Social Contagion of Violence

Columbia Law School Scholarship Archive

Faculty Scholarship Faculty Publications

2006

Jeffrey Fagan Columbia Law School, [email protected]

Deanna L. Wilkinson [email protected]

Garth Davies [email protected]

Follow this and additional works at: https://scholarship.law.columbia.edu/faculty_scholarship

Part of the Criminal Law Commons, and the Law and Society Commons

Recommended Citation Jeffrey Fagan, Deanna L. Wilkinson & Garth Davies, Social Contagion of Violence, THE CAMBRIDGE HANDBOOK OF VIOLENT BEHAVIOR AND AGGRESSION, DANIEL J. FLANNERY, ALEXANDER T. VAZSONY & IRWIN D. WALDMAN, EDS., CAMBRIDGE UNIVERSITY PRESS, 2007; COLUMBIA PUBLIC LAW RESEARCH PAPER NO. 06-126 (2006). Available at: https://scholarship.law.columbia.edu/faculty_scholarship/1428

This Working Paper is brought to you for free and open access by the Faculty Publications at Scholarship Archive. It has been accepted for inclusion in Faculty Scholarship by an authorized administrator of Scholarship Archive. For more information, please contact [email protected]. Columbia Law School Public Law & Legal Theory Working Paper Group

Paper Number 06-126 (Version: 02/27/07)

SOCIAL CONTAGION OF VIOLENCE in THE CAMBRIDGE HANDBOOK OF VIOLENT BEHAVIOR Daniel Flannery, A. Vazsonyi, & I. Waldman (Eds.) Cambridge University Press, 2007

BY:

PROFESSOR JEFFREY FAGAN COLUMBIA LAW SCHOOL - AND - ASSOCIATE PROFESSOR DEANNA L. WILKINSON THE OHIO STATE UNIVERSITY -AND- ASSISTANT PROFESSOR GARTH DAVIES SIMON FRASER UNIVERSITY

This paper can be downloaded without charge from the Social Science Research Network electronic library at:

http://ssrn.com/abstract=935104

Electronic copy of this paper is available at: http://ssrn.com/abstract=935104 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

CHAPTER 36 Social Contagion of Violence

Jeffrey Fagan, Deanna L. Wilkinson, and Garth Davies

Introduction increase and decline as distinct phenomena with unique causes. Moreover, these Like many large American cities, New York causes are typically regarded as exogenous City experienced a sudden and dramatic to the people or areas affected. For exam- increase in homicides beginning in 1985. ple, the onset and severity of the homicide The homicide run-up was highest for ado- trend were attributed to the sudden emer- lescents, but rates increased quickly for older gence of unstable street-level crack mar- persons as well (Fsagan, Zimring, & Kim, kets, with high levels of violence between 1998). Unlike many other cities, in which sellers (Baumer, 1994; Baumer, Lauritsen, homicides declined gradually over the next Rosenfeld, & Wright, 1998; Fagan & Chin, decade yet remained above their pre-1985 1991; Fryer, Heaton, Levitt, & Murphy, 2005; levels, the increase in New York was fol- Grogger & Willis, 2000). Others suggested lowed by an even larger decline over the that drug markets created a demand for guns next 5 years. By 1995, homicide in New that in turn trickled down from drug sell- York City had dropped below its 1985 level; ers into the hands of adolescents (Blum- by 1996, it was lower than the 1985 rates; stein, 1995; Fagan, 1992). Structural theorists by 1998, homicide rates were 25% lower implicated race-speciﬁc economic deﬁcits in than the 1985 levels; and by 1998, homi- inner cities (Krivo & Peterson, 2000; Peter- cide rates were lower than three decades son & Krivo, 1993) or racial residential seg- earlier. The drop in crime generally, and regation (Massey, 1995). There have been not just homicide, was an order of mag- many claims regarding the sources of the nitude greater than any observed in large decline, including changes in police strat- American cities since the 1950s (Zimring, egy (Kelling & Cole, 1996), demographic 2007). changes (Cook & Laub, 1998; Eckberg, Explanations of this roller-coaster pat- 1995), incarceration (Blumstein & Beck, tern of violence beginning in the 1960s 1999), and lower demand for illegal drugs have tended to partition the periods of (Curtis, 1998).

