. . tin!)
. . 9:1...
. rJ.
. V u%! u:
. ‘ . o ‘
1'54".“ "1'11, 0: ‘
‘ ~. u .
"1*”..qu v
1. .‘I‘ a. 5?? ’-
13. if: L...

a... . at. 9..

r

‘Y 1: ‘

- _. i.
.néa

X.
2?...
3.1%.). ’1ﬁun

1b-
. .v. I}! Zr‘ Q. ’l ‘r-f‘lh
. . 5 in ’5:- I ﬂ .
#51! i3 aim» mﬁhuaﬁd . u.
h... .5 1‘ ‘. 1.... .1 ﬂy.
’1. n‘h I l“; vat. I
t a. ‘
I
s: I:

$541...»st
irilx‘
.. 3
a. . 1...‘ f‘.
.t 5‘.

.( A).
‘91:}.Vae
3 0’. 3
31.19;;

:3 Ev‘tiﬂlw
é. 5.1.3:.»«37.
. €5.35}:
.;
:3. .. .u.+

5..

Ir
3‘ ta «1.

Samoa:

 

 

 

3007

This is to certify that the
dissertation entitled

The Neighborhood Context of Police Use of Force Behavior

presented by

Cedrick G. Heraux

has been accepted towards fulﬁllment
of the requirements for the

PhD. degree in Criminal Justice

 

 

M

WMajor %fessor’s Signature

wj/wog

Date

MSU is an Afﬁrmative Action/Equal Opportunity Institution

 

I“ 1
LIBRARY
I Michigan State '
University
w...— "1

 

- - -‘—---‘---0-0—0-0-‘-0-0-v-l-O-l-O-I-I-l-O-‘-l--i.

.I-0-0--I-0-0-0-I-O-c-n-u-n-.--—~ -

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

DCT312

n
ﬂu?

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 p:/ClRC/DateDue.indd-p.1

 

THE NEIGHBORHOOD CONTEXT OF POLICE
USE OF FORCE BEHAVIOR
BY

Cedrick G. Heraux

A DISSERTATION
Submitted to
Michigan State University

in partial fulﬁllment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
School of Criminal Justice

2006

ABSTRACT

THE NEIGHBORHOOD CONTEXT OF POLICE
USE OF FORCE BEHAVIOR

By

Cedrick G. Heraux

Police use of force has been an important topic of research within the ﬁeld
of criminal justice, and studies over the past several decades have attempted to
explain this form of behavior. Social control theory and social disorganization
theory are put forth as the most appropriate theoretical framework within which to
analyze police use of force behavior. This research compares two different
conceptualizations of use of force in attempt to determine which individual-level
and neighborhood-level factors inﬂuence this behavior.

The focus of this research was to determine if interactions between
individual-level variables and neighborhood-level variables are signiﬁcant
predictors of use of force behavior by police ofﬁcers. More speciﬁcally, it is
argued that both the prevalence and severity of force are affected by interactions
between neighborhood levels of concentrated disadvantage and ofﬁcer race,
suspect race, and suspect demeanor. In addition, it is also argued that the
prevalence and severity of force are affected by interactions between
neighborhood levels of crime and ofﬁcer race, suspect race, and suspect
demeanor. The prevalence of force was deﬁned as either the absence or
presence of force, while the severity of force was deﬁned as the maximum

amount of force (as deﬁned by ofﬁcers) within an encounter.

Findings were consistent across both dependent variables. In both
instances, none of the interactions between individual-level and neighborhood-
level variables exerted a statistically signiﬁcant effect. However, evidence for the
statistically signiﬁcant effect of suspect behavior was strong across both models.
Speciﬁcally, if suspects were antagonistic toward ofﬁcers, or provided physical
resistance, both the prevalence and severity of force increased. Implications for

both theory and policy are discussed.

 

 

Copyright by
CEDRICK G. HERAUX
2006

 

This work is dedicated to my late paternal grandmother, Gertrude Heraux, who
always wanted a Dr. in the family, and to my parents, Robert and Helyett Heraux,
who always allowed me to follow my dreams. There are no words for how much

your love and support mean to me.

 

ACKNOWLEDGEMENTS

Any work of this magnitude owes its completion in large part to numerous
individuals whose encouragement and support are instrumental. I would ﬁrst like
to acknowledge all of my dissertation committee members, both past and
present. Dr. Peter Manning, Dr. Stephen Mastrofski, and the late Dr. Lawrence
Messe each provided me with insights which helped to shape my initial thoughts.
Dr. Bill Davidson, Dr. Christina DeJong, Dr. Edmund McGarrell, and Dr.
Christopher Maxwell all spent countless hours reading and re-reading, as well as
providing guidance. I am also indebted to Dr. Joel Garner, who allowed me to
use his data from the Police Use of Force Study. I am proud to have known all of
these ﬁne individuals.

I have also made countless friends during my time at Michigan State
University. Of those who have gone before me, I am especially thankful for the
friendship of Dr. Jeffrey Cancino (one of my ﬁrst ofﬁce-mates), Dr. Meghan
Stroshine, Dr. Sean Varano, Dr. Beth Huebner, Dr. Brandon Kooi, Dr. Justin
Patchin, and Dr. Sameer Hinduja. You have all been missed these past few
years, and I have many fond memories of our time together. Of those who
remain, I can only hope that l have been able to touch your lives, as both
someone to learn from (after all, who enjoys SPSS syntax as much as I do?),
and as someone with whom to have fun. Ryan Martz, Jason Ingram, Nicholas
Corsaro, Hao Lu, Karen Ream, and Tim Homberg have all provided countless

hours of support and socializing, keeping me sane throughout this entire process.

vi

 

My MSU “family” would not be complete without Amanda Burgess-Proctor (and
Bryan and Anna), whose emotional support has truly been unending. Seeing her
handle motherhood and her academic work was inspirational, and is directly
responsible for helping me produce the ﬁnished product that you see here.

I would like to thank my family for providing me with examples of how hard
work and love can be the two most powerful forcesin the world. My father,
Robert, and my mother, Helyett, have spent their entire lives caring for one
another and our family while continually working countless hours to allow me to
always follow my dreams. My sister, Caroline, has also provided me with an
example of academic success to strive for, while also demonstrating that love
and family are truly more important. Her children, Isabelle and Trevor, remain
one of the brightest lights of my life. To the rest of my family, particularly my
maternal grandmother Mireille, my late maternal grandfather Aime, and my late
paternal grandfather Georges, I love you all, and I hope I have made you proud.

Finally, I would like to thank my wife, Summer. I have been working on this
dissertation throughout our entire lives together, and your love and support has
been overwhelming. In addition to all of your hard work at yourjob, you have
worked hard at helping me with mine. The conclusion of my journey at Michigan
State University demonstrates your efforts as much as mine. I love you for being

my best friend, sounding board, and motivator as you made that journey with me.

vii

TABLE OF CONTENTS

LIST OF TABLES ................................................................................. xi
CHAPTER 1: INTRODUCTION ................................................................. 1
Why Study Neighborhoods? ............................................................................. 4
General Theoretical Dimensions of Neighborhoods ............................... 4
The Importance of Context .............................................................. 6
The Chicago School and Social Disorganization .............................. 7
Donald Black’s Theory of Social Control ....................................... 10
Theoretical Deﬁnitions of Neighborhoods and Criminal Justice
Research .......................................................................... 1 1
Neighborhoods as the Unit of Analysis ............................................. 17
A Brief Introduction to the Literature .................................................... 26
Summary ....................................................................................... 29
CHAPTER 2: LITERATURE REVIEW ...................................................... 30
Studies Using Neighborhoods as a Variable of Interest ............................ 30
Offending Behaviors by Adults and Juveniles .................................... 30
Police Behavior ........................................................................... 35
Use of Force Research ..................................................................... 47
Studies of Association Examining Use of Force Behavior ..................... 48
Multivariate Studies on Use of Force Behavior ................................... 60
Use of Force Research and Neighborhood Context ............................. 75
Limitations of Prior Studies ................................................................. 80
Variables of Interest from Prior Studies ................................................. 83
Summary ....................................................................................... 89
CHAPTER 3: DATA AND METHODS ...................................................... 90
The Police Use of Force (PUF) Study Data ........................................... 91
Data Collection Procedures ........................................................... 91
Sample Characteristics ................................................................ 97
The Operationalization of Neighborhood ....................................... 100
Speciﬁc Hypotheses ........................................................................ 104
Speciﬁcation of Variables of Interest ................................................... 106
Dependent Variables .................................................................. 106
Prevalence of Force ............................................................... 106
Severity of Force .................................................................... 108
Independent Variables ................................................................ 1 10
Neighborhood-Level Variables .................................................. 1 1 1
Encounter-Level Variables ....................................................... 1 12

viii

Suspect Characteristics ...................................................... 1 13

Ofﬁcer Characteristics ......................................................... 114

Nature of the Encounter and Offense ..................................... 115

Nature of the Encounter Location .......................................... 117

Analysis Procedures ....................................................................... 1 18
Spatial Autocorrelation ................................................................ 1 18
Hierarchical Linear Modeling ......................................................... 122
Summary ...................................................................................... 130
CHAPTER 4: RESULTS ...................................................................... 131
Results ........................................................................................ 138
Dependent Variable Analyses for Maximum Force ............................ 139
Signiﬁcant Findings from the Maximum Force Model ..................... 145
Dependent Variable Analyses for Physical Force plus Threats ............. 151
Signiﬁcant Findings from the Physical Force plus Threats Model ...... 156
Summary ...................................................................................... 164
CHAPTER 5: CONCLUSIONS .............................................................. 166
Discussion .................................................................................... 169
Re-Visiting the Theoretical Foundation for the Current Research ......... 169
Theoretical Foundations of Current Hypotheses ............................... 172
Impact of Variables Concerning Neighborhood-Level Effects ............... 173
Impact of Variables Concerning Interaction Effects ............................ 175
Impact of Individual-Level Variables ............................................... 180
Impact of Variables Concerning the Nature of the Location ............. 180

Impact of Variables Concerning the Nature of the Encounter ........... 181

Impact of Variables Concerning Police Mobilization ....................... 183

Impact of Variables Concerning Ofﬁcer Characteristics .................. 185

Impact of Variables Concerning Suspect Characteristics ................ 187
Interpretation of Results ............................................................... 189
Implications of the Current Research .................................................. 196
Theoretical Implications ............................................................... 196
Research Implications ................................................................. 199
Practical Implications .................................................................. 200
APPENDIX A .................................................................................... 202
APPENDIX B ..................................................................................... 204
APPENDIX C .................................................................................... 208
APPENDIX D .................................................................................... 256
APPENDIX E .................................................................................... 261

ix

APPENDIX F .................................................................................... 267
REFERENCES .................................................................................. 284

LIST OF TABLES

Table 1: Measures of Force, Methodologies and Sample Sizes ..................... 73
Table 2: Characteristics of Six Jurisdictions ............................................... 92
Table 3: The Force Continuum Across Jurisdictions .................................... 96
Table 4: Items within Prevalence of Force Measure ................................... 107
Table 5: Items within Maximum Force Measure ........................................ 109
Table 6: Sample Characteristics (Level-1 Independent Variables) ................. 133
Table 7: Sample Characteristics (Level-2 Independent Variables) ................. 138
Table 8: Full Model for Maximum Force ................................................... 142
Table 9: Full Model for Physical Force Plus Threats ................................... 153
Table 10: Summary of Results ............................................................... 190
Table A-1: Signiﬁcant Predictors of Use of Force Behavior Across Multivariate
Studies ............................................................................................. 203
Table B-1: Sample Characteristics for Variables used in Factor Analysis ....... 205
Table B-2: Correlation Matrix for Variables used in Factor Analysis ............... 206
Table B-3: Factor Loadings for Variables used in Factor Analysis ................. 207
Table C-1: 6-Site Correlation Matrix ........................................................ 209
Table D-1 : Descriptive Statistics (NUMERSO, NUMBERPO, OFF 1AGE,
SUSPAGE, REPEAT) ........................................................................................ 257
Table E-1: Maximum Force Model ......................................................... 262
Table E-2: Prevalence of Force Model .................................................... 264
Table F-1: Mean Differences in Arrest — No Arrest Census Tracts ................ 269
Table F-2: Logistic Regression Analysis .................................................. 270

xi

Table F-3: Negative Binomial Regression Analysis ................................... 272

Table F-4: Mean Differences in 1-4 Arrests - 5 or More Arrests in Census

Tracts .............................................................................................. 274
Table F-4: Poisson Regression Analysis for Prevalence of Force .................. 275
Table F-5: Negative Binomial Regression Analysis for Maximum Force ......... 277

xii

LIST OF FIGURES

Figure 1: Charlotte Census Tract Arrests ................................................. 279
Figure 2: Colorado Springs Census Tract Arrests ...................................... 280
Figure 3: Dallas Census Tract Arrests ..................................................... 281
Figure 4: St. Petersburg Census Tract Arrests .......................................... 282
Figure 5: San Diego Census Tract Arrests ............................................... 283

xiii

 

Chapter 1: Introduction

One of the most fundamental research problems in the criminal justice
literature has been the examination of police behavior. It is not only important to
understand what it is the police do while on patrol, but also why they do these
things. In efforts spanning the past thirty years, numerous researchers have
examined not only police behavior, but also the causes of that behavior, in order
to increase our range of knowledge in this area. Within this relatively abundant
literature there is general agreement concerning the actual actions of police
ofﬁcers (see Bayley 1994; Cordner 1979; and Manning 1997 for examples of
studies of how police ofﬁcers spend their time), but there is much less agreement
concerning the causes of these behaviors (see Riksheim and Cherrnak 1993 for
a comprehensive discussion of various explanatory factors, and how the latter
have been studied). This is perhaps understandable, given the variety of
methods employed and personal biases of the researchers towards particular
theoretical orientations, but it severely restricts a proper understanding of police
behavior, which in turn limits the effectiveness of policy recommendations
stemming from this research. In these times when the public seems to have
developed a mistrust of police ofﬁcers due to negative publicity associated with
police misconduct (Bell 2000; Cao et al. 1996; Tuch and Weitzer 1997), this is a
particularly troubling observation. If we do not fully comprehend the causes of
police behavior, it becomes extremely difﬁcult to address misbehavior’, thus

ensuring that the gap between the police and the public will remain intact. For

 

' Note, however, that while police misconduct is a very important topic of study, this particular research
will analyze all police use of force, regardless of the appropriateness of that behavior.

every stride made through community policing programs (Hickman et al. 2000;
Mastrofski et al. 1995), we move three steps back for every instance, whether
widely publicized or directly experienced by a citizen, of racial proﬁling, general
harassment or the use of force. While the former two instances of police behavior
are certainly a point of concern, it is the latter which has the greatest potential for
harm, both in a physical sense to the citizen as well as in terms of damage to
police-community relations. Accordingly, this dissertation will use data collected
in six jurisdictions located in large American cities during ofﬁcer and suspect
surveys conducted subsequent to an adult arrest within these jurisdictions.2
Using ideas from Donald Black’s theory of social control, as well as from social
disorganization theory, this research will examine all aspects of the police use of
force during arrest encounters with adult citizens.

Within the theories mentioned above, this research focuses on a very
speciﬁc issue regarding the use of force. Previous research has provided
measures of: (1) how often police engage in the application of force within a
speciﬁc encounter (Langan et al. 2001; Walker and Graham 1998;); (2) the
prevalence of each method of force applied within a speciﬁc encounter3
(Crawford and Burns 1998; Geis and Binder 1990); and (3) the maximum level of
force applied within a speciﬁc encounter (Garner et al. 1995). More importantly,

while most research has focused on only one of these three aspects of the use of

 

2 The methodology of the study and the data itself will be discussed in greater detail in Chapter 3 of this
work.

3 That is, the use of strictly verbal commands is vastly different than the ﬁring of a weapon, and some
research has differentiated between the various methods of force employed by officers. Note that this has
often been difﬁcult due to the fact that some methods of the use of force are utilized in a statistically rare
manner.

force, these studies have also only focused on one class of explanatory factors
(i.e. psychological, sociological, or organizational). In an attempt to be
comprehensive, this research will rely on a combination of psychological (largely
individual), sociological (largely situational), and organizational factors used to
explain the use of force. However, rather than merely relying on what previous
research has held regarding these explanations, this research attempts to
integrate these perspectives within a neighborhood context framework. The
approach used here allows one to explain variation in police use of force by
demonstrating the importance of neighborhood context as a determinant of that
behavior. Focusing solely on organizational constraints (Alpert and MacDonald
2001; Engel 2000), or on the individual characteristics of ofﬁcers and/or citizens
(Scrivner 1994), or on the speciﬁc situations at hand (i.e. the details of particular
encounters) (Sorensen et al. 1993), previous research has largely ignored the
social-psychological effect of neighborhood contextual cues in shaping how
police ofﬁcers perform in the ﬁeld. By focusing on this aspect, therefore, this
research will provide a comprehensive view of police use of force behavior,
allowing us to develop a fuller understanding of that behavior. Using multilevel
modeling techniques, this dissertation will estimate the effects of neighborhood
context on police use of force, above and beyond the traditional explanatory
factors located within psychological, sociological and organizational frameworks.
Accordingly, the remainder of this chapter provides a discussion of why

neighborhood context is important in the study of police behavior.

Why Study Neighborhoods?

Some of the recent literature in the ﬁeld of criminal justice has focused on
ecological factors which are theorized to affect police and/or offender behavior. In
general terms, this body of research has examined how neighborhood-level
variables interact to inﬂuence these behaviors in a variety of contexts. It is
important to note, however, that when engaging in such research, one must take
the preliminary step of deﬁning “neighborhood” in order to maintain validity. In
theoretical terms, researchers must properly operationalize such an important
variable in order to ensure that the results are both valid and generalizable, while
still adhering to a parsimonious model. The following discussion will be based on
a theoretical description of neighborhoods, then, rather than a methodological

one.‘

General Theoretical Dimensions of Neighborhoods:

When studying the ecological basis of police behavior, the neighborhood
can be examined from several different perspectives; those looking in may have
a different view than those looking out. However, in all of these contexts, there
remain a number of agreed-upon deﬁning characteristics. First, there is typically
a sense of geographic concentration. Bursik and Grasmick (1993: 6) note that,
“most basically, a neighborhood is a small physical area embedded within a
larger area in which people inhabit dwellings.” There are two consequences of

such a geographically-based deﬁnition: (a) business districts, while often

 

4 Note that while this discussion focuses on theoretical characteristics which deﬁne neighborhoods, Chapter
3 will focus more closely on the methodology of how neighborhoods were deﬁned for the current work,
with an emphasis on how physical boundaries of neighborhoods are determined.

technically deﬁned by geographic restrictions, do not exhibit other characteristics
of true neighborhoods; and (b) neighborhoods differ from a sense of community.
With regards to the latter concept, it is important to note that while members of
ethnic and racial groups may make claims to an overall sense of community,
Hispanics in a neighborhood in New York differ in small, but important, ways from
Hispanics in a neighborhood in Miami. More importantly for the current research,
it is also likely that Hispanics residing in the Greenwich Village neighborhood of
New York differ from Hispanics residing in the Spanish Harlem neighborhood of
New York.5 Correlated with this geographic concentration is the well-noted
phenomenon of increasing segregation of minority populations within urban
centers. While a neighborhood may technically be rural in nature, it is the growth
of urban ghettoes which has concerned sociologists and criminologists for the
past few decades, and it thus upon these that this research shall concentrate.
Indeed, Jackson (1989: 62) notes that “region, as a sociohistorical construct
inﬂuencing the relationships between racial and ethnic groups, is a ﬁlter through
which. . .groups are viewed.”

A second characteristic of neighborhoods is the “collective life that
emerges from the social networks that have arisen among the residents and the
sets of institutional arrangements that overlap these networks” (Bursik and
Grasmick 1993: 6). That is, there is a shared sense of values amongst the

residents of the neighborhood. In order for coherence to truly emerge, the

 

5 Throughout this research some common terminology will be used to maintain consistency regarding racial
and ethnic groups. In particular, the term “Hispanic” will be used in place of “Latino” or “Chicano”, and
the term “Black” will be used in place of “African-American.” These terms are used for simplicity, and do
not reflect any disregard for the nominal preferences of the groups mentioned.

individuals in the collective must subscribe to similar beliefs. As Blau (1964: 255)
states, “normative standards that restrict the range of permissible conduct are
essential for social life.” This arises out of culture, which Black (1976: 61) notes
“includes conceptions of what ought to be, what is right and wrong, proper and
improper...”

The ﬁnal characteristic used to deﬁne neighborhoods in a general sense is
that of a shared experience. Burisk and Grasmick (1993: 6) note that “the
neighborhood is inhabited by people who perceive themselves to have a
common interest in that area and to whom a common life is available...[That is,]
the neighborhood has some tradition of identity and continuity over time.”
Individuals who have similar occupational distinctions (i.e. white-collar vs. blue-
collar) and salary levels, and thus are located within the same social class, are
more likely to reside in the same, or similar, neighborhoods, and thus have
similar experiences. In addition, individuals who display the same levels of family
stability and mobility (i.e. geographic consistency) will also have similar
experiences due to these factors. Thus, while shared beliefs are a reﬂection of
internal consistency, shared experiences allow individuals to form a bond within

their neighborhood through external consistency.

The Importance of Context:

The discussion of how neighborhoods can be deﬁned theoretically has
provided a starting point for understanding research regarding the ecological

basis of behavior. Therefore, it now becomes necessary to examine how

ecological research ﬁrst emerged within the ﬁelds of criminology and sociology.
By understanding the origins of such research, we are better able to structure the
current research to focus on the importance of ecology on police use of force

behavior.

The Chicago School and Social Disorganization

An ecological understanding of human behavior has its origins in the work
of Park and Burgess (1924), who developed the concentric zone theory of cities
in order to explain crime causation. The authors noted that cities provided the
perfect natural laboratory setting in which to study the process of invasion,
dominance and succession of various areas within those cities. Describing cities
as consisting of: (1) the central business district; (2) a transition zone; (3) the
workingman’s district; (4) the residential district; and (5) a commuter zone, Park
and Burgess (1924) argued that each of the ﬁve zones had a separate structure
and organization, with distinct populations and characteristics. Most importantly,
in terms of a contextual explanation, the authors found that the transition zone
experienced the greatest amount of delinquency, and that this was due to the
presence of social disorganization, the latter in turn being due to contextual
characteristics, including: (1) population makeup; (2) economic factors; and (3)
housing factors.

Shaw and McKay (1942) chose to build on the work of Park and Burgess,
arguing that the crime rate was distributed throughout the city, with delinquency

being most prevalent in the transition zones nearest the business district. As with

previous research, Shaw and McKay found that these areas were characterized
by populations that were largely: (1) immigrant; (2) minority; (3) lower class; and
(4) accepting of unconventional norms. Using over 50,000 juvenile court records
collected between 1900 and 1933, the authors determined that delinquency was
an ecological, rather than an individual, phenomenon, as it was the normal
response of normal individuals to abnormal social conditions (Shaw and McKay
1942). The authors noted that transition zones, with high rates of residential
mobility and racial heterogeneity, had enormous difﬁculty avoiding becoming
socially disorganized.‘5 Most importantly, in a theory of cultural transmission,
Shaw and McKay argued that traditions of delinquency in transition zones are
passed down to successive generations, regardless of changing racial
composition and the rate of mobility. Thus, in essence, the authors argued that
delinquency is more zone-speciﬁc than population-speciﬁc, with the development
and acceptance of criminal values being a self-perpetuating phenomenon.

After the work of Park and Burgess, and Shaw and McKay, the next
advancement in social disorganization theory came from Stark (1987), who
developed a series of thirty propositions regarding crime and deviance to outline
a theory of deviant places. Many of the latter propositions reﬂected speciﬁc
aspects of previous research on social disorganization, and Stark (1987: 893)
went so far as to argue that “high rates of crime and deviance can persist in
speciﬁc neighborhoods despite repeated, complete turnovers in the composition

of their populations. . . [and this] suggests that more than ‘kinds of people’

 

6 Bursik (1988: 521), following the work of Shaw and McKay (1942), deﬁned social disorganization as
“the inability of local communities to realize the common values of their residents or solve commonly
experienced problems.”

explanations are needed to account for the ecological concentration of
deviance...” Stark went on to note that, just as Shaw and McKay had found,
areas exhibiting high levels of deviance were characterized by: (1) high
population density; (2) high levels of poverty; (3) mixed use; (4) transience; and
(5) high degrees of dilapidation. Even more importantly, in the context of the
current research, Stark (1987: 902) stated, in Proposition 25, that “stigmatized
neighborhoods will suffer from more lenient law enforcement.”

The ﬁnal major research piece regarding social disorganization to be
explored here is that of Sampson and Groves (1989), who attempted to directly
test the theory as set forth by Shaw and McKay. The authors noted that the
general hypothesis was that poverty, racial and ethic heterogeneity, residential
mobility, and family instability all lead to community social disorganization, which
then leads to crime and delinquency. Using two separate surveys conducted in
1982 and 1984 throughout over 225 areas with approximately 11,000 residents
of Great Britain, Sampson and Groves examined the link between community
characteristics and criminal behavior. Most importantly, Sampson and Groves
found that in addition to the factors described by Shaw and McKay, other
external factors which affect social disorganization included: (1) community
supervision of teenage gangs; (2) informal friendship networks; and (3)
participation in formal organizations. The authors found that between-community
variations in these factors, and, by extension, social disorganization, were
responsible for much of the effect on rates of both criminal victimization and

criminal offending. Sampson and Groves (1989: 775) conclude, then, that

“previous macro-level research in crime and delinquency has relied primarily on
census data that rarely provide measures for the variables hypothesized to

mediate the relationship between community structure and crime.”

Donald Black’s Theory of Social Control

In addition to the work of the Chicago School, Black (1976) also provided
a foundation for ecological research through the promotion of a sociologically-
based model of the behavior of social control. In arguing for his theory of social
control, Black (1976: 107) noted that “law is stronger where other social control is
weaker”, and deﬁned the latter as “informal social control in the form of inﬂuence
exerted by parents, peers, religious and community leaders, and various
organizations.” Although not explicitly discussed as such, the lack of such
informal social control can be considered a form of social disorganization as
deﬁned previously. Thus, Black’s (1976) model hypothesized that formal social
control (i.e. police behavior) would be more prevalent in areas with low informal
social control (i.e. those that experience social disorganization). Truly, Black’s
(1976) theory sought to be the ﬁrst systematic examination of the application of
social control, allowing one to predict police behavior based on the dynamics,
including the contextual cues, of an encounter. The argument presented was that
factors such as relational distance and social stratiﬁcation inﬂuenced how an
ofﬁcer responded to situations. Bayley and Mendelsohn (1969), agreed, also
arguing that increased social distance between police ofﬁcers and the poor

results in more aggressive or punitive police practices in lower-class areas, as

10

well as in high-crime areas, due to the effects of contextual characteristics.
Indeed, Black (1980: 7-8) notes that “as the encounter proceeds [the ofﬁcer] may
learn a great deal more about the location and direction of the incident in social
space [as] patrol work is different for middle-class people in the suburbs than for
lower-class people in the inner city.” Black (1980: 11) continues, arguing that
“dispute settlement by the police varies with its location and direction in social
space, including its relation to race, social class, [and] the social structure...”
Again, in light of the current research, it is important to note that Black (1976)
argued that the police will exert more coercion against those individuals who
have a lower status, such as lower-class individuals, minorities and juveniles.7
As we have seen, the work of the Chicago School and later researchers,
as well as that of Donald Black, provided a ﬁrm foundation for development of the
ecological perspective. It is also instructive, however, to examine how more
generalized criminal justice research has dealt with the theoretical deﬁnitions of

neighborhoods as described in the previous section.

Theoretical Deﬁnitions of Neighborhoods and Criminal Justice Research

As we have provided a theoretical deﬁnition of various aspects of
neighborhoods, as well as a discussion of how social disorganization and social
control theory have addressed neighborhood I ecological factors, it is now
necessary to discuss the burgeoning role of neighborhoods in a general sense

throughout the criminal justice and sociology literature. As noted previously, while

 

7 Black (1976) also noted that this category of lower-status individuals included those who were impaired
(either mentally, or due to drugs or alcohol), and those who were disrespectful of an ofﬁcer’s authority.

11

neighborhoods obviously can be located in rural areas, this research will
concentrate on their development within large cities in America. The population
growth in such areas has been almost exponential in nature, resulting in intense
geographic concentration (Walker et al. 1996; Wilson 1993). This, in turn, has
had a deleterious effect on the individuals residing in these areas, increasing
virtually every social-structural problem already present in the inner city. Yet,
these areas remain a neighborhood in every sense of the word, as they exhibit
the three theoretical characteristics previously described.

With respect to the ﬁrst theoretical characteristic of neighborhoods,
geographic concentration, a number of researchers have taken to using the term
“underclass” to describe low-income individuals concentrated in the inner city
(see Mann 1993 and Wilson 1993, for examples). Wilson (1993: 14) goes further,
noting that the proportion of the poor who reside in the ghetto varies signiﬁcantly
by race, as “almost a third of all metropolitan blacks lived in a ghetto in 1980.” As
middle-class minority group members leave the city in increasing numbers, these
areas have become overwhelmingly populated with those who cannot afford to
leave. Goldberg (1993: 188) identiﬁes the postmodern city as a product of the
processes of urban renewal and gentriﬁcation. While urban renewal revitalizes
the “salvageable” areas of the city, lower-income individuals are forced into
smaller and smaller “pockets” of ghetto housing, resulting in a concentration of
poverty in speciﬁc neighborhoods and ensuring the continued presence of the
ghetto in American cities. The latter neighborhoods, therefore, are a direct result

of the former phenomenon. Goldberg (1993: 191) continues, arguing that “the

12

 

 

racial slum is doubly determined, for [it] bears the added connotations of moral
degeneracy, natural inferiority and repulsiveness...[and while] the slum locates
the lower class, the racial slum [locates] the underclass.” Kelling and Stewart
(1990: 469) agree, arguing that “the communities of the underclass are plagued
by massive joblessness, ﬂagrant and open lawlessness, and low-achieving
schools, and...the residents of these areas, whether women and children...or
aggressive street criminals, have increasingly been socially isolated from
mainstream patterns of behavior.” Wade (1993: 52) also notes this phenomenon
of perpetuating isolation, noting that “cultural geography is not a natural cultural
construction, but derives from dominant ideologies. . . propagated by the most
powerful...” While the latter was used to describe the various regions of
Colombia, this description could easily be applied to America’s urban
environments in which lower class individuals ﬁnd themselves increasingly
concentrated. Indeed, Bell (1992: 4) has argued that “what we now call the “inner
city’ is, in fact, the American equivalent of the South African homelands.” One
important consequence of this segregation is noted by Rossi (1968: 105), who
argues that many members of these communities view the police as ‘occupation
forces’ in place to exert the inﬂuence of the power elite, and enforce adherence
to their normative structure. This is echoed by Jackson (1989: 2), who argues
that “police-community relations are fraught with problems of authority...” causing
resentment by members of these communities. Walker et al. (1996: 89-90)
concur, noting that “low-income people, regardless of race, are...far more likely

to see or have contact with the police [as] police departments routinely assign

13

more patrol ofﬁcers to minority and low-income neighborhoods.” This is further
supported by Cox (1984: 174), who notes that “[police] administrators allocate a
substantial amount of resources to areas inhabited by [lower-class] group
members [while] at the same time efforts such as community policing...are rarely
implemented in those areas most in need of such efforts.” We can see, then, that
the phenomenon of geographical concentration of certain populations is one
experienced throughout the world, as well as throughout the United States. It is
also clear that this phenomenon presents a situation which is vulnerable to police
abuses and misinterpretations of legitimate police behavior. The population that
remains in these “pockets” is thus localized in an area that is, in the words of
Mann (1993: 87), "likely to constitute meaningful frames of reference for social
comparisons.”

The second characteristic of shared values and beliefs is likely to be the
most common conception of neighborhood norms. All societies are composed of
individuals that occupy various positions which are deﬁned by the normative
structure (eg. criminal, police ofﬁcer, “deadbeat dad”). Indeed, it is only though
this normative structure that a collective consciousness can be manifested in
action (Chambliss and Seidman 1971: 7). As Blau (1964: 60) states, “group
cohesion promotes the development of consensus on normative standards and
the effective enforcement of these shared norms. While it is true, as
McNamara (1967: 163) has claimed, that urbanization is associated with
heterogeneity of normative structures, within particular neighborhoods these

values remain relatively homogeneous. Wilson (1987: 14) notes that “values

14

emerge from speciﬁc social circumstances and life chances, and reﬂect one’s
class and racial position.” Thus, certain values adhered to by those residing in
low-income minority neighborhoods may not be a part of the structure within low-
income white neighborhoods, which, in turn, may differ signiﬁcantly from the
belief system of upper-income white neighborhoods. It has been noted that the
black community is “characterized by personal and social disorganization...[and]
black ghettoes typically have high crime rates that are disproportionate to white,
lower-socioeconomic areas” (Reasons and Kuykendall 1972: 227). Reiman
(2001: 160) argues that this perpetuates “an ideological message that: (1) the
threat to ‘law-abiding Middle America’ comes from below them on the economic
ladder, not above them; [and] (2) the poor are morally defective, and thus their
poverty is their own fault...” However, in general, lower class urban communities
operate within norms and systems of social control that reﬂect a life-style
accommodating both conventional and illegal behavior (Reiss 1986: 12). Some
researchers have gone further, positing that a cultural tolerance of violence exists
within these areas, although this has been disconﬁrrned in other work. Sampson
and Bartusch (1998), for example, found that while these areas actually exhibited
lower levels of tolerance, they also experienced greater cynicism regarding police
services, indicating a sense of frustration within their social circumstances. As
Blau (1964: 231) notes, it is only when this frustration, even opposition, to the
power of the criminal justice system is experienced throughout the collective that
social values legitimating opposition to that dominance will emerge. Just as

values serve to legitimate the social order and the various arrangements that

15

sustain it, so too can they become internalized to validate the principles that are
in the best interests of the collective (i.e. opposition to the perceived differential
nature of the application of force by police ofﬁcers).

The third deﬁning theoretical characteristic of a neighborhood is a sense
of shared experiences. Wilson (1980: x) argues that “it is difﬁcult to speak of a
uniform experience when the...p0pulation can be meaningfully stratiﬁed into
groups whose members range from those who are afﬂuent to those who are
impoverished.” However, it is equally true that within particular neighborhoods,
be they low-income or high-income (or somewhere in between), the majority of
individuals share the same daily experiences. In fact, the negative aspects of
low-income neighborhoods may make these shared experiences even more
salient, as individuals faced with such ovenrvhelming negativity strive to form
signiﬁcant social bonds with one another to provide support. These areas face
enduring hardships in the form of weak labor force attachment and
unemployment, residential isolation, and poverty (McLanahan and Garﬁnkel
1993). In order to deal with this situation, ghetto residents (the so-called
underclass) must develop a normative structure which allows for positive
contacts within the neighborhood. Meares and Kahan (1998: 810-811) support
this, noting that “although the characteristics of individuals may have a direct
effect [on behavior], the importance of the characteristics of people residing in a
neighborhood lies largely in the implications this has for the social organization of
a community.” Yet, while individuals face enormous external structural pressures,

they are also exposed to pressures within the community which act against the

16

development of a normative structure. As McGahey (1986: 247) notes, these
areas are disproportionately populated with unstable households in which family
members are unable to exercise authority and control over younger generations.
Black (1976: 135) mentions the same phenomenon, arguing that traditional
family ties have been loosened to the point of falling apart, weakening the
neighborhood and making intimacy situational, rather than communal. This leads
to the development of norms which accommodate illegal behavior, as discussed
earlier. However, while these shared experiences within these areas tend to be

negative, they are still shared, and thus provide a sense of neighborhood.

Neighborhoods as the Unit of Ana_lvsis:

The previous discussion of the theoretical elements of neighborhoods has
been a precursor to the fundamental question of this introductory chapter: why
study neighborhoods (as opposed to individuals) at all? It is thus time to explore
the consequences of the context of neighborhoods as it relates to the
relationships between individuals. As Chambliss and Seidman (1971: 500) note,
the events that shape the law (or its application in the form of the use of force)
are an outgrowth of the relationship between the legal order and the social
setting. The latter, in turn, provides the context for individual relationships, be
they between citizens or between citizens and law enforcement ofﬁcers. The
impact of this contextual inﬂuence can be considerable, as research indicates
that the ecology of criminal justice decision making has a direct (typically

negative) effect on outcomes, particularly for minority group members (McNeely

17

and Pope 1981: 21). Support for this idea is indicated by Black (1998: 35), who
notes that individuals may be held collectively liable due to their neighborhood,
social class, race, or ethnicity. Thus, law-abiding citizens exhibiting the “wrong”
characteristics can become tainted through a process of ecological
contamination (Werthman and Piliavin 1967: 79). This was observed by Bittner
(1970: 10) as well, who stated that “[this] inevitably entails the consequences that
some persons will receive the dubious beneﬁt of extensive police scrutiny merely
on account of their membership in those social groupings which...social
comparisons locate at the bottom of the heap.” Van Maanen (1973) also notes
this phenomenon, arguing that police behavior is often based on the generally
held (among ofﬁcers) notion that few of the citizens encountered on the street in
such neighborhoods are worthy of respect, and that many of the latter are
undoubtedly guilty of some crime. Indeed, it has been noted that “from the front
seat of a moving patrol car, street life in a typical lower class neighborhood is
perceived as an uninterrupted sequence of suspicious scenes” (Werthman and
Piliavin 1967: 56). As Jacobs and Helms (1997: 1366) note, “one result of these
uncertainties is that conventional street crime is difﬁcult to control in low-income
areas because criminals and the innocent share many characteristics.”8

Kohfeld and Sprague (1990) were some of the ﬁrst criminologists to
articulate this idea more speciﬁcally, arguing that the responses of both police
ofﬁcers and criminals are threshold-triggered behaviors, with the police

responding to certain types of crimes only after the activity had reached an

 

3 The corresponding corollary here is, of course, that police ofﬁcers should be more effective at
determining criminality in affluent neighborhoods due to the visible differences between residents of those
neighborhoods and lower-class criminals.

18

imaginary line of ‘incivility’ visible only to the ofﬁcers themselves.9 This threshold
concept is supported by other researchers, who note that “most illegality is
tolerated” (Black 1989: 77) and that “policemen often do not arrest persons who
have committed minor offences in circumstances in which the arrest is technically
possible” (Bittner 1967: 702). Klinger (1997), in agreement with the latter
research, also argues for the idea of a threshold for police response, noting that
ofﬁcers” perceptions of a high-crime area result in fewer arrests for minor illegal
activity, as the victims are often considered to be potentially guilty of some other
crime. In addition, Klinger (1997) notes that due to the high volume of crime in
such areas, the ofﬁcers assigned to these neighborhoods are more selective in
their responses to crime due to time and manpower constraints. lmportantly, this
has the undesirable effect of increasing the average seriousness of ofﬁcer-
initiated encounters, thus increasing the potential for conﬂict and violence within
such encounters. For police ofﬁcers, the neighborhood characteristics that are
most important for triggering this threshold appear to be ethnicity and
socioeconomic status (Dunham and Alpert 1988: 521). Smith et al. (1984: 243)
support this, ﬁnding that “socioeconomic status rather than suspect race is the
axis around which [police behavior] revolves”, and they go on to note that police
ofﬁcers are more likely to be punitive toward offenders encountered in lower

status neighborhoods. This is particularly signiﬁcant due to the fact that ofﬁcers

 

9 Obviously, the idea of a threshold-triggered response from police ofﬁcers contains the caveat that this
only applies to less serious (typically property) crimes. No researcher has ever seriously argued that the
police are willing to accept a certain amount of rape, aggravated assault or murder before engaging in
enforcement efforts.

19

act as “gatekeepers” for the criminal justice system, in large part determining who
is subject to various forms of social control.

There is an inherently dangerous fallacy associated with the process of
ecological contamination - the criminal justice system comes to see negative
members of certain neighborhoods as representative of the overall neighborhood
when this is clearly not the case. Swett (1972: 38), for example, notes that “the
propensity for police suspicion to increase according to ethnocentric perception
of cultural differences is reinforced by stereotypes” regarding lower-class
tolerance of criminal behavior. Yet, Sampson and Bartusch (1998: 784) found
that an individual “can be highly intolerant of crime, but live in a disadvantaged
context bereft of legal sanctions and perceived justice.” Thus, while contextual
effects lead actors within the criminal justice system to treat all neighborhood
residents as homogeneous based on a few negative contacts, the reality is that
the majority of these residents are intolerant of the negative members. However,
because the contextual effect, and its corresponding ecological fallacy, is so
strong, these individuals remain as targets, albeit improper ones, for the criminal
justice system. This leads to cynicism regarding the system, causing these mis-
targeted individuals to rely on informal, rather than formal, social control when
their need arises. As Dunham and Alpert (1988: 506) note, “police strategies and
practices incongruent with the basic culture and values of [a neighborhood]
would likely be ineffective and perhaps even counterproductive to maintaining
order and controlling crime.” The injurious effects of the context of communities,

then, result in a breakdown in the consistency of the system of social control.

20

Clearly, then, it has been demonstrated that neighborhood context, in and
of itself, is important for a variety of reasons. It is therefore instructive to examine
why neighborhoods are the most appropriate unit of analysis for developing an
understanding of police behavior. That is, what does a neighborhood-level
analysis provide that an individual-level or county-level analysis does not?10

To begin with, numerous researchers have found that introducing
neighborhood characteristics into analyses can reveal the importance of linking
micro- and macro-social processes together. For example, Sampson and
Woolredge (1987) found that, controlling for individual-level effects, burglary
victimization was directly related to community-based measures such as
unemployment, housing density and residential stability.11 They conclude that
“[ilmportant individual-level differences in lifestyle notwithstanding, the
community context of everyday activities is also a crucial theoretical factor in
explaining victimization risk” (Sampson and Woolredge 1987: 372). These results
were bolstered by Smith and Jarjoura (1989), who noted that burglary
victimization varied with characteristics of individual households at the individual
level, while victimization rates varied with characteristics of social areas at the
aggregate level. The authors go on to state that “[d]ata on 9,006 households in
57 residential neighborhoods...indicate that a more complete understanding of

factors inﬂuencing victimization risk emerges when both household and

 

'° Note that while this section discusses neighborhoods as the appropriate unit of analysis from a theoretical
standpoint based on prior literature, Chapter 3 will provide further discussion concerning the
Operationalization of neighborhoods as the unit of analysis, as well as information on how neighborhoods
were chosen.

” Note that although the authors were not explicitly testing social disorganization theory, many of their
community constructs can be found in the literature regarding that theory as it has been discussed
previously in this chapter.

21

neighborhood characteristics are included as independent variables” (Smith and
Jarjoura 1989: 621).

With respect to speciﬁc police behaviors, researchers have also noted the
importance of using neighborhood context in conjunction with individual-level
variables. Lizotte et al. (1993: 1) note that “[s]ocial scientists see communities as
being more than the sum of the individuals that comprise them [as the]
community provides the context in which Individuals organize their social lives,
thus helping to pattern their behavior.” It should be clear that this phenomenon
extends to police ofﬁcers who work in these communities, thus inﬂuencing their
behavior as well. Bittner (1967: 699) was one of the ﬁrst researchers to explore
this complex issue, arguing that “patrolmen have a particular conception of the
social order of skid-row life that determines the procedures of control they
employ.” This clearly illustrates the idea that police ofﬁcers are thought to behave
differently when interacting with marginal populations (i.e. minorities and the
poor), reserving one set of police tactics for the well-to—do and another for those
less well-off. Meehan and Ponder (2002: 402) conﬁrm this, arguing that “[s]ocial
psychological studies provide evidence that the police...apply a ‘cognitive
schema’ that views the ambiguous behaviors [of residents] as suspicious and
potentially criminal.” Cox and Frank (1992) also ﬁnd neighborhood effects, noting
that over one-quarter of all ofﬁcers studied made changes in their policing style
as neighborhood context changed.

The need for the inclusion of neighborhood context variables, in

conjunction with the typically used individual-level variables, has been described

22

above. The ﬁnal discussion centers on the decision to use neighborhoods as a
unit of analysis rather than a larger unit of aggregation (i.e. cities or counties).
One of the most prominent examples of the impact of the aggregation decision is
found in the work of Ouimet (2000), who studied offending rates and social
disorganization theory. Comparing data gathered at two (theoretically) different
levels of aggregation (495 census tracts vs. 84 neighborhoods), Ouimet (2000)
found that analyses at the neighborhood level provided stronger coefﬁcients and
increased predictive power. This is supported by Peterson et al. (2000: 38), who
note that “census tracts do not necessarily correspond to neighborhoods in a
socially meaningful sense.”12 More importantly, however, Ouimet (2000) went on
to note that aggregating to a level higher than neighborhoods would result in
problems similar to those found in using a level of aggregation that was too
small.” As Peterson et al. (2000: 33) note, “[t]he general theoretical rationale for
exploring the institutional context of neighborhood...stems from social
disorganization theory...which has its foundation in broad social conditions...[and
a] local institutional base.” The latter, clearly, are an element of neighborhoods,
rather than some larger level of aggregation, by their very deﬁnition. This is
supported by Messner and Tardiff (1986: 297) who argue that “neighborhoods
are more appropriate units of analysis...than are larger political and statistical
units because neighborhoods are more likely to constitute meaningful frames of

reference for social comparisons.”

 

'2 Discussion of why this was inappropriate, given that census tracts are the most valid and reliable measure
of neighborhoods, will be presented in Chapter 3.

'3 The same problems stem from using a proxy for neighborhoods, such as in the work of Mastrofski et al.
(2002), who use police beats as an Operationalization for neighborhoods in two different cities. These issues
will be discussed further in Chapter 3.

23

Given the previous discussion, it can be argued that neighborhoods are
large enough to have an impact on, yet small enough to provide a context for,
both citizen and ofﬁcer behavior. Coulton et al. (2001) conﬁrm this, noting that
residents experienced a reasonable degree of consensus when asked to deﬁne
the boundaries of their own neighborhood, indicating that the latter were capable
of exerting an inﬂuence on their behavior.“ However, it is unlikely that such
consensus would be obtained with regards to the deﬁnition of county or city
boundaries, as most individuals ﬁnd it necessary to resort to a map or physical
markers to determine the latter. Kelling and Stewart (1990: 460) ﬁnd the same
phenomenon, stating that “residents...construet ‘cognitive maps’ in which they
allocate distinctive places as ‘theirs’ — their neighborhood.” Taylor et al. (1984:
303) argue that these cognitive maps are important due to the fact that “block-
level linkages between social ties and territorial attitudes clarify how territorial
attitudes reﬂect, and may contribute to, the development of group-based norms
regarding appropriate behaviors in on-block settings.” This extends to police
ofﬁcers as well, as Ratcliffe and McCullagh (2001: 333) note that “[t]he
geographical nature of policing [in the form of] individual beats, means that an
ofﬁcer has to become intimately familiar with their patrol area. The authors go
on to note that this familiarity inﬂuences ofﬁcer behavior by affecting their
perception of the safety of various parts of their beat, and thus the tactics (i.e. the

appropriate behaviors) they use within those areas. McGarrell et al. (1997: 489)

 

‘4 It is interesting to note that residents of urban neighborhoods deﬁned areas that were geographically
much smaller for their neighborhoods than did residents of suburban neighborhoods for their
neighborhoods, although both groups gave relatively detailed descriptions of boundaries (Haney and
Knowles 1978).

24

support this as well, noting that “the neighborhoods categorized [in the study] as
high-disorder neighborhoods are those identiﬁed by the police and in local lore as
the centers of crime and disorder.” Walklate and Evans (1999: 21) concur, ﬁnding
that there was consensus between supervisors, patrol ofﬁcers, and community
members when asked to describe the worst crime problems and the worst crime
areas. While a speciﬁc city or county can certainly be construed as dangerous,
the inﬂuence of a dangerous context on an ofﬁcer’s behavior extends only to the
immediate surroundings, for it is only in the present location at the present time
that an ofﬁcer is concerned for their safety. The latter study also makes clear that
citizen behavior may be inﬂuenced by perceptions of neighborhood activities.
Indeed, several researchers (Gould and White 1974; Perkins et al. 1992; and
Suttles 1972) have found that the construction of cognitive maps can be
inﬂuenced by an individual’s perception of safety or danger within a speciﬁc area.
It has been demonstrated that neighborhood context can have a
signiﬁcant, and often negative, impact on the relationships between individuals in
lower income areas. Representatives of the criminal justice system, as wielders
of social control and the ability to use force, too often respond to the fallacy of
ecological contamination. It is thus important both to deﬁne neighborhoods
appropriately, and to examine the theoretical basis behind the context of
neighborhoods. In doing so, researchers can shed light on the role that race and
social class, at a neighborhood level, play in determining the subjects of social
control, and, more speciﬁcally, the use of force. This, in turn, will allow us to

make changes on a structural level to the determinants of the police use of force.

25

A Brief Introduction to the Literature

While Chapter 2 will present a more detailed review of the prior literature
regarding research on both neighborhoods and police behavior, it is instructive to
ﬁrst introduce some concepts which have shaped that literature. Many prior
studies on the inﬂuence of geography have focused on a larger area or region,
rather than a neighborhood. These studies identiﬁed a subculture of violence as
regions in which attitudes toward using force to resolve problems were positive
(Hawley and Messner, 1989; Simpson, 1985; Wolfgang, 1978). In contrast, a
series of studies emphasized the importance of structural position within those
regions as a primary factor inﬂuencing violent behavior (Black, 1976; Cao et al.,
1998; Luckenbill and Doyle, 1989). Subsequent research went further, arguing
that individuals with a lower structural position in such regions were more likely to
experience disputatiousness, which necessitated contact with higher levels of
violence (Black, 1980; Luckenbill and Doyle, 1989; Perez, 1994; Toch, 1995).

In development of theory tied more closely to neighborhoods as a unit of
analysis, prior research has focused on victimization, with the latter studies
ﬁnding that social disorganization was closely related to criminal victimization
(Sampson, 1984; Sampson and Woolredge, 1987; Taylor and Covington, 1988;
Velez, 2001). In relation to those works, research has also focused on offending
trajectories, ﬁnding that social disorganization leads to increased rates of
offending for various types of crime (Jang and Johnson, 2001; Land, 2000; Smith
and Jarjoura, 1989; Wikstrom and Loeber, 2000). In addition, research focusing

speciﬁcally on violent crimes has found that neighborhoods higher in social

26

disorganization had higher rates of violent crime (Baumer, 2002; Messner and
Tardiff, 1986; Morenoff et al., 2001; Reisig and Parks, 2000; Stewart et al., 2002;
Warner and Rountree, 1997).

