence
of anti-Black or anti-Hispanic disparity.

However, averaging across shootings may provide an incom-
plete picture if racial disparities vary across types of fatal shoot-
ings. The remaining models (1–20) separate different types of
shootings to test for this variation. No model showed significant
evidence of anti-Black or -Hispanic disparity, although evidence
for anti-Black and anti-Hispanic disparities was stronger when
civilians were young (Model 1 vs. 2). Evidence for anti-Black dis-
parities was also stronger when civilians were not suicidal (Model
7 vs. 8). Overall, there was considerable variation in racial

Table 3. Racial Disparity in Civilian Race by Shooting Type

Black Hispanic

Model and variable Level OR 95% CI OR 95% CI

0. Typical shooting — 0.15 0.08, 0.26 0.30 0.20, 0.46
1. Civilian age Low 0.25 0.14, 0.47 0.54 0.33, 0.88
2. Civilian age High 0.09 0.05, 0.16 0.17 0.10, 0.27
3. Civilian armed No 0.20 0.10, 0.37 0.28 0.15, 0.53
4. Civilian armed Yes 0.11 0.05, 0.24 0.32 0.17, 0.61
5. Civilian MH issue No 0.21 0.11, 0.39 0.47 0.25, 0.86
6. Civilian MH issue Yes 0.11 0.06, 0.20 0.19 0.11, 0.32
7. Civilian suicidal No 0.27 0.16, 0.48 0.27 0.16, 0.46
8. Civilian suicidal Yes 0.08 0.03, 0.18 0.33 0.17, 0.75
9. Civilian attacking No 0.10 0.03, 0.30 0.30 0.13, 0.68
10. Civilian attacking Yes 0.21 0.12, 0.36 0.30 0.17, 0.53
11. Officer number Low 0.14 0.08, 0.26 0.27 0.18, 0.41
12. Officer number High 0.15 0.08, 0.28 0.33 0.20, 0.56
13. Officer % Black Low 0.14 0.07, 0.26 0.29 0.18, 0.48
14. Officer % Black High 0.16 0.09, 0.29 0.31 0.19, 0.49
15. Officer % Hispanic Low 0.12 0.07, 0.22 0.24 0.15, 0.38
16. Officer % Hispanic High 0.18 0.10, 0.34 0.38 0.24, 0.59
17. Officer % women Low 0.14 0.08, 0.26 0.32 0.20, 0.51
18. Officer % women High 0.15 0.08, 0.28 0.28 0.17, 0.46
19. Officer experience Low 0.14 0.08, 0.26 0.30 0.18, 0.49
20. Officer experience High 0.15 0.08, 0.28 0.30 0.19, 0.49

Model 0 represents the typical shooting (all variables are effect coded
or centered). Models 1–20 are coded to indicate certain types of shootings.
Level indicates at what level of the variable racial disparity is tested. MH,
mental health. n = 917.

disparities (OR ranges from 0.09 to 0.54) across different types
of shootings.

Discussion
Concerns that White officers might disproportionately fatally
shoot racial minorities can have powerful effects on police legit-
imacy (31). By using a comprehensive database of FOIS during
2015, officer race, sex, or experience did not predict the race of
a person fatally shot beyond relationships explained by county
demographics. On the other hand, race-specific violent crime
strongly predicted the race of a civilian fatally shot by police,
explaining over 40% of the variance in civilian race. These results
bolster claims to take into account violent crime rates when
examining fatal police shootings (20).

We did not find evidence for anti-Black or anti-Hispanic dis-
parity in police use of force across all shootings, and, if anything,
found anti-White disparities when controlling for race-specific
crime. While racial disparity did vary by type of shooting, no
one type of shooting showed significant anti-Black or -Hispanic
disparity. The uncertainty around these estimates highlights the
need for more data before drawing conclusions about disparities
in specific types of shootings.

