The University of Chicago Press

Murder by Structure: Dominance Relations and the Social Structure of Gang Homicide
Author(s): Andrew V. Papachristos

Source: American Journal of Sociology, Vol. 115, No. 1 (July 2009), pp. 74-128
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/597791 .
Accessed: 23/08/2011 14:09
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to
American Journal of Sociology.
http://www.jstor.org
Murder by Structure: Dominance Relations
and the Social Structure of Gang Homicide1
Andrew V. Papachristos
University of Massachusetts, Amherst
Most sociological theories consider murder an outcome of the dif-

ferential distribution of individual, neighborhood, or social char-
acteristics. And while such studies explain variation in aggregate
homicide rates, they do not explain the social order of murder, that
is, who kills whom, when, where, and for what reason. This article
argues that gang murder is best understood not by searching for its
individual determinants but by examining the social networks of
action and reaction that create it. In short, the social structure of
gang murder is defined by the manner in which social networks are
constructed and by people’s placement in them. The author uses a
network approach and incident-level homicide records to recreate
and analyze the structure of gang murders in Chicago. Findings
demonstrate that individual murders between gangs create an in-
stitutionalized network of group conflict, net of any individual’s
participation or motive. Within this network, murders spread
through an epidemic-like process of social contagion as gangs eval-
uate the highly visible actions of others in their local networks and
negotiate dominance considerations that arise during violent
incidents.
INTRODUCTION
Most sociological theories of murder seek general “laws” or social “causes”
and rely heavily on heterogeneity or stratification as the main correlates
1
Of the many people who provided feedback on this project, I would like to especially
thank Mark Mizruchi, Tracey Meares, Jeff Fagan, Richard Taub, James Short, Damon
Phillips, Ron Burt, John Padgett, Edward Laumann, Robert Sampson, Jason Beckfield,
David Kirk, Andrea Leverentz, Don Tomaskovic-Devey, Emily Erikson, and the AJS
reviewers for their extensive comments. Part of this research was supported by National
Science Foundation grant SES-021551 to the National Consortium on Violence Re-
search (NCOVR). Direct correspondence to Andrew Papachristos, Department of So-
ciology, University of Massachusetts, Thompson Hall 728, 200 Hicks Way, Amherst,
Massachusetts 01003. E-mail: andrewp@soc.umass.edu
䉷 2009 by The University of Chicago. All rights reserved.

0002-9602/2009/11501-0003$10.00
74 AJS Volume 115 Number 1 (July 2009): 74–128

Murder by Structure
of homicide. The usual study attempts to explain differences in aggregate

homicide rates at the neighborhood, city, or state level by using individual
or contextual attributes such as age structure, economic conditions, eth-
nicity, or neighborhood composition. And indeed, despite some variation,
we know from these studies that individuals of minority groups, especially
males between the ages of 17 and 28, who live in poor, isolated neigh-
borhoods bereft of social and human capital are the most likely perpe-
trators and victims of murder. Gang membership is considered just an-
other risk factor that increases one’s likelihood of killing or being killed.
Yet neither individual characteristics nor social conditions kill people.
“Youth” does not pull a trigger nor anomie strangle a victim. Murder is
an action. And, in fact, the vast majority of people whom our models
identify as at risk of committing homicide never resort to it. That is,
aggregate-level theories cannot explain why particular people kill, how
their victims are selected, or what drives them to murder in the first place.
Our aggregate approach sees motive and murder as almost random in-
stantiations of demographic and social conditions that we call risk factors.
In such a view, individuals muddle through a sort of Hobbesian chaos,
albeit in particular types of places and under certain circumstances. As
a result, there is little rhyme or reason to who kills whom.
But we know that murder is not in fact such a random matter. It is
first and foremost an interaction between two people who more often than
not know each other: approximately 75% of all homicides in the United
States from 1995 to 2002 occurred between people who knew each other
prior to the murder (Federal Bureau of Investigation, selected years). We
also know that the victim and offender tend to resemble each other socially
and demographically (e.g., Wolfgang 1958; Luckenbill 1977). Young peo-
ple kill other young people, poor people kill other poor people, gang
members kill other gang members, and so on. Thus, contrary to strati-
fication theories, a particular murder is not so much the outcome of the
differential distribution of attributes as it is an interaction governed by
patterns of social relations between people similar in stature and status.
But if murder is an interaction, then we must ask how such interactions
are patterned; what types of social relations produce murder; and how,
if at all, these individual interactions organize larger patterns of murder.
In this article, I argue that gang murder is best understood not by
searching for its individual determinants but by examining the social
networks of action and reaction that create it. Gang members do not kill
because they are poor, black, or young or live in a socially disadvantaged
neighborhood. They kill because they live in a structured set of social
relations in which violence works its way through a series of connected
individuals. The gang qua group carries with it a set of extra-individual
adversaries and allies that shape individual choices of action, including
75
American Journal of Sociology
the selection of murder victims. As corporate actions between groups,

gang murders do not end with the death of the victim but persist in the
organizational memory of the gang, which is governed by norms of re-
taliation and violent mechanisms of social control. Gang murder occurs
through an epidemic-like process of social contagion as competing groups
jockey for positions of dominance, and aggregate patterns of murder arise
as these individual disputes create a network of group relations that shape
future patterns of conflict, collective action, and murder.
I shall demonstrate this interactional quality of gang murder by ap-
plying a network approach to incident-level homicide records. With these
data, I shall reveal the social structure of gang homicide in Chicago by
analyzing the mechanisms through which dominance disputes give rise
to stable networks of violence. My findings suggest that gangs are not
groups of murderers per se, but rather embedded social networks in which
violence ricochets back and forth. Individual murders between gangs
create an institutionalized network of group conflict—sustained patterns
of interaction—net of any individual’s participation or motive. Within
this network, murders spread through a process of social contagion as
gangs respond to threats by evaluating the highly visible actions of others
in their local networks. Individual murders, especially those public in
nature, directly threaten the social status and ranking of groups, thus
signaling to the gang and others in the social context that a threat has
occurred. Additionally, norms of reciprocity intimately link matters of
social status with vengeance-seeking behavior and the desire to avoid
subjugation to other gangs. Gangs must constantly (re)establish their social
status through displays of solidarity—in this case, acts of violence—which,
in turn, merely strengthen these murder networks.
I begin by developing a general framework of homicidal transactions
to show how the consequences of individual violent episodes shape future
patterns of murder. Then, I discuss how such interactions in the gang
context create murderous networks by placing adversaries in positions
where each must attempt to defend, maintain, or repair their reputation
by constantly negotiating dominance disputes. Finally, I use descriptive
and statistical network techniques to analyze such claims. Individual gang
murders, especially those that are reciprocal in nature, create and sustain
group relations by continually defining the nature and direction of inter-
group relations. Violence spreads to other gangs as status disputes arise
and are settled with violence. What begins as a single murder soon gen-
erates a dozen more as it diffuses through these murder networks.
76
Murder by Structure
MURDER AS INTERACTION: CONSEQUENTIALITY AND SOCIAL

STATUS
Interaction-based studies show a rather consistent sequential patterning
of interpersonal violence that coincides with Goffman’s (1967) concept of
“face” and its associated processes of face work (e.g., Felson 1981, 1982;
Fagan and Wilkinson 1998).2 In general, violence unfolds as a dynamic
process of exchange and interpretation between disputants and, quite
often, third parties. Homicide in particular has a standard interactional
form, irrespective of age, race, gender, location, or motive: opponents in
a confrontation seek to establish or maintain face at the other’s expense
by remaining steady in a contest in which violence is or has become an
acceptable means for settling the dispute (Luckenbill 1977; Polk 1999).
As an interaction, then, murder results from a give-and-take process in
which the victim and offender negotiate appropriate face-saving behav-
iors.3 Forward-looking actors know that the consequences of such violent
exchanges extend beyond the immediate transaction and must consider
the consequentiality of their actions as they negotiate competing individ-
ual interests—such as self-preservation or monetary gain—versus context-
specific behavioral expectations—such as norms of toughness and respect
(e.g., Felson 1981; Fagan and Wilkinson 1998; Anderson 1999; Phillips
2003; Jacobs 2004).
The most obvious consequence of a lethal exchange is, of course, death.
Yet, actors in violent exchanges often seek not only to save their lives but
also to minimize future victimization and to assert their social standing—
behavioral patterns common in dominance hierarchies of animal and hu-
man small groups (Mazur 1973a; Chase 1980; Gould 2003). Ivan Chase’s
(1980) classic example of “pecking orders” among chickens provides a
useful example. Chase finds that winning or losing a pecking contest
affects the social position of the victorious and vanquished chickens. A
chicken that wins a pecking contest is less likely to become a victim in
2
Goffman (1967) defines face as “the positive social value a person effectively claims
for himself by the line [verbal and nonverbal acts] others assume he has taken during
a particular contest” (p. 5). Processes of face work—those of maintaining, presenting,
or saving face—are by definition reciprocal and interactive. A sequence is formed in
which each interaction influences the subsequent one within the given transaction and
beyond, or what Goffman calls “consequentiality” (p. 157).
3
Even murders motivated by instrumental or material matters tend to follow the
microprocesses described here insofar as the interaction between the victim and per-
petrator is dynamic and may contribute to the lethal outcome. For example, compliance
with a robber, such as quickly relinquishing one’s wallet, tends to decrease the esca-
lation of violence. Moreover, expressive matters such as status or face are often attached
to instrumental motives; i.e., one commits a robbery not to put the proverbial bread
on the table but for the cash in hand needed for culturally prescribed consumption
patterns (Jacobs and Wright 1999; Topalli, Wright, and Fornango 2002).
77
future attacks precisely because it has proven its dominance; often, a

victorious chicken is never challenged a second time. However, a losing
chicken has two possible pecking-order trajectories. On the one hand, a
defeated chicken can become a victim in subsequent attacks; that is, other
chickens observe the first defeat, see the defeated chicken as vulnerable,
and target that chicken as an easy opponent. Such a chicken is a “double
loser” and is relegated to the bottom of the pecking order. On the other
hand, a defeated chicken can pursue a more aggressive position against
a nondominant chicken with the hopes of reestablishing dominance over
at least one other chicken. In this case, a one-win/one-loss chicken ends
up in the middle of the pecking order. Although this latter attack is in
no way retaliation—rarely do chickens seek revenge—a losing chicken
can regain some status relative to others in the group through outward
acts of aggression and displays of dominance.
In the world of chickens, the consequences and actions of violent ex-
changes are rather straightforward: one’s capacity to successfully domi-
nate another is directly related to one’s social standing. And while to a
certain degree this type of self-preservation is true of human societies, the
consequences of violent exchanges become increasingly complex when
symbolic matters—such as honor or reputation—are at stake. Unlike ma-
terial objects such as a loaf of bread or a parcel of land, which are useful
to their possessors independent of any context-derived meaning, concepts
such as honor, status, and reputation are entirely symbolic in that they
are predicated almost entirely on recognition by others in a social context.
In short, symbolic matters such as honor are manifested in face-to-face
interaction and bestowed by the group in accordance with their evaluation
of said interactions: to have honor is to be thought honored by others
(e.g., Bourdieu 1966; Pitts-Rivers 1966; Gould 2003).
In the United States, notions of honor are generally associated with
acts of hypermasculinity, especially the use of violence to protect one’s
reputation (Wyatt-Brown 1982; Katz 1988; Nisbett and Cohen 1996; An-
derson 1999; Polk 1999; Wilkinson 2003). Any honorable man protects
his reputation, and by extension his perceived social status and self-worth,
by exhibiting a “ferocity of will” that includes a willingness to engage in
violent face-saving behavior when his honor is assaulted (Miller 1993;
Stewart 1994). In this way, insults become deadly precisely because their
consequences directly influence the social standing of the disputants. For-
malized systems of honorific and status-conferring violence—such as a
duel or vendettas—more frequently emerge in social contexts in which
(1) formal institutions of social control are absent or impotent and (2)
violence is condoned or promoted as an acceptable form of social control
78
Murder by Structure
or “self-help” (Black 1983).4 And, modern street culture produces just such
a context for the criminal underworld and the urban gang, where violence
serves as a primary currency of honor and social status.5
MURDER AS A GIFT: RECIPROCITY, DOMINANCE, AND THE

SOCIAL CONTAGION OF VIOLENCE IN MODERN STREET
CULTURE
Life in many inner-city neighborhoods is governed by a “code of the
streets”—a set of informal behavioral precepts organized around a search
for respect that regulates public interaction (Anderson 1999; see also Wil-
kinson and Fagan 1996). A strong reputation is displayed through inter-
actions that demonstrate one’s willingness to use violence to maintain
face or gain respect. Quick wits and quick fists confer status. The timid
or trigger shy become suckers, punks, or victims. Violence is proof positive
of the respect one claims to have. Like pecking contests among chickens,
public displays of one’s willingness to use violence (as well as other non-
violent gestures, such as attire, body language, eye contact, etc.) are signals
of one’s street savvy and toughness with the main purpose of deterring
would-be attackers. On the street, the outcome of any single dispute is
consequential and cumulative. Failure to act in—or win—a given contest
not only diminishes one’s social standing vis-à-vis one’s opponent but
also makes one appear weak, a potential target for future street
interactions.
One of the street code’s most pervasive norms is that of retribution, a
perversion of the “golden rule” stipulating that personal attacks (verbal
or physical) should be avenged (Anderson 1999; Jacobs 2004; Jacobs and
Wright 2006). This threat of vengeance lingers over street interactions,
promising lex talionis payback for offenses made to one’s safety or honor.
As in other honor cultures, retaliatory street violence binds social status
4
The rich ethnographic and historical literature on conflict and violence supplies nu-
merous examples to this point, such as duels in the American South or 18th-century
England (Greenberg 1990; MacAleer 1994; Otterbein 1994); vendettas in Mediterranean
honor cultures (Blok 1974; Gambetta 1993; Gould 1999, 2000); codes of honor among
Japanese samurai (Ikegami 1995); status disputes in tribal and rural societies more
generally (Gluckman 1955; Bourdieu 1966; Pitts-Rivers 1966; Ginat 1987; Greenberg
1989; Villarreal 2002; White 2004); and codes of respect and honor among young males
in urban America (Horowitz 1983; Bourgois 1995; Anderson 1999).
5
Examples of the use of honor codes and violence in the criminal underworld include
organized crime (Gambetta 1993; Baxter and Margavio 2000); drug dealers and users
(Padilla 1992; Bourgois 1995; Jacobs 1999); professional thieves, robbers, and burglars
(Wright and Decker 1994; Jacobs and Wright 1999; Topalli et al. 2002); and urban
street gangs (Short and Strodtbeck 1965; Horowitz 1983; Vigil 1988; Decker and Van
Winkle 1996; Hughes and Short 2005).
79
and honor together through the norm of reciprocity. In this sense, inter-
personal violence can be thought of as a “gift,” a gesture that, if accepted,
demands to be reciprocated. Regardless of whether a gift is positive or
negative, an honorable person returns gifts, thus pointing to the “rude
truth that not only hospitality engenders obligations to reciprocate, but
that offenses and assaults do too” (Miller 1993, p. 5). On the street, those
who reciprocate the gift of violence are better equipped to maintain their
social standing as “reputable” and “honorable” persons, as well as deter
future transgressions (Jacobs 2004).6
Disputes about honor and reputation may ultimately be founded on
issues of dominance—that is, claims between disputants concerning who
lords power or status over whom.7 In the posthumously published Col-
lision of Wills, Roger Gould (2003) argues that interpersonal violence most
often arises from disputes over dominance, especially in relations in which
social ranking is unclear or cannot be solved using external cues. Ac-
cording to Gould, lethal violence occurs more frequently in symmetric
relations, rather than asymmetric or hierarchical ones, because social
status, normative behaviors, mechanisms of social control, and rules of
deference (basic parameters of any social transaction) are more ambiguous
and therefore more subject to dispute. In hierarchical relations—such as
employer-employee, parent-child, police-citizen, and teacher-student—
dominance is well defined and backed by social norms and institutions
of social control. In contrast, symmetric relations—such as friend, neigh-
bor, classmate, co-worker, or acquaintance—are by definition equitable;
that is, one party cannot unilaterally claim dominance over the other.8 A
dominance claim by one party in a symmetric relation is consequential
in that it potentially sets precedent for future interactions, including the
subjugation of one party or the ascension of the other party to a position
of power.
Dominance contests are exacerbated in the group context because mur-
6
George Fenwick Jones (1959) captures this ethic in his rephrasing of biblical aphorisms
into the language of honor: “Blessed are those who wreak vengeance, for they shall
be offended no more, and they shall have honor and glory all the days of their lives
and eternal fame in ages to come” (p. 40).
7
I use the term “dominance” similarly to Gould (2003) in reference to the general
distribution of power in a relation, i.e., who decides the overall direction of or action
in a relationship.
8
As Gould (2003, p. 71) writes, “In relations approximately this type [symmetric], it
is not easy to say how differences will most likely be resolved. Because each party to
a dispute has an equivalent basis on which to claim the right to have a say, neither
has an agreed-on right to override the views of the other. It makes no sense to say,
‘You must stand aside as I walk by because, as your neighbor, I demand that you do
so.’ Absent some other criterion such as seniority or status, people who are neighbors
cannot demand unilateral deference from each other.”
80
Murder by Structure
derous interactions and the networks they create are more likely to be
sustained over time and space. In the group context—that is, warring
countries, feuding families, or fighting street gangs—individual murders
can be framed as a threat to group solidarity and lead to social patterns
that elicit further violence (e.g., Short and Strodtbeck 1965; Decker 1996;
de la Roche 1996; Gould 1999). Disputes often become intrinsically col-
lective because the group regards an offense against a member as an
offense against all, a sentiment that fosters in-group cohesion as a function
of confronting external threats (Hughes [1948] 1991; Shils and Janowitz
1948; Pfautz 1961; Sherif 1967). Collective and, by extension, individual
honor derives from the ability of the group to successfully retaliate and
from its capacity to protect its members.9 Back-and-forth retaliation could
occur indefinitely with groups defining their existence in reference to their
enemies and allies, thus producing enduring conflict structures.
As individuals and groups get caught up in these types of dominance
contests, violence spreads through a process of social contagion that is
fueled by normative and behavioral precepts of the code of the street.
The social contagion of violence occurs either when (a) disputes are sus-
tained over time or violent acts between opposing parties escalate or when
(b) acts of aggression diffuse outward from the original source of conflict
to incite new violent acts between or toward nondisputants (Loftin 1986;
Fagan, Wilkinson, and Davies 2000; Jacobs 2004). The first mode of
contagion refers to direct retaliation, as in the types already discussed:
the retribution for a past dispute by the aggrieved or a member of the
aggrieved’s group against the person or group responsible for the original
affront. The second mode of contagion refers to a process of generalized
violence, the spread of violence to those not directly involved in the initial
dispute (Jacobs 2004). As in the one-win/one-loss chicken example, gen-
eralized violence functions as a way to restore one’s face by drawing
attention to acts that, although they may not be a part of the original
dispute, serve as status-conferring actions—a sort of collective liability in
which innocent victims (nondisputants) serve as proxies for past wrongs
(Black 1983; Gould 1999; Jacobs and Wright 2006).
I maintain that the emotional and practical consequences of a homicide
unfold through such contagion processes as individuals and groups get
caught up in struggles over dominance and social status. Furthermore, I
argue that the spread of violence and its consequences create a network
structure between disputants—sustained patterns of animosity, conflict,
and interaction. Put another way, individual homicides link together to
9
Anderson (1999) describes it this way: “Part of what protects a person is both how
many people can be counted on to avenge his honor if he is rolled on in a fight and
who these defenders are—that is, what their status on the street is” (p. 73).
81
form an overarching contentious network that acts as a conduit for future