Prior research on police behavior has also examined the inﬂuence of
ecological context. For example, studies on police-citizen contact have noted that
these contacts occur most often in neighborhoods with low SES, or that have
been identiﬁed as “bad” by ofﬁcers (Black, 1980; Crank, 1992; Fyfe, 1997;
Mastrofski et al., 1995; McGarrell et al., 1999; Sampson, 1986). In fact,
neighborhood composition has been found to affect: (1) rates of trafﬁc stops
(Mastrofski et al., 1998; Meehan and Ponder, 2002); (2) level of police patrol
(Cox and Frank, 1992; Greenberg et al, 1985); and (3) ofﬁcer behavior toward
citizens (Rossi, 1968; Smith and Frank, 2000; Weitzer, 2000). In addition, the
police subculture has also been argued to affect ofﬁcer behavior, particularly in
certain types of neighborhoods, due to an emphasis on ‘face-saving’ behavior
and the maintenance of authority (Black, 1980, 1998; Chevigny, 1995; Goldstein,
1990; Herbert, 1996, 1998; Manning, 1997; Muir, 1977; Skolnick and Fyfe, 1993;
Wilson, 1968).

Research has also focused on how neighborhood context can affect
citizen behavior toward police ofﬁcers, including the ﬁling of ofﬁcial complaints.
Much of this literature has emphasized dissatisfaction with the behavior of police
ofﬁcers, noting that concentrated disadvantage has a negative effect on
satisfaction with the delivery of police services (Cao et al., 1996; Sampson and

Laub, 1993; Reisig and Parks, 2000; Weitzer, 1999). A signiﬁcant proportion

27

have also noted that poor citizen demeanor can inﬂuence ofﬁcer behavior
(Klinger, 1994, 1997; Lundman, 1994; Worden et al., 1996; Worden and
Shepard, 1996). Studies on citizen complaints have conﬁrmed the negative
effects of concentrated disadvantage, noting that neighborhoods high in this
measure experienced a greater number of complaints for behavior ranging from
disrespect from ofﬁcers to excessive force (Kane, 2002; Kappeler et al., 1998;
Lawton et al., 2001; Mastrofski et al., 1999).

The prior literature on use of force, while generally well-developed, has
failed to focus on neighborhood effects. Many of these studies have presented
only a simple analysis of associations, and have found that the base rate of force
varies widely (Alpert and Dunham, 1995; Edwards, 2000; lACP, 2001; Klinger,
1995; Langan et al., 2001). In multivariate studies of use of force behavior,
ﬁndings have varied on the inﬂuence of numerous ofﬁcer and suspect
characteristics, with the majority ﬁnding that suspect antagonism or physical
resistance signiﬁcantly inﬂuences the use of force (Bayley and Garafalo, 1989;
Engel et al., 2000; Friedrich, 1980; Garner et al. 1995, 2002; Kavanagh, 1997;
Phillips and Smith, 2000; Terrill and Mastrofski, 2002; Worden, 1995).

Only two multivariate studies have focused explicitly on neighborhood
context and use of force behavior. Smith (1986) found that neighborhood social
class inﬂuenced arrest, but not use of force, while Terrill and Reisig (2003) found
that concentrated disadvantage increased use of force. These two studies

represent the foundations of this dissertation.

28

Summary

This introductory chapter has focused on a discussion of why it is
important to consider the effects of neighborhood context on police use of force
behavior. However, the research examined here has been of a more general
nature than the current study requires. While relevant theoretical positions and
concepts have been identiﬁed, emphasizing social disorganization theory and
Black’s theory of social control, the focus was on how the idea of neighborhoods
has been developed in prior theories, rather than on how neighborhood context
has been utilized in analyses stemming from well-deﬁned research questions.
The following chapter will provide a comprehensive review of the literature
regarding various neighborhood context studies, as well as relevant use of force

research.

29

Chapter 2: Literature Review

As the introductory chapter has outlined the necessity of using
neighborhoods as a variable of interest in well-deﬁned research, the current
chapter will identify previous studies which have used neighborhoods in such a
manner. While the current research focuses on use of force behavior by police
ofﬁcers, the use of neighborhood as a variable is more common in other areas of
criminal justice research. In particular, there has been a relatively recent
emphasis on neighborhood found in the literature on adult offending and juvenile

delinquency, with some research on police behavior also taking this approach.

Studies Using Neighborhoods as a Variable of Interest

As mentioned above, the number of studies focusing speciﬁcally on use of
force behavior while also emphasizing the effects of neighborhood on that
behavior is very small. In keeping with the beginnings of the emphasis on
neighborhoods stemming from social disorganization theory, much of the
research after the work of the Chicago School has maintained a focus on adult

offending and juvenile delinquency behaviors.

Offending Behaviors by Adults and Juveniles:
Although the Chicago School focused explicitly on neighborhoods,

identiﬁed as “zones” in much of the early literature on social disorganization,
studies that followed in the wake of the Chicago School research initially

emphasized higher levels of aggregation. The most concrete examples of this

30

take the form of theoretical descriptions of the subculture of violence, also
described as a Southern construct of violence. Wolfgang (1978) noted that this
subculture developed in vaguely deﬁned “areas” characterized by residential
mobility and a commitment to the use of force by its residents to solve problems.
Thus, offending in these areas, particularly violently, is a natural outgrowth of the
tolerance, and even outright encouragement, of violence. Simpson (1985), in
researching the same issues, found that while social class and levels of
inequality did not predict violent crime rates, regional culture and social
disorganization factors had a powerful inﬂuence on these rates.” Luckenbill and
Doyle (1989) examine the same phenomenon, noting that the culture of violence
is not exclusively the domain of urban residents, nor that of minorities of lower
social class. Cao et al. (1996: 379) concur, noting that “a more fruitful search for
the root causes of black violence may be...in the structurally disadvantaged
position of blacks in the US. society.” Thus, it is structural position (as argued by
Black 1976) that leads to an emergence of violence, rather than individual
characteristics such as race. Arguing that disputatiousness, that is, “the
likelihood of being offended by a negative outcome and seeking reparation
through protest...”, is the primary factor explaining a culture of violence in
particular areas, Luckenbill and Doyle (1989: 419) noted that research was more
effective when conducted at smaller levels of aggregation. More importantly, the

latter research noted that structural position was an important factor in how

 

‘5 Note, however, that these results are disputed by Braithwaite (1981), who, in a meta-analysis of previous
studies of social class and criminality, ﬁnds that there is a signiﬁcant relationship between social class of an
area and ofﬁcial rates of both adult and juvenile crime. Liska and Chamlin (1984) noted the same
phenomena~

31

 

disputatiousness developed. That is, the authors argued that individuals whose
positions necessitated contact with higher levels of violence were more likely to
experience disputatiousness, particularly when in a public setting. However, this
research failed to identify police ofﬁcers as individuals in such a position, and
therefore the opportunity to examine use of force behavior from this perspective
was lost. Perez (1994: 41) makes an attempt, however, noting that ‘the paradox
of face’ requires “a believable threat backed up by a little bit of history” of ofﬁcer
violence in the face of experiencing disputatiousness. Black (1980: 31-32) also
discussed use of force as an attempt at ‘face-saving’, noting that “the most
extreme violence seems to occur when a man refuses to submit totally to an
ofﬁcer’s authority.” Toch (1995: 124; italics in original) concurs, ﬁnding that “the
ofﬁcer becomes irritated by what he views as an unforgivable deﬁance of his
authority [leading to] rep defending, where violence is the fate entailed in [the
ofﬁcer’s] role.” Like Luckenbill and Doyle (1989), Hawley and Messner (1989)
focused on a subculture of violence and found that smaller units of aggregation
were more appropriate for research. In addition, they noted that while “violent
behavior is a way to fulﬁll cultural expectations. . . [and is] demanded in certain
interactions”, a more complete explanation of how violent behavior occurs
requires a theoretical model integrating cultural and structural factors (Hawley
and Messner 1989: 486).

Parallel to the studies of large units of aggregation, a smaller group of
researchers focused on neighborhoods as the unit of analysis, although the latter

studies occasionally emphasized theoretical propositions rather than focusing on

32

 

statistical analyses. Representative of this class of literature is the work of
Sampson (1984), who found that interracial criminal victimization was positively
(and very strongly) related to neighborhood heterogeneity.16 Sampson and
Woolredge (1987) also argue that neighborhood factors are related to
victimization. Noting that previous research at the neighborhood level inferred,
rather than directly measured, theoretical concepts, the authors ﬁnd that
residential mobility, housing density, and family disruption were all important
factors inﬂuencing victimization rates. These later studies signaled the beginning
of the use of statistical analyses to examine the effects of neighborhood context
on various human behaviors. Velez (2001), using a social disorganization model,
provides an example, ﬁnding that increased levels of public (i.e. informal) social
control within the neighborhood lead to decreased numbers of victimizations.
Taylor and Covington (1988), in a study of neighborhood context and violence,
found that neighborhood SES inﬂuenced behavior, noting that neighborhoods
with increasing lower class populations17 experienced increasing violence as
stability (used as a proxy for social disorganization) declined.

In addition to the research on generalized victimization risks and offending
behavior, there has also been an emphasis on speciﬁc types of offenses and
offending trajectories. Smith and Jarjoura (1989) ﬁrmly established the

importance of examining both individual and neighborhood characteristics when

 

'6 Interestingly, Shihadeh and Steffensmeier (1994) found that income inequality within a racial group (and,
presumably within a racially homogeneous neighborhood) can also lead to increased violence and
victimization.

'7 Taylor and Covington (1988: 553) used the term “underclass” to describe the minority population in
neighborhoods which were increasingly being marginalized after gentriﬁcation of surrounding
neighborhoods, following the example of Wilson (1993).

33

 

studying offending behavior. In a study encompassing over 9,000 households in
57 different residential neighborhoods, the authors determined that burglary rates
were related to attributes of individual households, as well as to neighborhood
characteristics. Jang and Johnson (2001), in contrast, study the issue of drug
use, and ﬁnd that increased neighborhood disorder leads to increased illicit drug
use while high levels of personal religiousity mediate these effects and result in
decreased illicit drug use. Wikstrom and Loeber (2000), in a study of offending
trajectories, found that low neighborhood SES had a direct impact on late onset
of offending for juveniles who had a mix of both risk and protective factors. Land
(2000) also studied offending trajectories and found similar results, noting that
neighborhood SES had a direct effect on neighborhood social disorganization,
which in turn affected the ability of informal social control methods to control the
behavior of juveniles in the neighborhood.

The greatest amount of research on speciﬁc offenses, however, has
focused on violent crimes. Stewart et al. (2002), in a multisite neighborhood
study across two states, ﬁnd that neighborhood afﬂuence decreases childhood
violence among the residents of that neighborhood. Interestingly, Baumer (2002:
579) found that neighborhood disadvantage did not affect the likelihood of
notifying the police for aggravated assault victimization, while the effect was
curvilinear for simple assault victims, with both high-income and low-income

victims less likely than middle-class victims to notify the police of victimization.18

 

‘3 Interestingly, Warner and Rountree (1997: 520) note that the rate of being assaulted also varied by
neighborhood, with social ties decreasing assault rates in predominantly white neighborhoods, yet having
no effect on assault rates in predominantly minority neighborhoods. The authors note that it is unclear if
these factors also affect reporting rates.

34

Messner and Tardiff (1986) also studied the theme of violence and neighborhood
context, focusing on homicide levels as related to characteristics of urban areas.
Examining 26 neighborhoods in New York City, the authors found that homicide
rates were highest in neighborhoods with extreme levels of poverty and high
rates of single-parent households. Morenoff et al. (2001) conﬁrm these results,
ﬁnding that homicide rates were highest in neighborhoods which experienced
concentrated disadvantage and low collective efﬁcacy. Reisig and Parks (2000:
6) also note the same phenomenon, ﬁnding that neighborhoods high in
concentrated disadvantage had a higher homicide rate, resulting in decreased

satisfaction with the police from neighborhood residents.

Police Behavior:

In addition to the research focusing on criminal behavior by both adults
and juveniles, studies examining the effect of neighborhood context on police
behavior have emerged. These studies bring us closer to our core questions
concerning use of force behavior and the effects of ecological context. The
research to be discussed here typically has focused on arrest practices and
police attitudes toward citizens (and vice versa). In light of this, it is interesting to
note that a study of patrol work by Whitaker (1982) found that of the two hours
per shift in which patrol ofﬁcers had encounters with citizens, only 45 minutes
were spent dealing with problems that were of a criminal nature. Bayley (1994:
17) noted that approximately ten percent, or 45 minutes, of the average shift is

spent dealing with dispatched incidents of a criminal nature. Mastrofski (1995:

35

383) goes further, arguing that overall (i.e. for dispatched and ofﬁcer-initiated
contacts), a relatively small amount of patrol work involves dealing with crime in
some manner. Yet, despite, or perhaps due to, the fact that relatively little time is
spent dealing with suspects and criminal activity, the issues of arrest practices
and police attitudes are viewed as important, as Goldstein (1990: 1) noted,
arguing that “efforts to improve policing should extend to and focus on the end
product of policing - on the effectiveness and fairness of the police in dealing
with the substantive problems that the public looks to the police to handle.” The
equity of police behavior is particularly important when considering the potential
effects of the use of force; however our review of prior works begins here with a
look at other police behaviors ﬁrst.

Sampson (1986), in a study examining delinquency and neighborhood
characteristics, found that individuals in neighborhoods with higher SES
experienced fewer contacts with police ofﬁcers, regardless of the amount of
criminal behavior. According to Sampson (1986: 877), these results indicated
that “a large part of any effect of individual SES on arrests is spurious and
reﬂects an ecological bias in police perceptions rather than a bias directed solely
at lower-class juveniles in actual police encounters.” This conforms to the
previous discussion of ecological contamination in urban neighborhoods
(typically Iow-SES, minority areas). Black (1980: 143) clariﬁes further, arguing
that “discretionary authority often carries with it the possibility of particularistic law
enforcement...[but] whether a system of mobilization is reactive or proactive

does not determine the probability of discriminatory enforcement [it only]

36

organizes that probability.” Smith (1987) replicates this result, ﬁnding that low
neighborhood SES leads to an increased use of arrest by ofﬁcers patrolling those
neighborhoods. Smith (1987: 768) further argues that this increased arrest rate is
a function of “a set of decision heuristics [based on neighborhood context] which
inﬂuence [an ofﬁcer’s] deﬁnition of situations...” McGarrell et al. (1997) provide
further evidence for the latter phenomenon, noting that police ofﬁcers are more
likely to identify neighborhoods with high levels of disorder as being “bad.”‘9
Liska and Chamlin (1984), as well as Crank (1992), continue in this vein, ﬁnding
that the percentage of non-white residents in a neighborhood is predictive of
increased arrest rates. In another example of a study focusing on neighborhood
SES, Seron and Munger (1996: 204) argue that “social control is organizaed
quite differently to deal with different social classes [and] through policing...the
poor, and especially the underclass, experience a special kind of ‘government of
the poor.”’ On the other hand, Sparger and Giacopassi (1986: 25) ﬁnd that
“police ofﬁcers see the wealthy as possessing different values and being
accorded a privileged status by the criminal justice system, resulting in some
resentment on the part of the police.” The latter study makes clear that police
ofﬁcers may engage in behaviors in high-income areas against its residents out
of resentment, a concept clearly at odds with conﬂict theory’s propositions that
only the poor are mistreated. This seems to correspond, however, to work
conducted by Fyfe (1997: 537), who noted that the overwhelming majority of

individuals in an ofﬁcer-initiated encounter, regardless of race or social class, are

 

‘9 Manning (1997: 199) agrees, noting that “the orientation of ofﬁcers” [behavior] is limited. . .by their
geographical knowledge.”

37

 

of poor demeanor because “those who come to police attention do not seek it,
but become unwilling clients through the intervention of [others].” It must be
noted, however, that the rate of ofﬁcer-initiated contacts can be affected by
departmental philosophy and other organizational constraints. Indeed, Mastrofski
et al. (1995) found that ofﬁcers in departments that emphasized community
policing were more selective in making arrests, and were much less inﬂuenced
by legal variables in behavior reminiscent of the threshold-triggered response
discussed earlier.20

Similar to arrest rates, trafﬁc stops provide a measure of police behavior.
Mastrofski et al. (1987), for example, found that overall, smaller departments
tended to initiate more trafﬁc stops per shift than larger departments. The authors
argued that this indicated that the pressure to perform on ofﬁcers was stronger in
smaller departments, thus increasing the amount of stops made in an attempt to
locate criminal activity. In a later study, Mastrofski et al. (1998) found that trafﬁc
stop rates are inﬂuenced by ofﬁcers’ perceptions of racial boundaries (i.e. the
geographic dividing lines between white and minority neighborhoods), with
minority drivers being stopped in mixed race areas at a much higher rate than
that for white motorists in those same areas. Meehan and Ponder (2002) note
similar results, ﬁnding that residential segregation patterns (presumably similar to
geographic boundary patterns as discussed by Mastrofski et al. 1998) inﬂuence

rates of trafﬁc stops performed by ofﬁcers in those neighborhoods. Departmental

 

2° Indeed, this has been one of the greatest criticisms of community policing - in ﬂeeing the ofﬁcer to make
decisions based on previous encounters with the individuals or on other situational variables, the officer has
also been granted the discretion to make decisions based on extralegal variables (including neighborhood
context) in a negative manner.

38

(and ofﬁcer) vigorousness can also be considered a measure of police behavior,
as Greenberg et al. (1985) demonstrate. The latter study found that
neighborhoods with high levels of non-white residents experienced an increase in
police strength.21 With respect to individual ofﬁcers, Brooks et al. (1993) found
that ofﬁcers assigned to slower (i.e. low-crime) patrol beats held more positive
attitudes about their jobs, and, more importantly, about the citizens in the areas
they patrolled. Thus, as noted previously, ofﬁcers in high-crime areas tend to
regard the residents of those areas with suspicion, while ofﬁcers in low-crime
areas tend to have more positive attitudes about the residents of those areas.
With respect to how police ofﬁcers and citizens interact within their neighborhood
contexts as an outgrowth of their attitudes regarding one another, Cox and Frank
(1992) found that high-crime areas experienced an increased level of consistent
behavior from the ofﬁcers assigned to those areas. Interestingly, Rossi (1968)
found that neighborhoods with a high percentage of Black residents, regardless
of their crime rates, were rated by police ofﬁcers as harder neighborhoods in
which to work, as well as being more hazardous. Smith and Frank (2000), as well
as Weitzer (2000), conﬁrm that neighborhood type and racial composition can
affect ofﬁcer behavior, leading to a difference in how ofﬁcers treat White and
Black residents of these areas, although these results were not signiﬁcant.
Weitzer (2000: 129) also notes that “there is substantial agreement
across...communities in the belief that police treat blacks and whites

differently...” Noting this inconsistency in behavior, Harring and Ray (1999: 70)

 

2' The study also found that neighborhoods with high levels of income inequality had higher levels of
police strength, although this result was not signiﬁcant (Greenberg et al., 1985).

39

 

argue that “the confrontation between...alert, aggressive police ofﬁcers and a
frightened Black man is inherently dangerous because of the police culture of
competitively ﬁghting ‘crime wars’ in unfamiliar minority neighborhoods.”
Focusing on these same issues regarding police culture, many
researchers have looked at the police subculture as indicative of poor attitudes,
and predictive of police behavior in certain neighborhoods. Indeed, research in
the ﬁeld of criminal justice has long identiﬁed the existence of a paramilitary
structure within police departments perpetuating an ‘us-vs.-them’ mentality (see
Wilson 1968; Muir 1977; and Goldstein 1990). Manning (1997: 4) was among the
ﬁrst to analyze the subculture, noting that “the driving force of policing is not the
regulations and policies, law, politics of public sentiment; the identifying feature is
the occupational culture in interaction with these forces.” Herbert (1996) concurs,
arguing that police behavior is governed by the normative structure of the
subculture. The author goes on to note that this structure, while making
allowances for bureaucratic regulations and the law, focuses mainly on the
realities of policing the street, and particularly on the dangers of policing in high-
crime neighborhoods. Thus, the ideas of machismo, safety / danger, competence
and morality become the main concern of patrol ofﬁcers operating in these
neighborhoods. Indeed, Weisburd et al. (2000: 3) note that approximately only
4% of police ofﬁcers believed that their fellow ofﬁcers used more physical force
than was necessary in making an arrest. Herbert (1998) notes that in reality there
is an inability of legal and bureaucratic inﬂuences to determine ofﬁcer behavior,

with only the ethos of the subculture affecting ofﬁcer performance and decision-

40

making?2 However, it should be noted that more recent research has argued
against the existence of an overriding police culture, and thus questions the
notion of all ofﬁcers responding to occupational stressors in similar ways
(Paoline, 2003, 2004).

The ideas present within the police subculture are often expressed in the
“paradox of saving face’. Skolnick and Fyfe (1993: 95 italics in original) elaborate
on this point, arguing that “the stmnger one ’s reputation for being mean, tough
and aggressive, the less iron-handed one actually has to be.” Yet, the authors
also note that the subculture sometimes demand forceful action from an ofﬁcer,
particularly in high-crime neighborhoods, noting that “[while] the written rule is
clear: cops are to use no more force than is necessary to subdue a suspect,
[when] a departmental subculture condoning [ofﬁcer aggressiveness] prevails,
the unwritten rule is: Teach them a lesson” (Skolnick and Fyfe 1993: 13). In a
continuing description of the police subculture, Skolnick and Fyfe (1993: 103)
also note that “[it is a] police cultural crime [and] a serious transgression in the
police cultural statute book...to talk back to a cop.” lmportantly, Terrill (2005:
110) argues that “ofﬁcers are socialized to ‘maintain the edge’ and be ‘one up’
on citizens not only to establish control, but to ensure proper respect.” Chevigny
(1995: xi) elaborates, noting that the effects of the police subculture can affect
ofﬁcer behavior to the point that “the police habit of charging the people they beat

with standardized crimes even got the name of a mock crime: ‘contempt of cop.”

 

22 Note, however, that Waddington (1999) has a somewhat different view, arguing that the police
subculture operates differently on the street than it does in concept. In particular, Waddington proposes that
ofﬁcers often engage in subcultural discussions (i.e. gloriﬁcation of violence, expressing a desire for
action) while at the station, but act much differently in the street.

41

Greenleaf and Lanza-Kaduce (1995) examine a different aspect of the
authoritative role of police ofﬁcers within the subculture, focusing on how
suspects respond to that authority. The authors note that “overt conﬂict [is] more
likely when authorities act congruently with their ofﬁcial norms” (Greenleaf and
Lanza-Kaduce 1995: 567). Stated differently, the authors propose that when
ofﬁcers are enacting their role of crime-ﬁghter, there is more likely to be
resistance on the part of the suspect, and corresponding conﬂict. Lanza-Kaduce
and Greenleaf (2000: 223) support this, noting that “police-citizen conﬂict will be
highest when social norms of deference counter positional authority.”23 This
again illustrates the dilemma police ofﬁcers face when confronting citizens in
high-crime neighborhoods: the ofﬁcers must be ﬁrm in order to maintain order
and ‘save face’ for themselves, but it is precisely this type of behavior which will
lead to conﬂict with citizens. It appears, then, that the informal, rather than
formal, aspects of the organization are important determinants of police behavior.
Patrol ofﬁcers respond to the demands of their peers voiced within the
subculture, and structure their behavior in certain neighborhoods accordingly.
Given Black’s (1980) theory of social control, these results regarding the effects
of the police subculture on ofﬁcer interactions with individuals in high-crime
neighborhoods are not surprising. Indeed, Black (1998: 40) noted that lower
status individuals (particularly minorities and the poor) experience less legal
protection overall from the police, and more scrutiny, due to their position within

the social stratiﬁcation structure.

 

23 Note, however, that Weidner and Terrill (2005) fail to support the hypothesis that conﬂict is more likely
when an ofﬁcer’s race, age, sex, and wealth deference norms counter their positional authority.

42

Just as neighborhood context can inﬂuence ofﬁcer behavior and
perceptions, so too can it affect how citizens of those neighborhoods themselves
perceive the behavior of ofﬁcers, or how those citizens react to that behavior.
Sampson and Bartusch (1998), for example, ﬁnd that concentrated disadvantage
within a neighborhood leads to increased dissatisfaction with police and an air of
legal cynicism. Reisig and Parks (2000) replicate this result, noting that
concentrated disadvantage has a signiﬁcant, negative effect on satisfaction with
police ofﬁcers. In further work, Reisig and Parks (2003: 211) also ﬁnd that
“aggressive patrol tactics may be viewed by some as intrusive and inconsistent
with community needs, thus further alienating residents in disadvantaged
neighborhoods who already report high levels of disaffection with police.” The
authors go on to note that concentrated afﬂuence within a neighborhood leads to
increased satisfaction with police, while the use of alternative patrol (as a
measure of community policing efforts) leads to increases in both satisfaction
with police ofﬁcers and in perceived quality of life. In disadvantaged
neighborhoods, however, Cao et al. (1996: 4) note that things get progressively
worse, arguing that “social and physical disorder send a message that law
enforcement has lost control over or consciously abandoned the community.”
Sampson and Laub (1993) underscore the fact that this “message” serves to
perpetuate a cycle of cynicism on the part of ofﬁcers and residents alike, and
leads to further stereotyped attributions that the residents of these areas are a
threatening group. The authors note that structural context, particularly the

obvious racial inequality in wealth and the concentration of “underclass” poverty,

43

 

deﬁnes the limits of appropriate police action. These limits encompass arrest, but
they are also reﬂected in numerous other police behaviors, including “hassling.”
Weitzer (1999) argues that neighborhood context therefore affects not only police
and citizen behavior, but also the attitudes of the citizens concerning their
encounters with the police. Within such a context, it is easy to understand why
every encounter between a citizen and a police ofﬁcer is fraught with tension and
angen

With respect to how neighborhood context has been presumed to affect
citizen behavior in response to police ofﬁcers, various studies have focused on
the effects of citizen demeanor. This research ties together aspects of a
threshold for police response, as well as aspects of the concepts of honor and
‘saving face’. In an ecological analysis of police behavior, Klinger (1997)
presented the hypothesis that ofﬁcers patrolling in high-crime areas were less
likely to exert their legal authority (e.g. make an arrest, interview witnesses) due
to the factors of: (1) time constraint; and (2) deservedness of victims. Although
not explicitly mentioned in the research, one implication of Klinger’s (1997) work
is that a citizen’s demeanor may inﬂuence the perception of their deservedness.
Indeed, Klinger (1994; 1996) did study whether such a demeanor effect occurs
during police-citizen interactions. Over the course of two separate research
efforts within in a single police department, Klinger (1994; 1996) found that
citizen demeanor was not a predictor of arrest behavior when criminal conduct
during the encounter was introduced into the analyses. However, using a

different data set, Lundman (1994) found that a demeanor effect on arrest was

44

 

present, and that the effect relied in large part on model speciﬁcation. This is
supported by Worden et al. (1996: 330) who note that much confusion stems
from the fact that “different studies have deﬁned demeanor, both conceptually
and operationally, in somewhat different ways.” However, Worden and Shepard
(1996), in a re-analysis of several data sets across domestic disturbances, trafﬁc
stops, and disputes found that the demeanor effect on arrest behavior (i.e. poor
demeanor on the part of suspects increased the chances of an arrest) persisted
even after considering criminal behavior within the encounter. The concepts of
honor and demeanor are even played out between juveniles and police ofﬁcers,
as Fagan and Wilkinson (1998) noted. The authors argued that the high demand
for guns in inner city areas was “fueled by an ecology of danger” (Fagan and
Wilkinson 1998: 105). In this environment, youths across two New York City
neighborhoods were seen to possess guns due to the respect afforded to them
by this possession. More importantly, this culture of respect for carrying and
using guns signiﬁcantly altered the interactions between police ofﬁcers and
juveniles, with the latter acknowledging that gun ownership lead to poor
demeanor on the part of suspects (Wilkinson and Fagan 2000). In a study which
focused more speciﬁcally on suspect resistance, Greenleaf and Lanza-Kaduce
(1995: 565) found that “after controlling for race, sex, and area of the city, overt
conﬂict between the police and citizens is related to the organization and
sophistication of the participants involved.” The former factor was composed of
elements of victim-suspect relationships, number of arrestees, and number of

bystanders, while the latter factor was composed of ofﬁcer experience, suspect

45

presentation, and the nature of the situation. In a later work, Lanza-Kaduce and
Greenleaf (2000: 227) examined the effects of resistance24 on the arrest
decision, and conﬁrmed their previous ﬁndings.

A ﬁnal grouping of research efforts regarding neighborhood context and
police behavior brings us even closer to our core research questions. While not
explicitly exploring the subject of police use of force, some researchers have
focused their efforts on other police behaviors which may be illegal or unethical.
Mastrofski et al. (1999), using Black’s (1980) theories on the behavior of law, as
well as again using the concept of concentrated disadvantage as a neighborhood
measure, ﬁnd that the latter phenomenon resulted in increased disrespect from
police ofﬁcers patrolling those neighborhoods.25 Kane (2002) found that
neighborhoods which experienced considerable structural disadvantage
(conceptually similar to concentrated disadvantage measures prevalent in other
research efforts) were far more likely to have reports ﬁled regarding police
misconduct within those neighborhoods. lmportantly, McCluskey and Terrill
(2005) ﬁnd that an ofﬁcer’s complaint rate for force and verbal discourtesy is
associated with higher levels of coercion in encounters with suspects. Also
regarding complaints against ofﬁcers, Kappeler et al. (1998: 127) note that “once
a justiﬁcation [for misconduct] has been accepted...subsequent deviance
becomes easier for the actor.” The authors go on to argue that this process of
acceptance of police deviance is more likely to occur when the deviance (i.e.

misconduct) has taken place in a high-crime area, noting that “characterizations

 

2‘ This was operationalized as the presence of any of the following: (1) verbal attack on the ofﬁcer; (2)
ghysical attack on the ofﬁcer; (3) refusal to obey a lawful order; or (4) resisting arrest.
It is important to note, however, that this relationship did not reach statistical signiﬁcance.

46

 

of victimized citizens as ‘drug users’, ‘deviants’, ‘criminals’, and ‘psychopaths’
usually serves [the ofﬁcer] well” (Kappeler et al. 1998: 126). Lawton et al. (2001),
in an analysis with a speciﬁc Geographical Information System (GIS) component,
found that disadvantaged neighborhoods had a signiﬁcant, positive effect on the
number of complaints ﬁled against police ofﬁcers patrolling those neighborhoods.
This latter study is particularly useful for the current research, as a GIS package
is also used here to identify neighborhoods and analyze the effects of their

context.

Use of Force Research

The police role is one that is inherently concerned with the threat and
application of violence, and violent encounters between police ofﬁcers and
citizens are carefully scrutinized to determine if the police used their powers of
coercion inappropriately. Incidents such as those involving Abner Louima and
Rodney King have decayed public conﬁdence in their police ofﬁcers, shattering
the mythologized image of the police ofﬁcer as the helping hand of the law.
Indeed, research shows that after a well-publicized incident of police use of force,
public opinion of the police takes a sharp downturn (Lasley 1994; Tuch and
Weitzer 1997). The effects of such incidents are typically felt more strongly
among non-whites, particularly those living in disadvantaged areas (Arthur and
Case 1994; Kaminski and Jefferis 1998; Son et al. 1997). Arthur and Case (1994:
167), for example, found that “in 1991 70% of white [and only] 43% of black

respondents approve of a policeman striking an adult male citizen under some

47

 

circumstances.” As has been discussed previously, the impacts of these
incidents are detrimental to the gains made by community policing programs, and
thus their study, and ultimately their prevention, is of great concern to criminal
justice researchers. The following section describes a variety of studies which
have focused on use of force behavior by police ofﬁcers in order to provide a

foundation for the current research questions.

Studies of Association Examining Use of Force Behavior:

While the issue of use of force behavior is relatively well-studied,
particularly in the past two decades, most of these studies either: (1) provide only
measures of association between that behavior and some variable; (2) are of a
purely descriptive nature, outlining only the base rates of force; or (3) provide
only simple regression equations for statistical analyses. Although there are
some studies which describe the effects of neighborhoods on behavior,
neighborhoods are typically categorized as a sociological variable, and are
studied within a more simplistic statistical analysis than the multi-level modeling
proposed here. However, these studies, while not as rich in detail as this
dissertation, provide a preliminary understanding of how use of force behavior
develops.

Much descriptive research has stemmed from government-sponsored
data collection efforts. The International Association of Chiefs of Police (IACP),
for example, developed the National Database Project on Police Use of Force in

order to “reﬂect operational realities of modern, street-level enforcement,

48

including the very meaning of “police use of force,’ deﬁned as the amount of force
required by police to compel compliance by an unwilling subject” (IACP: 2001).
These efforts found that for the 26 agencies reporting for the years 1997 and
1998, the ovenivhelming majority of incidents of use of force behavior occurred in
arrest-related situations.26 In addition, of the 2,264 use of force incidents reported
to the lACP, 909 were intraracial and 1,335 were interracial (IACP: 2001). More
importantly, a larger sampling of agencies (serving over 30% of the population of
the United States) over the years 1991 through 2000 found that the police used
force 3.61 times per 10,000 calls for service, a base rate of .04% (IACP: 2001).
These results were replicated in the 1999 Police Contact Survey (distributed as
an addendum to the National Crime Victimization Survey during that year), which
found that of the 44 million people reporting a face-to-face contact with police
ofﬁcers, 422,000 (.96%) experienced either the threat or use of force by the
ofﬁcer (Langan et al. 2001). Note that although the base rate is different (incident
reports vs. calls-for-service), the percentage of individuals experiencing use of
force in some manner remains below 1%.27 National data collection efforts also
extended to other countries, with the Queensland (Australia) Criminal Justice
Commission (Edwards 2000: 1) ﬁnding that “20% of the respondents to the 1999
Defendants Survey reported that police had used some kind of force against

them.” lmportantly, the Commission (Edwards 2000: 1) went on to note that “of

 

2‘ A discussion of why the current research chose to sample only arrest events will be presented in detail in
Chapter 3.

27 It is important to note also that this increase from .04% to .96% is possibly due to the extrapolation
required by the survey. The PCS was distributed to a nationally representative sample of 6,421 individuals
over the age of 12, representing (through extrapolation) 216 million individuals. Of these, 1,308 had a face-
to-face contact with police officers (representing 44.6 million individuals), and 14 experienced use of force
behavior (representing approximately 450,000 individuals) (Greenfeld et al. 1997).

49

those respondents who reported that police had taken some physical action
(either ‘force’ or ‘restraint’), about a quarter acknowledged that they had resisted
arrest...”

Aside from national data collection efforts, some researchers have chosen
to focus on one city or a set of cities in order to examine the issue of use of force
behavior. One of the earliest efforts was conducted by Milton et al. (1977), who
found that in 7 large (i.e. over 750,000 people) cities, while the rate of minorities
shot by police was much greater than their proportion in the general population,
this rate was consistent with the arrest rate of minorities for serious crimes. In
fact, Geller and Scott (1992: 153), note that many earlier studies of use of force
behavior found a strong relationship between minority arrest rates for serious
felonies and the percentage of shooting victims who are minorities. They go on to
note, however, that many of these studies have failed to include important
contextual information. It is also important to note that more recent studies have
found that minorities are disproportionately the victims of police shootings.
Goldkamp (1982), for example, ﬁnds that low-income suspects, regardless of
race or minority arrest rates, were more likely to be shot by police ofﬁcers. This is
supported by Locke (1996: 135), who argues that “persons of color are
disproportionately represented among those subjected to police use of force
where the discharge of a ﬁrearm is involved.” In an analysis of the Metro-Dade
Police Department in Miami, Alpert and Dunham (1995: 19) found that while in
31% of the cases of a ﬁrearm discharge a white ofﬁcer shot at a Black suspect, in

only 1.37% of the cases did a Black ofﬁcer ﬁre a shot at a white suspect.

50

Interestingly, in an analysis of the same department, Klinger (1995) found that
the base rate of physical force was 17%, a rate almost double that of any
previously published studies. Holmes (2000: 343), in an analysis of civil rights
complaints against police departments, also found an effect for race, noting that
“measures of the presence of threatening people (percent Black, percent
Hispanic [in the Southwest], and a majority/minority income inequality) were
related positively to average annual civil rights criminal complaints.”

Cloninger (1992) also conducted analyses using aggregate, city-level
data. In the analysis, the author found that there were a number of city
characteristics which accounted for the rate of deadly force, notably: (1) the
number of police ofﬁcers in the city per violent offense; (2) the probability of
arrest and conviction in the city; (3) the number of non-homicide violent offenses;
(4) the number of homicides per capita; and (5) the number of police killed per
violent offense. Using a much larger data set, two separate sets of researchers
also examined use of force (notably deadly force) using city-level data. Jacobs
and O’Brien (1998) conducted an analysis of 170 cities to examine police killings
between 1980 and 198628 in all cities with a 1980 population of greater than
100,000, while Sorensen et al. (1993) perform analyses on 169 cities to study
police killings between 1980 and 1984 in all cities with a 1980 population greater
than 100,000. With a total of 1,23129 cases, the latter authors ﬁnd four variables

to be statistically signiﬁcant, namely: (1) Southern geographic location; (2) the

 

2‘ The authors report that the mean number of police killings for all 170 cities in this period is 9.68, while
the mean rate per 100,000 for this period is 1.99 (Jacobs and O’Brien 1998: 846-847).

29 These cases represent all instances where the use of force by a police officer resulted in the death of a
felon.

51

violent crime rate; (3) percent Black in the city; and (4) the degree of economic
inequality. Jacobs and O’Brien (1998) also ﬁnd four variables (out of 12 inserted
in the model”) to be signiﬁcant: (1) city population size; (2) the divorce rate in the
city; (3) the murder rate in the city; and (4) the increase in the proportion of Black
to White populations.

In addition to those studies focusing on city-level or national data, there
are numerous efforts which present data only on non-lethal force options. In an
example of the latter, Klinger (1995: 175) studied over 100 ofﬁcers from three
patrol districts of the Metro-Dade County (Florida) Police Department and found
that in slightly over 60% of police-citizen encounters, the ofﬁcers did not use any
force. Klinger (1995) went on to note that in most instances where force was
used, the ofﬁcer used no level greater than simple verbal commands. In a study
speciﬁcally focused on verbal commands, MacDonald et al. (1985) found that if
one considers the potential for abuses based on the average number of
encounters multiplied by the number of ofﬁcers multiplied by the number of shifts,
physical and verbal abuse if not a problem. However, the authors also found that
two groups of ofﬁcers contributed disproportionately to the negative incidents
reported to the department. The ﬁrst group, making up between 10 to 20% of all
ofﬁcers, was characterized by consistently negative attitudes toward the public,
resulting in verbally abusive and rude behavior on the part of the ofﬁcers in their
contacts with the public (MacDonald et al. 1985: 299). The second group,

comprising 10 to 15% of all ofﬁcers, was characterized by an intense involvement

 

3° Note that Jacobs and O’Brien (1998) estimate two separate models — one for police killings of Black
suspects, and one for police killings of all suspects.

52

with the policing function in conjunction with disenchantment with the
organizational restraints of police work. This group was found to be
overrepresented in incidents of ‘curbside justice’, where suspects where
subjected to force but not arrested (MacDonald et al. 1985: 309). Upon further
reﬂection, then, one could say that between one-ﬁfth and one-third of police
ofﬁcers in the study engaged in some form of abuse of citizens, and thus it is
unusual that MacDonald et al. (1985) would claim that abuse was not
problematic within this department. Indeed, White (1994) argues that while most
police agencies ofﬁcially prohibit obscene language in contacts with citizens, the
concept of ‘command presence’ is taught at the police academy as an effective
means of controlling the situation. The author goes on to note that “profanity and
obscenity [are used] by ofﬁcers in. ..contacts as a form of aggression” (White
1994: 230).

Of the studies focusing on non-lethal force options, some researchers
have moved up the continuum of force from verbal commands to look at the use
of chemical tactics. Holmes et al. (1998) used vignettes in a survey of 662
ofﬁcers to assess the effects of training on the ofﬁcers’ ability to predict threat
from a suspect and react accordingly. The authors estimate three separate
models — the ﬁrst with the ability to predict threat as the dependent variable, the
second with the number of verbal warnings given to the suspect as the
dependent variable, and the third with the application of force as the dependent

variable (with the perceived threat level and the number of warnings included as

53

independent variables) (Holmes et al. 1998: 90-91).31 After testing both the
Threat Level and Verbal Warning models, the authors ﬁnd that the majority of
independent variables found to be statistically signiﬁcant in the ﬁrst model remain
so in the second model, and thus they estimate the Force Level model.32 In the
latter, Holmes et al. (1998) found 7 of 13 variables to be signiﬁcant, namely: (1)
the number of warnings given to a suspect; (2) the perceived level of threat from
the suspect; (3) suspect resistance; (4) suspect gender; (5) the ofﬁcer carried
chemical spray; and (7) years of service in the department. In another study
regarding these tactics, Kaminski et al. (1999) estimated three separate models,
with dependent variables of: (1) “OC spray eased arrest”; (2) “the suspect was
incapacitated”; and (3) “OC has minimally effective effects”. Focusing solely on
the use of oleoresin capsicum (OC) spray, the authors collected surveys from
ofﬁcers regarding 174 incidents of use between July 1993 and March 1994,
which they supplemented with 878 ofﬁcial reports from incidents between April
1994 and December 1996.33 After estimating all three models, Kaminski et al.
(1999) determine that the only variables which are statistically signiﬁcant are the
following: (1) the suspect was drinking; (2) the suspect was on drugs; (3) the age
of the suspect; and (4) the suspect was between 5 and 20 feet away from the

ofﬁcer at the time of OC use. In an attempt to determine the effectiveness of

 

3‘ The authors use the standard dictionary deﬁnition of threat to deﬁne their dependent variable as
“involving the intention on the part of the suspect to injure the ofﬁcer or other citizens” (Holmes et a1.
I998: 90).

32 It is important to remember that Model 3 only asks officers how they would respond with force given
certain situations, thus measuring their attitudes and beliefs regarding their own actions, rather than
measuring those actions directly.

33 The authors note that they eliminated from the analysis encounters involving crowd situations, animals,
misses, or OC canister malfunctions. More importantly, they also note that incidents with multiple ofﬁcers
using 0C, or multiple suspects on whom 0C was used, were randomly sampled so that either only one
ofﬁcer or one suspect was used in the analysis. The resulting number of cases in the ﬁnal analysis was 690.

54

defensive tactics in a more general sense, including the use of OC spray,
Kaminski and Martin (2000) surveyed 600 ofﬁcers from a large West Coast
municipal department in June 1998. The authors estimated six different models
focusing on ofﬁcer attitudes towards defensive training tactics and use,
particularly in relation to assaultive and resistive suspects. Over these six
models, the following variables emerge as signiﬁcant: (1) the rank of the ofﬁcer;
(2) the age of the ofﬁcer; (3) the ofﬁcer received additional academy training; and
(4) the ofﬁcer had been assaulted during the previous year.

What is most important to note about the group of studies on non-lethal
force options just discussed here is their lack either of including context variables
or of ﬁnding these variables to be signiﬁcant. Indeed, the closest any of these
research efforts come to a contextual analysis is the inclusion of a “location”
variable in the regression equation (Klinger 1995; Holmes et al. 1998). In these
instances, not only is the variable included as a nod to sociological inﬂuences on
ofﬁcer behavior in a ﬂat regression equation, rather than a more appropriate
multilevel model, the variable is not even found to be signiﬁcant (most likely due
to the use of inappropriate statistical techniques), thus obscuring any possible
real effects that context might have on non-lethal force behaviors by police
ofﬁcers.

The studies described above concerned themselves with less-than-Iethal
force options, yet they used a variety of measures to deﬁne their dependent
variables. In addition, as Desmedt (1984: 170, italics in original) notes, the use of

these options is often not clearly deﬁned by departmental policies. He goes on to

55

point out, however, that “there are clear limits placed on ofﬁcer’s right to use
deadly force”, stemming mainly from the Supreme Court decision in Tennessee
v. Garner, widely known as the “ﬂeeing felon” rule. One of the implications of this
statement is that criminal justice scholars are more likely to study the use of
lethal force as it is easier to identify and delineate as justiﬁable or excessive. In
one such study, Sherman and Blumberg (1981) found no differences in the use
of lethal force based on ofﬁcer education, with a sample consisting of 36% of
ofﬁcers with one year of college and 6% of ofﬁcers having completed an
undergraduate degree. Importantly, the authors go on to argue that “the legal
opportunities for police ofﬁcers to kill people probably occur far more often than
either the rate of shooting or the number of citizens killed suggests, which would
make the decision to shoot a highly discretionary one” (Sherman and Blumberg,
1981: 318). In an attempt to identify the factors affecting such a highly
discretionary decision, some research has argued that it is how ofﬁcers perceive
and respond to their work environment which structures the use of deadly force.
Matulia (1985), for example, argues that the use of deadly force is related to the
levels of crime and violence in the community. Studying homicides by police
ofﬁcers in 57 cities over a 14-year period, the author ﬁnds that the use of lethal
force is signiﬁcantly correlated with the violent crime rate. Interestingly, Matulia
(1985) also ﬁnds that the application of deadly force is signiﬁcantly correlated to
the number of police ofﬁcers murdered in the line of duty. The latter ﬁnding is
reminiscent of the earlier discussion regarding the police subculture, which often

justiﬁes police use of force as a defensive tactic in dangerous areas. In addition,

56

Geller and Karales (1981) have referred to the use of force as a “split-second”
decision, and further research has consistently supported the notion that ofﬁcers
rarely have the luxury of examining all of the aspects of a particular encounter
before engaging in the use of force (Copeland 1986; Fyfe 1982b; Vila and
Morrison 1994; Waegel 1984b). Thus, the use of deadly force can be seen to be
affected by general (i.e. contextual) characteristics, rather than by the
characteristics of speciﬁc encounters.

The ﬁnal group of preliminary research studies of the use of force, as
opposed to the multivariate analyses to be discussed in the following section,
includes numerous instances of deﬁnitions of the continuum of force, studies
which merely provide base rates of force, and some purely theoretical works.
Many of these studies focus primarily on the deﬁnition of force behavior itself, as
this is a discussion which is critical to any research on the topic. The IACP
(2001), for example, deﬁnes force as: (1) physical force (the use of hands or
feet); (2) chemical force (the use of OC spray or mace); (3) electronic force (the
use of Tasers or stun guns; (4) impact force (the use of batons and ﬂashlights);
and (5) lethal force (the use of a ﬁrearm. Note that while relatively detailed, this
study does not consider verbal commands by the ofﬁcer to be a use of force
behavior, nor does it attempt to cover the use of K-9 ofﬁcers or patrol vehicles as
instruments of force.“ In the Police Contact Survey, Langan et al. (2001) deﬁned
force as any contact in which an ofﬁcer pushed, grabbed, kicked or hit another

individual. Importantly, this research differed from that of the IACP by including

 

3‘ Most notably, the IACP explicitly states that it does not consider routine handcufﬁng during transport,
ﬁeld questions or investigation, as a use of force behavior. This will be further explored in Chapter 3.

57

bites from a K-9 unit, as well as the threat by an ofﬁcer to use any type of force.
Typically, however, the force continuum has been regarded as consisting of: (1)
ofﬁcer presence; (2) verbal commands; (3) control I restraints techniques; (4)
chemical agents; (5) impact weapons; and (6) deadly force, in increasing order of

potential for injury to the target.35

Klinger (1995: 172), for example, notes “that as
one moves from verbalization to deadly force the level of physical discomfort or
injury that citizens may likely experience as the result of ofﬁcers’ actions
increases.” Carter (1984: 226) expands upon this notion, arguing that in addition
to the continuum of force behavior, there are various types of inappropriate force
behaviors, speciﬁcally: (1) physical abuse / excessive force; (2) verbal /
psychological abuse; and (3) legal abuse. Alpert and Dunham (1995: 2) present
a somewhat different model, arguing that “to calculate the force factor, one must
measure both the suspects’ level of resistance and the ofﬁcers’ level of force,
both measured on the same scale.”

When considering those studies which provide base rates of the use of
force, one notes a signiﬁcant amount of variation. Indeed, Geller and Scott (1992:
23) note that “perhaps the greatest pitfall in interpreting studies on [use of force
behavior] is attempting to compare data [as] the reporting categories of different
police agencies vary, and the methodologies and deﬁnitions of key terms and
events differ or are unstated in many studies.” In a study of complaints against

ofﬁcers, for example, Perez (1994: 129) notes that a civilian review board found

that there was ofﬁcer misconduct in 17% of all cases, and in 20% of all cases

 

3’ Note that the continuum has also been deﬁned in terms of legality, as Adams (1997) proposes, ranging
from: (1) deadly force; (2) police brutality; (3) excessive force; (4) excessive use of force; (5) illegal use of
force; and (6) improper, abusive, illegitimate, unnecessary use of force.

58

involving allegations of excessive force?6 These results are similar to those
found by Holmes (2000). Yet, these studies focus only on complaints, and are
thus not instructive in analyzing the totality of use of force behavior. In examining
police contacts, Herz (2001: 58) found that “51% of police contacts that involved
the threat of or use of force involved juveniles between the ages of 12 and 19
[yet] juveniles ranked fourth (19%) in overall contacts with police.” It is important
to note, however, that the use of force was not the primary focus of this study,
and thus the deﬁnitions of force behavior were vague and widely deﬁned. In
studying arrests, Adams (1996: 61) notes that approximately 6% of all arrests
involve the application of some amount of force by police ofﬁcers. Garner et al.
(1995), in contrast, found that ofﬁcers used some amount of physical force in
22% of all arrests. Overall, previous studies have found base rates of force
ranging from some type of force in under 5% of incidents or arrests (Friedrich
1980; Croft 1985; Worden 1995; Engel et al. 2000; Langan et al. 2001) to some
type of force in over 20% of incidents or arrests (Smith 1986; Garner et al. 1995;
Klinger 1995; Edwards 2000; Terrill and Mastrofski 2002). The lowest base rate
of force noted was .8% (Langan et al. 2001), while the highest base rate of force
was 58.4% (Terrill and Mastrofski 2002). Clearly, then, there are signiﬁcant
issues regarding measurement and methodologies among previous studies of
the use of force. While this is one of the main weaknesses of studies on use of
force behavior, it is one that extends to both quantitative and qualitative studies,

both old and new research, and both studies of association and multivariate

 

36 It is interesting to note that these rates for the civilian review board were not much higher than those
from the internal affairs review board (Perez, 1994: 129).

59

research efforts. Nonetheless, the multivariate research studies discussed in the
next section provide a signiﬁcant advantage in that their measures of force are
used in conjunction with regression equations which take into account the effects

of the sum total of all independent variables.

