Comparing Safety Outcomes in Police Use-Of-Force Cases for Law Enforcement Agencies, NIJ, 2012

• Feb. 24, 2016 • Locations: United States of America • Topics: Statistics/Trends, Stun Guns/Tasers

Download original document:
Brief thumbnail

This text is machine-read, and may contain errors. Check the original document to verify accuracy.

The author(s) shown below used Federal funds provided by the U.S.
Department of Justice and prepared the following final report:

Document Title:

Comparing Safety Outcomes in Police Use-OfForce Cases for Law Enforcement Agencies
That Have Deployed Conducted Energy Devices
and A Matched Comparison Group That Have
Not: A Quasi-Experimental Evaluation

Author:

Bruce Taylor, Ph.D., Daniel Woods, Bruce Kubu,
Chris Koper, Ph.D., Bill Tegeler, Jason Cheney,
Mary Martinez, James Cronin, Kristin
Kappelman

Document No.:

237965

Date Received:

March 2012

Award Number:

2006-IJ-CX-0028

This report has not been published by the U.S. Department of Justice.
To provide better customer service, NCJRS has made this Federallyfunded grant final report available electronically in addition to
traditional paper copies.

Opinions or points of view expressed are those
of the author(s) and do not necessarily reflect
the official position or policies of the U.S.
Department of Justice.

Comparing safety outcomes in
police use-of-force cases for
law enforcement agencies that have
deployed Conducted Energy Devices
and a matched comparison group
that have not:
A quasi-experimental evaluation
September 2009
Report submitted to the National Institute of Justice
PERF Project Staff:
Bruce Taylor, Ph.D., Research Director | Daniel Woods, Associate |
Bruce Kubu, Senior Associate | Chris Koper, Ph.D., Deputy Research Director |
Bill Tegeler, Deputy Director, Management Services | Jason Cheney, Associate |
Mary Martinez, Associate | James Cronin, Senior Associate | Kristin Kappelman, Associate

This document is a research report submitted to the U.S. Department of Justice. This report has not
been published by the Department. Opinions or points of view expressed are those of the author(s)
and do not necessarily reflect the official position or policies of the U.S. Department of Justice.

September 2009
Report submitted to the National Institute of Justice

PERF Project Staff:
Bruce Taylor, Ph.D., Research Director
Daniel Woods, Associate
Bruce Kubu, Senior Associate
Chris Koper, Ph.D., Deputy Research Director
Bill Tegeler, Deputy Director, Management Services
Jason Cheney, Associate
Mary Martinez, Associate
James Cronin, Senior Associate
Kristin Kappelman, Associate

This study was funded by the National Institute of Justice (Grant # 2006-IJ-CX-0028).
The views expressed are those of the authors and do not necessarily represent the views
or the official position of the National Institute of Justice or any other organization.
Police Executive Research Forum, Washington, D.C. 20036
Copyright 2009 by Police Executive Research Forum
All rights reserved
Edited by Craig Fischer
Cover and interior design by Dave Williams

Table of Contents

Abstract ......................................................................................1
Executive Summary ...................................................................2
Chapter 1: Introduction .............................................................9
Chapter 2: Literature Review...................................................12
Chapter 3: Research Design and Methods ...............................21
Chapter 4: Study Results..........................................................35
Chapter 5: Discussion and Conclusion ....................................59
References................................................................................72
Appendices...............................................................................79

Abstract

ow law enforcement agencies (LEAs)
manage the use-of-force by officers is
perhaps one of the most important tasks that
they will undertake. One weapon that has
been advanced as a way to reduce injuries
for officers and suspects is the Conducted
Energy Device (CED). The purpose of our
project, conducted from late 2006 to 2008,
was to produce scientifically valid results
that will inform LEA executives’ decisions
regarding CED use. The goal of our study
was to produce practical information that
can help LEAs establish guidelines that
assist in the effective design of CED
deployment programs that support increased
safety for officers and citizens. We
conducted one of the first quasi-experiments
to compare LEAs with CED deployment
(n=7) to a set of matched LEAs (n=6) that
do not deploy CEDs on a variety of safety
outcomes, controlling for a variety of
incident factors (force used by officer, time
frame of incident, suspect race/gender/age,
suspect resistant behavior, and suspect
weapon use) and agency-level factors
(agency policy on CEDs, size/density of
LEA, and population density for
jurisdiction). For the LEAs that deployed
CEDs, we collected two years of data before
CED deployment and two years of data after
CED deployment. For the non-CED sites,
we collected four years of data over a
similar period.
Overall, we found that the CED sites
were associated with improved safety
outcomes when compared to a group of
matched non-CED sites on six of nine safety
measures, including reductions in (1) officer

injuries, (2–3) suspect injuries and severe
injuries, (4–5) officers and suspects
receiving injuries requiring medical
attention, and (6) suspects receiving an
injury that resulted in the suspect being
taken to a hospital or other medical facility.
(We refer to this last category as
“hospitalization,” although we have no data
on the extent to which officers or suspects
who went to a hospital or other medical
facility were admitted and stayed overnight,
as opposed to simply receiving an evaluation
or treatment and being released.)
Also within CED agencies, in some
cases the actual use of a CED by an officer
is associated with improved safety outcomes
compared to other less-lethal weapons. For
five of the eight comparisons, the cases
where an officer used a CED were
associated with the lowest or second lowest
rate of injuries, injuries requiring medical
attention, or injuries officer was taken to a
medical facility such as hospital or medical
clinic for treatment of an injury due to a useof-force incident requiring “hospitalization”
(see comment in previous paragraph). There
were no differences between the CED and
the non-CED sites on the outcomes of the
number of suspect deaths, officer severe
injuries, and officer injuries requiring
hospitalization.
The evidence from our study suggests
that CEDs can be an effective weapon in
helping prevent or minimize physical
struggles in use-of-force cases. LEAs should
consider the utility of the CED as a way to
avoid up-close combative situations and
reduce injuries to officers and suspects.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Executive Summary

he management of police officers’ use
of force is perhaps one of the most
important tasks that a law enforcement
agency (LEA) will undertake. LEA
executives have to make important policy
decisions on the types of force that will be
authorized, technologies to deliver that
force, and when and how often various types
of force can be used. One of the key
objectives in managing force is designing
approaches to reduce incidents of police use
of force and the injuries associated with
force. One weapon that has been advanced
as a way to reduce injuries for officers and
suspects is the Conducted Energy Device
(CED). Law enforcement executives have
been overwhelmed with questions about the
effectiveness of CEDs and the safety of
these devices. The lack of available
information and a full understanding of the
effects of using CEDs has hampered the
ability of police executives to make
informed policy decisions about these
devices. Police executives have been
provided with little independent scientific
evidence and guidance on the impact of
using CEDs. While decades of research have
documented the nature and extent of the
force used by police and the conditions and
correlates that affect its application (Smith et
al., 2007), little research has been done
isolating the effects of using CEDs on
injuries to suspects and officers.

Project purpose, goals
and objectives:
The purpose of our project was to produce
scientifically valid results that will inform
LEA executives’ decisions regarding the use
of CEDs. The goal of our study was to
produce practical information that can help
law enforcement executives make good
decisions about whether to deploy CEDs,
and if a decision is made to deploy them, to
help the agencies develop CED policy and
procedural guidelines that provide increased
safety for officers and citizens. In order to
accomplish this goal, our objective was to
conduct an evaluation comparing LEAs that
have deployed CEDs to a matched group of
LEAs that have not deployed CEDs in terms
of officer and suspect safety during use-offorce incidents.

Research design:
Our team used a quasi-experimental design
(QED) to compare departments with CED
deployment (n=7) to a set of matched
departments (n=6) that do not deploy CEDs
on a variety of outcomes. With our QED, we
are able to isolate the safety outcomes to be
expected if a department deploys CEDs,
controlling for a variety of related
organizational and individual/incident-level
factors.
A key element for all QEDs is the
process used to select a comparison group.
In our study, we used a matching design.
CED (n=7) and non-CED (n=6) sites were
matched based on violent crime levels,

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

police activity, agency size, and population
size of jurisdiction. The inclusion of 13
departments allows us not only to assess
incident-level factors, but also some
important departmental/organizational-level
factors that could affect outcomes. Our
study is one of the first to examine LEAs
that use CEDs to matched LEAs that do not
use CEDs.
We collected four years of data on all
incidents of use of force for all of the
participating departments. For the LEAs that
deployed CEDs, we collected at least two
years of data before CED deployment and
two years of data after CED deployment.
For the LEAs that did not deploy CEDs, we
collected at least four years of data over a
similar period. While the focus of our study
was on the use CEDs, we also collected data
on all use-of-force incidents (not just CED
cases) and examined the range of weapons
and unarmed tactics that the police employ
in exerting force to arrest suspects.

Site participants:
Our selection of cities was based on a
matching analysis using a PERF nationally
representative survey on use of force. We
obtained our data from seven sites that have
deployed CEDs and six sites that have not
deployed CEDs.
Overall, we believe our CED and nonCED sites are comparable. We collected
data from fairly comparable periods for the
CED and non-CED sites, within a year or
two. And while some differences emerged in
our assessment of the comparability of our
CED and non-CED sites, most of the
differences were relatively small and did not
seem to introduce any substantively
important biases. When combined with our
multivariate analyses, we believe that we
have a reasonably comparable group of CED
and non-CED sites with results that are
interpretable.

Data analytic approach:
We conducted a series of analyses
comparing CED and non-CED sites,
including bivariate analyses to describe the
basic raw differences between the CED and
non-CED sites on our outcome measures,
and a variety of multivariate analyses to
attempt to assess the viability of the
bivariate results and control for possible
alternative explanations for the earlier raw
differences. Our first multivariate analyses
were done using logistic regression to isolate
the effects of CED deployment on our
safety-related outcomes where we included
the following independent/control variables:
Whether the agency deploys CEDs, the time
frame of the incident, an interaction of CED
multiplied by time-frame, suspect race,
suspect gender, suspect age, whether the
suspect used resistant behavior, and whether
the suspect had a weapon at the force
incident.
One of the concerns with examining
multi-site data is that the individual use-offorce cases we analyze are clustered within
13 departments. In our study, individual
cases of weapon use by officers are nested
within specific police departments that have
various policy guidelines on the use of force.
Ignoring the nested structure of our data can
potentially lead to biased estimates. To
address this clustering issue we used two
approaches. First, we conducted a modified
logistic regression with a robust variance
estimator to adjust for within-cluster
correlation. However, with this approach we
do not get aggregate-level coefficients to see
the exact effects of aggregate-level
conditions on our individual results. To
examine and observe the effects of
aggregate-level factors, we conducted a
multi-level analysis using Hierarchical
Linear Modeling (HLM). While we
recognize our limited statistical power to
conduct HLM analyses (n=13 LEAs), we are

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

mainly using HLM to assess the robustness
of our findings from our earlier analyses and
take an initial step at assessing the possible
problem of aggregate-level nesting. We
focus our analyses of the HLM results on the
direction and magnitude of the effects (as
opposed to a focus on the statistical
significance of the results).

Study Results:
Overall, we found that the CED sites were
associated with improved safety outcomes
when compared to a group of matched nonCED sites on six of nine safety measures,
including reductions in:
• Officer injuries
• Suspect injuries
• Suspect severe injuries
• Officers receiving injuries requiring
medical attention,
• Suspects receiving injuries requiring
medical attention, and
• Suspects receiving an injury that resulted
in their being sent to a hospital or other
medical facility. (We refer to this as
“hospitalization,” but it does not
necessarily mean that suspects were
admitted and stayed overnight at a
hospital; we were unable to obtain data on
the extent to which officers or suspects
who went to a hospital or other medical
facility were admitted and stayed
overnight, as opposed to simply receiving
an outpatient evaluation and/or
treatment.)
There were no differences between the
CED and the non-CED sites on the
outcomes of the other three measures:
number of suspect deaths, officer severe
injuries, and officer injuries requiring
hospitalization.
For the six of nine significant
outcomes, our data suggest that the

magnitude of the effects of the improved
safety outcomes for the CED sites (relative
to the non-CED sites) was impressive. We
found a strong effect of CEDs on reducing
officer injuries based on our raw results (8%
officer injuries in the post period to 20% for
the non-CED sites), and our three
multivariate models. For agencies that
deploy CEDs, our data suggest that the odds
of an officer being injured are reduced by
over 70%. Also, for our CED-only site
analyses, when officers actually use CEDs
our data suggest that there is a 76%
reduction in officer injuries. Similar
reductions were observed for the CED sites
on our measure of suspect injuries, as
confirmed by our raw results (26% suspect
injuries in the post period to 43% for the
non-CED sites), and our three multivariate
models. For an agency that deploys CEDs,
our data suggest that the odds of a suspect
being injured are reduced by more than
40%.
Along the same lines, our data suggest
that CED sites were related to reductions in
suspect severe injuries based on our raw
results (5% suspect severe injuries in the
post period to 7% for the non-CED sites),
and our three multivariate models. For an
agency that deploys CEDs, our data suggest
that the odds of a suspect being severely
injured are reduced by over 40%. For our
CED-only site analyses, our data suggest
that CEDs were associated with the lowest
levels of suspect severe injuries compared to
other forms of force.
Our data suggest that CED sites were
related to reductions in injuries to officers
requiring medical attention based on our
raw results (8% for officer medical attention
in the post period to 16% for the non-CED
sites), and our three multivariate models. For
an agency that deploys CEDs, our data
suggest that the odds of an officer receiving
an injury requiring medical attention is
reduced by at least 80%. For our CED-only

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

site analyses, when officers actually use
CEDs our data suggest that there is a 63%
reduction in the probability of an officer
receiving an injury requiring medical
attention.
Similarly, our data suggest that CED
sites were related to reductions in injuries to
suspects requiring medical attention based
on our raw results (40% for suspect medical
attention in the post period to 53% for the
non-CED sites) and our three multivariate
models. For an agency that deploys CEDs,
our data suggest that the odds of a suspect
receiving an injury requiring medical
attention in the post period is reduced by
more than 45%.
Our data suggest that CED sites were
related to reductions in injuries to suspects
requiring hospitalization (defined as being
sent to a hospital, clinic, or other medical
facility for evaluation or treatment, not
necessarily being admitted for an overnight
stay) based on our raw results (16% for
suspect medical attention in the post period
to 36% for the non-CED sites), and our three
multivariate models. For agencies that
deploy CEDs, our data suggest that the odds
of a suspect receiving an injury requiring
hospitalization in the post period is reduced
by 52% for the logistic regression model or
only 11% for the HLM models relative to
agencies without CEDs. While there is a
wide gap in these estimates, both models
suggest that CED sites are associated with a
reduced probability of suspects receiving
injuries requiring hospitalization. For our
CED-only site analyses, our data suggest
that CEDs (30%) had the highest levels of
suspects receiving injuries requiring
hospitalization. Our data suggest that when
officers use CEDs there was a 139%
increase in the probability of a suspect
receiving injuries requiring hospitalization
(0.87, p<.001). This may reflect an informal
police practice of sending suspects who have
been subjected to a CED activation to a

hospital as a precautionary measure—for
example, to ensure that the skin punctures
caused by the CED darts do not become
infected. PERF’s guidelines for use of
CEDs, for example, developed in 2005 with
support from the U.S. Justice Department,
recommend that “all persons who have been
exposed to a CED activation should receive
a medical evaluation.” (See further
discussion of this in Chapter 5, “Discussion
and Conclusion.”) While overall, the CED
sites led to better outcomes than the nonCED sites on this measure, this result needs
to be explored further in future research.
Another concern raised by critics of
CEDs is that they may lead to higher death
rates for agencies that deploy CEDs. We
found no support for this concern. CEDs
seem to have a neutral effect on the number
of suspect deaths related to officer use-offorce cases. Before implementation of
CEDs, our data suggest that the CED sites
had less than one percent of their cases
(0.2%) involving a suspect killed by an
officer. After CED implementation, our data
suggest that this number remained about the
same statistically (0.4%). During the same
period, our data suggest that the non-CED
sites did not change either statistically. The
non-CED sites observed about one percent
of their cases (0.9%) involving a suspect
killed by an officer at the pre-test period,
and observed no change in the number of
suspects killed in force incidents at the postperiod (0.9%). Our data suggest that we
basically have a flat line for the CED sites
(0.2% to 0.4%) and a flat line for the nonCED sites (0.9% at both time points). On
balance, our study did not reveal a
significant effect of CEDs on suspect deaths,
but with a sample of only 44 suspect deaths
we do not have a high level of statistical
power to uncover statistically significant
findings.
While our study did not reveal
evidence of higher death rates for agencies

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

that use CEDs, concerns still remain
regarding a number of deaths that have
followed use of CEDs. One of the most
recent and influential studies of deaths
following CED use, conducted by a highlevel panel of medical experts for the
National Institute of Justice (NIJ) and
released in 2008, found that “the purported
safety margins of CED deployment on
normal healthy adults may not be applicable
in small children, those with diseased hearts,
the elderly, those who are pregnant, and
other at-risk individuals,” and that “the
medical risks of repeated or continuous CED
exposure are unknown and the role of CEDs
in causing death is unclear in these cases.”
The NIJ panel also found that not all of the
people who have died after being subjected
to a CED activation were chemically
dependent or had heart disease or mental
illness; “some were normal healthy adults.”
Additional research should be conducted to
explore these issues.
All in all, our data suggest that we
found consistently strong effects for CEDs
on increasing officer and suspect safety. Not
only are CED sites associated with improved
safety outcomes compared to a matched
group of non-CED sites, but also within
CED agencies, in some cases the actual use
of a CED by an officer is associated with
improved safety outcomes compared to use
of other less-lethal weapons. For five of the
eight comparisons, the cases where an
officer uses a CED were associated with the
lowest or second lowest rate of injury,
injuries requiring medical attention, or
injuries requiring hospitalization.

Implications of PERF results:
As other researchers have generally found in
use-of-force studies, we found that most of
our cases involved low levels of force and
few if any injuries. However, our study also
documented an important number of cases

when officers had to use more force to gain
control of a noncompliant suspect and take
the person to the ground. These types of
ground struggles carry an increased risk of
injury for officers and suspects. According
to our results, police equipment that allows
officers to avoid these up-close struggles,
such as CEDs and OC spray, hold the
promise of preventing injuries for officers as
well as suspects. These findings are
consistent with the work by Smith and
colleagues (2008) that CEDs and OC spray
allow officers to control suspects from a
distance without engaging in the hand-tohand struggles that often result in injuries.
The evidence from our study suggests
that CEDs can be an effective weapon in
helping prevent or minimize physical
struggles in use-of-force cases. LEAs should
consider the utility of the CED as a way to
avoid up-close combative situations and
reduce injuries to officers and suspects.
Similar results were obtained in a study by
Smith et al. (2008), who recommended that
CEDs should be authorized as a possible
response in cases where suspects use
defensive resistance (e.g., suspect struggles
to escape physical control of officer) or
higher levels of suspect resistance, in order
to avoid up-close combative situations.
We do not take a position on the
specific circumstances when an LEA should
authorize the use of the CED. We believe
such a policy decision needs to be made at
the local level. It is not appropriate, based on
a single study, to make a firm
recommendation on when a CED should be
authorized to be used. Each LEA has to
consider a multitude of factors in assessing
when to authorize use of CEDs, working
closely with its full set of community
partners to consider a range of local factors.
However, our study provides
important data points to inform these policy
decisions that LEAs need to make. For
example, there is little support in our data

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

for authorizing the use of CEDs in cases of
passive resistance by suspects, because those
types of cases rarely result in injuries to
officers. Also, in terms of reducing injuries,
there is little to gain by permitting use of
CEDs against certain special populations
(pregnant women, elderly citizens, and
others who are clearly physically impaired);
in our study, few of these persons were
involved in force cases where officers were
injured.
More work is also needed in the area
of officer training in the use of CEDs. There
is little attention in the CED literature to
training of officers and sheriffs’ deputies in
the proper use of CEDs. While some CED
manufacturers have developed CED training
curricula and some have even provided CED
training, there are few independent sources
for agencies to consult for guidance on
developing a CED training program (see
Smith et al., 2008). As a result, there is little
consensus on what training should be
required, what it should encompass, or what
its purpose should be beyond familiarization
with the device (see Smith et al., 2008).
More research is needed to identify which
types of CED training are most effective
(see Smith et al., 2008). Another training
issue is the inappropriate use of the CED.
Misuse can range from outright abusive or
illegal use of the weapon to less obvious
cases of officers turning to a CED too early
in a force incident. These problems can be
managed with policies, training, monitoring
and accountability systems that provide
clear guidance (and consequences) to
officers regarding when and under what
conditions CEDs should be used and when
they should not be used (see Smith et al.,
2008).

Conclusions and next steps
The management of officer use of force is
perhaps one of the most important tasks that

a law enforcement agency (LEA) will
undertake. Our study has documented an
important role for CEDs in this management
task. Overall, our data suggest that CED
sites were associated with improved safety
outcomes when compared to a group of
matched non-CED sites on six of nine safety
measures. And within CED agencies, in
some cases the actual use of a CED by an
officer is associated with improved safety
outcomes compared to other less-lethal
weapons. The evidence from our study
suggests that CEDs can be an effective
weapon in helping prevent or minimize
physical struggles in use-of-force cases.
LEAs should consider the utility of the CED
as a way to avoid up-close combative
situations and reduce injuries to officers and
suspects. Furthermore, use of CEDs by law
enforcement agencies is a relatively recent
development, and our research reflects the
early experience with CEDs (early- to mid2000s for most of the agencies in our study).
Over time, it seems reasonable to conclude
that LEAs will gain important insights into
the use of CEDs and may be able to improve
the safety outcomes associated with this
weapon.
Despite the utility of our findings,
many policy questions with the use of CEDs
remain: Where on the body a CED should be
used, the maximum safe number of CED
activations and the duration of shocks, and
the role of CEDs in contributing to deaths of
suspects. In addition, the results of our study
prompt further questions. For example, how
generalizable are our results? Our study was
made up of LEAs from urban areas. Would
our results be replicated in smaller
communities? While we were able to
include some incident-level and agencylevel control variables in our analyses, how
would our results hold up if we included
additional variables that might be available
in the future? To answer these and other
questions, a better approach to the collection

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

of police use-of-force data is needed. One of
the greatest barriers to conducting use-offorce research is the absence of uniformity
and comprehensiveness in the collection of
force data by LEAs across the country. We
observed limitations in content (information
about many of our areas of interest was not
collected by the LEAs), and timing (many of
the LEAs were limited in how long they
kept their force records – limiting our team
to no more than four years of analysis).
Also, the use-of-force tracking systems we
observed lacked a common architecture or
set of definitions, making comparative
analysis very difficult. We believe that a
national use-of-force database, as
recommended by Smith and colleagues
(2008), would greatly assist the law
enforcement community to produce reliable
answers to the above and other questions.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

CHAPTER 1:
Introduction

ur society gives law enforcement the
power to use force, even deadly force,
against citizens. Law enforcement agencies
make important decisions regarding this use
of force, including: the types of force to use,
technologies to deliver that force, when
various types of force can be used, and how
officers should be trained to use force. Many
elected leaders and police executives have
expressed a desire to reduce incidents of
police use of force within their communities,
especially deadly force. These leaders have
sought to identify innovative strategies and
use modern technology to achieve this
objective.
One alternative that has been advanced
is the use of a relatively new less-lethal
weapon called the Conducted Energy
Device (CED).1 Police chiefs and sheriffs
have communicated to the Police Executive
Research Forum (PERF) that they need
guidance in deciding whether to adopt CEDs
or other less-than-lethal weapons. The
purpose of this project was to produce
scientifically valid results that will inform
decisions about use of CEDs.
Compared to firearms, CEDs offer the
promise of helping officers to control violent
suspects without killing them or running the
risk of a stray bullet killing a bystander.

1 CED technology includes traditional stun guns and
projectile weapons sold under the trade names
Taser® and Stinger™ Handheld Projectile Stun
Guns (see http://www.ojp.usdoj.gov/
nij/topics/technology/less-lethal/conductedenergy-devices.htm).

However, there is uncertainty within the law
enforcement community about deployment
of CEDs, especially with regard to deaths
that have occurred following the use of
CEDs. Law enforcement executives have
been deluged with questions about the safety
and effectiveness of CEDs, and some have
been forced to explain a number of
controversial tactical uses of CEDs by their
officers. The lack of reliable information
and a full understanding of CEDs and
whether or how to best use them has
hampered the ability of police executives to
make informed policy decisions about the
devices. Police executives have been
provided with little independent scientific
evidence and guidance on the impact of
using CEDs, forcing them to make policy
and operational decisions without being
fully informed.
While several studies have examined
the relationship between CED usage and
arrestee handling, little work has been
completed regarding the relationship
between policies governing CED
deployment and the risk of injuries. This gap
exists despite the fact that the subject of
police use of force has been studied for more
than four decades (Smith, Kaminski, Rojek,
Alpert and Mathis, 2007). While decades of
research have documented the nature and
extent of the force used by police and the
conditions and the correlates that affect its
application (Smith et al., 2007), little
research has been done isolating the effects

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

of using CEDs on injuries to suspects and
officers.
This Police Executive Research Forum
(PERF) study, conducted from September
2006 to November 2008, is one of the first
to compare law enforcement agencies
(LEAs) that use CEDs with matched LEAs
that do not use CEDs. The purpose of this
study was to complete an objective analysis
of the effects that department-wide
deployments of CEDs by LEAs have on
injuries. Our primary aim was to evaluate
the effect of CED deployment on injuries
and death to police and suspects, associated
medical attention, and the need for
hospitalization.
Overall, our goal was to produce
practical information that can help LEAs
establish policy and procedural guidelines
that assist in the effective design of CED
deployment programs that support increased
safety for officers and citizens. In order to
accomplish this goal, we examined the
outcome of CED deployment in terms of
officer and suspect safety. We compared
outcomes for LEAs that have incorporated
the use of CEDs (n=7) to outcomes in LEAs
that have not incorporated the use of CEDs
(n=6). This study contains important
scientific information isolating the safety
outcomes to be expected if a department
deploys CEDs, controlling for a variety of
related organizational and individual level
factors. The focus of our study on injuries to
police officers and suspects during use-offorce events should bring some clarity to
this relatively understudied field.
Because the news media tend to
provide heavy coverage of serious uses of
force by police, it is easy to get the
impression that police use of force is
commonplace. But prior research suggests
that these types of encounters are rare. Only
1.5 percent of police-citizen contacts involve
the threat or application of physical force by

the police2, and 14 percent of these cases
involve subjects who claim they sustained
an injury (Durose, Schmitt, & Langan,
2005). Similar low levels of suspect injuries
sustained during use-of-force encounters
have also been found in single-agency
analyses based on surveys of law
enforcement officers (Kaminski,
DiGiovanni, & Downs, 2004; Smith &
Petrocelli, 2002). Alternatively, studies
using agency records found higher rates of
injuries to citizens during use-of-force
encounters, with injuries reported in
approximately 40 percent of the incidents
(e.g., Alpert & Dunham, 2004; Henriquez,
1999). Despite these differences based on
varying data sources, there seems to be
agreement that most suspect injuries are
relatively minor, typically consisting of
consisting of bruises, abrasions, and muscle
strains and sprains (Alpert & Dunham,
2000; Henriquez, 1999; Kaminski et al.,
2004; Smith & Petrocelli, 2002).
The data on the prevalence of officer
injuries in use-of-force encounters is less
clear. Some studies have found that one in
10 officers were injured during use-of-force
incidents (Henriquez, 1999; Kaminski et al.,
2004; Smith & Petrocelli, 2002).
Alternatively, analysis of force records from
the Miami-Dade Police Department and the
Baltimore County (Maryland) Police
Department revealed substantially higher
rates of officer injury, 38 and 25 percent,
respectively (Alpert & Dunham, 2000; 2004;
Kaminski, & Sorensen, 1995). Nevertheless,
as with findings regarding suspect injuries,
research on force-related officer injuries
found that most injuries were relatively
minor (Alpert & Dunham, 2000; Brandl,
1996; Brandl & Stroshine, 2003; Kaminski
et al., 2004; Smith & Petrocelli, 2002).