Policy Implications. Overall, officer demographics such as sex and
experience were not related to racial disparities in fatal shoot-
ings. Although officer race was related to racial disparities, the
fact that Black and Hispanic civilians were more likely to be
shot by same-race officers was largely explained by similarities
between officer and county demographics. Because racial dis-
parities in FOIS do not vary based on officer race, hiring more
diverse officers may not reduce racial disparities in FOIS. This
is not to say that increasing officer diversity is without merit, as
increasing officer diversity may broaden understanding of diverse
communities and increase trust in law enforcement. However,
these data suggest that increasing racial diversity would not
meaningfully reduce racial disparity in fatal shootings (32).

One of our clearest results is that violent crime rates strongly
predict the race of a person fatally shot. At a high level, reducing
race-specific violent crime should be an effective way to reduce
fatal shootings of Black and Hispanic adults. Of course, this is no
simple task—crime rates are the result of a large and dynamic
set of forces. However, the magnitude of these disparities speaks
to the importance of this idea. In counties where minorities com-
mitted higher rates of violent crime, a person fatally shot was
3.3 times more likely to be Hispanic than White and 3.7 times
more likely to be Black than White. This suggests that reduc-
ing disparities in FOIS will require identifying and changing
the socio-historical factors that lead civilians to commit violent
crime (20).

One limitation of our results is that they only focus on officers
who fired at a civilian that was fatally wounded. Not all officers
responding to such calls fire their weapons. Therefore, charac-
teristics such as officer race, sex, or experience may impact racial
disparities in FOIS through whether officers fire their weapons.
Testing this will require additional information about responding
officers who do not fire their weapons.

What Is the Evidence for Racial Disparity? When considering all
FOIS in 2015, we did not find anti-Black or anti-Hispanic dis-
parity. How do we explain these results? Our data are consistent
with three possible explanations.

One police-centered explanation is that these disparities
reflect depolicing (33, 34). Depolicing occurs when police offi-
cers’ concerns about becoming targets in civil litigation and the
media spotlight impede officers from enforcing the law. Such
concerns have been heightened due to recent high-profile shoot-
ings of Black men (35). The disparities in our data are consistent
with selective depolicing, where officers are less likely to fatally

15880 | www.pnas.org/cgi/doi/10.1073/pnas.1903856116 Johnson et al.

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
 ,  July 10 2020

https://www.pnas.org/cgi/doi/10.1073/pnas.1903856116


P
SY

C
H

O
LO

G
IC

A
L

A
N

D
C

O
G

N
IT

IV
E

SC
IE

N
C

ES

shoot Black civilians for fear of public and legal reprisals. All
else equal, this would increase the likelihood that a person fatally
shot was White vs. Black. However, depolicing might be limited
to areas with high-profile shootings (36). This explanation also
does not explain the disparity observed when comparing White
and Hispanic civilians. Future research could test for depolic-
ing more rigorously by using a quasiexperimental time-lagged
study investigating police use of force in cities before and after
high-profile shootings where racial issues are prominent.

On the other hand, a civilian-centered explanation for these
disparities is that White civilians may react differently toward
police than racial minorities in crime-related situations. If White
civilians present more threat toward police, this could explain
why a person fatally shot was more likely to be White than Black
or Hispanic. Among those fatally shot by police, Whites are more
likely (relative to racial minorities) to be armed and pose a threat
(26). We attempted to control for civilian threat level by mea-
suring whether they were armed and attacking, but found these
variables unrelated to the race of a person fatally shot. These
issues illustrate a broader challenge in inferring civilian char-
acteristics during fatal shootings. The newspaper databases we
analyzed contained at least some errors (e.g., in whether civil-
ians are coded as armed; ref. 37). There are likely more false
positives and negatives in these databases, such as when sepa-
rating individuals committing suicide who are not experiencing
a mental health crisis from those who are experiencing a men-
tal health crisis. Another challenge is that dichotomous variable
codes may not capture the complexity of these interactions (e.g.,
a person is coded as attacking, but they had stopped struggling
before they were fatally shot). One solution is to code civilian
threat level in a more continuous way (e.g., ref. 10). But this
will only be realistic if better records of FOIS are kept at the
federal level. For this reason, we urge caution when interpreting
the impact of civilian characteristics on racial disparities in fatal
shootings.