action. I hypothesize that individual murders—especially those occurring
in group contexts that stress the use of violence and the norm of reciprocity
as status-conferring mechanisms—create or sustain such structures since
the consequences of murder trigger normatively prescribed (re)actions
from connected groups involved in dominance disputes. Yet, how do such
murder networks arise? What do they look like? Are they enduring struc-
tures or merely temporal aberrations? And, what mechanisms produce
them? To answer these questions, the present study employs formal net-
work methodology to analyze one such contagious network—gang
conflict.
FROM GANG MURDER TO SOCIAL STRUCTURE: A NETWORK

APPROACH
The group nature of gang activity generates patterns of interaction, daily
behaviors, and belief systems that are particularly well suited to the cre-
ation of contentious networks and the social contagion of violence. Struc-
turally, the gang demands certain microbehavior of its members (e.g.,
hanging out, dealing drugs, fighting, partying, and so forth) that de facto
increase a member’s exposure to potentially violent situations (e.g., Short
and Strodtbeck 1965; Decker and Winkle 1996; Thornberry et al. 2003).
More important, however, honorific and status concerns are amplified in
the gang context not only because of the prevalence of broader street
culture but also because of internal group processes. In short, the gang
exists in a social milieu that emphasizes the idea of honor and promotes
the use of violence as a key status-conferring behavior.
At the individual/gang member level, status considerations are directly
associated with hypermasculine ideals of toughness and fighting prowess,
as well as a general penchant for mayhem and mischief (Miller 1958;
Short and Strodtbeck 1965; Horowitz and Schwartz 1974; Vigil 1988;
Hughes and Short 2005). At the group/gang level, collective honor is a
function of a group’s cohesion and ability to fend off perceived threats
(Short and Strodtbeck 1963; Kobrin, Puntil, and Peluso 1967; Decker
1996). Similar to the omnipresent threat of violence in the wider street
culture, the threat of attack creates a perpetual “myth” in which the gang
must constantly protect itself and its members (Katz 1988; Decker and
Winkle 1996). In turn, threats from outside the gang can bolster group
processes, generate internal cohesion, and solidify gang identity, structure,
and morale (Thrasher 1927; Short and Strodtbeck 1965; Klein and Craw-
ford 1967; Suttles 1968; Hagedorn 1988; Jankowski 1991). Furthermore,
perceived threats may in fact be an important causal dimension in gang
82
Murder by Structure
formation since mutual protection is one of the most frequently cited

reasons for joining a gang (e.g., Esbensen and Huizinga 1993; Decker and
Curry 2000; Thornberry et al. 2003).
The consequentiality of such threats stems from the fact that a gang’s
identity and social status are, in large part, defined by and through its
relationships to other neighborhood actors, but especially to other gangs.
For my purposes, I rely on Klein and Maxson’s (2006) definition of a
“street gang” as “any durable, street-oriented youth group whose involve-
ment in illegal activities is part of its group identity” (p. 4).10 A gang’s
reputation—its main currency of social standing—is assessed through a
dynamic process of exchange with those in its local context. In general,
an “honorable” gang is one that protects its members, acts as a cohesive
group, and abides by the “golden rule” of the street that requires retri-
bution for slander against the gang’s name, reputation, or turf (Short and
Strodtbeck 1965; Kobrin et al. 1967; Horowitz 1983; Vigil 1988; Decker
1996).11 External threats call into question whether or not a gang can or
will abide by such ideals, and a gang’s response to such threats will guide
10
The definitional debate in the gang literature is exhaustive to say the least (see Bursik
and Grasmick 1993; Spergel 1995). The pivotal issue in this debate is the extent to
which gangs should be defined by their criminal and deviant activity. Some scholars,
such as Short (1997), purposefully exclude deviant behavior as a defining characteristic
to avoid the problem of including in the definition the behavior that is being explained.
Such definitions tend to focus on activities and processes that define the gang as a
collective: hanging out, movement through space, group-defined organizational bound-
aries, membership criteria, etc. Other scholars include such collective attributes but
use levels of violence as the vital dimension that distinguishes unique gang types (Klein
1995; Knox 2001). In the present study, this debate is somewhat moot. On the one
hand, I am explicitly looking at collective processes in the gang that define any in-
dividual group. By these standards, all the gangs I study possess the internal qualities
of process-oriented definitions. On the other hand, it is violent—and, by legal defini-
tions, criminal—action on which I focus. But, as my analysis shows, violence is dynamic
and reciprocal. Thus, violence is both a defining trait and the result of such group
processes. The two are inseparable, although we like to think that they can be pulled
apart for pedagogical purposes (see Klein and Maxson [2006] for a recent discussion
along similar lines). I argue that focusing on observable behaviors and interactions
can show us something more tangibly useful about the group processes that make the
gang a distinct social entity.
11
A unique study of the social status of street gangs by Kobrin and colleagues (1967)
underscores this point. Kobrin et al. asked community residents, criminal justice of-
ficials, and gang members in a Chicago neighborhood to rank all the gangs in that
area according to various status criteria. The study finds that the social standing of
any individual gang is a function of perceived fighting ability, notoriety for being
“tough,” and a reputation for “looking for fights.” More impressively, evaluations of
high- and low-status gangs are consistent across the criminal justice officials, com-
munity residents, and gang members, suggesting that those inside and outside the gang
world recognize that the social status of a gang is directly related to its willingness to
use violence for honorific purposes and adherence to behavioral rules such as those
of contemporary street culture.
83
the status evaluations of others. This implies that a gang must consider
not only its immediate opponent but also how its actions or inactions will
be evaluated by others in the social context, including future adversaries
and allies.
Drawing on Gould’s theory of violence, my argument is that violence
among gangs is centered on concerns of dominance. Gang murder is, to
say the least, a conflictual interaction between members of two groups
that potentially threatens the social status and dominance positions of
those involved. Failure to respond to a threat may result in subjugation
to an enemy and, thus, lessens one’s social standing, honor, or reputation.
In contrast, a willingness to use violence—especially retaliatory violence—
brands a gang as worthy of respect and can possibly restore a damaged
reputation. If a gang frames an action as a threat, it must coordinate
group action to redress the threat in order to (re)establish the social order.
Otherwise, the group loses status vis-à-vis its opponents and others in the
network and, in turn, increases the likelihood of subsequent attacks or
future victimization. As a result, gang murder is often retaliatory in nature,
an attempt at achieving a sort of street justice (Maxson 1999; Kubrin and
Weitzer 2003).
Consistent with the framework put forth here and the dynamics of gang
violence more generally, I hypothesize that individual gang murders create
a lasting social network that influences subsequent patterns of violence.
I further hypothesize that dominance disputes between groups are largely
responsible for the creation of these murder networks. In other words,
when gangs get caught in dominance contests, norms of reciprocity and
the use of violence as status-conferring behavior fuel the social contagion
of violence. If, as I maintain, the gang defines itself through such con-
flictual interactions and dominance contests, then such structures should
persist over time. In a sense, there is a “gang effect” to the degree that
being a member of a particular gang carries with it a structured pattern
of relations.12 Gang members inherit this network and use it as a schematic
for their violent actions. While an individual member pulls the trigger, it
is the structure that determines who kills whom.
This approach posits that the gang exerts its influence on its members
12
By “gang effect” I mean some emergent property of the gang net of differences in
attributes of its members that influences individual and collective behavior. Prior
research shows a facilitative effect associated with gangs: members engage in a greater
range and amount of deviant activity during periods of membership as compared to
non–gang members (Short and Strodtbeck 1965; Esbensen and Huizinga 1993; Thorn-
berry et al. 2003). Yet, these studies face many of the problems associated with meth-
odological individualism in that the unit of analysis tends to be exclusively the gang
member. Such studies rarely measure properties, structures, or group processes of the
gang that facilitate such behavior.
84
Murder by Structure
precisely because the existence of the group is predicated on the inter-

dependence of its members. To the extent that the gang exists as a group,
its patterns of interaction must exist beyond the actions of any single
individual. In other words, the persistence of the social group demands
that certain behavioral patterns, norms, culture, and activities should be
sustained across time and space (Simmel 1898). The transmission of the
group manifests itself in similarities (or differences) in the behavior of its
members at different times. If group processes are at work—if there is a
gang effect—the structure it creates should have an influence on member
behavior above and beyond individual attributes or motives.
Reciprocal or persistent murders between groups would support these
hypotheses by suggesting that the organizational memory or collective
capacity of the gang is strong enough to incite future violence beyond the
individual event or the participation of any single member. The absence
of sustained or reciprocal interactions between groups falsifies such a view.
The alternative hypothesis implies a sort of isolated murder case: gang
members kill each other with limited or little effect on other member or
group activities. In this case, a murder would appear in a network at any
one point in time but would not appear in subsequent time periods or
spark reciprocal activities between groups. If such interactions were the
result of isolated interactions or individual motives, they would not affect
larger group-level actions and perhaps even question the existence of the
gang as a unique social group (e.g., Fleisher 2005; Sullivan 2005). In other
words, homicide would be a manifestation of individual motives and
therefore would not necessarily elicit further collective (group) action or
produce an effect on larger patterns of gang relations. Murder would not
create a “structure” in the network sense of the word used here. As such,
and consistent with the framework just discussed, the ensuing analysis
begins with the homicide incident—the specific details of an individual
homicide, in particular, who kills whom—and creates the global structure
from the bottom up.
DATA
The main source of data used in this article is two independent data sets
provided by the Chicago Police Department.13 The first set of data includes
records of all homicides in Chicago from 1994 to 2002. Data are recorded
at the incident (dyad) level and contain demographic, geographic, motive,
13
Data were provided by the Chicago Police Department’s Division of Research and
Development. The analysis of the data reflects the opinions of the author and in no
way represents the views of the Chicago Police Department or the City of Chicago.
85
and gang information on both the offender and victim as recorded by

homicide detectives.14
In the present analysis, I select all cases in which either the victim or
the offender in the homicide is identified as a gang member. This choice
requires some clarification since a debate exists surrounding the classifi-
cation of a “gang related” homicide (for a review, see Maxson and Klein
[1996]). Some municipalities rely on a conservative definition that classifies
a homicide as gang related only if the crime itself was motivated by gang
activity, such as turf defense, drug dealing, or prior gang conflicts. Other
municipalities use a member-based definition that classifies any homicide
as gang related if either the victim or the offender was a gang member.
Because I am interested in group violence, the former strategy errs on
the side of sampling too heavily on the dependent variable by capturing
only those cases in which a group motive was determined. In contrast,
the latter definition errs on the side of capturing too many incidents,
including those motivated purely by individual member purposes—after
all, gang members can and do act of their own accord.
Fortunately, the data allow me to distinguish between these two defi-
nitions. In addition to the official motive classification given by the
investigating detective, each case also contains information on the dis-
putants (including gang membership), other potential motivating circum-
stances (such as whether or not an altercation preceded the homicide),
and a short description of the chronology of the event. I employ the
member-based approach because it casts a broader net and potentially
yields more false-positive results that can be detected through additional
coding; it also minimizes potential selection on the dependent variable.
Indeed, as will be seen in the analysis, murders involving persons who
are not gang members (referred to here as “non–gang members”) differ
significantly on issues of dominance and reciprocity. Although not reported
here, I ran all analyses according to both definitions. As it stands, results
from both definitions are similar with respect to the direction, magnitude,
and significance of parameter estimates, although the conservative defi-
nition yields somewhat stronger results (as one might expect). Therefore,
I utilize the member-based definition in order to capture a wider range
of individual and collective behaviors.
Each homicide is coded according to the gang affiliation of the offender
and victim, creating a dyadic data set of homicide incidents. Coding the
14
Although the use of official crime statistics as a source of information is a perpetually
debated topic (for a discussion, see Thornberry and Krohn [2003] and Kirk [2006]),
homicide records have proven to be a considerably more reliable data source than
other official crime indicators for two reasons. First, considerable resources and man-
power are directed at homicide investigations, and, second, homicides are more likely
to be reported than other types of crimes.
86
Murder by Structure
data in this way has two major benefits. First, it allows me to recreate
each homicide at the incident level, that is, who kills whom, where, and
in what year. Second, and perhaps more important, it allows me to ag-
gregate individual interactions to the gang level. So, in the case of gang
wars, I can trace the number of deaths between groups, the timing of
such deaths, and the length of such exchanges.
The second set of data is derived from geographic maps of gang turf
for all police beats in Chicago. These maps were created by gang intel-
ligence officers, not homicide detectives, and are maintained and managed
by a different division of the Chicago Police Department than that pro-
viding the homicide data. In total, there are 281 police beats in Chicago
with a median resident population of approximately 6,000 residents per
beat. Police beats are distinct geopolitical entities generally bounded by
major intersections, thoroughfares, and other ecological markers. As I
describe in the next section, I use these maps to ascertain (1) the overall
size of a gang as a function of the total space it occupies, (2) how many
uncontested pieces of turf a gang controls, and (3) the extent to which
the turfs of any two gangs in a homicide overlap—that is, geographic
points that might serve as contested ground for dominance disputes or
conflict more generally.
Decoding the meaning of gang homicide networks and the interactions
themselves is extremely difficult without a qualitative understanding of
Chicago’s gang world, its key actors, and the subjective meaning they
give to it. To clarify and extrapolate the network and regression findings,
I also draw from field notes and interview material derived from ethno-
graphic fieldwork with gangs and gang members in several Chicago neigh-
borhoods. The research was designed to gather data on (a) how gang
members understand gang violence, (b) what their subjective reasons were
for engaging in inter- and intragang violence, and (c) how, if at all, they
linked seemingly disparate violent acts.15
To these ends, I designed the fieldwork to bring me as close as possible
to gang violence or, at least, close to the actual “facts” of the violent events.
I began by locating two “street gang workers,” essentially social workers
and caseworkers whose job it is to provide services directly to gang mem-
bers and to mediate gang disputes. One worker is employed by a major
religious organization that operates mainly in three African-American
housing projects on the South and West sides of the city, and the second
street worker is a former Hispanic gang member who works for a small
nonprofit organization on the West Side. Combined, these two men have
15
A full description of my ethnographic methods is presented elsewhere (Papachristos
2007).
87
more than 30 years’ experience working with gang members in Chicago.16