Multivariate Studies on Use of Force Behavior:

Despite the presence of numerous studies of use of force behavior
covering the past two and a half decades, there are few studies that use
multivariate statistical techniques. Of these, only two focused speciﬁcally on a
neighborhood analysis, and thus these will be discussed separately. The
remaining studies (Bayley and Garafalo 1989; Engel et al. 2000; Friedrich 1980;
Garner et al. 1995; Garner et al. 2002; Kavanagh 1997; Phillips and Smith 2000;
Terrill and Mastrofski 2002; Worden 1995) will be discussed here, in order to
identify: (1) speciﬁc measures of use of force behavior; (2) theoretical
orientations of the research; (3) samples used in the research; and (4) results
concerning independent variables found to have an effect on use of force
behavior.

The ﬁrst multivariate study of police ofﬁcer use of force was an analysis by
Friedrich (1980) of data from Reiss’ (1971) study conducted in high-crime areas
in Chicago, Boston and Washington, DC. Using 1,091 cases of the larger
sample of 5,391 encounters, Friedrich focused only on a dichotomous “force - no
force” distinction for the dependent variable. Finding a base rate of 5.1% for any

type of force used by an ofﬁcer, Friedrich used an ordinary least squares

60

regression analysis. Testing variables from psychological, sociological and
organizational perspectives, Friedrich (1980) found the following independent
variables to exhibit a statistically signiﬁcant, and positive, effect on use of force
behavior: (1) poor suspect demeanor;37 (2) “agitated” suspect; (3) suspect was
not sober (i.e. under the inﬂuence of either drugs or alcohol); (4) the number of
citizens present at the encounter; and (5) the number of ofﬁcers present at the
encounter.

Bayley and Garafalo (1989), observing only the 4pm-midnight shift over
350 hours in 3 different precincts in New York City, describe their dependent
variable as Potentially Violent Mobilizations (PVMs). The latter is further deﬁned
as: (1) police-citizen encounters involving disputes; (2) intervention of the police
to apply law against speciﬁc individuals; and (3) all police attempts to question
suspicious persons. The authors also noted that they were speciﬁcally interested
in examining any differences in behavior between ofﬁcers identiﬁed by their
peers as “skilled” and other ofﬁcers. However, while the former used force in 9%
of PVMs, the latter did so in only 6% of PVMs. More importantly, Bayley and
Garafalo (1989) noted that all of the variables found to be statistically signiﬁcant
predictors of force behavior were signiﬁcant for both skilled and control ofﬁcers,
thus indicating only minor differences between the two groups. Of the
independent variables included in their ordinary least squares regression model,

the following were those found to be statistically signiﬁcant: (1) the suspect was

 

37 Of the suspect sample, 186 individuals were considered “deferential”, 1011 were considered “civil”, and
295 suspects were considered “antagonistic”.

61

verbally antagonistic; (2) the suspect had a weapon; and (3) there was conﬂict at
the scene when the ofﬁcer arrived.

Garner et al. (1995) examine ofﬁcer use of force behavior, using 1,585
ofﬁcer surveys completed by ofﬁcers in the Phoenix Police Department after an
arrest.” While the authors operationalize force in three different ways (physical
force, a continuum of force, and maximum force), they note that some form of
physical force occurred in 22% of all arrests studied (Garner et al. 1995: 157).
After coding over 50 independent variables into six categories (the arrest
situation, the arrest location, ofﬁcer mobilization, ofﬁcer characteristics, suspect
characteristic, and an interaction of ofﬁcer and suspect characteristics), the
authors go on to estimate a regression equation for each of the three dependent
variables. After estimating these regression equations, Garner et al. (1995) ﬁnd
11 of the independent variables to be statistically signiﬁcantly related to the use
of force. Speciﬁcally, these are: (1) the use of a contact and cover tactic; (2) an
increase in the number of police ofﬁcers; (3) the arrest was ofﬁcer-initiated; (4)
the presence of a male ofﬁcer; (5) the presence of a male suspect; (6) the
presence of bystanders; (7) the suspect had committed a violent offense; (8) the
suspect was impaired by alcohol; (9) the suspect was known to be resistive or
carry a weapon; (10) the suspect was known to be involved with a gang; and (11)
the suspect used force against the ofﬁcer. It is important to note that all of these

independent variables are consistent predictors (i.e. either across 2 or all 3 of the

 

38 The authors note that they only sampled arrests, acknowledging that the use of force can obviously occur
in numerous situations where an arrest is not made. The authors go on to state that ofﬁcial records indicate
that 1,826 adults were taken into custody during the study period, indicating a response rate of 86% for the
surveys.

62

models tested) of either police use of force, suspect use of force, or both.
However, as the suspect’s use of force is the strongest predictor of the use of
force by an ofﬁcer, these variables can all be considered to predict use of force
by ofﬁcers (albeit some of them in an indirect manner).

In an attempt to examine variables from the three theoretical perspectives
(psychological, sociological and organizational), similar to that of Friedrich
(1980), Worden (1995) used observational data from 5,688 police-citizen
encounters collected during the Police Services Study in Rochester, New York,
St. Louis, Missouri and Tampa I St. Petersburg, Florida (Ostrom et al. 1977).
Restricting the sample to only those encounters between ofﬁcers and individuals
identiﬁed as suspects, Worden (1995) examines 1,528 such encounters with a
three-category force variable - no force, reasonable force, and unreasonable
force. Overall, the author ﬁnds a base rate of 3.9% for some force used during an
encounter, with 2.4% being reasonable force and 1.5% being unreasonable
force. Using a multinomial logistic regression analysis, with separate models for
reasonable and unreasonable force, Worden (1995) ﬁnds the following variables
to be statistically signiﬁcant predictors of either one or both types of force: (1) the
suspect had committed a violent crime; (2) the encounter involved a car chase;
(3) the suspect was a minority; (4) the suspect was male; (5) an increased
number of bystanders; (6) the suspect was intoxicated; (7) the police department
was considered “bureaucratic”; and (8) the suspect was hostile toward the police
ofﬁcer. The latter was expressly noted by Worden (1995) as having the greatest

effect on police use of force behavior within the sample.

63

Kavanagh (1997) examines instances of resisting arrest in 1,108 police-
citizen arrest encounters at the Port Authority Bus Terminal in New York City
between July 1990 and July 1991. Using data from arrest reports, call logs,
police injury reports, ofﬁcer personnel ﬁles, ofﬁcer surveys and personal
observations, the author developed a model with “any physical force”39 as the
dependent variable. Note, however, that the deﬁnition of any physical force
includes force used by either the ofﬁcer or the arrestee (1997: 19)‘°. The author
notes that the occurrence of physical force during the encounter was measured

by the presence of the crime of “resisting arrest”41

among those with which the
suspect was charged, in addition to an indication in the ofﬁcer’s report that the
resistance was active, rather than passive. In terms of operationalization of the
dependent variable, Kavanagh also notes that it was required that the arrest
come at the hands of an on-duty, uniformed patrol ofﬁcer in situations in which no
supervisor was present. Overall, Kavanagh ﬁnds a base rate of 17.2% of arrests
involving the use of force by either an ofﬁcer or a suspect. In all, Kavanagh
(1997) estimates 39 separate logistic regression models, substituting a variety of
independent variables in and out of the main model. After such extensive

analyses, the author concludes that none of the 38 models present any

statistically signiﬁcant differences over the main model, and thus it is most

 

39 Kavanagh (1997) notes that “any physical force” was chosen as the dependent variable because it is
simpler to deﬁne and measure objectively than illegal or unnecessary force.

‘0 While it is unlikely that an officer would refrain from using force after a suspect has used physical force,
it is less clear whether a suspect would necessarily provide physical resistance after an ofﬁcer has used
force. In either case, Kavanagh (I997) fails to provide descriptive statistics concerning the correlation
between both types of force.

“ Kavanagh (1997: 19) states that the crime of resisting arrest is deﬁned in the New York State Penal Code
as “intentionally prevent[ing] or attempt[ing] to prevent a police officer from effecting an authorized
arrest.”

instructive to examine the latter more carefully. The main model contains 22
independent variables, of which 10 were found to exhibit a statistically signiﬁcant
effect on the dependent variable, with 3 of those only statistically signiﬁcant at
the .10 level. Speciﬁcally, the presence of physical force during an arrest
encounter (by either the ofﬁcer or the arrestee) was predicted by the following:
(1) an ofﬁcer’s belief in the need to “save face” (.10 level); (2) the height of the
arrestee (.10 level); (3) the arrestee is Hispanic (.10 level); (4) the seriousness of
the crime”; (5) the presence of other violence during the encounter; (6)
disrespect from the arrestee“; (7) the arrestee was intoxicated; (8) the arrest
took place during the day“; (9) the arrest was ofﬁcer-initiated; and (10) there
were other arrestees present (Kavanagh 1997: 22).

More recently, Engel et al. (2000) conducted another multivariate analysis
using data from the Police Services Study (as did Worden 1995) to examine the
effects of encounter-speciﬁc variables in interactions with suspect demeanor.
Here the authors make a point to note that the sample is not random, with the
three cities (Rochester, New York; St. Louis, Missouri; and Tampa I St.
Petersburg, Florida) chosen to represent a cross-section of various
organizational styles and service conditions in urban neighborhoods. Focusing
on 60 different neighborhoods spread across these departments, with 15 shifts
sampled in each neighborhood, the study originally compiled data on 5,688

police-citizen encounters. As did Worden (1995), Engel et al. (2000) narrow the

 

‘2 Operationalized as felony vs. misdemeanor.

‘3 Disrespect was considered as the suspect either verbally abusing the ofﬁcer, refusing to stop and talk to
the ofﬁcer, or reﬁising to be handcuffed.

“ In an unexpected turn, the presence of a night-time arrest actually decreased the incidence of the use of
force.

65

sample, focusing only on those encounters between ofﬁcers and individuals
identiﬁed as suspects which were not trafﬁc stops, resulting in a ﬁnal sample of
1,461 cases. Using a dichotomous dependent variable of “force - no force”, the
authors ﬁnd a base rate of 3.4% of encounters involving the use of some amount
of force.45 While Engel et al. (2000) go on to estimate nine different logistic
regression equations to test the demeanor interactions, none of the latter prove
signiﬁcant, and thus the analysis is best understood by examining the direct
model.“6 Deﬁning the dependent variable as instances in which an ofﬁcer: (1) hit
or swung at the suspect with a weapon that was not a ﬁrearm; (2) used force to
make the suspect comply; or (3) used physical force that involved more than
handcufﬁng, Engel et al. (2000: 243) estimate a direct effects logistic regression
model.47 From this direct effects model, the authors ﬁnd only 5 variables to be
statistically signiﬁcant predictors of the use of force: (1) alcohol or drug use by
the suspect; (2) poor demeanor exhibited by the suspect“; (3) an increasing
number of bystanders; (4) the suspect ﬁghting with the ofﬁcers; and (5) the
seriousness of the offense.49 lmportantly, while noting that verbally disrespectful

or resistant suspects are almost 6 times more likely to have force used against

 

‘5 Note that despite the difference in sub-samples, these results correspond closely to those of Worden

(1995), who also used the PSS data. Interestingly, in a later work, Engel (2000: 276) ﬁnds that of the entire

sample of 5,179 citizens, 22% were arrested, and 9% had some force used against them.

‘6 Note also that the variables found to be signiﬁcant in the direct model remain signiﬁcant in every other

model tested; thus, considering the non-signiﬁcance of the interaction terms, nothing is gained and model

parsimony is lost in estimating anything other than the direct model. Engel et al. (2000), however, argue

that collinearity may exist within the sex/demeanor interaction term, thereby obscuring its effects.

‘7 While the authors themselves note that the assumption of independence has been violated by using

encounter-level variables in an analysis based at the individual (suspect) level, they argue that the larger

sample size diminishes any potential biases (Engel et al. 2000: 245).

‘8 The authors deﬁne a hostile demeanor as any noncompliance or verbal (to the exclusion of physical)

resistance on the part of the suspect.

‘9 Offense seriousness was broken down by the authors into a ﬁve-category variable, with 0=”no crime”,
=”minor property crime”, 2=”minor violent or major property crime”, 3=”moderate violent crime”, and

4=”major violent crime” (Engel et al. 2000: 246).

66

them than are suspects who are civil, the authors failed to include contextual (i.e.
neighborhood characteristic) variables which may have an effect both on suspect
demeanor, as well as how that demeanor is perceived by the ofﬁcer given the
characteristics of the neighborhood in which the encounter occurred.

Phillips and Smith (2000), in a multivariate analysis based on structuration
theory of time and space dynamics, looked at 217 complaint forms ﬁled by
citizens against on-duty police ofﬁcers between July 1990 and June 1994 in
Queensland, Australia.50 It is important to note, however, that these were only
allegations, with the citizen asserting that force was used improperly by the
ofﬁcer in question. However, the authors noted that they only included those
allegations which were judged to exhibit strong evidence. The criteria for such
evidence was as follows: (1) the alleged assault was referred to Queensland
Police Service Misconduct Tribunal or the Director of Prosecution; (2) formal
action was taken against the ofﬁcer(s); (3) the ofﬁcer(s) admitted to the alleged
assault; (4) ofﬁcer(s) who witnessed the alleged assault supported the account
provided by the victim; (5) citizen(s) who witnessed the assault supported the
account provided by the victim; or (6) there was physical evidence of an assault
(Phillips and Smith 2000: 482). Phillips and Smith focused their efforts on the
effects of time and space on the use of force, dividing the former into a
dichotomous “day - night” variable, while separating the latter into a three-
category “public, private or police” locale. The dependent variable was also

dichotomous, with the presence of any of the following types of force: (1) push /

 

5° Note that the original data set included 973 complaints where assault was the major allegation. Of these,
a random sample of 350 cases was selected, which was subsequently reduced to 217 cases through the
exclusion of cases where the evidence for the allegations was considered to be weak.

67

poke; (2) grab / grapple, coded as “0”, and any of the following: (3) punch; (4)
kick; or (5) baton hit, coded as “1" (Phillips and Smith 2000: 487).51 The authors
go on to estimate two separate logistic regression models, the ﬁrst with time and
space included only as separate main effects, and the second with time-space
interactions. Other than this distinction, the two models are identical with respect
to the variables included in the analysis. In both models, 17 other variables are
presented as predictors, with 9 of these achieving signiﬁcance. Speciﬁcally, the
authors ﬁnd that the following variables are positively related to the use of force
and statistically signiﬁcant: (1) the presence of 4 or more additional ofﬁcers; (2)
an increasing number of bystanders (broken down into 1 bystander, 2-3
bystanders, or 4 or more bystanders); (3) the ofﬁcer was a plainclothes ofﬁcer;
(4) all of the involved citizens were male (although this was signiﬁcant only at the
.10 level); (5) all of the involved citizens were under the age of 25; (6) the citizen
had a mental or physical disorder; (7) the citizen fought with the police ofﬁcers
(Phillips and Smith 2000: 488).” With respect to their primary hypotheses, the
authors found that there were no signiﬁcant main effects for the time and space
variables in the ﬁrst model. They did, however, ﬁnd support for their notion that
“citizens deﬁne night-time public space as a zone of fun, while police deﬁne it as
a zone of danger” (Phillips and Smith 2000: 490). Speciﬁcally, the authors note

that the night-public interaction is a statistically signiﬁcant predictor of the use of

 

5' Note here that the authors explicitly excluded the use (either unholstering or ﬁring) of ﬁrearms, as well
as that of 0C spray. The authors argue that this particular dichotomy, coupled with the operationalization
described above allowed them to both claim that force was used, and that its use was at least perceived as
imwarranted.

’2 The authors note that they excluded officer age, length of service, and ofﬁcer duties from the analysis
due to an unacceptably high number of cases with missing data for these variables.

68

force. However, it is important to note that in this instance the authors have
relaxed the standard of signiﬁcance, as the interaction is only signiﬁcant at the
.10 level. Notably, even obtaining such a result (albeit with a less stringent
requirement of signiﬁcance), Phillips and Smith (2000) fail to extend the “space”
argument to the logical conclusion of a neighborhood contextual effect on use of
force behavior.

Terrill and Mastrofski (2002) used data from an observational study
conducted during the summer of 1996 in Indianapolis, Indiana and the summer of
1997 in St. Petersburg, Florida._The Project on Policing Neighborhoods (POPN)
studied over 200 ofﬁcers for over 5,700 hours through approximately 1,000 patrol
shifts in 24 police beats encompassing 80 different neighborhoods. In an analysis
of 3,116 police-suspect encounters, Terrill and Mastrofski (2002) used a four-
category measure of force (no force, verbal force, restraint techniques, and
impact techniques) and ﬁnd that some force was used in 58.4% of all encounters
studied, with verbal force accounting for 37.4%, restraint techniques accounting
for 18.9% and impact techniques accounting for 2.1 %.53 The authors went on to
use an ordered probit analysis with almost 30 independent and control variables.
Of the latter, Terrill and Mastrofski (2002) found the following variables to be
statistically signiﬁcant predictors of increased use of force behavior: (1) the
suspect was a minority; (2) the suspect was young; (3) the suspect was of low

social class; (4) the suspect was intoxicated; (5) the suspect resisted the ofﬁcers;

 

’3 It is important to note here that Terrill and Mastrofski (2002) include “pat-downs” and handcuffmg as
use of force behaviors falling within the restraint technique category of force. This may be a signiﬁcant
reason for the resulting extremely high base rate of force, which doubles the next highest base rate found in
previous research. Reasoning for why the current research chose to exclude these behaviors will be
discussed in Chapter 3.

69

(6) the ofﬁcer had fewer years of education; (7) the ofﬁcer had fewer years of
experience; and (8) the ofﬁcer worked in Indianapolis. The last ﬁnding is
important, in that it demonstrates a contextual effect (albeit based on city, rather
than neighborhood), but Terrill and Mastrofski (2002) do not use multilevel
modeling to disentangle the effects of neighborhood level variables.54

The ﬁnal multivariate study to be discussed here is that of Garner et al.
(2002), who used a similar methodological approach to Garner et al. (1995) to
examine use of force behavior. Garner et al. (2002) collected data from every
precinct across every shift in the following six cities: (1) Charlotte, North Carolina;
(2) Colorado Springs, Colorado; (3) Dallas, Texas; (4) St. Petersburg, Florida; (5)
San Diego, California (Police Department); and (6) San Diego County, California
(Sheriff’s Department). Between the summer of 1996 and early spring 1997,
Garner et al. (2002: 721) collected data on 7,512 adult custody arrests. The
authors constructed two measures for the use of force, with the ﬁrst being a
measure of prevalence (i.e. a “force - no force” dichotomy), and the second
being a measure of severity. The former measure included the use of weapons,
weaponless tactics, and restraint techniques, while excluding threats of force,
gentle holds, weapon displays, and handcufﬁng. The latter measure was
constructed through a series of surveys given to ofﬁcers in which those officers
ranked numerous police behaviors on a scale of 1 to 100 based on their own
experiences. Garner et al. (2002: 724) reported that across all six jurisdictions,

17.1% (1,283 arrests) of arrest encounters included the use of force as deﬁned

 

5‘ In a later work, Terrill and Mastrofski (2003) estimate separate ordered probit models for Indianapolis
and St. Petersburg to examine coercion, and ﬁnd thirteen independent variables to be statistically
signiﬁcant in both models, with only four variables differing between the two models.

70

by the prevalence measure.55 With regards to the severity measure, the authors
noted that “across all 7,512 arrests, the average rank of the maximum-force
measure is 30.4” (Garner et al. 2002: 726). Noting that they were focusing on
situational characteristics, Garner et al. (2002) grouped independent variables
into ﬁve domains: (1) the type of location; (2) the nature of the offense; (3) police
mobilization; (4) the characteristics of the ofﬁcer; and (5) the characteristics of
the suspect. The authors tested the effects of these variables using three
regression models — a logistic regression analysis of the prevalence measure
without consideration of suspect resistance was Model 1, a logistic regression
analysis of the prevalence measure with suspect resistance included was Model
2, and a generalized linear regression analysis of the severity measure with
suspect resistance included was Model 3. Overall, Garner et al. (2002) found that
15 of the independent variables were predictors of increases in both the
prevalence and severity of force. These variables were as follows: (1) the ofﬁcer
was in the St. Petersburg Police Department; (2) the arrest occurred on a
weekend; (3) the suspect was not already in custody upon arrival of the ofﬁcer;
(4) the ofﬁcer was dispatched to a priority call; (5) the ofﬁcer approached using
lights and sirens; (6) an increasing number of ofﬁcers; (7) the ofﬁcer called for
backup; (8) the ofﬁcer was younger; (9) the ofﬁcer was male; (10) the ofﬁcer had
received prior injuries on-duty; (11) the suspect was male; (12) there were
bystanders of an unknown relationship to the suspect present; (13) the suspect

had a reputation for carrying weapons; (14) the suspect was antagonistic; and

 

5’ While this is the overall proportion, within sites the proportion varied from 12.7% in Colorado Springs to
22.9% in St. Petersburg.

71

(15) the suspect used physical resistance against the ofﬁcer. The latter two
ﬁndings are the most instructive, as Garner et al. (2002: 738) note that “when
suspects who display an antagonistic demeanor toward the police (but no
physical force) are compared with suspects who display a civil demeanor, the
odds of the police using physical force increase by 163% [and] when suspects
who use physical force against the police are compared with suspects who
display a civil demeanor, the odds of the police using physical force increase by
1800%.” Most importantly, given the emphasis of the current research, Garner et
al. (2002: 721-722) noted that although differences between the six sites were
tested and found to be signiﬁcant, there was no attempt to explain the cause of
this variation through the inclusion of neighborhood differences.56

As we have seen, prior research has used varying methodologies (i.e.
sampling), statistical techniques, and measures of force to study the effects of
numerous types of independent and control variables (see Table 1). Of the
multivariate studies presented here, however, there remain some consistent
effects of independent variables despite these differences. In general, those
variables found to have the most consistent effects on use of force behavior are:
(1) suspect race; (2) suspect gender; (3) suspect demeanor or hostility; (4)
suspect sobriety; (5) number of bystanders; (6) number of ofﬁcers; (7)
seriousness of the offense; and (8) suspect physical resistance. Yet, even given
the consistency of these effects, researchers have found an astoundingly wide

range of base rates for use of force behavior (see Appendix A, Table A-1). Of

 

’6 Garner et al. (2002) correctly note that multilevel modeling could not be used to discern differences
between cities due to the small sample size of only 6 cities.

72

these independent and control variables presented in Table A-1 (Appendix A), a

suspect’s demeanor and physical resistance are relatively well-documented as

having an effect. In fact, all eight of the multivariate studies discussed in this

section ﬁnd that either poor suspect demeanor (i.e. antagonism or verbal

resistance on the part of the suspect) or physical resistance by the suspect

signiﬁcantly increase the prevalence of use of force behavior by police ofﬁcers.

 

Table 1: Measures of Force, Methodologies and Sample Sizes

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Study Measure of Force Statistical Sample Size
Method
Friedrich (1980) No force — force OLS 1 ,091
(dichotomy) encounters
Bayley and No force - force 467 potentially
Garafalo (1 989) (dichotomy) OLS violent
mobilizations
Garner et al. (1995) (1) No force - force LR I OLS 1,585 ofﬁcer
(dichotomy) surveys
(2) Continuum of
force
(3) Maximum force
ranking
Worden (1995) No force I Reasonable MNL 1,528 police-
force / improper force citizen
encounters
Kavanagh (1997) No force — force LG 1,108 arrests
(dichotomy)
Engel et al. (2000) No force — force MNL 1,461 non-trafﬁc
(dichotomy) police-citizen
encounters
Phillips and Smith No force — force LR 217 complaints
(2000) (dichotomy)
Terrill and No force I verbal force/ OP 3,116 police-
Mastrofski (2002) restraint techniques I citizen
impact techniques encounters
Garner et al. (2002) (1) No force — force LR / GLM 7,512 arrests
(dichotomy)
(2) Maximum force
ranking

 

Note: OLS = Ordinary Least Squares, LR = Logistic Regression, MNL = Multinomial
Logistic Regression, LG = Logit, OP = Ordered Probit, GLM = Generalized Linear Model

73

 

 

In addition, a majority of the studies ﬁnd that an increasing number of bystanders
(Engel et al. 2000; Friedrich 1980; Garner et al. 1995; Garner et al. 2002; Phillips
and Smith 2000; Worden 1995) or a suspect in an altered state (i.e. drunk or on
drugs) (Engel et al. 2000; Friedrich 1980; Garner et al. 1995; Kavanagh 1997;
Terrill and Mastrofski 2002; and Worden 1995) also signiﬁcantly increase the
prevalence of the use of some method of force by police ofﬁcers. The effects of
the remaining variables outlined in Table A-1 (Appendix A) are less clear, with
only four studies reporting a statistically signiﬁcant effect for: (1) a violent or
felony offense (Engel et al. 2000; Garner et al. 1995; Kavanagh 1997; and
Worden 1995); (2) an increasing number of police ofﬁcers at the scene (Friedrich
1980; Garner et al. 1995; Garner et al. 2002; and Phillips and Smith 2000); and
(3) a male suspect (Garner et al. 1995; Garner et al. 2002; Phillips and Smith
2000; and Worden 1995). The evidence for a race effect is even less clear, with
only three studies reporting a statistically signiﬁcant increase in force behavior
when encountering a minority suspect (Kavanagh 1997 [only for Hispanics];
Terrill and Mastrofski 2002; and Worden 1995). It is important to note here,
however, that these effects were obtained by estimating individual-level models
with encounter-level variables included. In the next section we will discuss two
research efforts which have made speciﬁc attempts to study the effects of

neighborhood variables in a neighborhood—level model.

74

Use of Force Research and Neighborhood Context:

As discussed previously, while prior research has focused on numerous
forms of use of force behavior using a variety of independent variables, the
overwhelming majority of these studies have focused on individual
characteristics, or inappropriately placed encounter-level characteristics in an
individual-level model. The following two studies are the only research efforts
identiﬁed in this literature review as having used neighborhood-level models to
examine use of force behavior by police ofﬁcers. Clearly, then, these studies are
the closest in design and analysis to the current research proposed here.

The ﬁrst of these studies was conducted by Smith (1986) using the Police
Services Study data (Ostrom et al. 1977), which collected data on over 450
ofﬁcers during 900 patrol shifts in 60 different neighborhoods in the cities of
Rochester, New York, St. Louis, Missouri, and Tampa / St. Petersburg, Florida
during the summer of 1977. Of the 5,688 police-citizen encounters originally
observed, Smith (1986) limited his study to 762 encounters with “non-dangerous”
(i.e. unarmed) suspects. Focusing on neighborhood context as the primary
orientation to the study, the author deﬁned "coercive authority” as any situation in
which “police use[d] force or threatenled] a suspect with arrest, surveillance, or
physical harm” (Smith 1986: 318), and found a base rate of 30% for the use of
coercion in all encounters with non-dangerous suspects. Smith (1986: 314) went
on to deﬁne “neighborhoods” as “small residential areas within cities that are
deﬁned on the basis of police beats, census block groups, or enumeration

districts.” However, while Smith (1986) initially estimated separate neighborhood-

75

level and encounter-level models, the main model tested included neighborhood-
level variables in an individual-level model. Using a maximum-likelihood probit
analysis, Smith (1986) found the following variables to be predictive of an
increased use of force: (1) the neighborhood had a higher proportion of minority
residents; (2) the suspect was antagonistic; (3) the suspect was male; and (4) the
encounter took place in a private setting. Most importantly, in terms of its relation
to the current research, Smith’s (1986) work found that in the neighborhood-level
analysis, suspects encountered by the police in lower-class neighborhoods were
three times more likely to be arrested (although not to have force used against
them) than those encountered in high-status neighborhoods, controlling for
seriousness of the crime, race of the suspect, and the suspect’s demeanor.
Although this result did not translate to use of force behavior, Smith (1986)
correctly noted that social context potentially had been masking the effects of
certain variables in previous research, while amplifying the effects of others.
Thus, where previous research has found an effect for suspect race, it is possible
that this relationship was spurious, with social context providing the real effect.
Similarly, where research has found a null effect for demeanor, it is possible that
context, rather than demeanor, drove these results. For example, an ofﬁcer might
be more willing to ignore the negative demeanor of a suspect encountered in a
lower-class neighborhood in order to allow the suspect to “save face” in front of
other citizens, thereby hopefully ensuring cooperation from that suspect in the
future. Alternatively, an ofﬁcer might ignore a display of negative demeanor from

a suspect encountered in an upper-class neighborhood due to the desire to avoid

76

a lawsuit alleging harassment on the part of the ofﬁcer. Yet, while the results of
the neighborhood-level model are interesting, and Smith (1986) argues an
intelligent point regarding the possibility of context either falsely obscuring or
amplifying relationships between other variables and the use of force, the lack of
a multi-level model (i.e. one using multi-level modeling techniques such as
hierarchical linear modeling) again causes one to question the validity of these
results.

The second, and only other, study to examine the inﬂuence of
neighborhood context on use of force behavior was that of Terrill and Reisig
(2003), who used data from the Project on Policing Neighborhoods (POPN)
conducted during the summer of 1996 in Indianapolis, Indiana and the summer of
1997 in St. Petersburg, Florida. Noting that the overall data sample was reduced
to 3,330 police-suspect encounters in 80 neighborhoods”, Terrill and Reisig
(2003: 300) deﬁned force as any of the following: (1) verbal (issuing commands
or threats; (2) physical restraint (pat downs, ﬁrm grip, handcufﬁngsa; and (3)
impact methods (pain compliance techniques, takedown maneuvers, strikes with
the body, and strikes with external mechanisms. Thus, force was measured as a

four-category construct, with 1=no force, 2=verbal force, 3=physical restraints,

 

’7 The authors note that of over 11,000 encounter / activities observed, only 3,544 involved a suspect. Of
the latter, 136 were eliminated because they could not be geocoded (and thus could not be assigned to a
neighborhood), and 78 were eliminated because they occurred in neighborhoods with too few cases overall,
leading to the exclusion of those neighborhoods (Terrill and Reisig 2003).

’8 The authors note that their deﬁnition of force is based on the National Academy of Science’s definition
of violence as “acts that threaten or inﬂict physical harm” on suspects (Terrill and Reisig 2003: 299).
Accordingly, they argue that the appropriate measures of force are not the same for studies of excessive
force as they are for studies of the prevalence of any type of force, and thus they include handcufﬁng as
falling under the NAS deﬁnition. Yet, clearly the mere act of handcufﬁng does not necessarily either
threaten or inﬂict physical harm, and thus it remains unclear why the authors chose to include it in their
measure. This will be further explored in Chapter 3.

77

and 4=impact methods, and the authors found a base rate of 58% for use of
force behavior (37% verbal, 19% restraint techniques, and 2% impact
techniques) (Terrill and Reisig 2003: 300). Noting that “the research reported
here examines the degree to which forceful authority toward suspects is
inﬂuenced by the type of neighborhood in which police-suspect encounters
occur", Terrill and Reisig (2003: 292) included two neighborhood-level variables —
concentrated disadvantage, a construct used in previous research (Sampson et
al. 1997), and the homicide rate.59 Terrill and Reisig (2003) also included a wide
variety of encounter-level variables, such as suspect characteristics, ofﬁcer
characteristics, suspect behavior (i.e. demeanor, physical aggression, sobriety),
and control variables used in the majority of previous use of force research. The
authors also noted that ﬁve of the predictor variables included in the model
(suspect disrespect, suspect resistance, the occurrence of an arrest, the
presence of a weapon, and conﬂict between citizens) were time-dependent, and
thus were coded according to when they occurred during the encounter in order
to ensure causal order. Terrill and Reisig (2003) then presented three separate
regression equations to examine neighborhood effects on force in a variety of
ways. In the ﬁrst model, a weighted least squares regression analysis found that
increased levels of force were used against suspects encountered in
neighborhoods that were high in concentrated disadvantage, as well as those
encountered in neighborhoods that had high homicide rates. Noting that this

model did not include encounter-level variables, the latter are added in Model 2,

 

’9 The authors note that they use the homicide rate as a measure of crime due to the fact that “it is
considered by criminologists to be the most reliable measure of crime that is least sensitive to
lmderreporting” (Terrill and Reisig 2003: 301).

78

which was tested using ordinary least squares regression analysis. These
analyses found that increased levels of force were used against suspects who
were: (1) male; (2) minorities; (3) younger; and (4) lower-class. Noting that the
results of the ﬁrst two models “support[ed] [the] contention that a true test
concerning the effects of neighborhood context on level of force should
include...encounter-level variables, such as suspect sociodemographic
characteristics, to control for within-neighborhood variance”, Terrill and Reisig
(2003: 303) then estimated a ﬁxed-effects hierarchical linear regression model.
Overall, the authors noted that the effects observed in Model 1 remain in Model
3, ﬁnding that increased levels of force are used against suspects encountered in
neighborhoods high in concentrated disadvantage, as well as those encountered
in neighborhoods with high homicide rates, independent of suspect and ofﬁcer
characteristics, or of suspect behavior.60 lmportantly, Terrill and Reisig (2003:
306) found that using the more appropriate multilevel modeling approach to
estimate the separate effects of neighborhood and encounter-level variables
reduced the effect of race on the use of force. That is, the authors noted that the
signiﬁcant effect of race found in Model 2 had, in reality, been confounded by the
effects of neighborhood context, as the effects of race were reduced below levels
of signiﬁcance in Model 3. Terrill and Reisig (2003: 306) also noted that when
neighborhood effects were estimated in the multilevel model, the effects of sex,
age and class persisted, with males, younger individuals and lower-class

individuals more likely to have force used against them. It is clear, then, that the

 

6° In later research, Reisig et al. (2004), using the same POPN data, note that: (1) elevated levels of police
force can induce suspect disrespect, which in turn can lead to increased levels of force; and (2) suspects in
disadvantaged neighborhoods are more likely to show suspect disrespect.

79

 

results of the work conducted by Terrill and Reisig (2003) provide a ﬁrm
foundation for the current work. The latter authors appropriately included both
encounter-level and neighborhood-level variables in a hierarchical linear
regression model in order to estimate the true effects of these variables given the

effects of neighborhood context on use of force behavior by police ofﬁcers.

Limitations of Prior Studies

While the studies discussed in this chapter have provided the ﬁeld of
criminal justice with a vast array of useful ﬁndings, they remain incomplete, both
theoretically and methodologically. In order to illustrate the value of the current
proposal, then, it is important to review these limitations. This allows us to identify
shortcomings of prior research (although still addressing their value) while
emphasizing the unique contributions of the current work.

As has been mentioned within the current work, previous research efforts
have tended to focus either on only one theoretical orientation (i.e. psychological,
sociological, organizational) in examining use of force behavior, ignoring
neighborhood context, or on neighborhood context, ignoring its potential effects
on use of force behavior. From a theoretical standpoint, then, only the efforts of
Smith (1986), and Terrill and Reisig (2003) have combined an emphasis on
neighborhood context with research on use of force behavior. Although many
scholars have made contributions in these areas, it is the latter two works which
stand out as the closest in theory to the current work. Yet, even these works are

lacking in certain aspects of theoretical development. Smith (1986: 314), for

80

example, deﬁned neighborhoods as areas similar to police beats or to census
block groups. This presents a confusing picture of the nature of neighborhood
context, as police beats are substantially dissimilar from census block groups,
and thus both types of areas cannot be considered “neighborhoods” (see the
previous discussion in this chapter for an explanation of how prior research has
operationalized neighborhoods). Terrill and Reisig (2003), on the other hand,
introduce a source of error in their theoretical understanding of use of force
behavior. The authors choose to include ‘pat-downs’ and handcufﬁng as
instances of the use of force within their study. This is puzzling, considering that
the International Association of Chiefs of Police, as well as the National Institute
of Justice, speciﬁcally exclude handcufﬁng from use of force studies (IACP 2001;
Greenfeld et al. 1997; Langan et al. 2001).

Perhaps more important than vague or puzzling theoretical orientation is
the presence of methodological shortcomings in the prior literature. As has also
been discussed previously, the majority of previous research efforts have used
multivariate (occasionally only bivariate) models to examine neighborhood-level
effects. This results in numerous difﬁculties ranging from biased coefﬁcients to
incorrect conclusions based on mis-speciﬁc models. Smith (1986), for example,
while detailing an explicit model of the effects of neighborhood context, fails to
use hierarchical linear modeling techniques. Terrill and Reisig (2003), in the sole
use of force study which does use HLM, fail in two aspects: (1) the authors
estimate a ﬁxed-effects model, which does not allow individual-level

characteristics to vary across the second level of the model; and (2) they fail to

81

 

 

 

 

_ _.

 

account for spatial autocorrelation. Prior research has also succumbed to a
confusing multitude of issues concerning the validity of measures. As noted,
Terrill and Reisig (2003) present two concerns regarding validity, in their
deﬁnition of force (with its inclusion of handcufﬁng), as well as in their choice of
police beats as a proxy for neighborhoods. In addition, Terrill and Reisig (2003)
use demographic data from the 1990 Census, a choice which is unusual in that
the data in the POPN study was collected during 1996 and 1997 (thus making
the 2000 Census more appropriate). Smith (1986), in addition to the confusion
regarding the operationalization of neighborhood, suffers from an extrapolation of
a random sample of neighborhood residents (interviewed by phone) to form
aggregate demographic statistics for each neighborhood studied.

The last methodological critique to be made of previous studies centers
around the selection criteria, both for the area of the study, as well as for the
individuals observed. Smith (1986) noted that the sample was limited only to
those suspects who were unarmed. While perhaps easier relative to sampling the
entire arrestee population, this nevertheless limits the study, particularly when
one considers that the presence of a weapon can have a signiﬁcant impact on
ofﬁcer behavior (Bayley and Garafalo 1989; Garner et al. 1995; Garner et al.
2002). The work of Terrill and Reisig (2003) suffers from two sampling issues
which further call into question the validity of their results. The ﬁrst issue stems
from the sampling criteria for suspects, as the latter were deﬁned as individuals
who: (1) were identiﬁed as suspects by the police; (2) were interrogated by the

police; (3) were searched by the police; (4) were threatened or warned by the

82

police; (5) were arrested or cited by the police; (6) admitted they were
wrongdoers; or (7) had force used against them to prevent or stop criminal
activity (Terrill and Reisig 2003: 298). As the authors themselves correctly note,
the latter deﬁnition is, in effect, sampling on the dependent variable, which
introduces a source of bias. In addition, the Project on Policing Neighborhoods
(POPN) Study, the data from which are used by Terrill and Reisig (2003)
purposefully over-sampled disadvantaged neighborhoods in order to ensure that
an adequate amount of data on police-citizen encounters was collected. While
this sampling method did indeed provide Terrill and Reisig (2003) with a
signiﬁcant amount of data, the fact that none of the most afﬂuent neighborhoods
in both cities were included in the sampling design limits the validity and

generalizability of their results.

Variables of Interest from Prior Studies

While a more detailed presentation of the variables to be included in the
current research will be presented in Chapter 3, it is useful to clearly delineate
the reasons for the inclusion of these variables. In particular, speciﬁc prior
research efforts (which have been discussed throughout Chapter 2) will be
summarized with attention to the theoretical orientation of their independent
variables. This allows the current work to present a theoretically-driven model of
dependent and independent variables within a multilevel modeling framework.

In terms of the dependent variable, the current research will focus on two

separate measures, one of prevalence and one of severity, in order to capture

83

the full range of use of force behaviors. This approach is more consistent with
recent research on use of force, which has acknowledged the need for multiple
dependent variables (Garner et al. 1995; Garner et al. 2002). With respect to the
independent variables, those used here fall within one of three categories: (1)
those based on social disorganization theory; (2) those based on Black’s (1976)
theory of social control; and (3) those introduced as control variables.

The ﬁrst group of independent variables to be examined is that which falls
under the domain of social disorganization theory. The primary variable of
interest here is concentrated disadvantage, which a variety of researchers have
used as a measure of social disorganization (Kane 2002; Morenoff et al. 2001;
Reisig and Parks 2000; Sampson and Bartusch 1998; Sampson et al. 2002).
Social disorganization theory, however, also discusses the ability of
neighborhood residents to organize themselves. In this regard, variables
concerning citizen organization are also introduced. For example, Greenleaf and
Lanza-Kaduce (1995) propose that the latter concept can be measured by: (1)
the relationship between the victim and the suspect; (2) the number of suspects;
and (3) the presence of bystanders. While including all three of these variables,
other researchers have identiﬁed other measures of organization, namely: (1) the
number of bystanders (Engel et al. 2000; Garner et al. 1995; Worden 1995); (2)
the demeanor of bystanders (Garner et al. 1995); (3) the number of ofﬁcers
(Friedrich 1980; Garner et al. 2002; Phillips and Smith 2000)“; and (4) the

relationship between the suspect and bystanders (Garner et al. 2002).

 

6‘ In later research, McCluskey-er al. (2005) noted that peer groups’ (i.e. other ofﬁcers’) attitudes toward
aggressiveness inﬂuenced use of force. Speciﬁcally, the authors noted that encounters involving “low-

84

Black’s (1976) theory of social control also provides a theoretical
foundation, both for prior research as well as the current work. Of primary
importance to the current study is the concept of social distance, with Black
(1976; 1980) arguing that ofﬁcers are more likely to be penal (i.e. coercive)
towards suspects who are located at the bottom of the scale in terms of social
distance. Thus, the current work examines the effects of suspect characteristics
on police behavior, as Black (1976) hypothesizes that males, younger
individuals, and minority individuals are located at the bottom of the social
structure. Numerous researchers have examined these issues, alternatively
focusing on one or more of the following: (1) gender (Phillips and Smith 2000;
Smith 1986); (2) age (Garner et al. 2002; Terrill and Reisig 2003); and (3) race
(Garner et al. 2002; Terrill and Mastrofski 2002; Worden 1995). However, social
control also hypothesizes that ofﬁcers will be more penal towards those
individuals who are farther away in relational distance from the ofﬁcers
themselves, and thus research has also focused on ofﬁcer characteristics,
notably: (1) gender (Garner at 1995); (2) age (Garner et al. 2002); and (3) race
(Kavanagh 1997). In addition to individual characteristics, there are a number of
other variables which can locate either suspects or ofﬁcers within the social
space. Greenleaf and Lanza-Kaduce (1995), for example, argue that the
sophistication of a suspect can be measured by: (1) the presence of conﬂict
when the ofﬁcer arrives; and (2) the suspect’s level of intoxication. The latter has

been found to be a signiﬁcant predictor of use of force behavior in several studies

 

aggressive patrol peer group ofﬁcers” were less likely to result in force when confronted with a
disrespectful suspect than encounters involving “high-aggressive patrol peer group ofﬁcers”.

85

(Engel et al. 2002; Friedrich 1980; Kavanagh 1997; Terrill and Mastrofski 2002).
Greenleaf and Lanza-Kaduce (1995) also note that a suspect’s sophistication
stems from issues of visibility and approach, as well as the seriousness of the
offense. With respect to visibility, Smith (1986) and Garner et al. (2002) have
tested the effects of actual visibility (i.e. the ability to perceive given the
surrounding lighting during an encounter), as well as of a public-private space
dichotomy (i.e. did the encounter occur within a suspect’s home). The concept of
approach is designed to explain how an ofﬁcer’s introduction to the situation may
affect how the ofﬁcer locates the suspect in the social space. Garner et al.
(1995), as well as Garner et al. (2002) focus on the distinction between proactive
and reactive encounters, as well as emphasizing how the ofﬁcer actually
proceeded to an encounter (eg. with lights and sirens on). Offense seriousness,
although somewhat discounted by Black (1976), has also emerged as a
signiﬁcant predictor of use of force behavior (Engel et al. 2002; Garner et al.
1995; Worden 1995). A ﬁnal process by which suspects are located within the
social space is through their behavior, and thus numerous research efforts have
focused on suspect presentation. The effects of poor suspect demeanor and
physical resistance on the part of the suspect, for example, are well-documented
(Bayley and Garafalo 1989; Engel et al. 2002; Friedrich 1980; Garner et al. 1995;
Garner et al. 2002; Kavanagh 1997; Phillips and Smith 2000; Smith 1986; Terrill
and Mastrofski 2002; Terrill and Reisig 2003; Worden 1995). However, suspect
presentation may also include the following: (1) whether the suspect is known to

carry a weapon; (2) whether the suspect is known to be assaultive; and (3)

86

 

whether the suspect is a known gang member (Garner et al. 1995; Garner et al.
2002).

The ﬁnal group of variables to be examined is that of the control variables,
often introduced by researchers into regression equations as ‘common sense’
variables. With respect to issues surrounding use of force behavior by police
ofﬁcers, many of these control variables have focused on the potential
dangerousness of speciﬁc areas or speciﬁc situations. For example, Garner et al.
(1995), as well as Garner et al. (2002), examine the effects of: (1) known
hazardous locations; (2) known criminal locations; (3) whether the offense
occurred on a weekend; and (4) whether the suspect was already in custody
when the ofﬁcer arrived on the scene. All four of these variables represent some
measure of opportunity for conﬂict, either by the potential consequences of
entering a previously violent location or by the sheer numbers of individuals
encountered by the ofﬁcer. The crime rate has been examined in even greater
detail, although its effects have been split among researchers studying the
homicide rate (Sampson and Raudenbush 1999; Terrill and Reisig 2003) and
those studying the violent crime rate (Sampson et al. 1997). The role of
jurisdiction has also been examined to determine any city-speciﬁc effects on use
of force behavior (Garner et al. 2002; Terrill and Mastrofski 2002; Terrill and
Reisig 2003). Garner et al. (2002) also focused on variables related to speciﬁc
ofﬁcers and their potential for conﬂict by examining: (1) prior medical attention to
the ofﬁcer; (2) the number of arrests made by the ofﬁcer; (3) the ofﬁcer’s duty

status (i.e. on-duty or off-duty at the time of the encounter); and (4) the ofﬁcer’s

87

 

use of back-up. The remaining control variable to be used in the current work is
that of ofﬁcer demeanor, which is meant to capture an ofﬁcer’s general attitude
toward the suspect. Previous research has found that an ofﬁcer’s demeanor
toward suspects, particularly in the form of ‘face-saving’ behavior on the part of
the ofﬁcer, can have a signiﬁcant effect on other aspects of ofﬁcer behavior (Fyfe
1997; Herbert 1996; Skolnick and Fyfe 1993), including the use of force

(Kavanagh 1997).

88

Summary

With only a few exceptions, the literature examined here has used only
one of the available perspectives to analyze the causes of police behavior. Even
more importantly, much of the latter research has either ignored or mis-speciﬁed
the inclusion of contextual variables, particularly within use of force studies. This
rigid adherence to a particular theoretical orientation, or the lack of such an
orientation at all, has resulted in a confusing picture of the factors affecting use of
force behavior by police ofﬁcers. By their lack of estimating neighborhood effects,
prior research has effectively closed a viable pathway of knowledge in this
important criminal justice arena. The limitations of the work examined in this
literature review are signiﬁcant enough that they must be dealt with in an
intelligent, empirically sound manner. The current research seeks to do so by
presenting an integrative theoretical model, using Black’s ideas regarding social
distance, as well as social disorganization variables, within a hierarchical linear
model in order to estimate the true effects of neighborhood context on use of

force behavior.

89

Chapter 3: Data and Methods

The current research proposes the theories of social control (as deﬁned by
Black 1976) and social disorganization as the appropriate frameworks for
analyzing police use of force behavior. Within these theoretical frameworks, an
emphasis is placed here on the effects of neighborhood context on that behavior,
in the hopes of establishing a more comprehensive model of use of force activity
by police ofﬁcers. As indicated at the beginning of the current work, this research
will not look to differentiate between reasonable and unreasonable (or excessive)
force in terms of the legality of those distinct behaviors. Moreover, as has been
noted throughout, this dissertation focuses only on use of force behavior, to the
exclusion of other police activities such as trafﬁc stops, arrests, or ﬁeld
interrogations. Rather, the purpose of this study is to test for the various factors
which affect police use of force behavior within an arrest encounter. After
determining the validity of using hierarchical linear modeling techniques, the
following research questions will be addressed: (1) Do encounter-level factors
affect use of force behavior?; (2) Do neighborhood-level factors affect use of
force behavior?; and (3) do the effects of encounter-level factors vary across
neighborhoods?‘52

This research represents an important extension of the previous literature
by analyzing the relationship between encounter-level and neighborhood-level
variables in determining use of force behavior. As noted previously, only two
studies (Smith 1986; Terrill and Reisig 2003) were found which speciﬁcally

examined neighborhood-level models and use of force behavior, with only one of

 

‘2 Speciﬁc factors and directions of expected effects will be discussed later within this Chapter.

90

those (Terrill and Reisig 2003) estimating an HLM equation. Yet, as previous
research has also demonstrated, it is important to consider these relationships,
as prior studies have generally found that neighborhood-level variables can have
signiﬁcant effects on the behavior of police ofﬁcers. Thus, given the importance
of the latter ﬁndings, coupled with the lack of comprehensive, methodologically-
sound studies examining this issue, the current work seeks to make an important

contribution to the literature on police use of force.

The Police Use of Force (PUF) Study Data

The following section will provide a detailed description of the data
collection procedures used during the Police Use of Force Study (hereafter
referred to as PUF), as well as a discussion of the characteristics of the data
sample. This introduction to the data will provide an understanding of the

dependent and independent variables to be used in the current study.

Data Collection Procedures:

After contacting 24 sites to determine interest in participation, PUF began
data collection in the six jurisdictions which were able to participate: (1) Charlotte,
North Carolina; (2) Colorado Springs, Colorado; (3) Dallas, Texas; (4) St.
Petersburg, Florida; (5) City of San Diego, California (Police Department); and
(6) San Diego County, California (Sheriff’s Department). Table 2 presents the
relevant characteristics of each jurisdiction and their corresponding law

enforcement agencies.

91

 

Table 2: Characteristics of Six Jurisdictions

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Charlotte Colorado Dallas St. San San
Springs Petersburg Diego Diego
PD SO
US Census
2000
Total 540,828 360,890 1,188, 248,232 1,223, 2,813,
population 580 400 833
Percent 32.7 6.56 25.91 22.36 7.86 5.63
Black
Percent <18 26.02 27.91 29.55 22.66 25.55 27.17
Percent on 2.58 2.75 2.93 3.27 3.95 3.57
public
assistance
Percent 3.96 3.09 4.33 3.21 3.79 3.61
unemployed
Percent 10.62 8.70 17.78 13.26 14.60 12.43
poverty
Percent 13.17 10.14 14.67 13.54 11.07 11.35
female-
headed
households
Square 242.3 185.7 342.5 59.6 324.3 4,204
Miles
UCR 1997
Violent 1 ,452 229 1 .404 589 624 295
crime rate
per 100,000
Index Crime 9,231 5,824 9,892 7,430 4,782 2,104
rate per
100,000
LEMAS
1997
Number of 1,286 528 2,817 511 1,964 1,861
sworn
ofﬁcers
Ofﬁcers per 237.7 146.27 236.92 206.05 160.59 66.16
100,000
citizens
Ofﬁcers per 5.31 2.84 8.22 8.57 6.06 .44
sgiare mile
Hours of 667 720 1,186 720 928 718
training

 

 

 

 

 

 

 

92

 

While the Uniform Crime Reports (UCR) and Law Enforcement Management and
Administration Statistics (LEMAS) were taken from 1997 to correspond to the
data collection timeframe, Census data was taken from the 2000 Decennial
Census in order to provide a closer estimate than that provided by the data from
the 1990 Decennial Census.