2 Also, it has been estimated that only 15–20
percent of arrests involve the use of force by police
on non-complying suspects.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

A key concern with the use of force by
the police is the possibility of injury to
suspects and officers, along with related
costs for medical bills for suspects, worker’s
compensation claims for injured officers, or
compensatory damages paid out in legal
settlements or judgments. However, until
recently, few revealing studies had been
done on the frequency, causes, or correlates
of force-related injuries (Smith et al., 2008).
Over the past couple of decades, new
technologies have emerged that offer the
promise of more effective control over
suspects who resist police, with fewer or less
substantial injuries (Smith et al., 2008).
These technologies include oleoresin
capsicum (OC or “pepper spray”) found in
use in most law enforcement agencies, and
CEDs (such as Tasers®), reported to be in
use in more than 11,500 LEAs (Smith et al.,
2008). As with OC spray, CEDs have
generated controversy (Amnesty

International, 2004) and have been linked
with in-custody deaths and allegations of
overuse and even intentional abuse (Smith et
al., 2008). The focus of our study is
objectively assessing the experience of
LEAs with CEDs and whether they can be
deployed safely and effectively.
To follow, in Chapter 2, we review the
prior work that has been done in the area of
police use of force generally, and less-lethal
weapons in particular. In Chapter 3, we
provide a detailed account of how we
conducted this study, including a review of
the strengths and weaknesses of our research
design, measures, and analytic procedures.
In Chapter 4 we present all of the
substantive data analytic results for the
project. In Chapter 5, we summarize our
main findings, discuss the implications of
our results for LEA policy and training, and
provide some recommendations for future
research.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

CHAPTER 2:
Literature Review

aw enforcement officers are legally
authorized to use force, including
deadly force, in carrying out their mandate
to preserve order and enforce the law. Ever
since the issuing of the first less-lethal
weapon—the truncheon or ‘short billy’ club
– to officers in London’s Metropolitan
Police Department, police have sought safer
and more effective tactics and technologies
for controlling and subduing resistive and
combative subjects. Over the last several
decades, the law enforcement field has
witnessed substantial improvements in
unarmed methods of defense and control, as
well as the development of new variants of
the baton (e.g., side-handled and expandable
batons), chemical irritants (OC spray, CS [2chlorobenzalmalononitrile] spray, CN tear
gas (Alphachloroacetaphenone) spray;
pepper spray), and Conducted Energy
Devices (e.g., Tasers®). These newer
methods and technologies have been
variously credited by some experts with
reducing police shootings, the incidence of
use of force generally, officer and suspect
injuries, and excessive force complaints
(Ederheimer, 2005). Their adoption and use
have not been without controversy. Some
have claimed that police use of various
tactics or technologies such as the lateralvascular (or carotid) neck restraint, OC
spray, or CEDs have directly caused or
contributed to deaths in police custody
during encounters in which deadly force
may not have been appropriate, or that these
types of force have been used

inappropriately to "punish" suspects.3 For
example, Amnesty International has
documented over 245 deaths that occurred
after the use of CEDs. Other civil liberties
organizations have argued that a moratorium
should be placed on CED use until research
can determine a way for them to be safely
used.
In this chapter, we provide (1) an
overview of the literature on less-lethal
weapons generally and then (2) a more
detailed review of the use of CEDs. Next,
(3) we review outcomes associated generally
with the use of less-lethal weapons by the
police, followed by (4) a more intensive
review of outcomes associated with CEDs.

Less-lethal weapons (LLW)
To carry out their job, police officers rely on
a range of weapons that are considered less
lethal alternatives to firearms. The challenge
for chiefs and sheriffs is to manage the use
of these various weapons and to provide
clear, firm guidance to officers/deputies in
making appropriate decisions when
choosing to use one of their weapons. Force
3 In 2003, Amnesty International called for a
moratorium on Tasers® until an independent inquiry
on the use and effects of Tasers is completed, and
in 2004 it reiterated this recommendation
(Amnesty International 2003, 2004). In the 2004
report, Amnesty International offers
recommendations to agencies that decline to
suspend Taser usage. One of its significant
recommendations is that police departments using
Tasers strictly limit their use to situations where
the alternative would be deadly force.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

must be used cautiously and judiciously, and
only to promote the safety of the community
and officers (Adams 1995). In most
instances, police officers are justified in
their use of force to protect themselves or
other citizens, but sometimes they use force
that is unwarranted by the situation (Gaines,
Kaune, and Miller, 2001).
For more than a century,
advancements in technology have greatly
changed the weapons used by police. In a
sense, it appears that police weaponry has
come full circle. During the mid-1800s,
police officers in Boston and New York
relied on LLWs (primarily wooden clubs).
In the late 1800s, in response to betterarmed criminals, police forces began issuing
firearms to officers (Allison and Wardman
2004, 116). While firearms are still
standard-issue tools, today’s police
departments once again emphasize the use
of LLWs (albeit more advanced weapons
than in the 1800s) rather than firearms in
most situations, where lethal force is not
justifiable.
For several decades, the LEA
community has been in search of LLWs that
would provide officers with the ability to
manage use-of-force incidents effectively
while at the same time reducing the potential
for injury to suspects and officers (Smith et
al., 2007). Over this period, policing experts
recognized that a perilous gap existed in the
types of weapons available to officers; that
is, there are situations in which batons may
be too weak an option, and guns are too
strong.4 This fact became clear in 1985
when the U.S. Supreme Court ruled in
Tennessee v. Garner that the use of deadly
force to apprehend apparently unarmed,
nonviolent fleeing felons was an
unreasonable seizure under the Fourth
Amendment (Pearson, 2003).
4

See http://www.iejs.com/TechnologyandCrime/
Law_Enforcement_Technology/
less_than_lethal_weapons.htm.

Today, LEAs have a wider range of
less-lethal weapons, including:
• Impact projectiles (e.g., rubber bullets,
bean bags, and other blunt trauma
projectiles launched from a pump-action
shotgun)
• Electrical shock weapons (e.g., CEDs and
stun guns),
• Chemical irritants (e.g., OC spray, tear
gas and stink bombs),
• Physical restraints (although they are not
often considered “weapons,” they are
often used in conjunction with less-thanlethal devices and include nets, wire
entanglement systems, sticky foams, and
handcuffs/flexible cuffs),
• Hard impact weapons (e.g., retractable
batons and flashlights),
• Weapons that use extreme light (e.g.,
bright white lights or lasers that produce a
“wall of light” that may deter an assailant
from attacking someone behind the light;
these distraction devices can also confuse,
frighten, or disorient violent suspects),
and
• Acoustic-based weapons (e.g., acoustic
energy, at both audible and inaudible
frequencies, has been examined for
potential use as a LLW, primarily for
halting the advance of an aggressive or
violent crowd in a riot scenario).
While LEAs have experimented with
many of the aforementioned weapons, OC
spray and CEDs are the most commonly
used of these weapons (Smith et al., 2007).
Similar to the current day controversy
surrounding CEDs, as referenced earlier, in
the early to mid-1990s, OC spray was
spreading rapidly among U.S. police forces
and concerns were being raised regarding its
overall safety and cases of misuse (Amnesty
International, 1997 and ACLU of South
California, 1995). As pointed out by Smith

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

and colleagues (2007), these concerns
prompted the National Institute of Justice
(NIJ) to fund a variety of studies on the
safety and effectiveness of OC spray
(Edwards et al., 1997; Granfield et al., 1994;
Petty, 2004), and several other researchers
examined its incapacitative effects and the
relationship between OC use and
officer/suspect injuries (Kaminski et al.,
1998, 1999; Morabito and Doerner, 1997;
Smith and Alpert, 2000; Lumb and Friday,
1997).
These studies found that the deaths
occurring after the use of OC spray were
generally the result of positional asphyxia,
pre-existing health conditions, or drugrelated factors (Granfield et al., 1994; Petty,
2004). The research data suggest that the use
of OC spray by officers was associated with
fewer attacks on officers and a reduction in
related injuries to suspects and officers
(Edwards et al., 1997; Gauvin, 1995;
Kaminski et al., 1999; Lumb and Friday,
1997; NIJ, 2003; Nowicki, 1993; Smith and
Petrocelli, 2002). Nevertheless, this above
research suffered from a number of
methodological problems, such as the lack
of comparable control groups, measurement
limitations, and the lack of statistical
controls for the level of suspect resistance
and the use of other tactics or weapons that
may have been used in conjunction with OC.
As a result, we are left with inconclusive
evidence on the independent effect of OC
spray on suspect and officer injuries after
holding constant other types of force and
resistance that may have been used (Smith et
al., 2007).

Conducted Energy Devices
(CEDs)
CEDs use electro-muscular disruption
technology to cause neuromuscular
incapacitation and strong muscle
contractions through the involuntary

stimulation of both the sensory nerves and
the motor nerves, causing the suspect to be
temporarily incapacitated and fall to the
ground.5 CEDs, such as Tasers®, use
compressed nitrogen to fire two barbed
probes/darts and 50,000 volts of electricity
along thin wires attached to the probes.6 The
innovativeness of the CED weapon is that it
is not dependent on pain compliance (like
traditional stun guns), making it highly
effective on suspects with high pain
tolerance.
Until recently, TASER International’s
products were the main electronic
incapacitating devices commercially
available for police officers to carry (IACP
2003). As a result, the most common
devices in use by law enforcement are the
Taser M26 and X26 models.7 According to
TASER International, by 2005 more than
6,000 law enforcement agencies (primarily
in the United States) were using Tasers, with
more than 1,150 agencies deploying them to
all officers on patrol. There are more than
100,000 Tasers in use by police officers in
the field (Kelly 2004). Current industry
figures place CEDs in the hands of more
than 11,000 LEAs nationwide.8
TASER is an acronym for the Thomas
A. Swift Electric Rifle, developed in the
1970s by Jack Cover. Swift was a fictional
5

See http://www.ojp.usdoj.gov/nij/topics/
technology/less-lethal/conducted-energydevices.htm.
6
See http://www.ojp.usdoj.gov/nij/topics/
technology/less-lethal/how-ceds-work.htm.
7
However, two companies have recently entered the
market. In January 2005, Stinger Systems™ Inc.
began offering its Stinger 4 Dart Less Lethal Gun,
and in March 2005, Law Enforcement Associates™
(LEA) began offering its LEA Stun Gun. It is likely
that more devices using this technology will be
developed for sale to law enforcement agencies in
the future.
8 Taser International reports that more than 12,800
law enforcement, correctional and military
organizations in 44 countries use its devices. Of
these agencies, more than 4,500 of them equip all
of their patrol officers with Tasers. Since 1998,
more than 260,000 Taser brand immobilizers have
been sold to law enforcement agencies.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

character in a 1930s series of science fiction
books by Victor Appleton (Sanchez 2004).
The Taser fires darts that attach to (or
penetrate) a person’s skin or clothing and
create an incapacitating electrical current.
The Taser has evolved over the years. In
1999, the company developed the Advanced
Taser M26, which was powered by an
alkaline battery and used nitrogen cartridges,
rather than gunpowder, which was used in
earlier models, to fire projectiles. Shaped
liked a handgun, the Advanced Taser M26
became popular with law enforcement
officers. In 2003, the company introduced
the Taser X26, more compact than the
Advanced Taser M26 and, according to the
company, more efficient. It is powered by a
lithium battery and uses nitrogen cartridges
to fire projectiles. Tasers are
microprocessor-controlled. There is an
onboard memory that records the dates and
times of the most recent 585 times the unit
has been fired (Nielsen, 2001). The M26 has
a Microsoft Windows-compatible data port
that allows the data to be downloaded to a
computer using a special adapter cable
(Nielsen, 2001). This allows an agency to
monitor usage patterns, and in some cases
helps police executives either to document
an officer’s unwarranted use of the CED, or
to defend against unfair allegations of abuse
of force. Tasers are laser-sighted and use
cartridges attached to the end of the
weapon’s barrel.
The Taser has two modes: “probe” and
“touch stun.” In the probe mode, the
cartridges project, via a set of wires, a pair
of barbs (or darts with hooks) that attach to
clothing or penetrate the skin after the Taser
is fired, delivering an electrical charge
(Association of Chief Police Officers, 2004).
The Taser sends an electrical current down
the wires and through the body between the
two barb points. In the touch stun mode,
electrical contacts on the Taser are pressed
directly onto a person; there is a similar but

reduced neuromuscular effect (Donnelly et
al. 2002). Taser specifications indicate that
the Taser is effective on persons up to 21
feet away; the ability of police to keep such
a distance from a suspect during a
confrontation improves their safety
significantly.

Outcomes associated with the
use of less-lethal weapons
(LLWs) by the police
There have been a number of studies
conducted over the past several decades
focusing on LLWs (e.g., Kingshott, 1992;
Edwards, Granfield, and Onnen, 1997;
Gauvin, 1994; Phillips, 1994; IACP, 1995;
Robin, 1996; Morabito and Doerner, 1997;
Kaminski, Edwards and Johnson, 1998;
Smith and Petrocelli, 2002; Kershaw, 2004;
Adang, Kaminski, Howell, and Tilburg,
under review). Of key concern to
practitioners is the relationship between
officer use of force and injuries to suspects
and to the officers themselves. As different
parts of the body differ in vulnerability, and
because people vary in weight and fitness,
any weapon powerful enough to incapacitate
can kill under certain circumstances. Many
weapons manufacturers and LEAs are now
using the term “less-lethal” in place of the
older terms “non-lethal” or “less-thanlethal,” to emphasize that these weapons
tend to kill or injure far fewer targets than
firearms, which primarily incapacitate by
killing or maiming.
Several studies have focused on the
extent to which LLWs are “effective” in
helping officers gain compliance over a
subject. One such study found that OC was
“effective” 70% to 85% of the time,
depending on the definition and measure
used. Earlier studies had found higher levels
of effectiveness—ranging from 90% to
100% (Kaminski, Edwards, and Johnson,
1998). Since the 1970s, approximately 12

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

deaths have been attributed to impact
weapons like the beanbag round (Wilmette,
2001). If used improperly, these rounds can
penetrate subjects and cause serious injury
and death. In one California case, a woman
was threatening police with a knife when
they shot her in the arm and torso with
beanbag rounds, and she fell to the ground
and died. The cause of death was cited as a
laceration of the heart, due to severe focal
blunt force trauma (Shin, 2002). Another
danger of less-lethal weapons is confusing
the less lethal ones with the lethal ones. One
case has been documented in which an
impact round shotgun was found to contain
a one ounce slug. This was discovered after
it was fired, and the slug severed the target’s
leg above the right knee (James, 2002). To
overcome confusion, some departments use
bright orange stocks on their less-lethal
shotguns to distinguish them from the
others.
A small number of studies have
examined the extent to which various
weapons cause injuries, as less-lethal
weapons are most valuable to law
enforcement if they can result in subject
compliance while minimizing injury to both
officer and subject. While there have been a
number of studies that have examined police
use of deadly force or officers killed in the
line of duty, less research has been
conducted on nonfatal injuries to suspects
and officers (Smith et al., 2008). In studies
by Alpert and Dunham (2000), Meyer
(1992), and Smith and Petrocelli (2002), the
researchers found that when officers used
bodily force (e.g., takedowns, wrestling, and
punching) to get control of a suspect, they
had the greatest chance of getting injured.
Other research also suggests that suspects
have a higher likelihood of injury when
officers use canines and impact weapons
(such as batons or flashlights) (Smith et al.,
2008). Overall, despite decades of research
on use of force, much of the research on

injuries related to police use of less-lethal
weapons remains descriptive in nature or
contains substantial data and analytic
limitations that limit the utility of this
research (Smith et al., 2008).

Outcomes associated with the
use of CEDs by the police
CED specific non-medical studies: While
CEDs are now in use by thousands of LEAs
(GAO, 2005), the research on CEDs has
been mostly descriptive and few have
examined the relationship between CEDs
and injuries (see Charlotte-Mecklenburg
Police Department, 2006; Jenkinson,
Neeson, & Bleetman, 2006; Seattle Police
Department, 2002). The LEAs themselves
conducted much of the early research on
injury rates before and after CED
implementation. LEAs in Austin, TX; Cape
Coral, FL; Charlotte-Mecklenburg, NC;
Cincinnati; Orange County, Phoenix; South
Bend; and Topeka, based upon use-of-force
reports, all reported substantial declines in
either officer injuries (3 and 93 percent) or
suspect injuries (between 40 and 79 percent)
following the adoption of CEDs (Smith et
al., 2008). Overall, these assessments
indicate generally that CEDs are effective,
but these estimates vary depending upon
whether one evaluates the effectiveness of
all instances in which CEDs are deployed
against subjects or only the CED
deployments that result in both darts making
contact with the subject.9 The Cincinnati
Police Department reported CEDs were
successful at gaining compliance from
resistive/combative suspects approximately
84% of the time (Streicher, 2005). Of these,
about 65% of suspects were “immobilized”
and another 18% complied or were only
partially affected (Streicher, 2005). In 102
instances where the CED was ineffective, an
9 A subject will not receive an activation unless both
darts make contact with the subject’s skin.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

inadequate circuit resulted in continued noncompliance (e.g., missed target, darts failed
to penetrate subject’s clothes), but there
were also 61 instances in which the threat of
CED use was enough to gain compliance
(Streicher, 2005).
Next, TASER International claims that
police departments have seen a decrease in
officer and suspect injury rates after the
introduction of the Taser. The company’s
website claims that injuries to suspects have
dropped between 40 to 68 percent after the
introduction of the Taser and injuries to
officers have declined 41 to 93 percent. The
site also reports a reduction in worker’s
compensation claims for one police
department. The Grant City, Illinois Police
Department introduced the Taser as part of a
multi-faceted risk management program.
The TASER International website reports
that police department worker’s
compensation expenses were $454,192 and
$740,172 for the two years prior to the
introduction of the Taser. According to
TASER International, based on two years of
data after deployment of the Taser, the Grant
City Police Department spent zero dollars on
worker’s compensation expenses10.
However, these results have not been
subjected to independent analysis, except for
one analysis of data from TASER
International that was subjected to the
scrutiny of peer review. Based on data
maintained by TASER International,
researchers (Jenkinson, Neeson, &
Bleetman, 2006) found a low level of injury
associated with CED use (8%) compared to
the use of CS spray (13%) and batons
(24%).
Overall, questions have been raised
about these CED studies because they are
not the product of research produced by
independent sources (Smith et al., 2008).
Also, pre-post designs are generally

considered weak research designs,
especially considering that these studies did
not statistically control for situational factors
and other types of force used in conjunction
with CEDs during any given force incident.
Without a comparison group, such pretest/post-test designs are not effective at
isolating the effectiveness of CEDs. That is,
there is no way of knowing if some other
factor in the environment might have led to
the observed changes between the “before”
and “after” period.
In one of the more rigorous
independent studies in this area, Smith,
Kaminski, Rojek, Alpert and Mathis (2007)
analyzed the relationship between CEDs and
officer and suspect injuries from two law
enforcement agencies while simultaneously
controlling for the effects of other types of
force used by officers as well as suspect
resistance and other factors. The use of
CEDs was associated with reduced odds of
officer and suspect injury and the severity of
suspect injury in one agency. In the other
agency, CED use was unrelated to the odds
of injury; however, the use of pepper spray
was associated with reduced odds of suspect
injury. Among other findings, in both
agencies the use of hands-on tactics by
police was associated with increased odds of
officer and suspect injury, while the use of
canines was associated with increased odds
of suspect injury. A major concern with this
study was the absence of comparison
agencies that have not deployed CEDs, and
this study was limited to only two CED
deploying LEAs.
In another rigorous study of this issue,
Smith, Kaminski, Alpert, Fridell,
MacDonald, and Kubu (2008) collected
more than 24,000 use-of-force records from
12 police agencies that have deployed
CEDs.11 These data were combined and
analyzed using multilevel and fixed-effects

10
See: http://www.taser.com/documents/
TASERS_saving_lives_compilation-short.pdf

11 One of these LEAs also participated in the current
PERF study.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

models to investigate the relationship
between policy-related factors and the
likelihood of injury to police and citizens in
use-of-force incidents, adjusting for the
demographic and situational differences
between police use-of-force incidents. While
controlling for the use of less-lethal weapons
(OC spray and CEDs) in force encounters,
they found that the use of physical force
(hands, feet, fists) by police increased the
odds of injury to suspects by more than 50
percent and substantially (by a factor of 3)
increased the chances of injury to officers.
Conversely, the use of OC spray or CEDs
decreased the probability of injury to
suspects by 65 and 70 percent respectively.
Injuries to officers were unaffected by the
use of CEDs, while the odds of officer
injuries increased somewhat (by about 21
percent in the 12 agency models) when OC
spray was used. Overall, CED use reduced
the probability of injuries to suspects across
the 12 agencies in the combined analysis
and in two out of the three agencies, whose
data were analyzed independently (MiamiDade and Seattle). Likewise, the relationship
between OC spray and suspect injuries in
the multi-agency analysis is consistent with
the injury reduction finding in Richland
County; in Seattle, OC spray had no effect
on suspect injuries, while the Miami-Dade
Police Department does not issue OC spray.
Our study, the results of which are
reported in Chapter 4, builds on the Smith et
al. (2008) study, using comparable measures
and including LEAs that have deployed
CEDs and matched LEAs that have not
deployed CEDs. The problem with using
data only from CED agencies is that we
have no counterfactual comparison to
agencies that did not use CEDs, and are left
with a simple pre/post design with all of its
well-known flaws. Also, we are limited in
observing the full effects of CEDS across
similar types of force situations. In addition,
some agencies reserve the use of CEDs only

for certain types of more serious situations
that justify higher levels of force, and tend
to involve more danger to the officer,
bystanders, or suspects. In these agencies,
comparing CED use against situations
involving lower levels of danger, in which
other types of weapons may be used, could
set up an unfair comparison. Thus, to the
extent that our study includes pre-post
analysis of agencies that have deployed
CEDs, we are very cautious in our
interpretation of these data.
CED-specific medical studies: A
number of controlled medical studies have
been conducted examining the physiological
effects of CEDs on animals and humans.
One of the vital issues regarding the use of
CEDs is whether exposure can induce
ventricular fibrillation (VF).12 To address
this issue, a number of controlled studies
using sedated animals were conducted (see
Smith et al. [2008] for a full summary of
these studies). These studies found no VF of
the heart using standard discharges of
relatively short duration (e.g., 5–15
seconds), but higher-output discharges (e.g.,
15–20 times the standard) or discharges of
longer duration (two 40-second exposures)
induced VF or increased heart rhythm
(ventricular tachycardia) in some pigs
(Dennis et al., 2007; Lakkireddy et al., 2008;
Stratbucker et al., 2003; McDaniel et al.,
2005; Walter et al., 2008), and longer
duration exposures led to VF-induced death
in three pigs (Dennis et al., 2007; Walter et
al., 2008). Research by Nanthakumar and
colleagues (2006) found that orienting
TASER barbs across the hearts of pigs
(simulating a “worst case scenario” of
creating a current vector that directly passes
through the heart) led to stimulation of the
heart muscle (but not VF), while placement
12 VF is a condition in which there is uncoordinated
contraction of the cardiac muscle of the ventricles
in the heart, making them tremble rather than
contract properly, leading in some cases to a
cessation of blood circulation and death.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

across the abdomen did not (see also
Lakkireddy et al., 2006; Roy & Podgorski,
1989). Although cardiac stimulation may be
of little concern for healthy subjects,
Nanthakumar et al. (2008) caution that heart
stimulation might induce VF if preexisting
conditions are present, such as heart disease,
drug intoxication, and so forth.
Several controlled studies using
healthy human subjects also have been
conducted. For example, examining the
impact of CEDs under highly controlled
conditions, Levine et al. (2005) found that
20 human subjects exposed to
approximately a 2.4-second shock from
Tasers experienced no cardiac dysrhythmias.
Also, Levine et al. (2007) monitored the
hearts of 105 police trainees before, during
and after exposure to the X-26 TASER for
approximately 1 to 5 seconds. Although
subjects experienced significant increases in
heart rate following exposure, none
experienced VF. An earlier study by Levine
et al. (2005) reached similar conclusions. In
a review of this literature, Smith and
colleagues (2008) summarized these
findings indicating that the evidence
suggests that CEDs are relatively safe when
used on healthy at-rest as well as
physiologically stressed subjects, but that
medical researchers caution that CEDs are
not risk-free (National Institute of Justice,
2008; Vilke & Chan, 2007). Strote and
Hutson (2008), for example, suggested that
CEDs might cause physiologic and
metabolic changes that are clinically
insignificant in healthy individuals but that
could be harmful or even life-threatening in
at-risk populations (e.g., obese subjects with
heart disease and/or those under the
influence of drugs). Also, there are the
secondary injuries that can occur from
falling after exposure to a CED. For
example, Kroll, Calkins, Luceri, Graham,
and Heegaard (2008) reported six deaths due

to head injuries suffered during falls
following CED exposure.
While studies with animals and
healthy volunteers are important, there is
also a need for field studies in the actual
population at risk of CED exposure. Below
we examine three case review studies that
explored cases involving suspects who were
subjected to a CED activation. In a 1991
study, Kornblum and Reddy examined 16
CED-related deaths and reported that in all
cases the subjects were behaving in a bizarre
manner and that more than 80% of the
subjects were under the influence of
cocaine, PCP or amphetamine. Kornblum
and Reddy (1991) found that in only one
case was a CED possibly associated with a
death of a suspect. They concluded that the
CED did not cause death in this single case,
but may have contributed to the death. In
that case, the suspect had a heart disease and
toxic levels of PCP in his system.
In a review of 37 CED-related deaths,
Strote and Hutson (2006) found that autopsy
reports indicated that CEDs were a possible
cause of death in six and were a contributory
cause in four of the 37 death cases. Strote
and Hutson (2006) concluded that the fatal
CED encounters involved subjects already at
risk for sudden death from other causes, and
that a common factor in the death cases was
extreme agitation, often accompanied by
stimulant drug use and/or preexisting heart
disease.
In another study, Bozeman and
colleagues explored whether CEDs
contribute to or cause death (National
Institute of Justice, 2008), and wrote that
they found no conclusive medical evidence
that indicates a high risk of death from the
direct effects of CEDs (Bozeman, Winslow,
Hauda, Graham, Martin, & Heck, 2008). In
the study, six LEAs participated over two
years. Independent physicians worked with
each LEA. Researchers evaluated 962
incidents in which a CED was used on a

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

suspect. While the large majority (99.7
percent) of the suspects in these 962 cases
exposed to CEDs suffered no injuries or
only mild injuries, a small number suffered
potentially lethal injuries. This study did not
observe any deaths occurring immediately
after CED use that might suggest that the
CED directly affected a suspect’s heart
rhythm.
Studies examining the positive and
negative impacts of less-lethal weapons are
critical for producing information to guide

policy-makers. While the studies described
above have produced some important
information about various outcomes of
CEDs, individually and as a group they are
insufficient to guide decision-making. Our
quasi-experiment provides comparative data
from LEAs with CEDs and matched LEAs
without CEDs. Our design allows us to
isolate and rigorously evaluate the effects of
using CEDs on officer and suspect safety
outcomes.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

CHAPTER 3:
Research Design and Methods

Overview
Our research design allows our team to
compare departments with CED deployment
(n=7) to a set of matched departments (n=6)
that do not deploy CEDs on a variety of
outcomes. Matching was based on criteria
such as violent crime levels, police activity
(violent crime arrests), agency size, and
population size of jurisdiction. The inclusion
of 13 departments allows us not only to
assess incident-level factors, but also some
important departmental/organizational-level
factors that could affect outcomes. To assure
a fair comparison we collected at least four
years of data on all incidents of use of force
for all of the participating departments. For
the LEAs that deployed CEDs, we collected
at least two years of data before and two
years after CED deployment. For the LEAs
that did not deploy CEDs, we collected at
least four years of data over a similar period.
While the focus of our study was on
the use of CEDs, we also collected data on
all use of force incidents (not just CED
cases) and examined the range of weapons
(including pepper spray and batons) and
unarmed tactics (e.g., joint locking
techniques) that the police employ in
exerting force to arrest suspects. Agencies
that do not deploy CEDs all have other
forms of less-lethal options, and our study
provides evidence on the relative
effectiveness of CEDs to these other
options, controlling for a variety of related
organizational and incident-level factors.

Five of the sites in the study (three
non-CED sites and two CED sites) did not
have electronic use-of-force databases. For
these five sites, we sent a team of three data
collectors to collect random samples of 50
cases per year per site for four years (for a
total sample of about 200 cases per site).
Two individuals independently coded data
from hard copies of use-of-force forms. The
third person (a research supervisor) checked
these data collectors’ work, resolved any
conflicts between the two coding sheets, and
entered a reconciled sheet into a research
database. Inter-rater reliability statistics
were high across all five sites (on average
.91 across all the sites).