Finally, the lack of anti-Black or anti-Hispanic disparity and
the impact of race-specific crime are consistent with an exposure
argument, whereby per capita racial disparity in fatal shootings is
explained by non-Whites’ greater exposure to the police through
crime. This explanation is consistent with studies that have used
violent crime as a benchmark for testing disparity (20, 23–25).
However, this does not mean that researchers should continue to
use benchmarking approaches, even if using violent crime over
population size. Rather, researchers can take one or both pre-
dictors into account with our approach. Moreover, unlike the
benchmark approach, our conclusions regarding racial disparity
do not depend on which predictors are used (SI Appendix).

What These Findings Do Not Show. Our analyses test for racial
disparities in FOIS, which should not be conflated with racial
bias (21). Racial disparities are a necessary but not sufficient,
requirement for the existence of racial biases, as there are many
reasons why fatal shootings might vary across racial groups that
are unrelated to bias on the behalf of police officers.

For example, we found that a person fatally shot by police
was much more likely to be White when they were suicidal.
This does not mean that there are department policies or officer
biases that encourage fatal shootings of suicidal White civilians.
A more plausible explanation is that White civilians are more
likely to attempt “suicide by cop” than minorities (38). Similarly,
Black and Hispanic officers (compared with White officers) were
more likely to fatally shoot Black and Hispanic civilians. This
does not mean that there are department policies encouraging
non-White officers to fatally shoot minorities. Rather, the link
between officer race and FOIS appears to be explained by offi-
cers and civilians being drawn from the same population, making
it more likely that an officer will be exposed to (and fatally shoot)
a same-race civilian.

We stress that these findings cannot incriminate or exonerate
officers in any specific case. Findings at the national level do not
directly speak to the presence or absence of bias in individual
shootings. In other words, whether a particular officer shows bias
in any individual case is a different question than whether officers
in general show bias. Claims of national bias in FOIS requires
examining fatal force in aggregate, and not just in one incident
or racial group (39).

Conclusion. Until now, researchers have been unable to test ques-
tions related to officer characteristics in fatal shootings. We
created a near-complete database of fatal shootings in 2015 to
test questions about racial disparities in FOIS. However, contin-
ued work on this issue will require more information about the
officers, civilians, and circumstances surrounding these events.
We encourage federal agencies to enforce policies that require
recording information about the civilians and officers in FOIS
to better understand the relationship between civilian race and
police use of force.

Materials and Methods
We began by creating a list of all 2015 FOIS of civilians by nonfederal
on-duty police officers, as this was the first year that news organizations
collected near-complete databases of FOIS. We obtained this initial list of
civilians by combining information from The Washington Post and The
Guardian databases on January 1, 2016. We limited our analyses to White
(n = 501), Black (n = 245), and Hispanic (n = 171) civilians, because there were
insufficient data to analyze other racial groups. The institutional review
board at Michigan State University deemed this study exempt, as it relied
on public information.

We next obtained officer information by contacting all 684 police depart-
ments who had officers involved in a fatal shooting. We initially sent letters
requesting the race, sex, and years of experience of each officer who fired
at the civilian. From this written request, we received information on 62%
of shootings. We next called police departments to request missing data.
Finally, we searched newspaper articles, court documents, and internet
sources to supplement the missing data. In all, we were able to obtain
complete officer information in 72% of shootings and partial information
in 96% of shootings. Rather than remove shootings with missing informa-
tion, we estimated the missing data with multiple imputation (ref. 40; SI
Appendix).

We included several factors to predict the race of a person fatally shot.
Officer characteristics included the total number of officers who fired in
the shooting, the percent of officers who were Black or Hispanic, the
percent of female officers, and the average experience across the offi-
cers in years. Civilian characteristics included civilian age and whether they
were armed, suffering from a mental health issue, suicidal, or attacking
the officer. County-level factors included county population size, median
income, income inequality, percent of the county that was urban, and
race-specific violent crime rates. Details and correlations are provided in
SI Appendix.

