at .h...“ 5
yyawwflrahfim
lift. «n 5.34..
c 3 17
.rm‘“ . ..
inn". ’1}... I

hqa‘
. :1

 

\

, aw.

 

 

' IBRARY
Michigan State
University

 

 

 

This is to certify that the
dissertation entitled

THE RELATIONSHIP BETWEEN LIFESTYLE. DEMOGRAPHIC,
SITUATIONAL. AND STRUCTURAL FEATURES WITH HOMICIDE
IN INDIANAPOLIS

presented by

Nicholas A. Corsaro

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Criminal Justice

 

 

fl/W I

, ( Major Professor’s Signature

December I]; 5200 7

Date

MSU is an affinnative-action, equal—opportunity employer

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

MAR 0 6 2009

A1 nun

n
IQUVU

 

 

 

OCT 3 1 ZUUH
'13-: 1 A 9% ‘\

 

 

WA) 9 $312281 .

, . . i, .I it
Q} U I} . f. 1‘

 

Mil-X 9 7‘ 2.011

 

 

 

 

 

 

 

 

 

6/07 p:/ClRC/DateDue.indd-p.1

THE RELATIONSHIP BETWEEN LIFESTYLE, DEMOGRAPHIC, SITUATIONAL,
AND STRUCTURAL FEATURES WITH HOMICIDE IN INDIANAPOLIS
By

Nicholas A. Corsaro

A DISSERTATION
Submitted to
Michigan State University

in partial fillfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
School of Criminal Justice

2007

ABSTRACT

THE RELATIONSHIP BETWEEN LIFESTYLE, DEMOGRAPHIC, SITUATIONAL,
AND STRUCTURAL FEATURES WITH HOMICIDE IN INDIANAPOLIS

By

Nicholas A. Corsaro

This study investigates the relationship between incident and structural
characteristics and gang homicide. Incident level characteristics include situational
features of the event, and the demographic and lifestyle attributes of the homicide
participants. Prior research has shown that gang homicides have tended to be firearm
related and drug motivated where multiple, younger, non-white and male actors are more
inclined to be involved. The association between many of these incident level
characteristics and gang homicide has been established in prior research that relied upon
data from chronic gang cities, which are cities that have an extensive and historic gang
problem. The generalizability of these correlates with gang homicide in emerging gang
cities, which are cities that have a developing gang problem, is still largely unverified.
Thus, the relationship between previously established incident level correlates of gang
violence and gang homicide in an emerging gang city, Indianapolis, is examined here.
Relying on binary logistic regression models, results show that correlates of gang
homicide from chronic gang cities also delineated homicide in an emerging gang city.
Implications support the generalizability of gang homicide research over time and across
cities.

In addition, this study contributes to the ecological literature that examines the

relationship between gang homicide and neighborhood context. In particular, two

structural level measures were incorporated into hierarchical generalized linear models
(HGLM): economic deprivation and disorder. Prior research has shown a mixed
relationship between economic deprivation and gang homicide. The disorder-gang
homicide relationship, to this point, has gone untested in statistical models. The current
research attempted to delineate gang homicide fiom non-gang homicide at the
neighborhood level while simultaneously controlling for incident level measures (i.e.,
level-l variables). Results suggested that the relationship between gang homicide and
economic deprivation was somewhat fragile and unstable, while the association between
gang homicide and disorder was much more robust, significant, and stable. These
findings have both theoretical and policy implications that contribute to the gang

homicide and ecological literature.

COPyright by
Nicholas A. Corsaro

2007

ACKNOWLEDGEMENTS

To my family: Mom, Dad, Tony, Kim, AJ, Gabe, and Bradyl simply want to let
you all know, completing the degree was never the carrot at the end of the stick. It was
going home to see you. ’

To my committee: I want to offer my gratitude toward you as a collective unit.
When I came to MSU, I knew this was the group I wanted to guide me through the
program. Thank you for allowing me to choose the program I wanted to pursue. I
couldn’t be happier with the way it has worked out. Individually (and in noiparticular
order), I want to thank Natalie Hipple for being a part of this dissertation. We’ve been
working together for a long time and I hope that you feel a sense of satisfaction as I wrap
up my time here, as you’ve been instrumental in my educational process. To Tim
Bynum, I have learned a tremendous amount from you over the years and appreciate you
taking the time out of your hectic schedule to help guide me through this process. To Bill
Davidson, your insights throughout this dissertation have been really appreciated. And I
want you to know that having the same person who guided Bill Selke all those years ago
also serve on this committee means the world to me. To Chris Maxwell, thank you for all
of your help and guidance throughout my selection of coursework, comprehensive
exams, and ultimately the dissertation process. You have been an influential member of
my committee and have helped shape me into a better researcher.

Specifically to Ed McGarrell, it is hard to put into words what working for you all
these years has meant to me. In many ways it has been a perfect fit. I cannot think of
anyone who has been better to me than you have in terms of allowing, and providing, me

the resources (all different types) I needed to accomplish my goals. But it is much more

than that. It has been the evolution of a friendship that I feel will always bethere. You
have been a great mentor, and it is something I will never forget. I have and will always
consider you a model from which to work.

To those who have impacted my life while here, it is difficult to leave, though this
is certainly not goodbye. To my fi'iends on the bottom floor (again in no particular order)
J inesong Cheong, Jason Ingram, Ryan Martz, Eric Grommon, Jeff Gruenewald, Penny
Fischer, Dae-Hoon Kwak, Alexis Non'is, as well as my friends on the top floor, Mary Lee
Vandermore, Steve Chermak, Carol Zimmerman, Carole Gibbs, J esy Pizzaro, Chris
Smith, and lets not forget Bootsy, and Kato, I want to let each of you know that
sometimes I came to Baker just to see you. I guess I am just trying to say thank you for
coming along for the ride with me. And to my buddy, Dan King, I can honestly say that
you have been more like a brother than a friend, and I feel fortunate you were around.

Finally, I want to let Bill Selke know that I am honored to have followed in his
path--fiom Connersville, to Bloomington, to East Lansing, to Illinois. I only hope and

pray I can have the same impact in life that you had on mine.

vi

TABLE OF CONTENTS

LIST OF TABLES ............................................................................................................. ix
LIST OF FIGURES .......................................................................................................... xii
Chapter 1: Introduction ....................................................................................................... 1
Chapter 2: Review of Gang Research ................................................................................. 6
Demographic Characteristics of Actors in Gang Homicide ........................................... 7
Situational Correlates of Gang Homicide ..................................................................... 11
Lifestyle Measures of Gang Homicide ......................................................................... 13
Recent Advancements in Gang Homicide Research .................................................... 16
Chapter Summary ......................................................................................................... 19
Chapter 3: Macro-Social Theories of Crime ..................................................................... 21
Cone of Resolution ....................................................................................................... 22
Level 4: Classical Ecological Theories ..................................................................... 24
Level 3: Contemporary Ecological Theories ............................................................ 26
Level 2: Ecological Theories Specific to Violence and Homicide ........................... 40
Level 1: Ecological Theories Specific to Gang Homicide ........................................ 44
Chapter Summary ......................................................................................................... 50
Chapter 4: Description of the Current Study .................................................................... 54
Research Objective #1 .................................................................................................. 54
Research Objective #2 .................................................................................................. 57
Research Questions ....................................................................................................... 60
Research Hypotheses .................................................................................................... 60
Sources of Data ............................................................................................................. 61
IVRP Data ................................................................................................................. 61
Project on Policing (POPN) Survey Data Description ............................................. 64
US. Census Data ...................................................................................................... 65
Chapter Summary ....................................................................................... - .................. 68
Chapter 5: Data Description .............................................................................................. 70
Dependent Variable ...................................................................................................... 70
Independent Variables .................................................................................................. 72
Situational Measures ................................................................................................. 73
Demographic Measures ............................................................................................ 77
Lifestyle Measures .................................................................................................... 81
Structural Measures .................................................................................................. 83
Analytic Strategy .................................................................... 87
Research Question #1 ............................................................................................... 88
Research Question #2 ............................................................................................... 92
Chapter Summary ......................................................................................................... 96

vii

Chapter 6: Results ............................................................................................................. 98

Research Question #1 ................................................................................................... 98
Univariate Results ..................................................................................................... 98
Bivariate Results ....................................................................................................... 99
Multivariate Analyses ............................................................................................. 103
Predicted Probabilities ............................................................................................ 109
Diagnostic Tests ...................................................................................................... 112
Summary of Level-1 Analyses ................................................................................ 115

Research Question #2 ................................................................................................. 116
Univariate Results ................................................................................................... 1 17
Bivariate Results ..................................................................................... p ................ 1 18
Multivariate Results ................................................................................................ 119
Predicted Probabilities ............................................................................................ 129
Diagnostic Tests ...................................................................................................... 130
Summary of Level-2 Analyses ................................................................................ 134

Chapter 7: Discussion ..................................................................................................... 137

Overall Summary of Results ....................................................................................... 137

Limitations .................................................................................................................. 141

Implications for Gang Research .................................................................................. 148

Implications for Ecological Criminology ................................................................... 152
Economic Deprivation ............................................................................................ 152
Disorder ................................................................................................................... 155

Policy Implications ..................................................................................................... 159
Actor Level Initiatives ............................................................................................ 159
Neighborhood Level Initiatives .............................................................................. 160

Conclusion .................................................................................................................. 162

Appendix A ..................................................................................................................... 167
Appendix B ..................................................................................................................... 170
Appendix C ..................................................................................................................... 173
Appendix D ..................................................................................................................... 181
Appendix E ..................................................................................................................... 186
References ....................................................................................................................... 1 8 8

viii

LIST OF TABLES

Table 1: Actor distribution across gang and non-gang homicides in Indianapolis ........... 71
Table 2: Total number of homicides in Indianapolis classified by type ........................... 72
Table 3: Distribution of multiple suspects involved in incidents across homicide subtype7s3
Table 4: Distribution of the incident motive across homicide subtypes ........................... 74
Table 5: Distribution of the drug-related incidents across homicide subtypes ................. 75
Table 6: Distribution of the method of death across homicide subtypes ........ A. .................. 76
Table 7: Distribution of firearm related offenses across homicide subtypes .................... 76
Table 8: Descriptive demographic measures across homicide subtypes .......................... 78
Table 9: Descriptive demographic measures of actors across homicide subtypes ........... 81

Table 10: Descriptive statistics for actors' lifestyle measures of prior drug and weapon

arrests across homicide subtypes .............................................................................. 83
Table 11: Factor analysis for concentrated disadvantage ................................................. 85
Table 12: Survey items used to create disorder measure .................................................. 86
Table 13: Factor analysis for disorder ............................................................ ’ .................. 87
Table 14: Descriptive statistics for variables used in incident level analysis ................... 99

Table 15: Pearson's r correlations between outcome and independent variables included

in analysis ................................................................................................................ 100
Table 16: T-test results for independent variables for homicide subtypes ..................... 101
Table 17: Summary of bivariate analyses ....................................................................... 102

Table 18: Unstandardized coefficients (Beta) and Odds Ratios (OR) for victim-only
logistic regression models ....................................................................................... 105

Table 19: Unstandardized coefficients (Beta) and Odds Ratios (OR) for joint-actor
logistic regression models ....................................................................................... 108

ix

Table 20: Predicted probabilities of gang homicides based on situational and
demographic measures for victim-only model ....................................................... 110

Table 21: Predicted probabilities of gang homicides based on situational and
demographic measures for joint-actor model ......................................................... 112

Table 22: Descriptive statistics for Variables used in structural-level analysis .............. 118

Table 23: Pearson's r correlations between outcome and independent variables included

in analysis ................................................................................................................ 119
Table 24: HGLM conditional fixed effects models ........................................................ 121
Table 25: HGLM mixed model with victim-only data ................................................... 123
Table 26: Descriptive statistics for the number of homicides that occurred in police beats

for the joint-actor model ......................................................................................... 126
Table 27: Mixed logistic regression model for joint-actor data ...................................... 127

Table 28: Predicted probabilities of gang homicides based on changes in neighborhood
disorder ................................................................................................................... 130

Table 29: OLS spatial regression model ......................................................................... 132

Table A-1: Missing data table for data sources used in the multivariate analyses in
Indianapolis ............................................................................................................. 168

Table A-2: J oint-actor logistic regression model including drug arrests but excluding
weapon arrests ......................................................................................................... 169

Table B-1: Pearson's r correlations between independent variables included in victim-
only logistic regression ........................................................................................... 171

Table B-2: Pearson's r correlations between independent variables included in joint-actor
logistic regression ................................................................................................... 172

Table C- 1: Additional HGLM model where both prior drug and weapon arrests were
included as level- 1 control measures ...................................................................... 174

Table C-2: Additional HGLM model where prior drug arrests were included as a level-1
control measure and both level-2 measures were included .................................... 175

Table C-3: Additional HGLM model where a combined lifestyle measure was included as
a level-1 control measure and both level-2 measures were included ...................... 176

Table C-4: Cross-level interaction models-disorder with incident level measures ........ 177

Table C-5: Mixed logistic regression model for joint-actor data where both drug and
weapon arrests were included as lifestyle control measures ................................... 178

Table C-6: Mixed logistic regression model for joint-actor data where only drug arrests
were included as a lifestyle control measure .......................................................... 179

Table C-7: Interaction terms added to the mixed logistic regression joint-actors model 180

Table D-l: Pearson’s r correlations between independent variables included in victim-

only HGLM model .................................................................................................. 182
Table D-2: Pearson’s r correlations between independent variables included in joint-actor
mixed logistic model ............................................................................................... 183
Table E-l: Additional HGLM model separating race and economic deprivation .......... 187

xi

LIST OF FIGURES

Figure 1: Framework for the macro-social theories of crime and place ........................... 23
Figure 2: Proposed theoretical model ............................................................................... 53
Figure 3: Example of census block aggregation to police beat level ................................ 67

Figure D-l: Histogram of standardized residuals from the victim-only HGLM mixed
model ....................................................................................................................... 1 84

Figure D-2: Histogram of standardized residuals from the joint-actor mixed logistic
model ....................................................................................................................... 1 85

xii

Chapter 1: Introduction

The harm that gang homicide inflicts on communities and citizens has been a
central focus of criminological research and criminal justice policy alike. Strategic
intervention programs, such as Project Safe Neighborhoods (PSN) have been created and
implemented on a national level in order to reduce gang and gun violence. As of 2007,
over $1.5 billion has been devoted to PSN in an effort to integrate and focus criminal
justice agencies and resources in order to reduce the pervasive harm of gun violence.I
This is particularly important because last year (i.e., Fiscal Year 2006), the United States
Department of Justice dedicated additional funding of over $30 million under the “Anti-
Gang Initiative” program. Government and law enforcement agencies are not alone in
the extended effort to understand, and undermine, gang homicide.

Criminal justice researchers began to focus specifically on gang violence, and in
particular gang homicide, in the late 1980’s. The augmentation of this specific type of
research was largely driven by scholars attempting to understand the drug epidemic that
swept through the nation, which has been attributed as a major vehicle for the increased
homicide rates that occurred during the same period (F agan, 1989; Vigil, 1988). Thus,
researchers began to study, and operationalize, gang homicide as a dichotomy: gang
homicide and non-gang homicide (Maxson, Gordon, and Klein, 1985).

Irnportantly, gang homicide research has informed the criminological literature.
Gang homicides have been shown to have a number of unique, and consistent,
characteristics. The most consistent correlates have been situational features of the

incidents and the demographic make-up of the participants. Prior research shows that

 

' This fiscal statistics described here were found on the PSN website (www.psn.gov).

gang homicides are more likely to be clustered, to occur in the street, involve the use of a
firearm, and have multiple, younger, non-white, and male actors (Decker and Curry,
2002). Many of these correlates were discovered using data that stem from chronic gang
cities, a term used to describe cities with an organized, long-tenn gang structure and
history (Spergel and Curry, 1993). However, most US. cities were classified as
emerging gang sites, which means that gangs were not a problem in these cities until the
1980’s, and that their gangs’ organization was less hierarchical and less formally
structured. Gangs have evolved over time and their organizational structure varies by
city (Decker and Curry, 2002). In order to contribute to this evolving body of research, it
is important to review the foundation on which emerging theories of gang homicide are
built. This study takes the approach that it is necessary to examine whether previously
established correlates of gang homicides are generalizable to emerging gang cities.

Chapter 2 serves as a review of the evolution of the gang homicide literature,
paying particular attention to the situational, demographic, and lifestyle explanations of
gang homicide. In addition to ascertaining the generalizability of incident level correlates
between chronic and emerging gang cities, this type of examination is also important
because the current research attempts to build upon the macro-social literature by
examining the effect of neighborhood level influences on gang homicide. Before an
examination of this scope can be accomplished, the extent to which gang homicide is
generalizable fi'om research based primarily on chronic gang cities to an emerging gang
city (i.e., the city of Indianapolis) must first be established.

Prior research shows that neighborhood violence undermines social networks and

cohesion among residents (Sampson, Raudenbush, and Earls, 1997). In terms of

neighborhood structure, the ecological literature suggests that high levels of disorder have
a direct effect on street crime, particularly offenses that are drug-market driven
(Blumstein, 1995; Sampson and Raudenbush, 1999; Wright and Decker, 1997). A
number of ecological theorists propose the neighborhood-crime relationship can be
explained by the “incivilities thesis”, which states that physical and social incivilities in a
neighborhood leads to increased fear and isolation among residents (Bursik and
Grasmick, 1993; Skogan, 1990). Residential isolation increases citizen fear, which in
turn promotes crime. Until recently, the extent to which neighborhood effects influence
gang homicide has been largely underdeveloped in criminal justice research, though a
growing body of literature in this area has begun to emerge in criminology (Curry and
Spergel, 1988; Kubrin and Wadsworth, 2003; Pizarro and McGloin, 2006; Rosenfeld,
Brey, and Egley, 1999).

Chapter 3 reviews the macro-social literature, paying particular attention to the
indirect theoretical link that has been established between disorder and gang homicide
through drug markets. More specifically, disorder has been shown to have a positive
correlation with robbery (Sampson and Raudenbush, 1999), and robbery has been shown
to have a positive correlation with drug markets (Wright and Decker, 1997), and drug
markets have been shown to have a positive correlation with gang homicides (Blumstein,
1995). This study examines whether there may be a more direct relationship between
disorder and gang homicide. This is important because to date no research has employed
statistical models to test the direct relationship between disorder and gang homicide.
Findings from this type of examination have the capacity to inform gang research as well

as macro-social criminology.

Chapter 4 details the data sources used in this study, while Chapter 5 describes the
measurement of the dependent and independent variables used in the multivariate
regression models. Data were derived from three primary sources. The first source is
incident level data, which includes situational features of the incident and demographic
characteristics of the actors. These data were collected at bi-monthly homicide incident
reviews in Indianapolis, and serve as independent variables for both the incident and
structural level analyses. In addition the dichotomous outcome measure, gang versus
non-gang homicide, was also collected at the homicide incident reviews. The second
source of data relied upon citizen surveys that were administered to residents in
Indianapolis police beats. This questionnaire was designed to gauge, among other items
of interest, perceptions of the local social and physical neighborhood characteristics.
Taken together, these items were aggregated to create a measure of disorder at the police
beat level, which served as the neighborhood boundary in this study. The third source of
data included US. Census measures that were aggregated to create a variable of
economic deprivation at the neighborhood level. Incident level measures were
incorporated into binary logistic regression models where the dichotomous gang/non-
gang homicide outcome measure was predicted (Long and Freese, 2003). Both incident
and neighborhood levels were analyzed together using hierarchical general linear model
(HGLM) analyses, in order to account for the violation of the assumption of independent
observations at the neighborhood level (Raudenbush and Bryk, 2002).

Chapter 6 describes the two research questions posed in this study, and results of
the multivariate analyses used to address these lines of inquiry. Research Question #1

asks, are there demographic, situational, and lifestyle distinctions between homicides that

were classified as gang related from those that were classified as non-gang related?
Binary logistic regression models were used as multivariate techniques to address this
research question. Research Question #2 asks, do structural measures of the
neighborhoods where the incident occurred delineate gang homicides from non-gang
homicides? Given the multi-level nature of the data used in Research Question #2,
Hierarchical Generalized Linear Modeling (HGLM) and mixed-level logistic regression
models were used in the multivariate analyses.

Chapter 7 reviews the limitations of the current study, the theoretical and
methodological contributions this research makes, and provides direction for fixture areas

of inquiry. In addition, policy implications are discussed.

Chapter 2: Review of Gang Research

One of the classical studies in criminal justice research demonstrating the
importance of understanding the relationship between gangs and general forms of crime
and delinquency was written eighty years ago (Thrasher, 1927). Since then, a large
volume of research has been devoted to the study of gangs in order to unwrap the multi-
dimensional components that make-up gang related crime. More recently, the rise in the
homicide epidemic that occurred in the United States in the late 1980’s and early 1990’s
is largely attributed to the rise in gang-related homicide due to the increase in the number
of gangs, as well as the number of gang members, and territoriality issues directly related
to the drug market (Decker and Curry, 2002; Klein, 1995; Miller, 2001). This is
important because the dynamics and processes previously established as factors that
explain gang related homicide may have also changed, given that it is widely accepted
that gangs have evolved over time. In order to build on this evolving body of theory, it is
important to review the foundation on which these emerging theories are built. Indeed, if
the foundation is not solid, emerging theories of gang homicide will not withstand either
the test of time or empirical scrutiny.

In this chapter, some of the previously established correlates of gang homicide are
first reviewed in detail. These include demographic, situational, and lifestyle correlates
that previously have been shown to delineate gang homicide from all other homicide
types. Following this review, a description of more contemporary definitional
advancements driving gang homicide research will be discussed. These include the
operational distinctions of gang-motivated and gang affiliated offenses. In addition,

contemporary research has shown that the dynamics explaining gangs vary across

location and over time. A large body of research explaining gang homicide has been
established in different areas (i.e., different cities) and across different points in time (i.e.,
different decades), which is important when attempting to contribute to the body of
knowledge in gang research.

These definitional advancements could potentially open gaps in the gang-
homicide foundation. With this in mind, this chapter concludes with a discussion of the
opportunity for research to contribute to the body of knowledge devoted to explaining
gang homicide in two ways. First, it becomes important to empirically re-establish
whether previously determined factors continue to delineate gang homicide from non-
gang homicide, given these recent definitional advancements. This is necessary in order
to assess the generalizability of the current study with findings presented in the prior
literature. Second, it becomes imperative to review the limited (but growing) body of
research that has been devoted to examining whether structural factors such” as
neighborhood effects have an impact on gang homicide, relative to other forms of

homicide. Taken together, this discussion sets the stage for the current investigation.

Demographic Characteristics of A ctors in Gang Homicide

The three main demographic correlates that have been consistent predictors of
gang homicides when compared to non-gang homicides include the age, race, and gender
distribution of victims and suspects. One of the most consistent findings in the empirical
research on gang homicides is that victims and suspects tend to be substantially younger
than actors involved in non-gang homicides. Spergel (1983) demonstrated long ago that

gang homicides in Chicago involved younger persons than did non-gang homicides.

Maxson and Klein (1985) conducted an even more in-depth analysis (i.e., reviewing over
700 homicide cases in Los Angeles) of the difference between gang and non-gang
homicides and found that there were substantial differences in the age, ethnicity, and the
number of participants of the actors involved. In terms of age, Maxson et a1. (1985)
found that suspects and victims involved in gang homicides were considerably younger
than victims and suspects involved in non-gang homicides. In a more recent study of
gang homicides in Chicago, Block (1991) demonstrated that the risk of victims and
suspects being involved in gang homicides peaked between the ages of fifteen and
nineteen years. In a St. Louis study, Decker and Curry (2002) found that the average age
of the victim in a gang homicide was considerably younger in gang homicides (average
age of 22.7 years for victims, and 19.2 years for suspects) than in non-gang homicides
(28.1 years for victims, 20.3 years for suspects). Pizarro and McGloin (2006) also found
that victims and suspects were younger in gang homicides in Newark, NJ. Thus, the
younger age distribution of victims and suspects in gang homicides is widely seen in the
literature on gangs.

Homicide research fiom a gang perspective has almost entirely focused on males.
This is largely because gang homicide is a male oriented crime. A consistent
characteristic of gang homicides is that they tend to involve male victims and suspects,
particularly in comparison with other types of homicides. Maxson et a1. (1985) found
that gang homicides were much more likely to involve males as both victims and
suspects. Maxson and Klein (1990) also concluded that gang homicides were more likely
to involve male victims and suspects. In terms of explaining this difference, Bowker,

Gross and Klein (1980) found that male gang members purposely kept their female

counterparts from participating in drive-by shootings and gun assaults. In addition, Joe
and Chensey-Lind (1995) suggested that violence is more a normative feature of male
gang involvement, than it is for female gang involvement. When explaining why, Joe
and Chesney-Lind (1995) posited that violence involving female gang members is more
of a consequence and response to both physical and sexual abuse.

Since the mid-1990’s, research on female gang participation has certainly become
more prominent within the literature (Cheney-Lind, Sheldon, and Joe, 1996; Curry, 1998;
Sikes, 1997). To date, only the Miller and Decker (2001) study has examined the role of
female gang participation in homicide. Miller and Decker concluded that females were
most likely to be involved in gang homicides when they “were in the wrong place at the
wrong time” (2001: 115). Thus, their in-depth examination showed that female
involvement in gang homicide is usually a product of chance. While this study makes no
contention that gang crimes are extensively male, research has consistently shown that
gang homicides have almost been entirely a male phenomenon.

In terms of race, Spergel (1983) found that the aggregated racial characteristics of
actors in gang homicides were different than in non-gang homicides. In a Chicago gang
study, Curry and Spergel (1988) found that the gang homicide rate was higher in
Hispanic communities, while the gang delinquency rate was higher in Afiican American
communities. Maxson et al. (1985) similarly found that gang homicides were distinct
from non-gang homicides in that the victims and suspects were more likely to be non-
white (Hispanic or African American). There have since been a vast number of studies

that have found a similar racial demographic distinction between gang and non-gang

homicides (Decker, 1996; Decker, Bynum, and Weisel 1998; Pizarro and McGloin, 2005;
Rosenfeld et al., 1999; Tita, 1998).

A number of classic and contemporary studies have devoted specific attention
toward explaining the increased risk of non-white participants’ involvement in gang
violence. Classical criminology attempted to capture this relationship through strain and
subculture theories of violence. Subculture of violence theories posit that a subculture
emerges in defiance of greater societal norms, often due to constraint, or strain (Cohen
and Short, 1958). Wolfgang and Ferracuti (1967) proposed that young, lower
socioeconomic class African Americans possessed a value system where violence was an
acceptable and normal part of everyday inner-city life. While this work has received
extensive criticism because it fails to explain the differential treatment of African
Americans and whites that has historically occurred in the United States, it set the stage
for understanding how violence is socially situated and affects members of racial groups
in distinct ways (Surratt, Inciardi, Kurtz, and Kiley, 2004). More recently, Parker and
McCall (1999) linked the association between Afiican American job inaccessibility and
high homicide rates across subcategories of homicide, which is a modern link to strain
and subculture of violence theories. The important point is that both empirical and
theoretical research has consistently found, and attempted to explain, that gang homicides
are more likely to involve non-white participants.

To summarize this section on the demographic makeup of gang homicides,
victims and suspects tend to be younger, male, and non-white when compared with non-
gang homicides. However, the growing body of research shows that these demographic

measures are by no means sufficient indicators of gang homicide alone. There are a

10

number of situational factors that make gang homicides distinct from other forms of

homicide.

Situational Correlates of Gang Homicide

There are several situational correlates that differentiate gang homicides from
other forms of homicide. Gang homicide research shows that these homicides are more
likely to include multiple suspects, involve the use of firearms as the method of death,
and are more likely to be drug related than are non-gang homicides (Maxson and Klein,
1990, 1996; Maxson et al., 1985). These measures have theoretical links to the drug
market, which will be discussed in more detail.

Maxson et a1. (1985) also found that the victim-offender relationship was an
important delineating factor when discriminating gang homicides fiom non-gang
homicides in Los Angeles. This research was an extension of Wolfgang’s (1958) 11-
point categorization of the victim-offender relationship across all homicide types in
Philadelphia. In particular, Maxson et a1. (1985) found that no prior contact between the
victims and suspects was more common in gang homicides than in non-gang homicides.
Riedel (1987) extended this work into a slightly different gang homicide typology:
expressive vs. instrumental homicides. Instrumental homicides are cases where the
offender seeks to improve his or her position through some rational calculation such as
minimizing risk, increasing gain, or both. Expressive events fail to include the
rationality, or the ‘cost-benefit’ comparison. Expressive homicides tend to center around
‘saving face’ or ‘character contests’ (Decker, 1993). An important point that must be

raised is that the current study has no measures of the victim-offender relationship. This

11

will be discussed in greater detail in the discussion section dealing with limitations and
directions for future research. However, this body of research continues to build the
argument that gang homicides are different than non-gang homicides.

In terms of the measures available in the current study, previous research has
demonstrated that gang homicide is likely to be motivated by the drug market. The
participation of gang members in illegal drug sales is well established across a variety of
studies (Decker and Van Winkle, 1996; Pagan, 1989, Maxson et al., 1985; Vigil, 1988).
Blumstein and Wallrnan (2000) linked the temporal sequencing of the rise in homicide in
the early 1990’s with the emergence of street drug sales.2 They concluded that this was
particularly true in crack cocaine sales, where distributors began arming themselves with
firearms as a means of safety and protection.

From this theoretical explanation (i.e., the relationship between crack cocaine
drug markets and gang homicide), it is plain to see how researchers have attempted to
explain the relationship between the increased number of suspects, the use of firearms as
the method of death, and the increased likelihood of drug-related motives as correlates of
gang homicide. The group dynamic is captured in the measure of multiple suspects in
that offenses are the target of the group rather than an individual perpetrator. The other
two common situational correlates of gang homicides are incidents that are drug
motivated and involve firearms as the method of death. In addition to these situational
correlates, some research has attempted to link lifestyle measures with gang homicide,

since lifestyle measures attempt to identify ‘risk factors’ that increase the likelihood of an

 

2 It should be noted that Martin, Maxwell, White, and Zhang (2004) found little to no relationship between
cocaine use and city wide violent crime in a study utilizing national data from the Arrestee Drug Abuse
Monitoring (ADAM) program. However, their f'nrdings did not test the relationship between drug
sellers/markets and crime trends but rather focused on arrestee drug use and crime trends.

12

individual becoming involved in a gang homicide. A review of lifestyle measures

follows.

Lifestyle Measures of Gang Homicide

Within criminology, lifestyle measures essentially deal with risk factors that
increase the likelihood of an individual being involved in crime. Hindelang, Gottfredson,
and Garofalo (1978) were among the first theorists to establish a lifestyle theory of crime.
They proposed that individual risk factors and characteristics influenced different risk-
levels of victimization and/or offending. According to their theoretical premise, there are
several propositions related to lifestyle that determine the likelihood of participating in
crime and being victimized. First, they found that personal victimization was directly
related to the amount of time an individual spent in public places, particularly during the
evening hours, which is basically an indicator of their lifestyle. They also found
individuals who were more likely to be involved in crime and victimization hung around
others with similar lifestyles, which also helped explain personal victimization.

In terms of a more narrow focus toward understanding violence, Anderson (1999)
extended lifestyle theory by relating it to the inner city African American population.
Anderson (1999) specifically argued that violence in the inner city is largely due to the
‘search for respect’ that often results in the use of violence to ensure that individuals,
particularly those who self-identify with the ‘street culture’ rather than ‘decent culture’,
avoid being disrespected at all costs. He demonstrated that many young inner city
African American men and women feel a sense of hopelessness and alienation, largely as

a result of joblessness and persistent racism. This is often the precipitating and

13

underlying factor that causes violent confrontations. Anderson (1999) concluded that this
type of violence further legitimizes the oppositional culture and the “code of the street” in
the eyes of many inner city African Americans. Similarly, Oliver’s (1994) work
demonstrated that a large proportion of inner city Black men define masculinity (i.e.,
manhood) as something contingent upon their willingness to resort to violence in order to
resolve disputes.

In terms of gang violence, Anderson (1999) additionally argued that lifestyle is an
important indicator of violent behavior at the street-level. In particular, Anderson found
that the use and threat of violence by drug dealers is a means to gain social control. This
is because drug dealing is the most prominent source of income in these areas. In this
setting, gangs who are involved in the drug market ofien use violence to maintain street
credibility; otherwise, they would be unsuccessful drug dealers. Overall, this structural
setting creates a constant threat of violence due to a combination of both the drug trade
and the locales where drugs are most heavily purchased and sold.

Lifestyle measures are very consistent with routine activities theory (Cohen and
Felson, 1979). A routine activities approach attempts to explain the relationship, and
convergence, between victims and offenders (and the lack of intervening factors).
Specifically, Cohen and Felson (1979) argued that drug dealing, gang involvement, and
nighttime activities outside the home are all part of routine activities that are related to
risk levels of violent offending and/or victimization.

The operationalization of lifestyle is discussed by Hindelang, Gottfredson, and
Garafolo (1978), who asserted that lifestyle can be measured by the amount of time an

individual spends with his or her family, since individuals more likely to be engaged in

14

deviant lifestyles tend to spend time away from their family. A proxy measure of deviant
lifestyle is the criminal history based on an individual’s arrest and/or conviction record.
Jensen and Brownfield (1987) specifically stated, “offense activity can be considered as a
characteristic of lifestyles or as a type of routine activity which increases the risk of
victimization because of the motives, vulnerability, or culpability of people involved in
those activities” (Jensen and Brownfield, 1987: 87).

This type of operationalization of lifestyle would fit Sampson and Lauritsen’s
(1994) lifestyle principle known as homogarny, which is a principle that states that
persons are more likely to be victimized when they disproportionately associate with or
come into contact with offenders. It is important to note that to date, no study has
examined lifestyle measures as offense histories with gang homicide, though their
potential is seen, both directly and indirectly, in a number of Other scholarly works
(Anderson, 1999; Cohen and F elson, 1979; Lauritsen and White, 2001; McGarrell,
Chermak, Wilson, and Corsaro, 2006; Sampson and Lauritsen, 1994).

Thus, offense histories have potential to serve as lifestyle measures in terms of
capturing the rate of risk of being involved in gang homicides, among both perpetrators
and victims, by examining prior drug and weapons arrests. Given that Blumstein and
Walhnan (2000) linked gang homicides with drug and weapons offending, these
constructs have the potential to serve as key measures of an offender’s prior lifestyle.
The lifestyle and routine activities theories also serve to integrate the situational measures
(i.e., group, weapon, drugs) with these lifestyle characteristics. In addition to the
contribution to lifestyle theory, research dealing specifically with gang-homicide also has

the potential to re-affirm the relationship between prior demographic and situational

15

measures and gang homicide, given the recent advancements made in the literature. A

review of these advancements follows.

Recent Advancements in Gang Homicide Research

In 1988, the Office of Juvenile Justice and Delinquency Prevention (OJ J DP)
conducted the National Youth Gang Survey. In this particular survey, the questionnaire
contained a variable that allowed researchers to make distinctions in terms of defining
each city’s localized gang problem. The surveys were administered to participants who
had some direct operational and policy experience in dealing with its local gang problem
(Spergel and Curry, 1993).