Quasi-Experimental Design
While it might be preferable to assess the
impact of CEDs through a randomized
clinical trial (RCT), this type of design is not
possible in this context. We are unaware of
any police department that would randomly
assign a CED (or any other weapon) to its
officers, due to ethical concerns. Ethical
considerations dictate that police chiefs
develop use-of-force policies based on their
best judgments of what will be safest and
most effective in their communities. At this
time, different chiefs have made different
determinations on the question of whether to
deploy CEDs; however, for each chief, there
is no room for randomness in those
calculations. An RCT with CEDs could

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

potentially endanger the lives of officers,
suspects, and bystanders.
One common alternative to the more
inferentially powerful RCT design is a
quasi-experimental design (QED). In this
context, “quasi” means that the design is
similar to an experiment, except that this
design is characterized by a comparison
group that receives either a different
treatment or no explicit treatment at all (see
Cook and Campbell, 1979). QEDs have a
similar purpose as RCTs in terms of testing
causal hypotheses and share many structural
details (e.g., pre- and post-tests and
comparison groups), but lack random
assignment.
QEDs require the researcher to
enumerate alternative explanations one by
one, decide which are plausible, and then
use logic, design and measurement to assess
whether each one is operating in a way that
might explain any observable effect
(Shadish, Cook and Campbell, 2002: 14).
Christensen (1988: 306) argues that many
causal inferences can be made without using
the RCT framework; they are made by
rendering other rival interpretations
implausible. For example, if someone
unknowingly stepped in front of an
oncoming car and was pronounced dead
after being hit by the car, you would
probably attribute the death to the moving
vehicle. The person might have died as a
result of numerous other causes happening
at that same point in time (e.g., a long-term
debilitating illness that finally killed the
person), but such alternative explanations
are not accepted because they are not likely
to be plausible. In a like manner, the causal
interpretations arrived at from quasiexperimentation are those that are consistent
with the data in situations where rival
interpretations have been shown to be
implausible. Our design allows our team to
isolate a number of injury/safety outcomes
to be expected if an LEA deploys CEDs,

controlling for a variety of organizational
and incident-level factors.
Selection criteria for inclusion in
study – We selected 18 police departments
nationwide using a careful selection process
to ensure comparability across these
departments and to ensure that each
department could provide the necessary
outcome data regarding injuries in use-offorce incidents. Our goal was to have at least
12 departments in our study, and we were
able to obtain 13. The selection criteria
included: (1) being able to provide data on
all incidents of use of force (including data
such as type of force used and injury
outcomes to both officer and suspects), (2)
having a written policy identifying CED and
other less-lethal weapon placement on the
force continuum, (3) a willingness to share
data with PERF for this study, and (4)
having at least 100 sworn officers (we
sought larger LEAs for participation in our
study in order to obtain sufficient numbers
of use-of-force incidents for a robust
analysis).
Next, we needed to ensure that
appropriate groups could be compared to
each other and that the time-series (pre- and
post-test) component of this study could take
place. The final criterion (5) was that the
departments in our study needed to have all
of the necessary data available for at least 4
years (2 years pre- and 2 years post-CED
deployment or 4 years of a comparable time
period for the non-CED sites).
Matching to comparison cities –
With QEDs, the comparison group is usually
a naturally formed group that is similar in
some ways to the experimental group but it
does not receive the intervention that is the
subject of the study. In our case, we have
departments that use CEDs and a matched
group of police departments that do not. The
main concern with the QED design is that
although differences that can be measured
can be statistically controlled, unmeasured

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

variables related to the outcome variable
cannot be controlled. Therefore, our above
selection criteria standards on data
availability and PERF access had to be
vigorously enforced to ensure the
availability of all data necessary to create
statistical controls for possible “pretreatment” differences between our study
groups. We originally had 18 potential sites,
but five had to be dropped because of data
availability issues.
To assure that we could identify sites
into the study that could meet our study
selection criteria (see above), we needed a
methodology to screen sites for possible
inclusion into our study. Fortunately, at the
time, PERF had recently completed a survey
that could be used for screening purposes.
Our selection of cities was based on a PERF
nationally representative survey on use of
force conducted in 2006–2007. This survey
was done as part of another NIJ-supported
project (Smith et al., 2008) called the Use of
Force survey. Briefly, this survey was
developed to collect information on the
current state of less-lethal weapon policy,
practice, training and usage; and to
empirically assess the positive and negative
outcomes associated with less-lethal
weapons (full details on the survey are
provided in Smith et al., 2008). The
University of South Florida, in collaboration
PERF and the University of South Carolina,
developed this survey. The final survey
instrument contained a series of both openand closed-ended questions. PERF drew a
nationally representative stratified sample of
LEAs using the 2005 National Directory of
Law Enforcement Agencies database. PERF
stratified the sample by type of LEA (i.e.,
local police departments and sheriffs’
offices), region of the country, and the size
of the population served by the department.
The Use of Force survey was sent to a
stratified random sample of law enforcement
agencies (N = 950). Respondents were able

to submit the survey via mail, facsimile,
email, Federal Express, or the Internet (just
under 60% of the sampled LEAs submitted a
completed survey).
The same Use of Force Survey was
used by Smith et al. (2008) to select 12 cities
for their use-of-force study. To maximize
the utility of the two studies, we selected a
different group of sites in our study (only
one site participated in both studies). Our
selection process started with identifying
LEAs that have full deployment of CEDs
and that place CEDs in a low position on
their use-of-force continuum. Then, we
identified matched LEAs with full
deployment of CEDs that place it high on
the force continuum. To increase the
generalizability of our study, we then
selected a mix of CED type-sites, with some
placing it low on the use-of-force continuum
and some placing it high. Finally, we
selected matched LEAs that do not use
CEDs. Matching was based on criteria such
as violent crime levels, police activity
(violent crime arrests), agency size, and
population size of jurisdiction. Table 1
shows the final list of LEAs (N = 13) that
participated in our study. The table lists the
participating LEAs, and the dates we
collected data from each of these sites. For
the CED agencies, we selected data two
years before CED implementation and two
years after CED implementation. For the
non-CED agencies, we attempted to collect
data from roughly comparable periods.
As can be seen from the table, the data
collected from the CED and non-CED sites
are from the same basic time period, within
a year or two. For example, Pre-Period 1
starts in 1998 for the CED sites and 1999 in
the non-CED sites, with an overall average
of mid 2000 for CED sites and mid 2002 for
non-CED sites. The second of our pre CED
implementation periods averaged just about
mid 2001 compared to year 2003 for the
non-CED sites. The first of the post-CED

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 1: Years for which data were collected in sites before and after
CED deployment
Non-CED Sites
Site 1
Site 2
Site 3
Site 4
Site 5
Site 6
Average
CED Sites
Site 7
Site
Site
Site
Site
Site
Site

8
9
10
11
12
13

Average

Pre-Period 1
1999

Pre-Period 2
2000

Post-Period 1
2001

Post-Period 2
2002

2002
2003
2003
2004
2005

2003
2004
2004
2005
N/A

2005
2005
2005
2006
2006

2006
2006
2006
2007
2007

2002.7

2003.2

2004.7

2005.7

Pre-Period 1
1998

Pre-Period 2
1999

Post-Period 1
2001

Post-Period 2
2002

1999
2000
2000
2001
2001
2004

2000
2001
2001
2002
2002
2005

2002
2003
2005
2005
2004
2006

2003
2004
2006
2006
2005
2007

2000.4

2001.4

2003.7

2004.7

implementation periods for the CED sites
averaged just about year 2004 compared to
just about year 2005 for the non-CED sites.
The second of the post-CED implementation
periods for the CED sites averaged just
about year 2005 compared to just about year
2006 for the non-CED sites. While it might
have been preferable to have exactly the
same start and end dates for the CED and
non-CED sites, that was not possible due to
data availability issues. However, given that
we are within a year or two in most cases,
we do not believe that any bias was
introduced into the study based on temporal
considerations.
It is worth noting that the CED sites in
our sample have used CEDs for a relatively
short time frame. All of the CED sites
started using the CED weapon in the 21st
century. While this is not surprising, given
that the modern Taser that uses a nitrogen

cartridge instead of gunpowder to fire the
probes has been in use only since 1999, it
does have implications about the nature of
our study. Any conclusions that we draw
from our research reflect the early
experience with CEDs. Over time, it seems
reasonable that LEAs will gain important
insights into the use of CEDs and will be
able to attain further improvements in safety
outcomes associated with this weapon.
Table 2 (see below) presents the fouryear average for police and census data for
each of the above years for each
participating LEA. The table includes data
on the size of the residential population
served by the LEA, number of officers in
each LEA, number of arrests for violent
offenses, number of violent crimes (using
the UCR definition of violent crime), and
number of homicides. We were able to find
data for most of these measures; in cases

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

where we were not able to secure these data,
we place a dash in the cell of the table.
The main difference between the nonCED and CED sites is the participation of
one CED site that is much larger than the
other sites in our study. To address this
issue, we estimated all of our models with
and without the one unusually large site, and
found no major differences in our results.
For our residential population measure, we
found that our non-CED sites had just over
600,000 people on average (612,354), and
our CED sites had almost 2 million people
on average, but just over 700,000 if the one
largest site was excluded.13 The non-CED
and CED sites both had two agencies with
populations below 500,000. The non-CED
and CED sites each had three sites in the
500,000 to 900,000 range in population. The
non-CED and CED sites each had one site in
the one million-population range.
Next, the non-CED and CED sites
were comparable in terms of the size of each
of the agencies. All agencies selected for
participation in our study were within the
top 3% of LEAs in the United States in
terms of number of officers, with our
smallest CED site having 445 officers and
our smallest non-CED site having 308
officers. We found that our non-CED sites
had 1,324 sworn officers on average, and
our CED sites had 2,123 sworn officers, but
only 1,271 sworn officers if the one
unusually large site was excluded.
The non-CED and CED sites were
comparable in terms of the number of arrests
for violent crime by the LEAs. Non-CED
sites on average made 1,973 arrests for
violent crimes compared to 1,638 violent
crime arrests for CED sites (excluding the
13
There are some complications with examining
residential population for cities that have large
commuter populations. For example, one of our
sites has a residential population of just under
600,000; however, due to commuters from the
surrounding suburbs, its population rises to over
one million during the workweek.

one unusually large site). Three of the nonCED sites and two of the CED sites had
between 2,400 and 4,400 violent crime
arrests and one of the CED sites had 1,437
violent crime arrests. Two non-CED sites
and one CED site had between 650 and 850
violent crime arrests. Both non-CED and
CED sites included one site each with fewer
than 300 violent crime arrests.
The non-CED and CED sites were
comparable in terms of the number of
reported violent crimes. Non-CED sites had
on average 4,374 violent crimes compared
to 5,771 violent crimes for CED sites (not
counting the especially large site, where we
were not able to secure reliable data on
violent crimes for this period). The nonCED and CED sites were also comparable in
terms of the number of homicides. NonCED sites had on average 56 homicides
compared to 72 homicides for CED sites
(again, not counting the especially large site,
where we were not able to secure reliable
homicide data for this period).
The final sets of comparisons were on
demographic variables collected from the
2000 U.S. Census (see Table 2). The nonCED and CED sites were very similar on a
full range of background aggregate-level
factors. The non-CED sites averaged 8.5%
of the population below the poverty level
(ranging from 3.6% to 16.7%) compared to
10.1% for the CED sites (ranging from 3.3%
to 23.5%). The non-CED sites averaged a
household income of $50,386 (ranging from
$37,752 to $61,768) compared to $48,190
for the CED sites (ranging from $23,483 to
$71,551). The non-CED sites averaged 3.6%
unemployment (ranging from 2.1% to 6.8%)
compared to 3.9% for the CED sites
(ranging from 2.2% to 5.9%). The
population per square mile for the non-CED
sites averaged 3,782 people per square mile
(ranging from 530 to 9,316) compared to
3,466 people per square mile for the CED
sites (ranging from 831 to 10,161). The

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

percent of female-headed households for
non-CED sites averaged 7.6% (ranging from
5% to 9.9%) compared to 7.4% for CED
sites (ranging from 5.9% to 9.2%). The
percent of owner-occupied dwellings (which
sometimes is used as a measure of an area’s
economic stability) for non-CED sites
averaged 56.5% (ranging from 40% to
75.5%) compared to 56.7% for CED sites
(ranging from 34.9% to 71.7%). Regarding
the racial make-up of the jurisdictions, the
percent of non-white residents for non-CED
sites averaged 41.6% (ranging from 19% to
69%) compared to 36% for CED sites
(ranging from 30.1% to 51.3%). The
percentage of males in the population for the
non-CED sites on average was 48.3%
(ranging from 47.1% to 49.8%) compared to
49.5% for CED sites (ranging from 47.9% to
50.5%). The percentage of young people in
the population between the ages of 15 and
24 for the non-CED sites on average was
13.4% (ranging from 11.4% to 15.7%)
compared to 13.1% for CED sites (ranging
from 10.9% to 14.6%).
On balance, we believe our CED and
non-CED sites are comparable. We collected
data from fairly comparable periods for the
CED and non-CED sites, within a year or
two. The main difference between the nonCED and CED sites is the participation of
the one unusually large CED site in our
study. However, when we estimated all of
our models with and without this agency, we
found no major differences in our results.
With this site excluded from our analyses,
there are no major aggregate-level
differences between the CED and non-CED
sites across a range of variables including:
size of the residential population, number of
officers, number of arrests for violent
offenses, number of violent crimes, and
number of homicides. Further evidence of
the comparability of the CED and non-CED
sites can be seen in our analyses of the
demographic variables from the 2000 U.S.

Census. The non-CED and CED sites were
similar on a full range of background
aggregate-level factors such as population
below the poverty level, household income,
unemployment, population density, femaleheaded households, residential stability,
racial heterogeneity, percentage of males,
and youths in the population. Overall, while
differences were evident in a number of the
variables, most of the differences were
relatively small and did not seem to
introduce any substantively important
biases. Given that we were making
comparisons based on U.S. Census data (i.e.,
population data from the same data
collection system), any observed differences
represent actual differences in the
population. It would not be possible to select
places with exactly the same characteristics.
Therefore, the question is not whether there
are differences (by definition quasiexperiments represent comparisons across
non-equivalent groups), but whether the
differences appear to be substantively
important. We believe that there are no
substantively important differences between
the groups on relevant background factors.
We believe that we have reasonably
comparable CED and non-CED sites. When
combined with our multivariate analyses,
where we control for differences between
the sites, we believe that the results of our
study are interpretable.
Furthermore, we observed few
contextual changes across the sites that
might have affected the comparability of
CED and non-CED sites. For example, in
our interviews with police personnel and
review of agency documents, all of the
agencies provided detailed training for their
officers on use-of-force issues. There is little
evidence that additional refresher training
was adopted after the CED weapon was
introduced to the agency or that training
efforts were otherwise intensified across the
board after adopting CEDs. All of the

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

agencies seemed to have sound training
programs in place on use-of-force issues
during the time frame of our study. All the
agencies in our study required officers to
report all use-of-force incidents before
CEDs were introduced and there is no
evidence agencies changed their reporting
requirements during the time frames of our
study (other than reporting on issues specific
to the CED, such as how many times CEDs
were activated against suspects).

Limitations and barriers to
research
Conducting use-of-force research is a
difficult undertaking. There are a number of
barriers to conducting rigorous multi-site
research in the area of police use of force.
First, some LEAs do not systematically
maintain use-of-force data. Due to the large
number of reported crimes in most U.S.
cities and the very large number of contacts
that the police have with the public, the
opportunities for the police to use force are
vast. If the police do not have a reporting
system for clearly documenting cases
involving police use of force, the task for
researchers to do this post hoc is very
difficult. In our work, we came across LEAs
that did not have separate use-of-force forms
for officers to complete in force cases.
Instead, the force incident was recorded
within the narrative of the crime report or
arrest report, with no data field or check box
indicating that a force incident occurred. We
were not able to use these agencies in our
study, because the task of reviewing the
narratives of hundreds of thousands of
reports to identify force incidents is not
possible in a typical research project, and
while the internal affairs departments for
these agencies have separate files on force
cases that they investigate for possible
officer wrongdoing, this is only a small
percentage of cases involving force for a

typical agency. Our study was interested in
all force cases, not just those that were
investigated by an agency internal affairs
department. Thus, if an agency does not
have a use-of-force tracking system of some
type, researchers we will not typically be
able to include them in a research study on
police use of force.
Another problem is the lack of
standardization of data collection methods
for different LEAs. Some LEAs collect only
a limited number of fields on use of force,
and do not capture important information,
such as the nature of the force incident, the
nature of any injuries, the weapons available
to the officer, suspect characteristics, and
suspect actions prior to the officer’s use of
force. We were unable to use agencies that
do not collect data on factors that were
critical to our study, such as officer and
suspect injuries. Some LEAs provided
measures of suspect level of resistance, but
many did not. Consequently, this variable
had to be excluded from our analyses. As
pointed out by Smith et al. (2008), there is a
tradeoff between retaining the maximum
number of agencies for analysis and the
precision of the measures and/or the number
of measures used in the analysis. As
experienced by Smith et al. (2008) in similar
research, the data analyzed in our study
represent only records routinely captured by
LEAs and are missing many qualitative
features of the force events, such as the
nature of the incident that spurred the initial
contact between the police and the citizen
(e.g., domestic disturbance, robbery, routine
traffic stop, etc.), whether the suspect was
under the influence of drugs, and the
duration of the incident. These factors have
been shown in prior research to be
correlated with differences in the
seriousness and consequences of force
incidents (Adams, 1999; Alpert & Dunham,
2004; Kaminski & Sorensen, 1995). The
consequence of this situation, as pointed out

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

by Smith et al. (2008), is that like all
analyses outside of an experimental setting,
our models are to some degree misspecified.
In other cases, the LEA collects the
general category of force data, but codes it
in such a way that it cannot be readily
compared with force data from other
agencies. For example, instead of being able
to use a precise scale of level of injuries an
officer endured, a research team might need
to code the data simply as whether any
injury occurred (yes or no) to achieve crosssite comparability. Having more detail
regarding injuries imparts a number of
important analytical benefits, such as the
ability to model predictors of injury severity
as opposed to a more limited analysis of
whether or not an injury occurred (Smith et
al., 2008).
Also, as was done in the Smith et al.
(2008) study, we conducted an examination
of the injury narratives as a validity check
on the other injury indicators in the same
dataset. As an example, some LEAs counted
skin irritation from pepper spray and CED
dart punctures as injuries. However, this is
inconsistent with how we operationalized
injuries from these devices in this study and
the way the Smith et al. (2008)
operationalized injuries. The additional
details in the narratives allowed us to recode
these cases. (CED dart wounds to
unapproved targets, such as the groin or
face, were counted as injuries, however.)
Unfortunately, this recoding could not be
done in all datasets, due to the lack of data
in some narratives regarding injuries.
Similar coding issues arose with one
of our CED variables. As discussed earlier,
CEDs can be used in touch-stun mode or
dart-mode, and because each mode has a
different effect and is activated from
different distances from subjects, injury
patterns could vary by the mode employed
(Smith et al., 2008). Ideally, we would have
been able to measure whether a CED use

was done in touch-stun or dart-mode.
However, as in the Smith et al. (2008) study,
this was not possible, for many of the LEAs
in our study did not provide this extra level
of detail. Our only alternative was to use a
simple yes/no variable on whether a CED
was used by an officer.
Next, a large number of LEAs only
have paper records of their force data. To
include these LEAs in our research required
PERF to send a team of researchers to the
LEA site to code these paper records into a
standardized database. In addition to being
time consuming, this approach increases the
chances for errors in the data (even though
our team used various quality checks). Also,
due to the time-consuming nature of such a
task, our team was limited to taking a
random sample of cases for selected years,
as opposed to having all of the data
available. In our study, for five of the sites
we needed to code data due to the absence
of an electronic use-of-force database. Three
of the five sites were non-CED sites and two
were CED sites. We conducted statistical
tests to assess whether the use of these
different methods of data collection might
affect our results. That is, we introduced an
additional covariate (use of a random sample
of cases=0 or use of the population of
cases=1) to our logistic regression and HLM
tests for each of our outcome measures.
None of the new covariates were statistically
significant, nor did the other variables in the
model change appreciably.
Finally, due to the sensitive nature of
police force data, some LEAs feel obligated
to decline to participate in this type of
research project. Some of the cases could be
in litigation or pending litigation, and some
LEA attorneys prefer not to have those or
possibly other cases involved in a research
study. Even when the research team can
demonstrate that the data will be protected
and handled confidentially (as we did), some
LEAs might still feel compelled to err on the

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

side of caution and decline participation.
Non-participation is an issue in almost all
aspects of social science research, and it can
be particularly salient in the arena of police
use-of-force studies, due to the sensitivity
associated with the requested LEA data.

Data Collection/Measures
PERF requested hard and electronic copies
of departmental data that included: use of
force policy (past and current policies);
specific documentation of the placement of
CEDs on the department’s “use of force
continuum” policy; and all use-of-force
incident data (including cases with and
without use of CED weapons by the police)
for at least four years. The PERF team
collected the force data in one of two ways:
(1) we sent a two-person research team to
the participating LEA to conduct onsite
archival review and coding of use of force
documents, or (2) we collected electronic
use-of-force data maintained by the
participating LEA. Our team also worked
with each site to collect crime and
demographic data for each participating city.
The sources of these data were the FBI’s
Uniform Crime Report (UCR) system and
the U.S. Census.
First, there was agreement across the
agencies in our study on the definition of a
use-of-force case. The agencies counted a
case as officer use-of-force if it included any
physical strike or instrumental contact with a
person by an officer or any significant
physical contact that restricted the
movement of a person by an officer,
including the discharge of firearms, use of a
Conducted Energy Device, use of chemical
spray, use of any other weapon, choke holds
or hard hands, taking of the subject to the
ground, and deployment of a canine.
From the above data sources, we
created both departmental/organizational-

level measures and incident-level measures
for our statistical models.
Our outcome measures, at the incident
level, included the following: (the first four
below are officer measures and the final five
are suspect measures)
1. Officer injuries. This was a dichotomous
yes/no variable for any impairment of
physical condition, or pain to an officer
due to the suspect’s actions, including
physical damage produced by the
transfer of energy, such as kinetic,
thermal, chemical, electrical, and radiant
energy.
2. Officer injury severity. This was a
dichotomous minor/severe variable, in
which broken bones, stab wounds, and
gun wounds were classified as severe,
and bruises, lacerations, and burns or
punctures were classified as minor.
3. Officer injury from a force incident
requiring medical attention. This was a
yes/no variable indicating whether the
officer was seen by any type of medical
professional, such as an on-scene
emergency medical technician or
medical personnel in a hospital, related
to an officer use-of-force incident.
4. Officer injury from a force incident
requiring hospitalization. This was a
yes/no variable indicating whether an
officer was taken to a medical facility
such as hospital or medical clinic for
treatment of an injury due to a use-offorce incident. By using the term
“hospitalization” we do not mean to
being admitted to a hospital for an
overnight stay; information was not
available regarding how many of these
incidents resulted in an overnight stay,
as opposed to an outpatient evaluation
and/or treatment.
5. Suspect injuries. As with the first
outcome measure, officer injuries, this
was a dichotomous yes/no variable for
any impairment of physical condition, or

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

pain due to an officer’s actions,
including physical damage produced by
the transfer of energy, such as kinetic,
thermal, chemical, electrical, and radiant
energy.
Suspect injury severity. This was a
dichotomous minor/severe variable, in
which broken bones, stab wounds, and
gun wounds were classified as severe,
and bruises, lacerations, and burns or
punctures were classified as minor.
Suspect deaths. This was a dichotomous
yes/no variable indicating whether the
suspect died during or as a result of an
officer use-of-force incident. We had no
officer deaths in our sample and
therefore did not assemble a similar
officer death measure.
Suspect injury from a force incident
requiring medical attention. This was a
yes/no variable indicating whether the
suspect was seen by any type of medical
professional, such as an on-scene
emergency medical technician or
medical personnel in a hospital, related
to an officer use-of-force incident.
Suspect injury from a force incident
requiring hospitalization. This was a
yes/no variable indicating whether a
suspect was taken to a medical facility
such as hospital, medical clinic or
medical facility within a custodial
environment for treatment of an injury
due to the use-of-force incident. Again,
our use of the term “hospitalization”
does not imply being admitted to a
hospital for an overnight stay;
information was not available regarding
how many of these incidents resulted in
an overnight stay, as opposed to an
outpatient evaluation and/or treatment.

One of the concerns was making sure
to standardize across the datasets for all the
agencies on the minimum occurrence that
would be defined as an “injury.” For

example, if one agency defines any “handson” activity by an officer as an injury, and
another has a minimum threshold of a
physical sign such as a bruise, cut, or scrape,
we would have a problem making
comparisons across these agencies using
different definitions. A review of each
agency’s reporting policies on use of force
showed general agreement that an injury
could be any impairment of physical
condition, or pain. We also confirmed this
definition by reading narratives at agencies
that collected narratives on injuries. None of
the agencies counted “mental injuries.”
Also, none of the agencies distinguished
between accidental injuries (e.g., the suspect
accidentally trips after being handcuffed)
and injuries caused by the officer’s
deliberate actions. All the agencies counted
both of these types of injuries. Therefore,
while the agencies varied in the level of
detail they collected and coded regarding the
nature of the injuries, there was agreement
on the basic definition of an injury used by
the agencies.
Our individual-level covariate
measures were intended to help control for
potentially important incident-level
differences across our participating
departments that might explain our outcome
measures. While we would have liked to
include in our statistical models a full range
of incident-level factors (e.g., suspect
demeanor, suspect alcohol/drug impairment,
and size of suspect relative to the size of the
officer) that have received empirical support
in prior use-of-force research (see Garner,
Maxwell, and Heraux, 2002), only a limited
number of variables were available to our
team based on agency records. However,
each of the variables that were available and
included in our models were either shown to
be important predictors of use of force in
prior research (see Garner et al., 2002) or
had the potential to be important. Two of the
variables (weapons and physical aggression

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

by the suspect) represent situational factors
that might influence whether an officer
might use force. The other three variables
are demographic factors that might be
associated with whether an officer might use
force (race, gender and age). Our
individual-level control/independent
variables included the following:
1. Suspect race (1= white and 0=nonwhite)
2. Suspect gender (male=1, female=0)
3. Suspect age (under 25 years old=1 and
over 25 years old=0)
4. Suspect behavior (whether the suspect
used physical aggression or physical
resistance of any type against the
officer)
5. Suspect use of any weapon (yes=1,
no=0)
Our aggregate-level measures were
intended to help control for contextual
differences across the participating sites that
might impact our outcome measures. Our
aggregate/departmental-level
(control/independent variable) measures
included the following:
1. Total # of sworn officers per 100,000
people in the population
2. Total # of arrests per 100,000 people in
the population
3. Total # of violent crime arrests per
100,000 people in the population
4. Total # of Part I UCR crimes per
100,000 people in the population
5. Total # of violent crimes per 100,000
people in the population
6. Total # of homicides per 100,000 people
in the population
7. Percentage of the population in the
jurisdiction below poverty
8. Median household income for the
population in the jurisdiction

9. Percent unemployed for the population
in the jurisdiction
10. Population size/density for the
population in the jurisdiction
11. Percent female-headed household with
children for the population in the
jurisdiction
12. Residential stability for the population
in the jurisdiction
13. Racial heterogeneity for the population
in the jurisdiction
14. Percent male for the population in the
jurisdiction
15. Percent aged 15 to 24 for the population
in the jurisdiction

Data Analysis
Descriptive/bivariate analyses – First, we
cleaned all the data using standard datacleaning processes to verify that the data are
correct and conform to a set of rules. We
wrote SPSS programs to remove errors and
inconsistencies in all data files. Our first sets
of analyses are descriptive statistics for all
the main study variables for the entire
sample (CED and non-CED site data
combined). Second, we provide a graphical
presentation of our bivariate results for the
CED versus non-CED site comparisons on
our outcome measures.
Multivariate analyses – Because
QEDs involve comparison groups of
unknown equivalence and tend to involve
many different but interlocking relationships
between variables, the development of
statistical models becomes a critical process.
Statistical models will control for possible
pre-treatment differences between
departments with CEDs and those without
CEDs that could affect our outcome
measures. A variety of modeling techniques
exist (see Asher 1983), and a major problem
in analyzing data from QEDs is model
misspecification that can lead to biased

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

estimates of treatment effects (Trochim,
Cappelleri, and Reichardt, 1991). Modeling
and theory will allow us to identify and
remove from our models spurious variables
that do not help predict the relationship
between CED use/policies and our
outcomes. It will also help find suppression
effects when part of a variable affects part of
another variable even though the bivariate
relationship is not statistically significant.
To address the incident-level part of
our data within each department, we will
begin with the use of logistic regression. As
discussed earlier, to assure the use of
standard measures across all of our sites we
were required to dichotomize our outcome
measures. Logistic regression is an
appropriate technique to assess such binary
outcome measures. Logistic regression
allows us to include an enormous amount of
information, which will be necessary to
control for all the potential confounding
factors between police departments.
One of the concerns in analyzing data
across multiple sites is the clustering/nesting
of data. Nesting occurs when a unit of
measurement is a subset of a larger unit and
the units clustered in the larger unit might be
correlated. In our study, individual cases of
weapon use by officers are nested within
specific police departments that have
varying policy guidelines on the use of
force. Ignoring the nested structure of our
data (e.g., conducting only logistic
regressions) can potentially lead to biased
estimates. In the past, hierarchical data were
analyzed using conventional regressions, but
these techniques can yield biased standard
errors and potentially spurious results (Hox,
2002). That is, using uni-level analysis
methods on multilevel data can lead to
parameter estimates that are unbiased but
inefficient, and the standard errors are
negatively biased, which results in
spuriously ‘significant’ effects (see De

Leeuw and Kreft, 1986; Snijders and
Bosker, 1999; Hox, 1998, 2002).
First, the background circumstances of
the individual cases of use of force may vary
appreciably from department to department.
Factors such as these can give rise to a
degree of dependency or similarity among
the observations nested within a department.
Ignoring such dependencies (i.e., the intraclass correlational structure of multi-site
data) can result in deflated standard errors
for treatment effect estimates (Hox, 2002).
Moreover, if we ignore the nesting of
individuals in different departments in our
analyses, we run the risk of inadvertently
concealing potentially substantial betweendepartment heterogeneity in the effects of
CEDs. Such heterogeneity is likely given
that the sites are likely to vary considerably.
We will use Hierarchical Linear
Modeling (HLM) to obtain more appropriate
standard errors for estimates of CEDs’
effects.14 HLM will be used to assess how
differences in agency-level and incidentlevel factors relate to differences in
outcomes across departments. HLM will be
used to assess differences in use of force and
respective outcomes pre- and post-CED
deployment.
We used HLM 6 software (developed
by Raudenbush et al., 2004). HLM provides
14 While our primary strategy was to run HLM
models to address this clustering issue, before that
we examined the results of using a logistic
regression with a robust variance estimate to adjust
for within-cluster correlation. We conducted these
analyses using Stata statistical software with the
vce (cluster clustvar) option. The robust variance
estimator comes under various names in the
literature and within the Stata software it is known
as the Huber/White/sandwich estimate of variance.
The names Huber and White refer to the seminal
references for this estimator (Huber, 1967; White,
1980). The main limitation with using this approach
is that we do not get aggregate-level coefficients
like those produced using HLM, but we are still able
to address the clustered nature of our data and
produce unbiased estimates (Rogers, 1993;
Williams, 2000; Wooldridge, 2002). We use this
approach to examine more closely the viability of
the standard logistic regression results, before
examining a full multi-level HLM approach.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

a conceptual framework and a flexible set of
analytic tools to analyze the special
requirements of our nested data. At level 1
of an HLM the analysis an outcome variable
is predicted as a function of a linear
combination of one or more level 1
variables, plus an intercept, as so:

where β0j represents the intercept of group j,
β1j represents the slope of variable X1 of
group j, and rij represents the residual for
individual i within group j. On subsequent
levels, the level 1 slope(s) and intercept
become dependent variables for level 2:

and so forth,
where

and

are intercepts, and

and
represent slopes predicting β0j and
β1j respectively from variable W1. Through
this process, we accurately model the effects
of level 1 variables (i.e., individual/incidentlevel variables) on the outcome (e.g.,
injuries), and the effects of level 2 variables
(i.e., aggregate/site-level variables such as
agency policy on use of CEDs) on the
outcome. We will be examining differences
in the above outcomes across two annual
points in time after the CEDs were
implemented, controlling for any observed
pre-test differences in the comparison
groups during the two year period before
CEDs were implemented.
Statistical power: Statistical power
provides an estimate of how often one
would fail to identify a relationship that in
fact existed (Weisburd, Petrosino and
Mason, 1991; Cohen, 1988). Power is
jointly determined by sample size and effect
size. One of the most widely accepted
methods of evaluating effect sizes is

Cohen’s formulation (treatment mean control mean/shared variance): small
effects= .25, medium effects= .75, and large
effects= 1.25. For our logistic regression
models, where we are working with just the
incident-level data, we have more than
enough statistical power. That is, we have
thousands of use-of-force cases to analyze.
At the individual/incident-level, our study
would be able to detect small effect size
differences (< 0.10) across the CED versus
non-CED comparison groups (assuming an
alpha of .05, a two-sided test and power of
.95). What this means is that with our
sample size, we have power of greater than
95% to yield statistically significant results
even when the differences in proportions
between the CED and non-CED sites are
less than 10%.
However, for our HLM analyses,
where we explicitly model the nested nature
of our data, we only have 13 higher-level
units. Given our need to use HLM, we
conducted statistical power calculations to
assess whether we had enough cases in our
analysis to find statistically significant
differences in CED (n=7) and non-CED
(n=6) agencies if they existed. We used
computer routines developed by
Raudenbush and Liu (2000) to calculate
statistical power for our HLM test. This
program calculates approximate standard
errors and optimal sample sizes for estimates
of fixed effect parameters with multiple
levels. Our study with 13 departments is
able to detect only large effect size
differences with a power level of .80
(assuming an alpha of .05, a one-sided test,
Level-1 residual variance of 25, Level-2
residual variance of 10, and an intra-class
correlation coefficient of .15). Overall, we
have less statistical power to assess
differences when explicitly modeling the
nested nature of our data through HLM than
when we conduct the logistic regressions
and examine only the incident/individual-

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

level data. That is, for the HLM analyses we
were only able to detect large effects as
opposed to our ability to detect very small
effect size differences when we conducted
our logistic regressions. Given the difficulty
of on-site data collection and associated
costs, we were not able to increase the
number of departments beyond 13. Our
purpose in using HLM is to assess the nested
structure of our data and assure that our

logistic regression estimates are not biased.
We were able to produce unbiased HLM
estimates – just with less statistical power
across departments than within departments.
In our HLM results, we were looking to
confirm the direction and magnitude of our
logistic regression results, and relied less on
our tests of statistical significance (where
our statistical power was fairly modest).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

CHAPTER 4:
Study Results

he first sets of analyses include
descriptive statistics for all the main
study variables. The second sets of analyses
are our multivariate analyses using logistic
regression. The third sets of analyses are our
multi-level analyses using HLM to address
potential nesting effects due to the fact that
the individual use-of-force cases we analyze
are clustered within 13 departments.