From October 2002 until December 2003, I followed the two men as they
went about their daily work including walking or driving through neigh-
borhoods, chatting with residents, and helping gang members with a
variety of life issues such as getting a job, procuring health care for a
sick child, urging them to stay in school, and supporting them in various
legal issues.17 As I had hoped, both workers were deeply entrenched in
mediating gang violence. On a weekly basis, I observed as gang members
would tell the men about ongoing disputes, seek their counsel, and enlist
them to defuse potentially hostile situations. This usually meant relaying
messages back and forth between the warring parties. At the extreme, it
meant literally standing between gang members, pulling them off one
another, or urging them to put down their guns.
The ethnographic data used in this article are derived from observa-
tions, conversations, and informal interviews during the field period as
captured in ethnographic field notes.18 During the earlier stages of the
fieldwork, I merely observed as the workers went about their normal
routine, asking questions only sporadically. After the first few months, I
began engaging both the workers and their gang clientele more directly
about given disputes and gang violence more generally. As a rule, I never
wrote anything down in front of gang members, lest I rouse suspicion.
16
Unlike the more common one-neighborhood or one-gang approach, I intentionally
sought out gang and nongang contacts from multiple neighborhoods to understand
variation in violent acts under different neighborhood conditions. My intention was
to sacrifice some depth about particular gangs in order to capture variation between
gangs and across neighborhoods.
17
I followed each worker approximately twice a week, staggering the weeks with each
worker in order to provide some continuity in my observations of specific incidents,
such as inquiring about incidents that arose earlier in the week.
18
The usual caveats of urban ethnography are also warranted. Undoubtedly, my own
personality and physical features may have influenced my observations and conver-
sations. Even though the street workers vouched for me, just about every time I was
introduced to a new group of gang members, I was as seen as a “narc” (undercover
police agent), a drug customer, a social worker, and once even a “repo” man. After
some ridicule and a bit of persistence, the workers would assure people that I was
“cool,” which appeared to put people at ease. At the very least, gang members were
amenable to talking with me and tolerating my questions. However, my presence as
an outsider or “stranger” allowed me to ask questions that seemed glaringly obvious
to the insider. In a few instances, my questions were met with outright laughter on
the part of the respondent, and then a sense of pity followed as the respondent pro-
ceeded to answer my questions in great detail. My constant prodding even seemed
welcomed as some respondents even rummaged through personal artifacts, such as
photos or gang documents, to “teach me” about their gang or show me family members
who had been killed. Perhaps more important, my association with the street workers
allowed me to talk with gang members from multiple groups and to cross neighborhood
or turf boundaries that would otherwise have been off-limits.
88
Murder by Structure
Instead, I took jottings in the field (i.e., in the car, bathroom, etc.), which
were then expanded into full field notes as soon as possible after the day’s
observations. When I did ask people if I could interview them or talk
with them about a specific subject, I also asked if I they would mind if
I scribbled a few notes. When possible, I tried to capture the cadence of
speech patterns as well as the use of expletives, the latter of which is
essential in understanding the proximal motives of murder. To protect the
anonymity of those I observed, I replaced any identifying individual in-
formation with a pseudonym; I have not, however, altered any names of
the gangs.
The analysis proceeds in two stages. The first stage uses a network
approach to recreate the social structure of gang homicides building up
from individual homicides to a global network structure. The goal of this
stage is to determine the network pattern of gang homicide, describe its
basic properties, and assess its stability. The second stage analyzes possible
mechanisms responsible for the creation of these murder networks. More
specifically, I use dyadic models and network autocorrelation models to
examine the mechanisms responsible for the observed patterns of murder
and, in particular, patterns of reciprocity. All of the empirical material
comes from the data just described, but analyzed in two different ways.
Therefore, I save the description of statistical methodologies for each
respective section.
THE SOCIAL STRUCTURE OF GANG HOMICIDE

This section focuses on the network structure of gang murder and begins
with the construction of the murder network for a single year, 1994. I
then use standard network techniques to analyze the properties of the
observed network. To assess the stability and influence of this network—
that is, the extent to which it truly represents a stable social “structure”—
I recreate murder networks in two additional years, 1998 and 2002, and
run a statistical test to detect the relationship between these different
networks. These years are selected for two reasons. First, they differ in
their rate of homicidal interaction. The 1994 overall homicide rate is 33.4
per 100,000. In contrast, the overall 1998 rate is 24.3 and the 2002 rate
is 22.3, the lowest in more than a decade. Differences in rates should
therefore account for the extent to which the sheer density of interaction
dictates the shape of the observed network. On average, gang homicides
account for 35% of all homicides in any given year.
Second, the four-year time intervals further ensure that the level of
analysis is the gang, not the gang member. Although tenure of gang mem-
bership varies, gang membership tends to be a transitory experience
89
(Thornberry et al. 2003). Individuals generally mature out of gang mem-

bership but are also removed from membership for varying lengths of
time for school, incarceration, marriage, or, as seen here, death. The like-
lihood that a single gang member is involved in multiple gang homicides
eight years apart is not known but is most likely extremely small.
Descriptive Analysis
In 1994 there were 931 homicides in Chicago, of which 31% (n p 284)
involved at least one gang member as a victim or offender. Forty-eight
different gangs were involved in these homicides, leaving a potential net-
work size of 2,256 possible ties between two unique gangs. Figure 1
displays the directed graph (digraph) of these murders.
Each node represents a unique gang, with the shape representing the
predominant race of the gang: squares signify black gangs, triangles His-
panic gangs, circles white gangs, and diamonds Asian gangs. Nodes are
labeled with a number representing each unique gang, and a table listing
all gang names is given in the appendix below. Arcs indicate a single
homicide, with the direction indicating the victim. Bidirectional arcs in-
dicate reciprocal homicides between two groups. The thickness of the line
represents the intensity of the conflict measured as the number of hom-
icides between gangs. Loops indicate murders within the same gang. Iso-
lates represent gangs in which the only homicide that occurred involved
members of the same gang.
At the incident level, homicides involving gang members share several
commonalities. The vast majority of homicides (98%) were intraracial and
intergang (88%). Moreover, 93% involved a firearm, 89% occurred in
public (especially on the street or in a public walkway), 38% involved
more than one perpetrator, and 37% were reciprocal. The levels of reci-
procity are of particular importance since they far exceed the expected
number that would be found in networks with similar in-degree and out-
degree distributions. Indeed, to test the null hypothesis of independence
of the proportion of reciprocal murders in the network, I generated 5,000
simulated networks conditioned on the observed in- and out-degree of
the 1994 murder network. The mean number of expected reciprocal dyads
in the simulated networks is 15.31 with a standard deviation of 2.81,
whereas the observed number of dyads in the network is 36, thus giving
strong evidence that reciprocity occurs more than would be expected (t-
statistic p 7.24; P p .000). In short, reciprocity is one of the strongest
characteristics of the observed murder networks.
Most of these homicides began over symbolic threats. As seen in table
1, 98% of all gang-related homicides in 1994 in which the motive could
be determined began over expressive factors, as compared with 67% of
90
Murder by Structure
Fig. 1.—The social structure of gang homicide in Chicago, 1994. Squares represent black
gangs; triangles represent Hispanic gangs; diamonds represent Asian gangs; circles represent
white gangs; direction of arrow indicates direction of homicides; and thickness of line in-
dicates number of homicides between gangs.
non-gang-related homicides.19 That is, the “spark” of most of these mur-

ders, that is, its proximate motivating action, was typically some argument
over nonmaterial matters. Even more so than homicides in general, then,
gang homicides occur for symbolic reasons—an important caveat given
the prevalence of Chicago gangs in underground economies and “cor-
porate-style” drug dealing (e.g., Padilla 1992; Levitt and Venkatesh 2000).
19
Event narratives were used to code these cases rather than the official motive given
by the detective. Missing data are those cases that did not include such narratives or
in which the actual motive could not be determined. I coded a homicide as expressive
if it began as an argument or altercation. Instrumental homicides are those that began
as disputes over property, money, or narcotics or were committed during the course
of another crime (e.g., robbery). Turf considerations for a gang are considered instru-
mental if they relate to drug dealing or other economic interests and are expressive if
the turf violation involves use by a nonmember or noneconomic disputes (e.g., dis-
respectful graffiti on turf). The “other” category for nongang homicides refers mainly
to those that occurred during a sexual assault. Gould (2003) uses a similar coding
schema.
91
TABLE 1
Frequency of Gang and Nongang Homicides by Motive, 1994
Motive
Expressive Instrumental Other Total
Non-gang-related homicides . . . . . . 289 124 18 431
% non-gang-related homicides . . . 67.1 28.8 4.2 100
Gang-related homicides . . . . . . . . . . . 237 6 0 243
% gang-related homicides . . . . . . . . 97.5 2.5 0 100
Total . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 526 130 18 674
% total . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 19.3 2.7 100
Note.—x2 p 84.27; P p .000; df p 2.
Although these homicides are similar in terms of the context and motive,
figure 1 shows distinct racial cleavages, as well as differences in density,
power, and exposure to murder within the larger network.
Racial and Ethnic Differences in Network Structure

Figure 1 shows a densely connected network among black gangs with
several highly active groups and a sparsely connected, star-like network
of Hispanic gangs with a single gang at the center. White gangs are only
peripherally connected in the larger network and Asian gangs are com-
pletely isolated, and, for these reasons, they are largely excluded from the
remainder of the analysis.20 On average, each gang in the network is
involved in approximately three homicides. It is here, however, that racial
differences become most apparent. Black gangs are three times more
actively involved in murders than Hispanic gangs: black gangs, on av-
erage, are involved in 5.21 homicides whereas Hispanic gangs are involved
in 1.46.
Density is a basic network property that reflects the overall connect-
edness of actors: the more connected the social units, the greater the
network density. The density of the total network, measured as the pro-
portion of ties present of all possible ties, is 0.065: approximately 7% of
all ties that are possible are present.21 Disaggregated by race, the black
20
White gangs act exactly like the Hispanic gangs in terms of their network properties.
However, an assessment of Asian gangs is much more difficult because of their small
number and complete isolation from the larger network; in this respect, they seem to
be an island unto themselves.
21
Density, D, of a network with g actors is measured as the sum of all entries in the
matrix, divided by the possible number of entries:
ip1g jp1g xij
Dp .
g(g ⫺ 1)
92
Murder by Structure
gang network is more than five times as dense (D p 0.304) as the Hispanic
network (D p 0.058).22 Such differences in density create local (intraracial)
networks that affect the structure of gang relations, as well as the diffusion
of violence within these networks.
In epidemiological terms, the black murder network looks similar to a
“core infection” model in which highly active and interrelated groups
continually (re)infect each other and then infect others in the periphery
(see Laumann and Youm 1999). Although some gangs may be more active
than others, as discussed in the next section, each group is engaged in
murders with multiple gangs. Moreover, more than 35% of all relations
are reciprocal. This suggests that interaction flows quickly because actors
consistently interact and are sensitive to changes in the local network.
The highly public nature of gang homicide and the density of such net-
works further imply that others in the network would be aware of failure
to respond to local threats. The failure to respond is therefore easily
conveyed and interpreted in the action (or inaction) of highly visible al-
ters.23 The result is a local network structure that looks like small feuding
nation-states with persistent conflict between multiple groups.
Gang members themselves seem to recognize this structure, as an ex-
ample from the West Side of Chicago illustrates. In the summer of 2003,
a vicious feud broke out among gangs belonging to an alliance known as
the Almighty Vice Lord Nation (AVLN), a cluster of gangs that share a
common ancestry and formal pledges of mutual protection.24 The gang
war was triggered by the release from prison of a high-ranking gang leader,
Slick. A founding member of the AVLN, Slick claimed to be dismayed
by the levels of violence in his neighborhood and among the AVLN. Slick
called for a truce among the warring factions and even tried to establish
a “board of directors” to settle AVLN disputes nonviolently. However,
other Nation leaders viewed Slick’s actions as an attempt to consolidate
power and subsume authority under his own flag. An intra-Nation war
ensued that included several shootings and at least two homicides. Leaders
of the various factions fought to keep the independence of their small
group without being subservient to Slick’s faction. One gang leader com-
mented: “He [Slick] ain’t got no damn right to claim shit about my gang
or another other member of the Nation. . . . I don’t give a fuck if he’s
22
While it is beyond the scope of this article, the structure of these networks might
be correlated with macrotrends such as segregation; it is here that stratification might
influence the structure of the network, suggesting an important area for future research.
23
In network terms, this implies contagion by competition or equivalence in which an
actor uses the actions or perceptions of its alters to determine courses of action (Burt
1987).
24
Several descriptions have been written about the origins and evolution of the AVLN
(Keiser 1969; Dawley 1973; Knox 2001).
93
[an AVLN] or not. He can’t just up and walk out of the joint [prison]
and start calling shots. I sure as hell don’t answer to him, and my crew
[gang] don’t answer to him.” The war waned only after an assassination
attempt on Slick that left him paralyzed—the third attempt on his life,
the second by his own Nation.
In network terms, the intra-Nation dispute just described is a micro-
representation of the larger structure of black gang homicide. The warring
factions of the AVLN alliance create small clusters of conflict, essentially
trading violent episodes as groups struggled for dominance. Indeed, it is
precisely these types of feuds that are most likely responsible for the form
of the observed network structure, a matter I take up in the regression
analysis.
The homicide networks of black gangs are remarkably resilient to ex-
ternal shocks since the activity of any single gang contributes only mod-
estly to the overall structure. The black gang network consists of a single
large component in which all gangs are either directly or indirectly con-
nected. The removal of any single gang from this structure reduces the
activity around that particular node, but it does little to alter the overall
structure of the network. One would have to remove the 11 gangs in the
core, or 70% of all the nodes, in order to fundamentally alter the structure
of the network. The removal of these gangs would break the network
into six smaller components, essentially isolates.
The less dense Hispanic network differs markedly. The star graph found
in the Hispanic network is a highly centralized structure with a clear and
dominant center. The intraracial Hispanic network has four components
consisting of a dyad, a triad, an isolate, and a large component of 20
gangs. The first three components represent classic small group conflict.
In the case of the dyad, if one of the gangs fails to reciprocate, the network
falls apart. In the triad, we see a single gang acting as an aggressor against
two alters. In this case, a single gang assumes the dominant role in at-
tacking two alters who do not retaliate during the observation period, a
classic pecking order or dominance relationship (Chase 1980).
Much of the activity in the Hispanic network happens within the large
component centered on a single gang, the Latin Kings (node 21), at the
center of the star-like structure. In short, the Latin Kings hold the network
together and act as a significant point source of diffusion. While such a
structure is highly efficient for the diffusion of violence, it is also extremely
fragile. The removal of the Latin Kings would break this graph into eight
smaller subgraphs, essentially reducing it to five isolates not involved in
any homicides, two dyads, and a small spanning tree component of 10
gangs that is itself extremely fragile.
Qualitatively, Hispanic gang members readily acknowledge the size and
influence of the Latin Kings, but do so in such a way as to neutralize the
94
Murder by Structure
Latin Kings’ influence relative to their own group’s position. For example,
Lito, a member of the Spanish Cobras (node 27), explains his gang’s
relationship with the Latin Kings as follows: “Man, those broken-ass
Crowns [a disrespectful name for the Latin Kings] are everywhere . . .
like fucking roaches, you know. Sometimes, you go to squash ’em, you
know, but most of the time you just leave ’em the fuck alone. We got
other problems to deal with. But, if those bitches [Latin Kings] come at
us, you better believe we gonna have something to say about it.” Lito’s
remarks allude to the overall size of the Latin Kings (they “are every-
where”), while at the same time slandering the group (i.e., calling them
“roaches” and “bitches”). Though a bit defensively, Lito does say that his
gang would retaliate should the Latin Kings instigate violence.
Network Exposure and Power