From this survey, a typology arose that distinguished cities that had a gang
problem into one of two types: chronic gang cities and emerging gang cities. ‘Chronic
gang cities’ were defined as cities that had developed gang problems prior to 1980. In
addition, chronic gang cities had to have at least some gangs that were organized and
engaged in serious and often violent criminal activities (Spergel and Curry, 1993). The
label of emerging gang cities was applied to cities that had developed gang problems
after 1980, and the organizational structure of the gangs in these cites was not of
particular importance to this designation. Decker and Curry (2002) noted that most of the
extensive research on gang homicide emerged from cities that were defined as chronic
gang cities. This is particularly true in studies that were conducted in Los Angeles
(Maxson et al., 1985; Spergel, 1983) and Chicago (Block, 1991; Curry and Spergel,

1988)

16

In addition to these distinctions, definitional differences began to emerge even
when comparing gang homicides in Los Angeles with those in Chicago, which were the
two major chronic gang cities. In Chicago, the designation was more restrictive in that
homicides were classified as gang homicides only when the motive for the offense was
gang related (Block, 1990). In Los Angeles, the classification was more loosely defined
in that homicides were classified as gang homicides if the event itself involved
individuals (either a victim or suspect) who were included in a list of known street gang
members (Rosenfeld et al., 1999). Maxson and Klein used the more restrictive Chicago
definition on gang homicide data in Los Angeles and found that gang homicides would
have been “halved” (Maxson and Klein, 1990: 190). Thus, it is necessary for research in
this area to first establish the specific type of gang homicide definition that is being used,
in order to compare results across previous studies.

Additional research added to this body of knowledge by further exploring the
definitional issues concerning gang homicide. Maxson and Klein (1996) conducted an
analysis examining an empirical distinction between gang affiliated and gang motivated
homicides. Based on narrative descriptions of the incidents, the researchers classified
gang affiliated homicides as incidents that involved gang members, but that the homicide
itself was not a product of gang activity. Gang motivated homicides were distinguished
as homicides that involved gang members where the homicide was a direct result from
gang activity. Maxson and Klein (1996) found virtually no distinctions across these
homicide subtypes and concluded that there was no fundamental difference between gang
affiliated and gang motivated homicides, at least concerning the intent and motivation

behind the offenses.

17

Rosenfeld et a1. (1999) replicated this study and found similar results when
comparing gang motivated and gang affiliated homicides. However, Rosenfeld et a1.
(1999) did establish that gang motivated and gang affiliated homicides did significantly
differ from non-gang youth homicides, which are often classified as other forms of
homicide. On a related note, Rosenfeld et a1. (1999) also compared the neighborhood
context across the various gang homicide typologies and did not find significant
differences in the structural (i.e., neighborhood) measures across homicide subtypes.

This is important because their research was one of the first studies to attempt to integrate
structural measures in the designation of gang homicides, which is one of the major
purposes of the current research. More detail on this contribution will be discussed in the
next chapter.

An additional measurement issue related to the gang designation relates to the fact
that most gang research relies on police designations of the incident and the actors as
“gang related.” Abundant criminological research has demonstrated the limitations of
officially defined crime measures (Sykes, 1974). Among the limitations is the fact that
law enforcement practices, public policies, and resource allocation can be contingent
upon the operational definition of gangs and gang-related crime. This can lead to both
under— and over-estimates of gang related incidents. Indeed, Esbensen, Winfree, He, and
Taylor (2001) proposed that inaccurate definitions of gang membership either under or
overestimate gang membership and affect public perception and research accuracy. In
order to address this limitation, Esbensen et a1. (2001) focused on self-reported gang
identification and found it to be a reliable measure. However, self-reports are not

available in the context of crime incidents. Thus, relying on a law enforcement

18

classification is typically the only available method when trying to assess the
characteristics of an actual crime incident (i.e., whether or not the offense involves gang

members).

Chapter Summary

Studies specifically explaining gang homicide have become more prominent in
criminal justice research. When emphasizing the importance as to why gang homicide
deserves special attention distinct from other forms of homicide, Decker and Curry
(2002) demonstrated that gang homicides have a number of unique, and consistent,
characteristics. These include: the spatial concentration of the event, weapon (i.e.,
firearm) use, race of victim and suspect, location, drug involvement, age of actors,
gender, and the victim-offender relationship (2002: 345).

However, these correlates that are mainly accepted within the gang homicide
literature were largely informed by research that was conducted in Chicago and Los
Angeles, which have been shown to be qualitatively distinct from other cities in terms of
their gang structure and history. In contrast, most US. cities were classified as emerging
gang cities, which meant that gangs were not a problem in those cities until 1980, and
that their gangs’ organization was probably less hierarchical and less tightly structured.

In addition to the potential differences between chronic and emerging gang cities,
some studies have defined homicides as gang affiliated (i.e., a known gang member was
an actor in the offense), while other studies have relied on a more gang specific
operationalization, which is gang motivated homicides (i.e., the homicide itself was a

direct result of gang activity). Thus, it is important to examine whether previously

19

established correlates of gang homicides work within an emerging gang city, where the
definition is more along the lines of gang affiliated offenses (i.e., the Los Angeles
definition). The differences in the gang homicide definition create the opportunity for
inconsistencies when comparing results across studies.

Examining whether previously established correlates of gang homicide provides
insight into the law enforcement designation of gang homicide is especially important. A
direct test of the validity and reliability of the gang homicide definition can be conducted
by cross-validating the law enforcement definition of gang homicide with measures that
have been previously established as correlates of gang homicide. There has been little to
no research to date that has questioned the validity of these previously established
incident level correlates of gang homicide. This ultimately allows for further theoretical
and empirical examination of additional measures that have not been established with
gang homicide, such as structural (i.e., neighborhood) effects.

Understanding gang homicide is also important because there is a limited, but
growing body of research attempting to link macro-social level theories specifically to
gang homicide. In order to incorporate macro-level theories into the explanation of gang
homicide, a concrete distinction between gang and non-gang homicides is first required.
The features between gang and non-gang homicide must be empirically and theoretically
validated at the incident level in order to integrate macro-level theories into this area of
research. The foundation (i.e., incident level theories of gang homicide) was the focus of
the current chapter. The next chapter deals with a detailed review of macro-social

theories.

20

Chapter 3: Macro-Social Theories of Crime

While research regarding gang homicide has grown exponentially since the early
1990’s, macro-social ecological theories explaining the relationship between crime and
place have a much longer and more extensive history in criminology. In order to capture
the breadth of knowledge in this body of research, it is necessary to start broadly at the
classical and contemporary levels that explain crime in general and eventually ‘funnels
down’ toward a more crime specific level. This layering process is intended to inform
the current research by showing how certain ecological measures are important to explain
crime in general, while other ecological variables may be more attuned at explaining
specific types of crime.

Taylor (1997) described how geographers use a ‘cone of resolution’ to organize
knowledge about spatial processes at different levels of analysis (Brantingham and
Brantingham, 1976). Essentially, the observed spatial pattern varies as you progress
down the cone to increasingly smaller scales of analysis. Of particular importance to this
funneling effect is the idea that the factors that explain crime change as our focus narrows
to specific crime types. Thus, the explanatory factors narrow as you progress down each
level of the cone of resolution (Brantingham and Brantingham, 1976) becoming ever
more refined. Taylor (1997) argued that macro-social theories are well supported as one
examines specific types of offenses.

Consistent with this theme, I will start broadly and ‘funnel down’ within the
macro-social framework by examining how environmental factors first explain crime in
general, then into more specific offenses, and eventually in terms of explaining gang

homicide. Throughout, I will discuss how macro-level factors can facilitate or inhibit

21

crime both generally and specifically. Ultimately, the purpose of this review of the
ecological literature is to demonstrate the potential relationship between neighborhood

deprivation as well as incivilities and gang homicide.

Cone of Resolution

A macro-level or ecological analysis examines how characteristics of delimited
geographic areas such as neighborhoods, census tracts, cities, counties, states or nations
are related to crime rates. Macro-level theories seek to explain why certain
characteristics of ecological areas, but not others, account for the distribution of crime
(Pratt and Cullen, 2005). Figure 1 displays the cone of resolution that is described in
more extensive detail throughout this chapter.3 The cone of resolution used in this
framework includes four distinct levels used to explain crime from a macro-social
process. Level 4 involves the examination of crime from classic social disorganization
theory, and is the highest level of resolution in this study. Classic social disorganization
theory was among the first criminological paradigms to examine crime from a structural
level by showing how demographics and social factors such as economic status, ethnic
heterogeneity, and residential instability explained crime in the same neighborhoods over
time. Level 3 includes contemporary social disorganization theories, and is more narrow
due to the refinements and advancements that continue to occur within the social
disorganization paradigm. Specifically, this section reviews the literature on how human

action such as weakened social control mechanisms, fear and incivilities, and political

 

3 Taylor’s (1997) use of the cone of resolution primarily focused on levels of geography moving from
nations to street blocks. I extend this heuristic by moving from classic macro-social theories to theories
focused on specific crime types (i.e., gang homicides). This often overlaps with a movement from larger
geographical areas to smaller areas, and is distinct from Taylor’s application.

22

and social factors converge to create crime in general. Level 2 funnels down even further
by examining violent crime, such as homicide as a specific type of offense. The premise
that is built in this section is that the social processes that explain violent crime and
homicide are somewhat more refined than are those theories that explain crime at a
general level (i.e., multiple types of crime). Finally, Level 1 explains gang homicide and
reviews the advances made in this body of research from a structural perspective. While
the research attempting to link macro-social processes to gang homicide is currently
limited, it offers a great opportunity for research to contribute to the literature in this area.
One of the major purposes of the current study is to advance our understanding of gang

homicide from a structural perspective.

Figure 1: Framework for the macro-social theories of crime and place

 

Level4

 

 

Level3

Level 4 = Classical Ecological
Theories

Level 3 = Contemporary Ecological
Theories

 

 

Level 2 = Ecological Theories Specific
to Violence and Homicide

Level 1 = Ecological Theories Specific
to Gang Homicide

 

 

 

23

Level 4: Classical Ecological Theories

Park and Burgess laid the foundation for urban sociology by defining
communities as 'natural areas' that developed as a result of competition between
businesses for land use and between population groups for affordable housing. A
neighborhood is both a collection of people and institutions occupying spatially defined
areas, influenced by ecological, cultural, and sometimes political forces (Park and
Burgess, 1924). These areas are natural, activity related, residential, and physical.
Natural areas include many different forms, such as ethnic enclaves or activity related
areas. The activity related areas include business districts, shopping districts,
manufacturing districts, and residential areas. The residential areas are different among
income groupings, such as middle class neighborhoods compared with ‘ghettos’. There
are also physical components that divide the city, such as rivers, lakes, railroad tracks,
and airports.

Park and Burgess (1924) developed a ‘concentric zone theory’, which was also
employed by Shaw and McKay. Zone I was the central business area, which included
mostly manufacturing areas. Zone H was described as the ‘zone in transition’, which was
also referred to as the ‘ghetto’. Zone 11 was marked by a high level of transition, with
people moving in and out of the area. It was hypothesized that this zone of transition led
to social disorganization. Zones III to V were residential areas, HI being mostly
comprised of working-class homes, IV being middle-class homes, and V being commuter
residential areas.

Theories that explained neighborhood influences of criminal behavior soon

emerged from this tradition. Social disorganization theory in this context focused on the

24

effects of ‘kinds of places’, or more specifically, the idea that different types of
neighborhoods create conditions favorable or unfavorable to crime and delinquency. The
theory of social disorganization was largely developed out of the research conducted in

Chicago by Shaw and McKay.

Classic Social Disorganization Theory

Upon studying Chicago's juvenile court records over a period of several decades,
Shaw and McKay (1942) noted that crime rates were not evenly dispersed across time
and space in the city. They showed there were marked variations in the rate of truancy,
delinquency, and recidivism between different areas, which could not be explained in
terms of population size and density. They used these findings to explain some of the
variation in juvenile delinquency patterns. Essentially, they found that the distribution of
juvenile delinquents follows the pattern of physical structure and social organization of
American cities.

Shaw and McKay (1942) also found that crime was concentrated in the poorer
(slum) neighborhoods and that the racial and ethnic make-up of these neighborhoods did
not particularly matter. They supported this conclusion with empirical evidence,
essentially by examining the succession patterns of the city's poorest neighborhoods. In
essence, as new immigrant groups replaced older ones, the delinquency rates within these
neighborhoods remained remarkably stable over time. In addition, they found that
ethnicity did not prove to be a significant factor related to delinquency across the city, or
over time. Each ethnic group produced similar rates of delinquency, depending on the

type of neighborhood in which they resided. They found that urban areas experiencing

25

rapid changes in their social and economic structure had the highest crime rates,
regardless of the different groups residing in these areas.

The zone in transition was characterized by low economic status, ethnic
heterogeneity, and residential instability. Shaw and McKay (1942) posited that these
factors lead to the dissolution of social bonds and fiiendships (i.e., low social control)
among residents in the community. Low economic status was seen as creating ‘strain’
for residents of the zone in transition because individuals were forced to live in socially
disorganized neighborhoods. Strain theories emphasize the cultural goals of economic
success that generate pressure on individuals who do not possess the legitimate means to
accomplish material success. The tension that occurs causes a strain that leads to crime
and deviance (Agnew, 1985; Cloward and Ohlin, 1960; Merton, 1959; Messner &
Rosenfeld, 1994).

In addition, Shaw and McKay’s neighborhood social disorganization theory relied
on tenants from culture conflict theory. From a conflict perspective, crime and deviance
is brought about by an inter-group struggle for dominance (Sellin, 1938). Specifically,
Shaw and McKay (1942) contended that new waves of immigrants contributed to the lack
of social cohesion because of the different values (and backgrounds), different traditions,
and lack of communication that occurred between citizens in these areas. These same
tenants from classical social disorganization theory can also be found in more

contemporary criminological theories of crime and place.

Level 3: Contermaorary Ecological Theories
Social disorganization theory started with (and still operates today) five major

tenants (Pratt and Cullen, 2005). First, humans are social creatures and their behavior is

26

the product of their social environment. Second, the social environment provides cultural
values and definitions to govern the behavior of those who live within them. Third,
urbanization and industrialization have created communities that have a variety of
competing cultures, thus breaking down older and more cohesive patterns of values.
Fourth, this breakdown, or disorganization, of urban life has resulted in the basic
institutions of family, fiiendships groups, and social groups becoming more impersonal.
Fifth, as the values provided by these institutions become fragmented, several opposing
definitions about proper behavior arise and come into conflict. Thus, continued
disorganization makes the potential for conflict even more likely. Given this foundation,
contemporary research finds that there are several components that make-up socially
disorganized neighborhoods. One robust theoretical explanation of socially disorganized

neighborhoods is the loss of informal social control mechanisms.

Factors that Inhibit Neighborhood Social Control

Social disorganization theory has experienced several advancements in the past
two decades. In particular, recent scholars have extended social disorganization theory to
include several dimensions shown to impede neighborhood social control mechanisms.
These dimensions include family disruption, participation in community organizations,
urbanization, and population density in addition to Shaw and McKay’s (1942) classic
components of ethnic heterogeneity, residential mobility, and low economic status. More
recently, social disorganization theory began to include the constructs of collective

eflicacy and social capital as measures of informal social control.

27

According to Reiss (1985), corrrrnunities are important to study in order to
understand crime for various reasons. He found that a higher concentration of offenders
and victims of “common crimes” lived in low status neighborhoods. In addition, Reiss
(1986) showed that the social status and delinquency rate of a neighborhood had an effect
on the likelihood of youths in terms of increasing the risk of delinquency due in large part
because many youths were joined in local social networks.

Reiss (1986) stated that these networks were based primarily in neighborhoods or
communities. From this perspective, Reiss (1986) argued that youths produce different
subcultures, depending on the distinctive characteristics of a neighborhood and its
integration as well as its opportunity structure (Cloward and Ohlin, 1960). Deviance or
criminality is often seen as the byproduct of the failure of personal controls to deter
deviant conduct and from the failure of formal and informal social controls to induce and
reinforce conformity.

Sampson (1986) demonstrated the importance of the family structure when
modeling social disorganization. From this perspective, Sampson (1986) argued that the
family is important in terms of understanding the social disorganization framework for
three major reasons. First, neighborhoods with pronounced family disruption are not as
likely to provide an effective network of Social control (e.g., participation in voluntary
organizations and local affairs). Second, informal social control is attenuated due to
family disruption. Third, since serious crime is mostly a male phenomenon, it may be
that divorce contributes to the pool of males’ unattached and therefore freed from

mechanisms of social control.

28

Sampson and Groves (1989) found that neighborhoods with low economic status,
ethnic heterogeneity, high instability, and disrupted families inhibited social control at the
neighborhood level. In addition, these elements impeded individuals from participating
in voluntary organizations. They found that these factors were particularly important in
urban communities (i.e., the urbanization component), and related to a decreased capacity
for social control. Thus, communities, particularly urban areas, with sparse friendship
networks, unsupervised teenage peer groups, and low organizational participation had
high crime and delinquency rates.

Coleman (1988) was one of the first scholars to introduce the concept of social
capital. Social capital refers to intangible resources produced through relations among
people who facilitate action for mutual benefit (Coleman, 1988). In other words, social
capital refers to the resources that individuals garner from their relationships with other
people. The intangible aspect of social capital refers to the resident’s personal investment
in community problems and his or her desire to intervene in conflicts within his or her
neighborhoods in order to fix them. Thus, it is the willingness to take action developed
by social capital that enables residents to work together as a collective unit to solve the
problems in their communities. In contrast, the absence of social capital affects the
willingness of local residents to address neighborhood crime and disorder.

In terms of the relationship between ethnicity and poverty, William Julius Wilson
(1987) argued that the out migration of the middle class (in particular the Black middle
class) resulted in neighborhoods with insufficient economic and social foundations for
effective social control. Communities characterized by j oblessness, lawlessness, and low

achieving schools are crucial to understanding the development of the underclass. He

29

contended that without a frnancially stable middle class, these communities have neither
the residents who socialize their youngsters to conventional values, nor the ability to
sustain local institutions (Wilson, 1987). I

Sampson and Wilson (1995) posited that macro social patterns of residential
inequality give rise to social isolation and ecological concentration of the ‘truly
disadvantaged’, which in turn leads to structural barriers and cultural adaptations that
undermine social organization and hence the control of crime. In other words, inequality
leads to concentrated disadvantage, which leads to cultural adaptations of ghetto related
behavior.

Consistent with Shaw and McKay’s original findings, it is not “skin
pigmentation” that is captured with a race variable, but rather social conditions that are
highly correlated with racial and ethnic groups. Blau and Blau (1982) showed that when
relative deprivation is controlled for (in their model for violent crime), the previously
significant race variable dropped out. Lauritsen and White (2001) concluded that racial
difference in risk might reflect processes of community control or exposure that are
beyond the control of a given individual. However, some scholars such as Parker (2002)
proposed the importance of disaggregating event types (e. g., particularly homicide)
because the impact of urban disadvantage and family structure is different for whites and
Blacks, thus strengthening the need for context-specific explanations of crime and in
particular those that have a racial or cultural influence.

In addition to structural factors, Kubrin and Weitzer (2003) stated that culture
needs to be included in future theoretical elaboration. They showed that empirical

evidence points to a model of the cultural order in disadvantaged neighborhoods as both

30

structurally conditioned and diverse. Social isolation occurs in socially disorganized
neighborhoods. At the same time, some residents who lack opportunities for economic
advancement through conventional means pursue alternative routes to gaining status and
prestige. This is similar to Durkheim’s concept of anomie, whereby culture mediates and
conditions differential access to legitimate means (Durkheim, 1897/ 1979).

The role of social control in explaining the impact of social disorganization on
neighborhood variation in crime was advanced considerably through the research of
Sampson, Raudenbush, and Earls (1997). Specifically, they introduced the concept of
collective eflicacy as an indicator of informal social control. Collective efficacy was
operationalized by two items: shared expectations for social control and social
cohesion/trust. The concept ‘shared expectations for social control’ was operationalized
from a survey as an additive five-point Likert-scale item. Residents were asked if they
would take action if: 3) children were skipping school and hanging out on a street comer;
b) children were spray-painting graffiti on a local building; c) children were showing
disrespect to an adult; (1) a fight broke out in front of their house; and e) the fire station
closest to home was threatened with budgetary cuts. Social cohesion/trust, was
operationalized as a four-point item: “People around here are willing to help their
neighbors; this is a close-knit neighborhood; people in this neighborhood can be trusted;
people in this neighborhood generally don’t get along with each other (reverse coded);
and people in this neighborhood do not share the same values (reverse coded). Both of
these items are then aggregated to the neighborhood level.

Using this construct, Sampson, Raudenbush, and Earls (1997) found that

collective efficacy was negatively correlated with violent crime (e. g., when collective

31

efficacy is high, violent crime is low and vice versa). They found that immigration
concentration and residential stability explained 70 percent of the neighborhood variation
in collective efficacy. In addition, measures of collective efficacy yielded aghigh
‘between neighborhood’ reliability and were negatively associated with variations in
violence when individual characteristics, measurement error, and prior violence were
controlled. Additionally, collective efficacy explained variation in violent crime even
after controlling for neighborhood income levels. These results show that social control
mechanisms are important when trying to explain crime at the neighborhood level. Thus,
it becomes important to understand why control mechanisms are linked to neighborhood

crime.

Crime Explained by Low Social Control

To further explain variation in crime, it is necessary to move further down the
cone of resolution from theories that explain crime from general neighborhood
characteristics to theories focused on dynamics that generate crime when social control is
low. Felson (1987) stated that there are three key ingredients necessary for crime to
occur. There must be a likely offender, a suitable target, and the absence of a capable
guardian. Under this umbrella of routine activities theory, Felson (1987) contended that
there is an inter-connection between the routine activities of a place (e. g., how often these
three elements converge) and crime. Theoretically, the idea of routine activities at the
neighborhood level is very much ‘in line’ with Travis Hirschi’s control theory of crime.

Hirschi (1969) argued that attachment through social bonds prevents delinquent
behavior. According to Felson (1987) this is also true of potential criminals at the

neighborhood level because the presence of an intimate handler or capable guardian

32

prevents the criminal activity. However, such prevention is difficult to accomplish by
remote (or distant) control. Both Hirschi and F elson’s theories assume motivated
offenders who will commit crime whenever the ‘constraint’ is removed.

Felson stated, "just as lions look for deer near their watering hole, criminal
offenders disproportionately find victims in certain settings or high-risk occupations"
(1987: 914). The very system that fosters easy movement and vast opportunity for good
experiences also interferes with informal social control of youths and protection of person
and property. Routine activities theory does not assume complex offender thought
processes. It assumes Zipf’s (1950) principle of least effort where people tend to find the
shortest route, spend the least time, and seek the easiest means necessary to
accomplishing a goal. In this case, the goal is choosing a potential target.

This process begins when potential offenders first shake loose from the parent or
handler, then find a target for a crime when unmonitored by a guardian. Via the street,
the offender finds victim just as the teenager evades parent (Felson, 1987 : 917). Felson
stated that the street belongs to everyone, hence is supervised by no one, except for an
occasional law enforcement officer who may not know who belongs there anyway. This
is directly associated with the level of social control that residents have over their
neighborhood.

In the Sampson and Morenoff (2004) Chicago study, the intent was to see how
neighborhoods fare as units of guardianship and collective efficacy when the outcome is
homicide and robbery victimization. They found that neighborhoods were not
independent of one-another but rather were interdependent and characterized by a

functional relationship between what happens in one point in space and what happens

33

elsewhere (Sampson and Morenoff, 2004). They concluded that neighborhoods
themselves can serve as guardians when the following occurs: there is high collective
efficacy in and across neighborhoods; there is commitment to neighborhood
organizations, such as the presence of a block organization, a community newspaper, and
crime-reduction groups; and when there are voluntary (grass roots) associations such as
religious, political, or ethnic organizations. Consistent with the notion of collective
efficacy, residents both within and across neighborhoods can serve as guardians, if they
are both invested and unaflaid to do so (McGarrell, Giacomazzi, and Thurman, 1997;
1999). This is directly linked to the research on neighborhood incivilities and fear, which

is often referred to as neighborhood disorder (Skogan, 1990).

Disorder

Fear and incivilities (i.e., neighborhood disorder) can serve as both indicators of
as well as impediments to informal social control mechanisms. Some theories attempt to
explain fear as an outcome (Garafolo and Laub, 1978) while others use fear as an
intervening factor in explaining crime (Skogan, 1990). Most macro-social theories of
crime are somewhere in-between (i.e., they control for fear when explaining crime). The

evolution of the theories on fear and incivilities is discussed below.

Fear

Wilson (1975) and Garafolo and Laub (1978) were among the first to focus on
fear of crime, urban unease, and psychological factors and their relationship with
community dynamics. The key to their focus was to examine why so many people are

fearful of crime and incivilities. Hunter (1974) extended this idea another step forward

34

by describing how residents may interpret signs of incivility. Garafolo and Laub (1978),
Wilson (1975) and Hunter (1974) all concluded that signs of physical incivility lead to
residential fear by serving as cues of danger and potential victimization.

Hunter (1974) found that residents attribute disorderly actions and deteriorating
physical conditions to two complementary sources. Internally, the perceivers attributed
conditions to local residents and organizations that are unable to manage or preserve the
neighborhood. Beyond the neighborhood, perceivers concluded that the external
agencies of control, which bear some responsibility for preserving order, were unwilling
or incapable of doing so in that locale. Hunter stated that residents believed no one or no
external agencies could help, which caused them to feel personally at risk for
victimization. Hunter asserted that it was the signs of incivility, and the meaning
attached to them. Hunter also proposed that signs of incivility and crime were reciprocal
and that they were attached to an exogenous cause: neighborhood disorder. In sum,
Hunter (1974) extended the explanation beyond psychological processes that cause fear,
to include neighborhood crime rates and mutual impacts of crime and incivilities, and
placed these factors within varying community contexts.

McGarrell et a1. (1997) examined three models in terms of their relationship with
fear of crime: victimization, disorder, and community concern/community control. The
victimization model assumed those who have been victimized would have higher levels of
fear, and that groups more vulnerable to crime (e.g., women and the elderly) would likely
be more fearful. The disorder model assumed that disorder indicated weakened local
social control and attenuated traditional norms because physical decay signaled the lack

of residential concern about the neighborhood. The least developed of the three was the

35

community concern model that held that fear arose as concerns about the neighborhoods
increased. The authors maintained that fear was indeed a consequence of the erosion of
social control, as urban residents perceived it. In essence, fear was low in well-integrated
neighborhoods. An important point is that the addition of a community dimensions
measure increased the variance explained (in fear) by over 11 percent. Consequently,
McGarrell et al. (1997) suggested the need to include the community concems/infonnal
social control model along with the victimization and disorder models in order to better
explain fear of crime.

Whereas victimization, disorder, and community concern acted as facilitators of
fear, control, community integration and community cohesiveness acted as inhibitors of
fear. Specifically, residents who perceived their neighborhood as a real home, where
people mostly help one another, and who perceived the overall neighborhood, the city,
and the police as responsive, tended to be less fearful (McGarrell et al., 1997).

Gibson, Zhao, Lovrich, and Gaffney (2002) also examined fear of crime, and the
mediating effect of collective efficacy on fear. The authors argued that social integration
is a first-step in the building of neighborhood collective efficacy. Consistent with
McGarrell et al.’s (1997) findings, their findings suggested that as social integration and
collective efficacy co-varied, the level of citizens' perceptions of collective efficacy
played a mediating role in the relationship between social integration and perceptions of
fear of crime.

Gibson et al. (2002) also used measures from victimization models (past
victimization), social disorder models (character of neighborhood environment), social

integration models (ability to identify strangers in the area and feeling a part of the

36

neighborhood) in their investigation. They also captured individual perceptions of
collective efficacy (how trustworthy they feel their neighbors are, whether neighbors will
help, etc).

Gibson et a1. (2002) found that social integration had the greatest impact (effect)
on individual perceptions of collective efficacy, holding sociodemographic factors, prior
victimization, and perceptions of neighborhood social disorder constant. Perceptions of
social disorder had the second most important effect on collective efficacy. Social ties
among neighbors may have led to attachments that resulted in the building of trust among
neighbors and expectations that neighbors would intervene as agents of social control
during appropriate situations. They found that increased perceptions of collective
efficacy had a relatively large impact on alleviating the fear of crime among residents.
Social integration may be an important initial factor in explaining fear of crime, but
developing a sense of trust among neighbors and cohesion around the sense of obligation
to act as agents of informal social control may have been more crucial in terms of
stabilizing or improving neighborhoods. Other researchers in ecological research began
integrating fear as a key component into a more dynamic process: neighborhood

incivilities.

Incivilities

Neighborhood physical incivilities are not problematic in and of themselves but
rather indicate a lack of neighborhood social control. Wilson (1975) examined why
urban residents are often fearful for their safety. He suggested that it is not simply crimes
that residents find troubling. Beyond physical signs of disorder, Wilson (1975) found

that the daily hassles they are confronted with on the street, which include ‘street people,

37

panhandlers, rowdy youths, or 'hey honey' hassles’--and the deteriorated conditions that
surround them made residents weary and fearful.

Wilson and Kelling (1982) extended this line of thought yet another step by
modeling a temporal sequence. They found that deterioration led to cues about the lack
of social control and increased the likelihood of offenders coming into a neighborhood
and victimizing others. This, in turn, increases residential fear. They moved beyond fear
as an outcome to also explain resident-based informal social control on the street, the
vitality of street life itself, and increasing neighborhood crime rates. Wilson and Kelling
(1982) asserted that disorder reduced informal social control and thereby increased crime.

Skogan (1990) extended Wilson and Kelling’s (1982) findings by modeling
changes in neighborhood structure as the ultimate outcome of interest. Skogan (1990)
found that disorder was a signal of the breakdown of the local social order. He found
disorder eroded neighborhood control in terms of the capacity of residents to maintain
local events and conditions. This in turn drove out residents for whom stable community
life was important while at the same time discouraging people with similar values from
moving into the neighborhood.

Skogan (1990) classified disorder into two types: social and physical. Social
disorder was measured as public drinking, loitering (particularly groups of youths
“hanging out” in the street for no particular reason), harassment, noisy neighbors,
prostitution, and open-air drug markets. Physical disorder was measured as vandalism,
dilapidated housing (i.e., abandoned property), and the presence of trash and litter.
Skogan (1990) asserted that when present, these types of disorder were signs that nobody

in the neighborhood cared. The failure to take action encouraged deviants to advance

38

into the community in order to commit crime. Thus he concluded that the failure to
repair the disorder(s) set into motion a downward spiral of deterioration. As the small
disorders became more prominent, people in turn felt a heightened sense of fear of crime
victimization. This led to a collective withdrawal from the public sphere, and individuals
began to lose their sense of ‘tenitoriality’. At this point, territoriality only included their
individual households, and the informal social control at the neighborhood level
disappeared.

In sum, the theories discussed above progress from a sole focus on fear of crime
(Wilson, 1975; Garafolo and Laub, 1978), to concern about neighborhood street life and
crime (Wilson and Kelling, 1982) to neighborhood structural decline (Skogan, 1990),
also known as the “disorder—decline” hypothesis, to social-structural explanations of fear
that serve as a mediator of neighborhood crime (Gibson et al., 2002; McGarrell et al.,
1997). The consistent finding is that physical and social incivilities are often the
mediating factor in theories that explain fear and crime at the neighborhood level.

It must be noted, however, that some research has recently called into question
whether disorder is a mediating factor between neighborhood structural characteristics
and crime or whether disorder is a co-varying outcome along with crime. Sampson and
Raudenbush’s (1999) extremely rigorous neighborhood level study in Chicago essentially
showed that neighborhood incivilities, defined as systematic social observation (SSO)
disorder, often were explained by many of the structural features that explain crime,
which included some outcomes of predatory crime. Thus, they proposed that disorder
and crime are outcomes of the same social processes. They came to this conclusion

because they demonstrated that when collective efficacy was introduced into the models,

39

the’magnitude of incivilities as a correlate of crime (including homicide and burglary)
often reduced to the level of insignificance (Sampson and Raudenbush, 1999).

However, Sampson and Raudenbush found that disorder did significantly predict
robbery rates across census tracts, net of other structural measures including collective
efficacy. They concluded that areas with greater cues of disorder appear to be more
attractive targets for robbery offenders (Sampson and Raudenbush, 1999). They also
asserted that this finding is very consistent with Wright and Decker’s (1997) research
showing that robbery offenders are attuned to the local drug markets due to the likelihood
of prime targets with cash ‘on hand’ (Sampson and Raudenbush, 1999). Skogan (1987)
also found that disorder was related to robbery victimization.

Thus, the funneling effect described by the cone of resolution can also be seen
within the “disorder literature”. Neighborhood disorder may be a structural correlate of
specific types of offenses as was the case when explaining robberies (Sampson and
Raudenbush, 1999; Skogan, 1987). The importance of examining the relationship
between structural correlates and specific types of crime is established in the above
review on disorder. A more substantive assessment of the specific ecological correlates

of violence and homicide follows.

Level 2: Ecological Theories Specific to Violence and Homicide

Since Wolfgang’s (1958) seminal work on the distribution of homicides in
Philadelphia, criminologists have sought to explain the social-structural correlates of
violent crime. Wolfgang found that race, gender, and age were associated with homicide,
in particular certain types of homicide. He showed that the characteristics of the victim,

as well as the relationship between victim and offender explained some of the different

40

homicide patterns. Since this work, much has been written on explaining homicide
patterns in cities.

In terms of structural measures, Hsieh and Pugh (1993) found that poverty and
income inequality were each positively associated with violent crime. They found that
resource deprivation (measured as poverty and income inequality) appeared to be more
closely associated with homicide and assault than with robbery or rape. This is
suggestive that there is a need to analyze offenses by type because certain structural
factors may be better at explaining specific outcomes.

Often, the explanation of violent crime and homicide is directly associated with
the distribution of drugs in a neighborhood. Zimring and Hawkins (1997) hypothesized
that the illicit drug market association with violent crime is significant in some social
contexts and not others. They used this hypothesis to explain why there were no apparent
illicit drug markets (and relatively few homicides) in European industrial countries at the
time of their study. They made an important link between drug sales and homicides by
stating that drug sales and drug use could be a mediating, or facilitating, construct in
violent crime neighborhoods. That is, structural conditions may cause drug use, which in
turn may lead to drug markets that generate more serious, violent offending.

Rengert (1996) also discussed how the distribution of drug markets influenced
violent crime. More specifically, Rengert (1996) showed that neighbors tended to react
violently to drug dealers and drug users in their community. In response, drug dealers
tended to use violence to control citizens in the neighborhood, as well as their customers,

and also the competition (i.e., other drug dealers). Finally, Rengert showed that some

41

drug users tended to become violent as a result of taking drugs or as a means to obtain
money to purchase drugs (Rengert, 1996).

Within the violent crime context, the most obvious and arguably most studied
type of offense is homicide. One of the main reasons homicide is so intensely studied is
that this offense type is not as plagued by the ‘dark figure’ (i.e., unreported) of crime
problems that are found in other types of offenses. This is because when a honricide
occurs, there is a body or a missing person that is highly likely to come to the attention of
the police.

Homicides tend to concentrate in certain neighborhoods. As far back as the
1940’s, Bullock (1955) showed that 40 percent of all Houston homicides in 1945 to 1949
occurred within one city block of the offender’s residence and 74 percent occurred within
two miles. Pokomy (1965) found similar patterns in homicides in Houston in 195 8-1961.
Thus, homicides are also clustered in certain areas, and are not ‘randomly distributed’
across space.

The research on homicide is generally consistent with the studies reviewed earlier
on social disorganization, routine activities, and lifestyle theories. From a structural
perspective, Pridemore (2002) showed that there are generally three explanatory
approaches relating social structure to homicide: culture, strain (usually economic
deprivation), and social control. Cultural explanations of homicide have historically
tended to focus on values and beliefs of social groups (such as subculture of violence
theories). In terms of homicide, much of the cultural explanations have geared around
the ‘southern’ states accounting for more homicide than other states. Cultural

explanations have shown “mixed support” in explaining homicide, but Pridemore (2002)

42

argued they should not be ignored. Strain theories have tended to focus mostly on
economic deprivation (either absolute or relative depending on the study), while social
control models propose that the breakdown in social bonds decreases a community’s
ability to control its members, thereby freeing them to commit crime (Pridemore, 2002).

In terms of the relationship between race and violent crime at the neighborhood
level, Lauritsen and White (2001) proposed that neighborhood disadvantage has an
independent influence on an individual’s risk of violence. They found that the magnitude
of the neighborhood disadvantage coefficient was similar across each of the race and
ethnic gender subgroups (Blacks, whites, and Latinos) that lived in disadvantaged areas.
The authors concluded that living in disadvantaged areas made individuals more likely to
be the victim of stranger and non-stranger violence, particularly when comparing them to
counterparts in more advantaged communities.