In defining what constitutes a mental health issue, we relied on The
Washington Post’s coding, which indicates whether the person was expe-
riencing a mental health crisis or if there was no known crisis. The Post
does not specify the nature of the crisis. We also used the Post’s coding
of whether an individual is armed. We used newspaper reports to code that
a civilian was suicidal if 1) they left an explicit suicide note; 2) a family mem-
ber reported the civilian was suicidal; or 3) police reported that the civilian
explicitly told officers to shoot him or her. We also used newspaper reports
to code civilians as attacking if they were armed or actively struggling with
an officer. Behaviors such as fleeing or advancing toward an officer were
not coded as attacking. More details about these codes are provided in SI
Appendix.

All multinomial regression models were estimated with MPlus (Version
8.0; ref. 41). Whites served as the referent category relative to Black and His-
panic civilians. We used clustering to correct standard errors due to county-
level nonindependence. Details can be found in SI Appendix. Estimates for
each predictor were converted to OR to facilitate interpretation.

ACKNOWLEDGMENTS. D.J.J. is a postdoctoral researcher in the Laboratory
for Applied Social Science Research at the University of Maryland. We thank
the police departments who completed our information requests for their
cooperation. We also thank William Chopik, David Clark, and William Terrill
for their helpful comments on earlier drafts of this manuscript.

Johnson et al. PNAS | August 6, 2019 | vol. 116 | no. 32 | 15881

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
 ,  July 10 2020

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental
https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1903856116/-/DCSupplemental


1. President’s Task Force on 21st Century Policing, Final Report of the President’s Task
Force on 21st Century Policing (Office of Community Oriented Policing Services,
Washington, DC, 2015).

2. A. C. Davis, W. Lowery, FBI director calls lack of data on police shootings “ridicu-
lous,” “embarrassing.“ The Washington Post, 7 October 2015. https://www.
washingtonpost.com/national/fbi-director-calls-lack-of-data-on-police-shootings-
ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16-11e5-b31c-d80d62b53e28 story.
html?utm term=.fffb4215364f.

3. Data from “Shot dead by police.” The Washington Post. https://www.
washingtonpost.com/graphics/national/police-shootings/. Accessed 1 January 2016.

4. Vice News, Shot by cops and forgotten. Vice News, 11 December 2017. https://news.
vice.com/en ca/article/xwvv3a/shot-by-cops.

5. C. M. Donner, N. Popovich, Hitting (or missing) the mark: An examination of police
shooting accuracy in officer-involved shooting incidents. Policing Int. J. 42, 474–489
(2018).

6. G. B. Morrison, Deadly force programs among larger US police departments. Police
Q. 9, 331–360, 2006.

7. R. J. Friedrich, Police use of force: Individuals, situations, and organizations. Ann. Am.
Acad. Polit. Soc. Sci. 452, 82–97 (1980).

8. J. H. Garner, T. Schade, J. Hepburn, J. Buchanan, Measuring the continuum of force
used by and against the police. Crim. Justice Rev. 20, 146–168 (1995).

9. J. P. McElvain, A. J. Kposowa, Police officer characteristics and the likelihood of using
deadly force. Crim. Justice Behav. 35, 505–521 (2008).

10. W. Terrill, S. D. Mastrofski, Situational and officer-based determinants of police
coercion. Justice Q. 19, 215–248 (2002).

11. R. E. Worden, The “causes” of police brutality: Theory and evidence on police use
of force. in, Criminal Justice Theory: Explaining the Nature and Behavior of Crimi-
nal Justice, E. R. Maguire, D. E. Duffee, eds. (Routledge, New York, NY, ed. 2, 2015),
chap. 6.

12. W. A. Geller, K. J. Karales, Shootings of and by Chicago police: Uncommon crises—
Part I: Shootings by Chicago police. J. Crim. L. Criminol. 72, 1813–1866 (1981).

13. C. E. Menifield, G. Shin, L. Strother, Do white law enforcement officers target
minority suspects? Public Adm. Rev. 79, 56–68 (2019).

14. Data from “The counted.” The Guardian. https://www.theguardian.com/us-news/
ng-interactive/2015/jun/01/the- counted-police-killings- us-database#. Accessed 1 Jan-
uary 2016.