During meetings with police administrators in each of the six jurisdictions,
Garner et al. (1995) developed two-page surveys63 for arresting ofﬁcers to be
completed after every arrest.‘54 Focusing on arrests as the unit of observation, the
researchers noted that this design maximized their ability to obtain a large
number of representative incidents in which force could potentially be used. This
has two implications for the current study: (1) a large proportion of all adult
custody arrests were sampled within each jurisdiction, ensuring that the samples
are representative of all adult arrests made in these jurisdictions; and (2) any use
of force which did not result in an arrest of the suspect is not captured within this
data.

Estimating that a sample of between 900 and 1,200 arrests was needed
from each jurisdiction to obtain reliable estimates of use of force behavior,
Garner et al. (1995) sampled arrests over the summer, fall and winter of 1996

through 1997. Data collection began in Colorado Springs in August of 1996, and

 

‘53 While most criticisms of survey research focus on the issue of response desirability (i.e. an ofﬁcer
responding that force was not used in order to appear to be a ‘ good cop’), it is unlikely that this was the
case, as Garner et a1 (1995) still found that some physical force was used in 22% of all arrests. In addition,
concerns about telescoping (the process of responding with an event that had occurred outside of the study
time frame) were alleviated by having the officers ﬁll out the survey almost immediately after having made
an arrest.

6‘ Note that these surveys included identifying information about the arrest incident, including the
identiﬁcation of the arresting ofﬁcer, and thus were not anonymous. However, the ofﬁcers were informed
that completed surveys were conﬁdential research materials protected from legal proceedings (i.e.
subpoena) under 42 U.S.C. §3789(g).

93

ended in Charlotte in February 1997. Overall, 7,512 usable surveys were
completed, taking just two weeks to capture data on 1,192 arrests in Dallas, and
almost two months to capture data on 1,249 arrests in Colorado Springs. After
determining that there were no signiﬁcant differences between arrests in which a
survey was completed and arrests without such a survey, the authors concluded
that the large size and representative nature of the sample provided a ﬁrm
foundation for an analysis of use of force behavior (Garner et al. 1995: 29).

Based on the surveys completed by arresting ofﬁcers, Garner et al. (1995)
identiﬁed speciﬁc characteristics of ﬁve variable domains: (1) Nature of the
Offense; (2) Location of the Arrest; (3) Police Mobilization; (4) Characteristics of
the Ofﬁcer; and (5) Characteristics of the Suspect. Using information from the
surveys, Garner et al. (1995) also identiﬁed ﬁve elements of force (weapons,
tactics, restraints, motion, and voice), and used these to construct four measures
of force: (1) physical force; (2) physical force plus threats; (3) the continuum of
force; and (4) maximum force.

The simple physical force measure was a dichotomy of whether or not
force was used at some point during the arrest, and included any use of a
weapon or weaponless tactic, including the use of severe restraints (e.g. prone
cufﬁng, hobbles, body cufﬁng, leg cufﬁng).65 While this measure is useful, Garner
et al. (1995) noted that many measures of criminal and police behavior, including
the F Bl’s Uniform Crime Reports, include threats of violence as an indication of

violence. Accordingly, the second measure (physical force plus threats) is a more

 

‘5 Note that a corresponding dichotomous measure of suspect force included any of the following as an
instance of force: (1) use of any weaponless tactic; and (2) use, threatened use, or possession of a weapon.

94

comprehensive one, and thus more suitable for the current analyses.“ This
measure includes all of those contained within the simple physical force
measure, as well as the addition of any display or threatened use of a weapon.
Many researchers, however, have argued that dichotomous measures have
several disadvantages, most notably that they assign the same relative
importance to the use of restraints as they do to the discharge of an ofﬁcer’s
ﬁrearm, with both being assigned a ‘1’ as indicative that some physical force was
used. Noting this, Gamer et al. (1995) also developed the third measure,
continuum of force, to address these issues. The latter measure more clearly
delineates the ranking of several different types of force, as deﬁned by the six
participating jurisdictions.67 As is evident in Table 3, these rankings vary between
the six jurisdictions, and thus it is not acceptable to combine cases from each
jurisdiction into one overall measure of police use of force on the continuum.
Notably, the continuum of force across all six jurisdictions ranges from a low of
ﬁve force options in Dallas and the City of San Diego to a high of nine force
options in St. Petersburg. The latter includes distinctions between restraint
techniques, takedowns and counterrnoves, while the former two agencies move
from a minimal control mechanism to the use of intermediate weapons. Clearly,
then, there are signiﬁcant differences in the measures available to ofﬁcers
through their department’s force continuum. In addition, as Garner et al,. (1995)

note, the primary disadvantage of the force continuum is that it implies that the

 

‘6 It is important to note that Garner et al ( l 995) eliminated the mere possession of a weapon as a threat,
due in large part to the fact that in the jurisdictions studied it was often legal for an arrested suspect to have
possessed a weapon, including concealed ﬁrearms.

7 Each arrest was coded only for the highest amount of force used within the continuum by both the
suspect and the ofﬁcer.

95

 

difference in force moving from ‘Ofﬁcer Presence’ to ‘Verbal Control’ is the same
as the difference in force moving from ‘lmpact Weapons’ to ‘Lethal Force’ (in the

case of Colorado Springs).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 3: The Force Continuum Across Jurisdictions
Charlotte Colorado Dallas St. San San
Springs Petersburg Diego Diego
PD SO
Ofﬁcer 1 1 1 1 1 1
Presence
Verbal 2 2 2 2 2 2
Direction l
Control
Soft Control 3 3 3 3
Chemical 4 4
Measures
Control and 4 3 3
Compliance /
Restraint
Transporter 4
Take Downs 5
Hard Control 5 5 6 5
l Pain
Compliance
Counterrnove 7
Intermediate 6 6 4 8 4 6
llmpact
Weapons
Lethal Force 7 7 5 9 5 7

 

 

 

 

 

 

 

In response to this, a fourth measure, that of maximum force, was created on a
scale of 1-100, with ‘1’ being the least forceful action and ‘100’ being the most
forceful action. This scale was created by asking 503 ofﬁcers to rank hypothetical
force behaviors on a scale from 1-100 based on their own personal experience.

The resulting scale had face validity, as ofﬁcer presence and the use of verbal

96

commands were located near the bottom of the scale, while the use of weapons
(particularly ﬁrearms) was located near the top of the scale. After determining if
each of these speciﬁc behaviors occurred within the sample of 7,512 arrests, any
behaviors present in the sample were then weighted according to the rankings of

the ofﬁcer-created scale.

Sample Characteristics:

As noted previously, the sample used in this dissertation is police-citizen
arrest encounters in the six jurisdictions. More speciﬁcally, information was
collected from ofﬁcer self-reports on all adult custody arrests, leading to a sample
size of 7,512 usable surveys representing every precinct and every shift from
within the six jurisdictions studied. Thus, instances in which ofﬁcers used force
against a suspect but did not subsequently arrest that suspect are not reported
here. However, the large sample size, as well as the practitioner-grounded
nature of the dependent variables, ensures that a representative range of
behaviors and encounters are captured within this research. In addition, it is
important to note that prior research has found that instances of the use of force
outside of arrest situations are an extremely rare phenomenon (IACP 2001).
This, taken with the fact that the current research seeks only to explain police
behavior during arrest situations, leads us to be conﬁdent in the contributions of
the current work.

Within the six jurisdictions, many ofﬁcers were responsible for multiple

arrests over the course of the data collection period. Thus, in St. Petersburg, 278

97

ofﬁcers returned a usable survey after a total of 1,547 arrests (the most arrests
made in a jurisdiction during collection), while in the city of San Diego, 466
ofﬁcers returned a usable survey after a total of 947 arrests (the fewest arrests
made in a jurisdiction during data collection). The numbers for the remaining four
jurisdictions are as follows: (1) in Charlotte, 371 ofﬁcers and 1,314 arrests; (2) in
Colorado Springs, 289 ofﬁcers and 1,290 arrests; (3) in Dallas, 626 ofﬁcers and
1,456 arrests; and (4) in San Diego County, 314 ofﬁcers68 and 958 arrests.

An additional aspect of the data sample deserves consideration here, and
that is the issue of deﬁning use of force behavior. As discussed previously, the
current work uses data collected from the PUF Study, during which the
researchers made an explicit decision to exclude handcufﬁng behaviors as an
example of the use of force. While there has been considerable debate over this
topic in recent years, the research efforts described here follow the lead of
Garner et al. (1995; 2002) in excluding handcufﬁng. This exclusion is due to a
variety of reasons. Adams (1997: 3), for example, notes that “broad deﬁnitions of
use of force, such as those that include grabbing or handcufﬁng a suspect, will
produce higher rates than more conservative deﬁnitions.” The author goes on to
note that “the BJS pretest of the 1996 Police-Public Contact Survey found close
to 500,000 people subjected to the threat or use of force [but] when handcufﬁng
was included, the number was close to 1.2 million” (Adams 1997: 3). This effect
is notable when one considers the multivariate studies on use of force behavior

described in the previous chapter. Using data from the POPN study, Terrill and

 

‘8 Although law enforcement agents employed by the San Diego County Sheriff‘s Department are legally
“Deputies”, the current research refers to all law enforcement agents in the six jurisdictions as “officers” for
the sake of semantic simplicity.

98

 

Mastrofski (2002), as well as Terrill and Reisig (2003), ﬁnd a base rate of force of
58% when including handcufﬁng in their deﬁnitions. In both instances, the
researchers state that they have followed the National Academy of Science
deﬁnition of violence as “acts that threaten or inﬂict harm” on suspects (Terrill
and Reisig 2003: 299), and they argue that handcufﬁng falls under the NAS
deﬁnition. Yet, it is unclear that simply handcufﬁng a suspect threatens or inﬂicts
physical harm on that individual. While verbal commands are rightly considered a
use of force in that there is an understanding that physical force can be used if
the command is not followed, and thus threatening to handcuff someone may
constitute use of force behavior, the act of handcufﬁng in and of itself neither
threatens nor inﬂicts bodily harm to the suspect (except, potentially, in cases
where handcuffs are applied too tightly). The IACP (2001: 20) agrees, speciﬁcally
noting that handcufﬁng during transport or questioning is not included in its
measures of force, although it also fails to include verbal commands. Langan et
al. (2001: 2) also exclude handcufﬁng behaviors, while including verbal
commands, grabbing, and bites from a K-9 ofﬁcer. While Terrill and Mastrofski
(2002: 231, fn15) argue that the inclusion of an ‘arrest’ variable allows them to
control for the effects of mandatory handcufﬁng when taking a suspect into
custody, it is unclear that this provides a true estimate of handcufﬁng as force
behavior, particularly when one considers that the authors themselves
acknowledge that handcufﬁng did not occur during every arrest, and that not
every act of handcufﬁng was followed by an arrest. Thus, absent data regarding

the actual number of arrests without handcufﬁng and handcufﬁng without arrest,

99

it is impossible to disentangle these effects. This is a serious problem when one
considers that Terrill and Mastrofski (2002) ﬁnd that the use of restraints (a
category which included handcufﬁng) occurred in 18.9% of all police encounters

with suspects, leading to an overall base rate of force of 58%.

The Operationalization of Neighborhood:

As has been discussed throughout this dissertation, the concept of
neighborhood is one that has varied in deﬁnition across numerous studies. In an
analysis of forty peer-reviewed studies that were published in the last half of the
1990s, Sampson et al. (2002: 457) found that there was “very little consistency
across studies...in the way neighborhood [units] were operationalized or
theoretically situated.” More speciﬁcally, the authors noted that there was
difﬁculty in comparing outcomes from different studies due to the wide variety of
‘neighborhood’ units of analysis used. Sampson et al. (2002) found that overall,
the use of US. Census tracts as a measure of neighborhood was the most
common, with 19 studies in their meta-analysis using that designation. Other
studies used the following as measures: (1) Neighborhood clusters - 7 studies;
(2) Postal sectors / Zip codes - 5 studies; (3) US. Census block groups — 4
studies; (4) Enumeration Districts I Political Districts - 4 studies; (5) Face blocks
— 1 study; and (6) Police beats -1 study. With regard to the latter measure,
Terrill and Reisig (2003: 298) noted that the sites within the Project on Policing
Neighborhoods (POPN) study (Indianapolis, Indiana and St. Petersburg, Florida)

had drawn their police beats to reﬂect existing neighborhood boundaries. Reisig

100

 

and Parks (2003: 46), also using the POPN data, elaborated, stating that the 62
observed neighborhoods contained 130 US. Census tracts. Again, however, it is
unclear that this use of police beats as a proxy measure for neighborhoods is the
correct one, or, at the very least, is generalizable to other jurisdictions. In
particular, given the description presented by Reisig and Parks (2003), one could
say that in those jurisdictions in the POPN data there were an average of two
census tracts per police beat (i.e. their ‘neighborhood’ measure). Yet, as
discussed earlier, Ouimet (2000) gathered data on 84 neighborhoods containing
495 census tracts in a different jurisdiction, for an average of almost six census
tracts per neighborhood. New York City, on the other hand, has 292 recognized
neighborhoods covering ﬁve counties and 2,281 census tracts, for an average of
nearly 8 census tracts per neighborhood. In addition, New York City also covers
75 police precincts and 185 zip codes within those 292 neighborhoods, for an
average of 30 census tracts per police precinct (lnfoshare 2004). Thus, it should
be clear that neighborhood deﬁnitions based on census tracts and police beats,
while both potentially valid measures, are widely varied within the literature. This
stems primarily from the fact that US. census tracts are deﬁned to include
approximately 5,000 residents, which in New York City is a relatively small
geographic area due to crowding conditions while in St. Petersburg and
Indianapolis this may be a signiﬁcantly larger area. In both instances, the census
tract is a well-deﬁned demographic unit of analysis. However, in situations where
the census tract is much larger due to the fact that the 5,000 residents of the tract

live well-spaced apart from one another, the tract may be divided into numerous

101

police beats for reasons of manpower allocation and response time. In contrast,
in situations where the census tract is much smaller due to the fact that the 5,000
residents of the tract live in relatively crowded conditions, one police beat may
actually encompass several census tracts.

Given the variation in neighborhood deﬁnitions, and even the variation
within those deﬁnitions as discussed above, Sampson et al. (2002: 445) note that
“administratively deﬁned units such as census tracts and block groups are
reasonably consistent with the notion of overlapping and nested ecological
structures.” In addition to the latter study, several other researchers have found
that census tracts provide the geographic unit closest in conception to a
neighborhood unit. Coulton et al. (2001), for example, note that when 140
residents of seven different census tracts were asked to draw maps of their own
neighborhood, these maps corresponded highly with the ofﬁcial boundaries of
those census tracts. Leventhal and Brooks-Gunn (2000) concur, noting that
residents’ reports of neighborhood boundaries are consistent with the size (in
square miles) of census tracts in those areas. More importantly, Duncan and
Aber (1997) note that census tract boundaries are typically drawn with the aid
and advice of local community boards and planning commissions, and include
prominent physical features, such as major thoroughfares, which are easily
recognizable by residents. Messner and Tardiff (1986: 303) also note this
phenomenon, arguing that “in almost all instances, there are readily observable
census tracts boundaries which can be matched very closely

with...neighborhood boundaries.” This is important, as Messner and Tardiff

102

(1986: 301) note, due to the fact that “widely recognized and labeled
neighborhoods, rather than being mere ‘statistical aggregates,’ are population
groupings with meaning to the constituents.” This is bolstered by numerous other
studies which ﬁnd that residents of particular neighborhoods are both capable of
deﬁning those neighborhood boundaries clearly (often in line with drawn census
tract boundaries), as well as developing a sense of meaning and purpose within
those neighborhoods (Bursik and Grasmick 1993; Kelling and Stewart 1990;
Taylor et al. 1984). In light of the research indicating that census tracts match up
very favorably both with local administrative boundaries as well as with residents’
conceptions of their own neighborhoods, the current research will use US.
Census tract data to represent neighborhood-level processes.

With respect to the use of US. Census tract data, the ArcView ArcMAP
8.2 software package, a Geographic lnforrnation System, was used to locate
arrest points within a speciﬁc neighborhood through a process known as
geocoding. A speciﬁc address or street intersection for each arrest made within
the PUF Study was provided by the arresting ofﬁcer on an arrest report. Using
commercially available street maps of the six jurisdictions as reference points,
these addresses were geocoded (i.e. placed on a map) within the appropriate
jurisdiction. Based on their spatial representation on these maps, with their
corresponding latitude and longitude (X and Y coordinates), ArcMAP then allows
one to merge a database of attributes to the database of mapped addresses. In
terms of the current research, then, ArcMAP has mapped the arrest addresses to

their corresponding census tracts, and were then used to merge demographic

103

E:.'°‘::'."i""':-':".:'.'.';'-9. ;"-v-.v-- u; 2" ~s~ " " . -‘ ..
. ..;:;.hr:..ewz.cietasiitit.r vmimtsmaimrwmtrthgaeldmm. . 1‘ “- .. ‘- -; ~

' 4.

a;.
j}
.i'i.
".
z
:4}
’r‘.
k,
;.

.
.ti

'.
.t2‘
7"?
.'
.0: ,
‘fé. .
3'0
3.- .
..
it"
-v.
.‘t-
‘3.
its;
,-
..
3',
:._i_
.
1‘1
o‘T .
9- ‘
9‘: i.
it V
.1.
‘n
,2
3:“
-:

 

data for those census tracts to the mapped arrest points. This process is
extremely efﬁcient, as ArcMAP allows the user to set the tolerance limits for
address matching as high as 100% (in which case the address as written down
for an arrest would have to match an address in the commercially available
database exactly). As a matter of practicality, tolerance is often set at 80%, and
any resulting ‘potential matches’ are then matched by the researcher against the

arrest address to ensure that the latter is mapped correctly.

Speciﬁc Hypotheses

The literature review presented in Chapter 2 suggests that use of force
behavior by police ofﬁcers is affected in large part by the nature of the arrest
encounter (i.e. how the citizen behaves, as well as the circumstances
surrounding the arrest). However, there have also been well-designed individual-
level studies which have found that suspect characteristics such as race and
gender affect the application of force. The most statistically sound study to date,
that of Terrill and Reisig (2003), has found that although the effect of race is likely
confounded by the effects of neighborhood SES and social disorganization, the
gender effect remains. However, the latter study has several methodological
problems which have been discussed. These problems are, most importantly: (1)
the deﬁning of simple handcufﬁng as a use of force activity; and (2) the use of
police beats, rather than census tracts, as proxies for neighborhoods. In addition
to these issues, however, Terrill and Reisig (2003) failed to make a signiﬁcant

distinction regarding other individual-level characteristics potentially varying at

104

the neighborhood level. With an emphasis on the impact of neighborhood socio-
demographic characteristics on police use of force behavior, the authors note
that they were unable to consider whether such variability at the encounter
(individual) level existed due to limitations of the data, and thus estimated a ﬁxed-
effects model (Terrill and Reisig 2003: 318, en13). Yet, given the discussion of
Black’s theories regarding social distance presented earlier (one which was also
touched upon by Terrill and Reisig 2003), it is possible that ofﬁcer and suspect
characteristics, when not mis-speciﬁed in a ﬁxed-effects model, may vary in their
effects on use of force behavior across different neighborhoods. In light of the
preceding discussion, the current research seeks to examine the effects of
neighborhood characteristics on use of force behavior, while acknowledging that
such individual characteristics may also play a role. Accordingly, the discussion
to this point leads us to expect the following outcomes:

H1: Ofﬁcers will use more force in neighborhoods which are higher in

concentrated disadvantage;

H2: Minority ofﬁcers will use less force than white ofﬁcers in

neighborhoods which are higher in concentrated disadvantage, due to

their decreased social distance from residents.

H3: Ofﬁcers will use more force against suspects with poor demeanor who

are encountered in high-crime neighborhoods, due to the ofﬁcer's need to

“save face’ and maintain control in such areas.

105

 

H4: Ofﬁcers will use more force against suspects with poor demeanor in
neighborhoods which are higher in concentrated disadvantage, due to the
ofﬁcer’s need to ‘save face’ and maintain control in such areas.

H5: Ofﬁcers will use less force against minority suspects in neighborhoods
higher in concentrated disadvantage, due to the ecological contamination

hypothesis and the concept of tolerance.

Speciﬁcation of Variables of Interest
Dependent Variables:

The dependent variables of interest are two of the measures of force
developed within the context of the multi-site PUF study. These two measures
are: (1) the use of physical force plus threats; and (2) the maximum amount of
force used. As noted previously, recent research on police use of force has
acknowledged that the threat of violence (i.e. force) is an act of violence in and of
itself. Indeed, legally, individuals may be arrested for making threats of several
types, and thus including threats as a form of violence by police ofﬁcers is

consistent with these legal deﬁnitions.69

Prevalence of Force
The ﬁrst dependent variable to be used in this research is simply a
dichotomous measure of the prevalence of force. This variable distinguishes

between arrests in which physical force, or the threat of such force, was or was

 

‘9 Note that although various deﬁnitions refer to force as ‘violence’, or use the terms interchangeably, the
occurrence of a violent act (i.e. forceful behavior on the part of a police ofﬁcer) in no way implies
wrongdoing on the part of the ofﬁcer.

106

not used by an ofﬁcer in one of the six jurisdictions. As is common with
dichotomous variables, an arrest was assigned a value of zero (0) if no force was
used or threatened during the arrest encounter, and a value of one (1) if physical
force was either used or threatened during an arrest. In the PUF study, an arrest
was coded as having contained an instance of physical force if any of the
following were used or threatened: (1) weapons; (2) weaponless tactics; and (3)
severe restraints. Overall, the use or threatened use of physical force so deﬁned
occurred in 17.1 % of all arrest encounters (N = 1,283). Table 4 presents a more
detailed listing of the speciﬁc actions encompassed by the use of force

instrument for the prevalence of force measure.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 4: Items within Prevalence of Force Measure
Tactic Date Item N %

Use of a Weapon
None 7,354 97.9
Baton 6 0.1
Canine 14 0.2
Chemical Agent 79 1.1
Flashlight 23 0.3
Handgun 6 0.1
Motor Vehicle 9 0.1
Rifle / Shotgun 2 0.0
Other 19 0.3

Weaponless Tactics
No physical contact 6,328 84.2
Bite / scratch 1 0.0
Carotid hold 18 0.2
Control hold 153 2.0
Grab 589 7.8
Hit 12 0.2
Kick 3 0.0
Pressure hold 48 0.6
Push / shove 80 1.1
Spit 21 0.3

 

 

 

 

 

 

 

107

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 4 (cont’d)
Twist arm 98 1.3
Wrestle 91 1.2
Other tactic 70 0.9
Use of Restraints
Leg cuffs 67 0.9
More severe restraints 29 0.4
Prevalence of Force
Weapons 1 58 2.1
Weaponless Tactics 1,184 15.8
Severe Restraints 96 1.3
Total Physical Force 1,283 17.1

 

 

 

 

 

Seventy of Force

The second dependent variable to be used in this research encompasses
a ranking of various police actions. As discussed previously, the PUF study
asked a total of 503 ofﬁcers across all six jurisdictions to rank hypothetical police
behaviors on a scale of 1 to 100 based on their personal experiences, and then
weighted these rankings based on the occurrence of these actions within the
actual data collected over 7,512 arrests. This measure of severity, then,
operationalized as the maximum amount of force used in the encounter, ranges
from zero (0) for no action reported, to ninety-nine (99) for a fatal discharging of a
ﬁrearm.” Overall, the average maximum force ranking for all 7,512 arrests was
30.4, a ranking between the use of leg restraints (hobbies) and the threatened
use of a ﬂashlight. lmportantly, this ranking was higher than the simple use of
handcuffs, which was given a ranking of 28.2 by the ofﬁcers, and occurred in

82.3% (N = 6,182) of all arrests, providing more impetus for the exclusion of

 

7° It is important to note that the list of behaviors produced by these ofﬁcers included nearly all of the
behaviors which previous research has deﬁned as use of force activity. The only notable exception was
‘pat-downs’, which Terrill and Reisig (2003), as well as Terrill and Matrofski (2002), used in their
research.

108

handcufﬁng as a use of force behavior by police ofﬁcers. Table 5 provides a

more detailed listing of the speciﬁc actions encompassed by the use of force

instrument for the maximum force measure.

 

Table 5: Items within Maximum Force Measure

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Police Action Rank N % Cumulative
%

No action reported 0.0 62 0.8 0.8
Conversational tone 15.6 153 2.0 2.8
Gently hold suspect 15.9 83 1.1 3.9
Two ofﬁcers present 20.6 668 8.9 12.8
Command suspect 22.0 99 1.3 14.1
Shout I curse at suspect 22.5 3 0.0 14.1
Spit on suspect 23.2 2 0.0 14.1
Chase suspect (helicopter) 24.0 1 0.0 14.1
Verbally threaten suspect 25.4 5 0.1 14.2
Push suspect 26.7 0 0.0 14.2
Use handcuffs 28.2 4,305 57.3 71.5
Chase suspect (foot pursuit) 29.3 95 1.3 72.8
Use leg restraints 30.0 14 0.2 73.0
Threaten to use ﬂashlight 30.9 0 0.0 73.0
Threaten to use chemical 31.7 1 0.0 73.0
a ent

Possess canine ofﬁcer 31.9 10 0.1 73.1
Threaten to use baton 32.0 1 0.0 73.1
Grab suspect 33.0 461 6.1 79.2
Display baton 34.6 4 0.1 79.3
Use pressure hold 34.7 10 0.1 79.4
Twist suspect’s arm 35.1 98 1.3 80.7
Use other tactic 35.2 32 0.4 81.1
Display chemical agent 37.0 7 0.1 81.2
Use severe restraints 37.1 17 0.2 81.4
Bite suspect 37.7 0 0.0 81.4
Display ﬂashlight 37.8 7 0.1 81.5
Use choke hold 38.9 78 1.0 82.5
Possess shopgun 40.2 640 8.5 91.0
Kick suspect 40.6 1 0.0 91.0
Hit suspect 40.8 2 0.0 91.0
Chase suspect (car) 41.4 137 1.8 92.8
Use chemical agent 45.9 31 0.4 93.2

 

109

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table 5 (cont’d)

Threaten to use car as 46.0 0 0.0 93.2
weapon

Threaten to use canine 46.1 5 0.1 93.3
ofﬁcer

Wrestle with suspect 48.2 184 2.4 95.7
Use ﬂashlight 49.9 23 0.3 96.0
Threaten to use riﬂe / 51.8 1 0.0 96.0
shotgun

Use canine 52.1 12 0.2 96.2
Threaten to use handgun 52.4 2 0.0 96.2
Use baton 53.0 6 0.1 96.3
Use other weapon 53.1 15 0.2 96.5
Display handgun 55.4 165 2.2 98.7
Use carotid hold 56.0 31 0.4 99.1
Display riﬂe / shotgun 57.4 23 0.3 99.4
Use car as weapon 69.4 10 0.1 99.5
Use riﬂe / Shotgun 79.2 2 0.0 99.5
Use handgun 81.7 6 0.1 99.6

 

Independent Variables:

The independent variables used in previous research have generally fallen
into one of several categories: (1) the nature of the encounter l offense; (2)
suspect characteristics; and (3) ofﬁcer characteristics. More recent research that
has examined contextual effects has also explored: (1) the nature of the arrest
location; and (2) neighborhood-level socio—demographic variables. Based on
these two sets of literature, the current research incorporates variables falling
within these ﬁve categories as well. Although the focus here is primarily on
strengthening previous results regarding the effects of neighborhood context, as
well as presenting evidence for the effects of social distance on the behavior of

minority police ofﬁcers, the consistent signiﬁcant ﬁndings regarding other

110

variables must be addressed in addition to an array of statistical control

variables.

Neighborhood-Level Variables

Neighborhood structure within the six jurisdictions was determined through
the use of 2000 US. Census data at the census tract level. Speciﬁcally, at the
census tract level, the following data items were submitted to a factor analysis in
order to create a measure of concentrated disadvantage: (1) percent Black; (2)
percent under 18 years of age; (3) percent below poverty level; (4) percent
unemployment; (5) percent female-headed households; and (6) percent on public
assistance. It should be noted here that the factor score was created across all
census tracts within all six jurisdictions for the current analysis. This is due to four
reasons: (1) the original PUF sample contained all possible neighborhoods within
the six jurisdictions, and thus there is 100% coverage within the sample despite
the fact that not all census tracts experienced an arrest during the study period;
(2) the demographic similarities of census tracts across the six jurisdictions
increase the likelihood of generalizability; (3) utilizing only those census tracts in
which an arrest was made would result in a loss of statistical power due to
smaller sample size; and (4) utilizing only those census tracts in which an arrest
was made results in the inability to calculate a Moran’s l as a measure of spatial
autocorrelation. due to the resulting geographic “holes”.71 In addition to

concentrated disadvantage, one neighborhood-level measure of crime is

 

7‘ Further description of the census tracts used for, and excluded from, the analyses can be found in
Appendix F.

111

included in the analysis. By aggregating crime data for all arrests within the
sample to the appropriate census tracts, a neighborhood measure of all Index I
crimes per 100,000 residents was computed. While this provides measures of
arrests, rather than total offenses reported, this aggregation is similar to the F Bl’s
Uniform Crime Reports data, with the added beneﬁt of being available at the
neighborhood level. In addition, the jurisdiction is included as a series of dummy-
coded variables, with each of the following coded as zero (0) for a ‘No’ response
and coded as one (1) for a ‘Yes’ response: (1) Colorado Springs, CO (as the
reference category); (2) Charlotte, NC; (3) Dallas, TX; (4) St. Petersburg, FL; and
(5) San Diego (city and county), CA.72 The jurisdiction variable is included to

control for the potential effects of agency-speciﬁc factors.73

Encounter-Level Variables

As discussed previously, a number of encounter-level variables have been
included in prior research, either as statistical controls or in order to estimate
their effects upon use of force behavior by police ofﬁcers. All of the variables
included in the current research were ﬁrst tested for their bivariate correlations
with both dependent variables (see Garner et al., 2002 for more detail). In
addition, multivariate analyses with these variables were conducted using only

those variables within each “domain” (i.e. suspect characteristics, ofﬁcer

 

72 Due to the geographic “holes” resulting ﬁom separating the San Diego Police Department and the San
Diego Sheriff’s Ofﬁce, GeoDa was unable to calculate a measure of spatial autocorrelation for either
'urisdiction, and thus these were combined into one jurisdiction for all subsequent analyses.

Note, however, that a true measure of the effects of jurisdiction would require a 3-level model with
jurisdiction as one level. Given the fact that only six jurisdictions were studied, there are not sufﬁcient
degrees of freedom to conduct such an analysis, and therefore jurisdiction is used as a neighborhood-level
variable.

112

characteristics, nature of the encounter and offense, and nature of the encounter
location) with both dependent variables. From these analyses, only those
independent variables which were statistically signiﬁcant at the .01 level were
selected for inclusion in the ﬁnal models. These variables will now be examined

more closely, as they are included in the current research as well.

Suspect Characteristics

Prior research on the use of force, including the only study to use
multilevel modeling (Terrill and Reisig 2003), has often found various suspect
demographic characteristics to have an effect on the application of force.
Accordingly, the current research will include the following suspect
characteristics as independent variables: ( 1) suspect gender; (2) suspect age;
and (3) suspect race. Suspect gender is a dichotomy, with females coded as
zero (0) and males coded as one (1), with females as the reference category.
Suspect age is the actual age of the suspect. Suspect race is a series of dummy
coded-variables coded as zero (0) for ‘No’ and one (1) for ‘Yes’ for: (1) Black; (2)
Hispanic; (3) Other; (4) Missing and (5) White (as the reference category).

In addition to these demographic variables, other aspects of the suspect’s
behavior and status are included as control variables, namely: (1) suspect
sobriety; (2) suspect disrespect / antagonism; (3) suspect physical resistance; (4)
suspect is known to carry a weapon; (5) suspect is known to be assaultive; (6)
suspect is known to be a gang member; (7) suspect’s relationship to the victim;

and (8) suspect’s relationship to bystanders. Suspect sobriety is coded as 0 if the

113

 

 

suspect is sober and as 1 if the suspect is intoxicated due to either drugs or
alcohol. Suspect antagonism and suspect physical resistance are both
dichotomous variables, coded as zero (0) for a ‘No’ response (i.e. suspect was
not antagonistic I suspect was not physically resistant) and coded as one (1) for
a ‘Yes’ response. The next three control variables (‘suspect known to carry a
weapon’, ‘suspect known to be assaultive’, and ‘suspect known to be a gang
member’) are all also dichotomous variables, coded as zero (0) for a ‘No’
response and as one (1) for a ‘Yes’ response. The suspect’s relationships to the
victim and to bystanders are both a series of dummy-coded variables, coded as
zero (0) for a ‘No’ response and as one (1) for a ‘Yes’ response, for: (1) strangers
(as the reference category for victims); (2) friends; (3) family; (4) unknown
relationship; or (5) none (as the reference category for bystanders, indicating that

there were no bystanders present at any time.

Ofﬁcer Characteristics

Ofﬁcer characteristics have typically been included as a matter of
comprehensiveness on the part of researchers, with little emphasis on their
potential effects. Indeed, as noted earlier, previous studies have failed to allow
the effects of ofﬁcer characteristics, notably race, to vary across neighborhoods.
The current research seeks to rectify the latter issue, while also including other
ofﬁcer characteristics. Ofﬁcer gender and ofﬁcer race are coded in the same
manner as the corresponding variables for suspects. Ofﬁcer age is also the

actual age of the ofﬁcer. In addition to these demographic variables, other police

114

ofﬁcer characteristics are included as controls, namely: (1) ofﬁcer demeanor
toward the suspect; (2) prior medical attention given to ofﬁcer; and (3) the
number of surveys completed by the ofﬁcer. Ofﬁcer demeanor is a dichotomous
variable, coded as zero (0) for a ‘No’ response (i.e. the ofﬁcer was not
antagonistic) and one (1) for a ‘Yes’ response. Prior medical attention is a
variable used to determine if the ofﬁcer has been injured in the line of duty on a
prior occasion, and is coded as zero (0) for a ‘No’ response (i.e. the ofﬁcer has
never received medical attention for an injury sustained while on duty) and as
one (1) for a ‘Yes’ response. The number of surveys completed by the ofﬁcer
provides an indication of the overall aggressiveness of the ofﬁcer in performing
policing functions, as ofﬁcers who have completed more surveys have made

more arrests than their peers.

Nature of the Encounter and Offense

In general, prior researchers have included variables regarding the nature
of the offense, as well as the nature of the encounter (i.e. how the encounter
came about), as control variables. The current research follows this protocol by
introducing the following variables into the model: (1) the offense type; (2)
whether the offense occurred on a weekend; (3) the number of suspects; (4) the
custody status of the suspect; (5) the ofﬁcer’s duty status; (6) the ofﬁcer’s
approach to the encounter; (7) the ofﬁcer’s use of back-up ofﬁcers; (8) the
ofﬁcer’s mobilization; (9) the number of ofﬁcers present; (10) the demeanor of

bystanders toward the ofﬁcer; and (11) the number of bystanders. Offense type is

115

a dichotomous variable coded as zero (0) for a non-violent offense and as one
(1) for a violent offense. The timing of the arrest was coded as zero (0) if the
arrest took place during the week, and coded as one (1) if the arrest took place
during the weekend, deﬁned as 6pm on Friday nights until 6am on Monday
mornings. The total number of suspects is included as a measure of
dangerousness during the encounter, as any situation in which suspects
outnumber the ofﬁcer increases the ofﬁcer’s potential for injury. The custody
status of the suspect is a dichotomous variable, coded as zero (0) if the suspect
was not in custody at the time the ofﬁcer arrived and as one (1) if the suspect
was already in custody at the time the ofﬁcer arrived. The ofﬁcer’s duty status is
also a dichotomous variable, coded as zero (0) if the ofﬁcer was on-duty at the
time of the arrest and as one (1) if the ofﬁcer was off-duty at the time of the
arrest. The ofﬁcer’s approach to the encounter is a series of dummy-coded
variables which describe how the ofﬁcer proceeded to the encounter after having
been dispatched, with each of the following coded as zero (0) for a ‘No’ response
and coded as one (1) for a ‘Yes’ response: (1) routine approach (as the reference
category); (2) priority call; (3) lights and sirens; and (4) unknown approach. The
ofﬁcer's use of back-up ofﬁcers is a dichotomous variable coded as zero (0) if the
ofﬁcer did not call for back-up at any point, and coded as one (1) if the arresting
ofﬁcer called for back-up. The arresting ofﬁcer’s mobilization is also a series of
dummy-coded variables which reﬂect the different ways in which a police ofﬁcer
may have come into contact with a citizen regarding a speciﬁc encounter, with

the following coded as zero (0) for a ‘No’ response and coded as one (1) for a

116

‘Yes’ response: (1) ofﬁcer was dispatched to the scene (as the reference
category); (2) citizen initiated the encounter; (3) ofﬁcer initiated the encounter;
and (4) unknown how the encounter was initiated. The number of ofﬁcers present
during the encounter, as well as the number of bystanders, is introduced as a
measure of the ofﬁcer’s level of safety. The ﬁnal variable to be introduced here

as a control is the demeanor of bystanders toward the ofﬁcer, coded as zero (0) if
the bystanders were not antagonistic, and coded as one (1) if the bystanders

were antagonistic toward the police ofﬁcers present during the encounter.

N_at_l_ire of the Encounter Location

The ﬁnal class of control variables deals with the nature of the location in
which the encounter took place. In terms of the current research, these variables
are: (1) location is known for criminal activity; (2) location is known to be
hazardous; (3) the arrest occurred inside; and (4) visibility at the arrest location.
The ﬁrst two variables (‘location known for criminal activity’ and ‘Iocation known
to be hazardous’) are both dichotomous variables, coded as zero (0) for a ‘No’
response (i.e. the location is not known for criminal activity, or the location is not
known to be hazardous) and coded as one (1) for a ‘Yes’ response. The location
of the actual arrest is also a dichotomous variable, coded as zero (0) if the arrest
took place at a location other than inside the suspect’s home, and coded as one
(1) if the arrest took place inside the suspect’s home. Finally, the visibility

variable is an ordinal variable describing the ofﬁcer’s ability to distinguish the

117

suspect’s movements and surrounding features, with categories from 1-10, with

1=poor, 4=moderate, 7=good, and 10=excellent.

Analysis Procedures
Spatial Autocorrelation:

As can be seen from the discussions of the data collection procedures and
the theoretical emphases of the current work, the effect of neighborhood context
on police use of force behavior is viewed as a spatial phenomenon. Accordingly,
GIS software is used to link US. Census data to the appropriate census tracts
(the operationalization of “neighborhood” in the current research) in order to
provide a graphical representation of all arrest cases within the study which can
also be used for analytical purposes. Given that the current work seeks to
determine neighborhood effects, prior research has suggested that the most
serious obstacle to the latter is the issue of spatial autocorrelation (Baller et al.
2001; Leventhal and Brooks-Gunn 2000; Morenoff and Sampson 1997; Morenoff
et al. 2001; Smith and Jarjoura 1989). At its essence, spatial autocorrelation
refers to the tendency of geographical areas to be susceptible to similar events
(in this case, use of force during arrests), based in large part on their proximity to
one another. This susceptibility violates the assumption of independence (i.e. the
random distribution) between events. As Baller et al. (2001: 562) note, “if spatial
processes operate and are not accounted for, inference will be inaccurate and
estimates of the effects of independent variables may be biased...” Messner et al

(1999: 427) note the same phenomenon, stating that “ignoring spatial

118

dependence and/or spatial heterogeneity in the model may lead to false
indicators of Signiﬁcance, biased parameter estimates, and misleading
suggestions of ﬁt.” Several researchers have noted that there is a distinction
between two types of spatial autocorrelation - spatial effects (also called a
spatial lag) and spatial disturbance (also called a spatial nuisance or spatial
error) (Anselin, 1988; Doreian 1980; Doreian 1982). Baller et al. (2001) note that
in the case of a spatial lag, a weighted average of values for the dependent
variable in adjoining neighborhoods is introduced as a spatial dependence
covariate in the model. In the case of a spatial nuisance, the spatial dependence
is incorporated into the regression error term (Baller et al. 2001: 563-566). In
estimation of their model regarding neighborhood inequality and collective
efﬁcacy, Morenoff et al. (2001: 522) explicitly state that “neighborhoods are
interdependent and characterized by a functional relationship between what
happens at what point in Space and what happens elsewhere.” However,
Magalhaes et al. (2000: 6) correctly point out that “units...have, for instance,
different sizes, shapes, densities, and these differences can generate
measurement errors that can cause heteroskedasticity.” Langford et al. (1999)
note that multilevel modeling can address these issues of heteroskedastic errors.
More importantly, Battage et al. (2001: 1) appropriately note that heterogeneity
across units (i.e. differences between neighborhoods) is modeled with an error
component (i.e. spatial nuisance) model. Thus, the current work uses a spatial
error (or spatial nuisance) model for one methodological reason (i.e. in

conjunction with multilevel modeling techniques, the spatial error model alleviates

119

concerns regarding heteroskedasticity), and one theoretical reason (i.e. the
characteristics of surrounding neighborhoods, as opposed to incidents of use of
force in those neighborhoods, are thought to inﬂuence use of force behavior
within a particular neighborhood).

Once one has determined which model of spatial autocorrelation will be
used, it is still necessary to discover if the problem exists within the current
dataset. The test most often used to determine the presence of spatial
autocorrelation is the Moran coefﬁcient, which is similar to a Pearson correlation
coefﬁcient. The Moran coefﬁcient is “based on the spatial autocovariance of a
variable standardized by the overall mean of that variable, [which] can be
compared to the mean value of the coefﬁcient under the assumption of a random
spatial pattern, [after which] a ‘2’ score can be calculated to assess the statistical
signiﬁcance of the observed spatial autocorrelation” (Messner and Tardiff 1986:
307). More importantly, Anselin and Kelejian (1997: 153-154) note that “Moran’s
l...is the only acceptable [test] in the presence of spatially lagged dependent
variables...[and furthermore] is an exact test.”

While the Moran coefﬁcient provides a measure of spatial autocorrelation,
an issue of great concern to researchers using neighborhood-level data, until
very recently it has not worked within the conﬁnes of hierarchical linear modeling
(HLM) software. In fact, Morenoff et al. (2001: 532) noted that “software that can
simultaneously handle. ..random effects of neighborhoods and spatial
dependence” was not available at the time of their research. This resulted in the

estimation of a hierarchical generalized linear model without a spatial

120

dependence term in order to “compute posterior modes74 of neighborhood-
speciﬁc log-homicide rates given the data, the grand mean estimate for Chicago,
and the estimated between-neighborhood variance in the true log-rates”
(Morenoff et al. 2001: 532). These posterior modes were then introduced into a
regression model as an independent variable. While this two-step process is
viable, Morenoff et al. (2001) correctly note that this provides only an
approximation of the effects of spatial dependence.

Recently, Anselin (2003a) has introduced a software package designed to
more faithfully incorporate spatial dependence terms within hierarchical linear
modeling (HLM) software. The GeoDa software package was designed to
combine the interactive capabilities of GIS software (such as ArcView) with the
statistical abilities of analytical software (such as SPSS or HLM). As Anselin

(2003b) notes, this produces a Moran autocorrelation statistic as follows:

“H = ZiijWiijl [Equation 1]

where z), and 2‘. are variables observed within the given neighborhoods, and Wjj is
the spatial weights matrix. The Moran coefﬁcient is then viewed as a scatterplot,
with the spatially lagged variable on the vertical axis and the original variable on
the horizontal axis (Anselin 2003b). In order to determine which data points are
within a critical distance (yet still in another neighborhood) of the initial event,

GeoDa uses contiguity-based rates constructed from the shape ﬁles present in

 

7‘ The posterior modes n,‘ for neighborhood i was calculated as a weighted average of the log-homicide rate
for each neighborhood, and the overall mode of the homicide rates estimated ﬁ'om all neighborhoods.

121

the mapping software. These rates can be based on either the rock criterion,
which is constructed from only those neighborhoods sharing a common linear
boundary with the neighborhood of interest, or on the queen criterion, which is
constructed from all neighborhoods sharing either a common linear boundary or
a common vertex with the neighborhood of interest. Prior research (Messner et
al. 1999; Morenoff et al. 2001; Reisig et al. 2004) has used the rock criterion
when examining spatial dependence within neighborhood effects. However,
given that the current research uses a spatial error model of spatial
autocorrelation, there is no theoretical reason to believe that this process is
stronger for neighborhoods sharing common linear boundaries than for
neighborhoods sharing common vertices. Thus, the queen criterion is used here,
as it is believed that the characteristics of all the immediate surrounding
neighborhoods will exert an effect on use of force behavior within the
neighborhood of interest. Given the strengths of the software, then, GeoDa will
be used to introduce a well-deﬁned spatial autocorrelation term (as a spatial

error) based on the queen criterion into the hierarchical linear model.

Hierarchical Linear Modeling:

Given the nested nature of the data (arrests nested within neighborhoods),
the appropriate statistical technique to be used is hierarchical linear modeling.
The latter will be used to simultaneously regress each dependent variable
(separately) on neighborhood-level and encounter-level independent variables.

As Bryk and Raudenbush (1992: 20) note, the initial step in estimating an HLM is

122

to determine whether hierarchical modeling is appropriate for these data. In its
most basic form, the relationship between the dependent variable and the

independent variables at the ﬁrst level is described as follows:

Yij = Bot + BlilXii‘ xi) + rii [Equation 2]

where ij is the ith dependent variable in the jth neighborhood, 80,- is the intercept
of the jth neighborhood, B1j is the slope of the jth neighborhood, er is the ith
independent variable in the jth neighborhood, X.,- is the mean of the independent
variable in j‘" neighborhood, and rt; is the error term for the ith independent variable
in the jth neighborhood.

The relationship between the dependent variable and the independent

variables at the second level can be described as follows:

I30] = Yoo + Y01Wj + UOj [Equation 3]

where [30; is the intercept of the jth neighborhood taken from Equation 1, yon is the
mean of the intercept across neighborhoods, Yo1 is the difference of the means of
the independent variables, Wj is the value of the independent variable in the jth
neighborhood, and lit)" is the unique effect of the jth neighborhood on the mean of

the intercept

123

and

B1]: Y10 + Y11Wi + Uri [Equation 4]

where 81,- is the slope of the j‘" neighborhood taken from Equation 1. Y10 is the
mean of the slopes of the individual-level independent variables (i.e. the main
effect of the independent variable from Equation 1), yrr is the interaction term
between the individual-level independent variables and the neighborhood-level
independent variables, Wj is the value of the independent variable in the jth
neighborhood, and U11 is the unique effect of the j‘“ neighborhood on the mean of
the slopes.

These equations in combination represent the simplest example of a
random-intercepts model, in which only the ﬁrst level intercept coefﬁcient. I301. is
viewed as random. In the current research, as described earlier, it is
hypothesized that the effects of ofﬁcer race will vary across neighborhoods, thus
requiring the use of a random-coefﬁcients model, in which both 80; and 81; (the
slope for ofﬁcer race) are allowed to vary, and both yor and yrr are constrained to
be null. The latter is required due to the fact that when cross-level interactions
are present, the regression coefﬁcient of the direct independent variable
estimates the effect of that variable when the other independent variable in the
interaction is zero (Bryk and Raudenbush 1992: 20). This leads to the following

simpliﬁed equations:

Bot = You + U0; and 131,- = yio + U1,- [Equation 5 and Equation 6]

124

where yon is the average intercept across neighborhoods, yro is the average slope
across neighborhoods, U0] is the unique effect of the jth neighborhood on the
intercept, and U1] is the unique effect of the j‘“ neighborhood on the slope.

Again, at the most basic level, then, all of these equations can be

combined to yield:

Yij -"'- Yoo + y1o(Xij- X.j) + Uoj + U1j(Xij- X.j) + fij [Equation 7]

This model implies that the dependent variable, ij, is a function of the average
regression equation, yon + y10(er — X.j), added to a random error term with three
components: UOj, the random effect of the j‘" neighborhood on the mean; U1j(ij—
X4), where U1j is the random effect of the jth neighborhood on the slope 81,-; and raj,
the encounter-level error term (Bryk and Raudenbush 1992: 21 ).

Having established the basic model used in hierarchical linear modeling, it
is now necessary to describe the process the current research will undertake in
order to estimate the proper model using the PUF study data. Preliminary models
will be estimated using a one-way ANOVA model for the dependent variable at
the encounter level in order to obtain descriptive statistics which will test whether
or not HLM techniques are appropriate for these data. This ANOVA model will

produce the following equation:

Yii = You + Uoi + rii [Equation 3]

125

which is a one-way ANOVA model with a grand mean (yon), a neighborhood

effect (UOj), and an encounter-level effect (rij), and has a variance of:

Var (Y ij) = Var(uo,- + I'ij) = Too + 02 [Equation 9]

where Too represents the within-neighborhood variability, and 02 represents the
between-neighborhood variability. The latter two parameters allow us to calculate

the intraclass correlation coefﬁcient as:

p = r00 / (Too + 02) [Equation 10]

where p is the intraclass correlation coefﬁcient. The latter is deﬁned as a
measure of group homogeneity. If the intraclass correlation is zero, the clustering
of the data has no effect on the relationships between the variables of interest,
and the assumption of independent observations is not violated. This, in turn,
indicates that linear modeling techniques may be used. If the intraclass
correlation coefﬁcient is not equal to zero, then the assumption of independence
is violated, and using linear modeling techniques would result in: (1) reduced
reliability of parameter estimates; (2) increased probability of Type 1 errors; and
(3) underestimation of the standard error of the coefﬁcients. In this situation,
linear modeling is clearly not appropriate, and HLM techniques must be used. In
effect, the intraclass correlation coefﬁcient reveals the proportion of variance

between neighborhoods in the dependent variable. In addition to using this value

126

to determine if HLM is appropriate, analysis of the x2 statistic will be used to
reject the null hypothesis that there is no difference in the dependent variables
between neighborhoods. The intraclass correlation coefﬁcient and the x2 statistic
together are expected to conﬁrm that the data captured in the PUF study are
amenable to analysis through HLM techniques.