Morenoff, Sampson, & Raudenbush (2001) found that concentrated disadvantage
(i.e., high strain) combined with low collective efficacy (i.e., low social control)
independently predicted homicide. They also found that local organizations, voluntary
associations, and friend/kinship networks promoted collective efficacy. And, homicide
events were not randomly distributed across a city. They concluded there is. a high
degree of overlap between the spatial distribution of collective efficacy and homicide.

Land, McCall, and Cohen (1990) found three robust and reliable main effects in
terms of explaining homicide patterns: resource deprivation (measured as an index),
population structure index (percentage divorced) and unemployment rate and percentage
of the population ages 15 to 29 (as a factor variable). Morenoff et al. (2001) also showed

that there is a spatial proximity effect when examining violence, collective efficacy and

43

alternative measures of neighborhood inequality. The ‘indices of concentrated extremes’
they used to capture these patterns emerged as the most consistent predictor of the
variation in homicide rates.

Thus, this section shows that certain structural factors have been consistently
correlated with macro-level homicide patterns, across a variety of studies. Measures of
economic disadvantage, or deprivation, unemployment, and percentage divorced have
been consistent neighborhood level correlates of violent crime and homicide. However, a
growing body of research has attempted to examine the relationship between structural
factors and a specific type of homicide, which is gang homicide. The next section

focuses on this limited, but gowing body of research.

Level 1: Ecological Theories Specific to Gang Homicide

 

Since Thrasher’s (1927) seminal research in Chicago that showed gangs were
more likely to commit crime in distinct geographic areas (i.e., known as “gangland”),
research examining the social and structural processes of gang crime has emerged.
Moving forward, longitudinal analyses revealed that the national homicide rate peaked
twice in the 20th century (National Research Council, 1993). The first peak occurred in
the early 1930’s and the rate then fell for the next 30 years. The national homicide rates
then began to increase in 1973 and peaked between 1979 and 1981 and again in the mid-
to late 1980’s.

The predominant change in homicide beginning in the mid-19803 has been
attributed to a dramatic growth in youth homicide that resulted from the recruitment of
youth into illicit drug markets (Blumstein, 1995). The characteristics of these criminal

offenses became much of the focal point in gang research. Blumstein (1995) showed that

44

because of the nature of the markets, participants felt the need to arm themselves for self-
protection and the resulting arms race among young people increased use of guns as
methods for dispute-resolution. Reiss (1993) showed that the increase of gun availability
and drug markets presented greater opportunities for violence. Specifically, in terms of
gun availability, firearms were involved in approximately 60 percent of all homicides.
Similarly, Fox and Zawitz (2002) also reported that in the late 1990’s, over 60 percent of
all homicides in the U. S. were committed with a firearm.

Zimring and Hawkins (1997) argued that the huge disparity between US.
homicide rates and other nations was largely attributable to the importance of guns as an
explanatory factor. They rejected the notion that race or the media coverage
(glamorization of violence) was associated with the large discrepancy between US. and
other nation’s homicide rates. Ousey and Lee (2002) revealed partial support for the
theory of systematic violence and contingent causation models. They found that between
1984-1997 the within city variation in illicit drug markets activity had a significant,
positive relationship with within city variations in rates of homicide offending. They
concluded that as illicit drug markets expanded, there was a corresponding increase in the
rate of lethal violence. In terms of structural measures, Ousey and Lee (2002) found that
the effect on homicide rates as well as the expansion in illicit drug market activity
depended on the neighborhood levels of resource deprivation. In cities where the pre-
existing level of resource deprivation was average or above average, illicit drug market
activity tended to have a significant positive impact on homicide rates. However, in
cities where the preexisting level of resource deprivation was below average, the drug

market homicide association became negative.

45

It can be seen then that the social explanation of homicide events became more
complex in the late 1980’s than before. Specifically, research showed an increase in
youth homicides involving firearms was linked to the rise in drug markets. Thus, several
studies have since examined the neighborhood-gang homicide relationship. The first of
these studies was an investigation conducted by Curry and Spergel (1988), where they
examined the gang and non-gang homicides that occurred in Chicago between 1978 and
1985. They found that gang homicides were more likely than non-gang homicides to
occur in socially disorganized areas. In particular, they concluded that economic
disadvantage, the percent of Hispanics living in a community, and poverty were
significantly related to gang homicides. This investigation became a focal point for
future research in this area. In particular, given the social changes that occurred in gang
homicides (Blumstein, 1995; Maxson, 1999) and the fact that Chicago was defined as a
chronic gang city (see Chapter 2 for more details), this conclusion needed to be validated
using a different time frame (i.e., more recent) and different cities (i.e., emerging gang
cities).

Rosenfeld et a1. (1999) made this type of contribution. They examined gang-
affiliated, gang-motivated, (see Chapter 2 for more details) and non-gang youth
homicides between 1990 and 1995 in St. Louis, Missouri, which was defined as an
emerging gang city (Spergel and Curry, 1993). In terms of structural measures,
Rosenfeld et al. (1999) examined the relationship between neighborhood disadvantage’,

neighborhood instabilitf, and race6 across sub-categories of homicide. Their models

 

’ Neighborhood disadvantage was measured as a combined score of poverty, public welfare assistance, and
female household items (Rosenfeld et al., 1999).

5 Instability was a combined measure of residential stability (length of homeownership more than 5 + years
and percentage of occupied housing units) transformed, multiplied by —1 (Rosenfeld et al., 1999).

46

showed that neighborhood disadvantage and poverty became insignificant predictors
when race (% African American) was included. In addition, residential instability was
more likely to predict non-gang homicides. They concluded that gang homicides were
highly concentrated in disadvantaged neighborhoods that were situated in predominantly
African American communities in St. Louis (Rosenfeld et al., 1999: 514). However, this
was also true of non-gang homicides. Concentrated disadvantage and racial isolation
explained homicide in general, but not across specific types of homicide (i.e., gang
homicide). It is important to note, the structural models employed in this study were
Maximum Likelihood Regression Models in order to distinguish the independent
neighborhood effects on homicide type. These models also included spatial weights to
control for observed spatial autocorrelation. In essence, the complex statistical models
employed were ‘neighborhood level’ models, which did not take into account incident
level measures (though their research found incident level distinctions in prior analyses).
In another St. Louis study, Kubrin and Wadsworth (2003) studied disaggregated
homicides that specifically involved Black perpetrators from 1985 to 1995. While race-
specific gang homicides were only one subtype of homicides included in this study, their
findings contributed to the body of research linking structural measures with gang
homicide. In particular, Kubrin and Wadsworth (2003) found a significant positive
relationship between Black disadvantage7 and Black gang homicides. In addition,

instability had a marginal (p < .10) negative relationship with Black gang homicides

 

6 Race was defined as percent African American. Some studies (Wilson, 1987) combine this measure with
neighborhood disadvantage to create a measure of economic deprivation, but the authors wanted to
examine its independent effect on gang homicide (Rosenfeld et al., 1999).

7Black instability was composite variable composed of percentage Black in poverty, Black per capita
income, percentage Blacks no working, percentage Black families that are female headed, and percentage
Blacks with no high school degree (Kubrin and Wadsworth, 2003: 17).

47

leading the authors to conclude that gang homicides involving Black perpetrators were
more likely to emerge in neighborhoods where the perpetrators had long-standing social
connections and social ties (Kubrin and Wadsworth, 2003).

The results from Kubrin and Wadsworth’s (2003) findings are difficult to
compare with other gang studies in this area because of three major reasons: first, they
limited their analysis to a race-specific subset that only involved Black perpetrators,
which makes generalization difficult. Second, the structural covariates here,
operationalized as factor scores, only included measures that captured the percentage
Black disadvantage, percentage Black residential instability, and Black population totals.
Third, the outcome variable (i.e., count of gang homicides) fits the gang-motivated
typology (Rosenfeld et al., 1999) because only homicides that were motivated from
disputes over the control of the neighborhood or drug markets were included in this
subgroup (Kubrin and Wadsworth, 2003: 11). These limitations are by no means a
criticism of this research because this study specifically set out to explain Black homicide
as a whole, of which Black gang-motivated homicide was but one subgroup. Kubrin and
Wadsworth’s (2003) findings are important because they lend empirical support to the
idea that there is a relationship between structural measures and homicide subgroups,
including Black gang homicides. From a structural perspective, subgroup homicide
analyses seem to yield divergent results, which continues to fuel interest and research in
this area.

Pizarro and McGloin (2006) extended this body of research by examining the
gang and non-gang homicides in Newark, New Jersey from January 1999 through July

2004. They included covariates at the incident level as well as macro-social level

48

measures. The incident level measure was an escalation measure (see Decker, 1996)
created fiom narratives in order to measure whether there was evidence that the homicide
occurred as a result of a ‘threat’ made against a person’s family, social group, or gang
where applicable. The macro-level variables included measures of poverty”, social
disorganization’, and percent African American. In their final statistical model, Pizarro
and McGloin (2006) combined the incident level variable (i.e., escalation measure) with
the three macro-level variables. Controlling for incident (a level 1 situational measure)
and percent African American (a level 2 structural measure) appeared to wash away the
effects of poverty. The social disorganization variable did not significantly differentiate
between gang and non-gang homicides in any model in this study. This research
supported Rosenfeld et al.’s (1999) findings that neighborhood disadvantage does not
significantly delineate gang homicides from non-gang homicides. In addition, this study
built upon Rosenfeld’s research by combining an incident level situational measure (i.e.,
escalation) with previously established macro-level correlates of homicide.

To summarize this section on Level 1 in the cone of resolution, the more
contemporary studies examining the ability of structural processes to delineate gang
homicides from non-gang homicides had similar findings: measures of neighborhood
poverty and instability did not delineate gang homicides from non-gang homicides when
neighborhood race and incident level measures were controlled. Both Rosenfeld et a1.

(1999) and Pizarro and McGloin (2006) concluded that the structural processes

 

8 Poverty was constructed as a corrrposite variable including the percent of families living in poverty,
percent unemployed, percent receiving public assistance, and percent of single-parent households (Pizarro
and McGloin, 2006).

9 Social disorganization was constructed as the percentage of residents who had lived less than five years in
the census tract, population size, and the number of different ethnic/racial groups residing in the tract
(Pizarro and McGloin, 2006).

49

measuring disadvantage explained gang homicides and non-gang homicides fairly
similarly. Homicides of all types tended to occur in disadvantaged neighborhoods. This
is particularly true when measures of race (percent African American) were incorporated
in the models (Curry and Spergel, 1988; Kubrin and Wadsworth, 2003; Pizarro and
McGloin, 2006; Rosenfeld et al., 1999). Combined, this research supports Wilson’s
(1987) and Sampson and Wilson’s (1995) propositions that crime occurs in areas where

economic deprivation is prominent.

Chapter Summary

This chapter reviewed the theories and empirical evidence relating to the
delineation of gang homicides from non-gang homicides using structural level measures.
These structural measures have been designed to capture what Rosenfeld et a1. (1999)
described as the social facilitation of gang violence. A consistent finding across research
has been that measures of neighborhood deprivation, as well as informal social control,
consistently correlate with crime. My strategy is to take advantage of these robust results
fi'om macro-level theories, but also use the cone of resolution as a framework to begin to
narrow the focus to determine the specific measures that predict gang homicide at the
structural level.

When reviewing the research examining the relationship between macro-level
ecological measures and gang homicide, two consistent findings emerge. First, there
does not appear to be any support that neighborhood deprivation, or social
disorganization, successfully delineate gang homicides from non-gang homicides. It

would appear that the current measures of neighborhood disadvantage do an equally good

50

job at predicting all types of homicide. Second, it becomes apparent that no research to
date has examined the relationship between disorder and gang homicide.

Research examining the relationship between disorder and gang homicide is
particularly important when considering the recent developments in the ecological
literature concerning gang “set space”. In particular, Tita et a1. (2005) examined the
structural features of communities that both facilitate and insulate gang activity. They
conducted interviews with gang members and found that individuals in gangs often chose
places to “hang out” in areas where diminished social control, such as the absence of
capable guardians and physical abandonment of place was prominent. Thus, Tita et al.
(2005) have demonstrated low informal social control mechanisms are important
structural dimensions when attempting to explain gang set space, or areas where gangs
hang out.

From a broken windows perspective, both fear (i.e., lack of trust among neighbors
and fear of victimization) and incivilities (i.e., physical signs of disorder) are indicators
of, and mediating factors in, the absence of informal social control mechaniSms. When
combined, incivilities and fear create a measure of neighborhood disorder (Sampson and
Raudenbush, 1999; Skogan, 1990). Sampson and Raudenbush’s (1999) extensive
analysis in Chicago assessed the ability of disorder to predict homicides (i.e., all
homicides), burglaries, and robberies at the neighborhood level. They found that disorder
did not significantly explain homicides (i.e., all homicides) or burglary rates in Chicago
neighborhoods when other important structural features were present in the model,

including measures of disorder and economic deprivation.

51

However, Sampson and Raudenbush (1999) did find that disorder was a
statistically significant covariate of robberies at the neighborhood level and asserted that
this finding was consistent with Wright and Decker’s ( 1997) prior research demonstrating
that armed robbers are attuned to the local drug markets due to the likelihood of prime
targets with cash ‘on hand’ (Sampson and Raudenbush, 1999: 630). Thus, research has
shown that disorder may have predictive power when attempting to explain predatory
crimes that are directly related to drug markets (Sampson and Raudenbush, 1999;
Skogan, 1987; Wright and Decker, 1997). This fits with the major tenants of broken
windows and “disorder-decline” theories, which posit that there is a relationship between
physical and social incivilities and low levels of informal social control (Skogan, 1990).
The theory also proposes that these incivilities cue offenders in ways that increase the
likelihood that those offenders will engage in criminal activity. While research has called
into question the temporal sequencing of this relationship (Sampson and Raudebush,
1999), the relationship still remains empirically strong.

As discussed earlier, a relationship between gang homicides and the drug market
has been suggested by Blumstein (1995) and supported by Ousey and Lee (2002). Given
that disorder (i.e., survey incivilities) is positively associated with robbery (Sampson and
Raudenbush, 1999), and robbery is positively associated with drug markets (Wright and
Decker, 1997), and drug markets are positively associated with gang homicides
(Blumstein, 1995), I theorize that there may be a more direct relationship between
disorder and gang homicide (see Figure 2). A test of the relationship between disorder

and gang homicide could contribute to the body of knowledge in this area.

52

Figure 2: Proposed theoretical model

Proposed Relationship

 

 

Gang

Drug <—-=I'——> Homicides
Markets

+ i

Robbery

Disorder

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Though disadvantage alone does not appear to explain gang homicides"), as
evidenced in prior research (Rosenfeld et al., 1999; Pizarro and McGloin, 2006),
Sampson and Raudenbush (1999) contend that disorder is explained by many of the
underlying social processes that explain crime in general, including economic
deprivation. Thus, any model attempting to assess the capacity of disorder to delineate
gang homicides fi'om non-gang homicides (i.e., the bottom of the cone of resolution)
should control for neighborhood level deprivation (i.e., a measure shown to be important
in the middle of the cone of resolution). Again, the analysis of this model is one of the

major goals of the current study.

 

'0 The lone exception in the literature is Kubrin and Wadsworth’s (2003) study that found disadvantage
significantly co-varied with the count of Black gang homicides. However, their results are difficult to
generalize with other gang research studies because both their independent and dependent variables were
race (i.e., Black) specific.

53

Chapter 4: Description of the Current Study

This study employs a theoretical framework, drawn from lifestyle, routine
activities, and social disorganization theory that examines the incident and structural
characteristics among homicide subtypes, specifically gang and non-gang homicides.
Using this fi'amework as a guide, this research has two main objectives. The first goal is
to assess the empirical congruity of the law enforcement designation of homicide as
gang-related with previously established indicators of gang homicide at the incident level,
based on prior gang research. The second goal is to determine the extent that measures of
neighborhood economic deprivation and disorder explain gang homicide, both
independent of, as well as in conjunction with, incident level measures. Both objectives

are discussed in detail.

Research Objective #1

This study looks at the gang/non-gang homicide classification from data in
Indianapolis where gangs were defined as loosely structured networks of known, chronic
offenders (McGarrell and Chermak, 2003). Based on previous research, incident level
measures should co-vary with the gang homicides because these incidents have been
shown to have different demographic, situational, and lifestyle characteristics than non-
gang homicides. Specifically, gang homicides are distinct from non-gang homicides in
that the offense is more likely to be carried out with a firearm, to be drug-related, and that
suspects are more likely to be non-white as well as younger (Decker and Curry, 2002).

A growing body of research from the gang literature shows not all measures of

gang incidents and structure are similar, particularly across cities. As discussed in

54

Chapter 2, Spergel and Curry (1993) distinguished cities as either chronic gang cities or
as emerging gang cities in terms of the timing and organization of the gangs in the cities.
In essence, only two cities fit the chronic gang city classification: Chicago (Block, 1996)
and Los Angeles (Maxson, 1999). The majority of the US. cities included in Spergel and
Curry’s study fit the ‘emerging gang cities’ typology.

A vast amount of the classic empirical research on gangs derived fi'om studies in
Los Angeles and Chicago (Decker and Curry, 2002), where many of the gangs were
classified as tightly structured and hierarchical (Skolnick, 1990). Consequently, many of
the underlying gang-related incident measures shown to predict gang homicide emerged
from research focused on chronic gang cities, which have qualitative differences in terms
of gang structure from other metropolitan areas (Klein, 1995). Emerging research has
shown that cities such as St. Louis (Decker et al., 1998) and Newark (McGloin, 2005)
have gangs that are more loosely structured.

This is important when examining the relationship between incident level
measures of demographic, situational, and lifestyle measures and homicide. It is
necessary for research in this area to examine whether these previously established
incident level correlates of gang homicide holds true in emerging gang cities where gang
behavior may not necessarily fit the previous recognized typology. In order for studies
that focus on gang behavior to be comparable, and generalizable, this classification must
be examined in an empirical manner.

This research will examine the statistical ability of demographic (age, race, and
gender of homicide participants), situational (firearm use, drug motivation, and the

number of suspects), and lifestyle measures (homicide participants’ prior drug and

55

weapons arrest histories) to predict the gang/non-gang homicide incident defined by law
enforcement agents in Indianapolis. This classification of gang homicide in the
Indianapolis data fits the gang-affiliated model, where a homicide is classified as a gang-
incident when at least one actor is part of a known gang networkl1 (Rosenfeld et al.,
1999)

This examination will serve several purposes. First, from a routine activities and
lifestyle theoretical perspective, it will test whether previously established incident level
correlates of gang homicide delineates those homicides in Indianapolis that were
classified as gang-related. As noted above, this is important because many of these
measures stem from research that focused in chronic gang cities (i.e., more highly
organized gangs), which is not seen in emerging gang cities, such as Indianapolis (i.e.,
more loosely structured gangs). In order to generalize findings across studies, this
empirical distinction should be consistent. Second, this research will enhance the
literature dealing with police decision-making. As Weisburd and Braga (2006) noted,
police decision-making and classification has traditionally followed a clinical model.
That is the case with the classification of gang homicides in Indianapolis. To the extent
that the classification of gang homicides are differentiated from non-gang homicides
based on variables drawn from lifestyle theory, then the clinical police designation may
have a theoretically grounded basis consistent with prior gang research. It will also help
identify factors that law enforcement agents look for in order to grasp, and operationalize,

their local gang-problem.

 

n The other classification is the ‘gang-motivated’ type (Rosenfeld et al., 1999), which is defined as the
incident being a confrontation resulting from gang-behavior rather than a violent incident that may or may
not be gang-initiated. Decker, et al. (1999) argued there is an inherent distinction between these two
offense classifications that should be defined before any generalizations can be made.

56

Research Objective #2

There exists a great body of research in the gang literature that explains gang
homicide at the incident level. However, there is a void in the research literature in terms
of distinguishing gang homicide from non—gang homicide at the macro-social level.
Researchers have attempted to fill this void by examining the effect that neighborhood
social disorganization has when examining the gang homicide/non-gang homicide
distinction (Curry and Spergel, 1988; Kubrin and Wadsworth, 2003; Pizarro and
McGloin, 2006; Rosenfeld et al., 1999). While research consistently shows that
neighborhood measures of social disorganization explain homicides in general, the
relationship between structural measures and gang-specific homicide is still vastly
understudied.

As seen in the review of the literature in Chapter 3, some studies have shown that
macro-level variables measuring neighborhood disadvantage and poverty significantly
explain gang homicide (Curry and Spergel, 1988; Kubrin and Wadsworth, 2003).
However, Curry and Spergel’s study examined the relationship between neighborhood
structure and gang homicide from data between 1975 and 1985, which was before the
significant change, and increase, that occurred in gang homicides due to the emergence of
drug and gun markets associated with crack-cocaine (Blumstein, 1995). In addition,
Kubrin and Wadsworth’s (2003) study only examined the effects of neighborhood
deprivation and social disorganization on race-related (i.e., Black) homicides. Thus,
these findings must be interpreted with caution. As discussed earlier, Rosenfeld et a1.

(1999) found that neighborhood disadvantage did not distinguish gang homicides from

57

non-gang homicides. In addition, Pizarro and McGloin (2006) found that neighborhood
social disorganization did not demarcate gang homicide from non-gang homicide, while
poverty did have a discriminating effect between these subgroups of homicide incidents.
Thus, the relationship between neighborhood disadvantage (i.e., economic deprivation)
and gang homicide appears to be mixed, at best.

In addition, there is little, to no, current research available that has examined the
relationship between neighborhood disorder and gang homicides. One of the most likely
reasons for this lack of inclusion of disorder in statistical models measuring is that these
data are not readily available at the macro-social level given the cost and difficulty in
obtaining survey data. Census measures of economic deprivation and concentrated
disadvantage are extremely important in that they are the most consistent and robust
indicators of social disorganization (Pratt and Cullen, 2005). However, ecological
research examining neighborhood disorder shows the importance of citizens’ perceptions
concerning the surrounding environment in terms of explaining local crime (Skogan,
1990)

From a broken windows perspective, the appearance of physical and social
disorder leads to a breakdown in social control, which promotes gang violence (Wilson
and Kelling, 1982). Both fear (i.e., lack of trust among neighbors and fear of
victimization) and incivilities (i.e., physical and social signs of disorder) are indicators of,
and mediating factors in neighborhood disorder. While Sampson and Raudenbush (1999)
found that disorder was not significantly associated with homicides (i.e., all homicides)
or burglaries, they did find disorder significantly co-varied with robbery rates at the

neighborhood level. They concluded this finding was consistent with prior robbery

58

research (see Wright and Decker, 1997) that armed robbers are attuned to the local drug
markets due to the likelihood of prime targets with cash ‘on hand’ (Sampson and
Raudenbush, 1999: 630).

Given the established relationship between gang homicides and the drug market
(Blumstein, 1995; Ousey and Lee, 2002), it is important to examine whether
neighborhoods exhibiting higher disorder are at a greater risk for gang homicide. A test
of the relationship between disorder and gang homicide will contribute to the body of
knowledge in this area. This type of examination is consistent with the major tenants of
broken windows and disorder-decline theories, which posit that there is a relationship
between the physical and social cues of a neighborhood, marked by low informal social
control, and the willingness of offenders to commit street crime in these areas (Skogan,
1990)

This research will examine the statistical ability of concentrated disadvantage (a
factor score obtained through Census data) and disorder (an aggregated citizen survey
factor score) to distinguish gang homicides from non-gang homicides, both
independently and in conjunction with important incident level measures. Models that
examine multi-level data, both at the incident and structural levels, have the potential to
contribute to the research that attempts to delineate gang homicides from non-gang
homicides. In addition, it may be possible to identify important factors that correspond
with gang homicides that have yet to be uncovered.

Although this second stage of the analysis is primarily a test of the utility of
theoretically driven neighborhood factors to account for gang crime, the analysis also

contributes to the assessment of police classification of gang and non-gang homicides.

59

Despite the fact that the findings of this impact of neighborhood level indicators on
gang/non-gang homicide in prior research are limited, theory would predict that both
individual lifestyle indicators and neighborhood level characteristics should relate to gang
homicides. A finding of no association between neighborhood characteristics and gang
homicides would not “disprove” the validity of the police classification. It would,
however, raise questions as to whether the police classification has a theoretical basis or
whether it is simply capturing clinical judgments based on individual characteristics of

the incident (Weisburd and Braga, 2006).

Research Questions

As noted above, this study will focus on two central research questions:

1) Are there demographic, situational, and lifestyle distinctions between homicides
that were classified as gang-related (i.e., homicides involving known groups of
chronic offenders) from those that were classified as non-gang—related?

2) Do structural measures of the neighborhoods where the incident occurred

delineate gang-related homicides from non-gang-related homicides?

Research Hypotheses

1) Incidents classified as gang-related homicides will involve multiple, younger,
non-white, males with more extensive criminal histories than will non-gang-
related homicides. In addition, drug motivation and firearm use will increase the

likelihood that homicides were classified as gang-related.

60

2) Measures of disorder and economic deprivation will be more predictive of

incidents classified as gang homicides compared to non-gang-related homicides.

Sources of Data

The data used in this study come from three primary sources: local incident level
data, POPN, and the US. Census. The first source consists of incident level data that
were collected at bi-monthly homicide incident reviews, which took place at the
Indianapolis Violence Reduction Partnership (IVRP) meetings. The IVRP was created as
a strategic problem solving initiative to combat homicide, and thus data were used as part
of the strategic process (McGarrell and Chermak, 2003). Second, POPN data (Mastrofski
and Parks, 1990) were used to create a structural level measure (i.e., neighborhood
disorder). These surveys were administered to citizens residing in police beats in
Indianapolis to gauge perceptions of the local social and physical neighborhood
characteristics. The third data source is from the 2000 US. Census, which is used to
create the economic deprivation measure (see Sampson and Raudenbush, 1999; Sampson
and Wilson, 1995; Wilson, 1987). A more in-depth review of these three sources of data

follows.

IVRP Data

The incident data derive from the IVRP, which was a multi-agency, collaborative
effort to reduce homicide and serious violence in Indianapolis, Indiana (McGarrell and
Chermak, 2003). The IVRP was part of the US. Department of Justice’s Strategic

Approaches to Community Safety Initiative (SACSI), of which Indianapolis was one of

61

the six original jurisdictions”. IVRP was based on a data-driven, ‘strategic problem
solving approach’ where the local firearms violence problem was analyzed, leading to
strategic interventions that were based on the problem analysis (McGarrell et al., 2006).
Included in this multi-agency collaboration was the Office of the Mayor, the Indianapolis
Police Department, Marion County’s Sheriff and Prosecutor’s Office, the Superior Court
Criminal and Juvenile Division, Office of Probation and Parole, the Indiana State Police,
Bureau of Alcohol, Tobacco, F irearrns and Explosives, Federal Bureau of Investigation,
US. Attorney’s Office, Drug Enforcement Administration, and US. Marshal’s Service.
In addition to these agencies that represented the practitioners of the criminal justice
system, researchers from Hudson Institute and Indiana University contributed to the task
force by providing feedback to the team, while simultaneously collecting process and
outcome measures for the evaluation component of the IVRP (McGarrell and Chermak,
2003)

As part of their strategic agenda, the task force conducted bi-monthly homicide
incident reviews. This process involved a case-by-case review of homicide incidents by
teams of detectives, street level officers, prosecutors, probation and parole officers, and
other criminal justice personnel. Researchers coded detailed information about the
homicides based on the review by criminal justice officials. Two distinct steps were
employed by the research team to ensure data reliability. First, data collection protocols
along with a data collection instrument (i.e., an incident review sheet) were used to

capture important homicide information, including information about the incident, as well

 

"The original six sites included Indianapolis, New Haven, Portland, Memphis, Winston-Salem, and
Rochester. Since then, a four other sites (i.e., St. Louis, Detroit, Albuquerque, and Atlanta) implemented
the SACSI model as a means to reduce violence, gun crime, and homicide. These studies served as the
foundation for the Project Safe Neighborhoods (PSN) initiative, a national firearm and violence crime
reduction strategy (www.psn. gov).

62

as the actors (i.e., victims and suspects) involved. Second, reliability checks were
completed after data collection to make sure there was consistent inter-coder reliability.
The working relationship between Hudson Institute/Indiana University researchers and
the IVRP working group provided a setting where rich information was obtained about
the homicide incidents (McGarrell and Chermak, 2003; McGarrell et al., 2006)

The data employed in the present study focused on the 563 homicides that
occurred between January 1, 1997 and June 30, 2001. Of the 563 incidents 155 occurred
in 1997, 146 in 1998, 110 in 1999, 102 in 2000 and 50 item January 1, 2001 to June 30,
2001. The following information was coded at the incident reviews: the incident
location, the names and demographic information of the actors (i.e., both the victim and
all suspects) involved, the motive of the offense, the method of death, and whether the
incident involved individuals who were associated with “a group of known, chronic
offenders” (McGarrell and Chermak, 2003: 61). The working group used the term ‘group
of known, chronic offenders’ to capture the chronic, persistent offending that appeared to
be the pattern of gang-members in Indianapolis.13 More specifically, consensus emerged
that the term ‘group of known, chronic offenders’ captured both structured gangs as well
as loosely affiliated gangs and neighborhood crews. This operational definition was
explicitly chosen to avoid controversies over local definitions of which groups
constituted formal gangs with a hierarchical structure and extra-local connections and
which groups constituted local gangs with less structure and hierarchy.

In addition to the information made available at the incident reviews, researchers

conducted an additional step and ran criminal histories on the victims and suspects

 

’3 By using this definition, the IVRP working group attempted to capture the irrrpact that a ‘group’ has on
individual actors, across victims and suspects, since it is well established that individuals tend to violate the
law in groups (Hindelang, 1971; Reiss, 1986).

63

involved in the homicides. This was achieved by using the offender history database
housed in the Indianapolis Police Department (IPD). The criminal histories are specific
to Marion County, Indiana. The variables obtained at the IVRP meetings make-up the

Level-1 incident measures.

Project on Policing (POPN) SurveyData Description

In 1996, the Project on Policing Neighborhoods (POPN) study was carried out in
both Indianapolis, Indiana and St. Petersburg, Florida. This multi-agency study was
designed to assess and strengthen the relationship between policing agencies and citizens
(Mastrofski and Parks, 1990). One of the components of this study was the collection of
citizen surveys at the neighborhood (i.e., police beat) level in order to provide appropriate
context for the types of different neighborhoods that exist in the city.14 Indiana
University’s Center for Survey Research (CSR) administered commrmity surveys to
residents across the fifty police beats in Indianapolis.

The ideal sampling fi'ame called for completion of 100 surveys in each IPD police
beat, from which at least 80 or more completions were obtained in 34 of the 50 beats. In
total, 5,389 citizens were surveyed in Indianapolis, equating to a 53 percent completion
rate of the households in the sample frame (Reisig and Parks, 2004). Reisig and Parks
also stated that refusals after at least two attempts in order to reverse a respondents initial
refusal accounted for 31 percent, while 16 percent were persistently unavailable after at
least eight callbacks at various times of the day and evening (Reisig and Parks, 2004).

Interviews followed a structured format. Most questions provided a set of response

 

" It is important to note that Taylor (1996) posited that strategies to reduce crime must be at the street
block level or smaller in order to facilitate a response from local residents. Taylor (1996) specifically
argued that fear and physical deterioration can be reduced by strengthening formal social control

mechanisms at the street block level, and that a neighborhood must be defined at this level of analysis.

64

options from which respondents were instructed to choose the one that most closely
corresponds to their experience or view. While no individual identifiers were available in
these data, the police beat of the respondent was coded making it possible to aggregate
responses to the neighborhood level.

Previous researchers have used these citizen surveys to construct quality of life
measures at the neighborhood level (i.e., police beat) by aggregating citizen responses to
select questions (Reisig and Parks, 2004). As an example, Reisig and Parks (2004)
created an index of physical and social incivilities by aggregating individual responses to
six survey items to the neighborhood level. These items reflected citizen’s judgments on
the neighborhood problems of litter/trash, loitering, vandalism, gangs, abandoned
buildings, and drug dealing. Responses to these items were based on a three-point
ordinal scale ranging from no problem to a major problem.

Thus, the POPN citizen surveys have previously been aggregated to construct
neighborhood level items in research, which is pivotal for the type of data required to
examine Research Question #2 (i.e., structural effects on gang homicides) in this study.
Specifically, this study seeks to examine the relationship between neighborhood disorder
and gang homicide. However, the aggregated citizen surveys alone do not capture many
of the important macro-level theoretical constructs that represent socially disorganized

neighborhoods. Therefore, US. Census data are utilized to complement the survey data.

US. Census Data
This study relies on data collected by the US. Census Bureau in order to capture
structural measures of the different neighborhoods in Indianapolis. Census data are

available on-line from the Social Assets and Vulnerability Indicators (SAVI) project

65

housed at Indiana University-Purdue University at Indianapolis (IUPUI). The SAVI
project is designed to better inform policy-makers of the various types of neighborhoods
in Indianapolis by making demographic, economic, and housing data public.

These data represent the latest Census conducted in 2000, which are available at
the Census block level. Many of the measures have been used in prior research to create
an index of structural disadvantage. For example, Reisig and Parks (2004) previously
created a structural disadvantage index by computing a factor score, using principle
component analysis, using the following items in Indianapolis: percentage of people
living in poverty, percentage of labor force unemployed, percentage of female-headed
families (as a measure for family disruption), and percentage of population that is Afiican
American.

In order to maintain consistency at the structural level, the data sources had to be
aggregated into the same unit of analysis, which were the fifty IPD police beats in
Indianapolis. This unit was chosen because the POPN data were already aggregated to
the police beat level, and it was not possible to disaggregate these data smaller than the
beat level. In terms of the Census block measures, data consistency was achieved by
aggregating each Census block, and its respective measure15 up to the police beat level.
For those Census blocks that were completely housed within a single police beat, the
aggregation was straightforward in that the total value for each measure at the block level
was included into the beat’s aggregation. However, for those Census blocks that were

housed within two or more police beats, the value of the Census block given to a

 

'5 These measures are firrther discussed in Chapter 5.

66

particular police heat was equal to the proportion (i.e., area)16 that the Census block was
encompassed within that beat.17 Thus, if a Census block was housed in two different
police beats, the value of the Census measure of that block was divided among the beats
it was housed within (see Figure 3 for a visual display). In the Figure 3 example, the
value of the Census block measure (e. g., number of people living in poverty) would be

.50 for IPD Beat #1 and .50 for IPD Beat #2.

Figure 3: Example of census block aggregation to police beat level

 

IPD Beat #1 Census IPD Beat #2
Block

 

 

 

 

 

 

Using police beats as surrogates for neighborhoods in order to capture contextual
effects is not unprecedented in criminal justice research, as a number of studies have
employed this practice (see Chermak, McGarrell, and Gruenewald, 2006; Reisig,

. McClusky, and Mastrofski, 1999; Reisig and Parks, 2004; Skogan, Hamett, DuBois,
Corney, Kaiser, and Loving, 1999). In terms of the specifies, the average number of
block groups aggregated to the IPD beat level was 7.78, the median was 7.37, the mode
was 7.0, and the standard deviation was 2.98. This means that, on average, an IPD beat
was roughly equal to 7.78 Census blocks, and that roughly 68 percent of the IPD beats

housed between 4.8 to 10.76 Census blocks.

 

’6 The area available for each Census block was compared against its own area actually housed in each
distinct police beat. Most of the shapes (i.e., beats and Census blocks) were either squares or rectangles.
Area was calculated as base x height for squares and rectangles, and '/2 base x height for triangular shaped
units (Bradford, 1987). Integration techniques were used where the shape of the perspective police beat did
not fit conform to the regular pattern.

'7 The proportion that each census block was housed in a particular police beat was rounded at cut-points
.25, .50, and .75 and 1.00.