15. E. Hehman, J. K. Flake, J. Calanchini, Disproportionate use of lethal force in policing
is associated with regional racial biases of residents. Soc. Psychol. Personal. Sci. 9,
393–401 (2018).

16. D. Jacobs, The determinants of deadly force: A structural analysis of police violence.
Am. J. Sociol. 103, 837–862 (1998).

17. C. T. Ross, A multi-level Bayesian analysis of racial bias in police shootings at the
county-level in the United States, 2011–2014. PLoS One 10, e0141854 (2015).

18. K. Scott, D. S. Ma, M. S. Sadler, J. Correll, A social scientific approach toward under-
standing racial disparities in police shooting: Data from the Department of Justice
(1980–2000). J. Soc. Issues 73, 701–722 (2017).

19. P. Takagi, A Garrison state in “democratic” society. Crime Soc. Justice 1, 27–33
(1974).

20. J. Cesario, D. J. Johnson, W. Terrill, Is there evidence of racial disparity in police use of
deadly force? Analyses of officer-involved fatal shootings in 2015–2016. Soc. Psychol.
Personal. Sci. 10, 586–595 (2018).

21. B. Tregle, J. Nix, G. P. Alpert, Disparity does not mean bias: Making sense of observed
racial disparities in fatal officer-involved shootings with multiple benchmarks. J.
Crime Justice 42, 18–31 (2019).

22. J. Nix, B. A. Campbell, E. H. Byers, G. P. Alpert, A bird’s eye view of civilians killed by
police in 2015: Further evidence of implicit bias. Criminol. Public Policy 16, 309–340
(2017).

23. J. J. Fyfe, Blind justice: Police shootings in Memphis. J. Crim. L. Criminol. 73, 707–722
(1982).

24. P. A. Goff, T. Lloyd, A. Geller, S. Raphael, J. Glaser, “The science of justice: Race, arrests,
and police use of force” (Center for Policing Equity, New York, NY, 2016; https://
policingequity.org/what-we-do/research/the-science-of-justice-race-arrests-and-
police-use-of-force).

25. A. Inn, A. C. Wheeler, C. L. Sparling, The effects of suspect race and situation hazard
on police officer shooting behavior. J. Appl. Soc. Psychol. 7, 27–37 (1977).

26. S. DeGue, K. A. Fowler, C. Calkins, Deaths due to use of lethal force by law enforce-
ment: Findings from the national violent death reporting system, 17 US states,
2009–2012. Am. J. Prev. Med. 51, S173–S187 (2016).

27. V. B. Lord, M. W. Sloop, Suicide by cop: Police shooting as a method of self-harming.
J. Crim. Justice 38, 889–895 (2010).

28. K. D. Kochanek, S. L. Murphy, J. Xu, B. Tejada-Vera, Deaths: Final data for 2014. Natl.
Vital. Stat. Rep. 65, 1–121 (2016).

29. B. A. Reaves, “Local police departments, 2013: Personnel, policies, and practices”
(NCJ 248677, Bureau of Justice Statistics, Washington, DC, 2015; https://www.bjs.
gov/index.cfm?ty=pbdetail&iid=5279).

30. J. Berger, Still on patrol after two decades, valued but rare. The New York Times, 13
December 2011. https://www.nytimes.com/2011/12/14/nyregion/most-police-officers-
retire-after-20-years-or-move-up-the-ranks.html.

31. T. R. Kochel, Explaining racial differences in Ferguson’s impact on local residents’ trust
and perceived legitimacy: Policy implications for police. Crim. Justice Policy Rev. 30,
374–305 (2017).

32. B. W. Smith, The impact of police officer diversity on police-caused homicides. Policy
Stud. J. 31, 147–162 (2003).

33. W. M. Oliver, Depolicing: Rhetoric or reality? Crim. Justice Policy Rev. 28, 437–461
(2017).

34. S. Dewan, Deconstructing the “Ferguson Effect.” The New York Times, 29 March 2017.
https://www.nytimes.com/interactive/2017/us/politics/ferguson-effect.html.