It is important to note here that the binary dependent variable of physical
force plus threats requires the use of hierarchical generalized linear models
(HGLM), which take a slightly different form than the basic HLM equations
described above. In HGLM, the Level-1 model consists of three parts: (1) the
sampling model for a binary outcome; (2) the link function; and (3) the structural

model. The sampling model takes the form:

E (Yij I ¢il) = mii¢ij [Equation 10]
and

Var (Yr l ¢r) = mint; (1- an) [Equation 11]

where i represents an individual arrest, j represents a speciﬁc neighborhood, mjj
represents number of trials for the ith arrest in the jth neighborhood (equal to 1 in
the present Bernoulli case, as use of force was only measured once per arrest),
and (Dr; represents the probability of “success” for the ith arrest in the jth

neighborhood.

127

The link function (the logit-link in this case) takes the form:

[1]] = log (49,,- I (1- Op) [Equation 12]

where nr represents the log-odds of success for the ith arrest in the jth

neighborhood.

The structural model takes the form:

nil = 130i + i31lX1ii + Bzixzii +---+ Bpixpii [Equation 13]

where "ii is the predicted log-odds of success for arrest i in neighborhood j, 80,- is
the log-odds of success for an arrest in neighborhood j when all of the
independent variables have a value of zero, [31; is the effect of the ﬁrst
independent variable on the log-odds of success in neighborhood j, X1 r,- is the
value for the ith arrest in the jth neighborhood for the ﬁrst independent variable, [32;
is the effect of the second independent variable on the log-odds of success in
neighborhood j, Xzii is the value for the ith arrest in the jth neighborhood for the
second independent variable, Bpj is the effect of the last independent variable on
the Iof-odds of success in neighborhood j, and Xprr is the value for the ith arrest in

the jth neighborhood for the last independent variable.

128

Furthermore, in HGLM, the intercepts and slopes of the Level-1 model
become the outcome variables of the Level-2 model, leading to the following
equaﬁons:

1301 = Yoo + U0] [Equation 14]

where Yoo is the average log-odds of success across neighborhoods for the
average arrest and UOj is the unique effect of neighborhood j on the log-odds of

success for the reference category, as well as:

i31j = Y1o 'i' U1j [Equation 15]

where Yro is the mean over neighborhoods of the effect of the ﬁrst independent
variable on the log-odds of success and U0] is the unique effect of neighborhood j
on the effect of the ﬁrst independent variable on log-odds of success in

neighborhood j, as well as:

321': Yzo + Uzi [Equation 16]

where Yzo is the mean over neighborhoods of the effect of the second

independent variable on the log-odds of success and UOj is the unique effect of

neighborhood j on the effect of the second independent variable on log-odds of

success in neighborhood j.

129

 

Summary

Overall, this chapter has presented a discussion of the data collection
procedures, sample characteristics, variables of interest, and plans for statistical
analyses. Based on the data sample collected in the PUF study, as well as the
consistency of effects of particular independent variables throughout prior
research, two dependent variables and an array of independent variables were
chosen for analysis. Based on the work of Smith (1986) and Terrill and Reisig
(2003) regarding neighborhood effects on use of force behavior by police
ofﬁcers, it is anticipated that the data in the current research will be best utilized
within an analytical scheme focused on hierarchical linear modeling techniques.
A description of this technique has been provided, as well as a discussion of
preliminary steps to be taken to ensure that these techniques are indeed
appropriate for the data. The following chapter will present the results of these

analyses.

130

Chapter 4: Results

The purpose of this dissertation was to assess the effects of
neighborhood-level variables on use of force behavior by police ofﬁcers. Using
hierarchical linear modeling techniques, the research questions test the extent to
which neighborhood characteristics, as well as individual characteristics allowed
to vary at the neighborhood level, increase or decrease the use of force by police
ofﬁcers during arrest situations. The research sample used 7,512 police ofﬁcer
surveys ﬁlled out subsequent to every arrest in the six study sites between
August 1996 and February 1997. Of these 7,512 ofﬁcer surveys, 656 were
eliminated from the analysis due to an inability to be geocoded using ArcView
ArcMAP software, leaving a sample size of 6,856. Of the latter cases, 924 were
then eliminated due to the fact that they occurred in a census tract where too few
arrests overall were observed, leaving the ﬁnal sample size of 5,932. 75 In
addition to the latter issues with the data, it should be noted that several cases
were missing data on one or more of the independent variables. Traditionally,
social scientists have dealt with missing data by: (1) constructing a dummy
variable for missingness; (2) performing logistic regression to predict
missingness using covariates; and 3) reporting whether any variables are
signiﬁcant predictors of missingness. This approach has a number of
disadvantages, including a lack of theoretical justiﬁcation, as well as biased
parameter estimates and standard errors. The approach of simple listwise

deletion is also often used, but in situations where more than 5% of the cases are

 

7’ The criterion for too few observed arrests within a census tract was 4 or fewer, based on previous
research efforts which have identiﬁed 5 or more cases within a Level-2 unit as an acceptable sample size in
hierarchical linear modeling (Cueto et al. 2003).

131

missing this leads to a serious loss of statistical power where the data are
missing-at-random (MAR) or missing-completely-at-random (MCAR), and to
biased parameter estimates where the missing data are observed-at-random
(OAR).76 In these situations, it is preferable to use any of the following methods:
(1) direct maximum-likelihood estimation; (2) Bayesian modeling with Markov
Chain Monte Carlo methods; (3) multiple imputation; and (4) regression-based
imputation (Allison 2002; Jones 1996; Little and Rubin 1987). Of the latter
methods, only regression-based imputation is currently available in widely-used
pre-packaged software such as SPSS, and thus it is this method which was used
for missing data analyses. Using an OLS regression model for continuous
variables, and a logit regression model for binary variables, regression-based
imputation uses cases in the sample with complete information to predict data for
the missing variables of interest.77 Thus, Table 6 presents the sample
characteristics for the individual-level independent variables for the 5,932 cases
used in the analyses, with all missing values replaced through a regression-

based imputation method.78

 

7" Data for this dissertation were missing on only 10 of the 31 variables used in the analyses. Although in
some instances data were only missing for 1.5% of all cases, the variable for officer race, which is of
considerable importance to the hypotheses of the current work, was missing data in 5.4% of all cases.

77 It should be noted here that regression-based imputation does suffer from slightly biased standard errors
due to the fact that it ignores signiﬁcant variability. However, considering the nature of the data, as well as
the availability of alternative imputation methods, it was felt that this method was adequate for the purposes
of conducting a hierarchical linear model analysis focusing primarily on the effects of Level-2 (i.e.
neighborhood) variables.

78 Note that in Table 6, the values for REPEAT (the number of surveys completed by the officer),
NUMBERSO (the number of suspects present at the completion of the arrest), NUMBERPO (the number
of ofﬁcers present at the completion of the arrest), OFF1AGE (the actual age of the arresting ofﬁcer), and
SUSPAGE (the actual age of the suspect), are not included, as this would require too much space within the
table. This information is available in the Appendix (Table D-l).

132

Table 6: Sample Characteristics (Level-1 Independent Variables)

VARIABLE NAME

NATURE OF THE LOCA TION

CATEGORY

Location Known for Criminal Activity

Location Known to be Hazardous

Arrest Took Place Inside

Visibility at Arrest Location

NATURE OF THE ENCOUNTER

Violent Offense

Weekend

No
Yes

No
Yes

No
Yes

Poor
Somewhat Poor
Lessthan
Moderate

Moderate
More than
Moderate
Acceptable
Good

Fairly Good
Very Good
Excellent

No
Yes

No
Yes

Bystander Demeanor toward Ofﬁcer

Number of Suspects

POLICE MOBILIZATION
Custody Status of Suspect

Not Antagonistic
Antagonistic

On Street
In Custody

133

VALUE

—L A

#00 N-l —I~

O‘DQVOUI

JO

40

do

5932
3405
2527

5932
4885
1 047

5932
3938
1994

5932
89
1 30

206
359

466
551
754
756
498
2123

5932
4838

1094

5932
3669
2263

5932
5602
330

5932
5932

5109
823

MEAN

0.43

0.18

0.34

7.6

0.18

0.38

0.06

1.4

0.14

SD

0.49

0.38

0.47

2.43

0.39

0.49

0.24

1.22

0.35

Table 6 (cont’d)

Officer Dispatched to Scene

No

Yes
Citizen Initiated Encounter

No

Yes
Police Initiated Encounter

No

Yes
Unknown Encounter Initiation

No

Yes
Routine Approach to Scene

No

Yes
Priority Call Approach to Scene

No

Yes

Lights and Sirens Approach to Scene

No

Yes
Unknown Approach to Scene

No

Yes
Duty Status of Ofﬁcer

On Duty

Off Duty
Ofﬁcer Used Back-Up

No

Yes
Number of Ofﬁcers
OFFICER CHARACTERISTICS
Ofﬁcer Age
White Ofﬁcer

No

Yes

134

e—L

A

—L

A

o—B

—l

A

—l

_l

A

—l

5932
3296
2636

5932
5670
262

5932
371 7
221 5

5932
5113
819

5932
1 333
4599

5932
491 0
1 022

5932
5371
561

5932
5256
676

5932
5726
206

5932
4432
1 500

5932

5932

5932
1 394
4538

0.44

0.04

0.37

0.14

0.78

0.17

0.1

0.11

0.04

0.25

2.52

32.4

0.77

0.5

0.21

0.48

0.35

0.42

0.38

0.29

0.32

0.18

0.43

1.81

6.57

0.42

Table 6 (cont’d)

Black Ofﬁcer
No
Yes
Hispanic Ofﬁcer
No
Yes
Other Race Ofﬁcer
No
Yes
Ofﬁcer Gender
Female
Male

Ofﬁcer Demeanor toward Suspect
Not Antagonistic
Antagonistic

Prior Medical Attention to Ofﬁcer
No
Yes

Number of Surveys Completed by Ofﬁcer

SUSPECT CHARACTERISTICS
Suspect Age
White Suspect
No
Yes
Black Suspect
No
Yes
Hispanic Suspect
No
Yes
Other Race Suspect
No
Yes
Missing Race Suspect
No
Yes

135

—8 A

40

A

40

40

.—L

A

.—I

5932
5131
801

5932
5490
442

5932
5781
1 51

5932
628
5304

5932
5892
40

5932
5327
605

5932

5932

5932
3638
2294

5932
3271
2661

5932
5072
860

5932
5815
1 17

5932
5678
254

0.14

0.08

0.03

0.89

0.01

0.1

4.81

31.19

0.39

0.45

0.15

0.02

0.04

0.34

0.26

0.16

0.31

0.08

0.3

3.8

9.81

0.49

0.5

0.5

0.14

0.2

Table 6 (cont’d)

Suspect Gender

Female

Male
Suspect Known to be Assaultive

No

Yes
Suspect Known to Carry Weapon

No

Yes

Suspect Known to be a Gang Member
No
Yes

Suspect is Intoxicated
No
Yes

Victim is Stranger to Suspect
No
Yes

Victim is Friend to Suspect
No
Yes

Victim Is Family to Suspect
No
Yes

Victim has Unknown Relationship to Suspect
No
Yes

No Bystanders Present
No
Yes

Bystanders have Unknown Relationship to Suspect
No
Yes

Bystanders are Strangers to Suspect

No
Yes

136

.30 .3

A

..so ..Lo .3 .30

A0

40

5932
1 172
4760

5932
5543
389

5932
5655
277

5932
5305
627

5932
2259
3673

5932
2886
3046

5932
4340
1 592

5932
4638
1 294

5932
2985
2947

5932
2626
3306

5932
5514
418

5932
5070
862

0.8

0.07

0.05

0.11

0.62

0.51

0.27

0.22

0.5

0.56

0.07

0.15

0.4

0.25

0.21

0.31

0.49

0.5

0.44

0.41

0.5

0.5

0.26

0.35

Table 6 (cont’d)

Bystanders are Friends to Suspect 5932
No 0 5055
Yes 1 877
Bystanders are Family to Suspect 5932
No 0 5463
Yes 1 469
Suspect Demeanor toward Ofﬁcer 5932
Not Antagonistic 0 4694
Antagonistic 1 1238
Suspect Physical Resistance 5932
No 0 521 1
Yes 1 721

0.15

0.08

0.21

0.12

0.35

0.27

0.41

0.33

In addition to the Level-1 variables described in Table 6, the analyses of

this dissertation make use of several Level-2 (i.e. neighborhood-level)

independent variables. As discussed previously, the following variables were

submitted to a factor analysis: (1) percent African-American population within the

census tract; (2) density of individuals 17 years and younger within the census

tract; (3) percent of individuals within the census tract living below the poverty

level; (4) percentage of individuals over age 16 in the labor force who are

unemployed within the census tract; (5) percentage of female-headed

households within the census tract; and (6) percentage of individuals within the

census tract receiving public assistance income.79

 

79 Results of the factor analytic process can be found in Appendix B, which provides the correlation matrix
for the six variables used in the process, the component matrix demonstrating factor loadings, and the
traditional Kaiser-Meyer—Olkin and Bartlett’s tests for measuring the adequacy of the factor analytic model.

137

Table 7: Sample Characteristics (Level-2 Independent Variables)

VARIABLE NAME

Concentrated Disadvantage Factor Score

Index 1 Crime Rate (per 1,000 Census Tract
Residents)

Census Tract Located in Colorado Springs, CO

Census Tract Located in the City of San Diego, CA

Census Tract Located in Dallas, TX

Census Tract Located in the County of San Diego,
CA

Census Tract Located in Charlotte, NC

Census Tract Located in St. Petersburg, FL

Results

VALUE

No
Yes

No
Yes

No
Yes

No
Yes

No
Yes

No
Yes

385

385

385
337
48

385
323
62

385
296
89

385
324
61

385
31 2
73

385
333
52

MEAN

0

11.96

0.12

0.16

0.23

0.16

0.19

0.14

SD

155.57

0.33

0.37

0.42

0.37

0.39

0.34

Having presented the sample characteristics for the data used in this

dissertation, it is now appropriate to focus on the results of the multilevel

analyses performed within the HLM 5.05 software package. The hierarchical

linear modeling process enables the researcher to use a stage modeling

procedure for linear regression models, such as that used for the dependent

variable of maximum force. These results will be discussed ﬁrst, followed by a

138

discussion of the non-linear regression model for the dependent variable of
physical force plus threats. It should be noted here that although results are
presented in the following tables for the 6-site model as well as for all ﬁve sites
individually, only the results for the 6-site model are discussed below. This is due
to the fact that the results are generally consistent across all models, and thus
model parsimony and issues of generalizability lead to a focus on the 6—site

model.

Dependent Variable Analyses for Maximum Force:

As is appropriate for multilevel data, this dissertation used hierarchical
linear modeling techniques to determine the effects of encounter-level and
neighborhood-level variables on the dependent variable, maximum force. In
order to estimate these effects properly and assess the utility of the hierarchical
modeling process, HLM 5.05 allows the user to essentially engage in step-wise
modeling of the dependent variable. The initial step in this procedure is to
estimate an unconditional means model, which is equivalent to a one-way
analysis of variance (ANOVA), for maximum force at the encounter level. This
model includes the intercept for the encounter-level model as the only parameter
(with the intercept of the Level-1 intercept, designated as Goo, as a Level-2
parameter)“, effectively testing neighborhood variance in the mean level of
maximum force. The unconditional means model provides this dissertation with

several multilevel diagnostic statistics which are necessary for determining the

 

8° Note that Goo, while included in the model, is rarely of interest as it only determines whether the grand
intercept of maximum force is signiﬁcantly different from zero.

139

validity of using a hierarchical model given the data. The ﬁrst of these diagnostic
statistics is the reliability estimate, which measures the average reliabilities
across neighborhoods (i.e. the Level-2 units) to determine if the sample mean is
a reliable indicator of the true neighborhood mean. This statistic is calculated as:
A = Too/[Too + (oz/n11]. where T00 is the neighborhood-level variance, 02 is the
encounter-level variance, and n,- is the nth arrest in the jth neighborhood. The
reliability estimate obtained for maximum force (A = .67) indicates that using HLM
with these data would result in modeling neighborhood-level effects with a high
degree of precision.

The second diagnostic statistic provided by the unconditional means
model is that of the intraclass correlation coefﬁcient (ICC), which must be
manually calculated as: p = Too/(Too + o’), where T00 is the variance component of
the neighborhood mean, and 02 is the variance component of the Level-1 effect.
For the maximum force model, the ICC was .17, indicating that 17% of the total
variance of maximum force is between neighborhoods. While this is a large
enough proportion of the variance to be of interest, Terrill and Reisig (2003: 305)
correctly note that the question remains as to whether there is enough variation
in maximum force between neighborhoods to accurately model as a function of
Level-2 variables (i.e. neighborhood characteristics). This question is answered
by the Chi-square statistic (X2 = 1648.59), which is statistically signiﬁcant (p <
.000) for the unconditional means model. This result allows us to reject the null

hypothesis of no difference in maximum force between neighborhoods. Taken

140

together, the reliability estimate, intraclass correlation coefﬁcient, and Chi-square
value clearly indicate that the data is amenable to HLM analyses.

After determining that the data are amenable to HLM analyses, the ﬁnal
step is to estimate the full model. The latter, then, predicts maximum force from
all of the Level-1 (i.e. encounter-level) variables and Level-2 (i.e. neighborhood-
level) variables, with special attention paid to the cross-level interactions between
neighborhood characteristics and ofﬁcer race, neighborhood characteristics and
suspect race, and neighborhood characteristics and suspect demeanor. This full
model, described as an intercepts-and-slopes-as-outcomes model, also provides
us with several diagnostic statistics. The ﬁrst of these is the test of homogeneity
of Level-1 variance, which determines if estimates are biased due to
homogeneous Level-1 coefﬁcients. The Chi-square statistic provided by HLM (X2
= 49.02) is statistically signiﬁcant (p < .000), indicating that the data are not
homogeneous, and thus there is signiﬁcant variation between neighborhoods.
Further conﬁrmation is provided by the summary of model ﬁt which provides a
comparison between homogeneous and heterogeneous variances. The Chi-
square statistic for this diagnostic test (X2 = 1112.04) is also statistically
signiﬁcant (p < .000), indicating that the model with heterogeneous variance is
preferable. Accordingly, the parameter estimates and standard errors reported
here are from the latter model. As noted throughout this dissertation, the effects
of ofﬁcer race, suspect race and suspect demeanor are hypothesized to vary
between neighborhoods for a variety of theoretical reasons. The ﬁnal diagnostic

statistic is the Chi-square of the variance component estimates for these random

141

effects. The Chi-square for Black ofﬁcers is not statistically signiﬁcant (X2 = 1.59;
p > .500), nor is the Chi-square for ofﬁcers of the “Other” racial category
signiﬁcant (X2 = 7.97; p> .500). However, the Chi-square value for Hispanic
ofﬁcers is signiﬁcant (X2 = 9.38; p = .050). With respect to suspect race, the Chi-
square value for Black suspects (X2 = 14.66; p = .006) is signiﬁcant, while the
Chi-square values for Hispanic suspects (X2 = 7.13; p = .128) and suspects of the
“Other” (x2 = 7.58; p = .107) or “Missing” (x2 = 5.92; p = .204) racial category are
not signiﬁcant. Finally, the Chi-square value for suspect demeanor (X2 = 28.67; p
< .000) is signiﬁcant. These results allow us to reject the null hypothesis that
there is no signiﬁcant variation among the slopes. This conﬁrms that these Level-
1 independent variables are appropriately modeled as having slopes that vary
randomly with respect to the Level-2 independent variables of interest.

As the validity of the full model with two Level-1 random effects has been
determined through a number of diagnostic statistics, Table 8 presents the

results of the HLM analyses for the appropriate model.

Table 8: Full Model for Maximum Force

VARIABLE S-SITE SD CMPD CSPD DPD SPPD
Level-1 MODEL
Intercept 25.49* 29.04* 21 .59* 20.30* 24.98“ 3237*
NATURE OF THE LOCA TION
Location known for criminal
activity 039* -0.1 -0.08 0.14 0.54 0.28
Location known to be hazardous -0.08 0.17 0.74 2.29" -0.52 0.25
Incident took place inside 0.13 -0.35 0.21 1 .15* 1.1 6* 0.09
Wsibiiity -0.14* -0.12 -0.11 «23* -0.1 -.18*

142

Table 8 (cont’d)
NATURE OF THE ENCOUNTER

Violent offense

Incident took place on a
weekend

Bystander demeanor

Number of suspects

POLICE MOBILIZATION
Suspect already in custody
Citizen-initiated incident
Police-initiated incident
Other initiation to incident

Priority approach to incident
Lights and sirens approach to
incident

Other non-routine approach to
incident

Ofﬁcer off-duty
Ofﬁcer called for back-up
Number of ofﬁcers

OFFICER CHARACTERISTICS

First ofﬁcer age
African-American ofﬁcer

Hispanic ofﬁcer
Other ofﬁcer race
Male ofﬁcer

Ofﬁcer demeanor toward suspect
Ofﬁcer received prior medical
auenﬂon

Number of surveys completed

SUSPECT CHARACTERISTICS
Suspect age

African-American suspect
Hispanic suspect

Other suspect race

Missing suspect race

Male suspect

Suspect believed to be
assaultive

Suspect known to carry weapon
Suspect known to be a gang
member

0.72*

0.15
-0.03

-0.01

-0.36
0.26

0.20
-0.41
1.14*

1 .22*

-0.22
1.08*
1 .05*
0.72*

-.06*
-0.20
-0.38

1.32

0.56*

2.12

0.33
-0.02

-0.02

.09
0.54
0.60
-0.02

0.93*

0.26
1.65*

0.52

1.19*

0.08
0.67

0.38*

-0.98
0.42

-0.4
-1.59*
1.82*

1 .79*

0.1
-0.7
1.31*
.54*

-0.02
-0.37

0.45
1.11
0.14
-1.51

-0.44
-0.03

0.002

-0.91
0.28
-0.2

-2.75

-0.1

0.11
3.83*

-0.51

143

0.83

0.28
-1

0.22

-1 .40*
0.62
1 .38*
-4.45*
2.38*

0.49

1.16
0.11
1.31*
.53*

-0.02
-0.26

-7.35*
12.41*
0.78
4.53

2.76*
0.14

-0.03
0.54

2.64

1.41*

-0.51
0.98

0.5

1 .72*

0.06
0.26
0.27

0.21
0.07

0.87
-0.19
0.37

1.41

-2.47*
1.52
1.19
1.63

-0.04
-0.9

-1.6
-0.21
1.33
-2.27

1.05
-0.008

0.02

0.08
0.62
0.37
0.07

0.7

-0.28
7.20*

1.87

-0.16

0.83
0.12

.40*

-1.68*
-0.16

0.28
0.1
1.69*

1 .84*

-0.01
1.69
1.14*
.34*

-0.03
-0.14

0.73

-2.2

0.55
3.87*

-0.56
0.07

-0.02

0.33
0.01

-1.95
0.93

1.14*

—0.36
2.31

-1.25

.88*

—0.36
0.41
-.41*

-0.17
0.48

0.07
-0.49
0.62

1 .25*

0.01
-1.73*
1 .15*

.58*

-.00*
0.18
-1.34
1.57
0.32

5.60*

0.61
-0.028

-0.02

.72*
0.28
1.42
-1.18

1 .20*

0.11
1.24

-1 .54*

Table 8 (cont’d)
Suspect intoxicated
Wctim is friend to suspect
Victim is family to suspect

Victim has unknown relationship

to suspect

Bystanders have unknown
relationship to suspect
Bystanders are strangers to

suspect
Bystanders are friends to

suspect

Bystanders are family to suspect
Suspect demeanor toward ofﬁcer

Suspect uses physical force

Level-2

Colorado Springs, CO
Charlotte, NC

Dallas, TX

St. Petersburg, FL
San Diego, CA

Violent crime rate
Concentrated disadvantage

Spatial Error Term

Interaction Effects
African-American ofﬁcer **
Concentrated Disadvantage
Hispanic ofﬁcer **
Concentrated Disadvantage
Other ofﬁcer race “
Concentrated Disadvantage
African-American suspect **
Concentrated Disadvantage
Hispanic suspect **
Concentrated Disadvantage
Other suspect race **
Concentrated Disadvantage
Missing suspect race **
Concentrated Disadvantage

Suspect demeanor toward ofﬁcer

Concentrated Disadvantage

Suspect demeanor toward ofﬁcer

Violent Crime Rate
* p < .05

0.29
-0.59*
-0.51*

~0.13
0.74
0.4-4*

0.23
0.01

1 .61*
5.92*

0.30
3.77*

10.32*

4.44*

-0.0005

-0.05
-0.04

0.30
-0.07
0.19
0.12
0.16
-0.06
0.46

0.002

0.003

0.08
-0.17
-1 .33*

0.1
-0.01
1.22*

0.77
0.59

0.79
5.82*

##ﬁﬂ:

#

-0.0002
0.03

-1.14*

-0.04
0.09
-0.62
0.42
0.13
0.21

1.24

0.001

0.29

# variable removed from model

144

0.47 0.54
-1 .42" 0.21
-1.88* -0.97

-0.98 -0.78

1.47 -2.32

0.63 0.76

0.54 0.21

0.17 -0.11

0.2 5.49*
8.67* 10.20*
# #

# #
# #
# #
# #
-0.0003 .17*

-0.87 -0.22

0.23 0.54

1.08 1.90

-3.32 0.37

-3.79 -1.20

0.75 1.08

# 0.50
-1.64 1.41
# -0.31
-0.0003 -0.40*
0.04 -1.34

0.55
-0.99
-0.55

0.42
1.97
0.56

0.51
0.89

1 .63*
8.09*

##1##“:

0.000002

0.39
-0.07

0.14
-0.83
-1.30
0.22
-0.13
0.47

-0.40

0.00008

0.03

0.53
-.69*
-0.45

-0. 13
-0.29
-0.29

-0.22
-0.44

.98*
1 .33*

%%¥%%

-.10*
0.51

-0.39

-0.29
-0.001
0.71
—0.09
0.07
-1.95
2.36

0.04

0.33

Signiﬁcant Findings from the Maximum Force Model

The coefﬁcients presented in Table 8 represent some interesting ﬁndings
concerning the predictors of the maximum level of force used by police ofﬁcers
during arrest situations. The intercept, that is, the level of maximum force when
all other variables are constrained to be null, is 25.49 which falls between “ofﬁcer
verbally threatens suspect” and “ofﬁcer pushes suspect” on the maximum force
ranking. With respect to neighborhood-level variables, the dummy-coded
variables for PUF study site predict that, relative to Colorado Springs, CO, the
four other study sites experience signiﬁcantly higher levels of maximum force
(Charlotte, NC = .30; Dallas, TX = 3.77; St. Petersburg, FL = 10.32; San Diego,
CA = 4.44). Clearly, the most notable of these effects is for the St. Petersburg, FL
police department, which had a level of maximum force 10.32 points higher than
Colorado Springs on the force ranking, equivalent to an average maximum force
of 35.81 (falling between “ofﬁcer uses other tactic” and “ofﬁcer displays chemical
agent”). More importantly, in the context of this dissertation’s hypotheses, the
coefﬁcients for the Index I crime rate (-.001) and concentrated disadvantage (-
.05) were not statistically signiﬁcant at the .05 level, nor was the spatial error
variable (-.04). These ﬁndings demonstrate that neighborhood characteristics
which represent social disorganization (e.g. high levels of crime, high levels of
concentrated disadvantage) do not, in and of themselves, increase the average
level of maximum force between neighborhoods, controlling for all encounter-

level independent variables.

145

With respect to the encounter-level variables used in this dissertation, the
HLM analyses provide some interesting ﬁndings as well. With respect to those
encounter-level variables representing the nature of the arrest location, there are
several signiﬁcant predictors of maximum force. The coefﬁcient for a known
criminal location (.40) was statistically signiﬁcant, demonstrating that as an
ofﬁcer’s sense of safety diminishes, there is a predicted increase in the average
level of maximum force. This is bolstered by the coefﬁcient for visibility at the
arrest location (-.14), which predicts that as visibility improves (thereby potentially
increasing an ofﬁcer’s sense of safety), the average level of maximum force
decreases. However, the coefﬁcient for the arrest occurring inside (.13) and for a
known hazardous location (-.08) were both not statistically signiﬁcant.

There were also several encounter-level variables which focused on the
nature of the encounter that were found to be statistically signiﬁcant. The
coefﬁcient for whether the arresting offense was a violent crime (.72) was
statistically signiﬁcant, again demonstrating the idea that an ofﬁcer’s concerns
regarding personal safety predict higher levels of force. The coefﬁcients for the
remaining variables related to the nature of the encounter were not statistically
signiﬁcant. Thus, whether the offense occurred on a weekend (.15), the
demeanor of bystanders toward the police (-.03), and the number of suspects (-
.01) did not have a signiﬁcant effect on the maximum level of force used during
the encounter.

The next grouping of variables is concerned with how the police ofﬁcers

were mobilized, approached the situation, and responded to the situation from an

146

organizational standpoint. The coefﬁcient for the custody status of the suspect (-
.36) was negative, predicting that when a suspect is already in custody when an
ofﬁcer arrives, the average level of maximum force decreases relative to when
the suspect is not already in custody, but was not statistically signiﬁcant. With
regards to how the ofﬁcers came to be involved in the encounter, the coefﬁcients
for citizen initiation of the encounter (.26), ofﬁcer initiation of the encounter (.20)
and an unknown method of mobilization (-.41) were not statistically signiﬁcant.
Regarding the ofﬁcers’ actual approach to the encounter, a priority approach
(1.14) and the use of lights and sirens (1.22) were both statistically signiﬁcant
predictors of an increase in the maximum level of force, while the coefﬁcient for
an unknown (non-routine) approach to an encounter (-.22) was not statistically
signiﬁcant. The duty status of the ofﬁcer (1.08) was a signiﬁcant predictor of the
level of force used in an encounter, predicting that ofﬁcers on duty experience
increased levels of force. In addition, the use of back-up ofﬁcers (1 .05) and the
total number of ofﬁcers at the encounter (.72) were both statistically signiﬁcant,
again predicting that when an ofﬁcer fears for their safety the level of force used
increases.

Within the maximum force model, the HLM analyses also revealed
several signiﬁcant ofﬁcer characteristics. The coefﬁcient for ofﬁcer age (-.06) was
negative and statistically signiﬁcant, predicting that older ofﬁcers use a lower
level of average maximum force across neighborhoods. With respect to ofﬁcer
race, the coefﬁcients for Black ofﬁcers (-.20), Hispanics ofﬁcers (-.38), and

ofﬁcers of the “Other” racial category (1.33) were not statistically signiﬁcant,

147

indicating that the race of the ofﬁcer did not have an effect on the average level
of maximum force. However, the coefﬁcient for male ofﬁcers (.56) was
statistically signiﬁcant, thus predicting that male ofﬁcers use higher levels of
maximum force than female ofﬁcers. Finally, the coefﬁcients for ofﬁcer demeanor
toward the suspect (2.12), whether or not the ofﬁcer had received prior medical
attention (.33), and the total number of surveys completed by the ofﬁcer (-.02)
were not statistically signiﬁcant, with the latter two demonstrating that an ofﬁcer’s
activity level was not related to their levels of maximum force.

In regards to suspect characteristics, there are several signiﬁcant
predictors of maximum force. The coefﬁcient for the suspect’s age (-.02) was not
signiﬁcant. More importantly, the coefﬁcients for a Black suspect (.09), a Hispanic
suspect (.54), a suspect in the “Other” racial category (.60), and a suspect of
unknown race (-.03) were also not statistically signiﬁcant, indicating that a
suspect’s age and race do not have an effect on the maximum level of force used
during an encounter. However, the coefﬁcient for a male suspect (.93) was
statistically signiﬁcant, predicting that male suspects have a higher level of
maximum force used against them than do female suspects. Other suspect
characteristics were also found to have a statistically signiﬁcant effect on the
average level of maximum force used in an encounter. In particular, if a suspect
was known to carry a weapon (1.65), was a statistically signiﬁcant predictor of
the maximum level of force, while if the suspect was intoxicated (.29), if the
suspect was believed to be assaultive (.26), and if a suspect was known to be a

gang member (.52) were not statistically signiﬁcant. In terms of their relationships

148

to victims and bystanders at the encounter, the coefﬁcients for a suspect and
victim who were friends (-.59), a suspect and victim who were family (-.51), and a
suspect and bystanders who are strangers (.44) were all statistically signiﬁcant
predictors of the maximum level of force. Speciﬁcally, relative to a suspect and
victim who were strangers, if the suspect and victim were friends or family, there
was a predicted decrease in the average level of maximum force. If the suspect
and the bystanders were strangers, however, there was a predicted increase in
the average level of maximum force used in the encounter. One should also note
that the coefﬁcients for a suspect and victim with an unknown relationship (-.13),
a suspect and bystanders with an unknown relationship (.74), a suspect and
bystanders who are friends (.23), and a suspect and bystanders who are family
(.01) were not statistically signiﬁcant. Most importantly in the context of this
dissertation, the coefﬁcients for suspect demeanor (1.61) and suspect physical
resistance (5.92) represent relatively large effect sizes, indicating that these
characteristics are predicted to exert the largest effects on the average level of
maximum force between neighborhoods.

The cross-level interactions in which Level-1 variables were allowed to
vary randomly across neighborhoods provided the main impetus for this
dissertation, and the HLM analyses again prove instructive. The effects of ofﬁcer
race were allowed to vary across neighborhoods, and these interactions are
examined here. The coefﬁcients for ofﬁcer race were .30 (Black ofﬁcers —
concentrated disadvantage), -.07 (Hispanic ofﬁcers — concentrated

disadvantage), and .19 (“Other” racial category ofﬁcers — concentrated

149

disadvantage). None of the coefﬁcients for the interaction between ofﬁcer race
and concentrated disadvantage reached statistical signiﬁcance, indicating that
ofﬁcer race does not have a signiﬁcant effect on the average level of maximum
force used across different types of neighborhoods. The second variable which
was allowed to vary randomly was suspect race, and this was examined in a
cross-level interaction with the measure of concentrated disadvantage. The
coefﬁcients for these interactions were .12 (Black suspects — concentrated
disadvantage), .16 (Hispanic suspects - concentrated disadvantage), -.06
(“Other” racial category suspects — concentrated disadvantage), and .46
(“Unknown” racial category suspects — concentrated disadvantage), none of
which reached statistical signiﬁcance, indicating that suspect race does not have
a signiﬁcant effect on the average level of maximum force used across different
types of neighborhoods. With respect to suspect demeanor, two cross-level
interactions were examined: (1) the effects of suspect demeanor in high-crime
neighborhoods, as measured by the Index I crime rate; and (2) the effects of
suspect demeanor in neighborhoods high in concentrated disadvantage. The
coefﬁcients for these two cross-level interactions were .002 (for suspect
demeanor - Index | crime rate) and .003 (for suspect demeanor — concentrated
. disadvantage), neither of which reached statistical signiﬁcance, thereby
indicating that suspect demeanor, although properly modeled as a random effect,
does not have a signiﬁcant effect on the average level of maximum force across

different types of neighborhoods.

150

Dependent Variable Analyses for Physical Force p_lus Threats:

Consistent with regression techniques for dichotomous dependent
variables, this dissertation used a subset of HLM analyses known as hierarchical
generalized linear modeling (HGLM) to estimate the model with the dependent
variable of physical force plus threats. The latter technique was used, as the use
of the standard HLM model would be inappropriate for the following reasons: (1)
given a binary outcome measure (i.e. force vs. no force), the random effect can
take on only one of two values, and thus cannot be normally distributed; (2) the
random effect cannot have homogeneous variance, as it depends upon the
predicted values; and (3) effect sizes would be uninterpretable, as there are no
restrictions on predicted values in the standard model but due to the presence of
a binary variable there can be no values less than zero or greater than one (Bryk
and Raudenbush 1992). In the output produced by HLM, therefore, the OLS
estimates and the estimates from the linear model with the identity link function
can be ignored, as there is no meaningful interpretation of linear model results for
a dichotomous dependent variable (Gaitanis, 2003).

For all analyses regarding this dependent variable, coefﬁcients and other
statistics were taken from the unit-speciﬁc model with the logit link function. The
latter contains the random effect from the level-2 model, and is therefore a
prediction of the prevalence of force in a neighborhood typical of the independent
variables in the model. Unit-speciﬁc models, then, are more appropriate in
situations where the researcher is interested in the unique effects of level-2 units

(i.e. neighborhoods) on the dependent variable. As the focus of this dissertation

151

is the effects of neighborhood characteristics on the severity and prevalence of
force, unit-speciﬁc models are used.81

As with the previous dependent variable, HLM was again used to produce
an unconditional means model for the physical force plus threats dependent
variable.82 In the unconditional means model, the reliability estimate for the
intercept was .192. As Gaitanis (2003: 10) notes, within HGLM analyses, a low
reliability coefﬁcient in the unconditional model indicates that the model is an
insufﬁcient predictor of the outcome variable, and is insufﬁcient to explain
variation in the dependent variable of physical force plus threats. Thus, a full
model is tested, with predictors at both Level-1 (arrest encounters) and Level-2
(neighborhoods). This model is estimated with a random effect term in the Level-
2 equation, which implies that the effect of Level-1 variables (speciﬁcally, ofﬁcer
race, suspect race and suspect demeanor) vary depending on the Level-2 unit
(i.e. the speciﬁc type of neighborhood) being considered. Unlike the basic HLM
model, the full model tested in HGLM does not produce diagnostic statistics
which can be used to assess the appropriateness of the model. However, Bryk
and Raudenbush (1992) note that the deviance statistic is often used as a proxy
for a goodness-of-ﬁt statistic in HGLM, and the magnitude of the current

deviance statistic (4017.40) indicates that the full model is preferable to the

 

8‘ The population-average model, in contrast, would be used where the researcher is interested in obtaining
results that have maximum generalizability to all possible level-2 units (i.e. all possible neighborhoods).
Note that the population-average model is less efficient when the full model with random variables within
the nonlinear function is the correct model, as is the case here.

82 Note that the dependent variable was modeled as a Bernoulli distribution, used to indicate a binary model
where the outcome is only measured once, as force was only measured as having occurred at any point
during the arrest and therefore was only measured one time per arrest encounter.

152

unconditional model.83 In addition, based on the previous reliability estimate for
the model which did not include Level-2 predictors, coupled with the fact that
HGLM is the most appropriate statistical technique to test the full model, we can
be reasonably certain in stating that the results obtained through the HGLM
analyses represent the most accurate model description for the dependent
variable of physical force plus threats. Therefore, Table 9 presents the results of

the HGLM analyses for the full (i.e. intercepts and slopes as outcomes) model.

Table 9: Full Model for Physical Force Plus Threats

6-SITE SD CMPD CSPD DPD SPPD

VARIABLE MODEL

Level-1

intercept 001* 0.05* 0.02* 0.04* 0.07* 0.01*
NATURE OF THE LOCATION

Location known for criminal

activity 1.17 1.29 0.92 1.09 1.03 1.54*
Location known to be

hazardous 0.97 0.65 1.13 1.13 0.85 1.20
Incident took place inside 1.02 0.69 0.78 0.93 1.30 1.37
Visibility 0.96* 0.98 0.97 0.97 0.97 0.96
NATURE OF THE

ENCOUNTER

Violent offense 1 .34* 1.59 1.42 1.1 1 0.86 1.15
Incident took place on a

weekend 1.15 0.95 1.10 1.11 1.26 1.09
Bystander demeanor 1.17 1.18 0.94 0.62 1.70 1 .80*
Number of suspects 1.00 1.17 1.08 1.03 1.05 0.80*

 

*3 The deviance statistic is calculated as D 1 — 0;, which approximates by X2 (P2 — P,), where D. = -210g
Likelihood for Model 1 and D2 = -Zlog Likelihood for Model 2.

153

Table 9 (cont’d)
POLICE MOBILIZA TION
Suspect already in custody
Citizen-initiated incident
Police-initiated incident
Other initiation to incident

Priority approach to incident
Lights and sirens approach to
incident

Other non-routine approach to
incident

Ofﬁcer off-duty
Ofﬁcer called for back-up
Number of ofﬁcers

OFFICER CHARACTERISTICS
First ofﬁcer age

African-American ofﬁcer
Hispanic ofﬁcer
Other ofﬁcer race

Male ofﬁcer
Ofﬁcer demeanor toward

suspect

Ofﬁcer received prior medical
attention

Number of surveys completed

SUSPECT
CHARACTERISTICS
Suspect age
African-American suspect
Hispanic suspect

Other suspect race
Missing suspect race

Male suspect
Suspect believed to be
assaultive

Suspect known to cany weapon
Suspect known to be a gang
member

Suspect intoxicated
Victim is friend to suspect

Victim is family to suspect
Victim has unknown relationship
to suspect

Bystanders have unknown
relationship to suspect

0.63*
1.23

1 .48*
1.04
1.61*

1 .41*

1 .45*
1.21

1.69*
1.16*

0.99
1.06
1.72*
1.32
2.02*

1.77
1.34*

1.00

1.00

1.09
1.13

1 .35
1 .46
1 .47*

1.03
1.91*

0.93
1.35*
0.93
1.00

0.97
1.71*

0.66
1.82

1 .37
.48*
2.60*

1.59

1.87*
0.90

1.72*
1 .13*

0.99
0.83
1.67
1.60
0.71

0.39
0.92

0.96

0.99

1.54
1.24

1.38
0.00
1.27

1.24
3.51*

0.92
0.99
0.99
.52*

0.97
0.94

154

0.81
0.99

1 .88*
2.02
2.27*

0.68

1.52
2.01

1 .58*
1 .09*

0.99
1.92*
1.10
4.15
1.36

7.25*
2.63*

1.03

0.99

1.27
#

4.15
#
1.89*

0.85
1.6

1.32
1.07
.64*
1.23

0.85
2.33*

0.74
1.54

1.26
1.20
1.05

1 .48

0.97
1.00

1.31
1.23*

1.00
0.76
1.53
1.47
1.56

0.86
1.41

0.98

1.00

1.31
0.92

2.74
1.29
0.99

1 .82
3.40*

0.86
1.32
1.15
0.78

0.80
0.76

.51*
0.71

0.77
1.05
1.35

1.36

1.15
1.30

1.30
1.03

0.99
1.29
2.08*
0.07
1.31

0.87
0.86

1.00

1.00

.55*
0.72

1.51
1.66
1.34

0.77
2.71*

0.32
1.14
0.99
0.92

1.13

1.44

0.69
0.68

1 .69*
0.95
1 .66*

1.34

1.46
0.90

1.70*
1 .13*

0.99
1.08
0.42
3.51
4.78*

6.97
1 .54*

1.00

0.99

0.93
1.97

1.32
1.88
1.32

1.00
1.36

0.75
1.72*
0.9
1.18

0.96

1.74

Table 9 (cont’d)
Bystanders are strangers to

suspect 1.30* 2.14* 1.77* 1.12 1.03 1.25
Bystanders are friends to

suspect 1.16 1.41 1.3 1.35 1.00 1.09
Bystanders are family to

suspect 1.30 1.57 1.47 1.19 0.89 1.29
Suspect demeanor toward

ofﬁcer 2.50* 2.35* 2.48* 3.16* 1.95* 2.46*
Suspect uses physical force 10.35* 1 3.65* 7.1 8* 1 0.56* 8.38* 8.95*
Level-2

Colorado Springs, CO # # # # # #
Charlotte, NC 1.18 # # # # #
Dallas, TX 0.73 # # # # #
St. Petersburg, FL 1.31 # # # # #
San Diego, CA 0.99 # # # # #
Violent crime rate 1.00 1.00 1.00 1.05* 1.00 0.96
Concentrated disadvantage 1.10 1.14 0.76 0.96 1.17 1.53
Spatial Error Term 0.95 0.66 1.04 0.85 1.04 0.83
Interaction Effects

African-American ofﬁcer **

Concentrated Disadvantage 1.15 0.90 1.03 1.68 1.11 1.06
Hispanic ofﬁcer **

Concentrated Disadvantage 0.95 1.63* 0.24 0.80 1.04 1.08
Other ofﬁcer race **

Concentrated Disadvantage 1.02 0.79 1.46 1.49 1.05 0.75
African-American suspect **

Concentrated Disadvantage 0.96 0.82 1.29 1.30 0.92 0.78
Hispanic suspect **

Concentrated Disadvantage 0.88 0.80 # 1.04 0.83 1.33
Other suspect race ** '

Concentrated Disadvantage 0.58 0.62 1.33 1.10 0.31 0.31
Missing suspect race **

Concentrated Disadvantage 1.00 9.50 # 0.99 0.89 0.46
Suspect demeanor toward

ofﬁcer **

Concentrated Disadvantage 1.00 1.00 1.00 0.91 1.00 1.06
Suspect demeanor toward

ofﬁcer **

Violent Crime Rate 0.92 1.22 0.88 0.86 1.20 0.82
*p < .05 # variable removed from model due to multicollinearity

155

Signiﬁcant Findings from the Physical Force Plus Threats Model

The coefﬁcients presented in Table 9, which represent the coefﬁcients and
log-odds for each variable, present some interesting ﬁndings concerning the
predictors of the use of physical force or threats of physical force by police
ofﬁcers during arrest situations. The intercept, that is, the odds of force for a
reference arrest encounter (i.e. one with zeroes on all of the independent
variables) in a reference neighborhood (i.e. one with a mean of zero and random
effects of zero), is .01, predicting that a reference arrest encounter is 99% less
likely to contain a use of force behavior by an ofﬁcer. It should be noted here that
the percentages presented in discussion of the odds ratios represent an increase
or decrease in relation to the base rate for the reference category of that variable,
rather than an absolute increase or decrease. Thus, with respect to
neighborhood-level variables, the dummy-coded variables for PUF study site
predict that, relative to Colorado Springs, CO, arrest encounters in Dallas, TX
were 27% less likely (.73) and in San Diego, CA were 1% less likely (.99) to have
force used within them, while arrest encounters in Charlotte, NC (1 .18, or 18%
more likely), and St. Petersburg, FL (1.31, or 31% more likely), were more likely
to have force used within them, although none of these provided a statistically
signiﬁcant effect. In relation to this dissertation’s hypotheses on neighborhood
effects, neither the log-odds for the Index I crime rate (1.00, or equally likely) or
concentrated disadvantage (1 .09, or 9% more likely) were statistically signiﬁcant,
nor was the log-odds for the spatial error term (.95, or 5% less likely). As with the

previous dependent variable, these results indicate that neighborhood

156

characteristics representing social disorganization do not, in and of themselves,
predict increased use of force behavior in arrest encounters.

Regarding the effects of individual-level variables, analysis of the HGLM
results again provide us with some interesting ﬁndings. It is important to note that
these results are interpreted as the effects of a particular variable on the
prevalence of force for a reference arrest (i.e. zeroes on all other independent
variables) in any neighborhood. The ﬁrst group of independent variables
represents the nature of the arrest location, and three of the latter were not
statistically signiﬁcant. The log-odds for a known criminal location was 1.17 (17%
more likely to have force used within the encounter), while the log—odds for a
known hazardous location was .97 (3% less likely). In addition, the log-odds for
an arrest occurring inside was 1.02 (2% more likely). The only statistically
signiﬁcant variable related to the nature of the arrest location was for the visibility
at the arrest location (.96, or 4% less likely), which predicts that as visibility
decreases force is less likely to be used during an encounter.

The second group of independent variables is that which focuses on the
nature of the encounter. Again we note that the majority of these variables were
not found to have a statistically signiﬁcant effect. The log-odds for the crime
occurring on a weekend (1.15, or 15% more likely), the demeanor of bystanders
toward the ofﬁcer (1 .17, or 17% more likely), and the total number of suspects
within the encounter (1 .00, or equally likely) all predicted an increase in the
prevalence of force, but none of these results were statistically signiﬁcant. The

only statistically signiﬁcant variable related to the nature of the encounter was for

157

a violent criminal offense (1 .34, or 34% more likely), which predicts that an arrest
encounter for a violent crime is more likely to have force used within that
encounter than is an arrest encounter for a non-violent crime.

The next group of variables is concerned with how police ofﬁcers were
mobilized, approached the situation, and responded to the situation from an
organizational standpoint. Within this subset of individual-level variables there
were several independent variables which were not statistically signiﬁcant. These
included whether the encounter was initiated by the citizen (1.23, or 23% more
likely), unknown initiation to the encounter (1.04, or 4% more likely), and the duty
status of the ofﬁcer (1.21, or 21% more likely). However, a large proportion of this
subset of variables was statistically signiﬁcant. The log-odds for the custody
status of the suspect was .63, predicting that an arrest encounter with a suspect
already in custody was 37% less likely to have force used within that encounter
compared to an arrest encounter where the suspect was not in custody when the
ofﬁcer arrived. If the encounter was initiated by the police ofﬁcer, the log-odds of
force was 1.48, predicting that force was 48% more likely than in an arrest
encounter where the ofﬁcer was dispatched to the scene. If the ofﬁcer engaged
in a priority approach (i.e. above normal speeds) to the encounter, the log-odds
of force was 1.61, predicting that force was 61% more likely than in an arrest
encounter where the ofﬁcer engaged in a routine approach. If the ofﬁcer engaged
in an approach with lights and sirens, the log-odds of force was 1.41, predicting
that force was 41% more likely than in an encounter where the ofﬁcer engaged in

a routine approach. In addition, if the approach of the ofﬁcer was unknown (non-

158

routine), the log-odds of force was 1.45, predicting that force was 45% more
likely than in an arrest encounter where the ofﬁcer engaged in a routine
approach. The log-odds for an ofﬁcer’s use of back-up ofﬁcers was 1.69,
predicting that force was 69% more likely in an arrest encounter where the ofﬁcer
called for back-up. Finally, the log-odds for the number of total ofﬁcers at the
scene was 1.16, predicting that for each increase in the number of ofﬁcers, the
use of force was 16% more likely. It should be noted here that all of the
statistically signiﬁcant independent variables in the nature of mobilization subset
are related to an ofﬁcer’s perceived safety responding to, or during, an arrest
encounter.

HGLM analyses also revealed several interesting ﬁndings regarding ofﬁcer
characteristics. The log-odds for ofﬁcer age was .99, predicting that for each one
year increase in ofﬁcer age, force is 1% less likely in an arrest encounter. In
addition, the log-odds for the number of surveys completed by the ofﬁcer (a proxy
for ofﬁcer activity) was 1.00, predicting that for each additional survey the ofﬁcer
had completed, force was equally likely to be used in that arrest encounter.
However, neither of these variables reached statistical signiﬁcance, nor did
ofﬁcer demeanor (1.77 or 77% more likely when an ofﬁcer was antagonistic
toward the suspect), a somewhat surprising result. Of those ofﬁcer
characteristics that did reach statistical signiﬁcance, the log-odds for an ofﬁcer
having received prior medical attention due to a law enforcement-related injury
was 1.34, predicting that ofﬁcers who had received prior medical attention were

34% more likely to use force within an arrest encounter than an ofﬁcer who had

159

never received medical attention for a job-related injury. In addition, the log-odds
for a male ofﬁcer was 2.02, predicting that male ofﬁcers are twice as likely
(102%) than female ofﬁcers to use force within an arrest encounter. With regards
to ofﬁcer race, the log-odds for a Black ofﬁcer (1.06, or 6% more likely), and
ofﬁcer of the “Other” racial category (1.32, or 32% more likely) were not
statistically signiﬁcant. However, the log—odds for Hispanic ofﬁcers (1.72) were
statistically signiﬁcant, predicting that Hispanic ofﬁcers are 72% more likely to
use force relative to White ofﬁcers.