Descriptive statistics:
Univariate results: In Table 4 (see below),
we present univariate results for each of our
main outcome measures, comparing our preand post-test results for our CED sites and
non-CED sites. In the section below, we
review the overall sample results (CED and
non-CED combined) to provide context for
our later analyses. Immediately after this
section, we conduct statistical tests to
compare the pre- and post-CED
implementation results for CED versus nonCED sites and present the results
graphically.
Generally, our data suggest that the
vast majority of officers are not injured in
use-of-force cases (see Table 4). For the
CED and non-CED sites, our data suggest
that 11% of officers were injured in use-offorce cases in the pre-period and 9% in the
post–period; and suspect injuries were more
common in use-of-force cases (for the CED
and non-CED sites 24% in the pre-period
and 29% in the post-period) than officer
injuries. Our data suggest that medical

attention for officer injuries (pre-period 11%
and 8% post-period) was much less common
than medical attention for suspect injuries
(pre-period 51% and 41% post-period).
Likewise, hospitalization for officer injuries
(pre-period 4.1% and 4.3% post-period) was
less common than hospitalization for suspect
injuries (pre-period 28% and 17% postperiod). Our data suggest that the proportion
of officers receiving severe injuries (4.5%
pre-period and 5.6% post-period) was
similar to the same measure for suspects
(6.7% pre-period and 5.6% post-period).
There were no recorded officer deaths in our
sample, and fewer than 1% of the use-offorce cases in our sample had a suspect
death (0.3% for the pre-period and 0.4% for
the post-period).
The proportion of white suspects was
just under one-third for the whole sample
(32.7% in the pre-period and 30.7% in the
post period) or conversely the non-white
sample was just over two-thirds. The
proportion of male suspects was over 85%
across both time periods for the whole
sample (85.1% in the pre-period and 86.3%
in the post-period). The proportion of
suspects under 25 years old was more than
one-third for the whole sample (38.7% in the
pre-period and 39.3% in the post-period).
Our data indicate that the proportion of
suspects using physical aggression against
officers was about one-third for the whole
sample (32.5% in the pre-period and 34.2%
in the post-period). Our data indicate that the
proportion of suspects with a weapon at the

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 4: Descriptive Statistics for non-CED versus CED sites and all sites: Percentages for all study variables
Non-CED Site

CED Site

All sites

N for all sites

Officer injury (pre period)
Officer injury (post period)
Suspect injury (pre period)
Suspect injury (post period)
Medical attention for officer injuries (pre period)
Medical attention for officer injuries (post period)
Medical attention for suspect injuries (pre period)
Medical attention for suspect injuries (post period)
Hospitalization for officer injuries (pre period)
Hospitalization for officer injuries (post period)
Hospitalization for suspect injuries (pre period)
Hospitalization for suspect injuries (post period)
Officer severe injury (pre period)
Officer severe injury (post period)
Suspect severe injury (pre period)
Suspect severe injury (post period)
Suspect deaths in force incidents (pre period)
Suspect deaths in force incidents (post period)
Suspects White/Caucasian (pre period)
Suspects White/Caucasian (post period)
Suspects male (pre period)
Suspects male (post period)
Suspects under 25 years old (pre period)
Suspects under 25 years old (post period)
Suspect used physical aggression against officer (pre period)
Suspect used physical aggression against officer (post period)
Suspect had weapon (pre period)
Suspect had weapon (post period)
Officer used CEDs only against suspect (pre period)
Officer used CEDs only against suspect (post period)
Officer used baton only against suspect (pre period)
Officer used baton only against suspect (post period)
Officer used OC spray only against suspect (pre period)
Officer used OC spray only against suspect (post period)

10.3
20.3
29.9
42.5
3.5
15.9
35.2
53.2
3.3
6.3
30.5
36.3
7.0
6.4
7.3
7.2
0.9
0.9
43.8
35.0
85.0
84.9
38.1
40.1
30.7
23.2
27.7
50.5
0.0
0.0
4.1
7.0
11.3
16.2

11.5
8.3
22.8
26.6
13.2
7.5
54.8
39.8
4.3
4.1
26.8
16.2
4.0
5.3
6.5
5.0
0.2
0.4
30.9
30.3
85.1
86.5
38.9
39.2
35.8
37.9
16.3
10.7
0.0
11.1
1.4
0.8
13.8
8.1

11.3
9.4
24.4
29.4
11.3
8.2
51.3
40.8
4.1
4.3
27.5
16.9
4.5
5.6
6.7
5.6
0.3
0.4
32.7
30.7
85.1
86.3
38.7
39.3
32.5
34.2
19.5
15.7
0.0
9.6
2.0
1.7
13.3
9.2

1,058
7,670
2,234
9,131
910
6,521
1,068
8,944
847
6,513
762
8,875
1,058
7,670
2,234
9,131
1,952
9,279
1,379
11,922
2,330
12,067
2,124
8,873
2,237
3,892
1,416
6,444
2,350
11,797
2,350
11,797
2,350
11,797

Officer used some weapon other than CEDs,OC,batons or used multiple
weapons involving a CED, OC, or baton (pre period)

55.4

27.6

33.5

2,350

Officer used some weapon other than CEDs,OC,batons or used multiple
weapons involving a CED, OC, or baton (post period)
Officer used other form of non-weapon force (pre period)
Officer used other form of non-weapon force (post period)

67.5
29.1
9.3

38.3
54.8
41.6

42.3
49.3
37.2

11,797
2,350
11,797

force incident was under 20% for the whole
sample (19.5% in the pre-period and 15.7%
in the post-period).
In terms of actual weapon use, no one
in the sample used CEDs in the pre-period.
In the CED sites, our data indicate that 11%
of their force cases involved use of a CED
only. For both CED and non-CED sites, our
data indicate that use of batons by
themselves is not common (2% in the preperiod and 1.7% in the post-period). For the
sample as a whole, our results suggest that
use of only OC spray is more common (13%

in the pre-period and 9% in the post-period)
than CED and baton use. For the sample as a
whole, our data indicate that solo use of
weapons other than CEDs, batons, or OC
spray (or multiple weapon use involving
CEDs, batons, OC spray or some other
weapon) occurs in over one-third of the
force cases (34% in the pre-period and 42%
in the post-period). We also found evidence
that officer use of non-weapon force (e.g.,
hands-on tactics) is common in force
incidents (49% in the pre-period and 37% in
the post-period).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Bivariate results: The next sets of
findings are for our bivariate results
comparing outcomes for the CED and nonCED sites. Below we present the bivariate
results graphically, and present chi-square
statistics (in the text within parentheses) to
assess the statistical significance of our
bivariate comparisons.
Officer injuries: Our first chart
explores differences between CED and nonCED sites on the proportion of use-of-force
cases where an officer was injured before
CEDs were implemented and after CEDs
were implemented. Before the CED sites
had deployed CEDs, our data suggest that
11.5% of the officers were injured in force
cases compared to a similar proportion of
officers in the non-CED sites (10.3%) over
the same reference period, representing no
statistical difference (X2= 0.78, df=1,
p=.38). However, we found that the CED
sites observed a reduction in officer injuries
(8.3%) after they began their deployment of
CEDs, while the non-CED sites observed an
increase in officer injuries to 20.3% (X2=
52.68, df=1, p<.001).
Suspect injuries: Before the CED
sites deployed CEDs, our data suggest that
22.8% of their suspects were injured in force
cases, compared to a slightly higher
proportion of suspects in the non-CED sites
(29.9%) over the same reference period,
representing a statistically significant
difference (X2= 23.68, df=1, p<.001). The
CED sites observed a small increase in
suspect injuries (26%) after they began their
deployment of CEDs, while the non-CED
sites observed a much larger increase in
suspect injuries to 42.5% (X2= 102.02, df=1,
p<.001). While the CEDs started out at a
slightly lower rate of suspect injuries
compared to the non-CED sites (22.8% to
29.9%), our data suggest that the CED sites
were substantially lower at the post-period
(26% to 42.5%), at a rate much greater than
the initial differences would predict.

Officer injury requiring medical
attention: Before the CED sites deployed
CEDs, our data suggest that 13.2% of their
officers received medical attention for
injuries in force cases compared to a lower
proportion of officers in the non-CED sites
(3.5%) over the same reference period,
representing a statistically significant
difference (X2= 45.07, df=1, p<.001). The
CED sites observed a large decrease in
officers receiving medical attention for
injuries (7.5%) after they began their
deployment of CEDs, while the non-CED
sites observed a large increase in officers
receiving medical attention for injuries to
15.9% (X2= 29.78, df=1, p<.001). While the
CEDs started out at a higher rate of officers
receiving medical attention for injuries
compared to the non-CED sites (13.2% to
3.5%), our data indicate that the CED sites
were substantially lower in the post-period
(7.5% to 15.9%).
Suspect injury requiring medical
attention: Before the CED sites deployed
CEDs, our data suggest that 54.8% of their
suspects received medical attention for
injuries in force cases, compared to a lower
proportion of suspects in the non-CED sites
(35.2%) over the same reference period,
representing a statistically significant
difference (X2= 72.68, df=1, p<.001). The
CED sites observed a large decrease in
suspects receiving medical attention for
injuries (39.8%) after they began their
deployment of CEDs, compared to the nonCED sites that observed a large increase in
suspects receiving medical attention for
injuries to 53.2% (X2= 33.97, df=1, p<.001).
While the CEDs started out at a higher rate
of suspects receiving medical attention for
injuries compared to the non-CED sites
(54.8% to 35.2%), our data suggest that the
CED sites were substantially lower at the
post-period (39.8% to 53.2%).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer injuries: CED vs Non-CED Sites

% of officers injured.

50.0%
40.0%
30.0%

CED site
Non-CED site
20.3%

20.0%
11.5%
10.3%

10.0%

8.3%

0.0%
Pre-period

Post-period

Suspect injuries: CED vs Non-CED Sites
50.0%
% suspects injured.

CED site

40.0%

42.5%

Non-CED site

30.0%

29.9%

20.0%

22.8%

26.6%

10.0%
0.0%
Pre-period

Post-period

% receiving medical attention.

Officers receiving medical attention
50.0%
CED site

40.0%

Non-CED site

30.0%
20.0%
15.9%

13.2%
10.0%
0.0%

7.5%
3.5%
Pre-period

Post-period

% receiving medical attention.

Suspects receiving an injury requiring medical attention

60.0%

54.8%

50.0%
40.0%
30.0%

35.2%

53.2%
39.8%

CED site

20.0%

NonCED site

10.0%
0.0%
Pre-period

Post-period

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer injury requiring
hospitalization: Before the CED sites
deployed CEDs, our data suggest that 4.3%
of the officers required hospitalization for
injuries in force cases, compared to a similar
proportion of officers requiring
hospitalization in the non-CED sites (3.3%)
over the same reference period, representing
no statistical difference (X2= 0.89, df=1,
p=.35). The CED sites observed a very small
decrease in officers requiring hospitalization
for injuries (4.1%) after they began their
deployment of CEDs, compared to the nonCED sites that observed an increase in
officer requiring hospitalization for injuries
to 6.3% (X2= 3.9, df=1, p<.05). The CEDs
started out at a similar rate of officers
requiring hospitalization for injuries
compared to the non-CED sites (3.3% to
4.3%), but the CED sites were significantly
lower at the post-period (4.1% to 6.3%).
Suspect injury requiring
hospitalization: Before the CED sites
deployed CEDs, our data suggest that 26.8%
of their suspects required hospitalization for
injuries in force cases, compared to a similar
proportion of suspects requiring
hospitalization in the non-CED sites (30.5%)
over the same reference period, representing
no statistical difference (X2= 2.57, df=1,
p=.11). The CED sites observed a large
decrease in suspects requiring
hospitalization for injuries (16.2%) after
they began their deployment of CEDs,
compared to the non-CED sites that
observed a small increase in suspects
requiring hospitalization for injuries to
36.3% (X2= 61.59, df=1, p<.05). The CEDs
started out at a similar rate of suspects
requiring hospitalization for injuries
compared to the non-CED sites (26.8% to
30.5%), but our data suggest that the CED
sites were significantly lower at the postperiod (16.2% to 36.3%).

Officer severe injuries: Before the
CED sites deployed CEDs, our data suggest
that 4% of their officers were severely
injured in force, cases compared to a similar
proportion of officers in the non-CED sites
(7%) over the same reference period,
representing no statistical difference (X2=
1.32, df=1, p=.25). Our data also suggest
that the CED sites observed no significant
change in officer severe injuries (5%) after
they began their deployment of CEDs,
compared to the non-CED sites that
observed no change in officer severe injuries
for the non-CED sites to 6.4% (X2= 0.20,
df=1, p=.66).
Suspect severe injuries: Before the
CED sites deployed CEDs, our data suggest
that 6.5% of their suspects were severely
injured in force cases, compared to a similar
proportion of suspects in the non-CED sites
(7.3%) over the same reference period,
representing no statistical difference (X2=
0.23, df=1, p=.63). However, our data
suggest that the CED sites observed a
reduction in suspects’ severe injuries (5%)
after they began their deployment of CEDs,
compared to the non-CED sites that
observed no change in suspect severe
injuries to 7.3% (X2= 3.75, df=1, p<.05).
Suspect deaths: Due to the absence of
any officer deaths in our sample, we were
only able to examine suspect deaths
occurring in relation to a police use-of-force
incident. We had 28 suspect deaths in our
sample, and our analysis of these cases is
limited to the bivariate results we present
below (multivariate modeling was not
possible with our suspect death variable).
Before the CED sites deployed CEDs, our
data suggest that 0.2% of the suspects were
killed in force cases, compared to a higher
proportion of suspects in the non-CED sites
(0.9%) over the same reference period,
representing a small statistical difference
(X2= 8.7, df=1, p<.01). The CED sites

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer injury requiring hospitalization

% hospitalization.

50.0%
40.0%
30.0%
20.0%

CED site

10.0%
0.0%

4.3%
3.3%
Pre-period

Non-CED site

6.3%
4.1%
Post-period

Suspect injury requiring hospitalization
50.0%
% hospitalized.

40.0%
30.0%

36.3%

30.5%
26.8%

20.0%

16.2%

10.0%

CED site
Non-CED site

0.0%
Pre-period

Post-period

Officer severely injured

% severely injured.

50.0%
40.0%
30.0%
20.0%
CED site

10.0%

7.0%

0.0%

4.0%
Pre-period

6.4%
5.3%

Non-CED site

Post-period

% suspects severely injured.

Suspect severely injured
50.0%
40.0%
30.0%
20.0%
10.0%

CED site

7.3%
6.5%

7.2%
5.0%

Non-CED site

0.0%
Pre-period

Post-period

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

observed about the same number of suspects
killed in force incidents (0.4%) after they
began their deployment of CEDs, compared
to the non-CED sites that also observed no
change in the number of suspects killed in
force incidents (0.9%). While the postperiod results for CEDs (0.4%) and nonCEDs (0.9%) represents a small statistical
difference (X2= 4.02, df=1, p<.05) we do not
believe this difference is necessarily
attributable to the presence of CEDs (more
than likely it is just random noise in the
data). The CED sites started out at a lower
rate of suspect deaths compared to the nonCED sites (0.2% vs. 0.9%), and this
difference simply held up over the post
period (CED sites= 0.4% to non-CED sites=
15
0.9%). On balance, our data suggest that
CEDs do not appear to have much of an
effect on suspect deaths. That is, while at the
post-test the CED sites had fewer suspect
deaths than the non-CED sites, this seems to
reflect the fact that the CED sites had fewer
suspect deaths prior to the deployment of
CEDs.

15 While we are very concerned about small sample
size for this analysis, we did attempt to estimate a
logistic regression to assess whether there was a
statistical change from pre to post for the CED
compared to non-CED sites. We found no statistical
difference (B= -.02, p=.98) (further substantiating
our conclusion above that there was no difference
between CED and non-CED sites on the outcome of
suspect deaths).

Intercept
Does Agency Deploy
CED (1=yes, 0=no)
Time frame of incident
(post-CED/comparable
period=1, pre-CED/
comparable period=0)
Interaction CED * Time
Frame (1=CED and
post period)
Suspect race (White=1,
Non-White=0)
Suspect gender
(Male=1, female=0)
Suspect age (1=<25
years old, 0=>25 years
old)

-5.04
-1.79

Odds
Ratio
0.01
0.17

SE
1.18
1.21

P
value
0.00
0.14

0.22

1.25

0.60

0.71

-0.02

0.98

0.77

0.98

0.21

1.23

0.39

0.60

0.80

2.22

0.74

0.28

-1.13

0.32

0.49

0.02

To confirm our conclusion on suspect
deaths, we also examined whether officer
use of guns in force cases (including cases
where suspects died as well as cases in
which suspects lived) changed for the CED
sites compared to the non-CED sites. We did
not find evidence that CEDs had an effect on
the proportion of use-of-force cases where
an officer used a firearm. Before
implementation of CEDs, our data suggest
that the CED sites had less than one percent
of their cases (0.6%) involving an officer
using a firearm. After CED implementation,
our data suggest that this number remained
the same statistically and below one percent
(0.9%). During the same period, the nonCED sites did not statistically change either
on this measure. The non-CED sites
observed about two percent of their cases
(2.2%) involving an officer using a firearm
at the pre-test period, and observed no
statistical change in the number of officers
using firearms in force incidents at the postperiod (2.9%). On balance, our data suggest
that CEDs do not appear to have much of an
effect on officer use of firearms in force
incidents, affirming the above finding
regarding suspect deaths. Due to the small
sample size of use of force cases involving
firearms (overall, only about 1% of our
cases involved an officer using a firearm)
we limit our analyses of the firearms data to
16
these bivariate results.

16 While we are very concerned about small sample
size for this analysis on officer gun use, we did
attempt to estimate a logistic regression to assess
whether there was a statistical change from pre to
post for the CED compared to non-CED sites on
officer use of guns. We found no statistical
difference (B= -.004, odds ratio= .996, SE= .44,
p=.99) (further substantiating our conclusion above
that there was no difference between CED and nonCED sites on the outcome of officer use of gun in
force cases).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

% of suspects killed by officers.

Suspect deaths
1.00%

0.75%

0.50%
CED site

0.25%

0.00%

0.09%
0.02%
Pre-period

Comparing types of use-offorce by the police for CED
sites only (and post-test only):
The next set of analyses focus on just the
participating CED sites. For these analyses,
we examine the period after CEDs have
been deployed, comparing the actual use of
CEDs by officers to other forms of use of
force. Obviously we can only examine the
use of CEDs in the post-period (after CEDs
were introduced within the LEA), but we
also limit our analyses of the non-CED force
cases to the post-period to remove any
potential temporal effects on our
comparisons. That is, it could be potentially
problematic to have the CED site data
covering only a two-year period and the
non-CED site data covering four years. Our
first sets of analyses are bivariate models.
Our second sets of analyses are multivariate
models to confirm the earlier bivariate
results. For these models, we coded our useof-force data into five categories: CED use
only, baton use only, OC spray use only,
other weapon use or multiple weapon use,
and non-weapon force by officers (hands-on
tactics and other non-weapon approaches).
Suspect injuries: For suspect injuries
(see chart below), non-weapon/hands-on
tactics were associated with the highest
levels of suspect injuries (45%), followed by
batons (44%), CEDs (44%), multiple
weapons (or weapons other than CEDs,

0.09%
0.04%

Non-CED site

Post-period

batons, or OC spray) (15.9%), and OC spray
(7.6%). For these analyses, our data suggest
that OC spray and multiple/other weapons
were associated with significantly lower
suspect injuries than the other forms of force
(X2= 570.25, df=4, p<.001).
Officer injuries: For officer injuries
(see chart below), batons (24.3%) and nonweapon/hands-on tactics (20.5%) were
associated with the highest levels of officer
injuries, followed by OC spray (6.3%),
CEDs (5.4%), and multiple weapons (or
weapons other than CEDs, batons, or OC
spray) (3.4%). For these analyses, our data
suggest that multiple/other weapons, CEDs,
and OC spray and were associated with
significantly lower officer injuries than the
other forms of force (X2= 264.97, df=4,
p<.001).
Injuries requiring medical attention
for the suspect: For suspect medical
attention (see chart below), batons (62.5%),
CEDs (58%) and non-weapon/hands-on
tactics (55.7%) were associated with the
highest levels of suspects requiring medical
attention. For these analyses, our data
suggest that multiple/other weapons and OC
spray were associated with significantly
lower number of cases where suspects
required medical attention than the other
forms of force (X2= 644.98, df=4, p<.001).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

44.0%

45.0%

44.0%

40.0%
30.0%
20.0%

15.9%
7.6%

10.0%
0.0%
CED

Baton

OC Spray

Multiple/other
weapons

Other than
weapon

Officer injury: Post-test only and CED sites only
50.0%
45.0%
40.0%
35.0%
Officer
injury %

30.0%
24.3%

25.0%

20.5%
20.0%
15.0%
10.0%
5.0%

6.3%

5.4%

3.4%

0.0%
CED

Baton

OC Spray

Multiple/other
Weapons

Other than
weapon

Suspect medical attention: Post-test only and CED sites only
100.0%
90.0%
80.0%

Medical attention %

70.0%
60.0%

58.0%

62.5%
55.7%

50.0%

44.2%

40.0%
30.0%

24.1%

20.0%
10.0%
0.0%
CED

Baton

OC Spray

Multiple weapons

Other than
weapon

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Injuries requiring medical attention
for the officer: For officer medical attention
(see chart below), OC spray (12.6%), batons
(12.3%), and other than weapons (8.9%)
were associated with the highest levels of
officers requiring medical attention. For
these analyses, our data suggest that
multiple/other weapons and CEDs were
associated with significantly lower number
of cases where officers required medical
attention than the other forms of force
(X2= 56.19, df=4, p<.001).
Injuries requiring hospitalization
for the suspect: For suspect hospitalization
(see chart below), CEDs (29.5%), batons
(19.7%), and non-weapon/hands-on tactics
(16.7%) were associated with the highest
levels of suspects requiring hospitalization.
For these analyses, our data suggest that OC
spray (11.2%) and multiple/other weapons
(12.3%) were associated with significantly
lower number of cases where suspects
required medical attention than the other
forms of force (X2= 126.77, df=4, p<.001).
Injuries requiring hospitalization of
the officer: For officer hospitalization (see
chart below), our data suggest that there
were no statistically significant differences
across the various forms of use-of-force for
officers requiring hospitalization from a
force-related injury (X2= 2.72, df=4, p=.61).

Suspect severe injuries: Our data
suggest that for suspect severe injuries (see
chart below), CEDs (2%) and OC spray
(2.5%) were associated with lower levels of
suspect severe injuries than multiple/other
weapons (6.3%) and batons (5.9%) (X2=
9.88, df=4, p<.05).
Officer severe injuries: For the
officer severe injury variable (see chart
below), our data suggest that there were no
statistically significant differences across the
various forms of use-of-force for officers
receiving severe injuries (X2= 8.49, df=4,
p=.08).
Overall, within our seven CED sites,
our data suggest that OC spray was
associated with the best outcomes. That is,
for six of the eight comparisons, the cases
where an officer uses OC spray were
associated with the lowest or second lowest
rate of injury or medical attention/
hospitalization. For five of the eight
comparisons, our data suggest that the cases
where an officer uses a CED were
associated with the lowest or second lowest
rate of injury or medical attention/
hospitalization. For three of the eight
comparisons, our data suggest that the cases
where an officer uses a baton were
associated with the highest rate of injury or
medical attention/hospitalization.

Officer medical attention: Post-test only and CED sites only
25.0%

20.0%

Medical 15.0%
attention %

12.3%

12.6%

10.0%

5.0%

8.9%

3.9%

3.4%

0.0%
CED

Baton

OC Spray

Multiple weapons

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Other than
weapon

Suspect hospitalization: Post-test only and CED sites only
50.0%
45.0%
40.0%
% Hospitalized

35.0%
30.0%

29.5%

25.0%
19.7%

20.0%

16.7%

15.0%

11.2%

12.3%

10.0%
5.0%
0.0%
CED

Baton

OC Spray

Multiple weapons

Other than
weapon

Officer hospitalization: Post-test only and CED sites only
50.0%
45.0%
40.0%
35.0%
30.0%

% Hospitalized 25.0%
20.0%
15.0%
10.0%
5.0%

7.0%
3.6%

2.9%

4.3%

4.2%

0.0%
CED

Baton

OC Spray

Multiple/other
Weapons

Other than
weapon

Suspect severe injury: Post-test only and CED sites only
10.0%
9.0%
8.0%
7.0%
% severe injury

6.3%

5.9%

6.0%
5.0%
4.0%

3.6%

3.0%
2.0%

2.5%
2.0%

1.0%
0.0%
CED

Baton

OC Spray

Multiple weapons

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Other than
weapon

Officer severe injury: Post-test only and CED sites only
10.0%
8.8%

9.0%
8.0%
7.0%
6.0%

Severe injury %

4.6%

5.0%
4.0%
3.0%
2.0%
1.0%
0.0%

0.0%

CED

Baton

Logistic regression model for “CED
sites only” analyses: Next, we explore the
above “CED sites only” results with a
multivariate logistic regression model where
we introduce variables to control for other
factors that might affect the relationship
between type of force used and various
safety outcomes (see Table 5 below).17 We
only include the following variables18:
suspect race, suspect gender, suspect age,
and a categorical variable for type of force
used (officer used a CED and no other type
of force, baton use only, OC spray only, no
weapon used/hands on tactics used, and a
reference group [officer used multiple
weapons or a weapon other than CEDs,
batons, or OC spray]).19
17 Due to our small sample size of CED only sites
(n=7), we were not able to estimate robust variance
estimates controlling for the nesting of standard
errors nor we were able to estimate HLM models for
our CED site only analyses.
18 We did not include our variable for CED site
because we only used CED sites in this analysis. We
did not include our variable for time-period because
we only used the post-period for this analysis when
officers had CEDs or the other weapons available to
use.
19 As pointed out by one reviewer, it would have
been preferable to control for different types of
force situations (e.g., suspects with weapons or
suspects exhibiting violent behavior) since different
types of weapons are used for different types of
situations officers might face. By excluding
variables on the nature of the incident the officer(s)
were facing, we run the risk of placing too much

0.0%
OC Spray

Multiple weapons

Other than
weapon

For four of the eight models, our data
suggest that OC spray was related to greater
safety (see Table 5 below). Our data suggest
that when officers use OC spray there is a
90% reduction in the probability of a suspect
injury (-2.32, p<.001), a 73% reduction in
the probability of an officer injury (-1.29,
p<.001), a 35% reduction in the probability
of a suspect receiving medical attention for
an injury (-0.44, p<.001), and a 45%
reduction in the probability of a suspect
being hospitalized for an injury (-1.29,
p<.001) compared to cases where other
weapons or multiple weapons are used. Our
data suggest that when officers use CEDs
there is a 76% reduction in the probability of
emphasis on the weapon the officer used in
reducing negative outcomes. Unfortunately, three
of the seven CED sites did not collect this
information, and attempting to include theses
variables in our model would drop our sample size
down to four agencies. Despite this concern, we did
explore this issue with the data we had for the four
CED sites. We ran additional models with two
additional control variables (did the suspect use
violence against the officers and did the suspect
use or threaten to use a weapon against the
officers). While these new control variables were
statistically significant in some of the models, the
findings we reported in Table 5 generally did not
change very much (the direction of the parameters
did not change nor the magnitude). Given that little
changes when we add these variables to our models
and the missing data problems which emerge when
we attempt to include these variables, we believe it
is preferable to examine the data as we did in Table
5.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

an officer injury (-1.45, p<.001), and a 63%
reduction in the probability of an officer
requiring medical attention for an injury (0.99, p<.001) compared to cases where other
weapons or multiple weapons are used.
However, our data suggest that when
officers use CEDs there was a 139%
increase in the probability of a suspect
requiring hospitalization (0.87, p<.001)
compared to cases where other weapons or
multiple weapons are used. This is one of
the few negative/adverse findings for CEDs,
and may reflect an informal police practice
of sending suspects who have been
subjected to a CED activation to a hospital
as a precautionary measure—for example, to
ensure that the skin punctures caused by the
CED darts do not become infected. This is a
concern requiring more attention in future
research (see discussion section for more
detail on this finding).