These network structures expose individual gangs to varying levels of
homicide; but gangs differ in their size, organizational capacity, and power
to respond to such exposure. Network measures of degree centrality cap-
ture the dynamics of exposure to homicide and power to respond to such
threats. Separate measures of in-degree and out-degree are used to account
for exposure and overall activity.25 In-degree is essentially a measure of
network exposure: the number of homicides in which the gang is victim.
In epidemiological terms, it is the rate of infectivity, or simply the exposure
to the behavior of interest. Out-degree measures the opposite—the number
of murders committed by a gang—and is especially important because it
demonstrates the capacity of a gang to engage in murder. In essence,
gangs with higher out-degree possess more power relative to other gangs
in that they exert more effort in potential status-conferring acts of violence.
Figure 2 displays a scatter plot of in-degree (x-axis) versus out-degree
(y-axis) for the 48 gangs in the 1994 network; the actual in- and out-
25
The in-degree, dI(ni), of an actor ni is the number of other actors in the graph that
are connected to the specified actor, i.e., the number of connections terminating at the
actor. In-degree is thus measured as the column totals of sociomatrix X, or
dI(ni) p 冘g
jp1
xji p x⫹i.
The out-degree, dO(ni) , of an actor ni is the number of connections that emanate from
the specified actor and can thus be measured as the row totals of sociomatrix X, or
dO(ni) p 冘g
jp1
xij p xi⫹,
where g is the number of actors in the network and xij refers to any single interaction
between two actors i and j (Wasserman and Faust 1994).
95
Fig. 2.—Scatter plot of degree of centrality in 1994 homicide network. Points are jittered
to improve visibility.
degree measures for all gangs are given in appendix table A1. A linear
relationship exists between the in-degree and out-degree (Pearson R p
.596); when the two most central gangs, the Gangster Disciples (node 8)
and the Latin Kings (node 21), are omitted, this correlation increases
(Pearson R p .651). This suggests that exposure to violence is associated
with the homicidal activity of any gang: namely, a gang is more likely to
commit murder if it is exposed to higher levels of violence. Because not
all homicides are reciprocal, this suggests that exposure to violence can
also lead to outward contagion in the form of generalized violence toward
other gangs and even non–gang members.
The outliers in figure 2 are of particular interest since they signal strong
point sources of violence. A single black gang, the Gangster Disciples, is
the most active in the network overall. In part, this may be a function
of size and/or organizational capacity. The Gangster Disciples is Chicago’s
largest and most organized street gang. Presently reported in half of all
police beats in Chicago, the Gangster Disciples was founded in the late
1950s as a small neighborhood-based gang on the South Side. By the mid-
1990s, the gang had evolved into a highly organized and criminalized
street organization, complete with a corporate-style hierarchy that or-
chestrated a multistate drug distribution system.26 Interestingly, even
26
See Perkins (1987) or Knox (2001) for more details on the Gangster Disciples.
96
Murder by Structure
though the Gangster Disciples is the most central gang in the network,
its immediate alters are also large organized gangs, especially the Vice
Lords (node 9), the Black Disciples (node 7), and the Black Stones (node
15). Moreover, these gangs are engaged in high levels of reciprocal murder
during this time period.
The Latin Kings—the second most active gang in the entire network—
is by far the city’s largest Hispanic gang and is also highly organized,
criminalized, and politicized (Knox 2001). Unlike the Gangster Disciples,
whose opponents are other “corporate-style” gangs, the organizational
capacity and size of the Latin Kings far surpass those of the other Hispanic
gangs. Consequentially, Hispanic gangs most often interact with just a
single alter, and usually that alter is the Latin Kings. Thus, the Latin
Kings appears to be one of the few gangs in the Hispanic network with
the ability to coordinate multiple violent relationships to any large degree.
The only group not to follow the linear trend found in figure 2 is the
non–gang member category. Collectively, non–gang members experience
the highest exposure to homicide but are rarely involved in the killing of
gang members. In other words, non–gang members are a source of out-
ward contagion as victims of homicide but do not perpetuate violence to
any considerable degree. This most likely results because non–gang mem-
bers do not experience the group processes that sustain gang activities;
that is, the consequences of previous murders fail to incite further action
because there is no group per se or because it is simply not framed as a
threat that demands redress in the form of further violence. The lines in
figure 2 show simple ordinary least squares regression predictions of out-
degree regressed on in-degree for the 1994 murder network. The dashed
line shows the predicted slope when non–gang members are included
(b p 0.498; SE p 0.201; t p 2.48; P ! .01), and the solid line displays
the slope when only gang members are included (b p 0.820 ; SE p 0.078;
t p 10.33; P ! .001). While both slopes are positive and significant, in-
cluding non–gang members in the regression decreases the slope by more
than a third.
Thus far, a clear network structure emerges from homicides between
gangs, one with distinct racial cleavages that exposes individual gangs to
varying levels of violence. Yet, how stable is the observed network? Is
such a network the result of the temporal or cultural peculiarities? Or, as
hypothesized here, does such a structure indeed provide a stable basis for
future interaction?
97
Network Stability
To answer these questions, I recreate the structure of gang homicide in
two additional years, 1998 and 2002, displayed in figures 3 and 4,
respectively.
The visual similarity between networks of different years is striking.
Black gangs are again involved in dense, symmetric networks, whereas
the Hispanic gangs are in a star-like network with the Latin Kings in the
center. The only apparent change in the overall network structure occurs
in 2002, when the prominence of the Latin Kings seems to diminish. As
predicted by the cut-point analysis presented above, this breaks the His-
panic network into several small components of dyads and triads.
To test the stability and predictive power of the earlier gang conflict
networks on later gang homicide, I use the quadratic assignment pro-
cedure (QAP) on the 1994, 1998, and 2002 matrices for all 66 gangs that
98
Murder by Structure
were reported as taking part in any murder during these years (Krack-
hardt 1987). QAP procedures provide a basic test of the null hypothesis
of independence among the gang homicide networks. QAP correlates
two network structures using Monte Carlo simulations that randomly
rearrange the nodes while preserving the structure of the network in
order to determine the strength of the relationship between two matrices.
QAP then assesses the probability of a correlation being as high as the
observed correlation within a network of a similar structure. Parameter
values range from zero to one: the larger the value, the greater the
correlation, where statistical significance is the probability of achieving
as high a correlation as that observed. I use the results from N p
5,000 simulations as the basis for significance tests on all estimates.
Table 2 lists the QAP results, with the row acting as the predictor
variable. The QAP correlations provide enough evidence to reject the null
hypothesis of independence for the overall networks, the black networks,
99
TABLE 2
QAP Correlations for the 1994, 1998, and 2002 Gang-Related Homicide
Network
Hispanic
Overall Network Black Network Network
1998 2002 1998 2002 1998 2002
1994 . . . .526*** .435*** .500*** .422*** .637*** .120
1998 . . . .561*** .621*** .072
Note.—N p 5,000 Monte Carlo simulations.
*** P p .001.
and the early Hispanic networks. For the overall networks, there is a
moderate correlation between the structures from 1994 and 1998 (.526,
P ! .001) and 1998 and 2002 (.561; P ! .001). The correlation is somewhat
smaller, although still statistically significant, between 1994 and 2002
(.435; P ! .001). Disaggregated by race, the patterns of the black network
are similar to the overall network, with higher correlations between closer
years. The 1994 and 1998 homicide networks for Hispanic gangs are
highly correlated (.637; P p .001), but the significance is lost in 2002 when
the structure dissipates. That is, there is not enough evidence to reject
the null hypothesis of independence for 2002. The 2002 Hispanic network
provides the exception to the general rule: the prior network structure
does not appear to determine the homicide patterns.27 These findings
suggest, even with the change in structure in 2002, that overall homicide
patterns are significantly correlated and that the effect is somewhat stron-
ger in years closer in time.
Overall, the QAP test suggests that the observed patterns of murderous
interactions are rather stable over time, even though individual gang
members (most notably the victim) come and go. Clearly, then, murder
creates an enduring structure of relations, one that appears to be insti-
tutionalized in that it creates lasting patterns of contention among groups.
DOMINANCE RELATIONS AND THE CREATION OF GANG MURDER

NETWORKS
Whereas the previous section described the social structure of gang ho-
micide, this section examines the mechanisms that produce these murder
27
One possible reason for this change in network structure is the effect of law en-
forcement initiatives on the Latin Kings over the past several years. While similar
strategies have been used against black gangs, the shape of the Hispanic network
suggests that such intervention strategies would be more effective against a diffuse
structure rather than a dense one.
100
Murder by Structure
TABLE 3
Means, SDs, and Partial Correlations among Dyad-Level Variables
Mean SD 1 2 3 4
1. Murder tie, 1996–97 . . . . . . . . . . . . . .072 .259 . . . .534 .039 .220
2. Black-on-black murder . . . . . . . . . .064 .246 .013 .064
3. Absolute size difference . . . . . . . . . 11.16 20.22 .104
4. Competition for dominance . . . . .096 .132 . . .
Note.—N p 2,415. Partial correlations with gang-specific effects controlled.
networks. More specifically, I analyze how dominance disputes, absolute

differences in gang size, and prior network exposure predict the presence
of a murderous tie between dyads and whether or not a murder is
reciprocal.
Predicting a Murderous Tie—Dyadic Analysis

The analysis begins by examining the question of who kills whom or,
more precisely, what predicts the presence of a murder between any two
gangs in the network. In the first set of models, I predict the presence of
a murder between any two gangs during the 24-month period from Jan-
uary 1996 to December 1997. In total, 466 murders were committed by
70 groups (69 gangs and a non–gang member category). The dependent
variable is measured as a binary indicator (1 p yes, 0 p no) of whether
or not a murder occurred between two specific gangs among all possible
2,415 dyadic ties.
I examine the effects of three dyad-level variables on the presence of
a murder between gangs: (1) the amount of competition for dominance,
(2) the racial and ethnic composition of the involved gangs, and (3) the
absolute size differences between the two gangs. Table 3 presents partial
correlations between these variables controlling for gang-specific effects.
Competition for dominance.—To capture the theorized notion of dom-
inance, I use the gang turf maps to generate a measure I call competition
for dominance. Competition for dominance is measured as the Jaccard
coefficient of contested turf of two gangs involved in a given homicide.
That is,
a
,
(a ⫹ b ⫹ c)
where, as seen in table 4, a is the amount of turf that gang A and gang
B jointly occupy, b is turf occupied by gang B but not by gang A, and c
is turf occupied by gang A but not by gang B. This measure is derived
using a two-mode adjacency matrix of the gangs and the turf they occupy
101
TABLE 4
Binary Table of Association for
Gang Turf Relationship
between Any Two Gangs (Dyad
Level)
Gang A
Gang B Yes No
Yes . . . . . . . a b
No . . . . . . . c d
in 1996–97, the graph of which is displayed in figure 5. Essentially, figure

5 visualizes the relations between gangs (circles) and their turf (squares),
where each link between nodes represents a parcel of turf that a gang
occupies. These data allow me to quantify the total number of neigh-
borhoods in which a gang has a piece of turf and, more important, which
neighborhoods are occupied by more than one gang. I make the assump-
tion that gangs that occupy the same space are more likely to come into
direct conflict.28 Group A in figure 5 gives a straightforward example in
which two gangs are simultaneously reported in the same neighborhood
and, on the basis of my assumption, are therefore more likely to compete
for status and resources. Moreover, in group A this is the only neighbor-
hood in which these two gangs are reported and they are the only gangs
in this area. The Jaccard coefficient literally measures the proportion of
turf overlap between two gangs involved in a homicide. The values of
the coefficient range from zero to one, where one is exact overlap.
The argument here is that turf is not simply a piece of economic capital,
but a crucial dimension of symbolic capital and gang identity. Indeed, of
the 20 different turf disputes I observed during my fieldwork, only two
involved economic issues, in those instances, disputes about the rights to
sell drugs on a particular street corner. All other turf disputes involved
symbolic matters, such as disrespectful gang graffiti or boasts of social
status and reputation. The most common turf transgression I observed
was the unauthorized use of a piece of turf by a rival gang member, such
as playing basketball at a park controlled by a rival gang or eating at a
restaurant that one’s rivals claim as a hangout.
Beyond controlling a parcel of land and its uses, gang turf is at the
foundation of gang identity and group processes (Short and Strodtbeck
28
This technique is commonly used in network analysis to infer social relations based
on affiliation data; methodologically, one transforms a two-mode matrix into a one-
mode matrix on the basis of joint affiliation. For example, two individuals who belong
to the same organization, sit on the same board of directors, or attend the same school
would be considered to have a link between them.
102
Murder by Structure
Fig. 5.—Two-mode graph of gang turf in Chicago, 1996–97
1965; Horowitz 1983; Vigil 1988). In some contexts, the name of the neigh-
borhood is synonymous with that of the local gang. Many of the gang
members in this study often switch back and forth between the group
name and the neighborhood name. The K-Town Kings, for example, is
a faction of Latin Kings who reside in an area of Chicago in which the
street names all begins with the letter K (Kilmore, Kedvale, Kilpatrick,
and so on); thus, while they are formally Latin Kings, the gang members
informally call themselves the K-Town Kings and are recognized as such
by other members of the Latin Kings. Likewise, many gangs add street
103
names to their proper gang surnames—such as the 55th Street Disciples,