67

A far more common strategy has been the reliance of US. Census tract data as
surrogates for neighborhoods (see Curry and Spergel, 1988; Kubrin and Wadsworth,
2003; Pizarro and McGloin, 2006; Rosenfeld et al., 1999; Sampson and Groves, 1989;
Sampson and Raudenbush, 1999). In order to provide a contextual comparison across
these different neighborhood constructs, this study utilized the clip function found in the
Geoprocessing Wizard in Arcview GIS (version 3.2). By using IPD police beats as the
merging unit, it showed that there were 164 Census Tracts and 539 Census blocks in the
same area. This means, on average, there were 3.2 Census blocks per police beat. Thus,
the IPD beats, which are the neighborhood surrogates used in this study, are roughly two
to three times larger than are the US. Census tracts in Indianapolis. While prior research
has used police beats as neighborhood surrogates, it is important to remember that the
neighborhood boundaries in this study are larger than in a majority of macro-level

studies.

Chapter Summary

This chapter describes the objectives of the current study, including the specific
research questions and hypotheses. As described earlier, the purpose this dissertation is
to delineate gang homicides from non-gang homicides using both incident (i.e., level-1)
and structural level measures (i.e., level-2). Data from the IVRP, POPN, and US.
Census are used in order to achieve this goal. The level-1 measures are the situational,
demographic, and lifestyle variables that were obtained in the IVRP. These items will be
included as level-1 measures in subsequent statistical analyses. The level-2 measures

come from the POPN citizen surveys and US. Census. While the unit of analysis is

68

different across these data sources, this study aggregated the census data to the IPD
police beat level in order to provide consistent structural measures. The next chapter
provides specific detail concerning operationalization of the dependent and independent

variables used in this study.

69

Chapter 5: Data Description

This study is based on a cross-sectional design, which relies on data from the
IV RP incident reviews, IPD criminal history database, survey questionnaires acquired
through the POPN study, and US. Census data. Thus, the unit of analysis for the two
research questions in the current study is the homicide incident. Although experimental
and quasi-experimental designs are often preferred methods for uncovering causation,
such designs are not always possible (Cook and Campbell, 197 9). The use of the data
sources employed here is consistent with prior research attempting to uncover the
situational, lifestyle, demographic, and structural dimensions that delineate gang
homicide from non-gang homicide (see Curry and Spergel, 1988; Kubrin and
Wadsworth, 2003; Pizarro and McGloin, 2006; Rosenfeld et al., 1999). This chapter
details the operationalization and distribution of the dependent and independent

variable(s) used in the current study.

Dependent Variable

The IVRP working group classified a gang homicide as any incident where either
the victim or suspect(s) were part of a group of known, chronic offenders (McGarrell and
Chermak, 2003). This classification is similar to prior gang research that classified “gang
affiliate ” homicides as incidents where the incidents themselves were not necessarily a
product of gang related activity, but simply involved suspected gang members (Maxson
and Klein, 1990). In terms of the coding of these offenses, the research team in

Indianapolis only coded an offense as gang related if there was confirmatory evidence

70

(e. g., if an actor was a confirmed gang member or the group described was part of a drug-
selling organization) or if at least two review participants independently provided
information indicating prior gang involvement (McGarrell and Chermak, 2003: 64). In
this case, a gang homicide refers to any homicide that involved either a victim or suspect
who was involved in a gang, or was part of a known periphery gang network (i.e., known
associates). In essence, this is a more ‘loose’ classification, which is often referred to as
the “Los Angeles” definition of gang homicides (Maxson and Klein, 1990).

In terms of the breakdown of gang member profile by actors, Table 1 displays the
distribution of gang homicides between January 1997 and June 2001.18 The table shows
that 48.6 percent of suspects were classified as belonging to a known group of chronic
offenders (i.e., gang homicide). In addition, 37.3 percent of the victims were classified as
belonging to a known group of chronic violent offenders.19

Table 1: Actor distribution across gang and non-gang homicides in Indianapolis

 

 

Victim information Suspect information
N °_/c_» N 2/2
Victim was a 210 37.3 One or more suspects 333 48.6
gang member was a gang member
Victim was not 353 62.7 No suspect was a gang 352 51.4
a gang member member
Total 563 100 Total 685 100

 

 

'8 One of the potential threats to the reliability of the gang homicide measure is that law enforcement agents
would be more likely to define homicides as gang related as they became more comfortable with the
classification procedure over time. A logistic regression model (0 = non-gang homicide, 1 = gang
homicide) was estimated including year as an interval independent measure. This rival hypothesis was not
supported in the model. Actually, homicides were less likely (i.e., negative estimate) to be classified as
gang homicides over time. This is supported by the intervention analysis conducted by McGarrell et a1.
(2006).

'9 Stockton, CA also relied on a similar law enforcement designation based upon homicide incident reviews
(Braga, 2008), which is very consistent with the approach used in Indianapolis. Between 1997-1999,
Stockton law enforcerrrent officials classified fifty-three percent of homicide victims and sixty-one percent
of homicide suspects as gang members, which is very similar to the distribution in Indianapolis (seen in
Table 1).

71

In terms of the incident, which is the unit of analysis in this study, Table 2
displays the distribution of gang homicides that occurred in Indianapolis between January
1, 1997 and June 30, 2001. Again, gang homicides included those homicides where
either the victim or the suspect was involved with a known group of gang offenders (i.e.,
gang affiliated homicide). Slight majorities of the total homicide number were classified
as non-gang offenses (51.9 percent) compared with gang homicides (48.1 percent).20

Table 2: Total number of homicides in Indianapolis classified by type

 

 

Homicide type N %

Gang Homicides 271 48.1
Non-Gang Homicides 292 51.9
Total 563 100

 

Independent Variables

As mentioned earlier, this study employs four theoretically based sets of groups of
measures in order to distinguish gang homicides from non-gang homicides: one that
examines the situational characteristics of the incident (i.e., situational measures); one
that examines demographic characteristics of the actors involved (i.e., demographic
measures); one that examines the prior drug and arrest histories of the actors involved
(i.e., lifestyle measures); and one that examines the structural characteristics where the

homicide took place (i.e., structural measures).

 

2° In order to assess the ‘face validity’ of the gang/non-gang homicide classification in Indianapolis, gang
homicide rates in other cities were compared. Since this study relied on the Los Angeles definition of gang
homicide, it was important to compare the proportion of gang homicides observed in Indianapolis with Los
Angeles. In 1994, nearly 45 percent of all homicides in Los Angeles were gang related between 1994-1995
(Maxson, 1999). In Newark, NJ homicides were gang related in 40 percent of the cases between 1999-
2004. Thus, the proportion of gang related homicides in Indianapolis (51.9 percent) are consistent with
prior gang research.

72

Situational Measures

There are three situational variables that are used in this study: the incident
number of suspects involved in the incident, the motive as drug related, and the method
of death as firearm related. The incident number of suspects is a continuous variable that
measures the number of suspects identified in each homicide incident. A homicide
incident is unique in that the presence of a single victim constitutes a new incident. Thus,
the number of suspects in each incident is always proportional to one homicide victim.

Of the five hundred sixty three homicides that occurred in Indianapolis, eighty-
four (14.9 percent) involved multiple suspects. Table 3 displays the measures of central
tendency and dispersion for the different homicide types in Indianapolis. The average
number of suspects in gang homicides was 1.35, while the average (mean) number of
suspects in non-gang homicides was 1.10. The standard deviation for the number of
suspects in gang homicides was .83 and .37 in non-gang homicides. Thus, gang
homicides averaged a higher number of suspects and also had more variability in the
number of suspects per homicide incident than non-gang homicides.

Table 3: Distribution of multiple suspects involved in incidents across homicide
subtypes

 

 

Homicide type Mean Median Mode SD
Gang Homicides (n = 291) 1.35 1.0 1.0 .83
Non-Gang Homicides (n = 264) 1.10 1.0 1.0 .37

 

Another situational independent variable is the homicide motive. The motive was
coded fiom the IPD’s homicide incident sheet as well as the bi-monthly incident reviews.
The code sheet had five distinct categories from which investigators and researchers had

to select in order to classify the offense: drug related, street fight/dispute, domestic,

73

robbery, or other. Table 4 displays differences in homicide type by the motive. Two
hundred fifty five of the five hundred sixty three homicides (45.2 percent) were excluded
from this descriptive analysis due to missing data. Given there were two stages to this
coding process (i.e., the bi-monthly incident review meetings and the incident sheet
coded by law enforcement at IPD), missing data in these cases were most likely due to
unknown motives.

This presence of missing data for this variable appears to be relatively similar
across the different homicide subtypes because the proportion of the total offenses was
somewhat similar for gang homicides (53.9 percent) as it was for non-gang homicides
(57.5 percent). The most probable known motive for gang homicides was a street
dispute/fight (44.1 percent), compared with non-gang homicides where a domestic
situation (44.9 percent) was the most common motive. Robberies were the second most
likely motive for gang homicides (20.4 percent), compared with street disputes/ fights in
non-gang homicides (31.4 percent).

Table 4: Distribution of the incident motive across honricide subtypes

 

 

Motive Gang homicides Non-gang homicides
E 3/_o H %
Drug 30 10.3 10 6.4
Street Disputes/F i ghts 67 44.1 49 31.4
Domestic 19 12.5 70 44.9
Robbery 31 20.4 22 14.1
Other 5 3.3 5 3.2
Total 152 100 156 100

 

However, this variable warranted recoding based on prior gang research. The
participation of gang members in illegal drug sales is well established across a variety of

studies (Decker and Van Winkle, 1996; Fagan, 1989, Maxson et al., 1985; Vigil, 1988).

74

Maxson and Klein (1990) also demonstrated that gang homicides were more likely to be
drug-motivated. Thus, this nominal level variable was recoded into a dummy variable
that captured a drug-motivated incident (0 if the incident was not drug motivated, 1 if the
incident was drug motivated). In addition, the missing motive data were included in the
reference group because this variable was specifically designed to capture homicides that
were drug motivated. Given that the distribution of unknown motives were similar across
gang and non-gang homicides, the possibility of creating a bias in the delineation of gang
and non-gang homicides appear to be minimal. Table 5 shows the distribution of drug-
motivated incidents across the different homicide types in Indianapolis. Gang homicides
were more likely to be drug motivated (10.3 percent) than were non-gang homicides (3.7
percent).

Table 5: Distribution of the drug-related incidents across homicide subtypes

 

Drug-related Gang homicides Non-gang homicides

 

N. % E °_/2
No 262 89.7 261 96.3
Yes 30 10.3 10 3.7
Total 292 100 271 100

 

The final situational variable is the method of death. Table 6 displays the method
of death distribution for the two homicide subtypes in Indianapolis. Ten homicides were
excluded from this analysis due to missing data. The most obvious difference is that
gang homicides were much more likely to involve the use of a firearm (87.9 percent) than
non-gang homicides (59.1 percent). In addition, the method of death in non-gang
homicides was asphyxiation (10.2 percent of the time), blunt force (12.9 percent) and
stabbing (14 percent). These methods were much less common in gang homicides (2.4

percent for asphyxiation, 4.5 percent for blunt force, and 4.8 percent for stabbing).

75

 

Table 6: Distribution of the method of death across homicide subtypes

 

 

Motive Gang homicides Non-gang homicides '
N. i’/_o .1! %
Arson 0 0.0 2 0.8
Asphyxia 7 2.4 27 10.2
Blunt Force 13 4.5 34 12.9
Gunshot 254 87.9 156 59.1
Stabbing 14 4.8 37 14.0
Other 1 0.3 8 3.0
Total 289 100 264 100

 

The method of death measure was recoded into a dummy variable that captured
firearm use (0 if the method of death was not firearm related, 1 if the method of death
was due to a firearm) based upon prior research that has shown gang homicides are more
likely to involve firearms (Maxson and Klein, 1990, 1996; Maxson etal., 1985). The
incidents where the method of death was unknown were included in the reference group
due to the low number of missing cases, and the likelihood that the incident being firearm
related would be easier to distinguish given the clues that accompany such an offense
(e. g., appearance of gunshot wounds on the victim). Table 7 displays the distribution of
firearm related offenses across homicide subtypes. Gang homicides were more likely to
involve firearm use in the method of death (86.9 percent) than were non-gang homicides
(57.6 percent).

Table 7: Distribution of firearm related offenses across homicide subtypes

 

 

Firearm-related Gang homicides Non—gang homicides
E a , N a
No 38 13.1 115 42.4
Yes 254 86.9 156 57.6
Total 292 1 00 271 100

 

76

Demographic Measures

The demographic measures of interest in this study include the actors’ (i.e., both
victims and suspects) age, race, and gender. As a first step, the description of these
measures is at the actor-level, which means the descriptive demographic data are
partitioned across victims and suspects. It is necessary to examine the demographic
distribution across the different actors because there are less missing data for victims than
for the suspects. The victim was unknown (i.e., unidentified) in only one case, though
the missing data varies by the age, race and gender variables. Comparatively, of the six
hundred eighty five total suspected offenders, one hundred eighty five were classified as
having unknown suspects. This means that at least 27 percent of the suspect
demographic information was missing. Thus, examining these descriptive statistics
across the different actors is an imperative first step in order to control for missing data,
particularly those data pertaining to suspects. As a second step, the demographic
information for both victims and suspects will be combined to create joint-demographic
measures at the incident level, which is consistent with prior research (see Decker, 1996).
These joint-demographic measures at the incident level will be used in the multivariate
analyses in the subsequent chapter.

The actors’ age, race and gender were coded at the IVRP bi-monthly incident
review meetings. Table 8 displays the descriptive statistics for the age, race, and gender
measures of the actors involved in both gang and non-gang hOmicides in Indianapolis.
The measures of central tendency are displayed for the age of the actors involved in the
different types of homicides in Indianapolis. Gang victims averaged 28.1 years of age,

while non-gang victims averaged 34.3 years of age. Suspects in gang homicides

77

averaged 24.0 years of age, while non-gang homicide suspects averaged 29.9 years of
age.

Table 8 also displays the percent of non-white21 actors involved in the different
homicide subtypes. In gang homicides the victim was non-white 83.9 percent of the
time, compared with non-gang homicides where 59.0 percent of the cases involved a non-
white victim. In gang homicides, the suspects were non-white 59.6 percent of the time,
while suspects in non-gang homicides were non-white 43.3 percent of the time. In
addition, victims were male in gang homicides 88.7 percent of the time. Comparatively,
in 68.8 percent of the non-gang homicides the victim was male. Finally, males were
suspects of gang homicides 95.9 percent of the time, while males were the suspects in
84.2 percent of the cases in non-gang homicides.

Table 8: Descriptive demographic measures across homicide subtypes

 

 

Gang homicide Non-gang homicide
Age N Mean Median S_D N Mean Megan; S_D
Victims 290 28.1 25 11.3 260 34.3 32 18.3
Suspects 279 24.0 22 8.8 213 29.9 26 13.5
N on-white N Percent N Percent
Victims 290 83.9 251 59.0
Suspects 266 59.6 197 43.3
Male N Percent N Percent
Victims 291 88.7 263 68.8
Suspects 267 95.9 203 84.2

 

 

2‘ The original race categories were white, African American, Hispanic, and other. The use of non-white as
a dummy variable serves both theoretical and empirical purposes. First, prior gang research has shown that
African American communities (Kubrin and Wadsworth, 2003) and Hispanic communities (Curry and
Spergel, 1988) are at greater risk for gang activity. Second, Hispanics made up a very small percentage of
both victims and suspects in Indianapolis homicides. Specifically where the race of the actor was known,
Hispanics made up less than 2.9 percent (29 of the 1,027 known victims and suspects combined) of the
cases. Thus, for theoretical and empirical clarity, Hispanics and African American actors were collapsed
into a non-white category.

78

In order to combine these measures into joint-demographic variables, this study
relied on prior research to serve as a guide in terms of the operationalization of the
measures. Prior research has demonstrated that intra-racial homicides are distinct across
racial subgroups (see Decker, 1993; Decker, 1996; Monis, 1981). As an example in a St.
Louis study, Decker (1993) showed that while most homicides were intra-racial, Black-
on-Black homicides were more likely to involve actors with ‘distant relationships’ (e. g.,
acquaintances) while white-on-white homicides were more likely to stem from more
intense relationships, such as romantic links (Decker, 1993: 601). This pattern of
homicide across relationship types is relevant when attempting to delineate gang
homicides from non-gang homicides. Prior research has also shown that non-whites are
at greater risk for gang activity (see Curry and Spergel, 1988; Kubrin and Wadsworth,
2003). Thus, the joint-demographic variable used in this study is labeled non-white intra-
racial homicide, which captures non-white on non-white homicide incidents.

Non-white intra-racial homicide is a dummy coded variable (0 or 1) where the
value is zero for the reference group and one for any homicide where both the victim and
the suspect are non-white. For those homicides involving multiple suspects, only
incidents where every participant is non-white is coded as a non-white intra-racial
homicide. Where the suspect information was known, inter-racial suspect affiliation was
an issue in less than 2.4 percent of the total cases (N = 8). Thus, the reference category
includes white-on—white homicide as well as inter-racial homicide. Table 9 displays the
joint-demographic measure of race across gang and non-gang homicides. Non-white
infra-racial homicides comprised 78.3 percent of gang homicides, compared to 57.7

percent of non—gang homicides.

79

In terms of a joint-demographic variable for gender, this study uses male-on-male
as a measure. Given that gang homicides are more likely to involve male offenders
(Maxson et al., 1985), this measure is more likely to capture this established
phenomenon. The male-on-male incident variable is dummy coded (0 or 1) where the
value is zero for the reference group and one for any homicide where both the victim and
suspect are male. For those homicides involving multiple suspects, only incidents where
every participant is male is coded as a male-on-male homicide. Inter-gender suspect
affiliation among multiple suspects was an issue in 3.4 percent of the cases (N=12).
Table 9 shows that male-on-male homicides comprised 81.4 percent of gang homicides.
Conversely, 50.8 percent of the non-gang homicides were male-on-male incidents.

Finally, a joint-demographic measure measuring the age of the actors is defined as
the combined mean age of the victim and suspect(s) involved. Again, prior research
shows that gang homicides are more likely to involve younger actors (see Maxson et al.,
1985). The combined mean age variable is a continuous variable that is measured as the
total age of the victim and suspects (averaged) involved. In terms of the average, each
incident only has one victim, and may or may not involve multiple suspects. For the 85.1
percent of the incidents where there is only one known suspect (N = 479), the average
age is equal to the known suspect’s true age. For the 14.9 percent of the cases where
there were multiple suspects (N =84), the age variable here becomes the average age for
the total number of suspects involved in each incident. Table 9 shows that the combined
age of gang homicides was 51.80 years, or a mean age of 25.9 years for both victims and
suspects. Non-gang homicides had a combined average age of 63.9 years, or a mean age

of 3 1 .9 years for both victims and suspects.

80

Table 9: Descriptive demographic measures of actors across homicide subtypes

 

 

Gang homicide Non-gang homicide

N Mean Median 52 N Mean Mediap S_D
Nonwhite
Infra-racial 171 .783 1.0 .412 168 .577 1.0 .495
non-white
Male
Male-on— 172 .814 1.0 .390 179 .508 1.0 .501
male
Age
Total 180 51.80 48.0 15.12 186 63.97 60.0 26.08

average age

 

Lifestyle Measures

This category includes two measures that capture whether the actors’ involved in
the different types of homicide were at greater risk of being involved in the local drug
market. Similar to the strategy used with the demographic measures, the lifestyle
measure employed here will be at both the actor-level as well as the incident level. In
order to capture a lifestyle measure concerning potential involvement in gang crime, the
total number of prior drug arrests as well as the total number of prior weapon arrests are
examined. The use of criminal offense histories as lifestyle measures has been used in
prior research (see Jensen and Brownfield, 1987). The criminal histories used in this
study are specific to Indianapolis, which means that prior criminal involvement outside of
Indianapolis is not captured with these measures. These two variables are based on the
idea that individuals who have a local history (i.e., in Indianapolis) of drug possession as
well as weapon carrying are more likely to have a stake in the Indianapolis drug market.

For the actor-level measures, these variables are discrete, continuous measures

that capture the total number of previous drug and weapon arrests, by actor, in Marion

81

County, Indiana. A drug arrest is any arrest where the actor was apprehended for
possession or distribution of an illegal substance. A weapon arrest is any arrest where the
actor was in possession of a weapon of any kind (e.g., gun or knife). In terms of the
victim, this measure is the true sum of the prior total drug and weapons arrests in Marion
County, Indiana. This is also true where there is only one suspect. When an incident
involves multiple suspects, these two variables are the sum of the average prior drug and
weapon arrests across multiple suspects. In terms of the sequencing, the average number
of drug and weapon arrests are obtained independently across the number of suspects.

If an actor had no prior drug and weapon arrest history in Indianapolis, he or she
received a zero for both measures. Again, the victims’ information had fewer missing
values due to the fact that the victim was known in all but one of the incidents (i.e.,
unidentified). Table 10 displays the descriptive statistics for the lifestyle measures for
both victims and suspects. Victims in gang homicides had an average of .74 prior drug
arrests and .36 weapon arrests, compared with non-gang victims who averaged .35 prior
drug arrests and .17 prior weapon arrests. Suspects in gang homicides averaged .80 prior
drug arrests and .42 prior weapon arrests, while non-gang suspects averaged .35 prior
drug arrests and .19 prior weapon arrests.

For the joint-lifestyle measures, the prior number of drug and weapon arrests for
both victims and suspects were combined into two joint measures using summation, for
both drug and weapon arrests. These variables are designed to capture the combined risk
that the actors were involved in the drug market. Table 10 shows that actors in gang
incidents averaged 1.4 combined drug arrests, compared with non-gang incidents where

actors averaged .72 prior drug arrests. Actors involved in gang homicides averaged .7 2

82

weapon arrests, compared with non-gang homicides where actors averaged .37 prior
22
weapon arrests.

Table 10: Descriptive statistics for actors' lifestyle measures of prior drug and
weapon arrests across homicide subtypes

 

 

Gang Homicide Non-Gang Homicide

N Mean Median SD N Mean Median S12
Drug
Victims 292 .74 0 1 .4 270 .35 0 .89
Suspects 182 .80 0 1.2 196 .35 0 .84
Joint 182 1.4 O 2.2 196 .68 O 1.2
Weapon
Victims 292 .36 O .91 270 .17 0 .57
Suspects 182 .42 0 .85 196 .19 0 .50
Joint 182 .72 0 1.3 196 .37 O .86
Structural Meaaures

 

Two structural measures were aggregated to the IPD police beat level:
concentrated disadvantage and disorder. The concentrated disadvantage measure was
culled fiom the 2000 US. Census, while the disorder measure was created fi'om the
POPN survey data. This section describes the specific dimensions that comprise these
two structural measures.

Concentrated disadvantage has been used to represent urban neighborhoods that
are structurally disadvantaged and racially segregated (Sampson and Raudenbush, 1999).
The use of this measure has been widely employed in prior macro-level research

(McGloin and Pratt, 2003; McNulty and Bellair, 2003; Parks and Reisig, 2004; Sampson

 

22 McGarrell et al. (2006) found that in Indianapolis homicide victims averaged 12.5 arrests and suspects
averaged 11.5 arrests. Consequently, I also examined the total number of prior arrests. However, these
were directly proportional to the age of the victims (i.e., increased risk of prior arrest). Given the younger
age of gang homicide victims and suspects, the total number of prior arrests was not theoretically or
empirically grounded as a measure of gang homicide.

83

and Raudenbush, 1999). Consistent with prior research, concentrated disadvantage is a
factor variable (i.e., principal component) comprised of four census measures: the percent
of families that reside in the incident neighborhood who live below the United States
Social Security Administration Poverty Line; the percentage of people who are 16 years
of age and older and who are unemployed within the neighborhood; the percent of single-
mother households with children under 18 years of age; and the percentage of the
population who are African American.

Factor analysis is used to find the existing commonality between variables
(DiLeonardi and Curtis, 1988). Variables that contribute the same information can be
formed into factor variables. Thus, the four census variables used to capture concentrated
disadvantage were included in a data reduction (i.e., principal component) factor analysis.
The distribution of the values in the loading of the factor variable must be sufficiently
large enough to warrant the use of factor analysis. No assumptions of factor analysis
were violated in this reduction technique.23 Table 11 shows the communalities and the
total variance explained for the factor variable concentrated disadvantage. The
standardized communality loading in the principal component was relatively similar
across the four census measures, though unemployment was the weakest relative
contributor to this measure. The cumulative loading of this variable was moderate (2.80)
and explains over seventy percent of the co-variance in these four measures. The factor
variable measuring concentrated disadvantage is therefore statistically and theoretically

appropriate for firrther statistical modeling.

 

2’ The KMO test was larger than .50 (.592) and the Bartlett’s test of Sphericity was statistically significant
(p < .05), which means that no statistical assumptions were violated in this analysis.

84

Table 11: Factor analysis for concentrated disadvantage

 

 

Variable Communalities
Percent single mother headed homes .851
Percent Afiican American .656
Percent living below poverty .714
Percent unemployed .581
Total sum of srurared loading 2.80

 

The second structural measure, disorder, has been used in a number of ecological
studies (see McGarrell et al., 1997; Reisig and Parks, 2004; Sampson and Raudenbush,
1999; Skogan, 1990). Skogan (1990) classified neighborhood disorder into two
dimensions: physical disorder and social disorder. Physical disorder deals with the
physical appearance of a neighborhood, which Skogan (1990) operationalized as
vandalism, dilapidated and abandoned properties, as well as the appearance of trash and
litter. These measures capture the physical cues that a neighborhood is un—kept, which
serves as a sign that there is a breakdown of the local social order. Skogan’s (1990)
definition of social disorder included the presence of public drinking, loitering (e. g.,
youths hanging out on street comers for no particular reason), harassment, noisy
neighbors, prostitution, and open-air drug markets. These signs of social disorder are
proposed to lead to residential withdrawal from the community because daily life in these
areas creates fear and anxiety, which ultimately leads to the loss of social capital
(Skogan, 1990).

Eight survey items were culled from the POPN citizen surveys. Five of these
items measured the residents’ perceptions conceming physical disorder, while three items
measured their perception of social disorder and cohesion. As mentioned in Chapter 4,

there were 5,3 89 surveys that were completed by residents in Indianapolis.

85

Table 12 displays the questions that were used to create the beat-level disorder
measure, as well as the completion rate for each item. The joint completion rate for these
select items was 93.7 percent. In addition, the Chronbach’s Alpha reliability was .765,
which means that there was a high degree of consistency between these eight survey

24

measures.

Table 12: Survey items used to create disorder measure

 

 

Disorder Item Completeness
Physical N Percent
How big of a problem is litter/trash in 5,350 99.2
neighborhood a

How big of a problem is vandalism in 5,136 95.3
neighborhood a

How big of a problem is gangs in 4,954 91.9
neighborhood a

How big of a problem is abandoned 5,244 97.3
buildings in neighborhood a

How big of a problem is drug dealing in 4,582 85.0
neighborhood a

Social

Respondent feels safe walking in 5,287 98.1
neighborhood alone after dark b

How many neighbors will cooperate with 5,195 96.4
police c

Who causes problems in your 4,221 78.3
neighborhood d

Combined item completion 25,266/26,945 93.7
Chronbach’s Alpha .765

 

a = l-not a problem; 3-a minor problem; 5-a major problem

b = l-very safe; 2-somewhat safe; 3-somewhat unsafe; 4-very unsafe

c = l-just about everyone; 2-more than half; 3-about half; 4-fewer than half; 5-almost no one
d = l-something else; 2-people outside; 3-people from inside and outside; 4-people inside

 

2’ George and Mallory (2003) state that an alpha level higher than .7 is moderately strong, which means
these items were highly interrelated.

86

Each question was aggregated into a beat level measure by taking the average
item response fiom residents living inside each police beat. Given the moderately strong
reliability, high item completion percentage, and previous established theoretical
relationship (Skogan, 1990), these eight averaged items were then combined into a single
factor score called disorder at the police beat level. All assumptions of the principal
component analysis were met.25 Table 13 shows the communalities and the total variance
explained for the disorder variable. The standardized communality loading into the
variable was sufficient for each of the eight averaged items. The crunulative loading of
this variable was moderate (6.09) and explained over seventy-six percent of the co-
variance in these eight measures. The factor variable measuring disorder was therefore
statistically and theoretically appropriate for further statistical modeling.

Table 13: Factor analysis for disorder

 

 

 

Variable Communalities
Litter is a problem .907
Vandalism is a problem .554
Gangs is a problem .803
Abandoned property is a problem .795
Drug dealing is a problem .817
Safe at. night .764
Neighbor cooperation with police .849
Who causes problems in neighborhood .605
Total sum of squared loading 6.09
Analytic Strategy

In order to answer research questions 1 and 2, there are two distinct levels of data
employed in this study: incident level data and macro-social data.26 Both research

questions have the same outcome variable, which is a dichotomous measure that

 

2" The KMO test was larger than .50 (.901) and the Bartlett’s test of Sphericity was statistically significant

< .05).
gEF or detailed information of the data by measure and source, see Table A-1, Appendix A.

87

classified an incident as either a non-gang homicide (0) or a gang homicide (1).
Regression models for binary, discrete, outcomes allow researchers to explore the
relationship between exogenous variables and the probability that a specified event (i.e.,
gang homicide) occurs (Long and Freese, 2003: 109). Thus, for both research questions
the use of maximum likelihood (ML) methods27 that rely on a logit link function, such as

logistic regression models, is well suited for this type of outcome measure.

Research Question #1

The logistic regression model assumes the outcome to be a binary measure, where
its values are either 0 or 1. By meeting this criterion, logistic regression models can be
used to: predict an outcome measure on the basis of continuous and categorical
independent variables; determine the amount of variance explained in the outcome by
covariates; rank relative importance of independents; and, understand the impact of
control variables (Aldrich and Nelson, 1984). The linear function that determines how a

set of independent variables affect a dichotomous outcome is written as follows:

ln(1 pr .) = 130 + .13le, + + 13m.- [Equation 1]

 

Where p, is the probability of an event being a gang homicide, 130 is the log-odds of a

gang homicide when all of the independent variables have a value of zero, 131 is the effect
of the first independent variable on the log-odds of a gang homicide, x] is the value of the
ith observation on the log-odds of a gang homicide, Bk is the effect of the last independent

variable on the log-odds of a gang homicide(k), and xx, is the value of the ith observation

 

‘7 Aldrich and Nelson (1984) describe maximum likelihood estimation as an iterative process that
determines the direction and size change in the logit coefficients by increasing the log-likelihood (i.e., the
odds that the observed values of the outcome are predicted by the covariates) of the model’s fit until no
significant increase occurs (i.e., convergence).

88

for the last independent variable(k).

Aldrich and Nelson (1984) described six assumptions for the logistic regression
model. First, the dependent variable is assumed to be binary, taking on only two values
(e. g., zero and one). Second, the outcomes are mutually exclusive and exhaustive. Third,
the variation in the probability of an event (i.e., gang homicide) depends on K observable
independent variables. This assumption is written as follows:

P = P(Y IX) [Equation 2]

Where P(Y) is the probability of a gang homicide, X is the set of K independent
variables. Fourth, that the unknown relationship between Y and X is unbiased. Fifth,
that all irrelevant variables are excluded from the model. Sixth, there is no exact linear
relationship between the independent variables.

For research question #1 (Are there demographic, situational, and lifestyle
distinctions between homicides that were classified as gang related from those that were
classified as non-gang related?) all of the independent variables are incident level (i.e.,
level-1) measures. The use of a logistic regression model, where the outcome is
measured as either a gang homicide or a non-gang homicide, is appropriate for this
research question. The null model (-2 Log Likelihood) is written as
-2 * ln(Io) [Equation 3]

Where L0 is the likelihood of obtaining the observations if the independent
variables had no effect on the outcome (i.e., gang homicides). The full model is written
as:

-2 * ln(L) [Equation 4]

89

Where L is the likelihood of obtaining the observations with all the independent variables
incorporated in the model. The difference in the Log Likelihood yields a Chi-square
statistic, which is a measure of how well the independent variables affect the outcome
measure (gang homicide) (Long and Freese, 2003). The first Chi-square statistic, which
will only model situational measures, will yield the difference between the null model
(i.e., that no independent variables had an affect on the outcome) and the full model (i.e.,
that the situational measures had an affect on the outcome). The second model will yield
the difference between the situational measures as the base model and the demographic
characteristics of the actors will represent the firll model. The final level-1 model will
yield the difference between both the situational and demographic measures combined
and the lifestyle measures. Model comparisons (p < .05) will be used to evaluate the
predictive ability of situational, demographic, and lifestyle measures at the incident level.
Examining and comparing the logistic regression’s model’s Chi-square statistics will
determine the most predictive model, given the number and type of independent
variables.

After the best-fitting model is discovered for both the victim-only and joint-actor
data, predicted probabilities will be displayed in order to provide context to the model’s
coefficients. Predicted probabilities are used to determine the impact each coefficient
has, both alone and in conjunction with other measures, in terms of predicting the change
in the outcome measure (Long and Freese, 2002). Probabilties are written as follows:

P* = (odds/(1+odds) [Equation 5]
Where p* is the predicted probability of an outcome and the odds are equal to one minus

the probability of the outcome. In a logistic regression equation, the odds are in

90

logarithmic form, which can be transformed based on the following formula suggested by
Liao (1994: 12):

P* = exp(Z1)/(l+exp(Zl)) [Equation 6]

Where P* is the predicted probability of an outcome (i.e., gang homicide) and Z. is the
logistic regression equation (see Equation 1).

The main issue of concern with this analysis is the relative high frequency of
missing suspect information. While the victim information is well known in almost all
measures, the suspect information is much less complete. Over twenty-seven percent of
the homicide cases involved unknown suspects. As described earlier, even in cases
where the suspect was known, measures capturing some of the demographic and lifestyle
measures were also missing and cannot be recoded. Therefore, for Research Question #1,
two models will be presented in the subsequent chapter: 1) a model that only includes the
victim’s demographic and lifestyle measures (due to the high level of complete data). 2)
A model that includes joint information where both the victim and suspect were known,
and where demographic and lifestyle measures were captured for both actors. Situational
measures (i.e., drug related motive, firearm use as the method of death, and number of
suspects) will be incorporated into both models. This approach allows for an
examination between gang/non-gang homicides when only the victim’s demographic and
lifestyle measures are modeled, and when both actors’ demographic and lifestyle
measures are modeled. By relying on both models, rather than simply choosing one or

the other, this study attempts to reduce potential bias due to missing suspect data.

91

Research Question #2

Research Question #2 (Do structural measures of the neighborhoods where the
incident occurred delineate gang related homicides from non-gang related homicides?)
attempts to distinguish gang homicide fiom non-gang homicide based on neighborhood
characteristics. Similar to the strategy used in Research Question #1, the analytic
procedure will focus on a binary, logistic regression model where non-gang homicides
have a value of 0 and gang homicides have a value of 1. However, a major difference in
these two research questions is that the latter question relies on structural level data,
which requires a modified statistical procedure. Given both the discrete, dichotomous
outcome measure and the use of structural level independent variables, a hierarchical
generalized linear model (HGLM) is an appropriate methodological strategy
(Raudenbush and Bryk, 2002). Essentially, HGLM models the variance explained in the
probability of an outcome (i.e., odds) by partitioning the covariates into level-1 and level-
2 components, or (logit) links, in order to reduce the shared error variance between the
structural measures (Hox, 2002). When both level-1 and level-2 sets of measures are
incorporated into a single model (i.e., a standard logistic regression model), the error
terms are too optimistic, or small, due to the shared error variance in the level-2 measures
(Hox, 2002). Often, this leads to a Type I statistical error, where researchers are more
likely to reject the null hypothesis when it is in fact true. This is also known as the error
of accepting the alternative hypothesis that one or more of the independent variables
correspond with the outcome measure, when in fact they are only related due to chance

(Cook and Campbell, 1979).