Multivariate analyses using
logistic regression at
individual/incident-level (CED
compared to non-CED sites):
Using logistic regression, we explore
differences between CED and non-CED
sites on the following outcome measures for
officers and suspects: injuries (yes/no),
severity of injuries (minor injury or severe
injury), injury requiring medical assistance
(yes/no), and injury requiring hospitalization
(yes/no). For each of these outcome
measures we built two basic logistic
regression models (called Model 1 and
Model 2). Both models are presented in
Appendix 1. Model 1 included the following
independent/predictor variables: CED
(whether the agency deploys CED: 1= yes,
0=no), time frame of incident (postCED/comparable period= 1, pre
CED/comparable period=0), interaction

CED * Time Frame (1= CED and post
period), and suspect race (White= 1, NonWhite=0), suspect gender (male=1,
female=0), and suspect age (1= < 25 years
old, 0= > 25 years old).
Model 2 included all of the variables
from Model 1 plus two additional
independent variables: (a) whether the
suspect used resistant behavior (1=physical
aggression by suspect, 0= non-physical
aggression [e.g., verbal attacks, assuming a
fighting stance, etc.]) and (b) whether the
suspect had a weapon at the force incident
(1=yes, 0= no). The introduction of Model 2
allows us to analyze the impact of
controlling for these two additional
variables, but three of the sites (2 CED sites
and 1 non-CED site) did not have any data
on these variables. Therefore, Model 2 is
based on 10 sites, not 13 sites. While we
would have liked to include even more
variables in our logistic regression model,
we would have lost additional sites from our
analyses if we attempted to include other
independent variables. For the most part,
despite the fact that the suspect resistant
behavior and suspect possession of a
weapon variables were generally statistically
significant, our multivariate results in Model
2 were similar to our earlier results in terms
of direction and statistical significance. The
fact that our Model 1 results held up, even
with the inclusion of two substantively
significant variables, provides more
confidence in our results. That is, even
controlling for additional variables that
might affect our outcome measures, the
CED sites were still associated with a
variety of positive outcomes. In the text
below, we focus on presenting the results
from Model 1, for the results from Model 2
are very similar both in the direction,
magnitude, and statistical significance of the
effects.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 5: CED only site logistic regression
Logistic Regression for Seven CED Sites Only (Post Period Only)

Suspect race (White= 1, Non-White=0)
Suspect gender (Male=1, female=0)
Suspect age (1= < 25 years old, 0= > 25 years old)
Type of force used
(Reference group= multiple weapon/other weapon)
CED use only
Baton use only
OC spray only
No weapon used (hands on tactics)
Constant

B
0.13
0.39
0.21

0.24
-0.21
-2.32
-1.47
-0.59

SUSPECT INJURY
S.E.
Sig.
Odds Ratio
0.08
0.11
1.14
0.11
0.00
1.48
0.07
0.00
1.23

0.11
0.34
0.18
0.08
0.12

0.00
0.06
0.53
0.00
0.00
0.00

B
0.02
-0.11
0.14

1.27
0.81
0.10
0.23
0.55

-1.45
0.30
-1.29
-2.01
-1.32

OFFICER MEDICAL ATTENTION
-0.07
0.14
0.65
0.94
0.07
0.20
0.72
1.08
-0.05
0.14
0.72
0.95

-0.99
0.56
0.22
-1.18
-2.04

0.29
0.43
0.21
0.22
0.21

0.00
0.00
0.19
0.29
0.00
0.00

0.28
0.62
0.38
0.22
0.27

0.67
0.76
0.42
0.31
0.51
0.00

0.37
1.76
1.25
0.31
0.13

0.26
0.16
-0.44
-1.12
-0.18

0.52
1.07
1.04
0.31
0.46

0.07
0.51
0.58
0.87
0.02
0.00

1.09
1.63
0.68
0.87
0.06

0.71
1.80
0.84
2.07
0.03

0.00
0.00
0.45
0.00
0.00
0.00

0.24
1.35
0.27
0.13
0.27

0.13
0.35
0.11
0.08
0.11

0.00
0.05
0.65
0.00
0.00
0.11

1.30
1.17
0.65
0.33
0.84

SUSPECT HOSPITALIZATION
0.18
0.09
0.04
1.20
0.16
0.12
0.17
1.18
-0.14
0.08
0.08
0.87

0.87
0.10
-0.60
-0.35
-1.52

SUSPECT SEVERE INJURIES
0.70
0.29
0.02
2.02
-0.27
0.40
0.49
0.76
-0.19
0.29
0.50
0.82

-0.34
0.59
-0.18
0.73
-3.40

0.26
0.40
0.21
0.14
0.17

SUSPECT MEDICAL ATTENTION
0.28
0.08
0.00
1.32
0.36
0.10
0.00
1.43
-0.14
0.07
0.03
0.87

OFFICER HOSPITALIZATION
-0.18
0.20
0.36
0.83
-0.10
0.26
0.71
0.91
-0.02
0.18
0.92
0.98

0.09
0.49
-0.38
-0.14
-2.74

OFFICER INJURY
S.E.
Sig.
Odds Ratio
0.13
0.87
1.02
0.16
0.49
0.89
0.12
0.25
1.15

0.12
0.39
0.16
0.09
0.13

0.00
0.00
0.80
0.00
0.00
0.00

2.39
1.10
0.55
0.71
0.22

OFFICER SEVERE INJURIES
-1.38
0.77
0.07
0.25
0.57
0.77
0.46
1.77
0.05
0.49
0.92
1.05

-18.55 9677.64
-18.54 13307.02
-18.26 7324.43
0.31
0.50
-3.14
0.78

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

0.98
1.00
1.00
1.00
0.53
0.00

0.00
0.00
0.00
1.37
0.04

Each of our outcome measures
includes the entire time frame (both the preCED/comparable period and postCED/comparable period). To assess our
outcome measures during the relevant postperiod for CED versus non-CED sites, we
introduce our CED variable and a variable
representing the time frame of each use-offorce incident (pre or post) to form an
interaction term. The main purpose of our
logistic regression model is to attempt to
isolate the effects of CED deployment on
our safety-related outcomes after the
implementation of CEDs, controlling for
other factors that might affect levels of the
various outcomes. The variable of main
interest in our logistic regression models is
the interaction variable of agency
deployment of CED multiplied by time
frame. A positive value on this interaction
term would indicate that an agency that
deploys the CED is associated with more
injuries in the post-period than agencies
without CEDs, controlling for other factors.
A negative value on this interaction term
would indicate that an agency that deploys
the CED is associated with fewer injuries in
the post-period than agencies without CEDs,
controlling for other factors.
Any injury: For our suspect injury
model, our results indicate that for an
agency that deploys CEDs, the odds of a
suspect being injured in the post-period is
reduced by 44% relative to agencies without
CEDs (β=-0.57, odds ratio= 0.56, p<.0001).
To understand the nature of our effects we
estimated prediction profiles for each of our
logistic regression models. Our data suggest
that for an average non-white male under the
age of 25 years old (a group with a higherthan-average likelihood of being subjected
to a CED activation), the predicted
probability of an injury at the post-period for
a non-CED agency equals 50% compared to
a lower probability of only 28% for CED
agencies. As we observed with most of our

other Model 2’s, our suspect resistant
behavior (β= 0.37, odds ratio= 1.45) and
suspect weapon (β= 0.86, odds ratio= 2.35)
variables were statistically significant
(suspects who used physical aggression
against officers were 55% more likely to be
injured than suspects who did not, and
suspects who had weapons were 135% more
likely to be injured than suspects who did
not have weapons).
For our officer injury model, our
results suggest that for an agency that
deploys CEDs, the odds of an officer being
injured in the post-period is reduced by 70%
relative to agencies without CEDs (β=-1.20,
odds ratio= 0.30, p<.0001). Our data suggest
that for an average non-white male under the
age of 25 years old, the predicted probability
of an injury at the post-period for a nonCED agency equals 25% compared to a
lower probability of only 9% for CED
agencies.
Medical attention for injuries: For
our suspect medical attention model, our
results indicate that for an agency that
deploys CEDs, the odds of a suspect needing
medical attention for an injury in the postperiod is reduced by 79% relative to
agencies without CEDs (β =-1.54, odds
ratio= 0.22, p<.0001). Our data suggest that
for an average non-white male under the age
of 25 years old, the predicted probability of
a suspect needing medical attention for an
injury at the post-period for a non-CED
agency equals 52% compared to a lower
probability of only 35% for CED agencies.
For our officer medical attention
model, our results indicate that for an
agency that deploys CEDs, the odds of a
suspect needing medical attention for an
injury in the post-period is reduced by 87%
relative to agencies without CEDs (β=-2.04,
odds ratio= 0.13, p<.0001). Our data suggest
that for an average non-white male under the
age of 25 years old, the predicted probability
of an officer needing medical attention for

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

an injury at the post-period for a non-CED
agency equals 19% compared to a lower
probability of only 9% for CED agencies.
Hospitalization required for
injuries: For our suspect hospitalization
model, our results indicate that for an
agency that deploys CEDs, the odds of a
suspect requiring hospitalization for an
injury in the post-period is reduced by 52%
relative to agencies without CEDs (β= -0.73,
odds ratio= 0.48, p<.0001). Our data suggest
that for an average non-white male under the
age of 25 years old, the predicted probability
of a suspect requiring hospitalization for an
injury at the post-period for a non-CED
agency equals 33% compared to a lower
probability of only 16% for CED agencies.
For our officer hospitalization model,
our results indicate that there were no
differences for agencies that deploys CEDs
and agencies that do not deploy CEDs in
terms of the odds of an officer requiring
hospitalization for an injury in the postperiod (β= -0.23, odds ratio= 0.79, p=0.54).
Our data suggest that for an average nonwhite male under the age of 25 years old
(again, the group with a higher-than-average
likelihood of being subjected to a CED
activation), the predicted probability of an
officer requiring hospitalization for an injury
at the post-period for a non-CED agency
equals 6.2% compared to a similar
probability of only 5.2% for CED agencies.
Severity of injury (Minor vs.
Severe): For our suspect injury severity
model, our results indicate that there were
no differences for agencies that deploys
CEDs and agencies that do not deploy CEDs
in terms of the odds of a suspect receiving a
severe injury in the post-period (β=-0.58,
odds ratio= 0.56, p=0.12). Our data suggest
that for an average non-white male under the
age of 25 years old, the predicted probability
of a suspect receiving a severe injury at the
post-period for a non-CED agency equals

7% compared to a similar probability of
only 3.6% for CED agencies.
For our officer injury severity model,
our results indicate that there were no
differences for agencies that deploys CEDs
and agencies that do not deploy CEDs in
terms of the odds of an officer receiving a
severe injury in the post period (β= 0.56,
odds ratio= 1.75, p=0.42). Our data suggest
that for an average non-white male under the
age of 25 years old, the predicted probability
of an officer receiving a severe injury at the
post-period for a non-CED agency equals
5.2% compared to a similar probability of
only 4.9% for CED agencies.

Multivariate analyses using
logistic regression adjusting
for nested standard errors
(CED compared to non-CED
sites):
As discussed earlier, one of the concerns
with examining multi-site data is that the
individual use-of-force cases we analyze are
clustered within 13 departments. While we
ran an HLM model next to address this
clustering issue, before that we examined
the results of using a logistic regression with
a robust variance estimate to adjust for
within-cluster correlation. We conducted
these analyses using Stata statistical
software with the vce (cluster clustvar)
option. The robust variance estimator comes
under various names in the literature, but
within the Stata software it is known as the
Huber/White/sandwich estimate of variance.
The names Huber and White refer to the
seminal references for this estimator (Huber,
1967; White, 1980). The main limitation
with using this approach is that we do not
get aggregate-level coefficients like those
produced using HLM, but we are still able to
address the clustered nature of our data and
produce unbiased estimates (Rogers, 1993;
Williams, 2000; Wooldridge, 2002). We use

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

this approach to examine more closely the
viability of our earlier logistic regression
results, before examining our full multi-level
HLM results. Also, given that we are
attempting to assess the potential bias
introduced by using nested data across
multiple sites, we only examine our Model 1
results with variables of CEDs, time period,
time period multiplied by CEDs, suspect
race, suspect gender, and suspect age (see
Appendix 2). We did not estimate another
set of Model 2 results, for this involves
dropping three sites from our analysis,
which would weaken our tests of this
nesting issue.
The full results of our logistic
regression with robust variance estimates are
presented in Appendix 2. Below (see Table
6 and 7) we present just our main variable of
interest (our interaction variable for
presence of CED at the post-period) for each
20
of the eight outcome measures. As seen
below (see Table 6 and 7), three of the five
statistically significant results from the
earlier models remained significant under
our logistic regression with robust variance
estimates (including the variables of officer
injury, suspect medical attention, and officer
medical attention). In all three cases, as
reported earlier, CED agencies were
associated with lower post-test rates of
officer injuries, suspects requiring medical
attention for injuries, and officers requiring
medical attention. Two of the five results
rose above the .05 level of statistical
significance, but remained in the predicted
direction (that is, CED sites were still
associated with fewer post-test suspect
injuries and fewer suspects requiring
20 As discussed earlier, a positive value on this
interaction term would indicate that an agency that
deploys CED is associated with more injuries in the
post-period than agencies without CEDs, controlling
for other factors. A negative value on this
interaction term would indicate that an agency that
deploys CED is associated with fewer injuries in the
post-period than agencies without CEDs, controlling
for other factors.

hospitalization from an injury than non-CED
sites, but the result was no longer
statistically significant). However, one of
the outcomes not previously statistically
significant became statistically significant
under the new regression model with robust
variance estimates (that is, CED sites under
this new model are associated with fewer
post-test severe injuries for suspects than
non-CED sites).

Multi-level analyses using
Hierarchical Linear Modeling
(HLM):
As discussed earlier, one of the concerns
with examining multi-site force data is that
the individual use-of-force cases we analyze
are clustered within 13 departments,
violating the independence assumption of
traditional regression approaches. Our final
approach is to use HLM modeling to
examine more closely the viability of our
earlier bivariate results and various logistic
regression results. To follow, we compare
our HLM models to our first set of logistic
regression models and our second set of
logistic regression models (which included a
correction for nested standard errors). Each
of the three types of statistical models
includes a similar set of covariates to control
for suspect age, gender, race, time period
(post period after CEDs were implemented),
and type of agency (CED sites or non-CED
site).21 We also included two additional
aggregate-level variables in our HLM
model: (1) the number of officers in the
LEA per 100,000 in the population in the

All three models included the following
independent/predictor variables: CED (whether the
agency deploys CED: 1= yes, 0=no), time frame of
incident (post-CED/comparable period= 1, preCED/comparable period=0), suspect race (white=
1, non-white=0), suspect gender (male=1,
female=0), and suspect age (1= < 25 years old, 0=
> 25 years old).

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 6: Comparison of two types of logistic regression models for injury and
medical attention outcomes
Outcome measures/dependent variables
Suspect Injury
Variables in logistic
regression model

Without standard error correction
β

Interaction CED *
Time Frame

-0.57

Odds
Ratio
0.56

P value

0.14

<.0001

With standard error correction
β

Odds Ratio

-0.57

0.57

P value

0.55

0.30

Officer Injury
Without standard error correction
β
Interaction CED *
Time Frame

-1.20

Odds
Ratio
0.30

P value

0.20

<.0001

With standard error correction
β

Odds Ratio

-1.20

0.30

SE
0.51

P value
0.019

Suspect Medical Attention
Without standard error correction
β
Interaction CED *
Time Frame

-1.54

Odds
Ratio
0.215

P value

0.147

<.0001

With standard error correction
β
-1.54

Odds Ratio
0.21

SE
0.68

P value
0.02

Officer Medical Attention
Without standard error correction
With standard error correction
Odds
SE
P value
Odds Ratio
SE
P value
β
β
Ratio
Interaction CED *
Time Frame

-2.04

0.131

0.298

<.0001

-2.04

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

0.13

0.96

0.03

Table 7: Comparison of two types of logistic regression models for
hospitalization severe injury outcomes
Outcome measures/dependent variables
Variables in logistic
regression model

Suspect Hospitalization
Without standard error correction
Odds
SE
P value
β
Ratio

Interaction CED *
Time Frame

-0.73

0.48

0.19

<.0001

With standard error correction
β

Odds Ratio

-0.73

0.48

P value

0.70

0.29

Officer Hospitalization
Without standard error correction
Odds
SE
P value
β
Ratio
Interaction CED *
Time Frame

-0.23

0.80

0.38

0.5439

With standard error correction
β
-0.23

Odds Ratio
0.79

SE
0.42

P value
0.59

Suspect Injury Minor vs. Severe
Without standard error correction
With standard error correction
Odds Std
Std
P value
Odds Ratio
P value
β
β
Ratio Error
Error
Interaction CED *
Time Frame

-0.58

0.56

0.37

0.118

-0.58

0.56

0.26

0.02

Officer Injury Minor vs. Severe
Without standard error correction
With standard error correction
Odds Std
Std
P value
Odds Ratio
P value
β
β
Ratio Error
Error
Interaction CED *
Time Frame

0.56

1.75

0.69

0.4166

0.56

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

1.75

0.48

0.24

jurisdiction and (2) the population density
per square mile (this aggregate level variable
could only be added into the HLM model,
for the HLM is the only multi-level
technique of the three approaches). We
present our full HLM results in Appendix 3.
While we considered estimating HLM
models with other additional independent
variables (whether the suspect used resistant
behavior and whether the suspect had a
weapon at the force incident), we are not in
a good position to calculate such a set of
models. The introduction of these two other
variables would allow us to analyze the
impact of controlling for these two
additional incident-level factors, but three of
the sites (2 CED sites and 1 non-CED site)
did not have any data on these variables.
Therefore, our sample size of aggregate
level cases would drop to only 10 sites and
further reduce the statistical power of our
HLM analyses. While not presented, we did
calculate HLM models with these additional
variables, and the main results related to
comparing CED sites to non-CED sites did
not change from that presented below. We
also examined other aggregate level
variables in various combinations in our
HLM model (e.g., measures of arrests and
crimes in the jurisdiction, and various
demographic measures of the jurisdiction)
and achieved the same basic substantive
finding. We included the two measures of
population density and number of officers
per 100,000, for we believe they serve as our
most relevant area-level control variables.
Prior to running our HLM testing, we
examined a variety of diagnostic plots and
checked our data in terms of outliers,
normality, linearity, and followed
Tabachnick and Fidell’s (2007) discussion
of checking HLM assumptions (function
forms are linear at each level, the model is
specified correctly, the error term is not
correlated with the independent variables,
level-1 residuals are normally distributed

and level-2 random effects have a
multivariate normal distribution, level-1
residual variance is constant, level-1
residuals and level-2 residuals are
uncorrelated, and observations at highest
level are independent of each other). Our
models generally met these underlying
assumptions. While there are no agreedupon exact standards on values for each of
the tests associated with these assumptions
(e.g., see Mass & Hox, 2002; Raudenbush &
Bryk, 2002; Raudenbush & Willms, 1995),
there is some evidence that HLM is fairly
robust for modest violations of its
assumptions (Delpish, 2006; Goldberger,
1991).
Officer injury: All three models (our
logistic regression model, our logistic
regression model with a correction for
nested standard errors, and our HLM
models) indicate that agencies that have
deployed CEDs are associated with fewer
injuries to officers compared to non-CED
agencies. Our data suggest that the
magnitude of the effect of CED agencies is
similar across the three models (odds ratios
.30 and .23) and in the same direction
(negative coefficients). Our results indicate
that for an agency that deploys CEDs, the
odds of an officer being injured in the postperiod are reduced by either 70% (logistic
regression odds ratio= 0.30) or 77% (HLM
odds ratio= 0.23) relative to agencies
without CEDs. While we reached statistical
significance with our first logistic regression
model (where the analyses were solely based
on the individual-level [p<.0001]) and our
second logistic regression model [p=.019],
we did not reach statistical significance for
our HLM model (p=.38). As discussed
earlier, due to our small sample size (n= 13
agencies) we did not expect to find
statistical significance for our HLM
findings.22 Our main interest in running the
22To assess model fit we conducted a joint test that
all coefficients are zero, and compared that to our

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

HLMs was to confirm the direction of our
findings (i.e., whether CEDs were associated
with an increase or a decrease in our safetyrelated outcome measures), and the
approximate magnitude of the effect. For
officer injuries, our HLM results have
confirmed our earlier logistic regression
findings.
Suspect injury: All three models
indicate that agencies that have deployed
CEDs are associated with fewer injuries to
suspects compared to non-CED agencies.
Our data suggest that the magnitude of the
effect of CED agencies is similar across the
three models (odds ratios .56 and .53) and in
the same direction (negative coefficients).
Our results indicate that for an agency that
deploys CEDs, the odds of a suspect being
injured in the post-period are reduced by
either 43% (logistic regression odds ratio=
0.56) or 47% (HLM odds ratio23= 0.53)
relative to agencies without CEDs. While
we reached statistical significance with our
first logistic regression model (where the
analyses were solely based on the
individual-level), we did not reach statistical
significance for either of the other two
models.
Officer injury requiring medical
attention: All three models indicate that
agencies that have deployed CEDs are
associated with fewer cases of officers
receiving medical attention for injuries
related to use-of-force compared to nonCED agencies. Our data suggest that the
specified HLM model outlined in Appendix 3. The
result of our comparison is distributed as a ChiSquare distribution with degrees of freedom equal
to the number of constrained coefficients. For our
officer injury model test, that all the coefficients are
zero, our result was statistically significant (X2=
139.01 [DF = 20], p<.001), providing evidence that
our model is a better fit than a fully constrained
model.
23
For our suspect injury model fit test, that all the
coefficients are zero, our result was statistically
significant (X2= 40.12 [DF = 20], p<.01), providing
evidence that our model is a better fit than a fully
constrained model.

magnitude of the effect of CED agencies is
similar across the three models (.13 and.18)
and in the same direction (negative
coefficients). Our results indicate that for an
agency that deploys CEDs, the odds of an
officer receiving medical attention in the
post-period are reduced by either 87%
(logistic regression odds ratio= 0.13) or 82%
(HLM24 odds ratio= 0.18) relative to
agencies without CEDs. While we reached
statistical significance with both of our
logistic regression models, we did not reach
statistical significance for our HLM model.
Suspect injury requiring medical
attention: All three models indicate that
agencies that have deployed CEDs are
associated with fewer cases of suspects
receiving medical attention for injuries
related to use of force compared to nonCED agencies. Our data suggest that the
magnitude of the effect of CED agencies is
similar across the three models (.21 and .54)
and in the same direction (negative
coefficients). Our results indicate that for an
agency that deploys CEDs, the odds of a
suspect receiving medical attention in the
post-period are reduced by either 79%
(logistic regression odds ratio= 0.21) or 46%
(HLM25 odds ratio= 0.54) relative to
agencies without CEDs. While we reached
statistical significance with both of our
logistic regression models, we did not reach
statistical significance for our HLM model.

For our officer medical attention model fit test,
that all the coefficients are zero, our result was
statistically significant (X2= 110.22 [DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.
25 For our suspect medical attention model fit test,
that all the coefficients are zero, our result was
statistically significant (X2= 114.72 [DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 8: Comparison of three multivariate models for injury and medical
attention outcomes

Outcome measures/dependent variables
Variables in
Logistic/HLM model

Officer Injury
Logistic w/out standard error correction
β

Interaction CED *
Time Frame

-1.20

Odds Ratio
0.30

SE
0.20

P value
<.0001

Logistic w/out standard error correction
β
Interaction CED *
Time Frame

-0.57

Odds Ratio
0.56

SE
0.14

P value
<.0001

Logistic w/out standard error correction
Odds Ratio SE
P value
β
Interaction CED *
Time Frame

-2.04

0.13 0.298

<.0001

Logistic w/out standard error correction
Odds Ratio SE
P value
β
Interaction CED *
Time Frame

-1.54

0.21 0.147

<.0001

Logistic w/ standard error correction
β
-1.20

Odds Ratio
0.30

SE
0.51

P value
0.019

HLM
β
-1.49

Suspect Injury
Logistic w/ standard error correction
β
-0.57

Odds Ratio
0.56

SE
0.55

P value
0.30

Officer Medicalization
Logistic w/ standard error correction
Odds Ratio SE P value
β
-2.04

0.13

0.96

0.03

Suspect Medicalization
Logistic w/ standard error correction
Odds Ratio SE P value
β
-1.54

0.21

0.68

0.02

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Odds Ratio
0.23

SE
1.59

P value
0.379

HLM
β

Odds Ratio

-0.64

0.53

HLM
Odds Ratio SE

-1.74

0.18

HLM
Odds Ratio SE

-0.61

0.54

P value

1.63

0.71

P value

2.84

0.56

P value

1.27

0.65

Suspect injury requiring
hospitalization: All three models indicate
that agencies that have deployed CEDs are
associated with fewer cases of suspects
having to be hospitalized for injuries related
to use of force compared to non-CED
agencies. Our data suggest that the
magnitude of the effect of CED agencies is
similar across the three models (.48 and .89)
and in the same direction (negative
coefficients). Our results indicate that for an
agency that deploys CEDs, the odds of a
suspect having to be hospitalized for injuries
related to use of force in the post-period are
reduced by either 52% (logistic regression
odds ratio= 0.48) or 11% (HLM odds
ratio26= 0.89) relative to agencies without
CEDs. While we reached statistical
significance with one of our logistic
regression models (p<.0001), we did not
reach statistical significance for our HLM
model (p=.75).
Officer injury requiring
hospitalization: All three models failed to
reach statistical significance for the officer
hospitalization. However, the logistic
regression models and HLM model27 had
negative coefficients (suggesting that
agencies that have deployed CEDs are
associated with fewer cases of officers
having to be hospitalized for injuries). On
balance, our data suggest that the CED
agencies do not seem to differ from the nonCED agencies in altering the number of
officers requiring hospitalization for an
injury during a force incident.
Suspect injury severity: All three
models indicate that agencies that have
deployed CEDs are associated with fewer
severe injuries to suspects compared to non26 For our suspect hospitalization model fit test,
that all the coefficients are zero, our result was
statistically significant (X2= 120.81[DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.
27 For our officer hospitalization model fit test, that
all the coefficients are zero, our result was
statistically significant (X2= 922.37 [DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.

CED agencies. Our data suggest that the
magnitude of the effect of CED agencies is
similar across the three models (.56 and .36)
and in the same direction (negative
coefficients). Our results indicate that for an
agency that deploys CEDs, the odds of a
suspect being severely injured in the postperiod are reduced by either 44% (logistic
regression odds ratio= 0.56) or 64% (HLM28
odds ratio= 0.36) relative to agencies
without CEDs. While we did not reach
statistical significance with our first logistic
regression model (where the analyses were
solely based on the individual-level), we did
reach statistical significance for our logistic
regression model with a correction for
nested standard errors (p<.02) and reached
statistical significance for the HLM model
(p=.05). Overall, the evidence suggests that
agencies that have deployed CEDs are
associated with fewer severe injuries to
suspects compared to non-CED agencies.
Officer injury severity: All three
models failed to reach statistical significance
for the officer injury severity outcome
measure. While all three models had
positive coefficients (suggesting that
agencies that have deployed CEDs are
associated with more cases of officers
receiving severe injuries) they were not even
close to statistical significance for our
logistic regressions and HLM29 models
(p=.42, p=.24, and p=.23). On balance, our
data suggest that the CED agencies do not
seem to differ from the non-CED agencies
in altering the number of officers receiving
severe injuries during force cases.