a branch of the Gangster Disciples who reside on or around 55th Street.
Even more than identification, defense of turf is often the raison d’être
for collective violence precisely because it links gang identity with social
status. Defending turf affords an opportunity to display group cohesion
and social status, namely, that the gang has the capacity to keep what it
values and fend off would-be transgressors. Therefore, a threat against
turf can be thought of as a threat against the lifeblood of the organization
and a prime opportunity for the group to act for its own sake. One gang
informant describes the loss of turf (or failure to protect its boundaries)
as the ultimate failure of a gang. Melo, a midranking leader of the Two-
Six gang, describes an ongoing war with the rival Latin Lovers that began
over a piece of disrespectful graffiti:
This is our ’hood, see? We got no choice but to protect it. If we back down,
we ain’t shit. Everyone will think we ain’t nothin’ but a bunch of punk
ass bitches. . . . How can we call ourselves 2-6, if we don’t got this corner?
We always had this spot. It’s ours, man, no matter what those fuckers
[Latin Lovers] come at us with. . . . Without that, what do we got? Nothing.
Might as well join the fucking Boy Scouts if you ain’t got a spot. If we
back down, we look weak, man. . . . Can’t let no slobs [Latin Lovers] try
and just take that spot away.
Melo’s colorful remarks point to two important aspects of gang turf. First,
that the piece of turf in question (literally, a street corner) is partially what
defines the Two-Six: it is a part of their history, collective memory, and
persona. Second, defense of the spot is connected not only to such identity
claims but also to positions of dominance. Notice that Melo does not raise
a single economic issue, only those of dominance: that the Two-Six will
look like “a bunch of punk ass bitches” if they fail to protect their spot.
In fact, such subservience to the Lovers would, in Melo’s eyes, make the
Two-Six the structural equivalent of a Boy Scout troop.
Given such turf considerations, I take the proportion of turf overlap
captured in the Jaccard measure as an indicator of competition for dom-
inance between two gangs in a homicide. The more alike in turf overlap—
the closer the coefficient is to 1—the more gangs have to lose when conflict
arises. In other words, more is at stake between groups with higher turf
overlap because of the greater potential loss of status should a gang fail
to defend its turf. Similarly to Gould, I hypothesize that gangs are more
likely to engage in homicide and the reciprocal exchange of homicides
because neither gang holds a clear, externally supported dominance po-
sition. The higher the coefficient, the more alike the gangs are in turf
overlap and, consequently, the more likely they are to kill each other and
reciprocate violence.
104
Murder by Structure
Table 3 shows that, at the dyad level, the dominance measure between
any two gangs is 0.096; that is, any two gangs in a dyad share approxi-
mately 10% of their turf. If one considers only those dyads that resulted
in a murder, then this percentage increases to roughly 25%. However, the
standard deviations of network dominance measures are greater than the
mean, most likely skewed because of highly active and large gangs such
as the Gangster Disciples and the Latin Kings. To account for these
skewed distributions and potential outliers, I log the measure and conduct
the analyses with and without outlying gangs.
Racial composition of the gangs.—I examine racial variation in mur-
derous ties using a binary indicator of whether the murder occurs between
two gangs of the same racial or ethnic composition, but, more specifically,
if the homicide occurred between two black gangs (1 p yes, 0 p no).29
Gangs are coded as black or Hispanic if more than 90% of the victims
or offenders were of that race. This variable is included as a control for
racially unique correlates of homicide found in other research, as well as
an indicator of the racial distinctions in the observed network structures.
Moreover, in all cases the race of the victim is identical to the racial
composition of the gang as classified by the police and prior research. As
seen in table 3, approximately 7% of all dyads occur between two black
gangs; however, approximately 57% of actual murders occur between two
black gangs, resulting in the strongest single partial correlation of the
covariates (.534), a finding consistent with other quantitative research on
gang homicide in Chicago (e.g., Block and Block 1993; Papachristos and
Kirk 2006).
Size of gang.—The size of the various gangs is controlled for as the
absolute size difference between the offending and victim gangs, where
size is as measured by the total number of neighborhoods in which the
gang was reported by the Chicago Police Department. My underlying
assumption is simple: the larger the gang, the more space it occupies. The
size of a gang offers both a competing explanation of murder and a
potentially important control variable. The absolute size difference be-
tween gangs ranges from zero (two gangs of the same size) to 115, with
an average size difference of 11.16. But, again, the considerable standard
deviation of 20.22 suggests that this average is heightened by the presence
of large gangs such as the Gangster Disciples and Latin Kings. Size does
not appear to be highly correlated with the other covariates.
As mentioned above, the analysis is presented in two stages. The first
set of findings presents the dyad-level regressions models that predict a
29
Given that more than 95% of these murders were intraracial, this variable produces
fundamentally the same results as a simple same vs. different race variable and, I
believe, retains the importance of racial and ethnic differences in the networks.
105
murderous tie between any two gangs. The second set employs network
autocorrelation models to predict reciprocity within the observed murder
networks.
Results.—Table 5 presents the results from a dyad-level logistic re-
gression of the three predictor variables on whether or not a murderous
tie exists between two gangs in the 1996–97 network. Model 1 presents
the baseline model considering only the racial composition of the gang
and the absolute size difference between gangs in a dyad. Theoretically,
model 1 tests the extent to which murder between any two gangs is driven
by size. Both variables in model 1 are positive and statistically significant,
suggesting that a murder is more likely among two black gangs (b p
3.88; P ! .001), as compared to two Hispanic gangs or two gangs of dif-
ferent races, and as absolute differences in size between gangs increase
(b p 0.016; P ! .001). As a general matter, then, gang murders are more
likely between two black gangs, a finding that holds in all remaining
models. With regard to size, model 1 suggests that murders are more likely
between gangs of different sizes. This finding runs counter to Gould’s
argument, which predicted that violence would be more likely among
gangs more alike in size; however, as will be seen in the remaining models,
the significance of this variable drops when controlling for the larger
Gangster Disciples and Latin Kings.
Model 2 adds the theoretical construct at the core of my argument,
competition for dominance, namely, whether gang murder is driven by
turf overlap and the perceived dominance associated with such overlap.
As hypothesized, the competition for dominance variable has a positive
and statistically significant effect on the presence of a murder between
two gangs (b p 0.753; P ! .001). Taking the log odds of the raw coefficient
(exp [0.753] p 2.12) suggests that for each additional percentage increase
in turf overlap, the odds of a murder between gangs increase by a factor
of two, with all other variables held constant. In short, murders are indeed
more likely when two gangs have more at stake. The black-on-black
variable (b p 4.12; P ! .001) retains its significance, whereas the signifi-
cance of the size variable drops in this and subsequent models. This
finding holds in all subsequent models in table 5.
The remaining models in table 5 test the robustness of this finding by
adding in fixed effects for outlying gangs and non–gang members. Models
3 and 4 add in dummy variables for non–gang members and the Gangster
Disciples and Latin Kings, respectively. The coefficient for non–gang
members is positive and significant in both models, suggesting that non–
gang members are more likely to be killed across all dyads than gang
members. This is in part due to the fact that non–gang members (as a
group) are more likely to interact across gang and racial lines. Put another
way, all different types of gangs kill non–gang members whereas dyadic
106
TABLE 5
Parameter Estimates of Dyad-Level Logistic Regression Model Predicting a Murder between Two Gangs in 1996 and 1997
Model
1 2 3 4 5
Black-on-black gang murder (1 p yes) . . . . . . 3.882*** 4.114*** 4.577*** 5.028*** 6.781***
(.210) (.301) (.380) (.411) (.640)
Absolute size difference . . . . . . . . . . . . . . . . . . . . . . . . .016*** ⫺.001 .001 .001 .015
(.003) (.004) (.004) (.008) (.022)
Log (competition for dominance) . . . . . . . . . . . . . .753*** .962*** 1.030*** .756***
(.070) (.096) (.100) (.080)
Non–gang member (1 p yes) . . . . . . . . . . . . . . . . . 5.307*** 5.879***
(.720) (.791)
Gangster Disciples (1 p yes) . . . . . . . . . . . . . . . . . . .983 .722
(.880) (1.35)
Latin Kings (1 p yes) . . . . . . . . . . . . . . . . . . . . . . . . . 2.187*** 1.711***
(.431) (.590)
a
67 gang dummy variablesa . . . . . . . . . . . . . . . . . . . .
Constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ⫺3.599*** .627** 1.143*** 1.132*** ⫺1.373
(.14) (.30) (.35) (.38) (1.25)
Log likelihood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ⫺410.1 ⫺293.9 ⫺258.6 ⫺241.8 ⫺208.9
x2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399.6 632.0 702.6 736.2 748.2
Note.—N p 2,415. Numbers in parentheses are SEs.
a
These variables garnered 11 positive and statistically significant coefficients and 57 nonsignificant coefficients (0 p non–gang member).
* P ! .10.
** P ! .05.
*** P ! .01.
confrontations between two specific gangs are more circumscribed. As a

result, murderous ties exist between the non–gang member category and
a greater number of specific gangs (of both races) than between any two
gangs. Adding in the dummy variables for the Latin Kings (b p 2.19;
P ! .001) and the Gangster Disciples (b p 0.983) in model 4 increases
overall model fit, although only the former is statistically significant. This
suggests that the odds of a murder in a dyad increase by a factor of nine
when a Latin King is involved (exp [2.18] p 8.85). All other variables
except absolute size differences are positive and statistically significant.
Finally, model 5 adds in fixed effects for each of 69 gangs through a
series of dummy variables and using the non–gang member as a reference
group in order to control for nonindependence among dyads as well as
gang-level effects (see, e.g., Mizruchi 1989). Each dummy variable is coded
as 1 if the gang is present in a given dyad, such that each dyad will have
values of 1 for those two gangs and zeros for all other variables. For the
sake of presentation, the coefficients of these variables are suppressed in
model 5, but 11 of the gang indicators and the Latin Kings variable are
positive and statistically significant. More important, the black-on-black
variable (b p 6.78; P ! .001) and the competition for dominance variable
(b p 0.756; P ! .001) retain their directionality and statistical significance,
suggesting that a murder is more likely between two black gangs and
when two gangs share a greater portion of turf overlap, even when con-
sidering individual gang effects. Model 5 also provides the best fit of the
models in table 5.
Predicting Reciprocity—Network Autocorrelation Models

The final set of models examines the reciprocal exchange of murders. By
reciprocal, I am referring to any murder between members of two distinct
gangs that is followed by a murder between the same gangs six months
after the first killing and that is initiated by a member of the first victim’s
gang. In short, reciprocal murders indicate an exchange between these
two groups during the observation period.30 While reciprocal murders
include the more specific case of retaliation, the data do not allow me to
directly link specific motives between incidents. That is, I cannot deduce
whether the actual reason for a homicide was a prior homicide. Reciprocity
is measured as a binary variable indicating if a murder is reciprocal (1 p
30
In other analyses, I also allowed for murders between two gangs to occur six months
prior to account for any left-hand censoring issues; the results are extremely similar.
I also conducted analyses using 8-, 10-, and 12-month intervals, which also produces
similar results.
108
Murder by Structure
reciprocal, 0 p not) for the 466 murders that occurred in the 1996–97
period.
In contrast to the dyad-level analysis, the analysis of reciprocity con-
siders only those incidents that actually occurred. As a result, I am able
to include additional incident-level covariates, thus allowing for more
variation in motive and context. In addition to the black-on-black, ab-
solute size, dominance measures, and controls for non–gang members and
the Gangster Disciples and Latin Kings, I include four additional binary
covariates: (1) whether or not a firearm is used (1 p yes), (2) whether or
not the murder occurs in public or on the street (1 p public/street), (3)
whether or not the murder is preceded by a verbal argument or altercation
(1 p yes), and (4) whether or not multiple perpetrators are involved (1 p
yes). Consistent with prior research (see Howell 1999; Maxson 1999), table
6 shows that nearly all homicides involving gang members are committed
with a firearm (96%) and occur in public (91%). Moreover, 79% of all
homicides are preceded by an argument or fight between the disputants,
and 33% involve more than one offender.31 Like Gould (2003), I take
murders preceded by a verbal argument or altercation to indicate a sym-
bolic threat and those murders involving a greater number of disputants
to indicate a higher level of collective action. This variable is also theo-
retically consistent with the argument described above in that it ensures
that the proximate interactions preceding a murder were not about in-
strumental or monetary considerations. Taken together, I hypothesize that
disputes that occur on the street, occur over symbolic reasons, and involve
collective action should be more likely to produce reciprocal homicides.
I model reciprocity using a network autocorrelation model (also known
as a network effects model) that accounts for the structural interdepen-
dence among network units by using an autocorrelation term (Leenders
2002). In the present analysis, I use a model commonly employed in the
study of the diffusion of ideas, innovations, and diseases (Valente 2005),
which takes the form
log
prob (yt p 1)
[1 ⫺ prob (yt p 1)]
pa⫹ 冘 bk Xk ⫹ b(k⫹1)Wy(t⫺1) ,
where y is a binary vector indicating whether or not a murder in 1996–

97 occurred, a is the intercept, b’s are the parameter estimates for vectors
k, and W represents a lagged social network of gang homicide, which I
describe below. I expand this model to a two-level population average
logistic regression model with the 466 murders at level 1 and the 70 gangs
31
As in table 1, the symbolic variable was coded 1 if the murder is motivated by a
verbal argument or altercation that was not about property, money, or drugs or was
committed during the commission of another crime.
109
TABLE 6
Means, SDs, and Partial Correlations among Incident-Level Variables
Mean SD 1 2 3 4 5 6
1. Reciprocal murder . . . . . . . . . .453 .498 . . . .029 ⫺.016 .001 .077 .147
2. Gun used . . . . . . . . . . . . . . . . . . . .961 .194 .198 ⫺.118 .151 ⫺.006
3. Occur in public . . . . . . . . . . . . .908 .228 .048 .228 ⫺.082
4. Black-on-black murder . . . .697 .459 .051 ⫺.021
5. Symbolic . . . . . . . . . . . . . . . . . . . . .791 .407 .282
6. Multiple offenders . . . . . . . . . .333 .472 . . .
Note.—N p 466. Partial correlations with gang-specific effects controlled.
at level 2, thereby treating murders as nested within gangs.32 Given the