As with ofﬁcer characteristics, the analyses of suspect characteristics
reveal numerous independent variables which were not statistically signiﬁcant. Of
the suspect’s demographic characteristics, the log-odds for suspect age (1 .00, or
equally likely), a Black suspect (1 .09, or 9% more likely), a Hispanic suspect
(1.13, or 13% more likely), a suspect of the “Other” racial category (1 .35, or 35%
more likely), and a suspect of unknown race (1 .46, or 46% more likely) were not
statistically signiﬁcant. Of the independent variables dealing with the suspect’s
behavior, the log-odds for a suspect believed to be assaultive (1.03, or 3% more
likely) and a suspect believed to be a gang member (.90, or 10% less likely) were
not statistically signiﬁcant. Finally, of those independent variables related to the
suspect’s relationships, the log-odds for a suspect and victim who are friends
(.93, or 7% less likely), a suspect and victim who are family (1 .00, or equally
likely), a suspect and victim with an unknown relationship (.97,'or 3% less likely),
a suspect and bystanders who are friends (1 .16, or 16% more likely), and a

suspect and bystanders who are family (1 .30, or 30% more likely) were not

160

statistically signiﬁcant. However, there were numerous suspect characteristics
which did reach the level of statistical signiﬁcance. The log-odds for a male
suspect was 1.47, predicting that force was 47% more likely to be used against a
male suspect than against a female suspect. The log-odds for a suspect who
was known to carry a weapon was 1.91, predicting that force was 91% more
likely to be used in an arrest encounter where the suspect is known to carry a
weapon than in an arrest encounter where the suspect is not known to carry a
weapon. The log-odds for suspect sobriety was 1.35, predicting that suspects
who were intoxicated (through the use of either alcohol or drugs) were 35% more
likely to have forced used against them than suspects who were sober. Of the
relationship variables, the log-odds for a suspect and bystanders with an
unknown relationship (1.71, or 71% more likely), and a suspect and bystanders
who are strangers (1.43, or 43% more likely) were statistically signiﬁcant,
predicting that when there are bystanders present who are strangers, or have an
unknown relationship, to the suspect, force is more likely than in an arrest
encounter where no bystanders are present. Finally, regarding overt suspect
behavior toward the ofﬁcer, the log-odds for suspect demeanor was 2.50,
predicting that when a suspect is antagonistic toward the ofﬁcer, force is 150%
(two and a half times) more likely to be used than in arrest encounter where the
suspect is not antagonistic. In addition, the log-odds for physical resistance on
the part of the suspect was 10.35, predicting that when a suspect provides

physical resistance to the ofﬁcer, force is 935% (slightly under ten and a half

161

times) more likely to be used than in arrest encounter where the suspect does
not provide physical resistance to the ofﬁcer.

The cross-level interactions in which Level-1 independent variables were
allowed to vary randomly across neighborhoods provided the main research
questions for this dissertation, and the HGLM analyses illuminate several
interesting phenomena. It is important to note here that these parameters
represent the effects of the Level-2 unit (i.e. neighborhood characteristic) on the
effect of the Level-1 unit (i.e. individual-level independent variable) on the log-
odds of force. The effects of ofﬁcer race were allowed to vary across
neighborhoods, and these interactions are examined here. The log-odds for a
Black ofﬁcer-concentrated disadvantage interaction (1 .15, or 15% more likely),
and an ofﬁcer of the “Other" racial category-concentrated disadvantage
interaction (1.02, or 2% more likely), were not statistically signiﬁcant. However,
the log-odds for a Hispanic ofﬁcer-concentrated disadvantage interaction (.95)
were statistically signiﬁcant, predicting that for every one unit increase in the
measure of concentrated disadvantage, Hispanic ofﬁcers are 23% more likely to
use force in an arrest encounter than are White ofﬁcers in those same
neighborhoods. The second variable which was allowed to vary randomly was
suspect race, and this was examined in a cross-level interaction with the
measure of concentrated disadvantage. The log-odds for a Black suspect-
concentrated disadvantage interaction (.96, or 4% less likely), a Hispanic
suspect-concentrated disadvantage interaction (.88, or 12% less likely), a

suspect of the “Other” racial category-concentrated disadvantage (.58, or 42%

162

less likely), and a suspect of unknown race-concentrated disadvantage
interaction (1 .00, or equally likely) were all not statistically signiﬁcant. Finally, with
respect to suspect demeanor, cross-level interactions were examined with both
the Index I crime rate and the degree to which the neighborhood displays
concentrated disadvantage. The log-odds for the suspect demeanor-Index I rate
interaction was 1.00, predicting that force was no more or less likely in high-crime
neighborhoods for suspects who were antagonistic than for suspects in those
same neighborhoods who were not antagonistic, but this result was not
statistically signiﬁcant. The log-odds for the suspect demeanor-concentrated
disadvantage was .92, predicting that for every one-unit increase in concentrated
disadvantage within the neighborhood, antagonistic suspects are 8% less likely
to have force used against them than suspects who are not antagonistic and are
arrested in those same neighborhoods, but once again this result was not

statistically signiﬁcant.

163

Summary

Throughout this dissertation several hypotheses were presented relating
to the effects of ofﬁcer race, suspect race and suspect demeanor on police use
of force in varying types of neighborhoods. Analysis of the results from the HLM
and HGLM models reveals that of the ﬁve hypotheses tested, none were
conﬁrmed within the 6-site model. Although it had been hypothesized that ofﬁcers
would use more force in neighborhoods high in concentrated disadvantage, this
was not the case for the maximum force or the physical force plus threats
dependent variable. In fact, this hypothesis was not supported in either the 6-site
model or across any of the individual sites for either dependent variable.
Similarly, the hypothesis concerning minority ofﬁcers using less force than white
ofﬁcers in neighborhoods high in concentrated disadvantage was also found to
not exhibit a statistically signiﬁcant relationship in an overwhelming majority of
situations. In the 6-site model, across both dependent variables, neither African-
American ofﬁcers, nor Hispanic ofﬁcers, nor ofﬁcers of the “Other" racial category
used less force than White ofﬁcers. The only statistically signiﬁcant relationship
was found in San Diego, California, where Hispanic ofﬁcers were less likely (log-
odds of .95, or 5% less likely) than White ofﬁcers to use force for the physical
force plus threats dependent variable. The third hypothesis tested was that more
force would be used against suspects with poor demeanor who were
encountered in high-crime neighborhoods. Once again this hypothesis was not
supported in any of the models (6-site or individual) across either dependent

variable. It was also hypothesized that ofﬁcers would use more force against

164

suspects with poor demeanor who were encountered in neighborhoods high in
concentrated disadvantage due to their need to “save face” and maintain
authority in potentially dangerous environments. In the 6-site model, and across
the majority of the individual models, across both dependent variables, there was
no statistically signiﬁcant effect. However, in the only statistically signiﬁcant
effect, in Colorado Springs, Colorado, ofﬁcers actually used less force against
suspects with poor demeanor in neighborhoods high in concentrated
disadvantage (.40) than against suspects with poor demeanor encountered in
neighborhoods lower in concentrated disadvantage. Finally, it was hypothesized
that ofﬁcers would use less force against minority suspects in neighborhoods
high in concentrated disadvantage due to ecological contamination. Once again,
across all models and both dependent variables, there was no statistically
signiﬁcant effect.

The ﬁve hypotheses tested in this dissertation provided some interesting,
if somewhat confounding, results. Although none of these hypotheses were
found to exhibit statistically signiﬁcant relationships within the 6-site model, two
hypotheses were supported within individual sites, with one of those in the
expected direction and the other in the direction opposite of what had been
expected. In the concluding chapter, the results of the analyses will be
interpreted in the context of neighborhood effects on police use of force, with

special attention paid to the meaning of the results for the ﬁve hypotheses.

165

Chapter 5: Conclusions

Researchers in the ﬁeld of criminal justice have, for the past four decades,
recognized that the use of force by police ofﬁcers is a serious and controversial
issue. While there remains signiﬁcant debate about what constitutes
unreasonable force, the presence or absence of force more generally, as well as
the rankings of different methods of force on a continuum, has been well-
documented (Garner et al. 1995; Garner et al. 2002; Riksheim and Cherrnak
1993; Terrill and Mastrofski 2002). However, of the previous research examining
use of force behavior, only Smith (1986) and Terrill and Reisig (2003) have
acknowledged the potential effects of neighborhood characteristics on that
behavior.

In seeking to prevent abuses of force, it is imperative that social scientists
understand all potential inﬂuences on this type of behavior in order to more
accurately prevent and respond to these situations. The ﬁrst step is therefore
examining use of force behavior in a general sense to identify which individual-
level and which neighborhood-level characteristics may exert a signiﬁcant
inﬂuence on the prevalence and severity of force. As society becomes
increasingly concerned with how police ofﬁcers perform their duties, and law
enforcement agencies become increasingly concerned with meeting the needs of
their constituents through community-policing efforts, reducing abuses of force
and understanding when and why force has been used properly can address

both of these concerns.

166

The purpose of this research has been to identify those individual-level
(Level-1) and, more importantly, neighborhood-level (Level-2) variables which act
as predictors of either the prevalence or the severity of use of force behavior
within arrest encounters. Previous research has demonstrated that individual-
level predictors such as suspect race (Kavanagh 1997; Terrill and Mastrofski
2002; Worden 1995), suspect gender (Phillips and Smith 2000; Garner et al.
1995; Garner et al. 2002; Worden 1995), suspect demeanor (Bayley and
Garafalo 1989; Engel et al. 2000; Friedrich 1980; Garner et al. 2002; Kavanagh
1997; Worden 1995), and suspect physical resistance (Engel et al. 2000; Garner
et al. 1995; Garner et al. 2002; Phillips and Smith 2000; Terrill and Mastrofski
2002) have a statistically signiﬁcant effect on use of force behavior. However, all
of these studies failed to include neighborhood-level predictors, thus resulting in
a severe limitation of their explanatory power. As behavior in general, and police
behavior more speciﬁcally, has been shown to be responsive to neighborhood
effects (Black 1980; Klinger 1997; Mastrofski et al. 1999; McGarrell et al. 1997;
Weitzer 2000), the current research has sought to address this limitation through
the inclusion of such neighborhood-level variables.

This dissertation, then, has had several goals. The ﬁrst was to follow the
development of the literature on police behavior and the use of force. As with any
social scientiﬁc endeavor, it is imperative to understand past research in order to
provide a sold contribution to the ﬁeld. The second goal was to identify the
limitations of these prior studies, in the form of arguing that the failure to include

neighborhood-level variables has restricted our understanding of use of force

167

behavior. The third goal was to introduce relevant theoretical principles in order
to provide an underpinning for the current hypotheses. Using social control
theory (Black 1980), social disorganization theory (Park and Burgess 1924;
Sampson and Groves 1989; Shaw and McKay 1942; Stark 1987), and the
concept of tolerance (Kohfeld and Sprague 1990; Klinger 1997), the current work
proposed ﬁve hypotheses regarding use of force behavior during arrest
encounters. The fourth, and ﬁnal, goal was to demonstrate how the interaction of
individual-level and neighborhood-level variables within a hierarchical linear
model conﬁrm or disconﬁrrn the latter hypotheses.

Using data collected as part of the Police Use of Force (PUF) Study in six
American cities, a framework was developed for testing the effects of
neighborhood characteristics on use of force behavior, with special attention paid
to how those neighborhood characteristics may exert inﬂuence through several
individual-level characteristics. This research acts as a substantial contribution to
the current literature in that it moves beyond simple linear analyses of use of
force behavior, or even neighborhood-level analyses which have been mis-
speciﬁed. These prior works have made assumptions that the inﬂuences on use
of force behavior are either: (1) at the level of the individual; or (2) at the level of
the neighborhood, but have neglected to include interaction effects. This

assumption, or rather its repudiation, lies at the core of this research.

168

Discussion

The theoretical models of social control and social disorganization, as well
as the concept of tolerance, provide the most advantageous foundation from
which to view police use of force behavior. The latter all posit a direct relationship
between ecological (i.e. neighborhood) characteristics and some type of human,
including police ofﬁcer, behavior. A discussion of these theories, and the
foundation they provide for the current research, will allow for a greater
understanding of the results of this dissertation. In order to further understand the
contributions of these theories, the results for all theoretically-driven, as well as

all control, variables will be more closely examined here.

Re—visiting the Theoretical Foundation for the Current Research:
The theory of social control, as posited by Donald Black (1976; 1980;

1989), provides the core of the theoretical foundation for the current work. In
essence, social control theory has two tenets which are relevant to research on
police use of force behavior. The ﬁrst is the argument that formal social control,
conceptualized here as use of force behavior within the context of an arrest, will
be more prevalent in situations (i.e. arrests) where there is greater relational
distance between participants (i.e. suspects and police ofﬁcers). The second
tenet argues that formal social control will be more prevalent in environments (i.e.
neighborhoods) where there is a lack of informal social control. Thus, where

neighborhood residents are unable, or unwilling to organize, or are incapable of

169

coordinating social monitoring, police ofﬁcers are more likely to be required to
engage in formal social control.

The second tenet of social control theory discussed above is closely linked
to social disorganization theory, which argues that areas (i.e. neighborhoods) in
which the citizens lack the ability to organize themselves are characterized by
populations that are: (1) immigrant; (2) minority; (3) lower class; and (4)
accepting of unconventional norms (Park and Burgess 1924; Sampson and
Groves 1989; Shaw and McKay 1942; Stark 1987). This concept has been
modiﬁed more recently to focus on concentrated disadvantage, which argues that
neighborhoods in distress are more aptly characterized by a factor which is
comprised of: (1) the percentage of the population which is African-American; (2)
the percentage of the population which is below the poverty level; (3) the
percentage of the population which is on public assistance; (4) the percentage of
female-headed households; (5) the percentage of the population which is
unemployed; and (6) the density of individuals under the age of seventeen (Kane
2002; Morenoff et al. 2001; Sampson and Bartusch 1998; Sampson et al. 1997;
Reisig and Parks 2000; Terrill and Reisig 2003). Thus, in areas that are higher in
concentrated disadvantage, and are therefore disorganized and in distress,
police ofﬁcers are more likely to be required to engage in formal social control.

The ﬁnal theoretical concepts which have informed the current research
are that of ecological contamination, tolerance and face-saving behavior. In
essence, the concept of ecological contamination (Bittner; Black 1998; Van

Maanen 1973; Werthman and Piliavin 1967) argues that individuals, regardless

170

of their own characteristics, become tainted when present in the “wrong” area
(i.e. neighborhood). Thus, individuals who are present in a neighborhood high in
concentrated disadvantage assume the negative associations of that
neighborhood and become acceptable targets of police behavior (i.e. use of
force). This concept has been modiﬁed by Klinger (1997; see also Bittner 1967;
Black 1989; Kohfeld and Sprague 1990) who argues that this police response is
only triggered when criminal activity has surpassed a threshold of frequency or
severity. Thus, police ofﬁcers in “bad” neighborhoods may be willing to overlook
a certain amount of illegal activity due in large part to concerns regarding time
constraints and the necessity of responding to more serious incidents. However,
it has also been argued that it is precisely in these neighborhoods in which
ofﬁcers must engage in face-saving behavior as a preventative measure,
particularly when confronted with aggressive or antagonistic suspects (Black
1980; Chevigny 1995; Fyfe 1997; Hawley and Messner1989; Kavanagh 1997;
Perez 1994; Skolnick and Fyfe 1993; Toch 1995). Thus, there is a balance which
be must struck in such neighborhoods between tolerating some illegal activity,
particularly since a certain amount of such activity is expected in these
neighborhoods, and maintaining authority in order to prevent such illegal

activities from escalating in frequency or severity.

171

Theoretical Foundations of Current Hypotheses:

This dissertation has proposed ﬁve hypotheses derived from the theories
and concepts discussed above, which were tested using hierarchical linear
modeling techniques. The ﬁrst hypothesis (H1) proposed that ofﬁcers would use
more force in neighborhoods which are higher in concentrated disadvantage.
This hypothesis is derived from the theory of social control, which argues that
formal social control (i.e. use of force) is more likely to be used in areas which
exhibit a lack of informal social control (i.e. are high in concentrated
disadvantage). The second hypothesis (H2) proposed that minority ofﬁcers would
use less force than white ofﬁcers in neighborhoods which are higher in
concentrated disadvantage. This hypothesis is also derived from the theory of
social control, which argues that formal social control is less likely between
individuals who have a decreased social distance, such as minority ofﬁcers and
citizens in disadvantaged neighborhoods. The third hypothesis (H3) proposed
that ofﬁcers would use more force against suspects with poor demeanor who are
encountered in neighborhoods with higher crime rates. This hypothesis is derived
from the concepts of tolerance and face-saving behavior. That is, although a
certain amount of illegal activity is tolerated in neighborhoods with higher crime
rates due to the need to respond to more serious crimes, ofﬁcers must still
maintain authority and “save face” in these areas by not allowing suspects to
exhibit poor demeanor. The fourth hypothesis (H4) proposed that ofﬁcers would
use more force against suspects with poor demeanor who are encountered in

neighborhoods which are higher in concentrated disadvantage. This hypothesis

172

is also derived from the concepts of tolerance and face-saving behavior. As with
H3, ofﬁcers must maintain authority and “save face” in these areas by not
allowing suspects to exhibit poor demeanor. The ﬁfth hypothesis (H5) proposed
that ofﬁcers would use less force against minority suspects who are encountered
in neighborhoods which are higher in concentrated disadvantage than against
White suspects encountered in these neighborhoods. This hypothesis is derived
from the concepts of ecological contamination and tolerance. In essence,
minority suspects encountered in these types of neighborhoods are assigned the
negative attributes of the neighborhood, but their activity is tolerated due to the
fact that it is precisely the type of behavior which is expected of them in these

neighborhoods.

Impact of Variables Concerning Neighborhood-Level Effects:

Neighborhood-level variables were the focus of this dissertation,
particularly in interaction with individual-level variables. However, the variables
concerning neighborhood characteristics are introduced to determine if ecological
variables, in and of themselves, are inﬂuences upon use of force behavior. These
variables are thought to act consistently across both dependent variables,
maximum force and physical force plus threats.

Within the 6-site maximum force model, three of the seven “neighborhood”
variables were statistically signiﬁcant. Notably, there was no statistically
signiﬁcant effect for the inﬂuence of the violent crime rate or the measure of

concentrated disadvantage. In addition, an arrest encounter taking place in

173

Charlotte, North Carolina showed no signiﬁcant difference in average maximum
force from an arrest in Colorado Springs, Colorado. However, if the arrest
encounter took place in a neighborhood located in Dallas, Texas, maximum force
was predicted to increase in relation to an arrest that took place in a
neighborhood located in Colorado Springs, Colorado. Similarly, if the arrest
encounter took place in a neighborhood located in St. Petersburg, Florida,
maximum force was predicted to increase in relation to the reference city. Finally,
if the arrest encounter took place in San Diego, California, maximum force was
predicted to increase in relation to the reference neighborhood. These results
present a signiﬁcant contrast to prior research which has found that the violent
crime rate and concentrated disadvantage (Terrill and Reisig 2003) can inﬂuence
use of force behavior. More importantly, these results disconﬁnn one of the
current work’s hypotheses (H1) that more force is expected in neighborhoods
that are high in concentrated disadvantage.

Within the 6-site physical force plus threats model, none of the seven
“neighborhood” variables were signiﬁcant. Again, this is in direct contrast to
previous studies which have found that the violent crime rate and concentrated
disadvantage (Terrill and Reisig 2003) can inﬂuence the likelihood of force
behavior. As with the previous dependent variable, the most relevant point is that
these results disconﬁrrn one of the current work’s hypotheses (H1) that more
force is expected in neighborhoods that are high in concentrated disadvantage.

The predictive power of the variables regarding neighborhood

characteristics was unexpected, both in relation to the current work and in

174

relation to previous research. Of the seven “neighborhood” variables, none were
statistically signiﬁcant across both models. Of these seven variables, three
(“arrest took place in Dallas, TX”, “arrest took place in St. Petersburg, FL”, and
“arrest took place in San Diego, CA”) were statistically signiﬁcant in the maximum
force model, while none were statistically signiﬁcant in the physical force plus
threats model. These results represent a strong departure from prior research
(Smith 1986; Terrill and Reisig 2003) which has found statistically signiﬁcant
effects for neighborhood variables. However, as has been noted previously, it is
believed that the current work has modeled neighborhood variables more

appropriately in interactions with several individual-level variables.

Impact of Variables Concerning Interaction Effects:

The variables concerning neighborhood characteristics are also
introduced to determine if neighborhood-level variables, in interaction with
individual-level variables, are inﬂuences upon use of force behavior. These
variables are thought to act consistently across both dependent variables,
maximum force and physical force plus threats, although they are clearly thought
to vary across different types of neighborhood. More importantly, these
interactions are the main focus of the analyses of this dissertation.

Within the 6-site maximum model, none of the nine interaction variables
were signiﬁcant. Although this is consistent with prior research which has not
found any statistically signiﬁcant effects (T errill and Reisig 2003), it is in direct

contrast to the hypotheses of the current work. The second hypothesis of this

175

dissertation (H2) predicted that minority ofﬁcers would use less force than White
ofﬁcers in neighborhoods that are higher in concentrated disadvantage, due to
decreased social distance. The results predicted that Black ofﬁcers would use
more force than White ofﬁcers (.30) in these neighborhoods, as would ofﬁcers of
the “Other” racial category (.19), while Hispanic ofﬁcers were predicted to use
less force than White ofﬁcers (-.07) in neighborhoods high in concentrated
disadvantage. However, all of these represent relatively small effect sizes, and
none of these are statistically signiﬁcant. The third hypothesis of the current work
(H3) predicted that more force would be used against suspects with poor
demeanor arrested in neighborhoods with higher crime rates than against
suspects with poor demeanor arrested in neighborhoods with lower crime rates,
due to the ofﬁcer’s need to “save face” and maintain authority in these
environments. The results predicted that more force would be used against
suspects with poor demeanor who were arrested in high-crime neighborhoods
(.003) than against suspects with poor demeanor who were arrested in lower-
crime neighborhoods. However, this again represented a very small effect size,
and was not statistically signiﬁcant. The fourth hypothesis of this dissertation (H4)
predicted that more force would be used against suspects with poor demeanor
arrested in neighborhoods higher in concentrated disadvantage than against
suspects with poor demeanor arrested in neighborhoods lower in concentrated
disadvantage, due to the ofﬁcer’s need to “save face” and maintain authority in
these environments. The results predicted that more force would be used against

suspects with poor demeanor who were arrested in neighborhoods high in

176

concentrated disadvantage (.002) than against suspects with poor demeanor
who were arrested in neighborhoods lower in concentrated disadvantage.
However, this again represented a very small effect size, and was not statistically
signiﬁcant. The ﬁfth, and ﬁnal, hypothesis of the current work (H5) predicted that
less force would be used against minority suspects arrested in neighborhoods
higher in concentrated disadvantage than against White suspects arrested in
these neighborhoods, due to the concepts of tolerance and ecological
contamination. The results predicted that more force would be used against
Black (.12), Hispanic (.16), and “Missing race” (.46) suspects arrested in
neighborhoods high in concentrated disadvantage than against White suspects
arrested in these neighborhoods. In contrast, less force was predicted against
suspects of the “Other” racial category (-.06) arrested in neighborhoods high in
concentrated disadvantage than against White suspects in these neighborhoods.
However, these effect sizes were also very small, and none of them were
statistically signiﬁcant.

Within the 6-site physical force plus threats model, again none of the nine
interaction variables were signiﬁcant. Although this is consistent with prior
research which has not found any statistically signiﬁcant effects (Terrill and
Reisig 2003), it also is in direct contrast to the hypotheses of the current work.
The second hypothesis of this dissertation (H2) predicted that minority ofﬁcers
would use less force than White ofﬁcers in neighborhoods that are higher in
concentrated disadvantage, due to decreased social distance. The results

predicted that Black ofﬁcers were more likely to use force than White ofﬁcers

177

(1.15) in these neighborhoods, as were ofﬁcers of the “Other” racial category
(1.02), while Hispanic ofﬁcers were predicted to be less likely to use force than
White ofﬁcers (.95) in neighborhoods high in concentrated disadvantage.
However, all of these represent relatively small effect sizes, and none of these
are statistically signiﬁcant. The third hypothesis of the current work (H3)
predicted that more force would be used against suspects with poor demeanor
arrested in neighborhoods with higher crime rates than against suspects with
poor demeanor arrested in neighborhoods with lower crime rates, due to the
ofﬁcer’s need to “save face” and maintain authority in these environments. The
results showed that force was predicted to be less likely to be used against
suspects with poor demeanor who were arrested in high-crime neighborhoods
(.92) than against suspects with poor demeanor who were arrested in lower-
crime neighborhoods. Although this not in accordance with H3, this again
represented a very small effect size, and was not statistically signiﬁcant. The
fourth hypothesis of this dissertation (H4) predicted that more force would be
used against suspects with poor demeanor arrested in neighborhoods higher in
concentrated disadvantage than against suspects with poor demeanor arrested
in neighborhoods lower in concentrated disadvantage, due to the ofﬁcer’s need
to “save face” and maintain authority in these environments. The results showed
that force was predicted to be equally likely to be used against suspects with
poor demeanor who were arrested in neighborhoods high in concentrated
disadvantage (1.00) than against suspects with poor demeanor who were

arrested in neighborhoods lower in concentrated disadvantage. Again, while this

178

was not in accordance with H4, this represented a very small effect size, and was
not statistically signiﬁcant. The ﬁfth, and ﬁnal, hypothesis of the current work (H5)
predicted that less force would be used against minority suspects arrested in
neighborhoods higher in concentrated disadvantage than against White suspects
arrested in these neighborhoods, due to the concepts of tolerance and ecological
contamination. The results shewed that, in accordance with H5, force was
predicted to be less likely to be used against Black (.96), Hispanic (.88), and
suspects of the “Other” racial category (.58) arrested in neighborhoods high in
concentrated disadvantage than against White suspects arrested in these
neighborhoods. In contrast, force was predicted to be equally likely to be used
against suspects of the “Missing race” racial category (1 .00) arrested in
neighborhoods high in concentrated disadvantage than against White suspects in
these neighborhoods. However, these effect sizes were also very small, and
none of them were statistically signiﬁcant.

The predictive power of the interaction variables was clearly less than was
expected, given that the hypotheses predicted a statistically signiﬁcant effect for
these interactions and none was found. While previous research has been
inconsistent with regards to ﬁndings concerning neighborhood variables and
interaction effects (Smith 1986; Terrill and Reisig 2003), the current work is
believed to have properly modeled these interactions by allowing the effects of
individual-level variables to vary across Level-2 units. However, despite using the
appropriate methodology, none of the ﬁve proposed hypotheses were supported

in the current research. Thus, although this work has been conducted rigorously,

179

the independent variables in the form of interactions have added almost no
predictive power to the 6-site models for the dependent variables of maximum

force and physical force plus threats.

Impact of Individual-Level Variables:

Although the focus of this dissertation is on the interaction effects between
individual-level and neighborhood-level variables, none of these produced
signiﬁcant results. Therefore, it is also instructive to examine the impact of the
numerous individual-level variables which were included as statistical controls

due to their statistical signiﬁcance in prior research efforts.

Impact of Variables Concerning the Nature of the Location

Police use of force behavior has been thought to be inﬂuenced by
variables concerning the nature of the location of the arrest encounter due to the
fact that there are certain physical characteristics of an area which may inﬂuence
an ofﬁcer’s perception of his/her safety, and thus increase the use of force.
These variables are thought to act consistently across both dependent variables,
maximum force and physical force plus threats.

Within the 6-site maximum force model, only two of the four “location”
variables were signiﬁcant. When a location was known for criminal activity,
maximum force was predicted to increase in arrest encounters at that location.
However, as visibility increased at that location, maximum force was predicted to

decrease. These results are consistent with the concept of ofﬁcer concerns

180

regarding safety inﬂuencing use of force behavior. More importantly, they are
also consistent with ﬁndings from prior research on use of force behavior (Garner
et al. 2002).

Within the 6-site physical force plus threats model, only one of the four
“location” variables was signiﬁcant. In an unexpected turn, as visibility increased
at the arrest location, the likelihood of use of force behavior was predicted to
increase. This is in direct contrast to the concept of ofﬁcer concerns regarding
safety inﬂuencing use of force behavior. In addition, this is also at odds with
ﬁndings from previous research (Gamer et al. 2002).

The predictive power of the variables regarding the nature of the location
was as expected. Of the four “location” variables, only the visibility at the arrest
encounter location was statistically signiﬁcant across both models. In addition,
the location being known for criminal activity was also statistically signiﬁcant, but
only in the maximum force model. Although the variables “arrest took place
inside” and “location was known to be hazardous” were not statistically signiﬁcant
in either model, this ﬁnding is consistent with prior research as well (Garner et al.

2002).

Impact of Variables Concerning the Nature of the Encounter

As with the previous grouping of variables, the variables concerning the
nature of the arrest encounter itself are all related to an ofﬁcer’s perceptions of
safety during the arrest encounter. Once again, the perception of danger within a

speciﬁc encounter is thought to increase use of force behavior. As before, these

181

variables are thought to act consistently across both dependent variables,
maximum force and physical force plus threats.

Within the 6-site maximum force model, only one of the four “encounter”
variables was statistically signiﬁcant. If the arrest was for a violent offense,
maximum force in the arrest encounter was predicted to increase. This is
consistent with an ofﬁcer’s concerns regarding safety leading to increased force.
Prior research has repeatedly noted that offense seriousness, taken to show an
arrestee’s capacity for violence, exerts a signiﬁcant inﬂuence on use of force
behavior (Garner et al. 20002; Worden 1995).

Within the 6-site physical force plus threats model, there was again only
one of the four “encounter” variables which was statistically signiﬁcant. As with
the previous model, if an arrest was for a violent offense, the likelihood of force in
that speciﬁc arrest encounter was predicted to increase. Once again, this is
consistent with prior literature on the likelihood of use of force behavior (Garner
et al. 2002; Worden 1995).

The predictive power of the variables regarding the nature of the arrest
encounter was more limited than was expected. Of the four “encounter”
variables, only the seriousness of the offense was statistically signiﬁcant across
both models. Of the three other variables in this subgroup (“arrest took place on
a weekend”, “bystander demeanor toward the ofﬁcer”, and “number of
suspects”), none were statistically signiﬁcant in either model. These results are in

contrast to previous research (Engel et al. 2000; Friedrich 1980; Garner et al.

182

 

2002; Worden 1995), which has found that bystander demeanor, as well as the

number of suspects involved in the arrest, can result in increased use of force.

Impact of Variables Concerning Police Mobilization

The variables concerning an ofﬁcer’s mobilization to the arrest encounter
are the ﬁnal subgroup which is all related to an officer’s perception of safety
during an arrest encounter. In essence, these variables describe an ofﬁcer’s
entry into the arrest, in reason, method, and response. As with the previous two
subgroups, the perception of danger within a speciﬁc encounter is thought to
increase use of force behavior. Once again, these variables are thought to act
consistently across both dependent variables, maximum force and physical force
plus threats.

Within the 6-site maximum force model, ﬁve of the ten “mobilization”
variables were statistically signiﬁcant. If the ofﬁcer responded to the scene in a
“priority approach” manner (i.e. with higher rates of speed, occasionally
disregarding trafﬁc signs and signals), maximum force was predicted to increase.
Also with respect to an ofﬁcer’s approach, if the ofﬁcer used lights and sirens en
route to the encounter, maximum force was predicted to increase. In addition, if
an ofﬁcer requested the presence of back-up ofﬁcers at the arrest encounter,
maximum force was predicted to increase. An increase in maximum force was
also predicted with each subsequent ofﬁcer who arrived at the arrest encounter.
Finally, if an ofﬁcer was off-duty when the arrest was made, maximum force was

predicted to increase. These results are once again consistent with ﬁndings from

183

 

prior research (Garner et al. 1995; Garner et al. 2002), which has found
statistically signiﬁcant effects for ofﬁcer mobilization.

Within the physical force plus threats model, seven of the ten
“mobilization” variables were statistically signiﬁcant. If the encounter was initiated
by the ofﬁcer (rather than having been dispatched, or ﬂagged down by a citizen),
the likelihood of force was predicted to increase. In addition, if the ofﬁcer
responded to the scene in a “priority approach” manner (i.e. with higher rates of
speed, occasionally disregarding trafﬁc signs and signals), the likelihood of force
was predicted to increase. Also with respect to an ofﬁcer’s approach, if the ofﬁcer
used lights and sirens en route to the encounter, the likelihood of force was
predicted to increase. In addition, if the ofﬁcer took some other unknown (yet
non-routine) approach to the encounter, the likelihood of force was predicted to
increase. Once again, if an ofﬁcer requested the presence of back-up ofﬁcers at
the arrest encounter, the likelihood of force was predicted to increase. An
increase in the likelihood of force was also predicted with each subsequent
ofﬁcer who arrived at the arrest encounter. However, in an unexpected ﬁnding, if
a suspect was already in custody when the ofﬁcer arrived at the encounter, the
likelihood of force was also predicted to increase. With the exception of the latter,
these results are consistent with previous ﬁndings (Garner et al. 1995; Garner et
al. 2002) which have demonstrated a statistically signiﬁcant relationship between
ofﬁcer mobilization and increased likelihood of force.

The predictive power of the variables regarding the mobilization of police

ofﬁcers was as expected. Of the ten “mobilization” variables, four (“ofﬁcer took a

184

 

priority approach”, “ofﬁcer used lights and sirens on approach”, “ofﬁcer called for
back-up”, and “number of ofﬁcers”) were statistically signiﬁcant across both
models. Of the eight other variables in this subgroup, one (“ofﬁcer was off-duty”)
was statistically signiﬁcant in the maximum force model, while three (“suspect
was already in custody”, “police initiated the encounter", “ofﬁcer took other (non-
routine) approach to encounter”) were statistically signiﬁcant in the physical force
plus threats model. These results are in agreement with previous research
(Engel et al. 2000; Garner et al. 2002; Worden 1995), which has found that an

ofﬁcer’s entry into the arrest encounter can increase use of force behavior.

Impact of Variables Concerning Oﬁicer Characteristics

The variables concerning ofﬁcer characteristics are clearly meant to
determine if an ofﬁcer's ﬁxed characteristics and variable behavior are inﬂuences
upon use of force behavior. As with the previous subgroups, these variables are
thought to act consistently across both dependent variables, maximum force and
physical force plus threats.

Within the 6-site maximum force model, two of the eight “ofﬁcer" variables
were statistically signiﬁcant. In relation to an ofﬁcer’s ﬁxed characteristics, with
each one unit increase in the age of the ofﬁcer, maximum force was predicted to
decrease. In addition, maximum force was predicted to increase if the ofﬁcer was
male. These ﬁndings are consistent with prior research (Engel et al. 2000;

Garner et al. 2002; Terrill and Mastrofski 2002; Worden 1995) which has found

185

 

that these ofﬁcer characteristics can have signiﬁcant effects on use of force
behavior.

Within the 6-site physical force plus threats model, three of the eight
“ofﬁcer” variables were statistically signiﬁcant. With respect to an ofﬁcer’s ﬁxed
characteristics, if an ofﬁcer was Hispanic, the likelihood of force was predicted to
increase in relation to White ofﬁcers. In addition, the likelihood of force was E
predicted to increase if the ofﬁcer was male. In relation to an ofﬁcer’s behavior, if ‘
an ofﬁcer had received prior medical attention for an injury sustained while at

work, the likelihood of force was predicted to increase. Once again, these

 

ﬁndings are generally consistent with previous ﬁndings (Engel et al. 2000; Garner
et al. 2002; Terrill and Mastrofski 2002; Worden 1995) which have demonstrated
a statistically signiﬁcant effect of ofﬁcer characteristics on use of force behavior.
The predictive power of the variables regarding ofﬁcer characteristics was
generally as expected. Of the eight “ofﬁcer" variables, only one (“ofﬁcer was
male”) was statistically signiﬁcant across both models. Of the eight other
variables in this subgroup, one (“ofﬁcer age”) was statistically signiﬁcant in the
maximum force model, while two (“ofﬁcer was Hispanic” and “ofﬁcer received
prior medical attention”) were statistically signiﬁcant in the physical force plus
threats model. These results are generally in agreement with previous research
(Engel et al. 2000; Garner et al. 2002; Terrill and Mastrofski 2002; Worden 1995),
which has found that ofﬁcer characteristics can inﬂuence use of force behavior.
However, it should be noted that it is unusual that ofﬁcer age was not found to be

signiﬁcant in the physical force plus threats model. In addition, although a race

186

effect was found for Hispanic ofﬁcers in the physical force plus threats model, it is
believed that in the current research race effects are more properly modeled in

an interaction with neighborhood-level variables.

Impact of Variables Conceming Suspect Characteristics

The variables concerning suspect characteristics are clearly meant to
determine if a suspect’s ﬁxed characteristics, variable behavior, and relationships
are inﬂuences upon use of force behavior. As with the previous subgroups, these
variables are thought to act consistently across both dependent variables,
maximum force and physical force plus threats.

Within the 6-site maximum force model, seven of the nineteen variables
were statistically signiﬁcant. In relation to a suspect’s ﬁxed characteristics, if the
suspect was male maximum force was predicted to increase. There were
numerous statistically signiﬁcant effects related to a suspect’s relationships. If the
suspect was friends with the victim, maximum force was predicted to decrease.
Similarly, if the suspect was a member of the victim’s family, maximum force was
predicted to decrease. Finally, if the suspect was a stranger to all of the
bystanders, maximum force was predicted to increase. There were also several
statistically signiﬁcant effects related to a suspect’s behavior. If the suspect was
known to carry a weapon, maximum force was predicted to increase. If the
suspect was antagonistic (i.e. had poor demeanor) toward the ofﬁcer, maximum
force was predicted to increase. Finally, if the suspect provided physical

resistance to the ofﬁcer, maximum force was predicted to increase. These results

187

 

are consistent with prior research (Bayley and Garafalo 1989; Friedrich 1980;
Garner et al. 2002; Kavanagh 1997) which has identiﬁed suspect characteristics
as inﬂuencing use of force behavior.

Within the 6-site physical force plus threats model, seven of the nineteen
variables were statistically signiﬁcant. With respect to a suspect’s ﬁxed
characteristics, if the suspect was male the likelihood of force was predicted to
increase. There were also two statistically signiﬁcant effects related to a
suspect’s relationships. It the suspect had an unknown relationship to the
bystanders, there was a predicted increase in the likelihood of force. In addition,
if the suspect was a stranger to all of the bystanders, the likelihood of force was
also predicted to increase. There were also several statistically signiﬁcant effects
related to a suspect’s behavior. If the suspect was known to carry a weapon, the
likelihood of force was predicted to increase. In addition, if the suspect was
intoxicated during the arrest encounter, the likelihood of force was predicted to
increase. More importantly, however, if the suspect was antagonistic (i.e.
exhibited poor demeanor) or physically resisted the ofﬁcer, the likelihood of force
was predicted to increase. As with the previous model, these results are
consistent with previous studies (Bayley and Garafalo 1989; Friedrich 1980;
Garner et al. 2002; Kavanagh 1997), which have identiﬁed these suspect
characteristics as increasing the likelihood of force.

The predictive power of the variables regarding suspect characteristics
was generally as expected. Of the nineteen “suspect” variables, ﬁve (“suspect

was male”, “suspect was known to carry a weapon , suspect was a stranger to

188

 

all bystanders”, “suspect was antagonistic”, and “suspect provided physical
resistance”) were statistically signiﬁcant across both models. Of the remaining
fourteen variables in this subgroup, two (“suspect was friends with the victim” and
“suspect was family to the victim”) were statistically signiﬁcant in the maximum
force model, while two (“suspect was intoxicated” and “suspect had an unknown
relationship to all bystanders”) were statistically signiﬁcant in the physical force
plus threats model. These results are generally in agreement with previous
research (Bayley and Garafalo 1989; Friedrich 1980; Garner et al. 2002;
Kavanagh 1997), which has found that suspect characteristics can inﬂuence use
of force behavior. Notably, suspect race was not found to be signiﬁcant, in
contrast to some prior studies (Kavanagh 1997; Smith 1986; Terrill and
Mastrofski 2002; Worden 1995). However, it is believed that the current work
more properly models suspect race in interaction with neighborhood-level

variables, and thus no effects were expected here.

Interpretation of Results:

Given that none of the ﬁve hypotheses of this dissertation were supported
by the analyses of both dependent variables within hierarchical linear models, it
is instructive to present a discussion of why these hypotheses were not
conﬁrmed. Table 10 presents a summary of the results obtained for each of the

ﬁve hypotheses.“

 

8‘ Appendix E contains a further description of how these results compare to the work found in Garner et al.
(2002), where the PUF data were analyzed only at the individual level.

189

Table 10: Summary of Results

Hypothesis

H1: Ofﬁcers will use more force
in neighborhoods that are
higher in concentrated
disadvantage than in
neighborhoods that are lower in
concentrated disadvantage

H2: Minority ofﬁcers will use
less force than White ofﬁcers in
neighborhoods that are higher
in concentrated disadvantage
African-American ofﬁcers
Hispanic ofﬁcers
"Other" ofﬁcers

H3: Ofﬁcers will use more force
against suspects with poor
demeanor who are arrested in
neighborhoods that have higher
crime rates than against
suspects who are arrested in
neighborhoods that have lower
crime rates

H4: Ofﬁcers will use more force
against suspects with poor
demeanor who are arrested in
neighborhoods that are higher
in concentrated disadvantage
than against suspects with poor
demeanor who are arrested in
neighborhoods that are lower in
concentrated disadvantage

H5: Ofﬁcers will use less force
against minority suspects who
are arrested in neighborhoods
that are higher in concentrated
disadvantage than against
White suspects who are
arrested in neighborhoods that
are higher in concentrated
disadvantage
African-American suspects
Hispanic suspects
"Other" suspects
"Missing" suspects

Max.

Force

-0.050

0.300
-0.070
0.190

0.003

0.002

0.120

0.160
-0.060

0.460

No

No
No
No

No

No

No

No
No

No

190

Direction

No

No
Yes
No

Yes

Yes

No

No
Yes
No

p
<.05 Hypothesis Physlcal
Force
Plus

Threats

1.100

1.150
0.950
1.020

0.920

1.000

0.960

0.880
0.580

1.000

No

No
No
No

No

NO

No
No
No
No

P
<.05 Hypothesis
Direction

Yes r

 

No
Yes
No

No

No

Yes
Yes
Yes

No

It can be seen from Table 10 that of the hypotheses examined within the
maximum force model, none were statistically signiﬁcant and only four
coefﬁcients were in the hypothesized direction. Within the physical force plus
threats model, none of the hypotheses achieved statistical signiﬁcance and only
ﬁve log-odds were in the hypothesized direction. There are several possibilities
for the failure of the current research to conﬁrm the hypotheses. In particular, the
current research, although believed to be methodologically and theoretically
sound, experienced some limitations which must be considered when interpreting
the results. There are data issues to be discussed (use of an arrest-only sample,
use of ofﬁcer self-reports, use of a community violent crime rate constructed by
aggregating arrests sampled during data collection, use of volunteer sites, and
small sample size for sites), as well as issues of variable choice (use of only two
of a wide variety of potential use of force measures, and lack of certain
variables), which may have impacted the results.

The ﬁrst challenge to this dissertation concerns the sampling of use of
force incidents. As has been previously discussed, the Police Use of Force (PUF)
Study sampled only arrest encounters, and thus does not capture any use of
force within an encounter which did not result in an arrest. As base rates of force
vary rather widely among previous research studies which have sampled arrests
and all police—public interactions (Engel et al. 2000; Friedrich 1980; Garner et al.
2002; Terrill and Mastrofski 2002; Worden 1995), it is difﬁcult to ascertain the
extent of use of force behavior which is not captured within the current study.

However, it remains a real possibility that there are use of force incidents which

191

occurred within the PUF sites which did not result in an arrest, and thus the
current data must be seen as incomplete as an explanation of police use of force
in all situations. However, it is important to note that as all arrests within the PUF
sites were captured, the current data does accurately describe police use of force
behavior in arrest situations.

The second concern to be discussed is the fact that the current data was Pf
taken from ofﬁcer self-reports. Although data collection occurred under the

guarantee of conﬁdentiality as provided by Federal regulations, there remains a

 

presumption of bias in situations where individuals are asked to report their own
behavior. While clearly ofﬁcers did report on their use of force actions against
arrestees, it is unclear if these reports are entirely accurate, as certain behaviors
may have been downplayed. In addition, ofﬁcers may have failed to disclose
certain activities for fear of facing departmental discipline. It should be noted,
however, that while there are numerous potential data sources (e.g. arrestees”
reports of force used against them, observational research, or police records), no
prior research studies have used these in combination, and each has distinct
advantages and disadvantages. More importantly, suspects in these arrest
encounters were randomly sampled and surveyed regarding their own behavior,
as well as that of the arresting ofﬁcer. The latter’s use of force actions were then
compared between suspect surveys and ofﬁcer surveys to assess the validity of
the ofﬁcer self-report. At the aggregate level, both suspects and ofﬁcers reported

a 20% base rate for use of force behavior, indicating that ofﬁcer self-reports were

192

an accurate measure of such behavior within the six jurisdictions studied in this
research.85

The third concern dealing with data issues is the fact that the
neighborhood violent crime rate was constructed using arrest charge codes for
arrests made during the data collection period. These arrest charges were coded
as either violent crimes or non-violent crimes, and were then aggregated up to L
the census tract level (the measure of neighborhood for the current study). In this

situation there is a potential concern with multicollinearity. However, this concern

 

is addressed in two ways. First, it should be noted that all arrests within each '
jurisdiction were sampled during the data collection period, and thus these
reports represent ofﬁcial crime statistics for each neighborhood. Second, and
most importantly, the variance inﬂation factor (VIP) and tolerance scores indicate
that multicollinearity was not present.

The fourth concern regarding data issues is related to the fact that the
PUF sites were enlisted in the study on a volunteer basis. Thus, there may be
characteristics of these departments themselves, or of the command staff within
the departments, which inﬂuence the use of force behavior of their ofﬁcers. One
could argue that these (unmeasured) characteristics make these sites distinct
and therefore the results are not generalizable. However, it is the characteristics
of the sites themselves, rather than of the law enforcement agencies within those

sites, which have been posited to inﬂuence use of force behavior. lmportantly,

 

'5 This information was obtained ﬁ'om a personal communication with Dr. Christopher D. Maxwell, one of
the principal investigators on the Police Use of Force (PUF) study.

193

the PUF sites are sufﬁciently varied across numerous characteristics to instill
conﬁdence in the generalizability of the results.

The ﬁfth, and ﬁnal, concern stemming from data issues is the fact that a
relatively small number of sites were involved in data collection for PUF. As
noted previously, the most statistically sound model would be a three-level

hierarchical linear model with PUF sites at the third level. However, the results

 

from within the PUF sites are substantially similar to the 6-site model, and thus a

two-level model with dummy-coded variables for sites is acceptable under the

 

current conditions.

III.'.-. . -.I'u"”i -
‘3 .
1'
Ir

The ﬁrst issue dealing with the choice of variables is related to the fact that
there are numerous operationalizations of use of force behavior, and thus the
current research has selected only two of those various options for study. As has
been previously discussed, prior research has used a vast array of measures of
use of force (Bayley and Garafalo 1989; Engel et al. 2000; Friedrich 1980;
Garner et al. 2002; Terrill and Reisig 2003; Worden 1995). However, the
measures used here have signiﬁcant strengths. In particular, it is important to
note that the maximum force measure is based on ofﬁcer rankings of actual
ofﬁcer behavior in a variety of situations. In addition, the prevalence of force
measure beneﬁts from simplicity in that distinctions are not made between
reasonable or unreasonable force, only in whether or not force has occurred.

The second, and ﬁnal, concern regarding the choice of variables has
already been brieﬂy discussed. In any scientiﬁc endeavor the choice of

dependent and independent variables is an important one, and out of necessity

194

 

 

most research studies do not include an exhaustive range of variables. While
there may be individual-level or neighborhood-level, including organizational,
variables which have been omitted from the current work, the variables which
have been included here represent a systematic effort to reproduce the effects of
prior research.

Having considered potential limitations of the current research, it is most
important to note that it remains a possibility that the hypotheses were not
conﬁrmed for the simple reason that neighborhood-level variables do not, in fact,
exert any inﬂuence on police use of force behavior. Prior research (Bayley and
Garafalo 1989; Engel et al. 2000; Friedrich 1980; Garner et al. 2002; Kavanagh
1997; Phillips and Smith 2000; Smith 1986; Terrill and Mastrofski 2002; Terrill
and Reisig 2003; Worden 1995) has either failed to examine neighborhood
effects or has suffered from inappropriate methodology. Thus, while Terrill and
Reisig (2003) ﬁnd statistically signiﬁcant effects for the homicide rate and
concentrated disadvantage, two neighborhood-level variables, it is possible that
those ﬁndings were confounded by the use of a ﬁxed-effects model. In essence,
then, it is possible that the current research, in modeling neighborhood-level
variables in a random-effects model, has correctly determined that these
variables do not exert any statistically signiﬁcant effect on police use of force

behavior.

195

Implications of the Current Research

As has been discussed throughout this dissertation, the study of police
use of force behavior is one of the most fundamental research issues in social
scientiﬁc inquiry. The current research, in analyzing this behavior through use of
multilevel modeling examining individual-level and neighborhood-level variables,
strives to make a substantial contribution to the literature by identifying how these
variables inﬂuence use of force. The results of this work have several
implications, of a theoretical, research-driven and practical nature, and these will

be discussed here.