28 For our suspect injury severity model fit test, that
all the coefficients are zero, our result was
statistically significant (X2= 931.52 [DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.
29 For our officer injury severity model fit test, that
all the coefficients are zero, our result was
statistically significant (X2= 441.85 [DF = 20],
p<.001), providing evidence that our model is a
better fit than a fully constrained model.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Table 9: Comparison of three multivariate models for hospitalization and
severe injury outcomes

Outcome measures/dependent variables
Suspect Hospitalization
Variables in
Logistic/HLM model

Logistic w/out standard error correction
β

Interaction CED *
Time Frame

-0.73

Odds Ratio
0.48

SE
0.19

P value
<.0001

Logistic w/ standard error correction
β

Odds Ratio

-0.73

0.48

SE
0.70

P value
0.29

HLM
β
-0.12

Odds Ratio
0.89

SE
0.36

P value
0.75

Officer Hospitalization
Logistic w/out standard error correction
β
Interaction CED *
Time Frame

-0.23

Odds Ratio
0.80

P value

0.38

0.5439

-0.23

Logistic w/out standard error correction
Interaction CED *
Time Frame

Odds Ratio Std Err

-0.58

0.56

0.37

P value
0.118

Logistic w/out standard error correction
β
Interaction CED *
Time Frame

0.56

Logistic w/ standard error correction

0.80

SE
0.42

P value
0.59

HLM
β
-0.06

Suspect Injury Minor vs. Severe
Logistic w/ standard error correction
β
-0.58

Odds Ratio Std Err P value
0.56

0.26

0.02

P value

1.75

0.4166

0.56

Odds Ratio Std Err P value
1.75

0.48

0.24

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Odds Ratio
0.94

SE
0.46

P value
0.91

HLM
β
-1.02

Officer Injury Minor vs. Severe
Logistic w/ standard error correction

Odds Ratio Std Err
0.69

Odds Ratio

Odds Ratio
0.36

SE
0.44

P value
0.05

HLM
β
1.07

Odds Ratio
2.92

SE
0.82

P value
0.23

CHAPTER 5
Discussion and Conclusion

he manner in which a policing agency
manages its use of force, including the
types of force it uses, technologies to deliver
that force, and when various types of force
can be used, are among the most important
decisions that a LEA executive will have to
make. One of the key objectives in
managing force is designing approaches to
reduce incidents of police use of force and
the injuries associated with force. One
weapon that has been advanced as a way to
reduce injuries for officers and suspects is
the Conducted Energy Device (CED). Police
chiefs and sheriffs charged with making the
decision whether to use CEDs or other lesslethal weapons need guidance about whether
the weapons are in fact effective. Law
enforcement executives have been deluged
with questions about the effectiveness and
safety of CEDs, and the lack of available
information and a full understanding about
the effects of using CEDs has hampered the
ability of police executives to make
informed policy decisions about the devices.
Police executives have been provided with
little independent scientific evidence and
guidance on the impact of using CEDs,
forcing them to make policy and operational
decisions without being fully informed.
While decades of research have documented
the nature and extent of other types of force
used by police and the conditions and
correlates that affect the application of force
(Smith et al., 2007), little research has been
done isolating the effects of using CEDs on
injuries to suspects and officers. The

purpose of this project was to produce
scientifically valid results that will inform
LEA executives’ decisions regarding the use
of CEDs.
Our study is one of the first to
compare LEAs that use CEDs to matched
LEAs that do not use CEDs. The problem
with evaluating data solely from CED
agencies is that the inferences that can be
made about the results are limited by the
usual problems with pre/post designs and
their inability to rule out rival explanations
for any impacts of the intervention, in this
case, the deployment of CEDs. That is, it is
hard to control for alternative factors that
could explain changes from the pre-test
period to the post-period in those types of
designs.30 We completed an objective
analysis of the effects that department-wide
deployments of CEDs by LEAs have on
injuries and deaths to police and suspects,
associated medical attention, and need for
hospitalization. The goal of our study was to
produce practical information that can help
law enforcement executives make good
decisions about whether to deploy CEDs,
and if a decision is made to deploy them, to
help the agencies develop CED policy and
procedural guidelines that provide increased
safety for officers and citizens. In order to
accomplish this goal, we examined the
outcome of CED deployments in terms of
30 While we also conduct a set of within CED site
analyses, we are very cautious in our interpretation
of these data and we rely more on our CED site to
non-CED site comparisons.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

officer and suspect safety. We then
compared the differences in outcomes
between police agencies that have
incorporated the use of CEDs (n=7) to those
found in police agencies that have not
incorporated the use of CEDs (n=6). This
study contains important scientific
information isolating the safety outcomes to
be expected if a department deploys CEDs,
controlling for a variety of related
organizational and individual/incident-level
factors.
The first major methodological
challenge in conducting our study was
finding a set of comparison LEAs that have
used CEDs and a matched group that did not
use CEDs. Our selection of cities was based
on a matching analysis using a PERF
nationally representative survey on use of
force. Overall, we believe our CED and nonCED sites are comparable. We collected
data from roughly comparable periods
(within a year or two) for the CED and nonCED sites. The main difference between the
non-CED and CED sites is the participation
of one CED site that is much larger than the
other sites in our study. However, when we
estimated all of our models with and without
this large site, we found no major
differences in our results. With this site
excluded from our analyses, there are no
major aggregate-level demographic
differences between the CED and non-CED
sites across a range of variables including:
population size, size of agency, number of
arrests for violent offenses, number of
violent crimes, and number of homicides.
The non-CED and CED sites were also
similar on a full range of background
aggregate-level factors measured through
the 2000 U.S. Census (even with the
especially large site included in the
analysis). Overall, while some differences
emerged in our assessment of the
comparability of our CED and non-CED
sites, most of the differences were relatively

small and did not seem to introduce any
substantively important biases. When
combined with our multivariate analyses, we
believe that we have a reasonably
comparable group of CED and non-CED
sites with results that are interpretable.
Another important point to recall is
that all of the LEA sites with CEDs in our
sample have had fairly limited experience
with using the CED. None of the CED sites
started using the CED weapon in the 20th
century.31 Therefore, any conclusions that
we draw from our research reflect the early
experience with CEDs. Over time, it seems
reasonable to expect that LEAs will gain
important insights into the use of CEDs and
will be able to further improve safety
outcomes associated with this weapon.
In the remainder of this section, we
review our results and summarize the main
findings. Following our review of our
results, we discuss the implications of our
results for LEA policy and training, and
provide some recommendations for future
research.

Review of results:
Earlier we presented a variety of analyses
comparing CED and non-CED sites,
including bivariate analyses to describe the
basic raw differences between the CED and
non-CED sites on our outcome measures,
and a variety of multivariate analyses to
attempt to assess the viability of the
bivariate results and control for possible
alternative explanations that might explain
the earlier raw differences. Our first
multivariate analyses were done using
logistic regression to isolate the effects of
CED deployment on our safety-related
outcomes, where we included the following
independent/control variables: whether the
agency deploys CED, the time frame of the
31 Only two sites started using the CED weapon in
the early 2000s.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

incident, an interaction of CED multiplied
by time frame, suspect race, suspect gender,
and suspect age.32
To examine the clustering issue
described earlier, we used two approaches.
First, we conducted a modified logistic
regression with a robust variance estimator
to adjust for within-cluster correlation.
However, with this approach we do not get
aggregate-level coefficients to see the exact
effects of aggregate level conditions on our
individual results. To examine and observe
the effects of aggregate-level factors, we
conducted a multi-level analysis using
Hierarchical Linear Modeling (HLM).33
While we recognize our limited statistical
power to conduct HLM analyses (n=13
LEAs), we are mainly using HLM to assess
the robustness of our findings from our
earlier analyses. We focus our analyses of
the HLM results on the direction and
magnitude of the effects (as opposed to a
focus on the statistical significance of the
results).
Below we review our bivariate and
multivariate results for each of our main
outcome measures for the CED versus nonCED site comparisons: officer and suspect
injuries, officer and suspect severe injuries,
officer and suspect injuries requiring
medical attention, officer and suspect
32 We also assessed two additional independent
variables to our logistic regression model: (a)
whether the suspect used resistant behavior and (b)
whether the suspect had a weapon at the force
incident. For the most part, despite the fact that
the suspect resistant behavior and suspect
possession of a weapon variables were generally
statistically significant, our multivariate results
were similar to our earlier univariate results in
terms of direction and statistical significance. The
fact that our Model 1 results held up, even with the
inclusion of two substantively significant variables,
provides more confidence in our results. That is,
even controlling for additional variables that might
affect our outcome measures, the CED sites were
still associated with a variety of positive outcomes.
33 We included two additional aggregate-level
variables in our HLM model: (1) the number of
officers in the LEA per 100,000 in the population in
the jurisdiction and (2) the population density per
square mile.

injuries requiring hospitalization, and
suspect deaths. For these outcomes, we
review our results comparing CEDs to nonCED sites, followed by our results for CED
sites only. For the analyses of only the CED
sites, we review both raw bivariate results
and multivariate logistic regression models
for the period after CEDs have been
deployed (comparing the actual use of CEDs
by officers to other forms of use of force).
Officer injuries: Our results across all
of our analyses suggest a strong effect of
CEDs on reducing officer injuries. Our first
set of raw bivariate results compared
differences between CED and non-CED
sites on the proportion of use-of-force cases
where an officer was injured before CEDs
were implemented and after CEDs were
implemented in CED sites, and during a
similar reference period for the non-CED
sites. Before the CED sites deployed CEDs,
the proportion of officers injured in force
cases (12%) was similar compared to nonCED sites (10%) over the same reference
period. The CED sites then went on to
observe a reduction in officer injuries (to a
level of 8%) after they began their
deployment of CEDs, compared to the nonCED sites that observed an increase in
officer injuries for the non-CED sites to
20%.
For our logistic regression officer
injury model, we found that for an agency
that deploys CEDs, the odds of an officer
being injured in the post-period is reduced
relative to agencies without CEDs
(p<.0001). Our results held up when we
estimated our logistic regression with robust
variance estimates (the results were still
statistically significant p<.02). While we did
not reach statistical significance for our
HLM model (p=.38), the effects were in the
34
same direction and of a similar magnitude.
34 As discussed earlier, due to our small sample
size (n= 13 agencies) we did not expect to find
statistical significance for our HLM findings. Our

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

All three multivariate models (our logistic
regression model, our logistic regression
model with robust variance estimates, and
our HLM models) indicate that agencies that
have deployed CEDs are associated with
fewer injuries to officers compared to nonCED agencies. The magnitude of the effect
of CED agencies is similar across the three
models (.30 and .23) and in the same
direction (negative coefficients). Our results
indicate that for an agency that deploys
CEDs, the odds of an officer being injured
in the post-period is reduced by over 70%
relative to agencies without CEDs.
Our final set of analyses on officer
injuries focused on just the participating
CED sites. Batons (24%) and nonweapon/hands-on tactics were associated
with the highest levels of officer injuries
(45%), followed by OC spray (6%), CEDs
(5%), and multiple weapons (or weapons
other than CEDs, batons, or OC spray) (3%).
For these analyses, multiple/other weapons,
CEDs, and OC spray and were associated
with significantly lower officer injuries than
the other forms of force (p<.001). Based on
our logistic regression model, when officers
use CEDs, there is a 76% reduction in the
probability of an officer injury compared to
cases where other weapons or multiple
weapons are used.
Suspect injuries: Our results, across
all of our analyses, demonstrate that CEDs
are related to reductions in suspect injuries.
Before the CED sites deployed CEDs, 23%
of the suspects were injured in force cases,
compared to a slightly higher proportion of
suspects in the non-CED sites (30%) over
the same reference period, representing a
small statistical difference (p<.001). The
CED sites observed a small increase in
suspect injuries (to 26%) after they began
main interest in running the HLMs was to confirm
the direction of our findings (i.e., whether CEDs
were associated with an increase or a decrease in
our safety-related outcome measures), and the
approximate magnitude of the effect.

their deployment of CEDs, compared to the
non-CED sites that observed a much larger
increase in suspect injuries for the non-CED
sites to 43% (p<.001). While the CEDs
started out at a slightly lower rate of suspect
injuries compared to the non-CED sites
(23% vs. 30%), the CED sites were
substantially lower at the post period (26%
vs. 43%), at a rate much greater than the
initial differences would predict.
For our logistic regression model, our
results indicate that for an agency that
deploys CEDs, the odds of a suspect being
injured in the post-period is reduced relative
to agencies without CEDs (p<.0001). Our
logistic regression with robust variance
estimates and HLM model also indicated
that agencies that have deployed CEDs are
associated with fewer injuries to suspects
compared to non-CED agencies. The
magnitude of the effect of CED agencies is
similar across the three models and in the
same direction. Our results indicate that for
an agency that deploys CEDs, the odds of a
suspect being injured in the post-period is
reduced by over 40% relative to agencies
without CEDs.
For our CED-only site analyses, nonweapon/hands-on tactics were associated
with the highest levels of suspect injuries
(45%), followed by batons (44%), CEDs
(44%), multiple weapons (or weapons other
than CEDs, batons, or OC spray) (15.9%),
and OC spray (7.6%). For these analyses,
OC spray and multiple/other weapons were
associated with significantly lower suspect
injuries than the other forms of force
(p<.001). Based on our logistic regression
model, there was no difference between
cases when officers use CEDs and cases
where other weapons or multiple weapons
are used in terms of suspect injuries.
Officer injury severity: Across all of
our analyses, our results demonstrate that
CEDs do not have an effect the severity of
officer injuries. Before the CED sites

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

deployed CEDs, 4% of the officers were
severely injured in force cases, compared to
a similar proportion of officers in the nonCED sites (7%) over the same reference
period, representing no statistical difference.
The CED sites observed no significant
change in officer severe injuries (5% from
4%) after they began their deployment of
CEDs, compared to the non-CED sites that
observed no change in officer severe injuries
(6% to 7%). All three multivariate models
failed to reach statistical significance for the
officer injury severity outcome measure.
The CED agencies do not seem to differ
from the non-CED agencies in altering the
number of officers receiving severe injuries
during force cases. For our CED-only site
analyses, we found no statistically
significant differences in the proportion of
officers receiving severe injuries across the
various use-of-force tactics. Based on our
logistic regression model, we found no
difference between cases when officers use
CEDs and cases where other weapons or
multiple weapons are used in terms of
officer severe injuries.
Suspect severe injuries: Before the
CED sites deployed CEDs, 7% of the
suspects were severely injured in force
cases, compared to a similar proportion of
suspects in the non-CED sites (7%). The
CED sites went on to experience a reduction
in suspect severe injuries (to 5% from 7%)
after they began their deployment of CEDs,
compared to the non-CED sites, which
observed no change in suspect severe
injuries. All three multivariate models also
indicated that agencies that have deployed
CEDs are associated with fewer severe
injuries to suspects compared to non-CED
agencies. The magnitude of the effect of
CED agencies is similar across the three
models (.56 and .36) and in the same
direction (negative coefficients). Our results
indicate that for an agency that deploys
CEDs, the odds of a suspect being severely

injured in the post-period is reduced by over
40% relative to agencies without CEDs.
While we did not reach statistical
significance with our first logistic regression
model (where the analyses were solely based
on the individual-level), we did reach
statistical significance for our logistic
regression model with a correction for
nested standard errors (p<.02) and reached
statistical significance for the HLM model
(p=.05). Overall, the evidence suggests that
agencies that have deployed CEDs are
associated with fewer severe injuries to
suspects compared to non-CED agencies.
For our CED-only site analyses, CED
(2%) and OC spray (3%) forms of force
were associated with lower levels of suspect
severe injuries than multiple/other weapons
(6%) and use of batons by officers (6%)
(p<.05). Based on our logistic regression
model, we found no difference between
cases when officers use CEDs and cases
where other weapons or multiple weapons
are used in terms of suspect severe injuries.
Officer injury requiring medical
attention: Before the CED sites deployed
CEDs, 13% of the officers in the sites
received an injury requiring medical
attention in force cases, compared to a lower
proportion of officers in the non-CED sites
(4%) over the same reference period
(p<.001). The CED sites observed a large
decrease in officers receiving an injury
requiring medical attention (to 8% from
13%) after they began their deployment of
CEDs, compared to the non-CED sites,
which observed a large increase in officer
receiving an injury requiring medical
attention (to 16% from 4%) ( p<.001). All
three multivariate models indicate that
agencies that have deployed CEDs are
associated with fewer cases of officers
receiving an injury requiring medical
attention related to use of force compared to
non-CED agencies. The magnitude of the
effect of CED agencies is similar across the

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

three models and in the same direction. Our
results indicate that for an agency that
deploys CEDs, the odds of an officer
receiving an injury requiring medical
attention in the post-period is reduced by
over 80% relative to agencies without
CEDs.
For our CED-only site analyses, OC
spray (13%), batons (12%), and tactics other
than weapons (9%) were associated with the
highest levels of officers receiving an injury
requiring medical attention. For these
analyses, multiple weapons/other weapons
and CEDs were associated with significantly
lower number of cases where officers
received an injury requiring medical
attention than the other forms of force
(p<.001). Based on our logistic regression
model, when officers use CEDs, there is a
63% reduction in the probability of an
officer receiving an injury requiring medical
attention ( p<.001) compared to cases where
other weapons or multiple weapons are
used.
Suspect injury requiring medical
attention: Before the CED sites deployed
CEDs, 55% of the suspects received an
injury requiring medical attention in force
cases, compared to a lower proportion of
suspects in the non-CED sites (35%) over
the same reference period, representing a
statistically significant difference (p<.001).
The CED sites observed a large decrease in
suspects receiving an injury requiring
medical attention (to 40% from 55%) after
they began their deployment of CEDs,
compared to the non-CED sites, which
observed a large increase in suspects
receiving an injury requiring medical
attention (to 53% from 35%) (p<.001). All
three multivariate models indicate that
agencies that have deployed CEDs are
associated with fewer cases of suspects
receiving injuries requiring medical
attention related to use-of-force compared to
non-CED agencies. Our results indicate that

for an agency that deploys CEDs, the odds
of a suspect receiving an injury requiring
medical attention in the post-period is
reduced between 79% (logistic) and 46%
(HLM) relative to non-CED agencies.
For our CED-only site analyses,
batons (62.5%), CEDs (58%) and nonweapon/hands-on tactics (55.7%) were
associated with the highest levels of suspects
receiving an injury requiring medical
attention. For these analyses, multiple/other
weapons and OC spray were associated with
significantly lower number of cases where
suspects received an injury requiring
medical attention than the other forms of
force (p<.001). Based on our logistic
regression model, we found no difference
between cases when officers actually use
CEDs to cases where other weapons or
multiple weapons are used in terms of
suspects receiving an injury requiring
medical attention.
Officer injury requiring
hospitalization: Before the CED sites
deployed CEDs, 4.3% of the officers
received an injury requiring hospitalization
in force cases, compared to a similar
proportion of officers who received an
injury requiring hospitalization in the nonCED sites (3.3%) over the same reference
period, representing no statistical difference
(p=.35). The CED sites observed a very
small decrease in officers receiving an injury
requiring hospitalization (to 4.1%) after they
began their deployment of CEDs, compared
to the non-CED sites, which observed an
increase in officers receiving an injury
requiring hospitalization (to 6.3%). The
CED sites started out at a similar rate of
officers receiving an injury requiring
hospitalization compared to the non-CED
sites (4.3% and 3.3%, respectively), but the
CED sites were significantly lower at the
post period (4.1% to 6.3%). All three
multivariate models failed to reach statistical
significance for the officer hospitalization

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

measure. While the logistic regression
models and HLM model had negative
coefficients, on balance, the CED agencies
do not seem to differ from the non-CED
agencies in altering the number of officers
requiring hospitalization for an injury during
a force incident.
For our CED-only site analyses, we
found no statistically significant differences
across the various forms of use-of-force in
terms of officers receiving injuries requiring
hospitalization. Based on our logistic
regression model, we found no difference
between cases when officers actually use
CEDs and cases where other weapons or
multiple weapons are used in terms of
officers receiving injuries requiring
hospitalization.
Suspect injury requiring
hospitalization: Before the CED sites
deployed CEDs, 27% of the suspects
received an injury requiring hospitalization
compared to a similar proportion of the
suspects in the non-CED sites (31%) over
the same reference period (p=.11). The CED
sites observed a large decrease in suspects
receiving an injury requiring hospitalization
(to 16%) after they began their deployment
of CEDs, compared to the non-CED sites,
which observed a small increase in suspects
receiving an injury requiring hospitalization
to 36% (p<.05). The CEDs started out at a
similar rate of suspects receiving injuries
requiring hospitalization compared to the
non-CED sites (27% to 31%), but the CED
sites were significantly lower at the post
period (16% to 36%).
All three multivariate models indicate
that agencies that have deployed CEDs are
associated with fewer cases of suspects
receiving an injury requiring hospitalization
compared to non-CED agencies. While the
direction of the effect of CED agencies is
similar across the three models, the
magnitude of the effect was quite different.
Our results indicate that for an agency that

deploys CEDs, the odds of a suspect
receiving an injury requiring hospitalization
in the post-period is reduced by 52% for the
logistic regression model or only 11% for
the HLM models relative to agencies
without CEDs. While there is a wide gap in
these estimates, both models suggest that
CED sites are associated with a reduced
probability of suspects receiving injuries
requiring hospitalization.
For our CED-only site analyses, CEDs
(29.5%), batons (19.7%), and nonweapon/hands-on tactics (16.7%) were
associated with the highest levels of suspects
receiving injuries requiring hospitalization.
For these analyses, OC spray (11.2%) and
multiple/other weapons (12.3%) were
associated with significantly lower number
of cases where suspects received injuries
requiring hospitalization than the other
forms of force (p<.001). Based on our
logistic regression model, when officers use
CEDs, there was a 139% increase in the
probability of a suspect receiving injuries
requiring hospitalization (0.87, p<.001)
compared to cases where other weapons or
multiple weapons are used.
We have explored this 139% increase
and attempted to disentangle this result. For
example, suspects who were subjected to a
CED activation were not different from
suspects who had other weapons used
against them on our injuries measure. Also,
these results apply only to agencies that
have CEDs (the matched non-CED sites
were not in these analyses). We discussed
this finding with some of the police
personnel at the sites. These personnel
indicated that some agencies may have an
informal practice in place where they send
suspects who have been activated by a CED
to a hospital more frequently compared to
other types of force cases (perhaps due to
the heavy news media coverage than can
sometimes emerge with a CED case).
Furthermore, as noted earlier,

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

“hospitalization” in this context does not
necessarily mean an overnight stay in a
hospital; rather, it signifies only that
suspects were sent to a hospital, clinic, or
other medical facility, and in many cases
may have simply received outpatient
evaluation and/or treatment. It is also worth
noting in this context that in 2005, PERF
issued a set of 52 CED guidelines that,
among other things, recommended that “all
persons who have been exposed to a CED
activation should receive a medical
evaluation,” and that “officers should not
generally remove CED darts from a subject
that have penetrated the skin unless they
have been trained to do so.” Unfortunately,
we do not have the case narratives on each
of the CED cases in this study to assess
exactly what is occurring, and that is why
more research is needed on this topic.
Alternatively, this could be just an
anomalous finding. Given that there is little
precedence for collecting the type of data we
collected, we were not aware of this
potential complexity in our data and were
not able to build this into the design of our
study. Future researchers will be able to
consider this finding and build in features to
be able to explore this issue in their
research.
Suspect deaths: Before
implementation of CEDs, the CED sites had
less than one percent of their cases (0.2%)
involving a suspect killed by an officer.
After CED implementation, this number
remained about the same (0.4%). During the
same period, the non-CED sites did not
change either, observing about one percent
of their cases (0.9%) involving a suspect
killed by an officer at the pre-test period as
well as the post-period. While the postperiod results for CEDs (0.4%) and nonCEDs (0.9%) represents a small statistical
difference (p<.05) we do not believe this
difference is necessarily attributable to the
presence of CEDs (it is likely just random

noise in the data). We basically have a flat
line for the CED sites (0.2% to 0.4%) and a
flat line for the non-CED sites (0.9% at both
time points). On balance, CEDs do not
appear to have much of an effect on suspect
deaths, but with a sample of only 44 suspect
deaths we do not have a high level of
statistical power to uncover statistically
significant findings. One of the concerns
that has been expressed by a number of
organizations regarding CEDs is that they
may lead to higher death rates for agencies
that deploy CEDs. We found no support for
this concern. CEDs seem to have a neutral
effect on the number of suspect deaths
related to officer use-of-force cases.

Summary of findings:
Overall, we found that the CED sites were
associated with improved safety outcomes
when compared to a group of matched nonCED sites on six of nine safety measures,
including reductions in (1) officer injuries,
(2–3) suspect injuries and severe injuries,
(4–5) officers and suspects receiving injuries
requiring medical attention, and (6) suspects
receiving an injury requiring hospitalization.
For the other three of nine measures, there
were no differences between the CED and
the non-CED sites on the outcomes of the
number of suspect deaths, officer severe
injuries, and officer injuries requiring
hospitalization.
For the six of nine significant
outcomes, the magnitude of the effects of
the improved safety outcomes for the CED
sites (relative to the non-CED sites) was
impressive. We found a strong effect of
CEDs on reducing officer injuries based on
our raw results (8% officer injuries in the
post–period, compared to 20% for the nonCED sites), and our three multivariate
models. For agencies that deploy CEDs, the
odds of an officer being injured are reduced
by over 70%. Also, for our CED-only site

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

analyses, when officers actually use CEDs
there is a 76% reduction in officer injuries.
Similar reductions were observed for the
CED sites on our measure of suspect
injuries, as confirmed by our raw results
(26% suspect injuries in the post–period,
compared to 43% for the non-CED sites),
and our three multivariate models. For an
agency that deploys CEDs, the odds of a
suspect being injured are reduced over 40%.
Along the same lines, CED sites were
related to reductions in suspect severe
injuries based on our raw results (5%
suspect severe injuries in the post–period,
compared to 7% for the non-CED sites), and
our three multivariate models. For an agency
that deploys CEDs, the odds of a suspect
being severely injured are reduced by over
40%. For our CED-only site analyses, CEDs
were associated with the lowest levels of
suspect severe injuries compared to other
forms of force.
CED sites were related to reductions in
injuries to officers requiring medical
attention based on our raw results (8% of
use-of-force cases requiring officer medical
attention in the post–period in CED sites,
compared to 16% for the non-CED sites),
and our three multivariate models. For an
agency that deploys CEDs, the odds of an
officer receiving an injury requiring medical
attention is reduced by at least 80%. For our
CED-only site analyses, when officers
actually use CEDs there is a 63% reduction
in the probability of an officer receiving an
injury requiring medical attention. Similarly,
CED sites were related to reductions in
injuries to suspects requiring medical
attention based on our raw results (40% of
cases requiring suspect medical attention in
the post-period in CED sites, compared to
53% for the non-CED sites), and our three
multivariate models. For an agency that
deploys CEDs, the odds of a suspect
receiving an injury requiring medical

attention in the post-period are reduced by
over 45%.
CED sites were related to reductions in
injuries to suspects requiring hospitalization
based on our raw results (16% resulting in
suspect medical attention in the post period,
compared to 36% for the non-CED sites),
and our three multivariate models. For
agencies that deploy CEDs, the odds of a
suspect receiving an injury requiring
hospitalization in the post-period is reduced
by 52% for the logistic regression model or
only 11% for the HLM models relative to
agencies without CEDs. While there is a
wide gap in these estimates, both models
suggest that CED sites are associated with a
reduced probability of suspects receiving
injuries requiring hospitalization. For our
CED-only site analyses, CEDs (30%) had
the highest levels of suspects receiving
injuries requiring hospitalization. When
officers use CEDs, there was a 139%
increase in the probability of a suspect
receiving injuries requiring hospitalization
(0.87, p<.001). This is one of the few
negative/adverse findings for CEDs, and
may reflect an informal police practice of
sending suspects who have been subjected to
a CED activation to a hospital as a
precautionary measure—for example, to
ensure that the skin punctures caused by the
CED darts do not become infected. While
overall, the CED sites led to better outcomes
than the non-CED sites on this measure, this
result needs to be explored further in future
research.
Another concern raised by proponents
of CEDs is that they may lead to higher
death rates for agencies that deploy CEDs.
We found no support for this concern. CEDs
seem to have a neutral effect on the number
of suspect deaths related to officer use-offorce cases. Before implementation of
CEDs, the CED sites had less than one
percent of their cases (0.2%) involving a
suspect killed by an officer. After CED

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

implementation, this number remained about
the same (0.4%). During the same period,
the non-CED sites did not change either.
The non-CED sites observed about one
percent of their cases (0.9%) involving a
suspect killed by an officer at the pre-test
period, and observed no change in the
number of suspects killed in force incidents
at the post-period (0.9%). We basically have
a flat line for the CED sites (0.2% to 0.4%)
and a flat line for the non-CED sites (0.9%
at both time points). On balance, CEDs do
not appear to have much of an effect on
suspect deaths, but with a sample of only 44
suspect deaths we do not have a high level
of statistical power to uncover statistically
significant findings. For officer severe
injuries and injuries to officers requiring
hospitalization, we also found no differences
between the CED and non-CED sites.
All in all, we found consistently strong
effects for CEDs on increasing officer and
suspect safety. Not only are CED sites
associated with improved safety outcomes
compared to a matched group of non-CED
sites, but also within CED agencies, in some
cases the actual use of a CED by an officer
is associated with improved safety outcomes
compared to other less-lethal weapons. For
five of the eight comparisons, the cases
where an officer uses a CED were
associated with the lowest or second lowest
rate of injury or medical attention/
hospitalization.
While our study is one of the first to
compare CEDs to matched non-CED sites,
such quasi-experimental designs (QEDs) are
not without limitations. As mentioned
earlier, in the limitations section of Chapter
3, QEDs are not as strong as randomized
experiments in isolating the effects of a
policy (in our case the policy to either
deploy or not deploy CEDs). The main
concern is that, as opposed to randomized
experiments, it hard to control for the many
unmeasured variables related to the outcome

variable (Shadish et al., 2002). Randomized
experiments are typically considered the
best method for eliminating threats to
internal validity in evaluating social policies
and programs (Berk et al., 1985; Boruch, et
al., 1978; Campbell, 1969; Campbell and
Stanley, 1963; Dennis and Boruch, 1989;
Riecken et al., 1974). However, it was not
possible in this study to randomly assign the
use of various weapons to police officers.
With QEDs, the key is to determine all
of the important covariates that might affect
our outcome measures and statistically
control for any observed differences on
these measures in our matched participating
agencies. We believe we have identified the
most important covariates that might
confound our comparison of CED and nonCED sites, and we have used these measures
to effectively isolate the effects of various
less-lethal weapons. We have considered
various alternative explanations for our
results, and believe the most plausible
explanation is that the availability of CEDs
to officers is a key factor in reducing injuries
to officers and suspects. For example,
differences between the CED and non-CED
sites could be attributable to differences in
time periods (this was controlled for in our
selection of data from similar time frames
across the sites), the presence of more
aggravating incident-level factors in some
agencies such as greater presence of weapon
use by suspects or resistance used by
suspects (we included incident-level factors
in our models that statistically control for
these factors), the absence of more detailed
measures (while there are some concerns
with the specificity of our measures, i.e. not
having enough detail limits the number of
additional outcomes we can assess, but does
not affect the validity of our dichotomous
outcomes), and variation in sampling across
the comparison sites (this was assessed and
ended up being non-significant in our
statistical tests where a variable was added

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

to our models to measure whether a sample
or population data were used).
On balance the effect sizes evident in
our results are substantively important and
should be carefully considered by law
enforcement executives. For example, for
agencies participating in this evaluation that
deploy CEDs, our results suggest that the
odds of an officer being injured in the postperiod are reduced by over 70% relative to
agencies without CEDs. Also, the effect
sizes are generally large enough to suggest
that even if the comparability of the CED
and non-CED sites is not perfect, there are
still likely to be important safety gains for
officers in agencies that deploy CED
compared to those that do not.
Next, we discuss the implications of
our results for LEA policy and training, and
provide some recommendations for future
research.