92

Type I error often occurs in this instance because observations within the same
neighborhoods are not independent of one another, and standard statistical tests are not
robust against the violation of the independence assumption (Hox, 2002). By partitioning
the level-1 and level-2 measures, HGML estimates reduce the likelihood of a type I error.
The HGLM level-1 structural model is written as follows:
nij = 301 + ‘31;er + + Bpjxpii [Equation 7]

Where "U is the log of the odds of a gang homicide i in neighborhood j, BOj is the log-odds
of success for a gang homicide in neighborhood j when all of the independent variables
have a value of zero, 1311- is the effect of the first independent variable on the log of the
odds of a success (gang homicide) in neighborhood j, )0 ij is the value of the ith gang
homicide in the jth neighborhood for the first independent variable, Bpj is the effect of the
last independent variable on the log-odds of success in neighborhood j, and Xpij is the
value of the ith gang homicide in the jth neighborhood for the last independent variable.

The intercept of the level-q model becomes the outcome variables of the level-2
model, which is given by the following formula:

130,- = 700 + UOj [Equation 8]
Where 700 is the average intercept across neighborhoods and um is the unique increment
to the intercept associated with neighborhood j.

If the intercept as an outcome model yields a statistically significant result, which
is a criterion before examining slopes as outcomes (Raudenbush and Bryk, 2002), the
next step is to estimate the slopes at level-1 using the structural measures at level-2. This
is given by the following formula:

131,- = 710 + Ulj [Equation 9]

93

Where 710 is the average regression slope across the neighborhoods on the log-odds of
success of the first independent variable and ul 1- is the unique increment to the slope of

the first independent variable associated with neighborhood unit j.

Spatial Autocorrelation

In addition to the HGLM procedure, this study will examine, and control if
necessary, the impact of spatial autocorrelation across police beats in Indianapolis.
Previous research has shown that spatial autocorrelation can be a serious obstacle when
trying to model the relationship between crime and place (Morenoff and Sampson, 1997;
Morenoff et al., 2001; Smith and J arjoura, 1989). More specific to this study, prior
research examining the impact of structural measures on violent crime, including gang
homicide, has shown that spatial autocorrelation has the potential to create bias in the
neighborhood level estimates (Kubrin and Wadsworth, 2003; Rosenfeld et al., 1999).
Spatial autocorrelation refers to the susceptibility of geographical units to be influenced
by similar events (e. g., homicides) because of spatial proximity. The observed value at a
particular location depends on the values observed at neighboring locations (Anselin,
2003). When this happens, the assumption of random spatial independence is violated,
which leads to biased estimates among independent variables (Baller, Anselin, Messner,
Deane, and Hawkins, 2001). There are two main types of spatial autocorrelation that can
create bias in statistical models: spatial lag and spatial error. Spatial lag refers to spatial
autocorrelation that occurs across space in the error term (Baller et al., 2001). Spatial
error refers to spatial autocorrelation where the dependent variable in a location is
affected by both independent variables in that location as well as neighboring locations

(Baller et al., 2001).

94

GeoDa software was created to identify, and control for spatial autocorrelation
(Anselin, 2003). GeoDa relies on six specific diagnostic tests to identify spatial
autocorrelation. Three tests determine the extent that spatial error is a problem, which
include the Moran’s I test, the Lagrange Multiplier error test, and the Robust Lagrange
Multiplier error test. Two tests determine the extent that spatial lag is a problem,
including the Lagrange Multiplier lag test and the Robust Lagrange Multiplier lag test.
One test, the Lagrange Multiplier SARMA test, examines whether both spatial lag and
spatial error exists simultaneously. These tests allow the researcher to identify whether
spatial autocorrelation exists, and which set of control variables need to be included in a
given statistical model to down-weight the bias that occurs when spatial autocorrelation is
present. These set of control variables are generated in GeoDa and can be linked into
common statistical packages such as SPSS or HLM files, if necessary (Anselin, 2003).

GeoDa software requires the dependent variable (i.e., number of homicides in
police beats) to be normally distributed and continuous because it generates estimates
using an Ordinary Least Squares (OLS) method. This study models all homicides in
Indianapolis, which include both non-gang and gang homicides. The outcome variable in
this instance is the count of homicides at the police beat level, which is often a highly
skewed distribution. A common strategy used in prior research (see Morenoff and
Sampson, 1997; Sampson and Raudenbush, 1999) is to transform the count into a
measure that is more normally distributed by using the following formula:
ln((number of homicides + population)*10,000)) [Equation 10]

Thus, the logged homicide rate as the outcome measure approximates a continuous,

normal distribution. The independent variables in this model include the two structural

95

measures: economic deprivation and disorder. GeoDa software will be used to determine

whether spatial autocorrelation is a problem in the Indianapolis data.

Chapter Summary

This chapter presents the description and distribution of the independent measures
and the dependent variable that will be included in statistical models in the subsequent
chapter. In addition, the analytic strategies that will be used for both research questions
were described. The outcome measure for the two research questions posed in this
dissertation is a dichotomous variable that captures whether an incident was a non-gang
or gang homicide. Logistic regression models will be used to determine the various
measures that best explain gang homicide in Indianapolis. For Research Question #1, a
series of level-1 variables including situational, demographic, and lifestyle measures will
be modeled in order to delineate gang homicides from non-gang homicides. This model
will be applied in two ways: 1) where the victim’s demographic and lifestyle measures
are taken into account and 2) where both the victim and suspect’s (i.e., joint)
demographic and lifestyle measures are modeled. While the latter approach paints a
more complete picture of the homicide incident, the presence of missing data (due in
large part to missing suspect information) cannot be ignored. This study attempts to
explain gang homicides with data that are complete and thorough, which requires this
multi-stage approach.

The second Research Question relies on neighborhood level data, which entails
the use of statistical techniques specifically designed to handle structural measures. This

study will first examine the extent that spatial autocorrelation exists in Indianapolis by

96

using GeoDa software to model whether there is a significant bias between
neighborhoods (i.e., spatial lag or spatial error). Where bias exists, a set of control
variables, which are generated in GeoDa, will be included in the HGLM statistical
models to regulate the effect of spatial bias. If spatial autocorrelation is not a significant
problem in the Indianapolis data, the validity and generalizability of the study will be
more clearly understood. The second step for Research Question #2 is to model the
effects of disorder and deprivation on gang homicides by controlling for the level-l
measures that best explain this offense type. Thus, the latter research question builds
upon the findings of the first research question. The following chapter presents the

results of these analyses.

97

Chapter 6: Results

The purpose of this dissertation was to assess the relationship between incident
and structural level measures and gang homicides. This chapter presents the results of
the univarite, bivariate, and multivariate statistical models used to address Research
Question #1 and Research Question #2 posed in Chapter 4. The analyses for Research
Question #1 relied upon on level-1 data (i.e., incident level measures). The analyses for
Research Question #2 relied upon both on level-1 and level-2 data (i.e., both incident and
neighborhood level measures). In addition, the appropriate level-2 structural model was
contingent upon results from the level-1 model. This was because it was essential to
control for the most important incident level measures, which were uncovered in the
analyses for Research Question #1. Thus, this chapter presents the results sequentially by

the research question.

Research Question #1

Are there demographic, situational, and lifestyle distinctions between homicides
that were classified as gang related from those that were classified as non-gang related?

The variables used to address this research question are presented below.

Univariate Results

 

Table 14 displays the measures of central tendency and dispersion for the gang
homicide dependent variable as well as the incident level covariates. Seven of the
measures here are dummy coded, including the gang homicide outcome measure. The

other measures, which include the age of the actors and the prior number of drug and

98

weapon arrests, are continuous variables. In addition, the variables that are labeled as
“joint” measures include data for both the victims and suspects combined.

Table 14: Descriptive statistics for variables used in incident level analysis

 

 

 

Variable N Mean Median S.D. Min Max
Dependent variable

Gang homicide 563 .518 1.0 .500 0 1
Situational rneasures

Firearm use 563 .728 1.0 .445 O 1
Number of suspects . 555 1.23 1.0 .666 l
Drug—motivated incident 563 .07 0.0 .071 0 1
Demographic measures

Victim age 550 31.0 28.0 15.3 < 1 95
Victim non-white ‘ 541 .748 1.0 .434 0 1
Victim male ‘ 554 .792 1.0 .405 0 1
Joint age (sum) 366 57.9 53.0 22.2 16 160
Joint intra-racial non-white " 339 .681 1.0 .466 o 1
Joint male-on-male I 351 .655 1.0 .476 0 1
Lifestyle measures

Victim prior # of drug arrests 562 .556 0.0 1.2 0 10
Victim prior # of weapon arrests 562 .268 0.0 .777 0 7
Joint prior # of drug arrests 378 1.05 0.0 1.85 0 11
J ointprior # of weapon arrests 378 .533 0.0 1.11 0 7
. dummy coded variable

Bivariate Results

Two bivariate analyses are presented here. First, for the seven dummy coded
measures Pearson’s r correlations were used to determine the relationship between the
variables. Pearson’s r correlation statistics measure the direction and strength of
association between two variables (George and Mallery, 2003). All covariance
relationships were assessed based upon their association with the outcome measure, gang

homicide. Second, a series of independent sample t-tests are displayed for the continuous

99

variables age and prior drug and weapon arrests among actors, where the group
classification was non-gang and gang homicide (i.e., zero or one). T-tests were used to
determine whether there was a mean difference for the continuous independent variables
between the groups classified as non-gang and gang homicides, provided the underlying
assumptions were met.28

Table 15 displays the Pearson’s r correlation statistics between the six dummy
coded independent variables and the gang homicide outcome measure. Each of the six
measures had a positive, statistically significant relationship with gang homicide. The
strongest standardized Pearson coefficient was seen in firearm use, a situational measure
(Pearson’s r = .330). The joint male-on-male incident variable, a demographic measure,
had the second strongest relationship observed in this analysis (Pearson’s r = .327). The
drug motivated incident variable had the weakest relative co-variance association with
gang homicides (Pearson’s r = .128), though this relationship was also statistically

significant.

Table 15: Pearson's r correlations between outcome and independent variables
included in analysis

 

 

Variable Gang homicide N p-value
Firearm use .330 563 < .001
Drug-motivated incident .128 563 < .01
Victim non-white .238 541 < .001
Victim male .244 5 54 < .001
Joint intra-racial non-white .221 339 < .001
Joint male-on-male .327 351 < .001

 

 

‘8 These assumptions of the t-tests include equality of variances, where the distributions are not
significantly different from a normal distribution, and that the samples are independent of one-another. The
former two assumptions are tested directly in the SPSS software program, while the latter is true given that
each case where homicide occurs is distinctly coded as a non-gang or a gang homicide. When the Levene’s
test for the equality of variances showed that the variances were significantly distinct, meaning the
variances are not equal, the more conservative t coefficient was used to adjust for this violated assumption
(George and Mallery, 2003).

100

Table 16 displays the independent sample t-tests for the continuous independent
variables used in this analysis, where the group classification was non-gang homicide and
gang homicide. Gang homicides averaged significantly more suspects (mean = 1.35
suspects) than non-gang homicides (mean = 1.10 suspects). Victims in gang homicides
were, on average, significantly younger (mean = 28.12 years) than victims than in non-
gang homicides (mean = 34.36 years). The joint age distribution of victims and suspects
combined was also shorter for gang homicides (mean = 51.8 years) than non-gang
homicides (mean = 63.9 years). Victims in gang homicides averaged more drug arrests
(mean = .746 drug arrests) and weapon arrests (mean = .363 weapon arrests) than victims
in non-gang homicides (mean drug arrests = .351; mean weapon arrests = .166). This
was also true where the victim’s and suspect’s prior drug and weapon histories were
combined into joint-actor measures.

Table 16: T-test results for independent variables for homicide subtypes

 

 

Gang compared with p- Mean SE.
non-gang homicides N T (If value Difference Difference
Number of suspects 555 4.67 409.5 < .001 .251 .053
Victim age 550 -4.73 422.7 < .001 -6.23 1.31
Joint age 366 -5.48 298.5 < .001 -12.16 2.22
Victim prior # of drug 562 3.89 486.9 < .001 .394 .101
arrests

Victim prior # of 562 3.06 491.7 < .01 .196 .064
weapon arrests

Joint prior # of drug 378 4.05 279.4 < .001 .771 .190
arrests

Joint prior # ofweapon 378 3.09 308.2 < .01 .356 .115
arrests

 

The purpose of the bivariate analyses was to determine whether measures from
situational, demographic, and lifestyle models were substantively different across

homicide subtypes (i.e., gang and non-gang homicides), at least at the bivariate level.

101

The prior bivariate analyses indicated there was a substantive and consistent distinction
between gang homicides and non-gang homicides. In terms of situational measures,
gang homicides significantly averaged more suspects, were more likely to involve a
firearm, and were more likely to result from a drug dispute. The demographic measures
showed that gang homicides were more likely to involve younger actors, non-white
actors, and male actors. Finally in terms of the lifestyle measures, gang homicide actors
averaged more drug and weapon arrests than non-gang homicide actors. Table 17
displays the summary of the bivariate relationships between the independent variables
and the gang homicide outcome measure.

Table 17: Summary of bivariate analyses

 

Statistical Relationship with Statistically

 

Variable Model Gang Homicide Significant
Situational measures

Firearm use Correlation Positive Yes
Number of suspects T-test Positive Yes
Drug-motivated incident Correlation Positive Yes

Demographic measures

Victim age T-test Negative Yes
Victim non-white Correlation Positive Yes
Victim male Correlation Positive Yes
Joint age (sum) T-test Negative Yes
Joint intra—racial non-white Correlation Positive Yes
Joint male-on-male Correlation Positive Yes

Lifestyle measures

Victim prior # of drug arrests T-test Positive Yes
Victim prior # of weapon T-test Positive Yes
arrests
Joint prior # of drug arrests T-test Positive Yes
Joint prior # of weapon T-test Positive Yes
arrests

 

102

Multivaflrte Aggy/ms

All of the incident level measures were incorporated into logistic regression
models. Two sets of logistic regression equations are presented in this section. The first
set of models examines the situational, demographic, and lifestyle measures where only
the victim’s demographic and lifestyle measures were included (i.e., victim-only models)
in order to minimize the impact of missing data. The second set of logistic regression
models relied upon the joint-actors (i.e., both victims and suspects combined)
demographic and lifestyle measures. The joint-actor models have substantially more
missing data, but also take into account both the theoretically relevant victim and suspect

information at level-1.

Victim-Only Models

Model 1 in Table 18 shows that when all of the situational level measures were
entered into the logistic regression, each of the covariates had a significant positive main
effect on gang homicide. The anti-logarithm (i.e., exponentiation), which is the
logarithm’s multiplicative inverse, transforms the coefficient from a log-odds scale into
an odds-ratio, after subtracting one (Long and Freese, 2003). This ratio shows the main
effect that each variable had in terms of predicting gang related homicides. Model 1 in
Table 18 shows that when homicides were drug motivated they were nearly 1.5 times
more likely (odds = 1.47) to be gang homicides. Similarly, when homicides were firearm
related, they were five times more likely (odds = 4.02) to be gang related. Finally, each
additional homicide suspect doubled the likelihood (odds = 1.09) that homicides were
gang related. When only the situational variables were included in the model, roughly 20

percent of the variance in gang homicides was explained. The logistic model’s Chi-

103

square was 91.26, which served as the base model for comparison in subsequent analyses
since these measures are independent of the relevant actor-level data. Thus, gang
homicides were significantly more likely to be drug motivated, involve firearms, and
have multiple suspects in the victim-only model.

Model 2 in Table 18 indicates that adding the victim-level demographic
information yielded a significant improvement in the explained variation of the gang
homicide outcome measure over and above the effects of the situational level variables

1.29 The variance explained in gang homicides

that were already included in the mode
rose from 20.2 percent when only the situational variables were included to 28 percent
when the demographic measures of the victim were modeled. As the age of the victim
increased by one year, the likelihood that the homicides were gang related declined
roughly 1.8 percent (odds = -.018). When victims were non-white, the likelihood that
incidents were gang homicides increased 1.4 times (odds = 1.42). Similarly, when
victims were male, it improved the odds that homicides were gang related by 1.31, or 131
percent. Thus, victims were significantly more likely to be younger, non-white, and male
in gang homicides. Each situational measure also retained a significant association with
gang homicides, net of the demographic measures. While the situational measures
retained their significant relationship with gang homicide, each variable’s estimated
effect on predicting gang homicide changed, compared with Model 1, with the inclusion

of the victim’s demographic measures."0

 

29 The difference in the Chi-square statistics showed the model significantly improved the explained
variance in gang homicides given the addition of the demographic variables. The difference in the model
fit (129.39 — 91.26 = 38.03) was greater than the critical Chi-square value (6 — 3 = 3; 7.81 critical value
where p < .05). Thus, the model that includes both the victim’s demographic information and situational
measures is a significant improvement over the situational effects model.

3° In Model 2, when homicides were drug motivated they were twice as likely (odds = 1.02) to be gang
homicides. Similarly, when homicides were firearm related, they over three times more likely (odds =

104

Model 3 in Table 18 shows that including the victim-level lifestyle measures did
not significantly increase the amount of variation explained in gang homicides. The
difference in the model Chi-square statistics was negative (127.9 — 129.3 = -1.4),
meaning the model was actually hindered by the inclusion of the demographic measures
of the victims. While the variance rose slightly from 28 percent to 28.7 percent when the
victim-level lifestyle measures were included, the minimal increase could not be
attributed to more than random chance. Alternative lifestyle measures were also created
and modeled, though none of these measures yielded a significant result.31 Thus, lifestyle
measures did not enhance the incident level explanation of gang homicides in
Indianapolis. The best victim-only model was Model 2, where the situational and
demographic measures explained 28 percent of the variance in gang homicide.

Table 18: Unstandardized coefficients (Beta) and Odds Ratios (OR) for victim-only
logistic regression models

 

 

Model 1 Model 2 Model 3
(N=555) (N=529) (N=529)
Beta OR Beta OR Beta OR
Variable (SE) (Expjbi-l) (SE) (Expjm-l) (SE) (Expij)
Intercept -2.05 -- -2.60 -- -2.57 --
(.302) (.506) (.505)
Firearm use 1.61 502*" 1.18 326*" 1.15 317*"
(.222) (4.02) (.244) (2.26) (.244) (2.17)
Number of .739 209*" .883 2.41 *** .891 243*"
suspects (.193) (1.09) (.223) (1.41) (.222) (1.43)
Drug-motivated .905 2.47" .698 2.01 * .620 1.85
incident (.400) (1.47) (.412) (1.01) (.414) (.85)
Victim age -.018 .982" -.019 .982"
(.007) (-.018) (.007) (-.018)

 

2.26) to be gang homicides. Finally, each additional homicide suspect improved the odds that homicides
were gang related by 1.41, or 141 percent.

3 ' Alternative lifestyle measures were created to control for the impact of age. Older actors had a greater
likelihood for more criminal arrests. This became problematic when trying to explain gang homicides, as
actors in gang homicides tended to be younger. This assertion was also supported in Model 2. Drug and
weapon arrest rates (# of arrests/age) and interaction terms (# of arrests*age) were included in logistic
regression models. No measure approached statistical significance (p < .10) in any of these alternative
measures.

105

Table 18 (cont’d)

Victim non- .887 242*" .828 228*“
white (.245) (1.42) (.247) (1.28)
Victim male .839 2.31*** .783 2.18“"
(.259) (1.31) (.261) (1.18)
Victim # of .168 1.18
drug arrests (.104) (0.18)
Victim # of -.011 .989
weapon arrests (.151) (-.011)
Nagelkerke (pseudo) R-square .202 .280 .287
z 2 model improvement (or) 91.26 129.39" 127.9
(3) (6) (8)

 

*p < .10, ”p < .05, ”*p < .01

J oint-Actor Models

Model 1 in Table 19 is the same for both the victim-only model and the joint-
actor model since the actor information was excluded from the first (i.e., baseline)
analysis. It shows that when all of the situational level measures were entered into the
logistic regression, each of the covariates had significant positive main effects on gang
homicide. Again, gang homicides were significantly more likely to be drug motivated,
involve firearms, and have multiple suspects. When only the situational variables were
included in the model, roughly 20.2 percent of the variance in gang homicides was
explained. The logistic model’s Chi-square was 91 .26, which was the base model for
comparison in subsequent analyses since these measures were independent of the relevant
actor-level data.

Model 2 in Table 19 shows that adding the joint-actor demographic information
yielded a significant improvement in the explained variation of the gang homicide

outcome measure over and above the effects of the situational level variables that were

106

already included in the model.32 The variance explained in gang homicides rose from
20.2 percent when only the situational variables were included to 34.2 percent when the
joint-actors demographic measures were modeled. As the combined-actors age increased
by one year, the likelihood that homicides were gang related declined roughly 1.6 percent
(odds = -.016). When both actors were non-white, the odds that incidents were gang
homicides doubled (odds = 1.02). Similarly when all the actors were male, it improved
the odds that homicides were gang related by 1.35, or 135 percent. Thus, actors were
significantly more likely to be younger, non-white, and male in gang homicides. In terms
of the situational measures, both the incident number of suspects and firearm related
situational variables retained their significant association with gang homicides, net of the
demographic measures. However, the significant relationship between the drug-
motivated homicide measure and gang homicides became statistically insignificant with
the inclusion of the joint-actor demographic data.

Model 3 of Table 19 shows that adding the joint-actor lifestyle measures did not
significantly increase the amount of explained variation in the gang homicides outcome
measure.33 The variance explained did rise from 34.2 percent to 35.9 percent when the

j oint-actor lifestyle measures were included, but this improvement was not a statistically

significant change. Including both lifestyle measures did not improve the incident level

 

’2 The difference in the Chi-square statistics showed the model significantly improved the explained
variance in gang homicides given the addition of the demographic variables. The difference in the model
fit (97.99 — 91 .26 = 38.03) was greater than the critical Chi-square value (6 — 3 = 3; 6.25 critical value
where p < .10). The significance threshold is reduced to an alpha level of 90 percent due to the missing
data (i.e., lower degrees of freedom). The model that includes both victim and suspect demographic
information and situational measures for is a significant improvement over the situational effects model.
3’ The difference in the model Chi-square statistics indicated that the model fit ( 103.8 — 97.99 = 5.81) was
less than the critical Chi-square value (8 - 6 = 2; 4.60 critical value where p < .10).

107

explanation of gang homicides in Indianapolis.34 The best model in the joint-actor

analysis is Model 2, where the situational and demographic measures explained 34.2

percent of the variance in the gang homicide outcome measure. The next step was to

examine the impact that specific measures had in terms of predicting gang homicides,

which can be seen using predicted probabilities.

Table 19: Unstandardized coefficients (Beta) and Odds Ratios (OR) for joint-actor

logistic regression models

 

 

 

Model 1 Model 2 Model 3
(N=555) (N=331) (N=331)
Beta OR Beta OR Beta OR
Variable (S.E.) -1 (SE) (Expfbl-l) (SE) (Exmm-l)
Intercept -2.05 -- -2.24 -- -2.29 --
(.302) (.675) (.681)
Firearm use 1.61 502*“ .939 255*" .945 2.57“”
(.222) (4.02) (.311) (1.55) (.316) (1.57)
Number of .739 209*“ 1.10 302*" 1.10 302*"
suspects (.193) (1 .09) (.262) (2.02) (.261) (2.02)
Drug-motivated .905 2.47" .505 1 .65 .428 1 .53
incident (.400) (1 .47) (.509) (0.65) (.517) (0.53)
Joint age -.016 .984" -.017 .983“
(.007) (-.016) (.007) (-.017)
Non-white .706 2.02" .653 1 .92"
intra-racial (.294) (1 .02) (.303) (0.92)
Male-on-male .857 235*" .788 2.19"”
(.291) (1.35) (.298) (1.19)
Joint actor # of .193 1.23"
drug arrests (.091) (0.23)
Joint actor # of -.033 1.21
weapon arrests (.148) (0.21)
Nagelkerke (pseudo) R-square .202 .342 .359
z 2 model improvement (dt) 91 .26 9799* 103.8
(3) (6) (8)

 

*p < .10, “‘p < .05, **“‘p < .01

 

3‘ However, the number of prior drug arrests was statistically significant in Model 3 in the joint-actors
model. Each drug arrest for both the victim and suspect combined increased the odds that homicides were
gang related by 23 percent (odds = .23). Based on this empirical finding, it became apparent that the best
joint-actor incident level model included drug arrests but excluded weapon arrests ( Z 2= 103.79, 7 df).

This model (Model 4) is displayed in Appendix A, Table A-2. Given that theory would predict both drug
and weapon arrests as lifestyle measures should co-vary with gang homicides, and that the bivariate results
for weapon arrests was statistically significant, this model is not displayed in the main-results section.

108

Predicted Probabilities

The substantive patterns in the data can be better revealed using predicted
probabilities because they allow researchers to understand the exact impact a measure, or
set of measures, have with an outcome of interest (Long and Freese, 2003). Predicted
probabilities are the estimated probability that the dependent variable takes its high value
(i.e., gang homicides) (Long and Freese, 2003). The predicted probabilities are the
percent of correctly predicted cases of those that were correctly classified in the logistic
regression equation. In the subsequent analyses, all variables were set to their mean
values, except those variables that were specifically modeled in order to determine their
exact impact on predicting gang homicides. This was done in order to assess the
relationship between both structural and demographic measures. For the victim-only
model the correct classification was 73 percent, while the joint-actor model was 73.7
percent. Thus, predicted probabilities were independently displayed for the victim-only
and the joint-actor models.

Table 20 displays the predicted probabilities for the victim-only model. In terms
of predicting gang homicides based on the relationship between the race of the victim and
firearm use, it became apparent that when an incident was firearm related the difference
in predicting gang homicides was minimal (4.4%) between whites and non-whites.
However, when an incident was not firearm related, the difference in gang homicide
prediction was 11.5 percent between non-whites and whites, which is substantively (i.e.,
relatively) larger. A similar distinction was observed in the gender-firearm relationship.
The difference in predicting gang homicides based on the victim’s gender was less than

half as large when an incident was firearm related (4.3%) compared with an incident that

109

was not firearm related (11 percent). Finally, for incidents that had two suspects, the race
of the victim predicted 3.3 percent of the gang homicide prediction. However, when an
incident had only one suspect, the race of the victim explained over twice that amount
(7.2 percent).

Thus, a consistent pattern emerged. Situational measures, such as a firearm
related incident and the number of suspects had more relative importance in terms of
predicting gang homicides than the demographic measures of the victim. However, when
the situational variables of interest were absent (i.e., an incident only had one suspect or
was not firearm related), the race (i.e., non-white) and gender (i.e., male) of the victim
were more important contributors to the gang homicide classification. Thus, race and
gender were not as important as the situational features that predicted gang homicide, but
in their absence demographics were twice as important.

Table 20: Predicted probabilities of gang homicides based on situational and
demographic measures for victim-only model

 

 

Predicted probability A
of gang homicide

Firearm & Race
Non-white victim, no firearm 89.8
White victim, no firearm 78.3 11.5%
Non-white victim, firearm 96.6
White victim, firearm 92.2 4.4%
Firearm & Gender
Male victim, no firearm 89.3
Female victim, no firearm 78.3 11%
Male victim, firearm 96.5
Female victim, firearm 92.2 4.3%
Number of Suspects
One suspect-non-white victim 94.2
One suspect-white victim 87.0 7.2%

110

Table 20 (cont’d)
Two suspects-non-white victim 97.5
Two suspects-white victim 94.2 3.3%

 

Table 21 displays the predicted probabilities for the joint-actors model. In terms
of the relationship between the race of the actors and firearm use, when an incident was
firearm related the difference in predicting gang homicides was 5.5 percent between
white actors and non-white actors. However, when an incident was not firearm related,
the difference in gang homicide prediction was 10.9 percent between non-white actors
and white actors, which is relatively larger. In terms of the gender-firearm relationship,
the difference in predicting gang homicides based on the gender of the actors was 6.8
percent when an incident was firearm related compared with an incident that was not
firearm related (13.3%). For incidents that had two suspects, the race of the actors only
explained 3.5 percent of the gang homicide prediction. However, when an incident had
only one suspect, the race of the actors explained over 8.5 percent of the gang homicide
prediction, which is over twice that amount.

The same consistent pattern emerged in both the joint-actors model and the
victim-only model. Situational measures, such as a firearm related incident and the
number of suspects had more relative importance in terms of predicting gang homicides
than the actors’ demographic measures involved in the joint-actors model as well. Again,
when an incident only had one suspect or was not firearm related the race and gender of
the actors was twice as important to the gang homicide classification. This finding was
consistent across both models, which adds to the reliability (i.e., consistency) of the

results.

111

Table 21: Predicted probabilities of gang homicides based on situational and
demographic measures for joint-actor model

 

 

Predicted probability A
of gang homicide

Firearm & Race
Non-white actors, no firearm 85.7
At least one white actor, no firearm 74.8 10.9%
Non-white actors, firearm 93.9
At least one white actor, firearm 88.4 5.5%
Firearm & Gender
All male actors, no firearm 86.6
At least one female actor, no firearm 73.3 13.3%
All male actors, firearm 94.3
At least one female actor, firearm 87.5 6.8%
Number of Suspects
One suspect-non—white victim 89.7
One suspect-white victim 81.2 8.5%
Two suspects-non-white victim 96.4
Two suspects-white victim 92.9 3.5%

 

Diagnostic Tests

Diagnostic tests were performed to ensure the logistic regression models
displayed in the previous section were not in violation of statistical assumptions (Menard,
1995). Model 2, which included situational and demographic measures, was the best
overall model fit, given by the model Chi-square statistics, for both the victim-only and
the joint-actor models. Thus, diagnostic tests were performed for both sets of logistic
regression models.

It is imperative to include all theoretically relevant variables in the analysis in
order to avoid model misspecification (Menard, 1995). While the prior analyses include

many of the previously established correlates of gang homicide, they do not include

112

incident measures of escalation (Decker, 1993; Pizarro and McGloin, 2006), expression
(Decker, 1996) or the victim-offender relationship (Ridel, 1987). Thus, not all of the
theoretically established correlates that gang research has uncovered were incorporated in
the current study. This will be discussed in more detail in the limitations section of the
Chapter 7. However, many of the measures included in the statistical models are
theoretically relevant indicators of gang homicide, which were discussed in Chapter 2.

Another important examination was to assess whether the independent variables
in the logistic regression models had a high degree of collinearity. Perfect collinearity
among the independent variables means that at least one independent variable is a perfect
linear combination of the others (Menard, 1995). While perfect collinearity is rarely a
problem, a high degree of inter-correlation (i.e., multicollinearity) between independent
variables can also be problematic. In the presence of multicollinearity, estimated
coefficients will be unbiased and efficient, but their standard errors will be inflated (Berry
and Feldman, 1985). Inflated standard errors leads to Type H statistical error, that is
researchers fail to reject the null hypothesis when in fact the null hypothesis is false.
According to Berry and Feldman (1985) multicollinearity is not a problem unless a
correlation exceeds some predefined cut-off value, which is typically around .80. The
highest observed correlation for the victim-only model was -.278 (see Appendix B, Table
B-l). The highest observed correlation for the joint-actor model was -.355 (see Appendix
B, Table B-2).

Thus, the diagnostic tests indicated that multicollinearity was not a problem with
either set of the logistic regression models. Heteroscedasticity, or an uneven error

variance distribution, can also be a problem in a logistic regression (Menard, 1995). In

113

order to detect the accuracy of the logistic regression models, two different tests were
performed.

First, the cut-values were altered in the logit model to assess whether the default
cut-value of .50 was the best in terms of overall model fit. The classification table for the
victim-only model showed that correct classification occurred nearly 73 percent of the
time with a cut-off value of .50. The cut-values were altered from .35 to .70 to see if the
classification could be improved. Correct classification occurred 68.8 percent of the time
in the victim-only model with a cut-value of .35 and 57.8 percent of the time with a cut-
value of .70. The cut-value of .50 was acceptable because this' yielded the highest correct
classification percentage. The cut-value of .50 was also the best delineating value for the
joint-actor model. When the cut-value was set at .50, correct classification of the
outcome measure occurred 73.7 percent in the joint-actor model. The joint-model’s
correct classification occurred 71 percent of the time with a cut-value of .35 and 62.5
percent of the time with a cut-value of .70. Thus, the cut-value of .50 was acceptable for
both the victim-only model as well as the joint-actor model. A correct classification of
over 73 percent for both models is moderately strong, meaning heteroscedasticity was an
unlikely problem.

The second test was to examine the standardized residuals, which are measures
that capture the difference between the predicted and observed values for both sets of
logistic regression models. The mean for the residuals was less than .001 for both the
victim-only and the joint—actor models, which indicates that the residuals did not have a
mean substantially different than zero. Thus, no test of multicollinearity or

heteroscedasticity showed that either set of logistic regression models presented in this

114

section violated any major statistical assumptions. These diagnostic tests indicate that
overall the models presented in this section fit both the victim-only and joint-actor data,

meaning the results can be interpreted with increased confidence.

Su__mm_rarLof Level-1 Analyses

This section was designed to assess the statistical relationship between level-l
situational, demographic, and lifestyle measures and gang homicides. Two sets of
logistic regression models were presented here: 1) a victim-only model, and 2) a joint-
actor model. The victim-only model had more complete data (roughly 94 percent
complete) than the j oint-actor model (roughly 59 percent complete), due in large part to
the high number of unknown suspects. Conversely, the joint-actor model incorporated
data from both the victim and the suspects. Thus, both sets of models included important
incident level information that neither model alone could achieve. When examining the
estimates, diagnostics, and predicted probabilities both sets of models had very similar
and consistent results. Just as important, the logistic regression diagnostic tests did not
reveal any violated assumptions. The results, then, can be interpreted with a relatively
high degree of confidence.35

The situational and demographic measures were significant predictors of gang
homicides in both sets of models. Gang homicides were more likely to involve firearm
use, involve multiple suspects, and have younger, non-white, and male actors. While

drug related motive was a significant covariate in Model 1 (i.e., situational variables

 

3SIt should be noted that missing data analysis (i.e, mean replacement of missing suspect information
combined with a control variable for missing data) indicated that there was a selection bias in the joint-
actor models. In particular, homicides were significantly more likely to be defined as gang homicides
when suspect demographic data were missing. This is not unexpected, since the victim was identified as
part of a known group of chronic offenders roughly forty-five percent of these cases. However, the
estimates observed in the missing data models were almost identical to those presented in this chapter in
terms of direction and statistical significance in the situational, demographic, and lifestyle measures.

115

only), its effect was diminished when the demographic measures were included in both
sets of regression equations. In addition, the lifestyle measures operationalized as the
number of prior drug and weapon arrests did not significantly improve the prediction of
gang homicides at the incident level. Thus, Model 2 was selected as the most robust and
parsimonious statistical model at level-1 based on the modeliChi-square comparison tests.
The predicted probabilities showed that there was an observed relationship pattern
between the situational and demographic measures. At least in terms of a relative-
relationship, the situational variables were more important to the gang homicide
classification than the demographic measures. In terms of predicting gang homicide,
when the incident was firearm related and when there were multiple suspects, there was a
marginal difference between the race and gender of the actors. However, when the
method of death was not firearm related, and when there was only one suspect, incidents
were more likely to be classified as gang homicides when the actors were non-white
and/or male. In either case, the predicted probabilities had a high percentage of correct
classification, which means these measures are statistically strong correlates of gang
homicides at the incident level. The next step was to assess whether structural measures
(i.e., economic deprivation and disorder) predicted gang homicides, net of the incident

level measures observed here.

Research Question #2

Do structural measures of the neighborhoods where the incident occurred
delineate gang homicides from non-gang homicides? In order to address this research

question, this study employs incident level measures used in Research Question #1 as

116

well as two neighborhood level measures: economic deprivation and disorder. Incident
level controls help ensure that any observed neighborhood level relationship (i.e., level-2)
with gang homicide is net of the incident factors (i.e., level-1) that have been shown to
delineate gang homicide from non-gang homicide. In particular, the situational and
demographic measures used in the analyses for Research Question #1 are incorporated
into the HGLM statistical models (i.e., Bernoulli models) as control variables. Again,
two models are presented in this section to control for missing suspect information: a

victim-only model and a joint-actor model.