Theoretical Implications:

The current research was built on the theoretical underpinnings of Black’s
social control theory and social disorganization theory, as well as the concepts of
tolerance and ecological contamination. The results presented previously clearly
have implications for these theoretical concepts, particularly in light of the fact
that none of the ﬁve hypotheses were supported. With respect to Black's theory
of social control, a suspect’s social status and relational distance from an ofﬁcer
are thought to affect use of force behavior. However, the current research found
no effects for either social status (i.e. suspect race in isolation) or relational
distance (i.e. ofﬁcer race in interaction with the measure of concentrated
disadvantage). While these results are consistent with prior research, it is at odds
with the proposed hypotheses of the current work. Yet, at this time it is not

prudent to declare that Black’s theory of social control is unable to explain police

196

 

use of force behavior for two reasons: (1) suspect status is perhaps better
measured by the suspect’s actions, rather than ﬁxed characteristics (that is, a
suspect may place themselves in a particular status position through their
behavior, such as physically resisting the ofﬁcer or becoming antagonistic); and
(2) relational distance is perhaps better measured through an interaction
between ofﬁcer race and suspect race. It should be noted, however, that the
latter concept was not tested due to the fact that the analyses of this dissertation
were aimed at understanding the effects of individual-level variables in interaction
with neighborhood-level variables, such as ofﬁcer race and concentrated
disadvantage, rather than those of individual-level variables in interaction with
other individual-level variables, such as ofﬁcer race and suspect race.

In regards to social disorganization theory, the current work tested the
effects of concentrated disadvantage on police use of force behavior. Again, this
dissertation found no statistically signiﬁcant effects for this concept, either alone
or in interaction with individual-level variables, although the effects were
generally in the hypothesized direction. These results are not consistent with
previous multivariate studies which have studied neighborhood effects on police
use of force (Smith, 1986; Terrill and Reisig, 2003), both of which found a
statistically signiﬁcant effect for neighborhood characteristics. More importantly,
these results are at odds with the hypotheses proposed within this dissertation. It
is possible that social disorganization theory is unable to explain police use of
force behavior for two reasons. First, social disorganization theory was

developed and primarily used to examine the behavior of criminal offenders

197

(mostly delinquents). Thus, the theory may be more appropriate for explaining
deviant behavior, such as unreasonable force by police ofﬁcers, rather than all
use of force behavior. Second, the concept of concentrated disadvantage may be
more appropriately constructed using the percentage of non-white residents of
the neighborhood, rather than the percentage of African-American residents of
the neighborhood. Although none of the hypotheses tested here were statistically r

signiﬁcant, numerous propositions regarding Hispanic ofﬁcers and Hispanic

 

suspects were in the hypothesized direction. Considering that the population of 2?
Hispanics is quite different in San Diego than it is in Colorado Springs, the use of
a concentrated disadvantage variable constructed using the percentage of non-
white residents may have led to statistical signiﬁcance.

The current work also used the concepts of tolerance and ecological
contamination to examine police use of force behavior. These theoretical
concepts are connected in that both purport to describe levels of acceptable
behavior by individuals in particular neighborhoods. Once again, this dissertation
found no statistically signiﬁcant effects for these concepts, a result that is
inconsistent with prior work on ofﬁcer behavior. The most likely explanation for
the lack of signiﬁcance is that previous research (Klinger, 1997; Kohfeld and
Sprague, 1990) has focused on anest behavior, rather than use of force
behavior. Considering that the current work samples only arrests, it is possible
that any variation in use of force behavior explainable by these concepts has

been eliminated due to the lack of non-arrest situations.

198

mn Implications:

The limitations of the current research, discussed above, as well as the
results of that research, lead to several implications for future research on the
use of force behavior of police ofﬁcers. This dissertation has demonstrated the
importance of measuring suspect and ofﬁcer behavior appropriately, as well as
that of modeling that behavior using the appropriate statistical methods. Future
research efforts would benefit from: (1) utilizing several measures of force; (2)
utilizing several sources of data; (3) developing theoretical models which explain
all police behavior, including use of force; and (4) focusing on suspect behavior.

Numerous research efforts have utilized a variety of measures of use of
force, including maximum force, the prevalence of force, and the continuum of
force. However, there has been a notable lack of research which has used more
than one measure of force. While the current work uses two of these measures,
there remains a great deal of variation to be explored. Similarly, previous studies
have typically only used one source of data when examining use of force
behavior. Although ofﬁcer self-reports, arrestee interviews, ofﬁcial police reports,
and observational data are all viable avenues of research, prior research has
restricted itself to only one of these within speciﬁc studies. Future research
efforts would greatly beneﬁt from matching observational data to ofﬁcer self-
reports, or from matching the latter to arrestee interviews.

Perhaps even more important than the data issues discussed here are
concerns regarding the theoretical bases of research on use of force behavior.

Throughout the previous literature, researchers have focused on either use of

199

force behavior or on police behavior more generally. Future researchers would
beneﬁt from developing a theoretical model which is capable of explaining all
police behavior. Also in relation to the theoretical understanding of use of force
behavior, the results of this dissertation have demonstrated the importance of
variables concerning suspect behavior. In particular, the variables for suspect
demeanor (i.e. antagonism) and suspect physical resistance had statistically
signiﬁcant and relatively large effects across both dependent variables. This
indicates that future research would benefit from focusing more speciﬁcally on

the role of these variables in relation to police use of force behavior.

Practical Implications:

The results of this dissertation also have several practical implications for
law enforcement agencies with respect to training and policy on use of force
behavior by their ofﬁcers. In particular, this work has demonstrated that speciﬁc
ofﬁcer behaviors can lead to increased average maximum force, as well as an
increased likelihood in the application of force. Thus, law enforcement agencies
would beneﬁt from focusing training efforts and policy revisions on these factors
in order to more closely regulate use of force behavior.

The current work found that ofﬁcers responding to a call with a “priority
approach” (i.e. to a call dispatched as an emergency) or with lights and sirens
engaged had higher levels of maximum force, and were more likely to use force
during arrest encounters. In addition, arrest encounters where the ofﬁcer called

for back-up, or which had higher numbers of ofﬁcers present, also experienced

200

higher levels of maximum force, and were more likely to have force used within
them. As discussed previously, these variables are related to an ofﬁcer’s sense
of safety when entering an arrest encounter. Thus, these results indicate that law
enforcement agencies can, and should, focus their training efforts and policies on
identifying situations which have the potential for an elevated sense of danger.
Evidence from the current work, as well as from prior studies, provide a ﬁrm
foundation from which agencies can assess the behavior of ofﬁcers across a
variety of arrest situations, shifts, and types of police jurisdiction.

In conclusion, then, it should be noted that although the results of this
dissertation did not conﬁrm the proposed hypotheses, they do provide some
interesting ﬁndings which support previous studies in relation to suspect behavior
and ofﬁcer safety. Future research efforts should continue a focus on these
issues, with a clear understanding of the importance of using a variety of

measures of use of force behavior, as well as numerous data sources.

201

Appendix A

202

Table A-1: Signiﬁcant Predictors of Use of Force Behavior Across

Multivariate Studies

Base Rate of Force
Minority Suspect

Male Suspect
Increasing II of
Bystanders

Increasing # of Ofﬁcers
Drunk l Drugged
Suspect

Violent I Felony Offense
Antagonistic I Hostile
Suspect

Suspect Resists I Uses
Force

Base Rate of Force
Minority Suspect

Male Suspect
Increasing # of
Bystanders

Increasing # of Ofﬁcers
Drunkl Drugged
Suspect

Violent] Felony Offense
Antagonistic I Hostile
Suspect

Suspect Resists I Uses
Force

Friedrich
1 980

5.10%

Kavanagh

1 997

17.20%
+

Bayley

and

Garafalo

1 989

6-9%

Engel
etaL
(2000)

3.40%

203

Garner

et al.
1995

22%

+

Worden
1995

3.90%
4.

+

+

Phillips Terrill and Garner

and
Smith
2000
NR

'1-

Mastrofski
2002

58.40%
4.

et al.
2002

17.10%

+

+
+

Appendix B

204

 

Table B-1: Sample Characteristics for Variables used in Factor Analysis

 

 

 

 

 

 

 

 

Variable Variable Label N M SD
Name

BLACKPOP % African-American within Census Tract 385 24.62 29.48

TRACTPOV % Below Poverty Level within Census 385 18.86 13.49
Tract

TRCTPASS % on Public Assistance within Census 385 4.82 4.62
Tract

TRCTFEMH % Female-Headed Families within Census 385 9.69 7.30
Tract

TRCTUNEM % Unemployed within Census Tract 385 5.04 6.50

TRACTDEN Density of Individuals under 17 Years Old 385 0.64 0.88

 

within Census Tract

 

 

 

 

205

 

 

 

Table B-2: Correlation Matrix for Variables used in Factor Analysis

 

 

 

 

 

 

 

 

Variable BLACK TRACT TRCTPASS TRCTFEMH TRCTUNEM TRACT
Names POP POV DEN
BLACKPOP 1 .000 .438 .393 .647 .246 -.1 1 5
TRACTPOV .438 1 .000 .685 .575 .606 .340
TRCTPASS .393 .685 1 .000 .739 .201 .451
TRCTFEMH .647 .575 .739 1 .000 .180 .256
TRCTUNEM .246 .606 .201 .180 1.000 .022

TRACTDEN -.1 1 5 .340 .451 .256 .022 1 .000

 

 

 

 

 

 

 

206

 

 

 

Table B-3: Factor Loadingsfor Variables used in Factor Analysis

 

 

 

 

 

 

 

 

 

 

Variable Variable Label Factor
Name Loadim_
BLACKPOP % African-American within Census Tract .653
TRACTPOV % Below Poverty Level within Census Tract .874
TRCTPASS % on Public Assistance within Census Tract .857
TRCTFEMH % Female-Headed Families within Census Tract .850
TRCTUNEM % Unemployed within Census Tract .497
TRACTDEN Density of Individuals under 17 Years Old within .406
Census Tract

 

KMO Measure of Sampling Adequacy = 0.66

Bartlett’s Test of Sphericity
x2 = 1177.88 (.000)

207

 

Appendix C

208

 

TL“... A. ; .""‘.’ ) _ .(UKI

Table C-1: 6-Site Correlation Matrix

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Wsibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRl02
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

209

LOCRIME LOCHAZRD

1

0.45
0

-0.09
8.40E-15

-0.09
1595-14

-0.09
1.53E-13

-0.03
0.01

0.07
2.42E-09

0.11
4.63E-19

-0.03
0.03

-0.01
0.46

0.16
3.60E-41

0.03
0.01

-0.05
5.52E-06

-0.01
1

0.01
0.47

0.01
0.67

-0.04
0.01

-0. 12
1 .48E-22

—0.03
0.04

-0.02
0.08

0.13
2.77E-28

0.1
7.76E-16

-0.06
9.18E—08

-0.01
0.7

0.12
1.17E-22

0.01
0.85

0.01
0.52

0.02
0.06

0.04
0.01

0.01
0.72

 

 

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTI-IAP2
Dispatch or Onview Not Reported

PRI02
Priority Call

SIRENZ
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

210

INSIDE

0.06
4.38E—06

0.14
7.59E-30

-0.06
0.63

0.01
0.7

0.01
0.53

0.17
1.20E-43

0.02
0.09

-0.19
5.56E-56

-0.15
5.87E-36

0.1
2.68E—1 5

-0.19
1.71 E-54

-0.01
0.33

0.01
0.58

VISIBLE

-0.03
0.01

0.01
0.84

-0.05
0.01

-0.04
0.01

0.08
3.48E-12

0.02
0.09

-0.03
0.01

-0.01
0.29

-0.07
1.82E-09

-0.03
0.02

0.04
0.01

0.01
0.3

Table C-1 (eont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

211

VIOLENT WEEKEND

0.07
3.18E-09

0.02
0.06

-0.03
0.01

-0.05
0.01

0.03
0.03

-0.15
8.96E-38

-0.08
3.22E-12

0.14
7.89E-33

-0.05
0.01

0.05
0.01

0.01
0.43

-0.01
0.85

-0.01
0.8

-0.04
0.01

-0.02
0.06

-0.07
1.62E-08

0.01
0.6

0.01
0.73

0.04
0.01

-0.01
0.25

0.01
0.38

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

212

BPDEMEAN NUMBERSO

0.04
0.01

-0.05
0.01

0.02
0.14

-0.01
0.52

0.02
0.04

0.03
0.01

0.03
0.02

0.01
0.23

0.01
0.85

-0.03
0.01

0.02
0.15

0.06
2.14E-06

0.01
0.51

0.03
0.03

0.02
0.15

0.04
0.01

0.01
0.9

 

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

213

CUSTODY

-0.01
0.45

-0.16
6.80E-40

-0.05
0.01

—0.01
0.49

-0.08
1.62E-10

0.02
0.11

0.03
0.02

CIT2

-0.16
7.84E-43

-0.09
3.45E-13

-0.01
0.27

-0.02
0.08

0.03
0.01

0.05
0.01

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

214

POL2

-0.31
1.99E-148

-0.19
1.13E-59

0.16
6.12E-41

0.02
0.15

0.01
0.26

OTHAP2

-0.11
1.03E-20

0.04
0.01

0.05
8.18E-06

0.01
0.64

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT 2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

215

PRIO2

0.04
0.01

-0.16
1.32E-41

-0.01
0.34

SIREN2

—0.12
1065-22

-0.01
0.68

 

Table C-1 (cont’d)

LOCRIME
Location Known for Criminal Activity

LOCHAZRD
Location Known to be Hazardous

INSIDE
Arrest Occurred Inside

VISIBLE
Visibility at Place of Arrest

VIOLENT
Violent Offense

WEEKEND
Friday 6pm to Monday 6am

BPDEMEAN
Bystander Demeanor Toward Police

NUMBERSO
Number of Suspects at Completion of
Arrest

CUSTODY
Custody Status

CIT2
Citizen Initiated Contact with Police

POL2
Police Initiated Contact with Arrestee

OTHAP2
Dispatch or Onview Not Reported

PRIOZ
Priority Call

SIREN2
Used Lights and Siresn

NOTROUT2
Approach Unknown

OFFDUTY
Ofﬁcer Duty Status

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

216

NOTROUT2 OFFDUTY

0.03 1
0.02

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1 AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

217

LOCRIME

0.05
0.01

0.1
9.61 E-17

-0.06
2.35E-07

-0.01
0.74

-0.01
1

0.03
0.03

0.01
0.43

0.02
0.13

0.02
0.06

0.01
0.21

-0.03
0.77

0.16
6.10E-41

0.01
0.81

-0.03
0.01

-0.02
0.15

-0.02
0.07

0.09
1.20E-13

LOCHAZRD

0.07
2.38E-08

0.13
5.06E-28

-0.06
1.56E-07

-0.01
0.81

0.01
0.75

0.02
0.08

0.02
0.13

0.04
0.01

0.04
0.01

0.01
0.26

0.01
0.68

0.15
1.20E-36

-0.02
0.04

—0.03
0.01

-0.02
0.12

0.01
0.87

0.13
2.55E-25

.9 0.5-Nan. w
3
a .

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

218

INSIDE

—0.02
0.11

0.01
0.42

0.04
0.01

0.03
0.01

-0.01
0.22

-0.01
0.29

-0.03
0.01

-0.01
0.34

-0.02
0.11

-0.04
0.01

0.01
0.68

0.01
0.6

-0.03
0.02

0.01
0.44

0.01
0.65

-0.08
2.31E-12

0.03
0.01

VISIBLE

-0.05
0.01

-0.08
1.22E-11

0.11
6.24E-21

0.03
0.01

-0.03
0.01

-0.01
0.7

0.01
0.96

-0.01
0.36

-0.03
0.02

-0.03
0.03

0.03
0.04

0.01
0.67

-0.04
0.01

0.01
0.45

-0.02
0.17

-0.03
0.02

0.01
0.86

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

219

VIOLENT

0.08
1.53E-11

0.02
0.16

-0.04
0.01

0.03
0.03

0.01
0.56

-0.02
0.11

-0.02
0.13

0.02
0.16

-0.01
0.83

-0.06
1.88E-07

-0.01
0.47

0.06
7.42E-08

0.01
0.23

-0.01
0.73

-0.08
3.12E-10

0.05
0.01

0.09
1.16E-12

WEEKEND

0.01
0.61

-0.01
0.35

-0.04
0.01

-0.01
0.4

0.01
0.29

0.02
0.08

0.02
0.09

0.01
0.72

0.03
0.02

0.02
0.08

0.03
0.01

0.01
0.96

0.05
0.01

~0.03
0.01

-0.11
7.69E-19

0.02
0.09

-0.03
0.06

 

 

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

220

BPDEMEAN

0.05
0.01

0.12
1.02E-22

-0.04
0.01

0.01
0.78

0.01
0.98

0.01
0.92

0.03
0.01

0.18
1.70E-50

0.02
0.09

-0.01
0.54

-0.05
0.01

0.08
8.51

-0.02
0.12

-0.01
0.92

-0.03
0.02

-0.01
0.98

0.04
0.01

NUMBERSO

0.07
2.90E-09

0.3
6.25E-142

-0.03
0.02

-0.03
0.01

-0.01
0.86

0.01
0.45

0.01
0.72

0.01
0.69

0.01
0.77

-0.01
0.42

-0.05
0.01

0.03
0.01

0.01
0.52

0.01
0.52

0.01
0.49

0.01
0.7

-0.01
0.93

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

221

CUSTODY

-0.08
2.39E-12

-0.05
0.01

0.01
0.76

0.06
4.695—06

0.01023682
0.4

-0.02
0.06

-0.01
0.41

-0.01
0.45

-0.01
0.67

-0.08
8.21 E12

-0.04
0.01

0.02
0.07

-0.01
0.66

-0.02
0.21

0.02
0.04

-0.06
1.21E-06

-0.04
0.01

CIT

0.04
0.01

0.02
0.06

0.05
0.01

0.01
0.83

-0.02
0.2

0.02
0.13

0.02
0.05

0.01
0.25

0.05
8.43E-06

-0.03
0.03

-0.01
0.89

-0.01
0.69

0.01
0.94

0.01
0.58

0.01
0.32

0.01
0.41

0.02
0.18

 

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

222

POL2

0.05
0.01

0.05
0.01

-0.02
0.09

-0.03
0.02

—0.03
0.01

0.01
0.29

0.01
0.45

-0.02
0.1

-0.01
0.56

0.06
3.06E-06

-0.03
0.02

0.05
0.01

-0.05
0.01

-0.01
0.34

-0.03
0.01

-0.02
0.07

0.03
0.01

OTHAP2

-0.08
2.37E-10

-0.01
0.94

0.03
0.03

0.01
0.52

0.05
0.01

0.02
0.09

0.03
0.01

-0.01
0.95

-0.02
0.05

0.02
0.11

0.02
0.13

-0.04
0.01
0.06

5.09E-06
0.04

0.01
0.04

0.01
0.02

0.1

-0.04
0.01

 

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1 AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

223

PRI02

0.11
1.72E-18

0.1
4.21 E-17

0.01
0.24

0.02
0.12

-0.01
0.43

-0.05
0.01

-0.04
0.01

0.03
0.02

0.01
0.24

0.02
0.1

-0.03
0.02

0.01
0.44

-0.02
0.09

-0.03
0.02

0.01
0.54

0.03
0.03

0.03
0.01

SIREN2

0.1
4.93E-16

0.1
2.41 E15

-0.05
0.01

-0.03
0.03

-0.01
0.3

0.02
0.07

0.01
0.42

0.01
0.98

0.01
0.54

—0.01
0.43

-0.05
0.01

-0.01
0.79

0.02
0.08

0.01
0.32

-0.03
0.01

0.04
0.01

0.01
0.96

0:. . “? no) r?“

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1 AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

224

NOTROUT2

0.21
8.82E-67

0.09
4.87E-14

-0.03
0.01

0.03
0.03

0.01
0.54

0.01
0.65

0.01
0.23

0.01
0.95

0.02
0.2

-0.02
0.05

-0.02
0.08

0.01
0.56

0.01
0.59

0.01
0.83

0.02
0.13

0.03
0.01

0.01
0.71

OFFDUTY

0.02
0.15

0.01
0.62

0.04
0.01

0.04
0.01

-0.02
0.1

0.01
0.97

0.12
0.21

-0.02
0.17

-0.01
0.95

-0.01
0.26

-0.01
0.34

0.01
0.4

0.02
0.14

0.01
0.67

-0.02
0.1

0.01
0.48

0.01
0.45

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Officers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
' Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

225

BACKUP

0.17
1.22E-47

-0.01
0.38

0.03
0.02

-0.03
0.01

-0.01
0.51

-0.01
0.72

0.01
0.29

0.05
0.01

0.02
0.05

-0.06
1.42E-07

0.04
0.01

-0.03
0.01

-0.01
0.31

-0.02
0.21

0.04
0.01

0.09
2.62E—13

NUMBERPO

-0.08
1 .50326E-10

-0.05
0.01

-0.01
0.3

0.01
0.91

-0.01
0.8

0.05
0.01

0.03
0.02

-0.02
0.05

-0.05
0.01

0.05
6975-06

-0.01
0.68

-0.02
0.16

-0.02
0.07

0.01
0.32

0.05
0.01

 

Table C-1 (eont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1 AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

226

OFF1AGE

-0.03
0.01

-0.03
0.02

-0.02
0.05

0.06
1.01 E-07

-0.01
0.27

0.02
0.04

-0.05
0.01

0.04
0.01

-0.09
1.09E—14

0.01
0.75

0.01
0.41

0.01
0.63

0.01
0.72

0.03
0.03

BLACK1

-0.12
7.43E—22

-0.07
4.20E-08

-0.04
0.01

0.05
0.01

-0.04
0.01

-0.04
0.01

-0.01
0.79

0.11
2.69E-18

-0.02
0.06

0.01
0.82

0.02
0.05

-0.01
0.82

0.01
0.27

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

227

HISP1

-0.05
0.01

0.03
0.01

0.01
0.33

-0.02
0.19

-0.06
8.58E-07

-0.02
0.05

-0.05
0.01

0.09
4.77E-15

0.01
0.32

0.04
0.01

0.02
0.12

-0.01
0.18

OTH1

0.02
0.05

0.02
0.18

0.01
0.95

-0.05
7.55E-06

-0.01
0.96

-0.03
0.03

0.03
0.01

0.05
9455-06

-0.01
0.5

-0.02
0.18

-0.02
0.1

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

228

MALE1

0.01
0.87

-0.01
0.26

0.02
0.05

—0.01
0.42

-0.01
0.94

0.02
0.06

-0.02
0.1

0.01
0.62

0.05
7.29E-06

0.01
0.39

PSDEMEAN

-0.01
0.91

-0.03
0.02

0.01
0.71

0.02
0.13

0.01
0.95

-0.01
0.83

0.01
0.51

0.02
0.08

0.018
0.13

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Officer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

229

MEDPRIOR

0.03
0.01

0.02
0.05

-0.04
0.01
0.05
0.01

-0.01
0.58

-0.04
0.01
-0.02
0.1

0.01
0.66

REPEAT

0.01
0.77

-0.01
0.83

-0.029
0.02

-0.03
0.02

-0.12
3.24E-25

-0.02
0.19

-0.02
0.09

I! 3'1" ‘- .. Inrn.‘

 

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

230

SUSPAGE

-0.02
0.06

-0.08
7.36E-11

-0.01
0.35

-0.03
0.02

0.03
0.02

0.01
0.46

BLACKS

—0.37
3335-217

014
1235-29

0.02
0.11

-0.01
0.34

0.04
0.01

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1 AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

231

HISPS

-0.07
2.38E-08

0.01
0.74

0.08
4625-12

-0.01
0.33

OTHS

-0.01
0.26

0.01
0.58

0.02
0.06

 

Table C-1 (cont’d)

BACKUP
Called for Backup

NUMBERPO
Number of Ofﬁcers at Completion of
Arrest

OFF1AGE
Actual Age of First Ofﬁcer

BLACK1
First Ofﬁcer Black

HISP1
First Ofﬁcer Hispanic

OTH1
First Ofﬁcer Other Race

MALE1
First Ofﬁcer is Male

PSDEMEAN
Police Demeanor Toward Suspect

MEDPRIOR
Prior Medical Attention to Ofﬁcer

REPEAT
Surveys Completed by this Ofﬁcer

SUSPAGE
Actual Age of First Suspect

BLACKS
Black Suspect

HISPS
Hispanic Suspect

OTHS
Other Race Suspect

SMISS
Missing Race Suspect

MALES
Suspect is Male

BELASSLT
Police Believe Suspect to be Assaultive

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

232

SMISS MALES

-0.01
0.44

-0.01
0.93

-0.01
0.84

BELASSLT

Table C-1 (cont’d)

KNOWEAPN

Police Believe Suspect Carries Weapon

GANG_RG

Police Knowledge of Gang Membership

INTOX

Suspect is Intoxicated

VFRIEND

Victim is Friend of Suspect

VFAMILY

Victim is Related to Suspect

VUNK

Victim and Suspect have Unknown

Relationship
BUNK

Victim and Bystanders have Unknown

Relationship
BSTRANGE

Bystander is Stranger to Suspect

BFRIEND

Bystander is Friend of Suspect

BFAMILY

Bystander is Related to Suspect

SPDEMEAN

Suspect Demeanor Toward Police

PHYSSUS

Suspect Uses Physical Force

PTHREAT

Police Use or Threaten Physical Force

MAXIMUM

Average Maximum Force by Police

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

233

0.12
4.36E-25

0.12
9.05E-23

0.06
1.83E-06

0.03
0.02

-0.1
5.77E-17

0.11
2.26E-20
-0.01
0.99

—0.01
0.6

0.04
0.01

-0.08
4.42E-12

0.07
4.78E-09

0.06
3.63E-07

0.07
1 .43E-09

0.07
3.00E-09

LOCRIME LOCHAZRD

0.2
2.39E-64

0.13
8.84E-28

0.06
1.68E—07

0.04
0.01

-0.02
0.07

0.08
8.35E-11
-0.01

0.6

-0.01
0.42

0.07
1.87E-08

-0.01
0.46

0.09
6.15E-15

0.09
4.49E-13

0.09
9.26E-13

0.07
2.76E-08

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

234

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

INSIDE

0.04
0.01

0.05
0.01

-0.09
2.40E-13

-0.03
0.01

0.14
1.64E-33

-0.17
3.51E-46

-0.02
0.06

0.07
3.23E-08

0.06
1.38E-07

0.14
7.39E-33

-0.04
0.01

-0.04
0.01

-0.03
0.01

-0.03
0.01

VISIBLE

-0.04
0.01

-0.01
0.67

-0.11
3.00E-20

-0.02
0.12

-0.03
0.02

-0.03
0.03

-0.01
0.34

0.06
1.62E-07

—0.01
0.48

-0.05
0.01

-0.09
9.34E-13

-0.09
3555-14

-0.1
1.24E-15

-0.07
7.17E-09

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

235

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

VIOLENT

0.09
3.83E-14

0.02
0.08

-0.01
0.71

-0.03
0.02

0.3
9.73E-139

-0.28
2.59E-119

-0.02
0.2

-0.01
0.3

0.06
4.26E-07

0.1
8.51E-16

0.09
4.70E-13

0.11
4.21E-19

0.11
7.28E-19

0.09
1.25E-13

WEEKEND

-0.01
0.31

-0.02
0.14

0.08
1.33E-11

-0.02
0.18

0.05
0.01

-0.06
1.72E-06

0.02
0.05

-0.01
0.33

—0.01
0.8

0.04
0.01

0.03
0.01

0.01
0.27

0.03
0.03

0.02
0.08

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
\ﬁctim is Friend of Suspect

VFAMILY
\ﬁctim is Related to Suspect

VUNK
Victim and Suspect have Unknown
Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

236

BPDEMEAN NUMBERSO

0.07
7.66E-10

0.06
2.13E-07

0.04
0.01

0.01
0.28

-0.01
0.86

0.02
0.06
-0.01
0.23

-0.01
0.88

0.13
1.51E—25

0.1
2.14E-17

0.19
9.99E-57

0.14
1565-30

0.13
8.26E-27

0.08
4.96E-12

0.13
1.93E-26

0.05
0.01

0.01
0.42

0.01
0.46

-0.04
0.01

0.05
0.01
0.06
3.11E—07

-0.02
0.15

-0.01
0.42

-0.06
4.31 E-06

0.02
0.06

-0.01
0.43

0.06
2.61 E-07

0.08
2805-12

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

237

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

CUSTODY

-0.03
0.01

0.01
0.46

-0.09
1.65E-12

-0.06
4.48E-07

-0.08
2.82E-11

-0.06
3.35E—07

-0.04
0.01

0.14
6885-30

—0.06
1.90E-07

-0.05
0.01

-0.04
0.01

-0.05
0.01

-0.08
9.47E-11

-0.09
9.07E-13

CIT2

0.01
0.83

-0.01
0.32

0.01
0.9

0.01
0.23

0.01
0.68

-0.01
0.24

0.02
0.1

0.03
0.03

0.01
0.24

-0.02
0.15

0.01
0.4

0.01
0.37

0.02
0.15

0.03
0.02

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

238

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

POL2

0.06
1.05E-06

0.05
0.01

-0.06
1.47E-06

0.06
1.96E—07

-0.14
6.81 E-30

0.18
1.60E-52

-0.05
0.01

-0. 14
5.99E-32

0.01
0.46

-0.05
0.01

-0.08
2.11E-10

-0.04
0.01

0.02
0.04

0.02
0.05

OTHAP2

-0.01
0.38

-0.05
0.01

0.06
3455-07

0.01
0.26

-0.06
2.31E-06

0.13
5235-27

0.1
1.60E-16

0.05
0.01

-0.08
1.57E-10

-0.08
3.62E-1 0

-0.01
0.22

—0.02
0.13

-0.04
0.01

—0.03
0.01

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

239

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-talled)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

PRIO2

0.04
0.01

0.01
0.88

0.02
0.07

—0.04
0.01

0.13
1.72E-26

-0. 16
2.42E-40

-0.07
1 .42E-09

0.03
0.02

0.05
0.01

0.08
1.86E-10

0.08
2.20E-11

0.07
2.00E-08

0.1
5.7OE-16

0.07
1.63E-08

SIREN2

0.04
0.01

0.03
0.02

0.02
0.12

0.01
0.89

-0.06
2.44E-07

0.06
8.58E-07

-0.02
0.19

-0.04
0.01

-0.01
0.83

-0.04
0.01

0.04
0.01

0.04
0.01

0.07
1.95E-09

0.07
5.20E-09

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

240

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

NOTROUT2 OFFDUTY

0.04
0.01

0.02
0.05

0.01
0.6

0.01
0.28

0.01
0.66

0.02
0.04

0.03
0.02

-0.03
0.04

-0.01
0.32

-0.02
0.1

0.01
0.26

0.01
0.3

0.06
1.72E-07

0.03
0.03

0.03
0.01

0.02
0.12

0.01
0.98

-0.01
0.98

-0.02
0.13

-0.02
0.08

0.04
0.01

0.01
0.93

0.01
0.82

-0.02
0.07

0.02
0.08

0.04
0.01
0.03
0.01

0.02
0.11

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

241

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-talled)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

BACKUP NUMBERPO

0.07
1.73E-09

0.06
4095—06

-0.03
0.01

-0.01
0.74

0.01
0.49

—0.06
2.48E-06

0.01
0.38

-0.01
0.48

0.06
7.86E—07

0.01
0.29

0.08
1.20E-11

0.08
5.28E-11

0.16
2.54E-42

0.18
4.53E-52

0.16
5455-40

0.06
1.67E-06

0.03
0.01

-0.02
0.09

-0.01
0.35

0.01
0.35

0.03
0.02

0.01
0.78

0.05
0.01

0.01
0.7

0.12
4.19E—23

0.12
6295-23

0.19
1.02E-59

0.22
2.25E-78

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

242

Pearson
Correlation
Sig. (2-tai'led)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

OFF1 AGE BLACK1

0.01
0.92

-0.04
0.01

~0.02
0.06

-0.01
0.22

0.01
0.66

-0.01
0.34

0.01
0.59

0.02
0.05

-0.02
0.14

0.01
0.59

-0.05
0.01

-0.03
0.01

-0.05
6.78E-06

-0.04
0.01

-0.02
0.1

0.03
0.02

0.01
0.8

0.01
0.36

—0.01
0.27

-0.01
0.88

0.01
0.65

0.01
0.73

-0.01
0.88

0.01
0.53

-0.01
0.69

0.01
0.76

0.01
0.67

0.01
0.58

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Wctim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

243

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

HISP1

-0.01
0.45

-0.04
0.01

0.03
0.03

0.01
0.62

0.03
0.02

0.01
0.45

0.02
0.07

-0.01
0.99

-0.02
0.09

-0.02
0.09

0.02
0.19

0.01
0.63

0.03
0.02

-0.03
0.02

OTH1

0.01
0.69

-0.01
0.93

0.01
0.91

0.02
0.09

-0.03
0.02

0.03
0.03

0.04
0.01

-0.03
0.03

-0.02
0.15

-0.02
0.05

-0.14
0.22

-0.02
0.07

-0.01
0.68

0.01
0.61

 

Table C-1y(cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

244

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

MALE1

0.01
0.54

0.02
0.07

0.01
0.65

-0.01
0.22

-0.02
0.15

0.02
0.18

0.02
0.05

0.01
0.72

0.01
0.5

-0.01
0.29

0.01
0.34

—0.01
0.58

0.05
0.01

0.03
0.01

PSDEMEAN

0.03
0.01

-0.01
0.59

0.02
0.21

-0.01
0.87

-0.01
0.6

-0.01
0.59

-0.01
0.98

-0.01
0.88

0.01
0.82

-0.01
0.28

0.1
2.12E-17

0.06
2.38E-06

0.06
2275-07

0.05
5.25E-06

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

245

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

MEDPRIOR REPEAT

0.01
0.22

-0.04
0.01

0.03
0.01

-0.01
0.48

0.03
0.03

-0.02
0.21

0.03
0.02

0.05
0.01

0.02
0.16

0.03
0.03

0.04
0.01

0.07
1215-09

0.06
3.36E-07

0.08
3.01 E-11

-0.03
0.01

-0.03
0.02

0.05
0.01

0.01
0.31

-0.02
0.17

0.06
1.73E-06

0.02
0.09

-0.03
0.01

0.04
0.01

0.02
0.11

0.01
0.33

0.01
0.89

0.02
0.1

0.1
1.23E-16

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

246

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

SUSPAGE

-0.07
1.13E-08

-0.08
6.97E-11

0.13
1.46E-25

-0.01
0.23

0.05
0.01

-0.01
0.57

0.02
0.06

0.01
0.47

-0.06
7.455-07

-0.02
0.09

0.02
0.05

-0.01
0.46

-0.02
0.09

0.01
0.58

BLACKS

0.05
0.01

0.1
2.95E-18

~0.1
1.19E-17

0.02
0.04

-0.01
0.45

0.04
0.01

-0.11
6.44E-19

-0.01
0.3

0.06
1 .38E-07

0.04
0.01

0.02
0.05

0.05
0.01

0.05
0.01

0.02
0.06

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Wctim is Friend of Suspect

VFAMILY
Wctim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

247

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)

HISPS

-0.01
0.68

-0.04
0.01

0.06
1.385-07

-0.02
0.05

0.02
0.09

-0.01
0.37

0.09
2.12E-14

-0.01
0.83

-0.03
0.02

0.01
0.73

0.01
0.97

0.01
0.31

0.01
0.37

0.01
0.33

OTHS

0.02
0.05

-0.01
0.46

0.02
0.1

0.02
0.17

-0.01
0.73

0.02
0.11

0.11
1.46E-21

-0.01
0.25

-0.01
0.95

-0.02
0.07

-0.02
0.1

-0.01
0.63

0.01
0.76

0.01
0.6

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
\ﬁctim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

248

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

SMISS

—0.02
0.09

-0.05
0.01

-0.01
0.71

0.01
0.42

0.01
0.71

0.03
0.01

0.04
0.01

-0.01
0.94

-0.01
0.52

-0.01
0.45

0.01
0.55

-0.01
0.63

-0.01
0.9

-0.01
0.65

MALES

0.06
3.69E-07

0.02
0.06

0.05
0.01

-0.02
0.09

0.03
0.03

-0.02
0.1

0.03
0.04

-0.01
0.27

-0.01
0.25

-0.01
0.47

0.04
0.01
0.01

0.3

0.06
2.38E-06

0.07
4.095-08

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

249

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

BELASSLT KNOWEAPN

0.32
4.655-161

0.17
2145-46

0.04
0.01

-0.02
0.07

0.04
0.01

-0.05
0.01

-0.01
0.51

-0.02
0.13

0.03
0.01

0.01
0.21

0.09
6.53E-13

0.07
2.41E-08

0.08
3.06E-11

0.07
3.49E-08

1

0.29
3.53E-132

0.01
0.25

0.02
0.18

-0.03
0.03

-0.01
0.95

0.01
0.4

-0.02
0.21

0.03
0.02

-0.01
0.79

0.05
0.01

0.04
0.01

0.12
4.01 E-22

0.11
4.68E-20

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

250

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

GANG_RG

-0.01
0.61

0.01
0.53

-0.05
0.01

0.01
0.68

-0.03
0.02

-0.02
0.12

0.01
0.28

-0.01
0.63

0.05
0.01

0.03
0.01

0.04
0.01

-0.01
0.26

INTOX

-0.01
0.97

0.02
0.09

0.05
0.01

0.08
4.16E-10

-0.01
0.23

-0.02
0.05

-0.02
0.19

0.18
2.11E-51

0.12
3.24E-23

0.11
4.84E-21

0.12
1.03E-22

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

251

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

VFRIEND VFAMILY

-0.33
1.68E-168

0.32
4.26E-159

0.01
0.83

-0.08
3.19E-12

0.08
4.41 E-11

0.01
0.63

-0.018
0.14

-0.03
0.02

—0.02
0.05

-0.02
0.07

-0.11
1.76E-19

0.01
0.84

-0.06
3.64E-06

0.03
0.01

0.19
6.57E-56

0.02
0.11

0.02
0.06

0.02
0.18

-0.01
0.48

 

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

252

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

VUNK

0.04
0.01

-0.13
4.28E-29

0.01
0.4

-0.06
2.30E-06

-0.03
0.01

-0.03
0.02

-0.03
0.07

-0.02
0.14

BUNK

-0.12
1.37E-23

-0.12
8.96E-24

-0.09
7.52E-13

0.02
0.09

0.03
0.01

0.02
0.04

0.05
0.01

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

253

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

BSTRANGE BFRIEND

-0.17
2.42E-45

-0.12
8.20E-24

0.03
0.02

0.02
0.04

0.03
0.02

0.04
0.01

-0.12
5.36E-24

0.04
0.01

0.06
4.97E-07

0.06
4.92E-06

0.06
0.01

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Wctim is Friend of Suspect

VFAMILY
\ﬁctim is Related to Suspect

VUNK
Victim and Suspect have Unknown Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

254

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

BFAMILY SPDEMEAN

0.01
0.26

0.03
0.01

0.03
0.01

0.02
0.12

0.48
0

0.35
8.67E-202

0.24
1.34E-91

Table C-1 (cont’d)

KNOWEAPN
Police Believe Suspect Carries Weapon

GANG_RG
Police Knowledge of Gang Membership

INTOX
Suspect is Intoxicated

VFRIEND
Victim is Friend of Suspect

VFAMILY
Victim is Related to Suspect

VUNK
Victim and Suspect have Unknown
Relationship

BUNK
Victim and Bystanders have Unknown
Relationship

BSTRANGE
Bystander is Stranger to Suspect

BFRIEND
Bystander is Friend of Suspect

BFAMILY
Bystander is Related to Suspect

SPDEMEAN
Suspect Demeanor Toward Police

PHYSSUS
Suspect Uses Physical Force

PTHREAT
Police Use or Threaten Physical Force

MAXIMUM
Average Maximum Force by Police

Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation

Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2—tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)
Pearson
Correlation
Sig. (2-tailed)

255

PHYSSUS PTHREAT MAXIMUM

0.48
0

0.32
3.16E-165

Appendix D

256

 

Table D-1: Descriptive Statistics
(NUMBERSO, NUMBERPO, OFF1AGE, SUSPAGE,

N
O

REPEAT)
VARIABLE NAME VALUE N MEAN SD
Number of Suspects at
Completion of Arrest 5932 1.39 1.18
1 5371
2 927
3 323
4 125
5 40
6 31
7 18
8 8
9 2
10 5
14 1
17 1
20 1
35 3
Number of Officers at
Completion of Arrest 5932 2.50 2.00
1 1536
2 3028
3 1 180
4 550
5 197
6 161
7 62
8 63
9 3
10 35
11 9
12 8
13 9
14 4
15 6
18 1
3
1

01
O

257

Table D-1 (cont’d)
VARIABLE NAME

Actual Age of First Ofﬁcer

VALUE

258

20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
60
70

5932
1
3
43
74
1 83
423
471
443
504
41 7
754
424
444
325
277
31 5
21 7
1 69
1 65
1 22
1 97
96
1 29
95
95
1 02
64
62
63
, 30
55
1 9
24
23
1 0
1 3
2

1
1
1

MEAN

32.46

SD

6.57

 

Table D-1 (cont’d)

VARIABLE NAME

Actual Age of Suspect

VALUE

259

16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60

5932
40
1 32
257
269
361
247
228
230
263
275
268
236
233
1 97
257
240
262
257
248
238
209
205
205
1 85
1 87
1 27
1 16
1 22
1 05
96
88
77
60
52
44
34
31
21
20
1 3
21
1 0
1 0
1 1
1 0

MEAN

31.17

SD

9.76

 

Table D-1 (cont’d)

Actual Age of Suspect 61
62
63
54
65
66
68
69
70
72
73
74
75
78

VALUE
Number of Surveys Completed
by this Officer

u—l
OCDCDNCDOIbOON—l

NM—l—L—l—L—L—t—L
w—INODOISAOON—I

260

AAAachow-xcocooicoco

N

5932
1445
1049

930
778
539
449
394
250
277
21 3
161
69
38
109
25
45
48
1 9
1 8

MEAN
4.60

SD
4.00

 

Appendix E

261

The data used within this dissertation were obtained from the Police Use
of Force Study. These data have been previously analyzed at the individual level,
with an emphasis on explaining how these encounter-level characteristics can
inﬂuence use of force behavior (Garner et al., 2002). The latter study noted that
the association between these independent variables and the dependent
variables was inﬂuenced by the incorporation of suspects’ resistance, as well as
the speciﬁc measure of force used. The tables presented below examine the
effects of introducing neighborhood-level variables into the models used in these

previous analyses.

Table E-1: Maximum Force Model

SEVERITY
Garner et
al. (2002) Heraux (2006)
Variable b sig b slg Consistency
SD Police 5.03 Yes NIA NIA Not Compared
DPD 3.95 Yes 3.73 Yes Both Signiﬁcant
SD Sheriff 6.99 Yes 4.44 Yes Both Signiﬁcant
Charlotte 3.43 Yes 0.30 No Garner et al. Sig / Heraux NS
St. Pete 3.65 Yes 10.32 Yes Both Signiﬁcant
Loc. Criminal Activity 0.21 No 0.39 Yes Garner et al. NS I Heraux Sig.
Loc. Hazardous 0.27 No -0.08 No Both Not Signiﬁcant
Arrest Inside N/A N/A 0.13 No Not Compared
Visibility -0.18 Yes 014 Yes Both Signiﬁcant
Violent Offense 1.25 Yes 0.72 Yes Both Signiﬁcant
Weekend 0.42 Yes 0.15 No Garner et al. Sig / Heraux NS
Bystander Demeanor 0.04 No 003 No Both Not Signiﬁcant
# Suspects 0.55 Yes -0.01 No Garner et al. Sig I Heraux NS
Patrol Division 0.37 No NIA NIA Not Compared
Suspect in Custody -0.54 Yes 036 No Garner et al. Sig I Heraux NS
Citizen Initiated 0.07 No 0.26 No Both Not Signiﬁcant
Police Initiated 0.22 No 0.20 No Both Not Signiﬁcant
Unknown Initiated -0.14 Yes 041 No Garner et al. Sig / Heraux NS
Priority Call Approach 1.56 Yes 1.14 Yes Both Signiﬁcant
Lights and Sirens
Approach 2.57 Yes 1.22 Yes Both Signiﬁcant
Unknown Approach -0.23 No 022 No Both Not Signiﬁcant

262

Table E-1 (cont’d)

Off-Duty NIA N/A 1.08 Yes Not Compared
Called for Backup 1.21 Yes 1.05 Yes Both Signiﬁcant

# of Ofﬁcers 2.61 Yes 0.72 Yes Both Signiﬁcant
Ofﬁcer Age -2.00 Yes 006 Yes Both Signiﬁcant
Black Ofﬁcer 018 No 020 No Both Not Signiﬁcant
Hispanic Ofﬁcer 055 No 038 No Both Not Signiﬁcant
Other Race Ofﬁcer 0.46 No 1.32 No Both Not Signiﬁcant
Male Ofﬁcer 1.10 Yes 0.56 Yes Both Signiﬁcant
Police Demeanor 2.75 Yes 2.12 No Garner et al. Sig I Heraux NS
Prior Medical

Attention 0.87 Yes 0.33 No Gamer et al. Sig I Heraux NS
# of Surveys N/A NIA -0.02 No Not Compared
Suspect Age 005 No -0.02 No Both Not Signiﬁcant
Black Suspect 0.02 No 0.09 No Both Not Signiﬁcant
Hispanic Suspect 0.34 No 0.54 No Both Not Signiﬁcant
Other Race

Suspect -0.93 No 0.60 No Both Not Signiﬁcant
Missing Race

Suspect 0.06 No 002 No Both Not Signiﬁcant
Male Suspect 0.85 Yes 0.56 Yes Both Signiﬁcant
Suspect Assaultive 0.14 No 0.26 No Both Not Signiﬁcant
Suspect Weapon 1.98 Yes 1.65 Yes Both Signiﬁcant
Suspect Gang NIA N/A 0.52 No Not Compared

Suspect Intoxicated 0.28 No 0.29 No Both Not Signiﬁcant
Victim and Suspect

Friends -1.17 Yes -0.59 Yes Both Signiﬁcant

Victim and Suspect

Family -1.47 Yes -0.51 Yes Both Signiﬁcant

Unknown Victim -0.82 Yes -0.13 No Garner et al. Sig I Heraux NS
Unknown

Bystander and

Suspect 0.91 Yes 0.74 No Garner et al. Sig I Heraux NS
Bystander and

Suspect Strangers 0.52 Yes 0.44 Yes Both Signiﬁcant

Bystander and

Suspect Friends 0.44 No 0.23 No Both Not Signiﬁcant
Bystander and

Suspect Family 0.10 No 0.01 No Both Not Signiﬁcant

Suspect

Antagonistic 1.11 Yes 1.61 Yes Both Signiﬁcant

Suspect Physical

Resistance 7.30 Yes 5.92 Yes Both Signiﬁcant

Of the 52 independent variables tested in the maximum force model either
by Garner et al. (2002) or in this dissertation, 6 were not directly compared due to
their removal in either the former or the latter research due to multicollinearity. Of

the remaining 46 variables, 36 produced the same result in both sets of research

263

 

(i.e. the variable was statistically signiﬁcant in both Garner et al. (2002) and in
this dissertation, or the variable was not statistically signiﬁcant in both Garner et
al. (2002) and in this dissertation). Of the 10 instances where a variable’s effects
were different between the two sets of research, in 9 of those instances the
variable was statistically signiﬁcant in Garner et al. (2002) and not statistically
signiﬁcant in this dissertation. In the remaining instance, the variable was not
statistically signiﬁcant in Garner et al. (2002), yet was statistically signiﬁcant in
this dissertation. These results indicate that the inclusion of neighborhood-level
variables in the current research has reduced the inﬂuence of numerous
individual-level variables from prior research (Garner et al., 2002) below
statistical signiﬁcance. Considering that Garner et al. (2002: 743) acknowledged
that the lack of neighborhood-level variables was a concern, it is believed that the
results of the current research are theoretically and methodologically sound.

The results for the prevalence of force model are presented below,

Table E-2: Prevalence of Force Model

PREVALENCE
Garner et al. Heraux
(2002) (2006)
Variable B sig B sig Consistency

SD Police 1.00 No NIA N/A Not Compared
DPD 0.85 No 0.73 No Both Not Signiﬁcant
SD Sheriff 0.87 No 0.99 No Both Not Signiﬁcant
Charlotte 0.83 No 1.18 No Both Not Signiﬁcant
St. Pete 1.78 Yes 1.31 No Garner et al. Sig I Heraux NS
Loc. Criminal Activity 1.32 Yes 1.17 No Garner et al. Sig I Heraux NS
Loc. Hazardous 0.94 No 0.97 No Both Not Signiﬁcant
Arrest Inside 1.00 No 1.02 No Both Not Signiﬁcant
Visibility 0.02 No 0.96 Yes Garner et al. NS/ Heraux Sig.
Violent Offense 1.17 No 1.34 Yes Garner et al. NS I Heraux Sig.
Weekend 1.18 Yes 1.15 No Gamer et al. Sig l Heraux NS
Bystander Demeanor 1.47 Yes 1.17 No Garner et al. Sig / Heraux NS

264

 

 

 

 

Table E-2 (cont’d)
# Suspects

Patrol Division
Suspect in Custody
Citizen Initiated

Police Initiated
Unknown Initiated
Priority Call Approach
Lights and Sirens
Approach

Unknown Approach
Off-Duty

Called for Backup

# of Ofﬁcers

Ofﬁcer Age

Black Ofﬁcer

Hispanic Ofﬁcer

Other Race Ofﬁcer
Male Ofﬁcer

Police Demeanor
Prior Medical Attention
# of Surveys

Suspect Age

Black Suspect
Hispanic Suspect
Other Race Suspect
Missing Race Suspect
Male Suspect

Suspect Assaultive
Suspect Weapon
Suspect Gang
Suspect Intoxicated
Victim and Suspect
Fnends

Victim and Suspect
Family

Unknown Victim
Unknown Bystander
and Suspect
Bystander and Suspect
Strangers

Bystander and Suspect
Fnends

Bystander and Suspect
Family

Suspect Antagonistic
Suspect Physical
Resistance

0.92
NIA
0.74
1.10
1.29
0.87
1.40

1.44
1.40
1.26
1.51
1.42
0.57
1.07
1.52
1.07
1.72
1.18
1.30
0.92
1.00
1.19
1.19
1.77
1.24
1.42
1.19
1.59
0.66
1.23

0.75

0.92
0.88

1.44
1.15
1.24

1.28
2.63

19.00

No
N/A
Yes

No
Yes

No
Yes

Yes
Yes
No
Yes
Yes
Yes
No
Yes
No
Yes
No
Yes
No
No
No
No
Yes
No
Yes
No
Yes
Yes
Yes

No

No
No

Yes
No
No

No
Yes

Yes

1.00
NIA
0.63
1.23
1.48
1.04
1.61

1.41
1.45
1.21
1.69
1.16
0.99
1.06
1.72
1.32
2.02
1.77
1.34
1.00
1.00
1.09
1.13
1.35
1.46
1.47
1.03
1.91
0.93
1.35

0.93

1.00
0.97

1.71
1.30
1.16

1.30
2.50

10.35

265

No
N/A
Yes

No
Yes

No
Yes

Yes
Yes
No
Yes
Yes
No
No
Yes
No
Yes
No
Yes
No
No
No
No
No
No
Yes
No
Yes
No
Yes

No

No
No

Yes
Yes
No

No
Yes

Yes

Both Not Signiﬁcant
Not Compared

Both Signiﬁcant
Both Not Signiﬁcant
Both Signiﬁcant
Both Not Signiﬁcant
Both Signiﬁcant

Both Signiﬁcant

Both Signiﬁcant

Both Not Signiﬁcant

Both Signiﬁcant

Both Signiﬁcant

Garner et al. Sig / Heraux NS
Both Not Signiﬁcant

Both Signiﬁcant

Both Not Signiﬁcant

Both Signiﬁcant

Both Not Signiﬁcant

Both Signiﬁcant

Both Not Signiﬁcant

Both Not Signiﬁcant

Both Not Signiﬁcant

Both Not Signiﬁcant

Garner et al. Sig l Heraux NS
Both Not Signiﬁcant

Both Signiﬁcant

Both Not Signiﬁcant

Both Signiﬁcant

Garner et al. Sig I Heraux NS
Both Signiﬁcant

Both Not Signiﬁcant

Both Not Signiﬁcant
Both Not Signiﬁcant

Both Signiﬁcant
Garner et al. NS l Heraux Sig.
Both Not Signiﬁcant

Both Not Signiﬁcant
Both Signiﬁcant

Both Signiﬁcant

Of the 51 independent variables tested in the prevalence of force model
either by Garner et al. (2002) or in this dissertation, 2 were not directly compared
due to their removal in either the former or the latter research due to
multicollinearity. Of the remaining 49 variables, 39 produced the same result in
both sets of research (i.e. the variable was statistically signiﬁcant in both Garner
et al. (2002) and in this dissertation, or the variable was not statistically
signiﬁcant in both Garner et al. (2002) and in this dissertation). Of the 10
instances where a variable’s effects were different between the two sets of
research, in 7 of those instances the variable was statistically signiﬁcant in
Garner et al. (2002) and not statistically signiﬁcant in this dissertation. In the
remaining 3 instances, the variables were not statistically signiﬁcant in Garner et
al. (2002), yet were statistically signiﬁcant in this dissertation. Once again, these
results indicate that the inclusion of neighborhood-level variables in the current
research has reduced the inﬂuence of numerous individual-level variables from

prior research (Garner et al., 2002) below statistical signiﬁcance.