Implications of PERF results
regarding when to use CEDs:
Prior research on police use of force,
including our results, consistently shows that
most use-of-force encounters involve low
levels of force and few if any injuries for
officers and suspects. However, it is not
uncommon for officers to have to use more
force to gain control of a noncompliant
suspect and take the person to the ground,
with the officer using the ground for
leverage (see Smith et al., 2008). These
types of ground struggles carry an increased
risk for injury for officers and suspects.
According to our results, police devices such
as CEDs and OC spray that avoid these upclose struggles hold the promise of avoiding
injuries for all concerned parties. These
findings are consistent with the work by
Smith and colleagues (2008) that CEDs and
OC spray allow officers to control suspects
from a distance without engaging in the

hand-to-hand struggles that typically cause
injuries.
The evidence from our study suggests
that CEDs can be an effective weapon in
helping prevent or minimize physical
struggles in use-of-force cases. LEAs should
consider the utility of the CED as a way to
avoid up-close combative situations and
reduce injuries to officers and suspects.
Also, for agencies that do not deploy CEDs,
our results suggest that they should consider
the possible value of deploying CEDs, and
the relevance of the CED for use by officers
in their community. Also, similar results
were also uncovered in a similar study by
Smith et al. (2008). Faced with similar
results, Smith et al. (2008) recommended
that CEDs should be authorized as a
possible response in cases where suspects
use defensive resistance (e.g., suspect
struggles to escape physical control of
officer) or higher levels of suspect
resistance, in order to help officers avoid upclose combative situations. We do not take a
position on the specific circumstances when
an LEA should authorize the use of the
CED. We believe such a policy decision
needs to be made at the local level. It is not
appropriate, based on a single study, to
make a firm recommendation on when a
CED should be authorized to be used. Each
LEA has to consider a multitude of factors
in assessing when to authorize use of the
CED, working closely with its full set of
community partners to consider a range of
local factors. However, our study provides
important data points to inform these policy
decisions that LEAs need to make. For
example, there is little support in our data to
consider authorizing the use of CEDs in
cases of passive resistance from a suspect;
these cases rarely results in injuries to
officers. Also, in terms of reducing injuries,
there is little to gain by permitting use of
CEDs against certain special populations
(pregnant women, elderly citizens, and

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

others who are clearly physically impaired),
for few of these persons were involved in
force cases where officers were injured in
our study. As pointed out in the Police
Executive Research Forum CED guidelines,
good CED policies and training will aid
officers in evaluating the totality of the
circumstances before using a CED, which
would include considering the following
factors: the age, size, gender, apparent
physical capabilities, and health concerns of
suspects, presence of flammable liquids, and
circumstances where falling would pose
unreasonable risks to the suspect.35
Many policy questions with the use of
CEDs still remain. Where on the body
should a CED be used? Do the number of
CED activations and the duration of shocks
impact safety? Does use of CEDs in
combination with a flammable substance
increase the possibility of ignition? Should
the use of CEDs against the very young,
pregnant women, and those suffering from
medical problems or other special
populations be prohibited? For example,
some have raised concerns about the use of
CEDs on seniors or individuals suffering
from osteoporosis. A deputy sheriff
suffering from this bone-weakening disease
reportedly sustained a fracture after he was
shocked during a training exercise (Anglen
2004).

Need for training for CEDs
There is little attention in the CED literature
to training of officers and sheriffs’ deputies
in the proper use of CEDs. While some CED
manufacturers have developed CED training
curricula and some have even provided CED
training, there are few independent sources
for agencies to turn for guidance on
developing a CED training program (see
Smith et al., 2008). As a result, there is little
35 See PERF guidelines on the use of CEDs (Police
Executive Research Forum, 2005).

consensus on what training should be
required, what it should encompass, or what
its purpose should be beyond familiarization
with the device (see Smith et al., 2008).
Officer training varies from familiarization
training with the CED (sometimes including
officers being shocked with the CED to
experience the weapon’s effects) to
comprehensive scenario-based training
where multiple weapons and other tools,
including the CED, are available to deal
with a simulated threat. However, research
to identify which of these approaches is
most effective has not been done (see Smith
et al., 2008).
Another training issue is the
inappropriate use of the CED. As with any
service weapon, officers can misuse CEDs.
Misuse can range from outright abusive or
illegal use of the weapon to less obvious
cases of officers turning to a CED too early
in a force incident (e.g., bypassing verbal
de-escalation skills and going right to the
use of the CED). These problems can be
managed with policies, training, monitoring,
and accountability systems that provide
clear guidance (and consequences) to
officers regarding when and under what
conditions CEDs should be used and when
they should not be used (see Smith et al.,
2008). Good CED policies and training
should also require that officers evaluate the
totality of the circumstances before using a
CED, which would include the age, size,
gender, apparent physical capabilities, health
concerns of suspects, presence of flammable
liquids, and circumstances where falling
would pose unreasonable risks to the
suspect.
Another issue that policing agencies
may consider in light of this study is a
phenomenon that has been called “weaponoption overload.” Some police practitioners
have expressed concern about officers
having “too many tools on their belt,” such
as a CED, a collapsible baton, OC spray,

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

nunchuks, and a heavy flashlight in addition
to a firearm. Some departments have
discontinued the carrying of OC spray due
to its potential for affecting persons other
than the intended subject, or have
discontinued use of the baton because it
requires close contact. Police departments
that provide CEDs for officers may consider
the possibility of officers, in a fast-moving,
highly charged situation, becoming
temporarily confused if they have too many
force options on their belts. A decision to
deploy CEDs may cause some departments
to discontinue use of other less-lethal
options.

Next steps for researchers
As discussed in the limitations sections of
this report, one of the greatest barriers to
conducting use-of-force research is the
absence of uniformity and
comprehensiveness in the collection of data
about uses of force by LEAs across the
country. Our team was only able to identify
a small group of LEAs that were able to
participate readily in this study. While our
team is very thankful that they participated,
even for these agencies, the data available
for analysis were limited. We observed
limitations in content (information about
many of our areas of interest was not
collected by the LEAs), and timing (many of
the LEAs were limited in how long they
kept their force records, limiting our team to
no more than four years of analysis). Also,
the use-of-force tracking systems we
observed lacked a common architecture or
set of definitions. Similar barriers were
reported by Smith et al. (2008).
One possible solution to this problem
has been advanced by Smith and colleagues
(2008) involving a federal incentive for
agencies to collect use-of-force data using a
common set of data elements and definitions
to define what information is captured.

Smith and colleagues (2008) suggest that
Congress, with advice from the National
Institute of Justice, could fund a grant
program to state or local law enforcement
agencies that collect and make available for
research purposes data on the use of force by
police. Smith and colleagues (2008)
recommend that such a program could be
jump-started by NIJ field-testing with
volunteer LEAs a model use-of-force data
collection protocol. The recommended
approach by Smith et al. (2008) model
would provide useful data from a select
number of agencies and a model of how data
collection and analysis can assist with
agency policies and training, as well as
providing critical information to the research
community.
Smith and colleagues (2008) further
suggest that LEAs could be encouraged to
apply for grant funds to build the systems
necessary to collect use-of-force data, which
then would be used to support research and
analysis aimed at reducing the need for and
harmful consequences of police use of force.
Smith et al. also call for developing a
common software platform for data entry,
storage, and transmission to a research team
that would advise agency participants, audit
the incoming data, and create a publicly
available and non-proprietary dataset for
research purposes.
We agree with Smith et al. (2008) that
such a strategy would result in an important
national-scale data source that could be
maintained and updated regularly as new
use-of-force technologies came online, and
would likely spur new and better research on
how to reduce the harm that can occur when
LEAs use force.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

References

Adams, K. (1995). Measuring the
prevalence of police abuse of force. In
W.A. Geller & H. Toch (eds.), And
justice for all: Understanding and
controlling police abuse of force (pp.
61–98). Washington, D.C.: Police
Executive Research Forum.
Adang, O. Kaminski, R., Howell, M. &
Mensink, J. (under review). Assessing
the performance of pepper spray in useof-force encounters: The Dutch
experience. Policing: An International
Journal of Police Strategies and
Management.
Allison, William Thomas and Wardman,
Robert C. (2004). To Protect and to
Serve: A History of Police in America.
New Jersey: Prentice Hall.
Alpert, G.P. & Dunham, R.G. (1995). Police
use of deadly force: a statistical analysis
of the Metro-Dade Police Department.
Washington, DC: Police Executive
Research Forum.
Alpert, G.P. & Dunham, R.G. (2000).
Analysis of police use of force data.
Washington, D.C. National Institute of
Justice.
Alpert, G.P. & Dunham, R.G. (2004).
Understanding police use of force:
Officers, suspects, and reciprocity.
Cambridge, NY: Cambridge University
Press.
American Civil Liberties Union of Southern
California. (1995, June). Pepper spray

update: more fatalities, more questions.
Retrieved October 15, 2008 from
http://www.aclusc.org/attach/p/Pepper_
Spray_New_Questions.pdf.
American Civil Liberties Union of Northern
California. (2005). Stun gun fallacy:
How the lack of Taser regulation
endangers lives. San Francisco, CA:
Author.
Amnesty International. (2006).
USA Amnesty International’s
continuing concerns about Taser use.
AI Index AMR 51/030/2006,
http://www.amnestyusa.org/
countries/usa/ document.do?id=
ENGAMR510302006. London:
Amnesty International.
Amnesty International. (2004). Excessive
and lethal force? Amnesty
International’s concerns about deaths
and ill-treatment involving police use of
tasers. Retrieved December 26, 2006
from http://web.amnesty.org/library/
index/ENGAMR511392004.
Amnesty International. (1997). USA: Police
use of pepper spray – tantamount to
torture. Retrieved December 26, 2006
from http://web.amnesty.org/library/
Index/engAMR510671997.
Anglen, Robert (2004). Officer’s injury tied
to Taser: Doctor’s memo contradicts
safety claim. The Arizona Republic.
December 26.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Asher, H. (1983). Causal Modeling. Beverly
Hills, CA: Sage.

Issues for Field Settings. Boston:
Houghton-Mifflin, pps. 37–94.

Association of Chief Police Officers (2004).
Independent Evaluation of the
Operational Trial of TASER.
PricewaterhouseCoopers LLP. April.

De Leeuw, J. and I.G.G. Kreft (1986),
Random coefficient models for
multilevel analysis, Journal of
Educational Statistics 11, 57–85.

Bozeman, W., Winslow, J., Hauda, W.,
Graham, D., Martin, B., & Heck, J.
(2008). Injury profile of Taser electrical
conducted energy weapons (CEWs).
Retrieved August 12, 2008 from
http://www.calgarypolice.ca/news/
pdf/2007-LLW%20Study%20ACEP
%20Poster%20(Hi%20Res).pdf.

Delpish, A. N. (2006). A comparison of
estimators in Hierarchical Linear
Modeling: Restricted maximum
likelihood versus bootstrap via
minimum norm quadratic unbiased
estimators. Unpublished dissertation:
Florida State University, Department of
Statistics.

Brandl, S. (1996). In the line of duty: A
descriptive analysis of police assaults
and accidents. Journal of Criminal
Justice, 24, 255–264.

Dennis, A.J., Valentino, D.J., Walter, R.J.,
Nagy, K.K, Winners, J., Bokhari, F.,
Wiley, D.E., Joseph, K.T., & Roberts,
R.R. (2007). Acute effects of TASER
X26 discharges in a swine model. The
Journal of Trauma, Injury, Infection and
Critical Care, 63, 581–590.

Brandl, S. G. & M. S. Stroshine (2003).
Toward an understanding of the physical
hazards of police work. Police
Quarterly, 6, 172–191.
Charlotte-Mecklenburg Police Department.
(2006). Taser project: First year—full
deployment study. Charlotte, N.C.:
Author. Retrieved December 26, 2006
from http://www.charmeck.org/NR/
rdonlyres/e2alrn6jzttfx35m2gwabbqjzhl
ahc567iwa eusye62e5iz6amtldfmv4
mel3ojqzq3qtzd375dhuii4ozio7y3estb/1
+year+taser+study.pdf.
Christensen, L. B. (1988). Experimental
methodology. Boston : Allyn and Bacon.
Cohen, J. (1988). Statistical power analysis
for the behavioral sciences, 2nd edition.
Hillsdale, NJ: Lawrence Erlbaum
Associates.
Cook, T.D and Campbell, D.T. 1979.
Validity. Chapter 2 of QuasiExperimentation: Design and Analysis

Donnelly, Tara, Karen Douse, Michelle
Gardner, and David Wilkinson. 2002.
PSDB Evaluation of TASER Devices.
Police Scientific Development Branch,
Home Office Policing and Crime
Reduction Group. Hertfordshire, United
Kingdom. September.
Durose, M.R., Schmitt, E.L. & Langan, P.A.
(2005). Contacts between police and the
public. Washington, D.C.: Bureau of
Justice Statistics.
Ederheimer, Joshua. 2005. Use of Force
Tools. Chapter 3 of Chief Concerns:
Exploring the Challenges of Police Use
of Force. Washington, DC: Police
Executive Research Forum. pps. 57–81.
Edwards, S.M., Granfield, J., & Onnen, J.
(1997). Evaluation of pepper spray.
Washington, D.C.: National Institute of
Justice.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Gaines, Larry, Michael Kaune, and Larry L.
Miller. Criminal Justice in Action: The
Core. Belmont, CA: Wadsworth, 2001
Gauvin, R. (1995). Oleoresin capsicum
spray: A progress report. The ASLET
Journal, May/June, 29–32.
Gauvin, R. 1994. Oleoresin Capsicum
Spray: A Progress Report. Portland,
Oregon: Portland Police Department,
Internal Affairs Unit.
General Accounting Office. (2005). Taser
weapons: Use of Tasers by selected law
enforcement agencies. Washington,
D.C.: Author.
Goldberger, Arthur S. (1991) A Course in
Econometrics, Harvard University Press,
Cambridge, MA.
Granfield, J., Onnen, J., & Petty, C.S.
(1994). Pepper spray and in-custody
deaths. Alexandria, VA: International
Association of Chiefs of Police.
Henriquez, M. (1999). IACP national
database project on police use of force.
In Use of force by police: Overview of
national and local data (pp. 19–24).
Washington, DC: National Institute of
Justice and Bureau of Justice Statistics.
Hox, J.J. (1998), Multilevel modeling: when
and why, in: I. Balderjahn, R. Mathar
and M. Schader (eds), Classification,
data analysis, and data highways,
Springer Verlag, New York, 147–154.
Hox, J. J. (2002), Multilevel analysis;
techniques and applications, Lawrence
Erlbaum Associates, Mahwah, NJ.
Huber, P. J. (1967). The behavior of
maximum likelihood estimates under
nonstandard conditions. In Proceedings
of the Fifth Berkeley Symposium on
Mathematical Statistics and Probability.

Berkeley, CA: University of California
Press, vol. 1, 221–223.
International Association of Chiefs of Police
(IACP) (1995). Pepper Spray
Evaluation Project: Results of the
Introduction of Oleoresin Capsicum
(OC) into the Baltimore County, MD,
Police Department. Final Report
submitted to the National Institute of
Justice, Alexandria, VA: International
Association of Chiefs of Police.
International Association of Chiefs of Police
(2003). The Advanced TASER. IACP
Training. Key No. 567. Alexandria, VA.
James, S. (2002). Less Lethal. Law and
Order, 8, 87–93.
Jenkinson, E., Neeson, C., & Bleetman, A.
(2006). The relative risk of police useof-force options: Evaluating the
potential for deployment of electronic
weaponry. Journal of Clinical Forensic
Medicine, 13, 229–241.
Kaminski, R. J., Edwards, S. M., & J. W.
Johnson. (1999). Assessing the
incapacitative effects of pepper spray
during resistive encounters with the
police. Policing: An International
Journal of Police Strategies and
Management, 22, 7–29.
Kaminski, R. J., Edwards, S. M., & J. W.
Johnson. (1998). The deterrent effects of
oleoresin capsicum on assaults against
police: Testing the velcro-effect
hypothesis. Police Quarterly, 1, 1–20.
Kaminski, R.J. & Sorensen, D.W.M. (1995).
A multivariate analysis of individual,
situational, and environmental factors
associated with police assault injuries.
American Journal of Police, 14(3/4),
3–48.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Kelly, Sean (2004). Less-than-Lethal
Weapons in Demand among Police.
Denver Post.
Kaminski, R., DiGiovanni, C., & Downs, R.
(2004). The use of force between the
police and persons with impaired
judgment. Police Quarterly, 7, 311–338.
Kershaw, S. 2004. As Shocks Replace
Police Bullets, Deaths Drop but
Questions Arise. The New York Times,
March 7.
Kingshott, B. 1992. Do we need a graded
response to violent situations? The
Police Journal, 65(4): 297–307.
Kornblum, R. & Reddy, S. (1991). Effects
of the Taser in fatalities involving police
confrontation. Journal of Forensic
Science, 36, 434–448.
Kroll, M.W., Calkins, H., Luceri, R.M.,
Graham, M.A., & Heegaard, W.G.
(2008). Electronic control devices.
Canadian Medical Association Journal,
179, 342–343.
Lakkireddy, D., Wallick, D., Ryschon, K.,
Chung, M.K., Butany, J., Martin, D.,
Saliba, W., Kowalewski, W., Natale, A.,
& Tchou, P.J. (2006). Effects of cocaine
intoxication on the threshold of stun gun
induction of ventricular fibrillation.
Journal of the American College of
Cardiology, 48, 805–811.
Lakkireddy, D., Wallick, D., Verma, A.,
Ryschon, K., Kowalewski, W., Wazni,
O., Butany, J., Martin, D., Tchou, P.J.
(2008). Cardiac effects of electrical stun
guns: Does position of barbs contact
make a difference? Pacing and Clinical
Electrophysiology, 31. 398–408.
Levine, S.D., Sloane, C., Chan, T.C.,
Dunford, J. & Vilke, G. (2007). Cardiac
monitoring of human subjects exposed

to the Taser. Journal of Emergency
Medicine, 13, S47.
Levine, S.D., Sloane C., Chan, T.C., Vilke,
G., & Dunford, J. (2005). Cardiac
monitoring of subjects exposed to the
TASER. Academic Emergency
Medicine, 12, S71.
Lumb, R.C. & Friday, P.C. (1997). Impact
of pepper spray availability on police
officer use-of force decisions. Policing:
An International Journal of Police
Strategy and Management, 20, 136–148.
Mass, C. J. & Hox, J. J. (2002). Robustness
of multilevel parameter estimates
against small sample sizes. Unpublished
Paper. The Netherlands: Utrecht
University.
McDaniel, W.C., Stratbucker, R.A.,
Nerheim, M., & Brewer, J.E., (2005).
Cardiac safety of neuromuscular
incapacitating defensive devices. Pacing
and Clinical Electrophysiology, 28,
s284–s287.
Morabito, E.V. & Doerner, W.G. (1997).
Police use of less-than-lethal force:
Oleoresin capsicum (OC) spray.
Policing: An International Journal of
Police Strategies & Management, 20,
680–697.
Meyer, G. (1992). Nonlethal weapons vs.
conventional police tactics: Assessing
injuries and liabilities. The Police Chief,
59, 10–17.
Nanthakumar, K., Billingsley, I.M., Masse,
S., Dorian, P., Cameron, D., Chauhan,
V.S., Downar, E., & Sevaptsidis, E.
(2006). Cardiac electrophysiological
consequences of neuromuscular
incapacitating device discharges.
Journal of the American College of
Cardiology, 48, 798–804.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Nanthakumar, K., Masse, S., Umapathy, K.,
Dorian, P., Sevaptsidis, E, & Waxman,
M. (2008). Cardiac stimulation with
high voltage discharge from stun guns.
Canadian Medical Association Journal,
178, 1451–1457.
National Institute of Justice. (2008). Study of
deaths following electro muscular
disruption: An interim report.
Washington, DC: National Institute of
Justice.
National Institute of Justice. (2003). The
effectiveness and safety of pepper spray.
Research for Practice. Washington,
D.C.: National Institute of Justice.
Nielsen, E. (2001). Taser International takes
the Taser to a higher level. Law and
Order, 5, 57–62.
Nowicki, E. (1993). Oleoresin Capsicum:
A non-lethal force alternative. Law
Enforcement Technology, 20, 24–27.
Pearson, C. (2003). Less-Lethal Weapons.
Retrieved April 5, 2003, from
http://www.nleta.com/articles/lesslethal
weapons.html
Police Executive Research Forum (2005).
PERF Conducted Energy Device
Policy and Training Guidelines for
Consideration. PERF Center on Force &
Accountability. Washington D.C.
October 20, 2005.
Petty, C.S. (2004). Deaths in police
confrontations when oleoresin capsicum
is used: Final report. Washington, D.C.:
National Institute of Justice.
Phillips, S.W. 1994. Oleoresin Capsicum in
Buffalo, Technical Report 04-95,
Canadian Police Research Centre,
Ottawa.

Raudenbush, S. W. & Bryk, A. S. (2002).
Hierarchical linear models:
Applications and data analysis methods
(2nd ed.). Thousand Oaks, CA: Sage
Publications.
Raudenbush, S.W., Bryk, A.S., Cheong,
Y.F. and R. Cogdon (2004). HLM 6:
Hierarchical Linear and Nonlinear
Modeling. Chicago: Scientific Software
International.
Raudenbush, S.W., & Liu, Xiao-Feng
(2000). Statistical power and optimal
design for multisite randomized trials.
Psychological Methods, 5(2), 199–213.
Raudenbush, S. W. & Willms, J. D. (1995).
The estimation of school effects.
Journal of Educational and Behavioral
Statistics, 20, 307–335.
Robin, G. 1996. The elusive and
illuminating search for less-than-lethal
alternatives to deadly force. Police
Forum, 62:1–8.
Rogers, W. H. 1993. Regression standard
errors in clustered samples. Stata
Technical Bulletin 13: 19–23. Reprinted
in Stata Technical Bulletin Reprints,
vol. 3, 88–94.
Roy, O.Z. & Podgorski, A.S. (1989). Tests
on a shocking device – the stun gun.
Medical and Biological Engineering
and Computing, 27, 445–448.
Sanchez, George. 2004. “TASER Use
Grows Despite Safety Questions.” The
Salinas Californian. September 4.
Seattle Police Department. (2002). The M26
taser year one implementation. Seattle,
WA: Author.
Shadish, W. R., Cook, T. D. and D.T.
Campbell, D.T. (2002). Experimental
and quasi-experimental designs for

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

generalized causal inference. New
York: Houghton Mifflin Company.
Shin, C. (2002). Less-lethal weapons could
be more lethal. Forty Niner, 8.
Smith, M.R. & Alpert, G.P. (2000). Pepper
spray: A safe and reasonable response to
suspect verbal resistance. Policing: An
International Journal of Police Strategy
and Management, 23, 233–245.
Smith, M.R., Kaminski, R.J., Alpert, G.P.,
Fridell, L.A., MacDonald, J. and B.
Kubu (2008). A multi-method
evaluation of police use of force
outcomes. Final report to the National
Institute of Justice.
Smith, M.R., Kaminski, R.J., Rojek, J.,
Alpert, G.P., & Mathis, J. (2007) The
impact of conducted energy devices and
other types of force and resistance on
police and suspect injuries. Policing:
An International Journal of Police
Strategies and Management, 30,
443–426.
Smith, M.R. & Petrocelli, M. (2002). The
effectiveness of force used by police in
making arrests. Police Practice and
Research, 3, 201–215.
Snijders, T. A. B. and R. Bosker (1999),
Multilevel analysis. An introduction to
basic and advanced multilevel modeling,
Sage, Thousand Oaks, CA.
Stratbucker, R., Roeder, R., & Nerheim, M.
(2003). Cardiac safety of high voltage
Taser X26 waveform. Engineering in
Medicine and Biology Society,
Proceeding of the 25th Annual
International Conference of the IEEE
EMBS, Cancun, Mexico, 1094–678X, 4,
3261–3262.
Streicher, Thomas H. (2005). Kennedy
School of Government, Harvard

University, and Office of Justice
Programs, U.S. Department of Justice
on-line conference on less-lethal force,
January 18.
Strote, J. & Hutson, H.R. (2008). Taser
safety remains unclear. Annals of
Emergency Medicine, 52, 84–85.
Strote J., & Hutson, H.R. (2006). Taser use
in restraint-related deaths. Prehospital
Emergency Care,10, 447–450.
TASER International, Inc. (2004). Press
releases. June. www.Taser.com/
press releases.
Trochim, W. (Ed.), (1986). Editor's Notes.
Advances in quasi-experimental design
and analysis. New Directions for
Program Evaluation Series, Number 31,
San Francisco, CA: Jossey-Bass.
Trochim, W., Cappelleri, J., and C.S.
Reichardt (1991). Random measurement
error does not bias the treatment effect
estimate in the regression-discontinuity
design: When an interaction effect is
present. Evaluation Review, 15:
571–604.
Vilke, G.M., & Chan, T.C. ( 2007) Less
lethal technology: Medical issues.
Policing: An International Journal of
Police Strategies and Management, 30,
341–357.
Vilke, G.M., Sloane, C., Bouton, K.D.,
Kolkhorst, F.W., Levine, S., Neuman,
T., Castillo, E., & Chan, T.C. (2007).
Physiological effects of a conducted
electrical weapon on human subjects.
Annals of Emergency Medicine, 26, 1–4.
Walter, R., Dennis, A., Valentina, D.,
Margeta, B., Nagy, K, Bokhari, F.,
Wiley, D., Joseph, K., & Roberts, R.
(2008). TASER X26 discharges in
swine produce potentially fatal

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

ventricular arrhythmias. Academic
Emergency Medicine, 15, 66–73.
Weisburd, D., Petrosino, A., & Mason, G.
(1991). When bigger is not better:
Design sensitivity in a sample of
criminal justice experiments. In N.
Morris & M. Tonry (Eds.), Crime and
justice: An annual review of research,
Vol. 15. Chicago: The University of
Chicago Press.
White, H. 1980. A heteroskedasticityconsistent covariance matrix estimator
and a direct test for heteroskedasticity.
Econometrica 48: 817–830.
Williams, R. L. 2000. A note on robust
variance estimation for clustercorrelated data. Biometrics 56: 645–646.
Wilmette, A. (2001). The Jaycor pepperball
system. Law and Order, 11, 92–95.

www.ojp.usdoj.gov/nij/topics/technology/
less-lethal/conducted-energydevices.htm

www.azcentral.com/specials/special43/
articles/1224taserlist24-ON.html.
www.aclu.org/police/abuse/19977
prs20051006.html).
www.cbc.ca/news/background/tasers/.
www.ojp.usdoj.gov/nij/topics/
technology/less-lethal/
conducted-energy-devices.htm.
www.ojp.usdoj.gov/nij/topics/
technology/less-lethal/
how-ceds-work.htm.
www.taser.com/documents/TASERS_
saving_lives_compilation-short.pdf

Wooldridge, J. M. 2002. Econometric
Analysis of Cross Section and Panel
Data. Cambridge, MA: MIT Press.