35. R. Rosenfeld, Documenting and explaining the 2015 homicide rise: Research
directions (NCJ 249895, National Institute of Justice, Washington, DC, 2016;
https://www.ncjrs.gov/App/Publications/abstract.aspx?ID=272055).

36. D. C. Pyrooz, S. H. Decker, S. E. Wolfe, J. A. Shjarback, Was there a Ferguson Effect on
crime rates in large US cities? J. Crim. Justice 46, 1–8 (2016).

37. D. A. Klinger, L. A. Slocum, Critical assessment of an analysis of a journalistic com-
pendium of citizens killed by police gunfire. Criminol. Public Policy 16, 349–362
(2017).

38. H. R. Hutson et al., Suicide by cop. Ann. Emerg. Med. 32, 665–669 (1998).
39. K. Downs, Why aren’t more people talking about Latinos killed by police? PBS

News Hour, 14 July 2016. https://www.pbs.org/newshour/nation/black-men-werent-
unarmed-people-killed-police-last-week.

40. C. K. Enders, Applied Missing Data Analysis (Guilford, New York, NY, 2010).
41. L. K. Muthén, B. Muthén, Mplus User’s Guide (Muthén and Muthén, Los Angeles, CA,

ed. 8, 1998-2017).

15882 | www.pnas.org/cgi/doi/10.1073/pnas.1903856116 Johnson et al.

D
o
w

n
lo

a
d
e
d
 b

y 
g
u
e
st

 o
n
 J

u
ly

 7
, 
2
0
2
0
 

See Retraction Published 
 ,  July 10 2020

https://www.washingtonpost.com/national/fbi-director-calls-lack-of-data-on-police-shootings-ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16-11e5-b31c-d80d62b53e28_story.html?utm_term=.fffb4215364f
https://www.washingtonpost.com/national/fbi-director-calls-lack-of-data-on-police-shootings-ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16-11e5-b31c-d80d62b53e28_story.html?utm_term=.fffb4215364f
https://www.washingtonpost.com/national/fbi-director-calls-lack-of-data-on-police-shootings-ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16-11e5-b31c-d80d62b53e28_story.html?utm_term=.fffb4215364f
https://www.washingtonpost.com/national/fbi-director-calls-lack-of-data-on-police-shootings-ridiculous-embarrassing/2015/10/07/c0ebaf7a-6d16-11e5-b31c-d80d62b53e28_story.html?utm_term=.fffb4215364f
https://www.washingtonpost.com/graphics/national/police-shootings/
https://www.washingtonpost.com/graphics/national/police-shootings/
https://news.vice.com/en_ca/article/xwvv3a/shot-by-cops
https://news.vice.com/en_ca/article/xwvv3a/shot-by-cops
https://www.theguardian.com/us-news/ng-interactive/2015/jun/01/the-counted-police-killings-us-database#
https://www.theguardian.com/us-news/ng-interactive/2015/jun/01/the-counted-police-killings-us-database#
https://policingequity.org/what-we-do/research/the-science-of-justice-race-arrests-and-police-use-of-force
https://policingequity.org/what-we-do/research/the-science-of-justice-race-arrests-and-police-use-of-force
https://policingequity.org/what-we-do/research/the-science-of-justice-race-arrests-and-police-use-of-force
https://www.bjs.gov/index.cfm?ty=pbdetail&iid=5279
https://www.bjs.gov/index.cfm?ty=pbdetail&iid=5279
https://www.nytimes.com/2011/12/14/nyregion/most-police-officers-retire-after-20-years-or-move-up-the-ranks.html
https://www.nytimes.com/2011/12/14/nyregion/most-police-officers-retire-after-20-years-or-move-up-the-ranks.html
https://www.nytimes.com/interactive/2017/us/politics/ferguson-effect.html
https://www.ncjrs.gov/App/Publications/abstract.aspx?ID=272055
https://www.pbs.org/newshour/nation/black-men-werent-unarmed-people-killed-police-last-week
https://www.pbs.org/newshour/nation/black-men-werent-unarmed-people-killed-police-last-week
https://www.pnas.org/cgi/doi/10.1073/pnas.1903856116