interest in who is doing the killing, I assign gangs in level 2 on the basis
of the offender’s gang.
Network exposure and social contagion.—Using a network autocorre-
lation model also allows me to include a final mechanism of theoretical
interest—a contagion parameter that models the influence of prior ex-
posure in murder networks. The stability of these murder networks sug-
gests that prior murders influence the selection of victims in subsequent
years. As hypothesized above, this may reflect a process of social contagion
in which communication and competition drive the diffusion of violence
between gangs. Communication drives the contagion of murder as ego
and alter are socialized into a shared understanding of animosity or con-
flict; that is, we are enemies and we both know it. Under these conditions,
small insults blow up into events that require a severe response because
of the prior interactions between groups. Competition drives contagion
when killing is necessary to prevent the loss of face: that is, if we do not
kill, then we lose status relative to others in the social network. If murders
are truly contests or exchanges in which status is won, lost, or gained,
then murder would be an important “winning” strategy. These two mech-
anisms are not necessarily mutually exclusive, however. In fact, the two
might directly influence each other: one is socialized into contentious re-
lations (communication) because of prior interaction and competition.33
I measure the social contagion of violence as a lagged adjacency matrix
of gang murder in 1994 and 1995, the Wy(t⫺1) parameter in the network
32
Traditional one-level models produce the same results, but with exaggerated standard
errors caused by gang-level effects.
33
Communication and competition roughly parallel Burt’s (1987) ideas of contagion
of cohesion and structural equivalence, respectively. Theoretically, cohesion and equiv-
alence refer to two potentially different mechanisms. Empirically, however, the two
might refer to similar social influence processes, especially in small homogeneous net-
works in which the majority of contact and socialization (cohesion) occurs between
those occupying similar roles (equivalence).
110
Murder by Structure
effects model. The matrix is lagged to account for the influence of past
gang murders on present decisions to reciprocate.34 When, as I do here,
a network is measured on direct contact, it captures social influence con-
veyed through overt transmission or direct pressure (Valente 2005, p.
103).35 This implies that the contagion of gang murder is influenced by
those in the immediate conflict network in the two prior years; namely,
gangs are influenced by interaction within their local network, in partic-
ular, their immediate network neighbors, or those with whom they were
involved in murders in the past. A significant b parameter provides evi-
dence of contagion. I hypothesize that gangs are more likely to engage in
reciprocal homicides with those whom they have killed or been killed by
in the past.
Results.—Table 7 examines the mechanisms responsible for the recip-
rocal murders in the 1996–97 networks by regressing the dependent var-
iable (whether a homicide is reciprocal) on incident-and gang-level var-
iables. Model 1 tests the effect of incident and contextual factors on the
likelihood of a reciprocal murder; in other words, this model assesses the
extent to which reciprocal murders are explained by variation in incident-
level circumstances. Model 1 shows nonsignificant effects of whether or
not a gun was used (b p ⫺0.202), whether or not the murder occurred
in public (b p ⫺0.599), and whether or not multiple offenders were in-
volved (b p 0.358). These variables lack statistical significance probably
as a result of the lack of variation in these murders: a great portion of
these incidents involve a gun, involve multiple offenders, and occur on
the street. The significant predictors in model 1 are the black-on-black
variable (b p 1.075; P ! .001), the symbolic variable (b p 0.923; P !
.05), and the non–gang member variable (b p ⫺3.63; P ! .001). These
findings suggest that reciprocal homicides are more likely when both the
victim and the offender are members of a black gang and when the killing
is preceded by a verbal argument or dispute. Additionally, reciprocal
34
This approach is similar to lagged spatial autocorrelation terms used to capture the
influence of past violent crime rates at the neighborhood level (e.g., Morenoff, Sampson,
and Raudenbush 2001). I use a two-year time frame in both instances to capture a
greater number of incidents and gangs. Analyses conducted on single-year networks
or with greater lags produce similar results, especially given the stability of these
networks. Furthermore, any time parameter used in such models creates some right-
hand censoring; i.e., gang disputes do end with the calendar year.
35
Consistent with other network research, I also measured W using measures of struc-
tural equivalence (measured as Euclidean distance) between gangs in the network.
However, the structural equivalence and adjacency matrices were highly correlated
(R p .871), suggesting that, in this case, the two were measuring the same social
influence processes. In other words, gangs interact with their equivalent alters. This
is not surprising given the small, nonoverlapping nature of the networks. I am grateful
to an AJS reviewer for reminding me of this point.
111
TABLE 7
Parameter Estimates of a Two-Level Population Average Logistic Regression Model Predicting Reciprocal Murders
Model
112
1 2 3 4 5 6
Incident-level characteristics:
Gun used (1 p yes) . . . . . . . . . . . . . . . . . . . . . ⫺.202 ⫺.112 ⫺.042 .081 .074 .035
(.671) (.666) (.786) (.701) (.762) (.704)
Occur in public (1 p yes) . . . . . . . . . . . . . . ⫺.599 ⫺.320 ⫺.427 ⫺.273 ⫺.377 ⫺.328
(.511) (.523) (.591) (.576) (.596) (.564)
Black-on-black murder (1 p yes) . . . . . . 1.075*** .631⫹ ⫺.0515 .469 .0983 ⫺.0714
(.277) (.351) (.522) (.311) (.428) (.441)
Symbolic (1 p yes) . . . . . . . . . . . . . . . . . . . . . . .923** .709* .816* .695* .765* .727*
(.328) (.336) (.381) (.357) (.370) (.361)
Multiple offenders (1 p yes) . . . . . . . . . . . .358 .381 .334 .266 .240 .232
(.255) (.271) (.290) (.285) (.293) (.290)
Non–gang member as victim or offender
(1 p yes) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ⫺3.633*** ⫺3.681*** ⫺2.242*** ⫺3.717*** ⫺2.623** ⫺2.909***
(.5411) (.675) (.621) (.593) (.980) (.698)
Gang-level characteristics:
Absolute size difference . . . . . . . . . . . . . . . . . .0231*** .0228*** .0141*** .0125* .009
(.005) (.005) (.004) (.006) (.011)
Lagged network, 1994–95 . . . . . . . . . . . . . . .234** .135⫹ .103*
(.078) (.070) (.052)
Log (competition for dominance) . . . . . . 1.216*** 1.246*** 1.322***
(.210) (.250) (.251)
Gangster Disciples (1 p yes) . . . . . . . . . . . .0851
(.883)
Latin Kings (1 p yes) . . . . . . . . . . . . . . . . . . . ⫺.335
(.571)
Constant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ⫺.417 ⫺1.222 ⫺2.086* 3.242** 2.891* 3.611**
(.77) (.76) (.97) (1.05) (1.27) (1.16)
Wald x2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47.64 53.81 55.81 72.38 79.94 87.32
Note.—N p 466. Numbers in parentheses are SEs.
⫹
P ! .10.
113
* P ! .05.
** P ! .01.
*** P ! .001.
murders are considerably less likely when a non–gang member is involved.

The significance of the symbolic variables supports the theory put forth
here and that of Gould (2003): murders are more likely to be reciprocated
when they are preceded by noneconomic/material disputes. This symbolic
variable retains its statistical significance in all the remaining models,
even when controlling for gang-level effects.
Model 2 adds the gang-level absolute size variable, which, as in the
dyad analysis, is positive and statistically significant (b p 0.023; P !
.001). This suggests that reciprocal murders are more likely between gangs
of different sizes. This effect is retained in all subsequent models (3–6),
except when dummy variables are added in model 6 for the Gangster
Disciples and Latin Kings. Thus, as in the dyadic models and counter to
Gould’s argument regarding violence among status equals, size differences
appear to influence reciprocal murders; but the effect of size appears to
be driven mainly to these two large gangs, in large part because they are
involved in a considerably larger number of turf disputes.
The social contagion parameter is added in model 3 to assess the effect
of prior network structure on reciprocal murders. The parameter is pos-
itive and statistically significant (b p 0.234 ; P ! .01) and slightly improves
overall model fit (Wald x2 p 55.81). In the case of reciprocal murders,
therefore, there is evidence of contagion: a murder is more likely to be
reciprocated if someone in your gang is killed in the prior six months by
a gang in your local network neighborhood. Put another way, the mur-
derous interactions of a gang’s network neighbors are positively associated
with a murder being reciprocated.
Model 4 presents the effect of the competition for dominance variable
on reciprocal murders, controlling for size and incident-level character-
istics. Just as in the dyad models, the dominance variable is positive and
statistically significant (b p 1.22 ; P ! .001 ). Thus, when two gangs share
a greater proportion of contested turf relative to their respective size, they
are more likely to exchange murders, net of incident-level characteristics
and absolute size differences. The log odds of this parameter
(exp [1.22] p 3.87) imply that a one-percentage-point increase in turf over-
lap increases the odds of a murder being reciprocated by a factor of nearly
four. Quite simply, when more is at stake between two gangs—when they
have a greater proportion of turf overlap that is often linked with dom-
inance and status considerations—the reciprocal exchange of murders is
considerably more likely. Additionally, this variable increases model fit
over the previous models (Wald x2 p 72.38).
Model 5 tests the robustness of the contagion (b p 0.135; P ! .10) and
dominance variables (b p 1.25; P ! .001) by considering both variables
simultaneously, both of which are positive and significant. The magnitude
and significance of the dominance variable are retained, although the
114
Murder by Structure
statistical significance of the contagion parameter drops to the most lax

significance level. The continued significance of these variables, as well
as the symbolic variable (b p 0.765; P ! .05), supports the idea that most
of the variation in reciprocal murders is explained by these gang-level
factors.
Finally, model 6 contains all incident- and gang-level predictors and
adds two dummy variables for the Gangster Disciples and Latin Kings.
As in the dyadic analysis, controlling for these two outlier gangs provides
the best overall model fit and mediates the effects of size of the gang
(Wald x2 p 87.32). More important, though, the influence of the symbolic,
social contagion, and dominance variables holds even when controlling
for the juggernaut Gangster Disciples and Latin Kings. However, neither
the Gangster Disciples nor the Latin Kings variables are statistically sig-
nificant, implying that other gangs are just as likely to retaliate.
Models 5 and 6 have several important implications. First, there is
considerable evidence of network contagion within these murder net-
works: one is more likely to reciprocate a murder with a gang it has been
involved with in murders in the past and/or when its prior network neigh-
bors are also engaged in murders. Second, competition for dominance, as
expressed in turf overlap, is a strong and consistent predictor of reciprocity
in murder between gangs, even when considering variation in incident-
level characteristics and size differences. In other words, when two gangs
have more at stake—regardless of the specific motivating event of a mur-
der—that murder is more likely to be reciprocated. Finally, it appears
that such immediate dominance considerations at least partially mediate
the effect of prior network exposure. Thus, even though there is evidence
for contagion, competition for dominance appears to have a stronger in-
fluence on reciprocity when controlling for other incident- and gang-level
characteristics.
DISCUSSION
These results demonstrate that gang murders create an enduring social
structure that is produced through dominance disputes and the social
contagion of prior interactions. At the incident level, homicides between
gangs most often result from conflict over symbolic threats within group
contexts that stress the use of violence as social control. The parameter
estimates of the symbolic and competition for dominance variables sup-
port such claims and Gould’s (2003) description of group violence: mur-
ders are more likely to involve symbolic circumstances among groups
caught up in dominance disputes. Contrary to Gould’s argument that
violence is more likely among near equals, however, murders are more
115
likely among gangs of different sizes, especially when two extremely large
gangs such as the Gangster Disciples and Latin Kings are taken into
consideration.
Descriptive network analysis reveals distinct racial cleavages of these
murder networks. Black gangs interact within a highly active and densely
connected network, whereas Hispanic gangs exist in a diffuse star-like
network with a clearly identified center of power. Both networks are
conducive to contagion, but the black network is more resilient to external
shocks. Furthermore, a positive and direct correlation exists between net-
work exposure and activity, regardless of network structure: the more a
gang is on the receiving end of murder (in-degree), the higher its total
murderous output (out-degree). This relationship holds even when con-
trolling for large and powerful gangs such as the Gangster Disciples and
the Latin Kings but does not apply to non–gang members.
The two-level models, as well as the QAP test, indicate that these
murder networks are institutionalized in that they persist over time and
influence individual action and collective behavior. In other words, there
is clear evidence of a gang effect that exists above and beyond individual
motives and any isolated murder. In the rational choice parlance, such
networks represent a considerable external constraint on gang member
behavior. Even though individual gang members undoubtedly pursue
their own self-interests, their murderous activities generally coincide with
the larger network structure between groups. In this manner, it is the
structure that kills.
Multivariate analysis provides evidence that prior network structure
and competition for dominance disputes significantly predict the presence
of a murder between gangs as well as reciprocity. With respect to dom-
inance, it appears that murder—reciprocal or otherwise—is more likely
to occur between gangs with greater amounts of turf overlap. As seen in
some of the qualitative evidence, turf disputes are less about a parcel of
land than about a gang’s status and perceived dominance, and thus,
greater overlap will have greater consequences on the social standing of
the conflicting gangs. Truck, a member of the New Breed, describes the
importance of turf this way:
Truck: Shit, if a nigga’ steps to you, tries to take what is yours, what he’s
really doing is seeing what you’re made of. Got it? It’s like this: say, some
crew [gang] rolls up and tries to take your spot [street corner hangout].
They’s disrespecting you on your ’hood, you gots to step up [retaliate].
That’s, like, the worst fucking thing they could do.
Author: Does the spot really matter that much?
Truck: Man, ain’t you listening? It ain’t about no fucking corner. I mean,
yeah, it’s about corner, but, fuck the corner—it’s about not looking like
a punk [weak]. If you a punk, a corner ain’t going to help, see. Plenty
116
Murder by Structure
of bitches [gangs perceived as weak or of lesser status] got corners they

can’t keep. What good is a corner to them? You can only keep a corner
if you down [loyal to the gang] and everyone knows you and your mob
[gang] are down too. If people know you’re solid [strong/willing to fight],
and that you and your boys [gang friends] [are willing to] throw down
[fight], then you be all right. You’re straight. You got your corner and
your rep [reputation]. That’s just the way it is.
Truck’s remarks and those of the gang members throughout this article
highlight the importance of turf not just as an economic or political re-
source, but as an essential determinant of social standing. Turf disputes
are occasions in which gangs struggle for dominance. What is more, turf’s
symbolic value is contingent on the ability of a group to fight and avoid
subservience to other groups. The regression findings demonstrate that
when two gangs share greater degrees of contested turf, they are more
likely to engage in lethal violence and reciprocate when murder does occur.
The analysis also provides evidence that prior network structure influ-
ences the social contagion of violence. The positive coefficient of the lagged
network parameter suggests that the reciprocal exchange of murder is
significantly more likely when a gang has had prior murderous relations
with its alter and it is pulled into a conflict. In this sense, the “gift” of
murder must be returned if only to establish one’s position in the network.
A member of the Spanish Cobras, Hector, explained his gang’s decision
to retaliate against the rival Latin Counts as not being a choice at all:
They always comin’ at [attacking] us. Everybody knows it, and they
watchin’ to see what you do. The thing is, what can a nigga’ do? If you
don’t get yourself some payback, you ain’t shit. People’ll see you [are] weak
and your mob’s [gang] weak. Then, next thing you know, everyone be
stepping at [attacking or insulting] you. You got to hold on to your rep.
And, the only thing to do is to go back and fuck some fools [members of
opposing group] up. If someone steps to you, you best roll right-the-fuck-
back-up on them! Get you some revenge. If we back down from those
motherfuckers [Latin Counts], they think they got the best of us. They’d
think they got more heart than we do.
The key here is the importance gang members attribute to reciprocity.

The norm of reciprocity that Hector describes demands retribution once
an attack has occurred, especially when the threat comes from someone
a gang has a prior contentious relationship with. Failure to retaliate
against a long-standing opponent amounts to honorific suicide. In either
case, gangs clearly seem to consider the actions of their alters, supporting
the notion of the social contagion of gang murder resulting from processes
of competition and contact.
In contrast to research that views homicide as a signal of disorder in
117
the gang milieu (e.g., Decker and Curry 2002), I find a clear and stable
ordering in the conflictual relations of gangs. What may seem like the
haphazard exchange of insults and bullets is in fact the product of an
overarching network of contention. The stability and symbolic nature of
these murder networks suggest that conflict is institutionalized and most
often is a product of dominance disputes and the evaluation of dominance
relations. A gang’s pattern of murder thus appears to reflect its position
in the network structure. Because of the public and interpretative nature
of gang murder, information is often not directly transmitted but refracted
and framed before action occurs (e.g., Podolny 1993). While the entire
murder network may not be visible to gangs because of the timing by
which retaliation occurs, these structures are decidedly real as they are
real in their consequences: “Like social facts, they are invisible yet con-
sequential macrostructures that arise as the product of individual agency”
(Bearman, Moody, and Stovel 2004, p. 60). The competition for dominance
measure used here implies that gangs are able to interpret not only their
immediate opponent but also their evaluation of others in the network.
One might argue that these findings are limited by the uniqueness of
the data and their context, a point frequently cited in regard to research
done on Chicago gangs (see, e.g., Maxson and Klein 2001, p. 247). While
the direction, magnitude, and significance of the findings may certainly
vary given the context, I would argue that many of the key structures
and processes would not. Similar methods and theoretical approaches
would reveal that dominance disputes to a great extent determine the
specific patterning of gang violence. Moreover, while murder itself is a
rather unique event, its interaction patterns are very similar to interper-
sonal violence more generally. That is, murder may not be a unique type
of action, but rather an action of greater degree in the continuum of
violence.36 Ultimately, the validity of these findings rests in empirical rep-
lication and extension to other types of violent interactions.
CONCLUSION
Gang murder in Chicago is not merely an outcome resulting from the
convergence of individual or ecological variables. It is a consequential
action that shapes intergroup relations, the product of dominance disputes
that perpetuate murderous interactions over time. The groups are not a
set of pedagogical constructs, but are readily identifiable by name, ge-
ography, and history. The Gangster Disciples engage the Four Corner
36
Zimring (1968) demonstrates that there is little difference in the characteristics of
fatal and nonfatal violent episodes, which implies perhaps a significant overlap in the
distribution of offenders and motives.
118
Murder by Structure
Hustlers in battles over social standing, and the Latin Kings and Two-
Six go to war over verbal and physical insults. In short, there is an order
to gang murder evident in the social networks created through individual
transactions that manifest themselves in the relations among and between
actors.
Gangs continually confront dominance considerations in their quest for
social status. A group can be seen as cohesive and provide its members
status and protection only if the group can continually assert its dominance
by protecting its members and turf and thereby avoid subjugation. Like
chickens in a pecking contest, gangs want to avoid defeat and subjugation
not only in the immediate lethal exchange but also in future exchanges.
This process generates contentious networks of interaction that may very
well define the existence of the gang as a unique sociological entity.
Whereas some prior research questions the collective capacity of the gang
as a social group (e.g., Short and Strodtbeck 1965; Decker and Curry
2002; Fleisher 2005; Sullivan 2005), this study demonstrates that patterns
of gang conflict are extremely stable over time—a rather remarkable fact
given the transitory nature of gang membership. Gang members come
and go, but their patterns of behavior create a network structure that
persists and may very well provide the conduit through which gang val-
ues, norms, and culture are transmitted to future generations. In this sense,
it is murder by structure.
Considering the consequential nature of murder also tells us something
useful about other types of collective behavior. In particular, the approach
put forth here has implications for the study of social contagion of ho-
micide and its abatement. Most theories explain the diffusion of violence
as a function of proximity to violence or heightened exposure to risk
factors. Recent studies have paid particular attention to the spatial dy-
namics of homicide examining the nonrandom clustering of high-homicide
neighborhoods as well as the spatial diffusion of violence between geo-
graphic units (Cohen and Tita 1999; Rosenfeld, Bray, and Egley 1999;
Morenoff et al. 2001). The central idea of such research is that neigh-
borhood social institutions and actors are interdependent; that is, the social
networks of individuals and groups (whether prosocial or deviant) reach
beyond neighborhood boundaries and therefore facilitate the spatial con-
tagion of homicide. In particular, certain deviant and criminal behaviors—
such as drug dealing, gambling rings, and gang behaviors—are especially
contagious.
Unfortunately, most aggregate studies of homicide fail to measure such
social networks and, consequently, remain largely agnostic regarding the
mechanisms responsible for the contagion of homicide. When they are
measured, networks are usually captured through indirect survey mea-
sures rather than formal network methodologies. The typical approach
119
entails using spatial proximity of neighborhoods and their associated levels