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Appendix 1: Logistic regression Model 1 and Model 2 both without
corrections for nested standard errors

Outcome measures/dependent variables
Suspect Injury

Officer Injury

Model 1
Variables in logisitc regression model

Odds Ratio

Model 2
P value

Odds Ratio

Model 1

P value

Odds Ratio

Model 2

P value

Odds Ratio

P value

Intercept

-1.45

0.23

0.12

<.0001

-1.39

0.25

0.17

<.0001

-2.26

0.10

0.18

<.0001

-3.00

0.05

0.30

<.0001

Does Agency Deploy CED (1= yes, 0=no)

-0.39

0.68

0.10

<.0001

0.25

1.28

0.13

0.0638

0.02

1.02

0.15

0.9131

-0.17

0.84

0.23

0.4576

0.77

2.17

0.13

<.0001

0.70

2.00

0.13

<.0001

0.89

2.43

0.19

<.0001

1.06

2.89

0.20

<.0001

-0.57

0.56

0.14

<.0001

-0.33

0.72

0.17

0.0511

-1.20

0.30

0.20

<.0001

-1.14

0.32

0.30

0.0001

Suspect race (White= 1, Non-White=0)

0.41

1.51

0.05

<.0001

0.18

1.20

0.09

0.053

0.23

1.26

0.08

0.0038

-0.25

0.78

0.17

0.1378

Suspect gender (Male=1, female=0)

0.55

1.73

0.08

<.0001

0.09

1.09

0.14

0.5086

0.03

1.03

0.11

0.7642

0.13

1.14

0.23

0.5708

Suspect age (1= < 25 years old, 0= > 25 years old)

0.14

1.15

0.05

0.0035

0.13

1.13

0.09

0.1419

0.23

1.26

0.07

0.002

0.08

1.08

0.15

0.6014

Suspect resistant behavior (1=physical aggression by suspect, 0= non-physical aggression)

0.37

1.45

0.09

<.0001

0.86

2.37

0.15

<.0001

Suspect had weapon (1=yes, 0= no)

0.86

2.35

0.11

<.0001

0.94

2.56

0.17

<.0001

Time frame of incident (post-CED/comparable period= 1, pre CED/comparable period=0)
Interaction CED * Time Frame (1= CED and post period)

Outcome measures/dependent variables
Suspect Medicalization
Model 1
Variables in logisitc regression model

Odds Ratio

Officer Medicalization
Model 2

P value

Odds Ratio

Model 1
β

P value

Odds Ratio SE

Model 2
P value

Odds Ratio

P value

-1.00

0.370 0.120

<.0001

-1.39 0.25021463 0.188

<.0001

-3.21

0.040 0.282

<.0001

-5.07

0.01

0.47

<.0001

Does Agency Deploy CED (1= yes, 0=no)

0.84

2.308 0.105

<.0001

0.464 1.58993453 0.141

0.001

1.23

3.432 0.257

<.0001

1.07

2.92

0.35

0.0022

Time frame of incident (post-CED/comparable period= 1, pre CED/comparable period=0)

0.80

2.224 0.137

<.0001

0.933 2.54212921 0.143

<.0001

1.62

5.042 0.283

<.0001

2.21

9.12

0.30

<.0001

-1.54

0.215 0.147

<.0001

-0.64

0.5246766 0.185

0.0005

-2.04

0.131 0.298

<.0001

-2.94

0.05

0.41

<.0001

Suspect race (White= 1, Non-White=0)

0.42

1.523 0.052

<.0001

0.282 1.32629518 0.103

0.0064

0.09

1.091 0.094

0.3584

-0.20

0.82

0.21

0.3479

Suspect gender (Male=1, female=0)

0.41

1.506 0.070

<.0001

0.306 1.35749933 0.144

0.0333

0.07

1.074 0.135

0.5952

0.43

1.53

0.32

0.18

-0.13

0.879 0.048

0.0076

-0.32 0.72594582 0.093

0.0006

0.06

1.061 0.090

0.508

-0.05

0.95

0.19

0.7798

Suspect resistant behavior (1=physical aggression by suspect, 0= non-physical aggression)

0.858 2.35868545 0.094

<.0001

0.80

2.23

0.19

<.0001

Suspect had weapon (1=yes, 0= no)

0.485 1.62375671 0.123

<.0001

1.84

6.27

0.21

<.0001

Intercept

Interaction CED * Time Frame (1= CED and post period)

Suspect age (1= < 25 years old, 0= > 25 years old)

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Variables in logisitc regression model

Outcome measures/dependent variables
Suspect Injury Minor vs. Severe
Officer Injury Minor vs. Severe
Model 1
Model 2
Model 1
Model 2
Odds Ratio Std Err P value
Odds Ratio Std Err P value
Odds Ratio Std Err P value
Odds Ratio Std Err
β
β
β

P value

Intercept

-2.50

0.08

0.36

<.0001

-3.28

0.04

0.46

<.0001

-2.35

0.10

0.58

<.0001

-2.76

0.06

0.79

0.0005

Does Agency Deploy CED (1= yes, 0=no)

-0.13

0.87

0.27

0.6215

0.50

1.65

0.34

0.1342

-0.62

0.54

0.49

0.2088

-0.24

0.79

0.64

0.7087

0.03

1.03

0.32

0.9169

-0.34

0.71

0.35

0.3317

-0.46

0.63

0.60

0.439

-0.55

0.58

0.63

0.3801

-0.58

0.56

0.37

0.118

-0.23

0.79

0.45

0.6107

0.56

1.75

0.69

0.4166

0.63

1.87

0.91

0.4888

Suspect race (White= 1, Non-White=0)

0.11

1.12

0.17

0.5068

0.13

1.14

0.23

0.5705

-0.14

0.87

0.34

0.679

0.56

1.75

0.49

0.2543

Suspect gender (Male=1, female=0)

0.04

1.04

0.27

0.8952

0.06

1.06

0.34

0.87

-0.08

0.92

0.42

0.8495

-0.07

0.93

0.65

0.9151

-0.14

0.87

0.17

0.4175

-0.34

0.71

0.24

0.1467

-0.01

0.99

0.31

0.9837

0.44

1.56

0.46

0.3334

Suspect resistant behavior (1=physical aggression by suspect, 0= non-physical aggression)

0.71

2.04

0.22

0.0014

0.05

1.05

0.45

0.92

Suspect had weapon (1=yes, 0= no)

1.21

3.37

0.27

<.0001

0.09

1.09

0.56

0.8731

Time frame of incident (post-CED/comparable period= 1, pre CED/comparable period=0)
Interaction CED * Time Frame (1= CED and post period)

Suspect age (1= < 25 years old, 0= > 25 years old)

Outcome measures/dependent variables
Suspect Hospitalization
Variables in logisitc regression model

Model 1
Odds Ratio SE

P value

Officer Hospitalization

Model 2
Odds Ratio SE

P value

Model 1
Odds Ratio SE

P value

Model 2
Odds Ratio SE

P value

Intercept

-0.85

0.43

0.14

<.0001

-0.70

0.50

0.18

0.0001

-2.91

0.05

0.32

<.0001

-3.37

0.03

0.43

<.0001

Does Agency Deploy CED (1= yes, 0=no)

-0.23

0.80

0.12

0.0536

0.54

1.71

0.15

0.0003

0.03

1.03

0.28

0.9077

0.02

1.02

0.35

0.9462

0.21

1.23

0.18

0.2417

0.12

1.13

0.18

0.5086

0.39

1.48

0.35

0.2645

0.41

1.50

0.36

0.2591

-0.73

0.48

0.19

<.0001

-0.70

0.49

0.21

0.0007

-0.23

0.80

0.38

0.5439

-0.48

0.62

0.43

0.2665

Suspect race (White= 1, Non-White=0)

0.18

1.20

0.06

0.0039

-0.19

0.83

0.10

0.0442

-0.37

0.69

0.15

0.0121

-0.38

0.68

0.23

0.0941

Suspect gender (Male=1, female=0)

0.21

1.24

0.09

0.0155

0.18

1.20

0.13

0.1738

-0.04

0.96

0.19

0.8204

0.11

1.11

0.30

0.7255

-0.27

0.76

0.06

<.0001

-0.24

0.79

0.09

0.0059

-0.15

0.86

0.13

0.2738

-0.02

0.98

0.19

0.9338

0.20

1.22

0.09

0.0227

0.74

2.10

0.19

0.0001

-0.12

0.88

0.14

0.3723

-0.27

0.77

0.32

0.4041

Time frame of incident (post-CED/comparable period= 1, pre CED/comparable period=0)
Interaction CED * Time Frame (1= CED and post period)

Suspect age (1= < 25 years old, 0= > 25 years old)
Suspect resistant behavior (1=physical aggression by suspect, 0= non-physical aggression)
Suspect had weapon (1=yes, 0= no)

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Appendix 2: Logistic regression model with a correction for
nested standard errors (Model 1 only)
Suspect Injury: Logistic Regression
Number of Observations = 9,324
Wald X2(6) = 138.30
Prob > X2 = 0.0000
Psuedo r2 = 0.024

Coefficient
-0.386

Robust Std.
Error
0.676

z
-0.57

P>|z|
0.569

0.774

0.523

1.48

0.139

-0.251

1.798

-0.572

0.554

-1.03

0.301

-1.658

0.513

Suspect White
Suspect Male

0.413
0.548

0.188
0.136

2.19
4.03

0.028
0.000

0.044
0.281

0.783
0.815

Suspect under 25 years of age

0.142

0.037

3.82

0.000

0.069

0.214

Constant

1.450

0.429

-3.38

0.001

-2.291

-0.608

Agency deploys CED
Post-test Period
Agency used CEDs in post period

95% Conference
Interval
-1.711
0.940

Officer Injury: Logistic Regression
Number of Observations = 7,963
Wald X2(6) = 52.22
Prob > X2 = 0.0000
Psuedo r2 = 0.0162

Coefficient
0.017

Robust
Std. Err.
0.526

z
0.03

P>|z|
0.975

0.887

0.479

1.85

0.064

-0.052

1.826

Agency used CEDs in post period

-1.203

0.512

-2.35

0.019

-2.207

-0.200

Suspect White
Suspect Male
Suspect under 25 years of age
Constant

0.032
0.230
0.231
-2.262

0.157
0.098
0.082
0.314

0.20
2.34
2.81
-7.22

0.838
0.019
0.005
0.000

-0.276
0.037
0.070
-2.877

0.340
0.423
0.392
-1.648

Agency deploys CED
Post-test Period

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

95% Conference
Interval
-1.015
1.048

Suspect Medical Attention: Logistic Regression
Number of Observations = 7,696
Wald X2(6) = 172.91
Prob > X2 = 0.0000
Psuedo r2 = 0.0334

Coefficient
0.836

Robust
Std. Err.
0.635

z
1.32

P>|z|
0.188

0.799

0.402

1.99

0.047

0.012

1.587

Agency used CEDs in post period

-1.539

0.685

-2.25

0.025

-2.881

-0.197

Suspect White
Suspect Male
Suspect under 25 years of age
Constant

0.420
0.410
-0.129
-0.996

0.107
0.178
0.093
0.635

3.92
2.30
-1.39
-1.57

0.000
0.021
0.164
0.117

0.210
0.061
-0.310
-2.241

0.630
0.759
0.053
0.250

95% Conference
Interval
0.465
2.002

Agency deploys CED
Post-test Period

95% Conference
Interval
-0.407
2.080

Officer Medical Attention: Logistic Regression
Number of Observations = 5,303
Wald X2(6) = 46.66
Prob > X2 = 0.0000
Psuedo r2 = 0.0165
Coefficient
1.233

Robust
Std. Err.
0.392

z
3.14

P>|z|
0.002

1.618

0.899

1.80

0.072

-0.144

3.380

Agency used CEDs in post period

-2.036

0.955

-2.13

0.033

-3.908

-0.164

Suspect White
Suspect Male
Suspect under 25 yearrs of age
Constant

0.087
0.072
0.059
-3.212

0.109
0.070
0.097
0.253

0.79
1.03
0.61
-12.72

0.427
0.304
0.539
0.000

-0.127
-0.065
-0.130
-3.707

0.301
0.208
0.249
-2.717

Agency deploys CED
Post-test Period

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect Hospitalization: Logistic Regression
Number of Observations = 6,996
Wald X2(6) = 173.45
Prob > X2 = 0.0000
Psuedo r2 = 0.0211

Agency deploys CED
Post-test Period
Agency used CEDs in post period
Suspect White
Suspect Male
Suspect under 25 yrs of age
Constant

Coefficient
-0.229
0.206
-0.732
0.182
0.214
-0.274
-0.853

Robust
Std. Err.
0.773
0.491
0.704
0.223
0.085
0.113
0.623

z
-0.30
0.42
-1.04
0.82
2.51
-2.42
-1.37

P>|z|
0.767
0.675
0.299
0.415
0.012
0.016
0.171

95% Conference
Interval
-1.745
1.286
-0.757
1.169
-2.113
0.649
-0.255
0.619
0.047
0.380
-0.496
-0.052
-2.073
0.367

P>|z|
0.852
0.341
0.586
0.001
0.706
0.221
0.000

95% Conference
Interval
-0.308
0.373
-0.412
1.191
-1.054
0.595
-0.589
-0.159
-0.270
0.183
-0.378
0.087
-3.287
-2.538

Officer Hospitalization: Logistic Regression
Number of Observations = 5,232
Wald X2(6) = 26.81
Prob > X2 = 0.0002
Psuedo r2 = 0.0051

Agency deploys CED
Post-test Period
Agency used CEDs in post period
Suspect White
Suspect Male
Suspect under 25 yrs of age
Constant

Coefficient
0.032
0.390
-0.229
-0.374
-0.044
-0.145
-2.912

Robust
Std. Err.
0.174
0.409
0.421
0.110
0.116
0.119
0.191

z
0.19
0.95
-0.54
-3.40
-0.38
-1.22
-15.24

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect Injury: Minor vs. Severe (Among those injured): Logistic Regression
Number of Observations = 2,929
Wald X2(6) = 57.53
Prob > X2 = 0.0000
Psuedo r2 = 0.0125

Agency deploys CED
Post-test Period
Agency used CEDs in post period
Suspect White
Suspect Male
Suspect under 25 yrs of age
Constant

Coefficient
-0.134
0.034
-0.582
0.112
0.036
-0.136
-2.505

Robust
Std. Err.
0.503
0.219
0.257
0.158
0.274
0.113
0.602

z
-0.27
0.15
-2.27
0.71
0.13
-1.20
-4.16

P>|z|
0.791
0.877
0.023
0.480
0.896
0.231
0.000

95% Conference
Interval
-1.120
0.853
-0.395
0.463
-1.086
-0.079
-0.198
0.421
-0.501
0.573
-0.358
0.086
-3.685
-1.325

Officer Injury: Minor vs. Severe (Among those injured): Logistic Regression
Number of Observations = 956
Wald X2(6) = 12.16
Prob > X2 = 0.0584
Psuedo r2 = 0.0046

Agency deploys CED
Post-test Period
Agency used CEDs in post period
Suspect White
Suspect Male
Suspect under 25 yrs of age
Constant

Coefficient
-0.621
-0.463
0.561
-0.139
-0.080
-0.006
-2.349

Robust
Std. Err.
0.425
0.353
0.482
0.488
0.259
0.241
0.322

z
-1.46
-1.31
1.17
-0.29
-0.31
-0.03
-7.29

P>|z|
0.144
0.189
0.244
0.775
0.757
0.979
0.000

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

95% Conference
Interval
-1.454
0.211
-1.155
0.228
-0.383
1.506
-1.095
0.817
-0.588
0.428
-0.478
0.465
-2.981
-1.718

Appendix 3: HLM Results
Officer injury: HLM model
--------------------------------------------------------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
--------------------------------------------------------------------------------------------------------------------For
Intercept1, B0
INTERCEPT2, G00
-2.411
0.411
-5.871
0.000
AGENCY DEPLOYS CED, G01
-0.031
0.559
-0.055
0.958
POPULATION DENSITY, G02
-0.000
0.000
-1.258
0.249
# OF OFFICERS per 100,000, G03
-0.000
0.000
-1.375
0.211
For
POST_PERIOD slope, B1
INTERCEPT2, G10
AGENCY DEPLOYS CED, G11
POPULATION DENSITY, G12
# OF OFFICERS per 100,000, G13

1.064
-1.494
-0.000
-0.000

1.116
1.591
0.000
0.000

0.954
-0.939
-1.095
-0.776

0.372
0.379
0.310
0.463

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
AGENCY DEPLOYS CED, G21
POPULATION DENSITY, G22
# OF OFFICERS per 100,000, G23

-0.648
0.819
0.000
-0.000

0.302
0.312
0.000
0.000

-2.142
2.621
1.259
-0.120

0.069
0.034
0.249
0.908

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
AGENCY DEPLOYS CED, G31
POPULATION DENSITY, G32
# OF OFFICERS per 100,000, G33

0.189
-0.430
-0.000
0.000

0.494
0.582
0.000
0.000

0.383
-0.738
-0.378
1.092

0.713
0.484
0.716
0.311

For
SUSPECT UNDER
25 YEARS OLD slope, B4
INTERCEPT2, G40
AGENCY DEPLOYS CED, G41
POPULATION DENSITY, G42
# OF OFFICERS per 100,000, G43

-0.005
-0.083
-0.000
0.000

0.425
0.567
0.000
0.000

-0.012
-0.146
-0.105
0.049

0.991
0.888
0.920
0.963

---------------------------------------------------------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect injury: HLM model
---------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT1, B0
INTERCEPT2, G00
-1.093
0.576
-1.896
0.099
AGENCY DEPLOYS CED, G01
-0.402
0.803
-0.500
0.632
POPULATION DENSITY, G02
-0.000
0.000
-1.052
0.328
AVERAGE NUMBER OF, G03
-0.000
0.000
-1.079
0.317
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
0.735
AGENCY DEPLOYS CED, G11
-0.639
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
-0.000
OFFICERS PER 100,000 IN POPULATION

1.180
1.632
0.000
0.000

0.623
-0.392
-1.074
-0.669

0.553
0.707
0.319
0.525

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
0.032
AGENCY DEPLOYS CED, G21
0.458
POPULATION DENSITY, G22
0.000
AVERAGE NUMBER OF, G23
-0.000
OFFICERS PER 100,000 IN POPULATION

0.216
0.258
0.000
0.000

0.147
1.778
1.718
-1.614

0.888
0.118
0.129
0.150

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.404
AGENCY DEPLOYS CED, G31
0.181
POPULATION DENSITY, G32
0.000
AVERAGE NUMBER OF, G33
0.000
OFFICERS PER 100,000 IN POPULATION

0.412
0.531
0.000
0.000

0.982
0.342
0.249
0.549

0.359
0.742
0.811
0.599

B4
0.168
0.183
0.000
0.000

0.535
-0.887
0.040
0.245

0.609
0.405
0.969
0.814

For SUSPECT UNDER 25 YEARS OLD slope,
INTERCEPT2, G40
0.090
AGENCY DEPLOYS CED, G41
-0.162
POPULATION DENSITY, G42
0.000
AVERAGE NUMBER OF, G43
0.000
OFFICERS PER 100,000 IN POPULATION

----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer Medical Attention: HLM model
----------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT1, B0
INTERCEPT2, G00
-3.554
0.801
-4.439
0.004
AGENCY DEPLOYS CED, G01
0.292
0.982
0.297
0.776
POPULATION DENSITY, G02
-0.000
0.000
-1.470
0.192
AVERAGE NUMBER OF, G03
-0.000
0.001
-0.397
0.705
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
-0.779
AGENCY DEPLOYS CED, G11
-1.744
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
-0.003
OFFICERS PER 100,000 IN POPULATION

2.940
2.842
0.000
0.003

-0.265
-0.613
-0.854
-1.184

0.800
0.562
0.426
0.282

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
-0.439
AGENCY DEPLOYS CED, G21
0.366
POPULATION DENSITY, G22
0.000
AVERAGE NUMBER OF, G23
0.000
OFFICERS PER 100,000 IN POPULATION

0.501
0.486
0.000
0.000

-0.875
0.753
0.694
0.260

0.415
0.480
0.514
0.804

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.794
AGENCY DEPLOYS CED, G31
-0.256
POPULATION DENSITY, G32
0.000
AVERAGE NUMBER OF, G33
-0.001
OFFICERS PER 100,000 IN POPULATION

1.017
1.030
0.000
0.001

0.781
-0.249
0.981
-1.368

0.464
0.812
0.365
0.220

For SUSPECT UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
-1.143
0.831
AGENCY DEPLOYS CED, G41
-0.012
0.753
POPULATION DENSITY, G42
-0.000
0.000
AVERAGE NUMBER OF, G43
-0.002
0.001
OFFICERS PER 100,000 IN POPULATION

-1.376
-0.016
-0.960
-2.241

0.218
0.988
0.375
0.065

----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect Medical Attention: HLM model
----------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
----------------------------------------------------------------------For
INTERCEPT1, B0
INTERCEPT2, G00
-0.063
1.448
-0.043
0.967
AGENCY DEPLOYS CED, G01
-0.895
1.976
-0.453
0.664
POPULATION DENSITY, G02
-0.000
0.000
-0.925
0.386
AVERAGE NUMBER OF, G03
0.000
0.001
0.170
0.870
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
0.584
AGENCY DEPLOYS CED, G11
-0.608
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
-0.001
OFFICERS PER 100,000 IN POPULATION

0.940
1.274
0.000
0.000

0.621
-0.478
-0.803
-1.909

0.554
0.647
0.449
0.097

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
-0.207
AGENCY DEPLOYS CED, G21
0.705
POPULATION DENSITY, G22
0.000
AVERAGE NUMBER OF, G23
-0.000
OFFICERS PER 100,000 IN POPULATION

0.272
0.323
0.000
0.000

-0.764
2.184
0.634
-0.086

0.470
0.065
0.546
0.934

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.112
AGENCY DEPLOYS CED, G31
0.649
POPULATION DENSITY, G32
0.000
AVERAGE NUMBER OF, G33
0.000
OFFICERS PER 100,000 IN POPULATION

0.605
0.779
0.000
0.000

0.185
0.834
0.224
0.053

0.859
0.432
0.829
0.960

For SUSPECTS AGE UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
-0.359
0.271
-1.327
0.226
AGENCY DEPLOYS CED, G41
0.088
0.335
0.262
0.801
POPULATION DENSITY, G42
0.000
0.000
0.845
0.426
AVERAGE NUMBER OF, G43
0.000
0.000
1.020
0.342
OFFICERS PER 100,000 IN POPULATION
-----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer Hospitalization: HLM model
----------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT1, B0
INTERCEPT2, G00
-3.262
0.291
-11.197
0.000
AGENCY DEPLOYS CED, G01
0.062
0.316
0.196
0.852
POPULATION DENSITY, G02
-0.000
0.000
-1.480
0.199
AVERAGE NUMBER OF, G03
0.000
0.000
0.506
0.634
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
0.304
AGENCY DEPLOYS CED, G11
-0.057
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
0.001
OFFICERS PER 100,000 IN POPULATION

0.402
0.465
0.000
0.000

0.758
-0.123
-0.829
1.873

0.483
0.908
0.445
0.119

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
-1.297
AGENCY DEPLOYS CED, G21
1.111
POPULATION DENSITY, G22
-0.000
AVERAGE NUMBER OF, G23
0.000
OFFICERS PER 100,000 IN POPULATION

0.591
0.601
0.000
0.001

-2.196
1.847
-0.129
0.687

0.078
0.123
0.902
0.522

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.077
AGENCY DEPLOYS CED, G31
-0.128
POPULATION DENSITY, G32
-0.000
AVERAGE NUMBER OF, G33
0.000
OFFICERS PER 100,000 IN POPULATION

0.625
0.737
0.000
0.000

0.123
-0.174
-1.489
1.183

0.908
0.869
0.196
0.290

0.040
-0.211
1.467
-1.459

0.970
0.842
0.202
0.204

For SUSPECT AGE UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
0.016
0.405
AGENCY DEPLOYS CED, G41
-0.101
0.479
POPULATION DENSITY, G42
0.000
0.000
AVERAGE NUMBER OF, G43
-0.000
0.000
OFFICERS PER 100,000 IN POPULATION

----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect Hospitalization: HLM model

----------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT1, B0
INTERCEPT2, G00
-0.386
1.263
-0.306
0.769
AGENCY DEPLOYS CED, G01
-1.256
1.719
-0.731
0.489
POPULATION DENSITY, G02
0.000
0.000
0.422
0.685
AVERAGE NUMBER OF, G03
-0.001
0.001
-2.222
0.061
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
0.331
AGENCY DEPLOYS CED, G11
-0.121
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
0.000
OFFICERS PER 100,000 IN POPULATION

0.289
0.362
0.000
0.000

1.145
-0.335
-0.646
0.279

0.290
0.747
0.539
0.788

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
-0.163
AGENCY DEPLOYS CED, G21
0.490
POPULATION DENSITY, G22
-0.000
AVERAGE NUMBER OF, G23
0.000
OFFICERS PER 100,000 IN POPULATION

0.253
0.272
0.000
0.000

-0.644
1.803
-2.031
0.555

0.540
0.114
0.081
0.596

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.206
AGENCY DEPLOYS CED, G31
0.038
POPULATION DENSITY, G32
-0.000
AVERAGE NUMBER OF, G33
0.000
OFFICERS PER 100,000 IN POPULATION

0.391
0.469
0.000
0.000

0.526
0.081
-0.888
1.619

0.615
0.938
0.404
0.149

-2.592
1.285
0.624
0.785

0.036
0.240
0.552
0.458

For SUSPECT AGE UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
-0.763
0.294
AGENCY DEPLOYS CED, G41
0.469
0.365
POPULATION DENSITY, G42
0.000
0.000
AVERAGE NUMBER OF, G43
0.000
0.000
OFFICERS PER 100,000 IN POPULATION

-----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Officer Severity of Injury Minor/Severe: HLM model
---------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT 1, B0
INTERCEPT2, G00
-2.130
0.508
-4.194
0.005
AGENCY DEPLOYS CED, G01
-1.069
0.619
-1.726
0.127
POPULATION DENSITY, G02
0.000
0.000
1.774
0.119
AVERAGE NUMBER OF, G03
0.000
0.000
1.253
0.251
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
-0.804
AGENCY DEPLOYS CED, G11
1.075
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
-0.000
OFFICERS PER 100,000 IN POPULATION

0.688
0.824
0.000
0.000

-1.170
1.305
-0.156
-0.891

0.281
0.233
0.881
0.403

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
2.304
AGENCY DEPLOYS CED, G21
-2.906
POPULATION DENSITY, G22
0.000
AVERAGE NUMBER OF, G23
0.000
OFFICERS PER 100,000 IN POPULATION

0.737
0.942
0.000
0.000

3.125
-3.084
2.147
1.860

0.018
0.019
0.068
0.105

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
-0.282
AGENCY DEPLOYS CED, G31
0.665
POPULATION DENSITY, G32
0.000
AVERAGE NUMBER OF, G33
-0.000
OFFICERS PER 100,000 IN POPULATION

1.024
1.131
0.000
0.000

-0.276
0.588
1.000
-0.739

0.791
0.574
0.351
0.484

0.992
-0.816
0.297
-1.407

0.355
0.442
0.775
0.202

For SUSPECT AGE UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
0.664
0.670
AGENCY DEPLOYS CED, G41
-0.624
0.764
POPULATION DENSITY, G42
0.000
0.000
AVERAGE NUMBER OF, G43
-0.000
0.000
OFFICERS PER 100,000 IN POPULATION

----------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Suspect Severity of Injury Minor/Severe: HLM model
---------------------------------------------------------------------Standard
Fixed Effect
Coefficient
Error
T-ratio
P-value
---------------------------------------------------------------------For
INTERCEPT 1, B0
INTERCEPT2, G00
-2.857
0.542
-5.268
0.000
AGENCY DEPLOYS CED, G01
-0.083
0.725
-0.115
0.912
POPULATION DENSITY, G02
0.000
0.000
0.353
0.734
AVERAGE NUMBER OF, G03
0.000
0.000
0.428
0.681
OFFICERS PER 100,000 IN POPULATION
For POST_PERIOD slope, B1
INTERCEPT2, G10
0.329
AGENCY DEPLOYS CED, G11
-1.028
POPULATION DENSITY, G12
-0.000
AVERAGE NUMBER OF, G13
-0.000
OFFICERS PER 100,000 IN POPULATION

0.354
0.437
0.000
0.000

0.930
-2.350
-1.287
-0.922

0.384
0.051
0.239
0.388

For
SUSPECT WHITE slope, B2
INTERCEPT2, G20
-0.315
AGENCY DEPLOYS CED, G21
0.334
POPULATION DENSITY, G22
-0.000
AVERAGE NUMBER OF, G23
-0.000
OFFICERS PER 100,000 IN POPULATION

0.461
0.541
0.000
0.000

-0.684
0.617
-0.884
-1.525

0.516
0.557
0.406
0.171

For
SUSPECT MALE slope, B3
INTERCEPT2, G30
0.894
AGENCY DEPLOYS CED, G31
-0.245
POPULATION DENSITY, G32
0.000
AVERAGE NUMBER OF, G33
-0.000
OFFICERS PER 100,000 IN POPULATION

0.940
1.236
0.000
0.000

0.950
-0.199
0.151
-0.383

0.374
0.848
0.885
0.713

-1.780
1.272
-0.779
0.718

0.118
0.244
0.461
0.496

For SUSPECT AGE UNDER 25 YEARS OLD slope, B4
INTERCEPT2, G40
-0.697
0.391
AGENCY DEPLOYS CED, G41
0.581
0.457
POPULATION DENSITY, G42
-0.000
0.000
AVERAGE NUMBER OF, G43
0.000
0.000
OFFICERS PER 100,000 IN POPULATION

----------------------------------------------------------------------

PERF’s Quasi-Experimental Evaluation on Deployment of Less Lethal Weapons

Police Executive Research Forum
1120 Connecticut Avenue, NW, Suite 930
Washington, DC 20036
202-466-7820
202-466-7826 fax
www.PoliceForum.org