Unim Results

Table 22 displays the measures of central tendency and dispersion for the gang
homicide dependent variable, the incident level control variables, and the neighborhood
level structural variables. It is important to note that the number of homicide incidents
reduced fi'om 563 to 452, which is a 19.7 percent decrease. The reason for this decline is
because 111 homicides occurred outside the fifty IPD police beats in Marion County (i.e.,
in a non-IPD jurisdiction).36 Since the police beat is the neighborhood boundary at the
structural level, the incidents that occurred outside this boundary were thus excluded
from the HGLM models. In addition, the structural measures are factor-scores, which
have a mean of zero and standard deviation of one. A constant (3.0) was added to both

neighborhood level measures in order to eliminate the negative values because

 

3" All homicides occurring in Marion County were included in the IVRP incident reviews and in the
database. However, despite Marion County having a unified city-county government, there were separate
law enforcement jurisdictions within the county (Marion County Sheriff’s Department and several small
township police departments) that were outside the police department’s beats. In 2007, the Indianapolis
Police Department and the Marion County Sheriff" 5 Department merged.

117

independent variables are not permitted to have a negative value when using logistic
regression (Long, 1997) or HGLM models (Hox, 2002).

Table 22: Descriptive statistics for variables used in structural-level analysis

 

Variable N Mean Median S.D. Min Max
Dependent variable

Gang homicide " 452 .555 1.0 .49 0 1
Incident level controls

Firearm use ‘ 452 .756 1.0 .429 0 1
Number of suspects 447 1.21 1.0 .649 1 8
Drug-motivated incident ’ 452 .084 0.0 .277 0 1
Victim age 446 31.3 28.0 14.9 < 1 95
Victim non-white " 444 .801 1.0 .399 0 1
Victim male " 451 .815 1.0 .387 0 1
Joint age (sum) 289 58.2 54.0 21.5 16 160
Joint intra-racial non-white . 270 .748 1.0 .434 0 1
Joint male-on-male ‘ 276 .673 1.0 .469 0 1
Neighborhood level covariates

Economic deprivation 49 3.02 2.78 .999 .86 4.99
Disorder 49 3.01 2.80 1.00 1.37 5.31

 

. dummy coded variable

Bivariate Results

Table 23 displays the bivariate relationship between the gang homicide outcome
measure and the independent variables used here. Each of the incident level control
variables had a significant bivariate association with gang homicide, which is consistent
with the findings in Research Question #1. In addition, the structural measures had a
significant positive correlation with gang homicides, suggesting that both neighborhoods
with higher levels of economic deprivation and disorder were more likely to have gang
homicides. Thus, both the incident and structural—level measures displayed in Table 23

met the criteria for inclusion in the subsequent HGLM models.

118

Table 23: Pearson's r correlations between outcome and independent variables
included in analysis

 

 

Variable Gang homicide N p-value
Incident level controls

Firearm use ‘ .333 452 < .001
Number of suspects .137 447 < .01
Drug-motivated incident ‘ .1 1 1 452 < .05
Victim age -.198 446 < .001
Victim non-white ' .223 444 < .001
Victim male " .221 451 < .001
Joint age (sum) -.259 289 < .001
Joint intra-racial non-white ‘ .218 270 < .001
Joint male-on-male “ .335 276 < .001
Neighborhood level covariates

Economic deprivation .078 452 < .10
Disorder .104 452 < .05

 

Multivariate Results
Similar to the strategy used in Research Question #1 , both victim-only models
and joint-actor models are presented in this section in order to control for missing suspect

data.

Victim-Only Models

Table 24 presents the results of the HGLM analysis37 for the baseline
neighborhood level models. Model 1 and Model 2 are conditional fixed-effects models
where the intercept at level—1 varies across neighborhoods as a function of the slope
estimates fi'om the level-2 variables, which are fixed structural measures. These models
show the average log-likelihood of a gang homicide outcome (i.e., the average intercept,

130,-) varies across neighborhoods as a function of the measured structural level estimates.

 

’7 It is important to note that there is no intra-class correlation coefficient (ICC) when the outcome is based
on a nonlinear link function (i.e., Bemoulli distribution). The ICC is the proportion of variance in the
outcome explained by the level-2 measures. According to Raudenbush and Bryk (2002), the level-1
variance becomes heteroscedasitic when the outcome is dichotomous, which creates bias in the variance
estimate. No pseudo-estimates have been created in the latest HLM sofiware package (HLM 6.2).

119

The relationship between economic deprivation and gang homicide has been examined in
prior studies (see Chapter 3). Thus, this relationship served as the baseline model for
comparison purposes. Model 1 shows the association between gang homicides and
economic deprivation (701). The average likelihood that homicides were gang related
significantly increased (p < .05) across neighborhoods as economic deprivation increased.
This relationship, independent of other incident and structural measures, has been
observed in prior research (see Pizarro and McGloin, 2006; Rosenfeld et al., 1999).

The current study is also concerned with the relationship between gang homicide
and neighborhood disorder, particularly since this association has not been examined in
prior research. Model 2 in Table 24 shows that disorder (701) had a significant positive
relationship (p < .05) with gang homicide, net of economic deprivation. As disorder
increased at the neighborhood level, there was an increase in the likelihood that
homicides were gang related across neighborhoods. In addition, the significant
relationship between economic deprivation and gang homicide remained when disorder
was included as a structural measure. The difference in the deviance statistics indicates
that Model 2, which included both disorder and economic deprivation at the
neighborhood level, fit the data better than Model 1.38 The next step was to assess
whether economic deprivation and disorder maintained their significant association with

gang homicide after controlling for incident measures at level-l.

 

38 The difference in the deviance statistics is calculated as Dl—Dz, which approximates a Chi-square
distribution (Pl—P2), where D. is the —210gLikelihood for Model 1 and D2 is the -2Loglikelihood for Model
2. Here, the difference in the deviance statistics is divided by the difference in the degrees of freedom (A
deviance statistics / A dt) and compared against a Chi-square distribution. Thus, (618.29 — 615.28) / 450 —
449) = 3.01(1 df) > Chi-square critical value of 2.705 where p < .10 .

120

Table 24: HGLM conditional fixed effects models

 

 

 

Model 1 Model 2
Variable (N=452) (N=452)
Beta OR Beta OR
(S.E.) (Expfbl-l) (SE) Ex -1
Intercept as outcome, I30,
700 (Intercept level-2) -.262 .769* -.684 .504*
(.152) (-.231) (.194) (-.496)
Level-2 measures
Economic deprivation, 701 .139 1.149" .084 1.088*
(.042) (.149) (.044) (.088)
Disorder, 702 .180 1.197"
(051) (.197)
Deviance Statistic (df) 618.29 (450) 615.28 (449) *

 

l"p < .10, "p < .05, "*p < .01

Table 25 displays the results of the HGLM mixed model, which shows the
variation in the gang homicide outcome measure (nij) as a function of both the incident
level data and neighborhood level data, simultaneously. This type of analysis focuses on
the difference in gang homicides across neighborhoods, net of situational and
demographic factors that are associated with gang homicides at the incident level. It is
important to assess whether the neighborhood level measures retained their significant
association with the outcome variable above and beyond the incident level factors in
order to establish whether the relationship between economic deprivation and disorder
with gang homicide is robust.

Model 3 in Table 25 shows that the average intercept (BOJ'), which is the estimated
likelihood that homicides were gang related when all the covariates are equal to zero,

increased as neighborhood disorder increased. The observed main effect between

121

disorder and gang homicides maintained a significant relationship above and beyond the
incident level controls that delineate gang homicide from non-gang homicide, and was
also found net of economic disorder at the neighborhood level. However, afier
controlling for the incident level variation in gang homicide, the effect of economic
deprivation (i.e., was no longer statistically significant). In addition, the effect that drug
motivation had at the incident level in terms of predicting gang homicide was diminished
(p = .186) when disorder and economic deprivation were included in the model.39 The
difference in the deviance statistics indicates that Model 3, which included both incident
and structural level measures, fit the data better than Model 2.40

Thus, when the incident level measures were included in the model, only disorder
as a neighborhood level variable retained a marginally significant association (p < .10)
with gang homicide.“ Homicides that occurred in neighborhoods with higher levels of
disorder were more likely to be gang homicides, holding incident level measures and
economic deprivation constant. This finding suggests there is empirical support for the

disorder-gang homicide hypothesis."2

 

’9 The wash out effect that disorder and economic deprivation had on drug motivation was not unexpected.
For example, Rengert (1996) showed that drug trafficking was highest where there was large neighborhood
instability, which is a key theoretical component of social disorganization (Shaw and McKay, 1942). Thus,
it is highly likely that drug motivated homicides at the incident level along with disorder and economic
deprivation at the structural level are both theoretically and statistically inter-related (i.e., autocorrelated).
4° The difference in the deviance statistics is 115.6 (615.28 - 499.62 = 115.6) > critical Chi-square statistic
35.2 with 23 degrees of freedom.

4‘ The significance level (i.e., p-value) for Research Question #2 is relaxed to a .10 threshold at level-2
since the disorder-gang homicide relationship is an exploratory test. Henkel (1976) asserted that it is
inappropriate to set a stringent significance level in exploratory research, and thus a .10 level is acceptable.
However, when examining the relationship between situational and demographic measures and gang
homicide, the more traditional significance threshold of .05 was employed.

’2 Given that drug motivation at level-1 was not statistically significant, I ran two additional HGLM models
where drug motivation was grand mean centered at level-1 and group mean centered at level-1. Neither
additional model obtained estimates where drug motivation was statistically significant at level-1 or made
disorder fit better (i.e., significance threshold) than the models displayed in the main results section.

122

Table 25: HGLM mixed model with victim-only data

 

Model 3
(N=435)
Beta OR

Variable (SE) ( Expf 131-1 ) p-value

Intercept, 130,

700 (Intercept level-2) -2.99 .050 < .001
(.734) (-0.95)

Incident level control measures

Firearm use, 131 1.35 3.84 < .001
(.279) (2.84)

Number of suspects, 132 .616 1.83 .010
(.238) (0.83)

Drug-motivated incident B3 .560 1 .74 . 1 86
(.422) (0.74)

Victim age, 134 -.018 .982 .035

. (.008) (—.018)

Victim non-white, 135 .677 2.01 .028
(.306) (1.01)

Victim male, 136 .857 2.38 < .01
(.294) (1.38)

Neighborhood level measures

Economic deprivation, 701 -.024 .978 .843
(.124) (-.022)

Disorder, 762 .251 1.28 .076
(.138) (0.28)

Deviance Statistic (df) 499.62 (426)

 

Given that this dissertation was focused on examining the relationship between
situational, demographic, structural, and lifestyle measures it was necessary to examine
whether the addition of lifestyle measures impaired the structural estimates found in the
HGLM model. Lifestyle measures were excluded from Model 3 (Table 25) because the
analyses for Research Question #1 found no empirical support for the lifestyle measures

ability to delineate non-gang homicide from gang homicide in the victim-only model.

123

However, given that disorder was supported net of the incident level measures, it was
desirable to include lifestyle measures as potential control variables at level-1.

The HGLM model would not converge when both prior drug and weapon arrests
were included with other level-1 control measures. Thus, a number of analytical steps
were performed in order to test the relationship between disorder and gang homicide, net
of the relevant incident level controls, including lifestyle measures. First, model
convergence did occur when drug and weapon arrests were incorporated at level-1 and
only disorder was included as a level-2 measure (i.e., economic deprivation was
excluded). In this case, disorder retained its significant association (p = .085) with gang
homicide (see Appendix C, Table C-l, Model 4). Second, model convergence occurred
when only prior drug arrests were included as a lifestyle measure at level-1 (i.e., prior
weapon arrests were excluded), which allowed both disorder and economic deprivation to
be incorporated as level-2 variables. Again, disorder retained its significant association
(p = .095) with gang homicide (Appendix C, Table C-2, Model 5). Finally, and perhaps
the best approach, the HGLM model converged when prior drug and weapon arrests were
combined (i.e., added) to create one lifestyle control variable, which allowed both
disorder and economic deprivation to be included as level-2 measures. In this model,
disorder maintained its significant association (p = .090) with gang homicide (see
Appendix C, Table C-3, Model 6). Thus, incorporating lifestyle measures as level-1
controls did not diminish the observed relationship between disorder and gang homicide,
net of other incident level control variables and holding economic deprivation constant.

The prior analyses utilized an intercept as outcome approach, which is appropriate

when trying to examine the direct relationship between gang homicides and disorder.

124

The next step was to examine whether disorder had an interactive effect with level-1
variables, which is also referred to as a cross-level interaction model. In this model, both
the intercept at level-l (BOj) and the slopes of the incident level measures (ij) vary as a
function of disorder at level-2. No cross-level interaction effects were statistically
significant where the slopes of the incident level measures were allowed to vary based on
the estimates from the disorder measure (see Appendix C, Table C-4). Thus, the
observed relationship between gang homicides and disorder appears to be a direct main
effect rather than an interactive association between disorder and incident measures at

level-1.

J oint-Actor Models

An HGLM mixed model was attempted where the joint-actors demographic
measures were included as demographic control variables at level-l. This was an
important examination, given that important suspect information was excluded from the
prior analyses due to missing data. However, the estimated HGLM model would not
converge after 100. iterations, which was most likely due the presence of missing actor
data. Where there were complete data for the situational, demographic and structural
measures, the number of cases available for the model was less than 60 percent (N =
264/452) of the original homicide incidents. HGLM models are designed to diminish the
shared error variance between the level-2 measures in order to create more reliable
standard errors, which reduce the likelihood of Type I error. In this instance, the shared
error variance between homicides that ocCurred in the same police beats appears to be of

minimal concern.

125

Table 26 shows that no single police beat accounted for more than 5.3 percent (N
= 14/264) of the total outcome measures, which includes both non-gang and gang
homicides at the neighborhood level. Most beats housed less than 2 percent (N = 5/264)
of the total number of homicides available where the joint-actor demographic measures
are included. Thus, a logistic regression model can be used here without seriously
violating the independent observation (i.e., shared error variance) assumption, though the
results should be interpreted with some caution.

Table 26: Descriptive statistics for the number of homicides that occurred in police
beats for the joint-actor model

 

Homicide distribution N Mean Median S.D. Min Max

 

Number of homicides per beat 48 5.50 5.0 3.71 l 14

 

Model 6 in Table 27 shows the results of the mixed logistic regression model for
the j oint-actor data. The variance explained in the gang homicide outcome measure was
31.9 percent when the situational, demographic, and structural measures were included in
the model. Similar to the victim-only model, disorder maintained a significant
association (p < .10) with gang homicide above and beyond the effects of the incident
level measures, and net of economic deprivation. For each unit change in the disorder
measure, the likelihood that homicides were gang related increased by 36 percent (odds =
.36). Economic deprivation no longer had a statistically significant effect on gang
homicides when incident level measures were included in the model (p = .533). In
addition, many of the situational and demographic measures retained their significant
association with gang homicides when measures of neighborhood disorder and economic

deprivation were included in the model. One obvious difference was the diminished

126

effect of non-white intra-racial homicide variable, which became statistically
insignificant (p = .226). Thus, one can tentatively assert that when both the victim and
suspect’s demographic measures are included as incident covariates, and neighborhood
measures of economic deprivation and disorder are also incorporated, the significant
main effect between the race of the actors and gang homicides is no longer observed.
However, this assertion should be interpreted with a degree of caution.

Table 27: Mixed ligistic regression model for joint-actor data

 

 

Model 6
(N=264)

Beta OR

Variable (S.E.) x -1 wm—e

Intercept -2.99 -- < .01
(909)

Incident level control measures

Firearm use 1.05 2.86 < .01
(.351) (1.86)

Number of suspects .763 2.14 < .01
(.273) (1.14)

Drug-motivated incident .3 12 1 .36 .549
(.521) (0.36)

Joint age -.017 .983 .032
(.008) (-.017)

Non-white intra-racial .438 1.55 .226
(.362) (0.55)

Male-on-male .893 2.44 < .01
(.326) (1.44)

Neighborhood level measures

Economic deprivation .093 1.10 .533
(.149) (0.10)

Disorder .305 1.36 .061
(.163) (0.36)

Nagelkerke (pseudo) R-square

Model x 2 (df) 72.167 (8)

 

127

Similar to the approach used in the HGLM victim-only data, a number of
additional mixed logistic regression models were estimated to assess whether disorder
maintained its significant relationship with gang homicide while controlling for actor-
level lifestyle measures. The results were mixed. When both the joint-actors prior drug
and weapon arrests were included as level-l controls, disorder no longer maintained its
significant association with gang homicide (p = .13). However, when only the combined
actors prior drug arrests was included as a single lifestyle control measure (i.e., weapon
arrests were excluded), disorder maintained its significant relationship (p = .069) with
gang homicide (see Appendix C, Table C-6, Model 8).

Although one model that included lifestyle measures distorted the observed
significant relationship between disorder and gang homicide, three other mixed logistic
regression models found support for the association. The one model where the statistical
significance between disorder and gang homicides washed away included two lifestyle
measures that were each statistically insignificant, which likely hindered all of the
model’s estimates. Given that disorder was only marginally associated with gang
homicide in all prior models (p < .10), this result was not entirely unexpected.

The above mixed logistic regression models only examined the direct relationship
between gang homicides and disorder. The next step was to examine whether disorder
had an interactive effect with the incident level control measures in the joint-actor model.
No interaction effects were statistically significant (see Appendix C, Table C-7). Similar
to the victim-only model, the observed relationship between gang homicides and disorder
appears to be a direct main effect rather than an interactive association between disorder

and incident level measures.

128

Predicted Probabilities

Predicted probabilities show the predictive effect that specific measures have with
an outcome of interest (Long and Freese, 2003). HGLM models are not well suited for
this type of analysis because the intercept at level-1 is a function of the level-2 slopes.
Thus, the equations that estimate the final model are extremely complex in that they vary
with changes in the level-2 data. This makes the use of predictive probabilities
impractical in this case because the level-2 measures are the focus of the current research
question and thus could not be set as fixed estimates. In addition, the current HLM
software package (version 6.2) does not contain a user-function that performs this task.
Thus, predicted probabilities will only be estimated for the joint-actor model, given its
coefficients were obtained in a logistic regression model. The results appear to be
generalizable to the victim-only model because the odds-ratios are relatively similar
between the two models.

The current analysis is interested in explaining the substantive relationship
between disorder and gang homicides. Disorder is a factor variable measured on a
continuous scale, which requires the variable to be set to both extreme values and
quartiles based on the observed distribution of the measure (Long and Freese, 2003).
Table 28 shows the predicted probability that homicides were gang related based on
neighborhood disorder changes. All of the incident level control variables as well as
economic deprivation were set equal to their mean values. Any change in the predicted
probability that homicides were predicted to be gang related is based solely on a change
in the disorder measure. The probability that homicides were gang related increased 5.8

percent when the neighborhood level disorder measure was changed fiom its lowest

129

observed value (1 .37) to its highest observed value (5.3), holding all other measures equal
to their mean. Thus, we can generalize that homicides were almost 6 percent more likely
to be gang related in the most disordered neighborhood compared to the neighborhood
with the least amount of observed disorder, holding all other factors constant. In terms of
the middle of the distribution, the probability homicides were gang related increased 2
percent when the disorder measure was changed from its value in the first-quartile to its
value in the third-quartile.

Table 28: Predicted probabilities of gang homicides based on changes in
neighborhood disorder

 

 

Predicted probability A
of gang homicide
Extreme values
Highest observed disorder value 97.3
Lowest observed disorder value 91.5 5.8 %
Quartile values
3rd quartile value 96.2
1St quartile value 94.2 2 %

 

Diagnostic Tests

One of the greatest threats to the validity of the findings presented in this section
is the presence of spatial autocorrelation (Anselin, 2003). Spatial autocorrelation can
affect the estimates at the neighborhood level due to the fact that geographic units are
susceptible to the influence of similar units in close proximity. Thus, beat level measures
could be autocorrelated with heat level measures of its nearest neighbor (i.e., adjacent
police beats). As discussed in Chapter 5, when spatial autocorrelation occurs the
assumption of random spatial independence is violated, which leads to biased estimates

among independent variables (Baller et al., 2001).

130

In order to assess whether spatial autocorrelation was an issue in the Indianapolis
data, a spatial regression model was estimated in GeoDa (0.95-i) software. Specifically,
first-order rook continuity weights43 were used in order to capture the shared
neighborhood variation in both the disorder and economic deprivation measures since
these were the structural level independent variables used in the previous logistic
regression models. The rook continuity weights were incorporated into a spatial
regression model where the homicide rate in each police beat was logged in order to
approximate a normal continuous distribution, which is a requirement of the GeoDa
software package. The outcome measure used in the previous logistic regression models
included both non-gang and gang homicides, which included all homicides at the
neighborhood level.

The major difference between the homicide outcome used in the logistic
regression models and the one used in the spatial regression model is the presence of a
denominator that controls for police beat size (i.e., the rate component). The Pearson r
correlation between disorder and police beat size was -.024 (p = .874). The Pearson r
correlation between economic deprivation and police beat size was -.016 (p = .917).
Thus, there appears to be no association with the two structural measures and police beat
size. This means the results are generalizable between the OLS and logit models, at least
in terms of the spatial autocorrelation between the structural measures and a beat’s
nearest neighbor (i.e., adjacent beat) value. Again, the relationship among the
neighborhood level independent variables across police beats is what is important in this

analysis.

 

‘3 Rook spatial weights are constructed from those neighborhoods that share a common linear boundary
with the neighborhood of interest. These are the most common type of spatial weights (Anselin, 2003).

131

Table 29 displays the results of the spatial regression analysis. Specifically, it
shows that both disorder and economic deprivation significantly co-vary with the
(logged) homicide rate at the neighborhood level, which is consistent with the base
HGLM model (see Model 2, Table 24). Most importantly, the model shows that spatial
autocorrelation, which includes measures of spatial lag, spatial error, and a combination
of both spatial lag and spatial error (i.e., the SARMA model) was not a problem in the
Indianapolis data. Given that one of the assumptions of the logistic regression model is
to exclude theoretically irrelevant independent variables, it would be inappropriate to
include spatial weights that reduce spatial autocorrelation in the multivariate models used
to address Research Questions 1 and 2 because spatial independence was not violated.
None of the spatial autocorrelation tests approached statistical significance meaning that
there was no observed bias among the police beats in terms of the two independent
variables: disorder and economic deprivation. This strengthens the validity of the
findings of the previous multivariate regression models.

Table 29: OLS spatial regression model

 

DV = ln(homicide rate)

 

 

 

(N = 50)
Beta
Independent Variable (St. Error) T-statistic P-value
Intercept 2.263 20.78 < .001
(108)
Disorder .298 2.30 .025
(129)
Economic Deprivation .408 3.16 < .01
(129)
Model F value (df) 15.90 (47) < .001
Adjusted R-square .378 --
Spatial Autocorrelation Tests
Moran’s I (error model) .067 1.05 .292

132

Table 29 (cont’d)

Lagrange Multiplier (error model) 1 1.16 .280
Robust LM (error model) 1 1.18 .277
Lagrange Multiplier (lag model) 1 0.44 .505
Robust LM (lag model) 1 0.46 .497
Lagrange Multiplier (SARMA) 2 1.62 .444

 

As previously mentioned, multicollinearity is not a problem unless a correlation
exceeds some predefined cut-off value, which is typically around .80 (Berry and
F eldman, 1985). In terms of the victim-only HGLM model, the highest observed
correlation for the incident level measures was -.305 (see Appendix D, Table D-l). The
highest observed correlation for the structural level measures was .353, which was the
relationship between disorder and economic deprivation. In terms of the joint-actor
logistic regression model, the highest observed correlation for the incident level data was
.319 (see Appendix D, Table D-2). The highest observed correlation between the
structural measures was .364, which was the correlation between disorder and economic
deprivation. The joint-actor correlation was based on complete actor data (N = 264).
Thus, the diagnostic tests indicate that multicollinearity among the independent variables
was not a problem with any of the multivariate models used in this chapter.

The level-1 standardized residuals of the HGLM model were assessed, which
included estimates fiom the level-2 structural measures. The mean was .08, the median
was 1.19, and the standard deviation was 2.42. The histogram of the level-1 residuals
shows the distribution approximated a normal curve (see Appendix D, Figure D-l).
Similarly, the standardized residuals from the joint-actor mixed logistic regression model
had a mean of .01 and a standard deviation of 1.0. The histogram of the residuals

indicates the errors approximated a normal distribution in this model as well (see

133

Appendix D, Figure D-l), which is an assumption of the test. Both histograms showed a
homoscedastic error distribution. Thus, influential cases do not appear to undermine
either the victim-only or the joint-actor models’ efficiency. In sum, no test of spatial
autocorrelation, multicollinearity, or heteroscedasticity showed that any of the
aforementioned models violated statistical assumptions. These diagnostic tests lend
weight that the estimates obtained from the multivariate models were not due to observed

bias in the data, which adds validity to the findings.

gum—min! of Level-2 Analyses

This set of analyses sought to address whether economic deprivation and disorder
at the neighborhood level affected the likelihood that homicides were gang related. A
number of steps were taken to ensure that the results presented in this chapter were not
compromised due to bias in the data. First, given that the suspect data were missing in
many of the homicide events, two separate sets of multivariate regression models were
estimated: a set of victim-only models, and a set of joint-actor models. Importantly, both
sets of models had consistent results in terms of the estimates, odds-ratios, and
significance levels. Second, HGLM models were used to control for the shared error
variance that often occurs when neighborhood level data are employed. This is
particularly true in the victim-only models, where a large majority of the observed
homicides were included in the models and thus many of the homicide incidents occurred
in the same neighborhoods. However, the HGLM models would not converge for the

joint-actor models, which was likely due to missing data.44 Fortunately, the fact that

 

4" As noted earlier, the joint-actor models excluded homicides where the suspects’ demographic measures
were unknown. Again, missing data analysis was conducted in order to assess the impact this selection bias

134

HGLM models would not converge where there were suspect data added strength to the
assertion that both sets of models (i.e., joint-actors and victim-only) were a necessity. In
addition, where the joint-actors data were used, there did not appear to be a major
problem with shared error variance because fewer than 60 percent of the original cases
were included in the models and thus most of the homicides occurred in different
neighborhoods. Third, a number of diagnostic tests were performed that examined the
extent that spatial autocorrelation, heteroscedasticity, and multicollinearity existed in the
observed data. None of these tests indicated that there was an inherent bias due to
extreme or influential cases, creating a sense of confidence in the results.

The results showed that when economic deprivation and disorder were modeled
independent of incident level control measures, both variables had a significant positive
relationship with gang homicide. When level-1 control measures were included, the
effect of economic deprivation on gang homicide washed away. This was true in both the
victim-only model (Table 25) and joint-actor model (Table 27) and for the majority of the
joint-actor models that tested different approaches to the inclusion of lifestyle measures.
Interestingly, the relationship between gang homicide and economic deprivation was seen
independent of other incident and structural level variables, but did not hold when other
theoretically relevant variables were included in the analysis. This exact finding between
economic deprivation and gang homicide has been observed in prior research (see
Chapter 3). Thus, the association between economic deprivation and gang homicide
appears to be one of a mediating effect (i.e., mediated through other measures), rather

than a direct effect.

 

created. As with the level-1 models, the level-2 estimates that relied on data where the missing values were
imputed had very similar estimates to the models presented in this chapter.

135

However, the direct relationship between disorder and gang homicide was much
more robust. Disorder maintained a significant positive relationship with gang homicide,
controlling for incident level control measures and net of economic deprivation. This
was true in a large majority of the multivariate models. Disorder did not have a direct
relationship with the slope estimates at level-1, suggesting its relationship with gang
homicides is more direct. Given that no previous study has relied upon multivariate
techniques to explain the disorder-gang homicide relationship, the results presented here
are suggestive that such a relationship exists. These findings have the potential to inform

future research, policy, and theory, which will be discussed in the next chapter.

136

Chapter 7: Discussion

This chapter focuses on the limitations, theoretical implications, directions for
future research, and overall results from the prior analyses. First, a brief summary of the
research findings will be presented. Next, a number of important limitations of the
current study are discussed in detail, which additionally provides direction for firture
research in this area. Finally, this chapter concludes by synthesizing the information
obtained from the results of the current study with a number of different bodies of
knowledge in criminal justice research. More specifically, a link between the findings
observed here and the implications for ecological criminology, gang research, directions

for research methodology, and public policy is discussed.

Overall Summary of Results

This study relied on two sets of statistical models (i.e., victim-only and joint-actor
models) at both level-l and level-2. Both sets of models had, very similar results in that
the situational, demographic, lifestyle, and structural measures were robust in their
direction and significance thresholds across all of the various sets of models. It begs the
question, which is the best set of models? Given the similarity between the joint-actor
and victim-only models, the best overall set of regression equations likely stems from the
victim-only models because excluding cases where suspects are unknown has the
potential to eliminate pertinent incident information. The victim-only models had far
fewer missing cases because it was extremely rare for homicide victims to remain
unidentified (i.e., one exception in four years). In addition, the level-2 analyses that

relied on HGLM techniques converged when relying on the victim-only data, which

137

controlled for the violation of independent observations (i.e., multiple events in the same
neighborhood). The joint-actor models did not converge and thus relied on mixed-
logistic regression estimates, which were more likely to have shared error variance
(Raudenbush and Bryk, 2002). Thus, the victim-only models reduce the likelihood of a
selection or statistical bias.

Situational and demographic correlates of gang homicides have been established
in prior gang research. However, the knowledge about the relationship between
situational and demographic correlates with gang homicides was ascertained in classical
gang studies that relied on data fiom chronic gang cities (Spergel and Curry, 1993). This
dissertation employed data from an emerging gang city, Indianapolis, which made it
possible to assess whether the current findings were consistent with prior gang homicide
research.

In terms of situational and demographic measures, gang homicides were more
likely to be drug motivated, involve multiple suspects, be firearm related, and have more
male, non-white, and younger actors. This was true in both the victim-only and joint-
actor models. Thus, at the incident level, this study was consistent with prior gang
research. Implications for the incident level findings suggest that there does not appear to
be a substantive difference between chronic and emerging gang cities in terms of
explaining gang homicide from level-1 measures. Even though many of these incident
level measures were ascertained in research that relied upon data from chronic gang
cities, the results presented here suggest that their explanatory power is robust across the
different gang city subtypes. This lends support to the idea that cross-city comparisons of

gang homicide research are indeed generalizable.

138

hi terms of lifestyle measures, prior studies have shown that individuals who
engage in gang activity are more likely to participate in serious, risky ‘street oriented’
behavior (Anderson, 1997; Oliver, 1994; Thomberry, 1998; Wright and Decker, 1997).
The use of criminal histories as a surrogate of lifestyle measures (i.e., criminal
propensity) has been established in lifestyle research (Jensen and Brownfield, 1987).
However, the efficacy of criminal histories as measures of violent gang behavior is still
widely unverified. This study examined the utility of criminal histories as surrogates of
gang violence, and the results were mixed at best.

Most of the models predicting gang homicide did not contain statistically
significant estimates from the prior number of drug and weapon arrests of the victim or
the combined actors. In the joint—actor model, the combined actors prior number of total
drug arrests did have a positive significant association with gang homicide in one of the
models. Thus, there appears to be minimal support for the use of criminal histories as
lifestyle surrogates for gang violence, but the results were not consistent or
overwhelming. This does not mean that lifestyle measures are unimportant in explaining
homicide. Indeed, both victims and suspects had extensive criminal histories. However,
these results suggest this is true for homicide generally but does not effectively
discriminate between gang and non-gang homicide.

At the neighborhood level, the established relationship between economic
deprivation and gang homicide was inconsistent. In prior research, economic deprivation
measures have been shown to co-vary with gang homicide until other structural and
incident level controls are modeled (Pizarro and McGloin, 2006; Rosenfeld etal., 1999).

Then, the relationship between economic deprivation and gang homicide seems to

139

diminish. The same result was observed in this study. Economic deprivation
significantly predicted gang homicide at the neighborhood level only when disorder was
included as a structural covariate. However, when incident level control variables were
included, the relationship between economic deprivation and gang homicides was no
longer statistically significant. This was true in both sets of multivariate regression
models.

The relationship between disorder and gang homicide has never been directly
tested in the criminological literature, at least using statistical techniques. A link between
disorder and gang homicide is theoretically plausible since disorder has been shown to
positively correlate with robbery (Sampson and Raudenbush, 1999), and robbery has
been shown to positively correlate with drug markets (Wright and Decker, 1997), and
drug markets have been shown to positively correlate with gang homicides (Blumstein,
1995).45 Therefore, it was necessary to test whether a direct relationship between
disorder and gang homicide existed.46 Controlling for relevant incident level factors, and
holding economic deprivation constant, disorder had a statistically significant positive
influence on gang homicide. This finding was consistently observed across a large
majority of multivariate regression models in this study.

In sum, the findings reported in Chapter 6 were fairly consistent with the research
hypotheses presented earlier. Before the implications of this study are discussed in detail,

the limitations of the current investigation are noted.

 

’5 Blumstein argued that that the crack cocaine market was distinct in terms of generating lethal violence.
This is an important qualification because the research on the link between drug trafficking and gang
homicides is mixed (see Block and Block, 1993).

‘6 Skogan (1990) contended that disorder generates a variety of criminal opportunities (e. g., public
drinking, drug use, prostitution, assaults and fights) in addition to robbery and drug markets.

140

Limitations

This research has a number of theoretical, operational, and methodological
limitations. While the obvious goal was to address as many limitations as possible, it was
unfeasible to alleviate all major issues. However this section, at a minimum, attempts to
discuss some of these concerns as a way of both cautioning against inappropriate
conclusions as well as directing prospective studies in this area. In particular, the
limitations discussed in this section have the potential to serve as guides for future areas
of inquiry.

While the incident level correlates used in this study have been shown to
distinguish gang homicide from non-gang homicide, they are by no means an exhaustive
list of important measures that have been shown to make this delineation. Rather, they
were measures that were generated at the IVRP incident reviews, and prior research has
shown their relative importance. Indeed, a wide number of studies have found that gang
homicides differ from non-gang homicides in a number of important ways that were not
captured in the data used here. Maxson et al. (1985) found that gang homicides were
more likely to take place in public settings, particularly on the street. This is important
because the gang literature shows that the street context is very important when
attempting to explain why gang homicides often occur. Researchers such as Riedel
(1987) and Decker (1993) distinguished the nature of a homicide based upon the victim-
offender relationship. For example, Riedel (1987) found that instrumental homicides
often occurred when the offender attempted to improve his or her position through some
rational calculation such as minimizing risk, increasing gain, or both. In a similar vein,

Decker (1993) found that expressive homicides tended to occur as a way of ‘saving face’

141

or character contests. While Anderson’s (1999) research was more concerned with
explaining the differences between street and decent culture, his research demonstrated
that drug dealers (i.e., those involved in street culture) have to save face in order to
maintain street credibility by being respected. Thus, these incident level measures that
capture the nature of the dispute that led to gang homicide should be included in gang
homicide models.

When trying to explain gang homicide using structural measures, Pizarro and
McGloin (2006) used a dummy coded escalation measure as an incident level control
variable in order to capture the common incident level characteristics of gang homicide.
Unfortunately, in the current research detailed narratives describing the victim-offender
relationship, or the manner in which the homicide escalated, were not gathered at the
IVRP incident review meetings. Rather the task force was more concerned with
identifying the individuals who had a pattern of chronic, persistent offending since they
were the targets of the IVRP deterrence-coercion strategy. Thus, future gang homicide
research in this area should attempt to gather measures of escalation, and/or the victim-
offender relationship when possible in order to include as control variables at the incident
level.

Another limitation was the use of drug and weapon arrests as lifestyle measures.
Undeniably, the use of criminal histories as a surrogate for an actor’s lifestyle is more the
exception than the rule. Most research concerning an actor’s lifestyle focuses on four
dimensions: irresponsibility, self-indulgence, interpersonal intrusiveness, and social rule
breaking (Walters, 2006). In terms of routine activities theory, Cohen and F elson (1979)

operationalized lifestyle as drug dealing, gang involvement, delinquent peer association,

142

and the amount of time spent away from home. They found that individuals who
engaged in these behaviors were more likely to be victimized since they were engaging in
activity that increased their risk of victimization.