of homicide (measured as aggregate rates) as indicators of such spatial
and social dependency of institutions and actors. Yet, we know that mere
exposure or proximity cannot fully explain the diffusion of homicide,
because murder rates themselves do not “move” across social or geographic
boundaries. Murder is not an airborne pathogen: one does not usually
become a victim or murderer simply by living in or next to a high-risk
area.37 The present study implies that both victims and offenders need to
be incorporated in (or at least proximate to) the social networks conducive
to sustaining such activities. Understanding how violent interactions gen-
erate meso- and macrostructures that act as the direct conduit of violence
is an essential first step in this direction. And, social network analysis
provides a promising avenue of investigation.
The theoretical framework and analytic approach advanced in the pre-
sent study also offer insight into the abatement of the social contagion of
violence. My results demonstrate that the “gang problem” in Chicago is
a structural problem with unique racial differences. What types of inter-
ventions might alter such structures? And how might social network anal-
ysis be used to identify the Achilles’ heel of such violent networks? The
evidence suggests that viable interventions would occur at the gang level.
That is, rather than simply target gang members writ large, interventions
should seek to disrupt links of conflict between gangs or else focus on
diminishing the organizational and violent capacity of high-impact
groups. The most obvious intervention entails data-guided enforcement
and legal efforts that target the organizational structure of specific gangs
(such as the Gangster Disciples and Latin Kings) rather than broad sweep-
ing policies that generically target gang members. In fact, some of the
disruptions in the observed networks may in fact have resulted from such
enforcement interventions. I urge data-driven strategies precisely because
of the distinct network differences observed here: attacking only a single
gang would not be as useful within the black gang network, whereas it
might be a devastating intervention within the Hispanic gang network.
Furthermore, social interventions that include dispute resolution and the
direct linking of key network actors with prosocial ties (such as employ-
ment services) might steer would-be violent gang members away from
more lethal forms of interaction. Social network techniques, such as those
used here, can map the topography of violent networks and, thus, provide
a guide to points of intervention.38
37
The obvious exception being the victim of a random act of violence, e.g., someone
caught in the crossfire of two rivals or hit by a stray bullet.
38
Some scholars are already examining the utility of a network approach in the abate-
ment of gang violence in Boston (Kennedy, Braga, and Piehl 1997), Los Angeles (Tita
et al. 2003), and New Jersey (McGloin 2005). For a review, see Papachristos (2006).
120
Murder by Structure
While the evidence presented here applies most directly to the case of
gang violence, it might also apply within the context of other sustained
violent and criminal interactions, such as drug markets, smuggling rings,
and terrorist activities, as well as other types of nonviolent interactions
more generally, especially competition, conflict, and reciprocal exchange.
For example, thinking of homicide as an interaction has implications for
network theory and analysis, particularly structural balance theory. Struc-
tural balance, and the related idea of transitivity, is one of the most basic
principles in social network analysis and implies that like-minded actors
also tend to share similar evaluations of alters, situations, events, and
ideas (Heider 1946; Holland and Leinhardt 1971). Within the context of
intergroup conflict, structural balance is commonly embodied in axioms
such as “the enemy of my enemy is my friend” (Mazur 1973b; Chase 1980).
Gangs in Chicago and elsewhere have been observed to create alliance
systems to further reinforce such pacts of mutual protection (Jacobs 1977;
Knox 2001). Thus, network structures similar to the ones presented here
provide a basic means to test the applicability of structural balance to
intergroup violence in a modern urban context. Do allies and friends
actually come to each other’s mutual aid? Does violence produce a pecking
order or confer status and rank? Moreover, homicides, as well as other
negative interactions such as hostile takeovers or warfare, may require a
reformulation and interpretation of standard network measures of cen-
trality, power, and cohesive subgroupings. And, the gang provides one
inroad into understanding such processes.
APPENDIX
A note on gang names.—Several of the gangs in this table have extremely
similar names, for example, the Conservative Vice Lords and the Traveling
Vice Lords, the Black Disciples and the Black Gangster Disciples, and
so on. Although at face value such names might suggest that these are
the same gangs, these groups are in fact qualitatively different social
entities. In the two examples just given, the gangs share a common an-
cestry but currently represent different factions that operate indepen-
dently. Such distinctions are readily recognized by both gang members
and law enforcement officials, and for these reasons, I treat them quan-
titatively and qualitatively as unique gangs.
121
TABLE A1
List of Node Labels, Gang Names, and Degree Centrality Measures for 1994,
1998, and 2002 Murder Networks
1994 1998 2002

Node In- Out- In- Out- In- Out-
Number Gang Name Degree Degree Degree Degree Degree Degree
1 ........ Ambrose 1 2 3 2 0 0
2 ........ La Raza 2 2 1 2 1 0
3 ........ Two-Sixers 2 1 0 0 0 0
4 ........ Bishops 0 2 0 0 1 0
5 ........ Latin Disciples 1 0 1 0 0 1
6 ........ Allport Lovers 1 0 0 0 0 0
7 ........ Black Disciples 3 3 3 7 1 2
8 ........ Gangster Disciples 12 9 10 11 8 6
9 ........ Vice Lords 6 6 4 6 4 5
10 . . . . . . . Black Gangster Disciples 2 2 0 3 2 2
11 . . . . . . . Black P. Stones 1 3 3 3 7 4
12 . . . . . . . Mickey Cobras 3 3 3 1 1 2
13 . . . . . . . Four Corner Hustlers 7 4 3 3 7 5
14 . . . . . . . Black Souls 2 1 1 2 0 1
15 . . . . . . . Black Stones 5 3 5 2 2 2
16 . . . . . . . Campbell Boys 0 1 0 0 1 0
17 . . . . . . . Maniac Latin Disciples 3 3 1 3 2 1
18 . . . . . . . Conservative Vice Lords 0 1 3 2 0 3
19 . . . . . . . Mafia Insane Vice Lords 2 3 2 2 4 0
20 . . . . . . . Cullerton Deuces 0 1 0 0 1 1
21 . . . . . . . Latin Kings 7 9 8 5 4 3
22 . . . . . . . Traveling Vice Lords 1 2 5 2 3 3
23 . . . . . . . Ebony Vice Lords 1 0 0 0 0 0
24 . . . . . . . Gaylords 0 1 0 0 0 0
25 . . . . . . . Imperial Gangsters 1 1 1 0 0 0
26 . . . . . . . Insane Unknowns 0 1 0 0 0 0
27 . . . . . . . Spanish Cobras 2 2 3 0 0 3
28 . . . . . . . Latin Counts 2 1 1 0 1 0
29 . . . . . . . Satan’s Disciples 2 2 2 2 5 2
30 . . . . . . . Latin Dragons 0 2 2 2 0 1
31 . . . . . . . Latin Eagles 1 1 0 1 1 1
32 . . . . . . . Latin Jivers 0 2 0 0 1 1
33 . . . . . . . C-Notes 1 0 0 0 0 0
34 . . . . . . . Puerto Rican Stones 1 0 0 0 0 1
35 . . . . . . . Spanish Gangster Disciples 1 0 0 0 1 1
36 . . . . . . . Brazers 1 0 0 0 0 0
37 . . . . . . . Two-Two Boys 1 0 0 0 0 1
38 . . . . . . . Latin Brothers Org. 1 0 0 0 0 0
39 . . . . . . . Insane Dragons 1 0 1 0 0 0
40 . . . . . . . Latin Lovers 1 0 0 0 0 0
41 . . . . . . . Pachucos 0 1 0 0 0 0
42 . . . . . . . Party People 0 1 0 1 0 3
Murder by Structure
TABLE A1 (Continued)
1994 1998 2002