266

 

Appendix F

267

The analyses in this dissertation used 350 Level-2 units (i.e. census tracts,
the measure of neighborhood in the current work). However, the initial data
collection encompassed 1,206 census tracts, distributed as follows: (1) Charlotte-
Mecklenberg, NC had 128 census tracts; (2) Colorado Springs, CO had 97
census tracts; (3) Dallas, TX had 304 census tracts; (4) St. Petersburg, FL had
72 census tracts; and (5) San Diego, CA (city and county combined) had 605
census tracts. Of this total, 604 were removed from the analysis due to the fact
that no arrest had occurred in that tract during data collection, and an additional
252 were removed from the analysis due to the fact that fewer than the
appropriate HLM threshold of 5 cases (i.e. arrests) had occurred in that tract
during data collection, leaving the ﬁnal sample size of 350 census tracts. In other
words, the limitation in this dissertation is that we are unable to include all census
tracts for two reasons: (1) not all tracts had an arrest during the data collection
period”; and (2) tracts with only one through four arrests present too little
variation. The following discussion evaluates these tracts and seeks to
determine: (1) their nature; and (2) the impact of removing them from the
analyses. 9

It is instructive to begin by examining the impact of removing census tracts
with no arrests by creating a dichotomous (i.e. arrest vs. no arrest) variable. As
no arrests occurred in 604 of these tracts, a comparison on the dependent
variables is not possible, since PUF only sampled use of force incidents in an

arrest encounter. However, all of these tracts were used to create the

 

’6 This is due in large part to the fact that the data was originally collected to examine Level-l effects, and
thus not enough time was spent to ensure that data was collected in all neighborhoods. Such issues are
inherent Iirnitations in using secondary data.

268

 

 

 

concentrated disadvantage factor score which was entered into the analysis as a
neighborhood-level variable. Therefore, differences in the variables used to
create this factor can be analyzed in order to determine if there are signiﬁcant

differences between census tracts with and without an arrest.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table F-1: Mean Differences in Arrest — No Arrest Census Tracts
Variable 6-SITE CMPD CSPD DPD SPPD SDPD SDSO
BLACK -14.22 4.08 -5.71 -5.49 -2.22 4.86 -.26

POP
(.00*) (.00*) (.00*) (.00*) (.03*) (.00*) (.80)
POV -12.29 4.23 -3.92 -7.78 -3.83 -9.13 .68
LEVEL
(.00*) (.00*) (.00*) (.00*) (.00*) (.00*) (.50)
PUB -7.27 -3.60 -.93 -5.11 -3.28 -7.51 .23
ASST
(.00*) (.00*) (.36) (.00*) (.00*) (.00*) (.82)
FEM -1 0.58 -2.98 -3.61 -5.65 4.20 -6.96 1.09
HOUSE
(.00*) (.00*) (.00*) (.00*) (.00*) (.00*) (.28)
TOT -7.67 4.33 -5.54 -5.66 4.30 -5.76 .47
UNEM
(.00*) (.00*) (.00*) (.00*) (.00*) (.00*) (.64)
DEN -2.58 -1 .84 4.58 4.54 4.66 -7.52 3.83
TOT18
(.01*) (.07) (.00*) (.00*) (.00*) (.00*) (.00*)
SPAT 45.86 48.80 -51.72 -35.11 -68.18 -102.39 -90.17
LAGZ
(.00*) (.00*) (.00*) (.00*) (.00*) (.00*) (.00*)

 

269

 

 

Of the 49 mean differences in these analyses, 42 were statistically
signiﬁcant, indicating that in the majority of situations there are signiﬁcant
differences in the demographic characteristics of census tracts with and without
arrests. The exception is the county of San Diego (excluding the city of San
Diego), in which there were signiﬁcant differences only in the density of
individuals under the age of 18, and in the spatial error correction term. Given
that such signiﬁcant differences exist between those census tracts with and
without arrests, it is possible that these demographic characteristics are in fact
better predictors of arrests than of use of force events. In order to explore this
possibility, a logistic regression was conducted using the dichotomous arrest

variable (i.e. arrest vs. no arrest) as a dependent variable.

 

 

 

 

 

 

 

 

 

 

 

 

Table F-2: Logistic Regression Analysis

Variable Name B Signiﬁcance
CONSTANT -5.78 .00*
BLACKPOP .02 .23
POVLEVEL .06 .01*
PUBASST .04 .44
FEMHOUSE -.04 .26
TOTUNEM -.07 .10
DENTOT18 -.15 .45
CSPD 3.07 .00*
DPD 1.91 .00“

 

 

 

 

 

270

 

Table F-2 (eont’d)

 

 

 

 

 

 

SPPD 2.09 .17
SDPD -1.08 .00*
$080 -1.01 .00*
SPATLAG2 .01 .00*

 

 

MODEL FIT: -2 Log Likelihood = 486.775

 

The results of the logistic regression indicate that although there are

signiﬁcant mean differences between census tracts with and without arrests,

these demographic characteristics are not generally predictive of the presence or

absence of those arrests. In fact, the classiﬁcation table from the logistic

regression indicates that the observed arrest-no arrest census tracts match their

predicted census tracts in 88.1% of the cases. It is also possible to examine the

inﬂuences on arrest behavior by using a continuous variable regarding the

number of arrests as the dependent variable. Using a negative binomial model in

this analysis, we ﬁnd that the results differ considerably from those of the logistic

regression.

271

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table F-3: Negative Binomial Regression Analysis

Variable Name B Signiﬁcance
CONSTANT -1.96 .00“
BLACKPOP .008 .01*
POVLEVEL .07 .00*
PUBASST -.02 .45
FEMHOUSE -.03 .03*
TOTUNEM -.03 .07
DENTOT18 -.01 .88
CSPD 1.65 .00*
DPD -.45 .01*
SPPD .98 .00*
SDPD .12 .60
$080 .34 .88
SPATLAGZ .42 .00*
ALPHA 1.65 .00*
MODEL FIT: Restricted Log-Likelihood = 4901.006

 

The alpha, or dispersion parameter, from the model conﬁrms that the

negative binomial model is appropriate for these data. With respect to the

independent variables, three of the variables that comprise the concentrated

disadvantage factor score (blackpop, povlevel, and femhouse) exhibit a

statistically signiﬁcant effect on arrests within a census tract. Thus, in contrast to

 

 

the logistic regression analyses, the results of the negative binomial regression
indicate that differences between independent variables established in Table 1
are, in fact, predictive of arrests within neighborhoods.

Considering that the results of the negative binomial regression indicate
that arrest, when measured as a count variable, is inﬂuenced by neighborhood
characteristics, it is instructive to now examine whether differences exist between
those census tracts with 1 to 4 arrests and those tracts with 5 or more arrests.
There were 252 of the former and 350 of the latter used in the analyses for this
dissertation. Given that an arrest occurred in all of these census tracts, we are
now able to compare them on the dependent variables. The physical force plus
threats (PTHREAT) dependent variable was normalized by the number of arrests
in the tract to account for the fact that tracts with more arrests present more
opportunities for the use of force. Both the original variable and the normalized
variable (NORMPREV) are used here for comparisons across tracts. Results for
these analyses can be found in the table below. It should be noted that there
were no census tracts with only 1 to 4 arrests in the city of San Diego, and thus

the SDPD is excluded from these analyses.

273

 

Table F4: Mean Differences in 14 Arrests - 5 or more Arrest Census Tracts

 

 

 

 

 

 

 

 

 

 

 

 

Variable 6-SITE CMPD CSPD DPD SPPD SDPD spso

BLACK -5.95 -5.99 -2.48 -3.70 -2.47 NIA -.47
POP

(.00*) (.00*) (02*) (.00*) (02*) (.64)

POV -9.55 4.11 4.11 -6.89 -3.35 NIA -2.64
LEVEL

(.00*) (.00*) (.00*) (.00*) (00*) (.01*)

PUB -8.64 4.02 4.91 4.59 -2.85 N/A -.89
ASST

(.00*) (.00*) (.00*) (.00*) (01*) (.38)

FEM -7.54 4.55 -2.39 -3.77 -3.18 NIA -1.71
HOUSE

(.00*) (.00*) (02*) (.00*) (.00*) (.10)

TOT -5.15 -3.33 -2.70 4.79 3.75 NIA -1.07
UNEM

(.00*) (.00*) (.01*) (.00*) (.00*) (.29)

DEN -3.99 -.80 .23 -2.79 -2.60 N/A -.98
TOT18

(.00*) (.43) (.82) (.01*) (02*) (.33)

SPAT 3.57 -1.75 -.57 4.79 2.16 NIA .62
um

(.00*) (.08) (.57) (00*) (04*) (.54)

PTHREAT 41.45 -5.44 4.16 -955 -2.90 N/A -6.06

(.00*) (.00*) (.00*) (.00*) (.01*) (.00*)

MAX .85 .87 -1.54 -.01 1.46 NIA .90
FORCE

(.73) (.39) (.13) (.99) (.17) (.38)

NORM -.95 .04 -2.36 -.30 .31 NIA -.24
PREV

(.34) (.97) (02*) (.76) (.76) (.81)

 

 

 

 

 

 

 

 

274

 

 

Of the 60 mean differences in these analyses, 39 were statistically
signiﬁcant. This indicates that there are signiﬁcant differences in the
demographic characteristics of census tracts with 1 to 4 arrests and those with 5
or more arrests. However, it is important to note that there was only 1 statistically
signiﬁcant mean difference among the 12 mean differences for both dependent
variables ((MAXFORCE and NORMPREV). Thus, it would appear that the
number of arrests made in a census tract does not inﬂuence either the
prevalence or severity of force. This is further explored through regression

analyses on both dependent variables.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table F-5: Poisson Regression Analysis for Maximum Force
Variable Name B Signiﬁcance
CONSTANT 3.25 .00*
BLACKPOP .0002 .69
POVLEVEL -.0003 .82
PUBASST .004 .18
FEMHOUSE -.002 .44
TOTUNEM .0008 .68
DENTOT18 -.0005 .73
CSPD -.02 .50
DPD .12 .00“
SPPD .30 .00*
SDPD .09 .02*

 

 

275

 

Table F-5 (cont’d)

 

 

 

SDSO .1 6 .00*
SPATLAGZ .00002 .65
ARREST2 -.002 .92

 

 

 

MODEL FIT: Restricted Log-Likelihood = -1 871.328

 

 

 

The regression analysis‘37 presented in Table F-5 indicates that: (1) these
neighborhood-level variables fail to explain the maximum amount of force used in
an arrest encounter; and (2) the effect sizes for the statistically signiﬁcant
variables (city dichotomies) are extremely small. These results are consistent
with those of this dissertation which used only the 350 census tracts with 5 or
more arrests. More importantly, the dichotomous arrest variable (14 arrests vs. 5
or more arrests) used as an independent variable was not statistically signiﬁcant,
indicating that there is no appreciable difference in census tracts with 1 to 4
arrests when compared to census tracts with 5 or more arrests with regards to
the maximum amount of force used in an encounter. Results for the prevalence

of force are found in the table below.

 

87 Note that Poisson regression was used, as the overdispersion parameter indicated that negative binomial
was not appropriate.

276

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Table F-6: Negative Binomial Regression Analysis for Prevalence of
Force

Variable Name B Signiﬁcance
CONSTANT 2.52 .00*
BLACKPOP .003 .56
POVLEVEL .003 .82
PUBASST .03 .44
FEMHOUSE -.02 .30
TOTUNEM .00004 .99
DENTOT18 -.10 .51
CSPD -.41 .26
DPD -.08 .74
SPPD .15 .64
SDPD .67 .87
SDSO .13 .75
SPATLAGZ .0002 .70
ARREST2 .17 .40
ALPHA 3.51 .00*
MODEL FIT: Restricted Log-Likelihood = -8526.979

 

 

 

The regression analysis presented in Table F-6 indicates that these
neighborhood-level variables fail to explain the prevalence of force. These results

are again consistent with those of this dissertation which used only the 350

277

census tracts with 5 or more arrests. More importantly, there is once again no
difference between census tracts with 1 to 4 arrests when compared to census
tracts with 5 or more arrests with regards to the prevalence of force.

Overall, the analyses of this appendix have demonstrated that: (1)
consistent with the results of this dissertation, demographic characteristics exert
a negligible effect on both the prevalence and severity of force; (2) these
demographic characteristics may be more effective predictors of arrest behavior
than of use of force behavior; and (3) the differences between census tracts with
1 to 4 arrests and those with 5 or more arrests are negligible regarding the effect
on the prevalence and severity of force. These results serve to strengthen
conﬁdence in the results of the current dissertation. The following pages present
maps of each city with an indication of the number of arrests in each census

tract.

278

 

Charlotte Census Tract Arrests

 

Legend

N_BREAK
1:] 0
-1-4
Illls-n4

 

279

 

Colorado Springs Census Tract Arrests

 

 

 

280

 

 

 

Dallas Census Tract Arrests

 

Legend

 

281

 

 

St. Petersburg Census Tract Arrests

 

 

282

 

 

 

San Diego Census Tract Arrests

 

at

 

..
a.

a...
A
.C

 

\.. ..

._ . ﬂu...

Legend

SantlegoCknsusThuxs

BREAK

-.-.

N

-5-34

 

 

283

REFERENCES

284

 

REFERENCES

Adams, K. (1996). Measuring the prevalence of police abuse of force. In W. A.
Geller & H. Toch (Eds), Police Violence: Understanding and Controlling
Police Abuse of Force. (pp. 52-93). New Haven, CT: Yale University
Press.

Adams, K. (1997). What we know about police use of force. In National Institute
of Justice (Ed), Use of Force By Police: Overview of National and Local
Data. (pp. 1-14. Washington, DC: National Institute of Justice.

Allison, P. (2002). Saga University Papers Series on Quantitative Applications in
the Social Sciences. Vol. 07: Missing Data. (136). Thousand Oaks, CA.:
Sage.

Alpert, G. P., & Dunham, R. G. (1995). Police Use of Deadly Force: A Statistical
Analysis of the Metro-Dede Police Department. Washington, DC: Police
Executive Research Forum.

Alpert, G. P., & MacDonald, J. M. (2001). Police use of force: An analysis of
organizational characteristics. Justice Quarteriy, 18(2), 393409.

Anselin, L. (1988). Spatial Econometrics: Methods and Models. Dordrecht:
Kluwer Academics.

Anselin, L. (2003a). GeoDa 0.9 User’s Guide. Urbana-Champaign, IL: Spatial
Analysis Laboratory, University of Illinois at Urbana-Champaign.

Anselin, L. (2003b). Spatial extemalities, spatial multipliers and spatial
econometrics. International Regional Science Review, 26(2), 153-166.

Anselin, L., & Kelejian, H. H. (1997). Testing for spatial error autocorrelation in
the presence of endogenous regressors. International Regional Science
Review, 20(1-2), 153-182.

285

 

Arthur, J. A., & Case, C. E. (1994). Race, class and support for police use of
force. Crime, Law and Social Change, 21(2), 167-182.

Baller, R. D., Anselin, L., Messner, S. F., Deane, G., & Hawkins, D. F. (2001).
Structural covariates of US. county homicide rates: Incorporating spatial
effects. Criminology, 39(3), 561-590.

Baumer, E. P. (2002). Neighborhood disadvantage and police notiﬁcation by
victims of violence. Criminology, 40(3), 579-616.

Bayley, D. H. (1994). Police for the Future. New York: Oxford University Press.

Bayley, D. H., & Garafalo, J. G. (1989). The management of violence by police
patrol ofﬁcers. Criminology, 27(1), 1-25.

Bayley, D. H., & Mendelsohn, H. (1969). Minorities and the Police: Confrontation
in America. New York: Free Press.

Bell, D. (1992). Faces at the Bottom of the Well: The Permanence of Racism.
New York: Basic Books.

Bell, D. (2000). Police brutality: Portent of disaster and discomforting
convergence. In J. Nelson (Ed.), Police Brutality: An Anthology (pp. 88-
101). New York: Norton and Co.

Bittner, E. (1967). The police on skid-row: A study of peace keeping. American
Sociological Review, 32(5), 699-715.

Bittner, E. (1970). The Functions of the Police in Modern Society: A Review of
Background Factors, Current Practices, and Possible Role Models. Chevy
Chase, MD: National Institute of Mental Health.

Black, D. (1976). The Behavior of Law. New York: Academic Press.

286

Black, D. (1980). The Manners and Customs of the Police. New York: Academic
Press.

Black, D. (1989). Sociological Justice. New York: Oxford University Press.

Black, D. (1998). The Social Structure of Right and Wrong. San Diego, CA.:
Academic Press.

Blau, P. (1964). Exchange and Power in Social Life. New York, NY: John Wiley.

Braithwaite, J. (1981). The myth of social class and criminality reconsidered.
American Sociological Review, 46(Feb.), 36-57.

Brooks, L. W., Piquero, A., & Cronin, J. (1993). Police ofﬁcer attitudes concerning
their communities and their roles: A comparison of two suburban police
departments. American Joumal of Police, 12(3), 115-139.

Bryk, A. S., & Raudenbush, S. W. (1992). Hierarchical Linear Models. Newbury
Park, CA: Sage Publications.

Bursik, R. J. (1988). Social disorganization and theories of crime and
delinquency: Problems and prospects. Criminology, 26, 519-551.

Bursik, R. J., Jr., & Grasmick, H. G. (1993). Neighborhoods and Crime: The
Dimensions of Eﬁective Community Control. New York: Lexington Press.

Cao, L., Frank, J., & Cullen, F. T. (1996). Race, community context and
conﬁdence in the police. American Joumal of Police, 15(1), 3—22.

Carter, D. L. (1984). Theoretical dimensions in the abuse of authority by police
ofﬁcers. Police Studies, 7(4), 224-236.

287

 

Chambliss, W. J., & Seidman, R. B. (1971). Law, Order and Power. Reading,
MA: Addison-Wesley Publishing.

Chevigny, P. (1995). Edge of the Knife: Police Violence In the Americas. New
York: The New Press.

Cloninger, D. (1992). Intercity variations in the police use of lethal response.
Journal of Economic Behavior and Organization, 17(3), 413422.

Copeland, A. R. (1986). Police shootings: The metropolitan Dade County
experience from 1956 to 1982. American Journal of Forensic Medicine
and Pathology, 7(1), 3945.

Cordner, G. (1979). Police patrol work load studies: A review and critique. Police
Studies, 2(2), 50-60.

Coulton, C. J., Korbin, J., Chan, T., & Su, M. (2001). Mapping residents'
perceptions of neighborhood boundaries: A methodological note.
American Journal of Community Psychology, 29(2), 371-383.

Cox, S. M., & Frank, J. (1992). The inﬂuence of neighborhood context and
method of entry on individual styles of policing. American Journal of
Police, 11(2), 1-22.

Cox, S. M. (1984). Race I ethnic relations and police: Current and future issues.
American Journal of Police, 3(2), 169-183.

Crank, J. P. (1992). Legalistic and order-maintenance behavior among police
patrol ofﬁcers: A survey of eight municipal police agencies. American
Journal of Police, 12(4), 103-126.

Crawford, 0., & Burns, R. (1998). Predictors of the police use of force: The
application of a continuum perspective in Phoenix. Police Quarterly, 1(4),
41-63.

Croft, E. B. (1985). Police Use of Force: An Empirical Analysis. [Dissertation].

288

 

Cueto, 8., Ramirez, C., Leon, J., & Pain, 0. (2003). Opportunities to Leam and
Achievement in a Sample of Sixth Grade Students in Lima, Peru.
Washington, DC: Global Development Network.

Desmedt, J. (1984). Use of force paradigm for law enforcement. Journal of Police
Science and Administration, 12(2), 170-176.

Doreian, P. (1980). Linear models with spatially distributed data: Spatial
disturbances or spatial effects? Sociological Methods and Research, 9(1),
29-60.

Doreian, P. (1982). Maximum likelihood methods for linear models: Spatial effect
and spatial disturbance terms. Sociological Methods and Research, 10,
243-270.

Duncan, G. J., & Aber, J. L. (1997). Neighborhood models and measures. In J.
Brooks-Gunn, G. J. Duncan & J. L. Aber (Eds), Neighborhood Poverty:
Context and Consequences for Children. New York, NY: Sage
Publications.

Dunham, R., & Alpert, G. (1988). Neighborhood differences in attitudes toward
policing: Evidence for a mixed-strategy model of policing in a multi-ethnic
setting. Joumal of Criminal Law and Criminology, 79(2), 504-523.

Edwards, A. (2000). Reported use of force by Queensland police: Findings from
the 1999 Queensland Defendants’ Survey. Brisbane, Australia: Criminal
Justice Commission.

Engel, R. S. (2000). The effects of supervisory styles on patrol ofﬁcer behavior.
Police Quarteriy, 3(3), 262-293.

Engel, R. S., Sobol, J., & Worden, R. E. (2000). Further exploration of the
demeanor hypothesis: The interaction effects of suspects' characteristics
and demeanor on police behavior. Justice Quarteriy, 17(2), 235-258.

289

 

 

Fagan, J., & Wilkinson, D. L. (1998). Guns, youth violence, and social identity in
inner cities. In M. Tonry & M. H. Moore (Eds.), Youth Violence, Crime and
Justice: A Review of Research (pp. 105-188). Chicago, IL: University of
Chicago Press.

Friedrich, R. J. (1980). Police use of force: Individuals, situations, and
organizations. The Annals of the American Academy of Political and
Social Science, 452(November), 82-97.

Fyfe, J. J. (1982). Blind justice: Police shootings in Memphis. Journal of Criminal
Law and Criminology, 73(2), 707-722.

Fyfe, J. J. (1997). The split-second syndrome and other determinants of police
violence. In R. G. Dunham & G. P. Alpert (Eds.), Critical Issues in Policing:
Contemporary Readings. (pp. 531-546). Prospect Heights, IL: Waveland
Press.

Gaitanis, J. A. (2003). An Application of Hierarchical Generalized Linear
Modeling (HGLM) Techniques for Investigation into the Effects of Co-
Mingling Delinquent and Non-Delinquent Youth in Juvenile Justice
Prevention Programs in Florida, Florida State University.

Garner, J. H., Maxwell, C. D., & Heraux, C. G. (2002). Characteristics associated
with the prevalence and severity of force used by the police. Justice
Quarteriy, 19(4), 705-746.

Garner, J.H., Schade, T., Hepburn, J., & Buchanan, J. (1995). Measuring the
continuum of force used by and against the police. Criminal Justice
Review, 20(2), 146-168.

Geis, G., & Binder, A. (1990). The future of non-lethal weapons. Joumal of
Contemporary Criminal Justice, 6(1), 143.

Geller, W. A., & Karales, K. (1981). Split-Second Decisions: Shootings of and by
the Chicago Police. Chicago, IL: Chicago Law Enforcement Study Group.

290

Geller, W. A., & Scott, M. (1992). Deadly Force: What We Know - A Practitioner’s
Desk Reference on Police-Involved Shootings in the United States.
Washington, DC: Police Executive Research Forum.

Goldberg, D. T. (1993). Racist Culture: Philosophy and the Politics of Meaning.
Cambridge, MA: Blackwell Publishers.

Goldkamp, J. S. (1982). Minorities as victims of police shootings: Interpretations
of racial disproportionality and police use of deadly force. In J. J. Fyfe
(Ed.), Readings on Police Use of Deadly Force. (pp. 128-151).
Washington, DC: Police Foundation.

Goldstein, H. (1990). Problem-Oriented Policing. New York, NY: McGraw-Hill,
Inc.

Gould, P., & White, R. (1974). Mental Maps. Middlesex, England: Penguin
Books.

Greenberg, D. F., Kessler, R. C., & Loftin, C. (1985). Social inequality and crime
control. The Journal of Criminal Law and Criminology, 76(3), 684-704.

Greenfeld, L. A., Langan, P. A., Smith, S. K., & Kaminski, R. J. (1997). Police
Use of Force: Collection of National Data. Washington, DC: Bureau of
Justice Statistics.

Greenleaf, R. G., & Lanza-Kaduce, L. (1995). Sophistication, organization, and
authority-subject conﬂict: Rediscovering and unraveling Turk's theory. of
norm resistance. Criminology. 33(4), 565-580.

Haney, W., & Knowles, E. (1978). Perceptions of neighborhoods by city and
suburban residents. Human Ecology, 6(2), 201-214.

Harring, S. L., & Ray, G. W. (1999). Policing a class society: New York City in the
1990s. Social Justice, 26(2), 63-81.

291

 

Hawley, F. F., & Messner, S. F. (1989). The southern violence construct: A
review of arguments, evidence, and the normative context. Justice
Quarteriy, 6(4), 481-511.

Herbert, S. (1996). The geopolitics of the police: Foucault, disciplinary power and
the tactics of the Los Angeles Police Department. Political Geography,
15(1), 47-57.

Herbert, S. (1998). Police subculture reconsidered. Criminology, 36(2), 343-369.

Herz, D. C. (2001). Improving police encounters with juveniles: Does training
make a difference? Justice Research and Policy, 3(2), 57-77.

Hickman, M. J., Piquero, A., & Greene, J. (2000). Does community policing
generate greater numbers and different types of citizen complaints than
traditional policing? Police Quarterly, 3(1), 70-84.

Holmes, M. (2000). Minority threat and police brutality: Determinants of civil
rights criminal complaints in US. municipalities. Criminology, 33, 343-368.

Holmes, S., Reynolds, K., Holmes, R., & Faulkner, S. (1998). Individual and
situational determinants of police force: An examination of threat
presentation. American Journal of Criminal Justice, 23(1), 83-106.

lnfoshare. (2004). Demographic Data for the City of New York. Retrieved from
www.infoshare.org.

lntemational Association of Chiefs of Police. (2001). Police Use of Force in
America 2001. Alexandria, VA: lntemational Association of Chiefs of
Police.

Jackson, P. l. (1989). Minority Group Threat, Crime and Policing: Social Context
and Social Control. New York: Praegar Publishing.

292

Jacobs, D., & Helms, R. (1997). Testing coercive explanations for order: The
determinants of law enforcement strength over time. Social Forces, 75(4),
1361-1392.

Jacons, D. & O’Brien, RM. (1998). The determinants of deadly force: A structural
analysis of police violence. American Journal of Sociology, 103(4), 837-
862.

Jang, S. J., & Johnson, B. R. (2001). Neighborhood disorder, individual
religiousity, and adolescent use of illicit drugs: A test of multilevel
hypotheses. Criminology. 39(1), 109-143.

Jones, M. (1996). Indicator and stratiﬁcation methods for missing explanatory
variables in multiple linear regression. Journal of the American Statistical
Association, 91, 222-230.

Kaminski, R. J., Edwards, S. M., & Johnson, J. W. (1999). Assessing the
incapacitative effects of pepper spray during resistive encounters with the
police. Policing: An lntemational Joumal of Police Strategies and
Management, 22(1), 7-29.

Kaminski, R. J., & Martin, J. A. (2000). An analysis of police ofﬁcer satisfaction
with defense and control tactics. Policing: An lntemational Journal of
Police Strategies and Management, 23(2), 132-153.

Kaminski, R., & Jefferis, E. (1998). The effect of a violent televised arrest on
public perceptions of the police: A partial test of Easton's theoretical
framework. Policing: An lntemational Journal of Police Strategies and
Management, 21(4), 683-706.

Kane, R. J. (2002). The social ecology of police misconduct. Criminology, 40(4),
867-896.

Kappeler, V. K., Sluder, R. D., & Alpert, G. P. (1998). Forces of Deviance:
Understanding the Dark Side of Policing. Prospect Heights, IL: Waveland
Press.

293

 

 

Kavanagh, J. (1997). The occurrence of resisting arrest encounters: A study of
police-citizen violence. Criminal Justice Review, 22(1), 16-33.

Kelling, G. L., & Stewart, J. K. (1990). Neighborhoods and police: The
maintenance of civil authority. Criminal Law Forum: An lntemational
Joumal, 1(3), 459476.

Klinger, D. A. (1994). Demeanor or crime? Why 'hostile' citizens are more likely
to be arrested. Criminology, 32(3), 475493.

Klinger, D. A. (1995). The micro-structure of nonlethal force: Baseline data from
an observational study. Criminal Justice Review, 20(2), 169-186.

Klinger, D. A. (1997). Negotiating order in patrol work: An ecological theory of
police response to deviance. Criminology, 35(2), 277-306.

Kohfeld, C. W., & Sprague, J. (1990). Demography, police behavior and
deterrence. Criminology, 28(1), 111-136.

Land, K. C. (2000). The Inﬂuence of Neighborhood, Pear, and Family Context:
Trajectories of Delinquent / Criminal Offending Across the Life Course.
Washington, DC: US. Department of Justice.

Langan, P. A., Greenfeld, L. A., Smith, S. K., Durose, M. R., & Levin, D. J.
(2001 ). Contacts between Police and the Public: Findings from the 1999
National Survey. Washington, DC: Bureau of Justice Statistics.

Langford, l., Leyland, A., Rasbash, J., & Goldstein, H. (1999). Multilevel modeling
of the geographical distributions of diseases. Applied Statistics, 48(2),
253-268.

Lanza-Kaduce, L., & Greenleaf, R. G. (2000). Age and race deference reversals:
Extending Turk on police-citizen conﬂict. Journal of Research in Crime
and Delinquency, 37(2), 221-236.

294

Lasley, J. (1994). The impact of the Rodney King incident on citizen attitudes
toward police. Policing and Society, 3(4), 225-245.

Lawton, B. A., Hickman, M. J., Piquero, A. R., & Greene, J. R. (2001). Using GIS
to analyze complaints against police: A research note. Justice Research
and Policy, 3(2), 95-108.

Leventhal, T., & Brooks-Gunn, J. (2000). The neighborhoods they live in: The
effects of neighborhood residence on child and adolescent outcomes.
Psychological Bulletin, 126(2), 309-337.

Liska, A. E., & Chamlin, M. B. (1984). Social structure and crime control among
macrosocial units. American Journal of Sociology, 90(2), 383—395.

Little, R., & Rubin, D. (1987). Statistical Analysis with Missing Data. New York,
NY: John Wiley and Sons.

Lizotte, A. J., Thornberry, T. P., Krohn, M. D., Chard-\Merschem, D., &
McDowalI, D. (1993). Neighborhood context and delinquency: A
longitudinal analysis. In H. Kemer & E. Weitkamp (Eds.), Cross-National
Longitudinal Research on Human Development and Criminal Behavior.
(pp. 217-227). Dodrecht: Kluwer Academic Press.

Locke, H. G. (1996). The color of law and the issue of color: Race and the abuse
of police power. In W. A. Geller & H. Toch (Eds.), Police Violence:
Understanding and Controlling Police Abuse of Force. (pp. 129-149). New
Haven, CT: Yale University Press.

Luckenbill, D. F., & Doyle, D. P. (1989). Structural position and violence:
Developing a cultural explanation. Criminology, 27(3), 419436.

Lundman, R. J. (1994). Demeanor or crime? The Midwest city police-citizen
encounters study. Criminology, 32(4), 631-656.

Magalhaes, A., Hewings, G.J.D., & Azzoni, CR. (2000). Spatial Dependence
and Regional Convergence in Brazil. Urbana-Champaign, IL: Regional
Economics Applications Laboratory, University of Illinois at Champaign.

295

Mann, C. R. (1993). Unequal Justice: A Question of Color. Bloomington, IN:
Indiana University Press.

Manning, P. K. (1997). Police Work: The Social Organization of Policing.
Prosepct Heights, IL: Waveland Press, Inc.

Mastrofski, S. D. (1995). The police. In J. Sheley (Ed.), Criminology. Belmont,
CA: Wadsworth Publishing.

Mastrofski, S. 0., Parks, R.B., Dejong, C., & Worden, RE. (1998). Race and
Every-Day Policing: A Research Perspective. East Lansing, MI: Michigan
State University.

Mastrofski, S. D., Reisig, M. D., & McCluskey, J. D. (2002). Police disrespect
toward the public: An encounter-based analysis. Criminology, 40(3), 519-
550.

Mastrofski, S. D., Ritti, R. R., & Hoffmaster, D. (1987). Organizational
determinants of police discretion: The case of drinking-driving. Journal of
Criminal Justice, 15(5), 387402.

Mastrofski, S. D., Worden, R. E., & Snipes, J. B. (1995). Law enforcement in a
time of community policing. Criminology, 33(4), 539-563.

Matulia, K. (1985). A Balance of Forces: Model Deadly Force Policy and
Procedure. Gaithersburg, MD: lntemational Association of Chiefs of
Police.

McCluskey, J.D., & Terrill, W. (2005). Departmental and citizen complaints as
predictors of police coercion. Policing: An lntemational Journal of Police
Strategies and Management, 28(3), 513-529.

McCluskey, J.D., Terrill, W., & Paoline III, E. A. (2005). Peer group
aggressiveness and the use of coercion in police-suspect encounters.
Police Practice and Research, 6(1), 19-37.

296

McGahey, R. M. (1986). Economic conditions, neighborhood organization, and
urban crime. In A. J. Reiss & M. Tonry (Eds.), Communities and Crime
(pp. 231-270). Chicago, IL: University of Chicago Press.

McGarrell, E. F., Giacomazzi, A. L., 8. Thurman, Q. C. (1997). Neighborhood
disorder, integration, and the fear of crime. Justice Quarterly, 14(3), 479-
500.

McLanahan, 8., & Garﬁnkel, l. (1993). Single mothers, the underclass and social
policy. In W. J. Wilson (Ed.), The Ghetto Underclass: Social Science
Perspectives. Newbury Park, CA: Sage Publications.

McNamara, J. H. (1967). Uncertainties in police work: The relevance of police
recruits' background and training. In D. Bordua (Ed.), The Police: Six
Sociological Essays. (pp. 163-252). New York, NY: John Wiley.

McNeely, R., & Pope, C. E. (1981). Race, Crime and Criminal Justice. Beverly
Hills, CA: Sage Publishing.

Meares, T. L., & Kahan, D. M. (1998). Law and (norms of) order in the inner city.
Law and Society Review, 32(4), 805-838.

Meehan, A. J., & Ponder, M. C. (2002). Race and place: The ecology of racial
proﬁling African-American motorists. Justice Quarteriy, 19(3), 399430.

Messner, S. F., Anselin, L., Baller, R. D., Hawkins, D., Deane, G., & Tolnay, S.
(1999). The spatial patterning of county homicide rates: An application of

exploratory spatial data analysis. Joumal of Quantitative Criminology,
15(4), 423450.

Messner, S. F., 8 Tardiff, K. (1986). Economic inequality and levels of homicide:
An analysis of urban neighborhoods. Criminology, 24(2), 297-317.

Milton, C., Halleck, J. W., Lardner, J., & Abrecht, G. L. (1977). Police Use of
Deadly Force. Washington, DC: Police Foundation.

297

 

 

Morenoff, J. D., Sampson, R. J., & Raudenbush, S. W. (2001). Neighborhood
inequality, collective efﬁcacy, and the spatial dynamics or urban violence.
Criminology, 39(3), 517-560.

Morenoff, J., & Sampson, R. J. (1997). Violent crime and the spatial dynamics of
neighborhood transition: Chicago 1970-1990. Social Forces, 76(1), 31-64.

Muir, W. k. (1977). Police: Streetcomer Politicians. Chicago: University of
Chicago Press.

Ostrom, E., Parks, R., & Whitaker, G. (1977). The Police Services Study.
Bloomington, IN: Workshop on Political Theory and Policy Analysis at
Indiana University.

Ouimet, M. (2000). Aggregation bias in ecological research: How social
disorganization and criminal opportunities shape the spatial distribution of
juvenile delinquency in Montreal. Canadian Journal of Criminololgy, 42(2),
135-156.

Park, R., & Burgess, E. (1924). Introduction to the Science of Sociology.
Chicago, IL: Chicago University Press.

Paoline III, E. A. (2003). Taking stock: Toward a richer understanding of police
culture. Journal of Criminal Justice, 31, 199-214.

Paoline III, E. A. (2004). Shedding light on police culture: An examination of
ofﬁcers” occupational attitudes. Police Quarterly, 7(2), 205-236.

Perez, D. (1994). Common Sense about Police Review. Philadelphia, PA:
Temple University Press.

Perkins, D. D., Meeks, J. W., & Taylor, R. B. (1992). The physical environment of
street blocks and resident perceptions of crime and disorder: Implications

for theory and measurement. Joumal of Environmental Psychology, 12,
21-34.

298

 

Peterson, R. D., Krivo, L. J., & Harris, M. A. (2000). Disadvantage and
neighborhood violent crime: Do local institutions matter? Journal of
Research in Crime and Delinquency, 37(1), 31 -63.

Phillips, T., & Smith, P. (2000). Police violence occasioning citizen complaint: An
empirical analysis of time-space dynamics. British Joumal of Criminology,
40(3), 480496.

Ratcliffe, J., & McCullagh, M. (2001). Chasing ghosts: Police perception of high
crime areas. British Journal of Criminology, 41(2), 330-341.

Reasons, C. E., & Kuykendall, J. L. (1972). Race, Crime and Justice. Paciﬁc
Palisades, CA: Goodyear Publishing Co.

Reiman, J. (2001). The Rich Get Richer and the Poor Get Prison: Ideology,
Class, and Criminal Justice. Boston, MA: Allyn and Bacon.

Reisig, M. D., McCluskey, J. D., Mastrofski, S. D., & Terrill, W. (2004). Suspect
disrespect toward the police. Justice Quarteriy, 21 (2), 241-268.

Reisig, M. D., & Parks, R. B. (2000). Experience, quality of life, and
neighborhood context: A hierarchical analysis of citizen satisfaction with
police. Justice Quarteriy, 17(3), 607—630.

Reisig, M.D., & Parks, RB. (2003). Neighborhood context, police behavior and
satisfaction with the police. Justice Research and Policy, 5(1), 37-65.

Reiss, A. J. (1971). The Police and the Public. New Haven, CT: Yale University
Press.

Reiss, A. J. (1986). Why Are Communities Important in Understanding Crime? In
A. J. Reiss & M. J. Tonry (Eds.), Communities and Crime. (pp. 1-33).
Chicago: University of Chicago Press.

Riksheim, E. C., & Cherrnak, S. M. (1993). Causes of police behavior revisited.
Journal of Criminal Justice, 21(4), 353-382.

299

Rossi, P. H. (1968). Between White and Black: The Faces of American
Institutions in the Ghetto. New York: Praegar Publishing.

Sampson, R. J. (1984). Group size, heterogeneity, and intergroup conﬂict: A test
of Blau's inequality and heterogeneity. Social Forces, 62(3), 618-639.

Sampson, R. J. (1986). Effects of socioeconomic context on ofﬁcial reaction to
juvenile delinquency. American Sociological Review, 51(6), 876-885.

Sampson, R. J., & Bartusch, D. J. (1998). Legal cynicism and (subcultural?)
tolerance of deviance: The neighborhood context to racial differences.
Law and Society Review, 32(4), 777-804.

Sampson, R. J., & Groves, W. B. (1989). Community structure and crime:
Testing social disorganization theory. American Journal of Sociology,
94(4), 774-802.

Sampson, R. J., & Laub, J. H. (1993). Structural variations in juvenile court
processing: Inequality, the underclass, and social control. Law and Society
Review, 27(2), 285-311.

Sampson, R. J., Morenoff, J. D., & Gannon-Rowley, T. (2002). Assessing
'neighborhood effects': Social processes and new directions in research.
Annual Review of Sociology, 28, 443-478.

Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighborhoods and
violent crime: A multilevel study of collective efﬁcacy. Science, 277, 918-
924.

Sampson, R. J., & Woolredge, J. D. (1987). Linking the micro- and macro-level
dimensions of lifestyle-routine activity and opportunity models of predatory
victimization. Joumal of Quantitative Criminology, 3(4), 371-394.

Sampson, R. J., & Raudenbush, S. W. (1999). Systematic social observation of
public spaces: A new look at disorder in urban neighborhoods. American
Journal of Sociology, 105(3), 603-651.

300

Scrivner, E. M. (1994). The Role of Police Psychology in Controlling Excessive
Force. Washington, DC: National Institute of Justice.

Seron, C., & Munger, F. (1996). Law and inequality: Race, gender..and, of
course, class. Annual Review of Sociology, 22, 187-212.

Shaw, C. R., & McKay, H. D. (1942). Juvenile Delinquency and Urban Areas.
Chicago: University of Chicago Press.

Sherman, L. W., & Blumberg, M. (1981). Higher education and police use of
deadly force. Joumal of Criminal Justice, 9(4), 317-331.

Shihadeh, E. S., & Steffensmeier, D. J. (1994). Economic inequality, family
disruption, and urban black violence: Cities as units of stratiﬁcation and
social control. Social Forces, 73(2), 729-751.

Simpson, M. E. (1985). Violent crime, income inequality, and regional culture:
Another look. Sociological Focus, 18(3), 199-208.

Skolnick, J. H., & Fyfe, J. J. (1993). Above the Law: Police and Excessive Use of
Force. New York: The Free Press.

Smith, B. W., & Frank, J. (2000). Explaining Police Activities across Urban
Neighborhoods. Washington, DC: US. Department of Justice.

Smith, D. A. (1986). The neighborhood context of police behavior. In A.J. Reiss
and M. Tonry (Ed.), Communities and Crime (pp. 313-341). Chicago: The
University of Chicago Press.

Smith, D. A. (1987). Police response to interpersonal violence: Deﬁning
parameters of legal control. Social Forces, 65(3), 767-782.

Smith, D. A., & Jarjoura, G. R. (1989). Household characteristics, neighborhood
composition and victimization risk. Social Forces, 68(2), 621-640.

301

 

Smith, D. A., Visher, C.A., & Davidson, LA. (1984). Equity and discretionary
justice: The inﬂuence of race on police arrest decisions. Journal of
Criminal Law and Criminology, 75(1), 234-249.

Son, I. S., Tsang, C.-W., Rome, 0., & Davis, M. (1997). Citizens' observations of
police use of excessive force and their evaluation of police performance.
Policing, 20, 149-159.

Sorensen, J. R., Marquart, J. W., & Brock, D. E. (1993). A comparison of factors
related to the killing of felons by police and citizens. Joumal of Police and
Criminal Psychology, 9(1), 20-33.

Sparger, J. R., & Giacopassi, D. J. (1986). Police resentment of the upper class.
Criminal Justice Review, 1 1(1), 25-33.

Stark, R. (1987). Deviant places: A theory of the ecology of crime. Criminology,
25(4), 893-909.

Stewart, E. A., Simons, R. L., & Conger, R. D. (2002). Assessing neighborhood
and social psychological inﬂuences on childhood violence in an African-
American sample. Criminology, 40(4), 801-830.

Suttles, G. D. (1972). The Social Construction of Communities. Chicago, IL:
University of Chicago Press.

Swett, D. (1972). Cultural bias in the American legal system. In C. Reasons & J.
Kuykendall (Eds.), Race, Crime and Justice. Paciﬁc Palisades, CA:
Goodyear.

Taylor, R. B., & Covington, J. (1988). Neighborhood changes in ecology and
violence. Criminology, 26(4), 553-589.

Taylor, R. B., Gottfredson, S. D., & Brower, S. (1984). Block crime and fear:
Defensible space, local social ties, and territorial functioning. Journal of
Research in Crime and Delinquency, 21 (4), 303-331.

302

 

Terrill, W. (2005). Police use of force: A transactional approach. Justice
Quarteriy, 22(1), 107-138.

Terrill, W., & Mastrofski, S. D. (2002). Situational and ofﬁcer-based determinants
of police coercion. Justice Quarteriy, 19(2), 215-248.

Terrill, W., & Mastrofski, S. D. (2003). Working the street: Does community
policing matter? In W. Skogan (Ed) Community Policing: Can It Work?
(pp. 109-135). Belmont, CA: Wadsworth Publishing.

Terrill, W., & Reisig, M. D. (2003). Neighborhood context and police use of force.
Journal of Research in Crime and Delinquency, 40(3), 291-321.

Toch, H. (1995). Violent Men: An Inquiry into the Psychology of Violence.
Washington, DC: American Psychological Association.

Tuch, S., & Weitzer, R. (1997). Racial differences in attitudes toward the police.
Public Opinion Quarteriy, 61, 642-663.

Van Maanen, J. (1973). Observations on the making of policemen. Human
Organization, 32(4), 407418.

Velez, M. B. (2001). The role of public social control in urban neighborhoods: A
multilevel analysis of victimization risk. Criminology, 39(4), 837-864.

Vila, B. J., & Morrison, G. B. (1994). Biological limits to police combat handgun
shooting accuracy. American Journal of Police, 13(1), 1-30.

Waddington, P. (1999). Police (canteen) sub-culture: An appreciation.
Criminology, 39(2), 287-309.

Wade, P. (1993). Blackness and Race Mixture: The Dynamics of Racial Identity
in Colombia. Baltimore, MD: Johns Hopkins University Press.

303

 

Waegel, W. R. (1984). How police justify the use of deadly force. Social
Problems, 32(2), 144-155.

Walker, 8., & Graham, N. (1998). Citizen complaints in response to police
misconduct: The results of a victimization survey. Police Quarteriy, 1(1),
65-89.

Walker, 3., Spohn, C. A., & DeLone, M. (1996). The Color of Justice: Race,
Ethnicity and Crime in America. Belmont, CA: Wadsworth Publishing.

Walklate, S., & Evans, K. (1999). Zero tolerance or community tolerance? Police
and community talk about crime in high-crime areas. Crime Prevention
and Community Safety: An lntemational Joumal, 1(1), 11-24.

Warner, B. D., & Rountree, P. W. (1997). Local social ties in a community and
crime model. Social Problems, 44(4), 520-536.

Weidner, R. R., & Terrill, W. (2005). Atest of Turk’s theory of norm resistance
using observational data on police-suspect encounters. Journal of
Research in Crime and Delinquency, 42(1), 84-109.

Weisburd, 0., Greenspan, R., Hamilton, E. E., Williams, H., & Bryant, K. A.
(2 000). Police Attitudes toward Abuse of Authority: Findings from a
National Study. Washington, DC: US. Department of Justice.

Weitzer, R. (1999). Citizens' perceptions of police misconduct: Race and
neighborhood context. Justice Quarteriy, 16(4), 819-846.

Weitzer, R. (2000). Racialized policing: Residents' perceptions in three
neighborhoods. Law and Society Review, 34(1), 129-155.

Werthman, C., & Piliavin, I. (1967). Gang members and the police. In D. Bordua
(Ed), The Police: Six Sociological Essays (pp. 59-68). New York: Wiley
Publishing.

Whitaker, G. P. (1982). What is patrol work? Police Studies, 4(4), 13-22.

304

 

White, R. (1994). Street life: Police practices and youth behavior. In R. White &
C. Alder (Eds.), The Police and Young People in Australia. Cambridge:
Cambridge University Press.

Wikstrom, P.-O. H., & Loeber, R. (2000). Do disadvantaged neighborhoods
cause well-adjusted children to become adolescent delinquents? A study
of male juvenile serious offending, individual risk and protective factors,
and neighborhood context. Criminology, 38(4), 1109-1142.

Wilkinson, D. L., & Fagan, J. (2000). What Do We Know about Gun Use among
Adolescents? Boulder, CO: Center for the Study and Prevention of
Violence.

Wilson, J. O. (1968). Varieties of Police Behavior: The Management of Law and
Order in Eight Communities. Cambridge, MA: Harvard University Press.

\MIson, W. J. (1980). The Declining Signiﬁcance of Race: Blacks and Changing
American Institutions. Chicago: University of Chicago Press.

Wilson, W. J. (1987). The Truly Disadvantaged: The Inner City, the Underclass
and Public Policy. Chicago: University of Chicago Press.

Wilson, W. J. (1993). The Ghetto Underclass: Social Science Perspectives.
Newbury Park, CA: Sage Publications.

Wolfgang, M. (1978). The sociology of aggression - crime and violence. The
Australian Journal of Forensic Sciences, 11(1), 3-32.

Worden, R. E. (1995). The causes of police brutality: Theory and evidence on
police use of force. In W. A. Geller & H. Toch (Eds.), And Justice for All:
Understanding and Controlling Police Use of Force. Washington, DC:
Police Executive Research Forum.

Worden, R. E., & Shepard, R. L. (1996). Demeanor, crime, and police behavior:
A reexamination of the police services study data. Criminology, 34(1), 83-
105.

305

 

Worden, R. E., Shepard, R. L., & Mastrofski, S. D. (1996). On the meaning and
measurement of suspects' demeanor toward the police: A comment on
'Demeanor and Arrest.'. Joumal of Research in Crime and Delinquency,
33(3), 324-332.

306

It‘t‘ i

   

lilliiiljjilijllilllljjljl