Jensen and Brownfeld (1987) proposed that criminal histories serve as a measure
of routine activity in that a higher number of offenses increase the risk of victimization
because of the motives, vulnerability, or culpability of people involved in those activities.
However, the weapon carrying measure employed in this study was limited. The
criminal histories in Marion County did not capture the weapon type the actor was
carrying at arrest. There is also the problem that not all suspects were arrested at the
scene of the homicide. A potential strategy would be to assess whether the weapon
obtained at the arrest was a firearm or not, given that firearm carrying among gang
members is common (Block and Block, 1993; Blumstein, 1995; Blumstein, Rivara, and
Rosenfeld, 2000; Blumstein and Rosenfeld, 1998; Bjerregaard and Lizotte, 1995; Huff,
1998; Lizotte, Tesoriero, and Thomberry, 1994; Maxson et al., 1985). This would be a
more likely surrogate for lifestyle behavior that may predict the likelihood of being
involved in a gang homicide.

In addition, Developmental Life Course (DLC) research shows that criminal
propensity is not constant throughout offenders’ lifecycle (Loeber and Farrington, 2000;
Sampson and Laub, 1993). In a similar vein, DeJong (1997) advanced deterrence theory
by showing that arrestees incarcerated for longer periods of time took longer to reoffend
(i.e., had a significantly longer time until failure) than arrestees incarcerated for shorter
periods of time. Lifestyle research attempting to predict victims and suspects

involvement in gang homicide might benefit from models that take into account the

143

amount of time (e.g., the number of days) from their most previous drug and weapon
arrest to the time of the homicide incident. Theory would predict that actors more
recently involved in the drug market would be at greater risk for gang homicide
involvement. This type of variable operationalization may capture this risk element.
Unfortunately, these measures were not captured in the Indianapolis data. Future
research could benefit from an empirical examination testing whether this measure of
lifestyle predicts gang homicide involvement at the individual level.

In terms of the structural data, Wilson and Kelling (1993) and Taylor (1997)
contend that the streetblock is the appropriate unit for neighborhood-level analysis,
particularly when trying to model neighborhood incivilities. The authors argue that
neighborhood structure is a relatively small group dynamic and that disorder (i.e.,
incivilities) is more likely to predict crime in a more concentrated area than at higher
units of analysis. Studies that have used more concentrated neighborhood data measuring
disorder have relied upon measures of physical and social incivilities from observational
data (Sampson and Raudenbush, 1999; Skogan, 1990). By using this methodology, these
studies constructed streetblock level estimates of disorder.

This study relied upon survey data from the POPN study to construct a measure of
disorder, which required aggregation to the police beat level. This is not overly
problematic because police beat borders have been used as neighborhood boundaries in
prior research (Chermak, McGarrell, and Gruenewald, 2006; Reisig, McClusky, and
Mastrofski, 1999; Reisig and Parks, 2004; Skogan et al., 1999). However, this makes a
more concentrated analysis impossible. Fortunately, Taylor’s (1997) main argument is

that incivilities cannot be aggregated to cities or police districts because these units of

144

analysis are theoretically misaligned. Given that prior research has relied upon police
beats and that the police beat unit is much smaller than a single city wide measure, the
use of police beats as a neighborhood level boundary is methodologically appropriate,
though limited.47

Another limitation was missing data. In 24 percent of the cases (N = 135) the
suspect(s) was unknown, and therefore the demographic and lifestyle measures of the
suspect were excluded where the offender was unknown. This was because, at least at
the time of the homicide incident review, these cases remained open given the presence
of an unknown suspect. Missing data has the potential to compromise the efficiency and
accuracy of the estimates obtained in statistical models (Shadish, Cook and Campbell,
2002). The use of both the victim-only and j oint-actor models was intended to control for
missing actor data. Missing data techniques showed that homicides were more likely to
be gang related when the suspect information was missing. Fortunately, the coefficients
were very consistent across both sets of models, suggesting that missing suspect
information did not bias the estimates in the joint-actor models. However, caution should
be taken when examining the joint-actor models because of missing suspect information.

An additional limitation is that the dependent variable in this study is only similar
to the gang affiliated homicide outcome measure used in prior gang research (Maxson
and Klein, 1990). Perhaps a better analytic strategy would be to perform an analysis
similar to Rosenfeld et al.’s (1999) approach, which involved the use of two outcome

measures: gang affiliated homicide and gang motivated homicide. This study found

 

’7 Researchers need to be aware that the same people are involved in multiple sources of data in this type of
analysis. As a potential way to capture the likelihood of covarying error, the addresses of the victims and
suspects should also be coded and aggregated to the neighborhood level in order to assess the degree that
actors commit offenses in their same (or different neighborhoods) where they live.

145

support for the disorder-gang homicide hypothesis, which is particularly relevant to this
limitation. Theory would predict that the disorder-gang homicide relationship would be
strongest where the incidents are a direct product of gang activity. In particular, gang
motivated offenses are more likely to occur from street disputes, as a likely product of
participation in the illegal drug market (Blumstein, 1995). Where possible, future
research in this area should attempt to assess the disorder-gang homicide relationship
with both sets of outcome measures.

It is also important to mention that this study is limited to the use of cross
sectional data, which is particularly significant when trying to explain the disorder-gang
homicide relationship. Taylor (1999) did not find support for temporal sequencing when
explaining the disorder-crime relationship with longitudinal data (i.e., the “incivilities
thesis”). Specifically, Taylor (1999) determined that incivilities at an earlier point in time
did not necessarily increase fear and crime in the same neighborhood at a later point in
time. Taylor’s work challenged the disorder-decline hypothesis (Bursik and Grasmick,
1993; Skogan, 1990), which posits that physical and social incivilities lead to crime.
However, this is not overly problematic here because this study attempted to simply
assess whether there was an observed disorder-gang homicide relationship. Thus, no
attempts were made in terms of explaining causation through temporal sequencing (i.e.,
whether or not more disorder leads to more gang homicide in the same neighborhood).
Indeed, no causal claims are attempted given the aforementioned limitations of the data
and the fact that the alpha level for disorder was only marginally statistically significant
(p < .10) throughout most models. Put simply, this study sought to address whether there

was an observed correlation between disorder and gang homicide, and the results are

146

indeed suggestive of a relationship. However, longitudinal data at the neighborhood level
would improve both the generalizability and validity of this finding.

Though the current study has a host of limitations, the results fi'om this research
address a number of gaps in criminal justice research. The advancements made in the

criminological literature are detailed throughout the remainder of this chapter.

Implications for T heavy

One of the challenges for gang homicide theoretical development is that many of
the same predictors of violence and homicide in general are thought to be predictors of
lethal violence could discriminate gang homicides from homicides generally. Essentially,
the proposition is based on the idea that situational, lifestyle, and ecological variables
should be even more robust predictors of gang homicide because they represent the type
of group-based street violence that these theories were designed to explain. In contrast,
homicide in general includes a more heterogeneous group of incidents that, though
predicted to relate to these theoretical constructs, may be less well-explained by routine
activities, lifestyle, and ecological theory.

Indeed, the findings of the analysis are consistent with this theoretical foundation.
Both incident-level and structural variables drawn from situational and ecological
theories of crime significantly delineated gang homicides from non-gang homicides. The
only exceptions were indicators of lifestyle measures drawn from arrest records and
neighborhood-level economic deprivation. Even with these variables, there was some
evidence of robust zero-order relationships between situational and ecological variables

and gang homicide. Indeed, this research shows that the theories designed to explain

147

group-based street violence do indeed explain a significant proportion of the variation
between violence in general (represented by non-gang homicide) and street violence
(represented by gang homicide). While this study was designed to examine this
relationship, future research is needed in this area in order to better understand gang

homicide from a theoretical perspective.

Implications for Gang Research

The analyses presented here tested the explanatory power of incident and
neighborhood explanations at multiple levels of analysis. One of the main goals of this
study was to assess whether previously established incident level correlates delineate
gang homicides from non-gang homicides in emerging gang cities, since many of these
correlates were revealed in research from chronic gang cities. This is important because
a number of studies have shown that the scope and nature of the gang problem has
changed dramatically over the years (Egley and Ritz, 2006; Webb and Katz, 2003).
Gangs have emerged as problems seen in large cities to problems in small and medium-
sized cities as well. Thus, gang homicides are a product of social, and in many cases
bureaucratic, operationalization. Law enforcement personnel are forced to define and
understand how gang structures exist, especially since gang dynamics vary over time and
across different cities. This was a potential threat to the generalizability of the
Indianapolis data since these measures were obtained in an emerging gang city, where
law enforcement does not have an extensive history in dealing with gangs.

The findings suggest that situational features of the incident and demographic
characteristics of the actors distinguish gang homicides from non-gang homicides in a

similar fashion in both chronic and emerging gang cities. Thus, the previously

148

established correlates of gang homicides continue to stand up to empirical scrutiny when
explaining different homicide subtypes. Gang homicides are more likely to involve
multiple, younger, non-white, male actors and the incidents tend to be firearm related,
and drug motivated. This was true in large, chronic gang cities in the early 1980’s and
held true in Indianapolis (i.e., an emerging gang city) in the late 1990’s. Thus, the
incident level correlates of gang homicide from this study are considered generalizable
with the results from prior gang homicide studies.

In terms of the law enforcement literature, Webb and Katz (2003) showed that
gang units specialize in intelligence gathering, investigation, and suppression techniques.
Also, gang units often differ across agencies because the local gang problems and issues
vary by locale. This is important because the tasks performed by the gang units are very
specialized across different cities and by police department. However, when defining the
relationship between homicide actors and gangs, many specialized gang unit officials
tend to look for similar network dynamics (i.e., relationships) among gang members and
offenders. For example, in Indianapolis, incidents were classified as gang homicides
when multiple law enforcement participants in the homicide incident reviews agreed
upon the relationship between an actor and a gang (i.e., the actor was part of a ‘group of
known, chronic offenders’) and when the individual describing the actor-gang
relationship provided extensive detail on the group of known chronic offenders
(McGarrell and Chermak, 2003).

As discussed earlier, Weisburd and Braga (2006) proposed that law enforcement
decision-making has tended to follow the clinical model, which is where police personnel

make decisions about the classification of offenses and strategies for reducing crime

149

based solely upon their own prior experience. Weisburd and Braga (2006) stated that this
model is frequently employed in a theoretically weak manner and is often not well suited
for strategic intervention approaches. In terms of understanding the gang homicide
problem, the IVRP team in Indianapolis appeared to extend an individually based clinical
classification approach through the group review as part of a strategic approach to reduce
violence. This conclusion is based upon three major factors.

First, when defining an incident as a gang homicide the IVRP working group and
researchers were diligent about internal consistency and accuracy. Thus, their definition
of gang homicides tended to be more conservative due to the strict standard they adhered
to when classifying offenders as members of a group of known chronic offenders
(McGarrell and Chermak, 2003). Again, offenders were classified as gang members’
only when there was an abundance of reliable information obtained about the actors at the.
homicide incident reviews, including when at least two law enforcement officials could
describe the gang or group.

Second, there was little empirical support for the lifestyle measures in terms of
predicting gang-affiliated homicides. Thus, a likely explanation is that law enforcement
agents did not simply define individuals as belonging to a group of known, chronic
offenders based upon their prior contact (i.e. arrests) with actors. A less strategic
approach would have likely relied upon official police records in terms of defining an
actor as someone who was a chronic, persistent offender. It would thus appear the gang
homicide task force was more concerned with understanding the relationship among

actors suspected to be involved in gang activity than solely using official police records.

150

Third, results from this study suggest that the Indianapolis definition of gang
homicide was consistent with prior gang research in both chronic and emerging gang
cities. This was true when modeling the situational features of the incident and
demographic characteristics of the actors. Although defining the gang problem is a local
decision, the classification developed by law enforcement personnel and researchers in
Indianapolis turned out to be consistent with other cities in terms of defining events as
gang homicides. Taken together, these three factors suggest that the Indianapolis team
defined gang homicides by utilizing an empirically driven investigation rather than
simply “going through the motions”, which is often the criticism of the clinical model
(Webb and Katz, 2003). Again, the IVRP was a replication of the Operation Ceasefire
intervention in Boston, which relied on a problem-oriented, strategic approach to reduce
gang violence. Thus, it would appear that extending the individually based clinical
classification approach through a group-based strategic problem solving process
enhances the validity and reliability of the gang homicide identification process.

In addition, the disorder-gang homicide relationship has implications for the law
enforcement literature as well. Again, this study indicates that disorder has a significant
positive relationship with gang homicide, and thus, two final implications are discussed.
First, the community policing and community-organizing approaches that have been
shown to reduce disorder have the potential to reduce the likelihood of gang violence
(Skogan, 1990; 2007). Indeed, the disorder-gang homicide relationship will be better
understood if future research shows that strategic approaches specifically designed to
reduce neighborhood disorder impacts gang homicide. Second, a rival hypothesis in

terms of explaining the disorder-gang homicide relationship is that groups in high

151

disorder contexts may get labeled as gangs. Policing research should attempt to gather
more information on the incident and whether structural cues are important to gang task

force members as they label incidents as gang or non-gang related.

Implications for Ecological Criminology

The relationship between neighborhood structure and gang homicide also informs
the ecological literature. This study utilized two neighborhood variables: economic
deprivation and disorder. The relationship between economic deprivation48 and gang
homicide has been examined in prior research. The disorder-gang homicide relationship
had not been specifically examined in statistical models until this study. Thus, the macro

level implications for the ecological literature vary by neighborhood measure.

Economic Deprivation

Based on the results from this study as well as the findings from prior research,
the relationship between economic deprivation and gang homicides appears fragile.
Pizarro and McGloin (2006) found that when escalation was included as an incident level

control variable, the once significant relationship between poverty and gang homicide

 

‘8 Pizarro and McGloin (2006) examined the relationship between poverty and gang homicide. They
constructed their poverty variable by summing the standardized value of the following four census
measures: percent of people living in poverty, percent of people unemployed, percent of people receiving
public assistance, and the percent of single parent households. Rosenfeld et al.’s (1999) research
constructed a neighborhood disadvantage variable, which was a four-item principal corrrponent of the
following measures: family poverty rate, percentage males unemployed, percent female-headed households
with children under 18, and percent of families receiving public assistance. Operationally, both of these
constructs are theoretically and methodologically very similar to the economic deprivation measure used in
this study. Of particular importance, Rosenfeld et a1. (1999) could not separate the influence of
neighborhood disadvantage on gang homicide from the influence of racial composition. In some of their
results, the authors exclude either racial composition or neighborhood disadvantage, as the two covariates
combined tended to confound their results. This lends support to Wilson’s (1987) contention that economic
deprivation should include a measure of percent Afiican American, as this construct is a better indicator of
structural disadvantage.

152

diminished. In this study, only when disorder and economic deprivation were isolated
fi'om incident level controls did economic deprivation exert a significant relationship with
gang homicide. However, when incident level correlates were incorporated into HGLM
models the once significant relationship between economic deprivation and gang
homicides observed in this study also diminished.

A comparison with Rosenfeld et al.’s (1999) findings is slightly more difficult to
disentangle because they attempted to isolate the independent effects between gang
homicide and neighborhood disadvantage and race. Of particular importance, the
researchers could not separate the influence of neighborhood disadvantage on gang
homicide from the influence of neighborhood racial composition (Rosenfeld et al., 1999).
In many of their models, the authors excluded either racial cemposition or neighborhood
disadvantage, as the two covariates combined tended to confound their regression
estimates. When racial composition was included in their multivariate models, but
neighborhood disadvantage was excluded, racial composition exerted a significant
influence on gang homicides at the neighborhood level. In contrast, when neighborhood
disadvantage was included and racial composition was excluded, neighborhood
disadvantage significantly predicted non-gang homicides. The authors concluded when
racial composition is controlled, neighborhood disadvantage does not significantly

increase the fiequency of gang homicide (Rosenfeld et al., 1999: 512).49

 

‘9 In an effort to replicate this specific facet of Rosenfeld et al.’s (1999) research, I re-estimated an
additional HGLM victim-only model using three structural measures: disorder, neighborhood deprivation
without percent African American included as part of the factor variable, and percent African American as
a stand-alone structural measure (see Appendix B, Table E-l). Incident level controls were also
incorporated into the model. Results from this analysis indicate that economic deprivation (minus percent
Black population) was positively associated with gang homicide (p = .04), while percent African American
was negatively associated with gang homicide (p = .05). Disorder retained its positive significant
association with gang homicide (p = .01) regardless of the deprivation-race delineation. In addition, when
either percent African American or economic deprivation was included as a structural measure, and the

153

Both Rosenfeld et al.’s (1999) findings as well as the results from this study
provide support for the use of a more conservative measure of economic deprivation that
incorporates race (i.e., the percent Afiican American population) as a dimension into the
structural variable. Combining the percent Afiican American population with other
measures of economic disadvantage into one overall variable has been done in a number
of prominent ecological studies (see Sampson and Raudenbush, 1999; Sampson and
Wilson, 1995; Wilson, 1987). Indeed, the results fiom this study were very consistent
with Rosenfeld et al.’s (1999) findings. Again, only when the neighborhood level models
included measures of disorder and economic deprivation did economic deprivation exert
a significant relationship with gang homicide. Once other important control variables
were incorporated, the relationship between economic deprivation and gang homicide
diminished.

Thus, prior research as well as the) results from this study suggests that the
relationship between economic deprivation and gang homicide is extremely unstable. It
would appear that, by itself, economic deprivation significantly predicts gang homicide at
the neighborhood level. Once other structural and incident level control variables are
incorporated, there is no significant association between economic deprivation and gang
homicide. On the surface it would appear that the most disadvantaged neighborhoods are

more likely to have gang homicide, but when other incident and structural factors are

 

other measure was excluded, disorder still retained its positive significant association with gang homicide.
Thus, this analysis finds empirical support for Rosenfeld’s et al.’s (1999) assertion that there is a
confounding relationship between the estimates of the measures percent African American and economic
disadvantage. Perhaps more importantly, the confounded relationship between deprivation and race had no
effect on the disorder-gang homicide estimates. In terms of interpretation, just as gang homicides can be
defined as gang affiliated (i.e., the Los Angeles definition) or gang motivated (i.e., the Chicago definition),
non-gang homicides are also likely to have varying operational components, such as dispute or domestic
related. This might explain why deprivation and racial composition explain non-gang homicides
differently. Implications for these findings suggest the need for additional research in this area, particularly
when attempting to understand non-gang homicides at the structural level.

154

controlled, this is not the case. Though it must be noted that these findings do not
diminish the importance of economic deprivation as a structural correlate of homicide in
general. Indeed prior research has shown that homicide in general is more likely to occur
in socially and economically disenfranchised areas (Hsieh and Pugh, 1993). However, it
would appear that economic deprivation as an ecological measure explains both non-gang
homicides and gang homicides in a similar fashion. These findings suggest that
economic deprivation’s impact on gang homicide seems to be mediated by both

individual and structural characteristics, including neighborhood disorder.

D_iSQI_dfl

The relationship between disorder and general forms of crime has been examined
in a number of studies. As discussed in Chapter 3, Sampson and Raudenbush (1999)
found that the disorder-crime relationship was not robust across different crime types. In
particular, they found that disorder lost its statistically significant relationship with
homicide (in general) and burglary when measures of collective efficacy were
incorporated into their multivariate regression models (Sampson and Raudenbush, 1999).
Sampson and Raudenbush (1999) did find that disorder maintained a statistically
significant relationship with robbery after controlling for collective efficacy at the
neighborhood level. This finding is consistent with the qualitative work of Wright and
Decker (1997), who showed that armed robbers are attuned at targeting dealers in the
local drug market, since these actors have access to cash and are unlikely to report to

police.

155

As discussed earlier, based upon longitudinal data, Taylor (1999) found little
support for the idea that neighborhood disorder at time 1 leads to crime at time 2. This
finding challenged the “incivilities thesis” proposed by Bursik and Grasmick (1993) and
Skogan (1990). Thus to this point, the ecological literature suggests that disorder is not
particularly well suited at explaining crime in general, either cross sectionally (i.e., the
relationship with different types of crime at a fixed point in time) or longitudinally (i.e.,
local disorder leads to increased crime). Rather, the prior research suggests that the
disorder-crime relationship is one that exists in specific types of crime, particularly
violent crime that is associated with drug markets. Given that gang homicides are more
likely to be drug-market driven (Blumstein, 1995) and take place in the street (Maxson et
al., 1985), the direct relationship between disorder and gang homicide was examined in
this study.

Ecological theory states that disorder is high when neighborhood informal social
controls are low (Bursik and Grasmick, 1993; Skogan, 1990). Street crime is higher in
neighborhoods with lower levels of informal social control (Morenoff, Sampson, and
Raudenbush, 2001; Sampson and Raudenbush, 1999; Sampson, Raudenbush, and Earls,
1997; Wilson and Kelling, 1982). This study found that disorder maintained a marginally
significant positive relationship with gang homicide net of economic deprivation and
after controlling for incident level measures of gang homicide. This suggests that the
relationship between disorder and gang homicides is relatively robust. At this point it
would be premature to assert that disorder exerts a strong relationship on gang homicides

at the neighborhood level, but the results presented here certainly suggest that such a

156

relationship may indeed exist. Thus, the observed relationship between disorder and
gang homicide has important implications for firture macro-social research.

Given that disorder maintained a significant relationship with gang homicide after
controlling for other theoretically important factors, a number of questions remain. Does
disorder maintain its effect with gang homicide when measures of collective efficacy are
included as neighborhood level variables? Skogan (1990) posited that low social control
leads to high disorder. Sampson et a1. (1997) proposed that low informal social control is
explained by low collective efficacy. Thus, criminological theory states that disorder and
collective efficacy are both inter-related with neighborhood informal social control
mechanisms. However, collective efficacy incorporates one key neighborhood level
element that disorder does not take into account, which is citizens’ willingness to take
action when there is a disturbance (Sampson et al., 1997). Sampson and Raudenbush
(1999) examined the predictive power of disorder on different types of crime when
measures of collective efficacy were incorporated and found mixed results based on
different types of outcomes (i.e., different types of crime). The question remains, is it
high disorder that is predicting gang homicides at the neighborhood level or is it low
informal social control mechanisms? In particular, are gang homicides more likely to
occur where residents are unwilling to take action, or where physical and social cues
promote gang homicide, or both? Only when measures of collective efficacy are
included as structural covariates can this relationship be untangled.

As discussed in the earlier limitations section, theory would also suggest that
disorder would have a stronger influence on gang-motivated homicides since these

incidents are more likely to occur as a direct product of gang specific activity (Maxson

157

and Klein, 1990). Specifically, gang motivated homicides are more likely to involve
public space, to be drug-market related, or to involve other types of street crimes
(Maxson and Klein, 1990). However, gang-affiliated homicides are incidents that
involve known gang participants, meaning that the incident itself does not necessarily
have to be gang-related (i.e., it is more of a lifestyle measure of the actors than a measure
of a gang-motivated incident). Since this study only relied on an outcome measure that is
consistent with gang-affiliated homicides, future studies should attempt to model the
gang motivated homicide-disorder relationship. Finally, the size of the neighborhood
boundary used here has important theoretical implications for ecological research. Most
police beats in Indianapolis were comprised of two to three census tracts. Given T ita et
al.’s (2005) findings that gang “set space” is likely to be much smaller than
neighborhoods or census tracts, future research attempting to unravel the disorder-gang
homicide relationship should rely upon more micro-level neighborhood boundaries.
Though many of these theoretical and methodological questions cannot be
addressed in the current study, the results from this study lend support to the idea that
disorder is more of crime specific structural factor than a neighborhood level correlate of
crime in general. In addition to its established relationship with residential fear, the
literature is beginning to suggest that neighborhood disorder'is a structural correlate of
violent, drug-market related crime. The findings presented here suggest that more
rigorous empirical examinations are certainly worthwhile because future studies can

address many of the aforementioned gaps in the literature.

158

Policy Implications

Taken collectively, the findings from this research have clear implications for
public policy. Evidence exists that strategic intervention programs have the potential to
reduce gang homicide (Kennedy et al., 1996; McGarrell et al., 2006; Webb and Katz,
2003). The incident level findings from this research suggest that gang homicides in
Indianapolis are similar to other chronic and emerging gang city homicides, at least in
terms of the situational features of the incidents and the actors’ demographic make-up.
The neighborhood level findings suggest that disorder has a significant effect on gang
homicide. Thus, this section describes some key policy implications based on the

empirical results observed in this study.

Actor Level Initiatives

In terms of the incident level findings, one of the most consistent predictors of
gang homicide was firearm use. A prominent policy intervention designed to reduce gun
violence by targeting youth gang offenders was the Operation Ceasefire program in
Boston (Braga, Kennedy, Waring, and Piehl, 2001). Kennedy (2006) described the ‘lever
pulling’ strategy implemented in Boston as a relatively narrow focus that utilized a wide
variety of data sources and analytic techniques, relied heavily on experiences and insight
of front-line law enforcement and community actors, and attempted to have a substantive
impact. Project research showed that the problem of youth homicide was largely
concentrated among a small group of chronically offending, gang-involved, youth
(Kennedy, Piehl, and Braga, 1996).

Organizational members (i.e., law enforcement and research personnel) in the

Operation Ceasefire project diagnosed the gang problem, and researchers concluded that

159

chronic disputes among gangs appeared to be the most significant cause of gang violence
in Boston (Braga et al., 1999). After implementing the focused-deterrence strategy
(Braga et al., 2001; Kennedy et al., 1999), the evaluation component of the Operation
Ceasefire project showed that Boston experienced a 63 percent reduction in youth
homicide and similar large reductions in nonfatal serious gun violence (Braga et al.,
2001). Because of its observed potential, the Operation Ceasefire project was replicated
in several cities including Baltimore, Minneapolis, and Indianapolis (Braga and
Weisburd, 2006: McGarrell et al., 2006). More recently, this strategy was a driving force
behind the PSN, a national program designed to reduce violent-gun crime.

The results from these strategic intervention programs are suggestive that
targeting gang members in a tactical manner can reduce overall homicide in emerging
gang cities. Multiple studies have now provided evidence that the prevalence of citywide
homicide has been shown to decrease when these programs are implemented. Given that
prior research, as well as the results from this study indicating that gang homicides are
more likely to be firearm related, strategic firearm reduction programs have the potential

to reduce the occurrence of gang homicide.

Neighborhood Level Initiatives

The neighborhood level results in this study suggest that gang homicides are more
likely in areas with higher disorder. Prior research has shown that disorder is more
prominent in areas with low informal social control mechanisms. Policy suggests when
informal mechanisms are low to non-existent it may be necessary to augment formal
social control mechanisms. This idea is based on the notion that there are areas where

citizens have a difficult time policing themselves. Therefore, these neighborhoods may

160

need formal assistance, by way of law enforcement, to help regulate behavior and build
informal social control mechanisms. This has been empirically observed because formal
social control mechanisms can also serve as a vehicle, or as a mediator, for strengthening
informal social control mechanisms (Greene, 2003).

From a theoretical perspective, strengthening informal social control by
enhancing formal social control is what Bursik and Grasmick (1993) refer to as public
level control. The idea behind public level control is to link citizens in disordered
neighborhoods to outside agencies, such as government organizations. In this sense,
public control is achieved because local organizations are able to secure public goods and
services that are allocated by agencies outside the neighborhood (Bursik and Grasmick,
1993). An example of this type of policy-driven intervention can be seen in broken
windows policing.

Broken windows policing is a law enforcement emphasis designed to reduce
disorderly behavior and minor offenses, often referred to as ‘quality of life’ offenses,
such as prostitution, public urination, and aggressive panhandling (Sousa and Kelling,
2006). Many of the behaviors targeted in broken windows policing are cues of
neighborhood level disorder (Skogan, 1990). This is important because Rosenfeld et al.
(2006) found that arresting offenders for minor crimes (i.e., vagrancy, vandalism, public
drunkenness, and aggressive panhandling) often ‘nets bigger fish’, such as offenders who
have outstanding warrants for more serious forms of crime. Further, Rosenfeld,
Fomango, and Rengifo (2007) and Messner, Galea, Tardiff, Tracy, et al., (2007) found
that arrests for less serious disorder-type offenses related to declines in the overall crime

rate, at least in New York City. Sousa and Kelling (2006) contended that only one part of

161

policing is crime reduction, but another component of policing is ‘order maintenance’.
Order maintenance as a policing strategy attempts to reduce neighborhood level instances
of accumulative harms and offenses. Reducing social and physical disorders in areas
marred with disorder has a worthwhile effect on citizens’ quality of life. Thus, it may be
possible to reduce gang homicides indirectly by improving neighborhood disorder, given
that study finds support for a disorder-gang homicide relationship.

However, critics argue that aggressive disorder policing actually increases harm at
the neighborhood level (Harcourt, 2001). Specifically, Harcourt (2001), Kennedy (2006),
and Skogan (2006) contend that increased arrests for minor offenses will likely create
neighborhood confusion as well as introduce even greater levels of racial inequality to the
criminal justice system, rather than changing views toward minor offending. Indeed,
some scholars see aggressive disorder policing as a form of social disorder itself (i.e.,
residents may not want to witness police arresting people in their neighborhood for minor
offenses). Future research in this area is vital to understanding the impact of aggressive
disorder policing at the neighborhood level. Whether aggressive disorder policing has a
positive or negative effect on neighborhood social order is still widely unknown, given
the low number of studies that have addressed this question. ‘More research is needed in
this specific area, especially given there is suggestive evidence that disorder and gang

homicide are related.

Conclusion

In conclusion, this study addresses gaps in the gang homicide literature. The

findings suggest that many of the significant incident level correlates that explain

162

homicide in chronic gang cities also hold in emerging gang cities. This was particularly
the casein Indianapolis, where the situational features of the incident and demographic
characteristics of the actors were significantly more likely to predict gang homicide. The
extent to which gang homicides are similar across chronic and emerging gang cities
remains largely unverified, though this research showed the similarities may be more
common than the differences. Although many of these incident level measures emerged
from research dealing with chronic gang cities, the results presented here suggest that
their explanatory power is robust across different city subtypes. However, several
important incident level explanations of gang homicide were not measured in this study.
Future research should examine explanations of gang homicide in terms of the victim—
offender relationship (see Decker, 1993; Riedel, 1987) across different city subtypes.
This type of empirical investigation could test the generalizability of gang homicide
across different locales.

The situational and demographic features that significantly predict gang homicide
alone are theoretically insufficient. Simply stated, these factors (i.e., multiple, young,
non-white males that carry firearms, and are involved in the drug market) do not explain
gang homicide alone mainly because these dynamics do not always result in a fatal
confrontation. Indeed, something else has to happen to explain how these dynamics all
converge to explain gang homicide. Decker (1996) found that homicides are often linked
through networks and described how these networks link victims, suspects and witnesses
to violence. In essence, a witness to one event can become the victim or suspect in
another event. Some researchers propose that once predictors from social learning theory

(i.e., measures that capture association with delinquent/criminal peers) are controlled, the

163

importance of demographic characteristics of the actors is superseded (Esbensen et al.,
2001: 20). Thus, it is vital to include theoretically relevant measures of peer association
and escalation when trying to explain gang homicides.

This research supports the hypothesized link between disorder and violent crime,
specifically gang homicide. Until the late 1990’s, there was an extensive limitation of
research that examined the neighborhood effect on gang homicide. Fortunately, a
growing body of literature has begun to disentangle the neighborhood-gang homicide
relationship. The results presented here are suggestive that a disorder-gang homicide
relationship exists, net of incident level features and other structural characteristics.
These findings do not contradict prior ecological research, but rather support the notion
that structural level disorder has a direct effect on violent neighborhood drug crime
(Sampson and Raudenbush, 1999; Wright and Decker, 1997).

Results from this study lend support to the idea that disorder is more of crime
specific structural factor than a neighborhood level correlate of crime in general (i.e.,
does not fit at the top of the cone of resolution). More specifically, neighborhood
disorder may be a specific structural correlate of violent, drug-market related crime (i.e.,
it appears to fit at the bottom of the cone of resolution). However, until future research
conducted in different cities examines the disorder-gang homicide relationship at a more
micro-social level (Tita et al., 2005) as well as including measures of collective efficacy
(Sampson et al., 1997) and incident level escalation (Decker, 1996), neighborhood level
explanations of gang homicide will be theoretically inadequate.

It is important to provide a firture research agenda that will take gang homicide

research in the appropriate direction. Future research in this area should focus on the

164

gaps in the gang homicide literature. Gang homicides need to be explained in terms of
the situational features of the event and demographic characteristics of the actors in
multiple gang city subtypes (i.e., whether they are chronic or emerging gang cities) and
by using different gang classifications (i.e., gang affiliated or gang motivated offenses),
while controlling for the victim-offender relationship (i.e., escalation measures).

hi addition, missing data analysis showed that there was a bias in the joint-actor
models where demographic information on the suspect was unknown. In particular,
where suspect demographic information was missing, homicides were more likely to be
gang related. Based upon censorial evidence from the researchers involved in the
incident reviews, it became apparent that this was consistent with social disorganization
theory (Sampson etal., 1997) and the ‘no snitch’ street culture (Anderson, 1999).
Indeed, Decker (1996) showed that individuals involved in the street culture were often
concerned about retaliatory responses from groups of violent offenders (i.e., gangs).
Criminological theory would suggest that individuals residing in socially disorganized
neighborhoods are less likely to cooperate with law enforcement, which is vital to solving
an open homicide. The fact that neighborhood disorder was a significant predictor of
gang homicide is thus consistent with theoretical assertions drawn from social
disorganization theory as well as ethnographic observations (Anderson, 1999). Future
gang homicide research in this area should rely heavily on the contextual information
about all homicide incidents in order to uncover the factors that predict gang homicide.
Based on the results of the missing data analysis it becomes apparent that unknown

suspect information is not data missing at random. Rather there appears to be a

165

relationship with unknown suspect information and gang homicide, which should be the
focus of future research in this area.

Most researchers would agree, demographic, situational, lifestyle, and structural
measures alone do not explain gang homicide. All of these factors have to come together
to create a combustible atmosphere where a fatal confrontation is more likely to result.
Continuing to examine the empirical validity of theoretically relevant measures that
explain the environment where gang homicides flourish is a vital step to criminological
research. Only then will our understanding of gang homicide become more valid and

generalizable. The research presented here has been a step in that direction.