Node In- Out- In- Out- In- Out-
Number Gang Name Degree Degree Degree Degree Degree Degree
43 ....... Red Scorpions 0 1 0 0 0 0
44 ....... Flying Dragons 1 0 0 0 0 0
45 ....... Simon City Royals 0 2 1 1 0 0
46 ....... Spanish Lords 1 1 0 0 0 0
47 ....... Unknown Vice Lords 1 2 0 0 0 0
48 ....... Titanic P. Stones 1 0 0 1 0 1
49 ....... La Familia 0 0 0 1 1 2
50 ....... Latin Saints 0 0 2 5 0 0
51 ....... Two-Six Boys 0 0 4 2 0 1
52 ....... Ashland Vikings 0 0 1 2 0 0
53 ....... New Breed 0 0 4 2 1 3
54 ....... Cicero Insane Vice Lords 0 0 1 1 1 1
55 ....... Insane Popes 0 0 0 1 0 0
56 ....... King Cobras 0 0 0 1 0 0
57 ....... Spanish Vice Lords 0 0 0 1 0 0
58 ....... Milwaukee Kings 0 0 1 0 0 1
59 ....... Latin Souls 0 0 1 0 0 0
60 ....... Undertaker Vice Lords 0 0 0 1 0 0
61 ....... Renegade Vice Lords 0 0 1 0 0 0
62 ....... Insane Deuces 0 0 0 1 0 0
63 ....... Insane Spanish Cobras 0 0 0 0 1 1
64 ....... Imperial Insane Gangsters 0 0 0 0 1 0
65 ....... Morgan Boys 0 0 0 0 1 0
66 ....... Orchestra Albany 0 0 0 0 1 0
REFERENCES
Anderson, Elijah. 1999. Code of the Streets. New York: Norton.
Baxter, Vern, and A. V. Margavio. 2000. “Honor, Status, and Aggression in Economic
Exchange.” Sociological Theory 18:399–416.
Bearman, Peter S., James Moody, and Katherine Stovel. 2004. “Chains of Affection:
The Structure of Adolescent Social Networks.” American Journal of Sociology 110:
44–91.
Black, Donald. 1983. “Crime as Social Control.” American Sociological Review 48:
34–45.
Block, Richard, and Carolyn R. Block. 1993. Street Gang Crime in Chicago. Research
in Brief report no. NCJ 144782. Washington, D.C.: U.S. Department of Justice, Office
of Justice Programs, National Institute of Justice.
Blok, Anton. 1974. The Mafia of a Sicilian Village, 1860–1960: A Study of Violent
Peasant Entrepreneurs. New York: Harper & Row.
Bourdieu, Pierre. 1966. “The Sentiment of Honor in Kabyle Society.” Pp. 191–241 in
Honor and Shame: The Values of Mediterranean Society, edited by John Peristiany.
Chicago: University of Chicago Press.
Bourgois, Philippe. 1995. In Search of Respect: Selling Crack in El Barrio. Cambridge:
Cambridge University Press.
123
Bursik, Robert, and Harold G. Grasmick. 1993. Neighborhoods and Crime: The Di-
mensions of Effective Community Control. New York: Lexington.
Burt, Ronald S. 1987. “Social Contagion and Innovation: Cohesion versus Structural
Equivalence.” American Journal of Sociology 92:1287–1335.
Chase, Ivan D. 1980. “Social Process and Hierarchy Formation in Small Groups: A
Comparative Perspective.” American Sociological Review 45:905–24.
Cohen, Jacqueline, and George Tita. 1999. “Diffusion in Homicide: Exploring a General
Method for Detecting Spatial Diffusion Processes.” Journal of Quantitative Crimi-
nology 15:451–93.
Dawley, David. 1973. A Nation of Lords: The Autobiography of the Vice Lords. Garden
City, N.Y.: Anchor.
Decker, Scott H. 1996. “Collective and Normative Features of Gang Violence.” Justice
Quarterly 13:243–64.
Decker, Scott H., and G. David Curry. 2000. “Addressing Key Features of Gang Mem-
bership.” Journal of Criminal Justice 28:473–83.
———. 2002. “Gangs, Gang Homicides, and Gang Loyalty: Organized Crime or Dis-
organized Criminals.” Journal of Criminal Justice 30:343–52.
Decker, Scott, and B. Van Winkle. 1996. Life in the Gang: Family, Friends, and
Violence. New York: Cambridge University Press.
de la Roche, Roberta Senechal. 1996. “Collective Violence as Social Control.” Socio-
logical Forum 11:97–128.
Esbensen, Finn-Aage, and David Huizinga. 1993. “Gangs, Drugs, and Delinquency in
a Survey of Urban Youth.” Criminology 31:565–89.
Fagan, Jeffrey, and Deanna L. Wilkinson. 1998. “Guns, Youth Violence, and Social
Identity in Inner Cities.” Pp. 105–88 in Youth Violence, edited by Michael Tonry
and Mark H. Moore. Vol. 24 of Crime and Justice: A Review of Research, edited
by Michael Tonry. Chicago: University of Chicago Press.
Fagan, Jeffrey, Deanna L. Wilkinson, and Garth Davies. 2000. “Social Contagion of
Violence.” Pp. 688–723 in Urban Seminar Series on Children’s Health and Safety.
Cambridge, Mass.: Harvard University, John F. Kennedy School of Government.
Federal Bureau of Investigation. Selected years. “Crime in America: Uniform Crime
Reports.” Washington, D.C.: Department of Justice.
Felson, Richard B. 1981. “An Interactionist Approach to Aggression.” Pp. 181–99 in
Impression Management: Theory and Social Psychological Research, edited by James
T. Tedeschi. New York: Academic Press.
———. 1982. “Impression Management and the Escalation of Aggression and Vio-
lence.” Social Psychology Quarterly 45:245–54.
Fleisher, Mark S. 2005. “Fieldwork Research and Social Network Analysis: Different
Methods Creating Complementary Perspectives.” Journal of Contemporary Criminal
Justice 21:120–34.
Gambetta, Diego. 1993. The Sicilian Mafia. Cambridge, Mass.: Harvard University
Press.
Ginat, Joseph. 1987. Blood Disputes among Bedouin and Rural Arabs in Israel: Re-
venge, Mediation, Outcasts, and Family Honor. Pittsburgh: University of Pittsburgh
Press.
Gluckman, Max. 1955. Customs and Conflict in Africa. Glencoe, Ill.: Free Press.
Goffman, Erving. 1967. Interaction Ritual: Essays on Face-to-Face Behavior. Garden
City, N.Y.: Anchor.
Gould, Roger V. 1999. “Collective Violence and Group Solidarity: Evidence from a
Feuding Society.” American Sociological Review 64:356–80.
———. 2000. “Revenge as Sanction and Solidarity Display: An Analysis of Vendettas
in Nineteenth-Century Corsica.” American Sociological Review 65:682–704.
———. 2003. Collision of Wills: How Ambiguity about Social Rank Breeds Conflict.
Chicago: University Chicago Press.
124
Murder by Structure
Greenberg, James B. 1989. Blood Ties: Life and Violence in Rural Mexico. Tucson:
University of Arizona Press.
Greenberg, Kenneth. 1990. “The Nose, the Lie, and the Duel in Antebellum South.”
American Historical Review 95:57–74.
Hagedorn, John. 1988. People and Folks: Gangs, Crime, and the Underclass in a
Rustbelt City. Chicago: Lake View.
Heider, F. 1946. “Attitudes and Cognitive Organization.” Journal of Psychology 21:
107–12.
Holland, P. W., and S. Leinhardt. 1971. “Transitivity in Structural Models of Small
Groups.” Comparative Group Studies 2:107–24.
Horowitz, Ruth. 1983. Honor and the American Dream. New Brunswick, N.J.: Rutgers
University Press.
Horowitz, Ruth, and Gary Schwartz. 1974. “Honor, Normative Ambiguity and Gang
Violence.” American Sociological Review 39:238–51.
Howell, James C. 1999. “Youth Gang Homicides: A Literature Review.” Crime and
Delinquency 45:208–41.
Hughes, Everett C. (1948) 1991. “The Study of Ethnic Relations.” Pp. 91–96 in Everett
C. Hughes: On Work, Race, and the Sociological Imagination, edited by Lewis Coser.
Chicago: University of Chicago Press.
Hughes, Lorine A., and James F. Short. 2005. “Disputes Involving Youth Street Gang
Members: Micro-social Contexts.” Criminology 43:43–76.
Ikegami, Eiko. 1995. The Taming of the Samurai: Honorific Individualism and the
Making of Modern Japan. Cambridge, Mass.: Harvard University Press.
Jacobs, Bruce A. 1999. Dealing Crack: The Social World of Streetcorner Selling. Boston:
Northeastern University Press.
———. 2004. “A Typology of Street Criminal Retaliation.” Journal of Research in
Crime and Delinquency 41:295–323.
Jacobs, Bruce A., and Richard Wright. 1999. “Stick-up, Street Culture, and Offender
Motivation.” Criminology 37:149–73.
———. 2006. Street Justice: Retaliation in the Criminal Underworld. New York:
Jacobs, James B. 1977. Stateville: The Penitentiary in Mass Society. Chicago: Uni-
versity of Chicago Press.
Jankowski, Martin Sanchez. 1991. Islands in the Street: Gangs and American Urban
Society. Berkeley and Los Angeles: University of California Press.
Jones, George Fenwick. 1959. Honor in German Literature. Chapel Hill: University
of North Carolina Press.
Katz, Jack. 1988. Seductions of Crime : Moral and Sensual Attractions in Doing Evil.
New York: Basic Books.
Keiser, R. Lincoln. 1969. The Vice Lords: Warriors of the Streets. New York: Holt,
Rinehart, & Winston.
Kennedy, D. M., Anthony A. Braga, and Anne M. Piehl. 1997. “The (Un)Known
Universe: Mapping Gangs and Gang Violence in Boston.” Pp. 219–37 in Crime
Mapping and Crime Prevention, edited by David Weisburd and Tom McEwen.
Monsey, N.Y.: Criminal Justice Press.
Kirk, David S. 2006. “Examining the Divergence across Self-Report and Official Data
Sources on Inferences about the Adolescent Life Course of Crime.” Journal of Quan-
titative Criminology 22:107–29.
Klein, Malcolm W. 1995. The American Street Gang. New York: Oxford University
Press.
Klein, Malcolm W., and Lois Y. Crawford. 1967. “Groups, Gangs, and Cohesiveness.”
Journal of Research in Crime and Delinquency 4:63–75.
Klein, Malcolm W., and Cheryl L. Maxson. 2006. Street Gang Patterns and Policies.
New York: Oxford University Press.
125
Knox, George W. 2001. An Introduction to Gangs, 5th ed. Peotone, Ill.: New Chicago
School Press.
Kobrin, Solomon, Joseph Puntil, and Emil Peluso. 1967. “Criteria of Status among
Street Groups.” Journal of Research in Crime and Delinquency 4:98–118.
Krackhardt, David. 1987. “QAP Partialling as a Test of Spuriousness.” Social Networks
9:171–86.
Kubrin, Charis E., and Ronald Weitzer. 2003. “Retaliatory Homicide: Concentrated
Disadvantage and Neighborhood Culture.” Social Problems 50:157–80.
Laumann, Edward O., and Yoosik Youm. 1999. “Racial/Ethnic Group Differences in
the Prevalence of Sexually Transmitted Diseases in the United States: A Network
Explanation.” Sexually Transmitted Diseases 26:250–61.
Leenders, Roger Th. A. J. 2002. “Modeling Social Influence through Network Auto-
correlation: Constructing the Weight Matrix.” Social Networks 24:21–47.
Levitt, Steven, and Sudhir Alladi Venkatesh. 2000. “An Economic Analysis of a Drug-
Selling Gang’s Finances.” Quarterly Journal of Economics 115:755–89.
Loftin, C. 1986. “Assaultive Violence as Contagious Social Process.” Bulletin of New
York Academy of Medicine 62:550–55.
Luckenbill, David F. 1977. “Criminal Homicide as a Situated Transaction.” Social
Problems 25:176–86.
MacAleer, Kevin. 1994. Dueling. Princeton, N.J.: Princeton University Press.
Maxson, Cheryl. 1999. “Gang Homicide: A Review and Extension of the Literature.”
Pp. 239–55 in Homicide: A Sourcebook of Social Research, edited by M. Dwayne
Smith and Margaret A. Zahn. Thousand Oaks, Calif.: Sage.
Maxson, Cheryl, and Malcolm W. Klein. 1996. “Defining Gang Homicide: An Updated
Look at Member and Motive Based Approaches.” Pp. 3–19 in Gangs in America,
2d ed. Edited by C. Ronald Huff. Thousand Oaks, Calif.: Sage.
———. 2001. “‘Play Groups’ No Longer: Urban Street Gangs in the Los Angeles
Region.” Pp. 235–66 in From Chicago to L.A.: Making Sense of Urban Theory, edited
by Michael J. Dear. Thousand Oaks, Calif.: Sage.
Mazur, A. 1973a. “A Cross-Species Comparison of Status in Small Established Groups.”
American Sociological Review 28:513–30.
———. 1973b. “Increased Tendency towards Balance during Stressful Conflict.” So-
ciometry 2:279–83.
McGloin, Jean Marie. 2005. “Policy Intervention Considerations of a Network Analysis
of Street Gangs.” Criminology and Public Policy 4:607–36.
Miller, W. I. 1993. Humiliation and Other Essays on Honor, Social Discomfort, and
Violence. Ithaca, N.Y.: Cornell University Press.
Miller, Walter B. 1958. “Lower Class Culture as a Generating Milieu of Gang Delin-
quency.” Journal of Social Issues 14:5–19.
Mizruchi, Mark S. 1989. “Similarity of Political Behavior among Large American
Corporations.” American Journal of Sociology 95:401–24.
Morenoff, Jeffrey D., Robert J. Sampson, and Stephen W. Raudenbush. 2001. “Neigh-
borhood Inequality, Collective Efficacy, and the Spatial Dynamics of Urban Vio-
lence.” Criminology 39:517–59.
Nisbett, R. E., and D. Cohen. 1996. Culture of Honor: The Psychology of Violence in
the South. Boulder, Colo.: Westview.
Otterbein, Keith. 1994. Feuding and Warfare: Selected Works of Keith Otterbein. Lon-
don: Routledge.
Padilla, Felix M. 1992. The Gang as an American Enterprise. New Brunswick, N.J.:
Rutgers University Press.
Papachristos, Andrew V. 2006. “Social Network Analysis and Gang Research: Theory
and Methods.” Pp. 99–116 in Studying Youth Gangs, edited by James F. Short and
Lorine A. Hughes. Lanham, Md.: AltaMira Press.
126
Murder by Structure
———. 2007. “Murder by Structure: A Network Theory of Gang Homicide.” Ph.D.

dissertation. University of Chicago, Department of Sociology.
Papachristos, Andrew V., and David Kirk. 2006. “Neighborhood Effects on Street Gang
Behavior.” Pp. 63–84 in Studying Youth Gangs, edited by James F. Short and Lorine
A. Hughes. Lanham, Md.: AltaMira Press.
Perkins, Useni Eugene. 1987. Explosion of Chicago’s Black Street Gangs: 1900 to the
Present. Chicago: Third World Press.
Pfautz, Harold W. 1961. “Near-Group Theory and Collective Behavior: A Critical
Reformulation.” Social Problems 9:167–74.
Phillips, Scott. 2003. “The Social Structure of Vengeance: A Test of Black’s Model.”
Criminology 41:673–708.
Pitts-Rivers, Julian. 1966. “Honor and Social Status.” Pp. 19–78 in Honor and Shame:
The Values of Mediterranean Society, edited by John Peristiany. Chicago: University
of Chicago Press.
Podolny, Joel M. 1993. “A Status-Based Model of Market Competition.” American
Journal of Sociology 98:829–72.
Polk, Kenneth. 1999. “Males and Honor Contest Violence.” Homicide Studies 3:6–29.
Rosenfeld, Richard, Timothy M. Bray, and Arlen Egley. 1999. “Facilitating Violence:
A Comparison of Gang-Motivated, Gang-Affiliated, and Nongang Youth Homi-
cides.” Journal of Quantitative Criminology 15:495–516.
Sherif, Muzafer. 1967. Group Conflict and Co-operation: Their Social Psychology. Lon-
don: Routledge & K. Paul.
Shils, Edward A., and Morris Janowitz. 1948. “Cohesion and Disintegration in the
Wehrmacht in World War II.” Public Opinion Quarterly 12:280–315.
Short, James F. 1997. Poverty, Ethnicity, and Violent Crime. Boulder, Colo.: Westview.
Short, James F., Jr., and Fred L. Strodtbeck. 1963. “The Response of Gang Leaders
to Status Threats: An Observation on Group Process and Delinquent Behavior.”
American Journal of Sociology 68:571–79.
———. 1965. Group Process and Gang Delinquency. Chicago: University of Chicago
Press.
Simmel, Georg. 1898. “The Persistence of Social Groups.” American Journal of Soci-
ology 3:662–98.
Spergel, Irving A. 1995. The Youth Gang Problem: A Community Approach. New York:
Oxford University Press.
Stewart, F. H. 1994. Honor. Chicago: University of Chicago Press.
Sullivan, Mercer L. 2005. “Maybe We Shouldn’t Study ‘Gangs’: Does Reification Ob-
scure Youth Violence?” Journal of Contemporary Criminal Justice 21:170–90.
Suttles, Gerald D. 1968. The Social Order of the Slum: Ethnicity and Territory in the
Inner City. Chicago: University of Chicago Press.
Thornberry, Terence P., and Marvin D. Krohn. 2003. “Comparison of Self-Report and
Official Data for Measuring Crime.” Pp. 43–94 in Measurement Issues in Criminal
Justice Research: Workshop Summary, edited by National Research Council. Wash-
ington, D.C.: National Academy Press.
Thornberry, Terence P., Marvin D. Krohn, Alan J. Lizotte, Carolyn A. Smith, and
Kimberly Tobin. 2003. Gangs and Delinquency in Development Perspective. New
York: Oxford University Press.
Thrasher, Frederic Milton. 1927. The Gang: A Study of 1,313 Gangs in Chicago. Chi-
cago: University of Chicago Press.
Tita, George, K. Jack Riley, Greg Ridgeway, Clifford Crammich, Allan F. Abrahamse,
and Peter W. Greenwood. 2003. “Reducing Gun Violence: Results from an Inter-
vention in East Los Angeles.” Santa Monica, Calif.: RAND Corporation.
Topalli, Volkan, Richard Wright, and Robert Fornango. 2002. “Drug Dealers, Robbery
and Retaliation: Vulnerability, Deterrence and the Contagion of Violence.” British
Journal of Criminology 42 (2): 337–51.
127
Valente, Thomas W. 2005. “Network Models and Methods for Studying the Diffusion
of Innovations.” Pp. 98–116 in Models and Methods in Social Network Analysis,
edited by Peter J. Carrington, John Scott, and Stanley Wasserman. New York:
Vigil, James Diego. 1988. Barrio Gangs: Street Life and Identity in Southern California.
Austin: University of Texas Press.
Villarreal, Andres. 2002. “Political Competition and Violence in Mexico: Hierarchical
Social Control in Local Patronage Structures.” American Sociological Review 67:
477–98.
Wasserman, Stanley, and Katherine Faust. 1994. Social Network Analysis: Methods
and Applications. Cambridge: Cambridge University Press.
White, Douglas R. 2004. Network Analysis and Ethnographic Problems: Process Mod-
els of a Turkish Nomad Clan. Lanham, Md.: Lexington.
Wilkinson, Deanna L. 2003. Guns, Violence, and Identity among African American
and Latino Youth. New York: LFB Scholarly.
Wilkinson, Deanna L., and Jeffrey Fagan. 1996. “The Role of Firearms in Violence
‘Scripts’: The Dynamics of Gun Events among Adolescent Males.” Law and Con-
temporary Problems 59:55–89.
Wolfgang, Marvin E. 1958. Patterns in Criminal Homicide. Philadelphia: University
of Pennsylvania Press.
Wright, Richard T., and Scott H. Decker. 1994. Burglars on the Job: Streetlife and
Residential Break-ins. Boston: Northeastern University Press.
Wyatt-Brown, Bertram. 1982. Southern Honor: Ethics and Behavior in the Old South.
New York: Oxford University Press.
Zimring, Frank. 1968. “Is Gun Control Likely to Reduce Violent Killings?” University
of Chicago Law Review 35:721–37.
128

The University of Chicago Press

Uploaded by

Copyright:

Available Formats

The University of Chicago Press

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The University of Chicago Press

Uploaded by

Copyright:

Available Formats

Murder by Structure: Dominance Relations and the Social Structure of Gang Homicide

Author(s): Andrew V. Papachristos

Most sociological theories consider murder an outcome of the dif-

䉷 2009 by The University of Chicago. All rights reserved.

74 AJS Volume 115 Number 1 (July 2009): 74–128

of homicide. The usual study attempts to explain differences in aggregate

the selection of murder victims. As corporate actions between groups,

MURDER AS INTERACTION: CONSEQUENTIALITY AND SOCIAL

future attacks precisely because it has proven its dominance; often, a

MURDER AS A GIFT: RECIPROCITY, DOMINANCE, AND THE

form an overarching contentious network that acts as a conduit for future

FROM GANG MURDER TO SOCIAL STRUCTURE: A NETWORK

formation since mutual protection is one of the most frequently cited

precisely because the existence of the group is predicated on the inter-

and gang information on both the offender and victim as recorded by

more than 30 years’ experience working with gang members in Chicago.16

THE SOCIAL STRUCTURE OF GANG HOMICIDE

(Thornberry et al. 2003). Individuals generally mature out of gang mem-

non-gang-related homicides.19 That is, the “spark” of most of these mur-

Racial and Ethnic Differences in Network Structure

Network Exposure and Power

DOMINANCE RELATIONS AND THE CREATION OF GANG MURDER

networks. More specifically, I analyze how dominance disputes, absolute

Predicting a Murderous Tie—Dyadic Analysis

in 1996–97, the graph of which is displayed in figure 5. Essentially, figure

Fig. 5.—Two-mode graph of gang turf in Chicago, 1996–97

names to their proper gang surnames—such as the 55th Street Disciples,

confrontations between two specific gangs are more circumscribed. As a

Predicting Reciprocity—Network Autocorrelation Models

where y is a binary vector indicating whether or not a murder in 1996–

at level 2, thereby treating murders as nested within gangs.32 Given the

murders are considerably less likely when a non–gang member is involved.

statistical significance of the contagion parameter drops to the most lax

of bitches [gangs perceived as weak or of lesser status] got corners they

The key here is the importance gang members attribute to reciprocity.

entails using spatial proximity of neighborhoods and their associated levels

1994 1998 2002

1994 1998 2002

———. 2007. “Murder by Structure: A Network Theory of Gang Homicide.” Ph.D.

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.