166

Appendix A

167

Table A-1: Missing data table for data sources used in the multivariate analyses in

 

 

Indianapolis
Variable Total % Narrative
Included

IVRP data

Firearm use 563 100 Complete data from incident reviews

Drug-motivated incident 563 100 Complete data from incident reviews

Number of suspects 555 98.6 6 cases involved missing suspect details

Victim non-white 563 100 Complete data from incident reviews

Victim male 554 98.4 7 cases where coding information was
lost

Victim age 550 97.7 13 cases missing data for victim’s age

Joint male—on-male 351 62.3 184 cases with unknown suspects
combined with missing victim data
(gender)

Joint age (sum) 366 65.0 184 cases with unknown suspects
combined with missing victim data
(age)

Joint intra-racial non- 378 67.1 184 cases with unknown suspects

white combined with missing victim data
(race)

Victim prior drug arrests 562 99.8 1 unknown victim, was unable to run
criminal history for this person

Victim prior weapon 562 99.8 1 unknown victim, was unable to run

arrests criminal history for this person

Joint prior drug arrests 378 67.1 1 unknown victim, 184 cases with
unknown suspects

Joint prior weapon 378 67.1 1 unknown victim, 184 cases with

arrests unknown suspects

POPN data

Disorder Measure 452 80.2 111 homicides fell outside IPD
jurisdiction. Overall survey completion
rate was 53 percent (5,389/8,245)

Census data

Economic Deprivation 452 80.2 111 homicides fell outside IPD

Measure

jurisdiction. Demographic measures are
estimated to be with 1-2 percent of
actual population (Freedman and
Wachter, 2002)

 

168

Table A-2: Joint-actor logistic regression model including drug arrests but

excludingweapon arrests

 

 

Model 4
(N = 331)
Beta OR

Variable (SE) (Exp(_b)—1)
Intercept -2.28 --

(677)
Firearm use .939 255*"

(.3 1 5) (1 .55)
Number of suspects 1.10 3.01***

(.260) (2.01)
Drug-motivated .437 1 .54
incident (.515) (0.54)
Joint age -.017 .983"

(.007) (-.017)
Non-white intra-racial .641 1.89"

(.298) (0.89)
Male-on—male .782 2.1 8"

(.296) (1.18)
Joint actor # of drug .184 1.20"
arrests (.081) (0.20)
Joint actor # of weapon -- --
arrests
Nagelkerke (pseudo) R-square .359
l 2 model improvement (df) 103.79

(7)

 

169

Appendix B

170

Table B-1: Pearson's r correlations between independent variables included in
victim-only logistic regression

 

 

Pearson r Pearson r Pearson r Pearson r Pearson r Pearson r
P-value P-value P-value P-value P-value P-value
N N N N N N
Variable Variable Variable Variable Variable Variable Variable
A B C D E F
A) # of .069 .036 -.052 -.043 .074
suspects -- .102 .396 .225 .320 .086
555 555 542 534 546
B) Drug- .069 .076 -.040 .082 .074
motivated .102 -- .073 .347 .056 .082
incident 555 563 550 541 554
C) Firearm .036 .076 -.278 .264 .191
use .396 .073 -- < .001 < .001 < .001
555 563 550 541 554
D) Age of -.052 -.040 -.278 -.183 -.116
victim .320 .347 < .001 -- < .001 < .01
534 550 550 536 548
E) Non- -.043 .082 .264 -.183 .226
white .320 .056 < .001 < .001 -- < .001
victim 534 541 541 536 541
F) Male .074 .074 .191 -.1 16 .226
victim .086 .082 < .001 < .01 < .001 --
546 554 554 548 541

 

171

Table B-2: Pearson's r correlations between independent variables included in joint-

actor logistic regression

 

 

Pearson r Pearson r Pearson r Pearson r Pearson r Pearson r
P-value P-value P-value P-Value P-value P-value
N N N N N N
Variable Variable Variable Variable Variable Variable Variable
A B C D E F
A) # of .069 .036 -.011 .118 -.128
suspects -- .102 .396 .838 .028 .014
555 555 339 351 336
B) Drug- .069 .076 .072 .108 -.039
motivated .102 -- .073 .187 .044 .459
incident 555 5 63 339 351 366
C) Firearm .036 .076 .269 .267 -.216
use .396 .073 -- < .001 < .001 < .001
555 563 339 351 366
D) Non- -.011 .072 .269 .214 -.211
white intra- .838 .187 < .001 -- < .001 < .001
racial 339 339 339 339 331
B) Male- .118 .108 .267 .214 -.355
on-male .028 .044 < .001 < .001 -- < .001
351 351 351 339 341
F) Total -.128 -.039 -.216 -.211 -.355
age of .014 .459 < .001 < .001 < .001 --
actors 366 366 366 331 341

 

172

Appendix C

173

> Table 01: Additional HGLM model where both prior drug and weapon arrests

were included as level-1 control measures

 

 

Model 4

Variable (N=435)
Beta OR
(SE) (ExpmH) p-value

Intercept, 130,

700 (Intercept level-2) -2.95 .052 < .001
(.703) (-.948)

Incident level control measures

Firearm use, 131 1.32 3.74 < .001
(.278) (2.74)

Number of suspects, 132 .623 1.87 .085
(.236) (0.87)

Drug-motivated incident B3 .51 1 1.66 < .01
(.424) (0.66)

Victim age, 134 -.018 .981 .033
(.008) (.019)

Victim non-white, 135 .635 1.88 .035
(.301) (0.88)

Victim male, 36 .818 2.26 < .01
(.296) (1.26)

Victim drug arrests, B7 .102 1.10 .326
(.103) (0.10)

Victim weapon arrests, Bg .001 1.01 .993
(.152) (0.01)

Neighborhood level measures

Economic deprivation n/a n/a n/a

Disorder, 701 .221 1.25 .085
(.127) (0.25)

Deviance Statistic (df) 498.3 (425)

 

174

Table C-2: Additional HGLM model where prior drug arrests were included as a
level-1 control measure and both level-2 measures were included

 

 

Model 5

Variable fl=435)
Beta OR
(S.E.) Ex -1 -value

Intercept, Bo)

yon (Intercept level-2) -2.91 .054 < .001
(.735) (-.946)

Incident level control measures

Firearm use, B] 1.32 3.77 < .001
(.279) (2.77

Number of suspects, B; .625 1.87 < .01
(.236) (0.87)

Drug-motivated incident 133 .509 1.66 .230
(.423) (0.66)

Victim age, 134 -.018 .981 .033
(.008) (-.019)

Victim non-white, 135 .648 1.91 .035
(.307) (0.91)

Victim male, 136 .817 2.26 < .01
(.295) (1.26)

Victim drug arrests, 137 .103 1.11 .263
(.092) (0.11)

Neighborhood level measures

Economic deprivation, 701 -.O3l .969 .803
(.124) (-.031)

Disorder, 702 .237 1.27 .095
(.139) (0.27)

Deviance Statistic (df) 498.2 (425)

 

175

Table C-3: Additional HGLM model where a combined lifestyle measure was
included as a level-1 control measure and both level-2 measures were included

 

 

Model 6

Variable (N=435)
Beta OR
(S.E. ) Ex -1 p-value

Intercept, Bo,

700 (Intercept level-2) -2.91 .055 < .001
(.734) (-.945)

Incident level control measures

Firearm use, 131 1.32 3.77 < .001
(.279) (2.77)

Number of suspects, 132 .620 1.86 < .01
(.236) (0.86)

Drug-motivated incident B3 .534 1.71 .207
(.422) (0.71)

Victim age, 134 -.018 .981 .033
(.008) (-.019)

Victim non-white, 135 .642 1.90 .037
(.307) (0.90)

Victim male, 136 .811 2.25 < .01
(.297) (1.25)

Total victim drug & weapon arrests, .065 1.06 .305

B7 (.064) (0.06)

Neighborhood level measures

Economic deprivation, 701 -.030 .969 .807
(.124) (-.31)

Disorder, 702 .241 1.27 .090
(.139) (0.27)

Deviance Statistic (df) 498.4 (425)

 

176

Table C-4: Cross-level interaction models-disorder with incident

level measures

 

 

Intercept Slope
Coefficient Coefficient
Variable“ ( st. error) (st. error)
Number of suspects, 710 & y” .697 -.022
(.407) (.921)
Drug-motivated incident, 720 & 72. .719 -.045
(1.69) (.465)
Firearm use, 730 & 731 .924 .122
(968) (267)
Victim age, 740 & 741 -.007 -.003
(.030) (.008)
Victim non-white, 750 & 751 1.52 -.257
(1.01) (.290)
Victim male, 760 8:761 1.23 -.403 _
(1.13) (.316)

 

*Each measure was estimated independent of the other interactive terms, and was included with the

measures from Model 3 in Table 25.

177

Table 05: Mixed logistic regression model for joint-actor data where both drug and

weapon arrests were included as lifester control measures

 

 

Model 7

Variable (N=2649
Beta OR
(S.E.) (Exp( 1_))-l) p-value

Intercept -2.86 .057 < .01
(918)

Incident level control measures

Firearm use 1.05 2.85 < .01
(354)

Number of suspects .768 2.15 < .01
(271)

Drug-motivated incident .269 1 .3 1 .609
(525)

Joint age -.017 .983 .034
(.008)

Non-white intra-racial .447 1.56 .228
(371)

Male-on-male .837 2.3 l .012
(334)

Prior drug arrests .129 1.14 .165
(.093)

Prior weapon arrests -.036 .964 .808
(149)

Neighborhood level measures

Economic deprivation .079 1.08 .601
(151)

Disorder .251 1.29 .136
(168)

Nagelkerke (pseudo) R-square .328

Model 1 2 (df) 74.37 (10)

 

178

Table C-6: Mixed logistic regression model for joint-actor data where only drug

arrests were included as a lifestyle control measure

 

 

Model 8

Variable QQ=264)
Beta OR
(S.E. ) Ex -1 p-value

Intercept -2.95 .052 < .01
(.915)

Incident level control measures

Firearm use 1.04 2.84 < .01
(352)

Number of suspects .761 2.14 < .01
(272)

Drug-motivated incident .323 1.38 .536
(521)

Joint age -.017 .983 .032
(.008)

Non-white intra-racial .418 1 .51 .25 3
(366)

Male-on-male .874 2.39 < .01
(330)

Prior drug arrests .049 1.05 .721
(137)

Neighborhood level measures

Economic deprivation .089 1.09 .551
(150)

Disorder .299 l .35 .069
(164)

Nagelkerke (pseudo) R-square .320

Model 1 2 (df) 72.29 (9)

 

179

Table C-7: Interaction terms added to the mixed logistic regression joint-actors
model

 

 

Variable" Coefficient St. error P-value
Number of suspects*disorder -.113 .252 .653
Drug-motivated incident*disorder -.708 .652 .277
Firearm use*disorder .027 .333 .935
Victim age*disorder -.002 .008 .998
Victim non-white*disorder -.168 .317 .846
Victim male*disorder -.195 .325 .823

 

*Each measure was estimated independent of the other interactive terms in Table C-4,
and was included with the measures from Model 6 in Table 27.

180

Appendix D

181

Table D-l: Pearson’s r correlations between independent variables included in

 

 

victim-only HGLM model
R R R R R R
P-value P-value P-value P-value P-value P-value
N N N N N N
Variable Variable Variable Variable Variable Variable Variable
A B C D E F
A) # of
suspects --
B) Drug- .046
motivated .327 --
incident 447
C) Firearm .028 .060
use .559 .200 --
447 452
D) Age of -.018 -.040 -.305
victim .707 .402 < .001 --
441 446 446
E) Non- -.025 .071 .271 -.210
white victim .598 .133 < .001 < .001 --
439 444 444 440
F) Male .036 .062 .197 -.088 .208
victim .449 .191 < .001 .062 < .001 --
446 45.1 45 1 446 444
G) Disorder .030 .050 -.007 .039 .104 .018
.525 .291 .887 .412 .029 .700
447 452 452 446 444 451
H) Economic -.048 .010 .084 -.014 .256 .051
Deprivation .312 .836 .076 .766 < .001 .277
447 452 452 446 444 451
(con’t)
G & H .353
< .001 -- -- -- --
452

 

182

Table D-Z: Pearson’s r correlations between independent variables included in
joint-actor mixed logistic model

 

 

R R R R R R
P-value P—value P-value P-value P-value P-value
(N = 264)
Variable Variable Variable Variable Variable Variable
Variable A B C D E F
A) # of
suspects --
B) Drug— .074
motivated .233 --
incident
C) Firearm .068 .071
use .272 .251 —-
D) Age of -.022 -.020 -.257
victim .724 .25 1 < .001 --
E) Non- .023 .047 .290 -.218
white victim .716 .448 < .001 < .001 --
F) Male .111 .069 .197 -.120 .190
victim .071 .264 < .01 .051 < .01 --
G) Disorder .070 .064 .008 .056 .134 .029
.254 .302 .901 .367 .029 .639
H) Economic -.054 .006 .070 .027 .297 .009
Deprivation .385 .920 .258 .667 < .001 .884
(con’t)
G & H .364
< .001 -- -- -- --

 

183

 

Figure D-l: Histogram of standardized residuals from the victim-only HGLM
mixed model

 

 

 

 

200
100*
Sld.Dev=2.42
Mean=.1
0 N=435.00
~ ~ 0 e v 0 <9 I I
’oo ‘90 ‘90 Yo 90 -o -o -o o -o 0.0 90
L1RES|D

184

Figure D-2: Histogram of standardized residuals from the joint-actor mixed logistic
model

 

50

Std. Dev = 1.00
Mean = .01
N = 264.00

 

 

 

191?“‘7‘/‘-0.".7.\'?<?‘?

Normalized residual

185

Appendix E

186

Table E-l: Additional HGLM model separatingace and economic deprivation

 

 

Model 9

Variable (N=435)
Beta OR
(S.E.) (Exp( l_1)-l ) p-value

Intercept, 1301

700 (Intercept level-2) -3.83 -- < .01
(815)

Incident level control measures

Firearm use, 131 1.27 3.58 < .01
(.281) (2.58)

Number of suspects, 132 .635 1.88 < .01
(.241) (0.88)

Drug-motivated incident B3 .585 1.79 .172
(.427) (0.79)

Victim age, 134 -.019 .980 .023
(.008) (-.02)

Victim non-white, 135 .444 1.55 .176
(.327) (0.55)

Victim male, 136 .876 2.40 < .01
(.295) (1.40)

Neighborhood level measures

Disorder, 101 .403 1.49 .013
(.154) (0.49)

Deprivation minus pct A.A., 702 .015 1.01 .043
(.007) (0.01)

Percent Afiican American, 703 -.380 .683 .056
(. 194) (-.317)

Deviance Statistic (df) 494.3 (425)

 

187

References

Aldrich, J. H., & Nelson, F. D. (1984). Linear, probability, logit, and probit models.
Newbury Park, CA: Sage Publications.

Anselin, L. (2003). GeoDa 0.9 user's aide. Urbana-Champaign, IL: University of Illinois
Urbana-Champaign.

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.

Berry, W. D., & Feldman, S. (1985). Multiple regression in practice. Newbury Park, CA:
Sage Publications.

Bjerregaard, B., & Lizotte, A.J. (1995). Gun ownership and gang membership. Journal of
Criminal Law and Criminology, 86, 37-58.

Blau, J. R., & Blau, P. M. (1982). Inequality and heterogeneity: A primitive theory of
social structure. Sociological Review, 47. 114-129.

Block, C. R. (1996). Street gangs and crime. Illinois Criminal Justice Information
Authority.

Block, R., & Block, CR. (1993). Street gang crime in Chicago. Research in Brief,
Washington, DC: US. Department of Justice, Office of Juvenile Justice and
Delinquency Prevention.

Blumstein, A. (1995). Youth violence, guns, and the illicit-drug industry. Journal of
Criminal Law and Criminolagy, 86. 10-36.

Blumstein, A., Rivara, F.P., & Rosenfeld, R. (2000). The rise and decline of homicide—
and why. Annual Review of Public fialth. 21 . 505-541.

Blumstein, A., & Rosenfeld, R. (1998). Explaining recent trends in US. homicide rates.
Journal of Criminal Law and Criminology, 88, 1175-1216.

Blumstein, A., & Wallman, J. (2000). The recent rise and fall of American violence. In
A. Blumstein, & J. Walhnan (Editors), The crime drop in America . New York,
MY: Cambridge.

 

Bowker, L. H., Gross, H. S., & Klein, M. W. (1980). Female participation in delinquent
gang activities. Adolescence, 15. 509-519.

Bradford, J. (1987). Geoboamegher's manual. Ft. Collins, CO: Scott Resources.

Braga, A.A. (2008). The prevention of gun homicide through pulling levers focused

188

deterrence strategies, J ournlof Criminal Justice, forthcoming.

Braga, A. A., Kennedy, D. M., Waring, E. J., & Piehl, A. M. (2001). Problem-oriented
policing, deterrence, and youth violence: An evaluation of Boston's operation
ceasefire. Journal of Research in Crime and Delinquency. 38. 195-225.

Brantingham, P. L., & Brantingham, P. J. (1993). Environment, routine, and situation:
Toward a pattern theory of crime. In R. V. Clarke, & M. Felson (Editors), Routine
Activity and Ratioaal Choice: Advances in Criminological Theogy (V 01. 5). New
Brunswick, NJ: Transaction Publishers.

Bullock, H. A. (1955). Urban homicide in theory and fact. Journal of Criminal Law 45,
565-575.

 

Bursik, R. J ., & Grasmick, H. G. (1993). Neighborhoods and crime: The dimensions of
effective community control. Lexington, New York: Lexington Publishers.

Chermak, S., McGarrell, E. F., & Gruenewald, J. (2006). Media coverage of police
misconduct and attitudes toward police. Policing, 29(2), 261-281.

Chesney-Lind, M., Sheldon, R. G., & Joe, K. A. (1996). Girls, delinquency, and gang
membership. In C. R. Huff (Editor), G_angs in America: 2nd Edition (V 01. 29pp.
185-294). Newbury Park, CA: Sage.

Cloward, R. A., & Ohlin, L. E. (1960). Delinquency and opportunity. Glencoe, IL: The
Free Press.

Cohen, J ., & Tita, G. (1998). The role of drug markets and gangs in local homicide rates.
Journal of @antitative Criminology, 2(3), 241-262.

Cohen, L. E., & Felson, M. (1979). Social change and crime rate trends: A routine
activity approach. American Sociological Review. 44. 588—608.

Coleman, J. (1988). Social capital in the creation of human capital. American Journal of
Sociology. 94. 95-120.

Curry, G. D. (1998). Female gang involvement. Journal of Research in Crime an_d
Delinquency. 35. 100-118.

Curry, G. D., & Spergel, I. A. (1988). Gang homicide, delinquency, and community.
Criminology, 26, 381-405.

Decker, S. H., & Van Winke, B. (1996). Life in the gang: famml. fiiends and violence.
New York: Cambridge Publishing.

Decker, S. H., Bynum, T. S., & Weisel, D. (1998). A tale of two cities: Gangs as
organized crime groups. Justice Quarterly. 15. 395-425.

189

Decker, S. H., & Curry, D. G. (2002). Gangs, gang homicides, and gang loyalty:
Organized crimes or disorganized criminals. Journal of Criminal Justice, 30. 343-
352.

DeJong, C. (1997). Specific deterrence and survival analysis: Integrating theoretical and
empirical models of recidivism. Criminology, 35, 561-576.

Durkheim, E. (1979). Suicide: A Stuy in Sociology. New York, NY: Free Press.

Egley, A.H. & Ritz, CE. (2006). Highlights of the 2004 National Youth Gang Survey.
OJJDP Fact Sheet. Washington, DC: United States Department of Justice, Office
of Juvenile Justice and Delinquency Prevention.

Esbensen, F., & Huizinga, D. (1993). Gangs, drugs, and delinquency in a survey of urban
youth. Criminology, 31, 565-586.

Felson, M. (1987). Routine activities and crime prevention in the developing metropolis.
Criminology. 25(4), 91 1-931.

Fox, J. A., & Zawitz, M. W. (2002). Homicide trends in the United States. U_._S.
Department of Justice: Bureau of Justice Statistics.

Garafolo, J ., & Laub, J. (1978). The fear of crime: Broadening our perspective.
Victimology, 3, 242-253.

George, G., & Mallery, P. (2003). SPSS for windows step by step: A simple guide an_d
reference, 11.0 update. Boston, MA: Allyn and Bacon.

Gibson, C. L., Zhao, J ., Lovrich, N. P., & Gaffirey, M. J. (2002). Social integration,
individual perceptions of collective efficacy, and fear of crime in three cities.
Justice Quarterly, 19(3), 537-564.

Harcourt, BE. (2001). Illusion of order: The false promise of broken windows policing,
Cambridge, MA: Harvard University Press.

Hindelang, M. J ., Gottfredson, M. R., & Garofalo, J. (1978). Victims of personal crime:
An empirical foundation for a theory of personal victimization. Cambridge, MA:
Ballinger Publishing.

Henkel, RE. (1976). Tests of significa_n§p. Thousand Oaks, CA: Sage Publications.
Hirschi, T. (1969). Causes of delinquency. Berkley, CA: University of California Press.

Hox, J. (2002). Multivariate analysis: Techniques and applications. Mahwah, NJ:
Lawerence Erlbaum Associates.

Hsiech, C.-C., & Pugh, M. D. (1993). Poverty, income inequality, and violent crime: A
meta-analysis of recent aggregate data studies. Criminal Justice Review, 18. 182-

190

202.

Huff, CR. (1998). Comparing the criminaclbehavgior of youth gangs and at risk youth.
Research in Brief, Washington, DC: US. Department of Justice, Office of
Juvenile Justice and Delinquency Prevention.

Hughes, L. A., & Short, J. F. (2005). Disputes involving youth street gang members:
Micro-social contexts . Criminology. 1. 43-76.

Hunter, A. (1974). Smbolic communities. Chicago, IL: University of Chicago Press.

J ang, S. J ., & Johnson, B. R. (2001). Neighborhood disorder, individual religiosity, and
adolescent use of illicit drugs: A test of multilevel hypothesis. Criminology,
2(1), 109-143.

Jensen, B. F., & Brownfield, D. (1987). Gender, lifestyles, and victimization: Beyond
routine activity. Violence and Victims. 1. 85-89.

Kennedy, D. (1997). Pulling Levers: Chronic Offenders, High Crime Settings, and a
Theory of Prevention. Valparaiso University Law Review, 31. 449-484.

Kennedy, D. M. (2006). Old Wine in New Bottles: Policing and the Lessons of Pulling
Levers. In D. Weisburd, & A. A. Braga (Editors), Police Innovation: Contrastig
Perspectives (pp. 155-170). Cambridge, New York: Cambridge University Press.

Kennedy, D. M., Piehl, A. M., & Braga, A. A. (1996). Youth Violence in Boston: Gun
Markets, Serious Youth Offenders, and a Use-Reduction Strategy. Law and
Contemporary Problems. 59. 147-196.

Klein, M. W. (1995). The American street ggg. Oxford: Oxford Publishing.

Kubrin, C. E., & Stewart, E. A. (2006). Predicting who reoffends: The neglected role of
neighborhood context in recidivsm studies. Criminology. 44(1), 165-188.

Kubrin, C. E., & Wadsworth, T. (2003). Identifying the structural correlates of Afiican
American killings: What can we learn from data disaggregation. Homicide
Studies 7, 3-35.

 

Kubrin, C. E., & Weitzer, R. (2003). New Directions in social disorganization theory.
Joumal of Research in Crime and Delinquency, 40(4), 374-402.

Land, K., McCall, P. L., & Cohen, L. E. (1990). Structural covariates of homicide rates:
Are there any invariances across time and space. American Journal of Sociology,
25, 922-963.

Lauritsen, J. L., & White, N. A. (2001). Putting violence in its place: The influence of
race, ethnicity, gender, and place on the risk for violence. Criminology and Public

Policy, 1(1), 37-60.

191

Liao, T. F. (1994). Integpreting probability models: Logjt, probit, and other generalized
models. Thousand Oaks, CA: Sage.

Lizotte, A.J., Tesoriero, J .M., Thomberry, T.P., & Krohn, MD. (1994). Patterns of
adolescent firearms ownership and use. Justice Quarterly, 11. 51-73.

Loeber, R., & Farrington, D. P. (2001). Child delinquents: Development, intervention,
an_d service needs. Thousand Oaks, CA: Sage Publications.

Long, J. S., & Freese, J. (2003). kgression models for categoricflependent variables
using Stata. College Station, TX: Stata Press.

Martin, S.E., Maxwell, C.D., White, H.R., Zhang, Y. (2004). Trends in alcohol use,
cocaine use, and crime: 1989-1998. The Journal of Drug Issues. 34. 333-359.

Mastrofski, S. D., & Parks, R. B. (1990). Improving observational studies of police.
Criminolggv. 28. 475-496.

Maxson, C. L., & Klein, M. W. (1985). Differences between gang and nongang
homicides. Criminology, 23, 209-222.

Maxson, C. L., & Klein, M. W. (1990). Street gang violence: Twice as great, or half as
great. In C. R. Huff (Editor), Gangs in Ameria (pp. 71-100). Newbury Park, CA :
Sage Publications. ‘

Maxson, C. L. (1999). Gang homicide. In D. Smith, & M. Zahn (Editors), Homicide
studies: A sourcebook of social research . Newbury Park, CA: Sage Publications.

Maxson, C. L., Gordon, M. A., & Klein, M. W. (1985). Differences between gang and
nongang homicides. Criminology, 23, 209-222.

McGarrell, E. F., & Chermak, S. (2003). Problem solving to reduce gang and drug-
related violence in Indianapolis. In S. H. Decker (Editor), Policing gangs and
youth violence (pp. 77-101). Behnont, CA: Wadsworth Publishing.

McGarrell, E. F., Chermak, S., Weiss, A., & Wilson, J. (2001). Reducing firearms
Violence through directed police patrol. Criminology and Public Policy. 1(1), 119-
148.

McGarrell, E. F., Giacomazzi, A. L., & Thurman, Q. C. (1997). Neighborhood disorder,
integration, and the fear of crime. Justice Quarterly. 14( 3), 479-500.

McGarrell, E.F., Giacomazzi, A.L., Thurman, Q.C. (1999). Reducing disorder, fear, and
crime in public housing: A case study of place specific crime prevention. Justice
Regarch and Policy. 1. 61-87.

McGloin, J. M., & Pratt, T. C. (2003). Cognitive ability and delinquent behavior among
inner-city youth: A life-course analysis of main, mediating, and interaction

192

effects. Intemiional Journal of Offender Therapy and Comparative Criminology,
fl, 253-271.

McNulty, T. L., & Bellair, P. E. (2003). Explaining racial and ethnic differences in
adolescent violence: Structural disadvantage, family well-being, and social
capital. Justice Quarterly. 20. 1-31.

‘Messner, S.F., Galea, S., Tardiff, K.J., Tracy, M., Bucciarelli, A., Piper, T.M., Frye, V.,
Vlahov, D. (2007). Policing, drugs, and the new homicide decline in New York
City in the 1990s. Criminology, 45, 385-414.

Miller, J ., & Decker, S. H. (2001). Young women and gang violence: Gender, street
offending, and violent victimization in gangs. Justice Quarterly. 18. 115-140.

Miller W B. (2001). The growth of youth gang problems in the United States.
Washington, DC: OJJDP.

Morenoff, J. D., Sampson, R. J ., & Raudenbush, S. W. (2001). Neighborhood inequality,
collective efficacy, and the spatial dynamics of urban violence. Criminology, 39,
517-560.

Morris, J. (1981). Why black-on-black homicide. Urban League Review. 6(1), 26-34.

Ousey, G., & Lee, M. (2004). Investigating the connections between race, illicit drug
markets, and lethal violence, 1984-1997. Journal of Research in Crime a_n_(_1
Delinquency. 40. 1-32.

Park, R. E., & Burgess, E. (1924). The Introduction to the Science of Sociology. Chicago,
IL: Chicago University Press.

Parker, R. N. (1989). Poverty, subculture of violence, and type of horrricide. Social
Forces 67, 983-1007.

 

Parker, R. N., & Smith, M. D. (1979). Deterrence, poverty, and type of homicide.
American Joumarl of Sociology, 85, 614-624.

Pizarro, J. M., & McGloin, J. M. (2006). Explaining gang homicides in Newark, New
Jersey: Collective behavior or social disorganization? J oumal of Criminal Justice
3, 195-207.

 

Pokomy, A. (1965). Human violence: A comparison of homicide, aggravated assault,
suicide, and attempted suicide. Journal of Criminal Law, Criminology, and Police
Science 56, 488-497.

 

Pratt, T. C., & Cullen, F. T. (2005). Assessing macro-level predictors and theories of
crime: A meta-analysis. hr M. Tonry (Editor), Crime and Justice: A Review of
Research (Volume 32 ed., pp. 373-450). Chicago : University of Chicago Press.

193

Pridemore, W. A. (2002). What we know about social structure and homicide: A review
of the theoretical and empirical literature. Violence and Victims 17(2), 127-156.

 

Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and
data analysis methods. Thousand Oaks, CA: Sage Publications.

Reisig, M. D., McCluskey, J. D., & Mastrofski, S. D. (2004). Suspect disrespect toward
the police. Justice Quarterly. 21(2 ), 241-268.

Reisig, M. D., & Parks, R. B. (2004). Can community policing help the truly
disadvantaged? Crime and Delinquency, 50. 139-167.

Reisig, M. D., & Parks, R. B. (2000). Experience, quality of life, and neighborhood
context: A hierarchical analysis of satisfaction with police. Justice marterly.
1_7(3), 607-630.

Reiss, A., & Tonry, M. (1986). Communities and crime. Chicago, IL: University of
Chicago Press.

Rengert, G. (1996). The geoggaphy of illegal drugs. Boulder, CO: Westview Press.

Riedel, M. (1987). Symposium stranger violence: Perspectives, issues, and problems.
Journal of Criminal LaL and Criminology. 78. 223-258.

Rosenfeld, R., Fomango, R., Rengifo, AF. (2007). The impact of order-maintenance
policing on New York City homicide and robbery rates: 1988-2001. Criminology,
4_5, 355-384.

Rosenfeld, R. B., Bray, T., & Egley, H. A. (1999). Facilitating violence: A comparison of
gang-motivated, gang-affiliated, and non-gang youth homicides. Journal of

Quantitative Criminology, 15, 495-516.

Sampson, R. J. (1986). Crime in cities: The effects of formal and informal social control.
In A. J. Reiss, & M. Tonry (Editors), Communities and Crime (V 01. 8pp. 271-
311). Chicago, IL: University of Chicago Press.

Sampson, R. J., & Groves, B. (1989). Community structure and crime: Testing social-
disorganization theory. American Journal of Sociolcgy. 97. 774-802.

Sampson, R. J., & Laub, J. H. (1993). Crime in the making: Pathwafi and turning points
through life. Cambridge, MA: Harvard University Press.

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(1), 443-478.

Sampson, R. J ., & Morenoff, J. D. (2004). Spatial (dis)advantage and homicide in
Chicago neighborhoods. In (Goodchild, Michael F., & Janelle, Donald G. ),

194

Spatially Integrated Social Science (pp. 145-170). New York: Oxford University
Press.

Sampson, R. J ., Raudenbush, S., & Earls, F. (1997). Neighborhoods and violent crime: A
multilevel study of collective efficacy. Science, 277. 918-924.

Sampson, R. J ., & Wilson, W. J. (1995). Toward a theory of race, crime, and urban
inequality. In J. Hagan, & R. D. Peterson (Eds), Crime and Inequality (pp. 37-
54). Palo Alto, CA: Stanford University Press.

Shadish, W. R., Cook, T. D., & Campbell, D. T. (2002). Experimental and guasi- ’
experimentgdesigpa for ganeralized causal inference. Boston, MA: Houghton
Mifflin Company.

Shaw, C. C., & McKay, H. D. (1942). Juvenile delinquency in urban areas. Chicago, IL:
University of Chicago Press.

 

Sikes, G. (1997). 8 Ball Chicks: A year in the violent world of girl gangsters. New York,
NY: Anchor Books.

Skogan, W. G. (1987). Disorder and community decline. Evanston, IL: Center for Urban
Affairs and Police Research, Northwestern University.

Skogan, W. G. (1990). Disorder and decline: Crime and the spiral decay of American
cities. New York: Free Press.

Skogan, W. G. (2006). The promise of community policing. In D. Weisburd, & A. A.
Braga (Editors), Police Innovation: Contrasting Perspectives (pp. 27-43).
Cambridge, New York: Cambridge University Press.

Skogan, W. G., Hartnett, S. M., DuBois, J ., Comey, J. T., Kaiser, M., & Lovig, J. (1999).
On the beatfi: Police and comqu problem solving. Boulder, CO: Westview
Press.

Skolnick, J. (1990). The social structure of street drug dealing. American Journal of
Police 9, 1-41.

 

Smith, D. A., & J arjoura, G. R. (1989). Household characteristics, neighborhood
composition and victimization risk. Socgl Forces, 68. 621-640.

Sousa, W. H., & Kelling, G. L. (2006). Of "Broken Windows," Criminology, and
Criminal Justice. In D. Weisburd, & A. A. Braga (Editors), Police lnnovflon:
Contrasting Perspectives (pp. 7 7-97). Cambridge, New York: Cambridge
University Press.

Spergel, 1., Curry, D., Chance, R., Kane, C., Ross, R., Alexander, A., Simmons, E., &

Oh, S. (1994). Gang suppression and intervention: Problem and responaa.
Washington, DC: Office of Juvenile Justice and Delinquency Prevention.

195

Surratt, H. L., Inciardi, J. A., Kurtz, S. P., & Kiley, M. C. (2004). Sex work and drug use
in a subculture of violence. Crime and Delinquency. 50. 43-59.

Sykes, G. (1974). The rise of critical criminology. J omnal of Criminal Law and
Criminology, 65, 206-213.

Taylor, RB. (2006). Incivilities reduction policing. In D. Weisburd, & A. A. Braga
(Editors), Police Innovation: Contrasting Perspectives (pp. 98-114). Cambridge,
New York: Cambridge University Press.

Taylor, R. B. (1999). The incivilities thesis: Theory, measurement, and policy. In R. H.
Langworthy (Editor), MeaaurirgWhaa Matters (pp. 65-88). Washington, DC:
National Institute of Justice/Office of Community Oriented Policing Services.

Taylor, R. B. (1997). Social order and disorder of street blocks and neighborhoods:
Ecology, microecology, and the systemic model of social disorganization. Journal
of Regarch in Crime ad Delinquency. 34(1), 113-155.

Taylor, R. B., & Harrell, A. V. (1996). Physigll environment and crime. Washington,
DC: National Institute of Justice.

Thomberry, TR (1998). Membership in youth gangs and involvement in serious violent
offending. In R. Loeber and D. P. Farrington (Editors), Serious and Violent
Juvenile Offenders: Rislfltoraami Successful Interventiopa. Thousand Oaks,
CA: Sage Publications

Thomberry, T. P., Krohn, M. D., Lizotte, A. J ., & Chard-Wierschem, D. (1993). The role
of juvenile gangs in facilitating delinquent behavior. J oumg of Resegch in Crime
and Delinquency. 30. 55-87.

Thrasher, F. (1927). The gang. Chicago, IL: University of Chicago Press.

Tita, G. E., Cohen, J ., & Engberg, J. (2005). An ecological study of the location of gang
"set space". Social Problema,_2_, 272-299.

Vigil, D. (1988). Banio gangs. Austin, Texas : University of Texas Press.

Wakeling, S. (2003). Ending gang homicide: Deterrence can work (Perspectives on
Violence Prevention No. 1). Sacramento, California Attorney General’s Office.

Walters, G. D. (2006). Risk-appraisal versus self-report in the prediction of criminal
justice outcomes. Criminal Justice and Behavior. 33( 3), 279-304.

Webb, V. J ., & Katz, C. M. (2003). Policing gangs in an era of community policing. In S.
H. Decker (Editor), Policig gangs and youth violence (pp. 17-43). Belmont, CA:
Wadsworth Publishing.

 

Weisburd, D., & Braga, A. A. (2006). Hot spots policing as a model. In D. Weisburd, &

196

A. Braga (Editors), Police innovation: Contrastingperspectives (pp. 225-244).
Cambridge, NY: Cambridge University Press.

Weisburd, D., & Braga, A. A. (2006). Hot spots policing as a model for police
innovation. In D. Weisburd, & A. A. Braga (Editors), Police Innovation:
Contpasting Perspectives (pp. 225-244). Cambridge, New York: Cambridge
University Press.

Wilson, J. Q. (1975). Thinking about crime. New York: Basic Books.

Wilson, J. Q., & Kelling, G. L. (1982). Broken windows: The police and neighborhood
safety. Atlantic 249(3), 29-38.

 

Wilson, W. J. (1987). The truly disadvantaged. Chicago, IL: University of Chicago Press.

Wolfgang, M. E., & Ferracuti, F. (1967). The subculture of violence: Towards an_
integgated theog) in criminology. London: Tavistock.

Wolfgang, M. (1958). Patterns in criminflromicide. Philadelphia, PA: University of
Pennsylvania Press.

Wright, R. T., & Decker, S. T. (1997). Armed robbers in arction. Boston, MA:
Northeastern University Press.

 

Zimring, F. E., & Hawkins, G. (1997). Crime is not the problem: Lethal violence in
America. New York, N.Y.: Oxford University Press.

197