688

Electronic copy of this paper is available at: http://ssrn.com/abstract=935104 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 689

None of the popular explanations of nomenon declines. This decline may occur either the increase or the decline is fully sat- because the density of contacts may decrease isfying. Moreover, the gap between the scale or because some form of resistance develops of demographic and policy changes and the that reduces the odds of transmission from scale of the crime decline suggests that there one person to the next, even in the pres- are processes at work other than these usual ence of exogenous contributing factors (Bai- suspects. Some have used the term “epi- ley, 1967; Burt, 1992). demic” metaphorically to describe the homi- In this chapter, we assess whether the cide run-up and decline, but with little pre- roller-coaster pattern of homicides in New cision and often conflating several features York City beginning in 1985 fits a con- of epidemics, such as social concentration, tagion model and identify mechanisms of spatial diffusion, and temporal spikes (see, social contagion that predict its spread across for example, Bailey, 1975). social and physical space. This framework Epidemic is a term used widely in for interpreting the homicide trends as an the popular and scientific literature to epidemic includes two perspectives. First, describe two quite separate components of the sharp rise and fall are indicative of a a phenomenon: an elevated incidence of the nonlinear pattern in which the phenomenon phenomena and its rapid spread via a con- spreads at a rate far beyond what would tagious process within a population in a be predicted by exposure to some exter- short period of time. For example, Gladwell nal factor and declines in a similar pattern (2000) describes how the incidence of an in which the reduction from year to year ordinary and stable phenomenon such as a exceeds what might be expected by linear seasonal flu can become epidemic when its regression trends. This leads to the second incidence increases in a very short time from perspective: the factors leading to its spread a predictable base rate to an elevated rate are not exogenous factors, as in the case of of infections. Moreover, epidemics need not contamination or disaster. Instead, the non- be contagious. Consider an outbreak of food linear increase and decline suggest that the poisoning from contaminated materials or a phenomenon is endemic to the people and cancer cluster near a polluted water supply. places where its occurrence is highest and These medical problems may occur at a rate that this behavior may be effectively passed well above an expected base rate, but are from one person to another through some not spread from person to person through process of contact or interaction. physical contact or an infectious process. In We assess the validity of these assump- contrast, an outbreak of influenza, the adap- tions in three ways. In the section “Violence tation of cultural fads, medical or industrial as Social Contagion,” we introduce a frame- innovation, or changes in the rates of antiso- work of social contagion that informs this cial behavior all reflect spread through inter- analysis. The diffusion and spread of social personal exposure to an “infectious” agent. behaviors are recurring themes in both the Although disease spreads through a host scientific and popular literatures. But there and agent (Robertson, 1990; Rothman, has been little theorizing on the mechanisms 1986), social contagion involves the mutual of social diffusion generally and specifically influence of individuals within social net- on the social interactions that may qualify works who turn to each other for cues and as a contagious process. This section reviews behavioral tools that reflect the contingen- the literature generally on social contagion cies of specific situations (Bailey, 1967; Burt, and then constructs an analytic framework 1987; Coleman, Katz, & Menzel, 1966). The to explain the spread of youth violence over contagious dimension is especially salient time and space. In the section “The Epidemi- during the upswing of an epidemic, when ology of Youth Homicide in New York City,” physical or social contact is critical to spread we analyze Vital Statistics data from the pursuant to exposure. But epidemics also New York City Department of Health and end, as the rate of new incidence of the phe- Mental Hygiene to construct simple time

Electronic copy of this paper is available at: http://ssrn.com/abstract=935104 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

690 jeffrey fagan, deanna l. wilkinson, and garth davies

series data that characterize the increase and this research, especially among adolescents: decline in homicides from 1985–2000.We trends in fashion and art (Gladwell, 2000), concentrate on homicides involving adoles- as well as problematic social behaviors, cents and young adults, populations who such as alcohol and drug use (Rowe & experienced the sharpest rise and decline in Rodgers, 1994), smoking (Rowe & Rodgers, homicide, both in New York and nationally 1991), teenage pregnancy (Crane, 1991), sui- (Cook & Laub, 1998; Fagan & Davies, 2004; cide (Berman, 1995; Gould, Wallenstein, & Fagan et al., 1998). Kleinman, 1990; Gould, Wallenstein, Klein- In the section “Neighborhood Effects man, O’Carroll, & Mercy, 1990), and delin- on Social Contagion of Youth Homicide,” quency (Jones, 1997). Common to each of we estimate models to identify the spa- these examples is the social structure of tial and social trends in youth homicide. transmission. Thus, the fundamental social Using growth curve or hierarchical mod- causes of disease – primarily social structural els, we disaggregate homicide and non- or social interactionist in nature – can be lethal injury data by neighborhoods over seen as pathways along which more micro- the 11-year period and fit models to demon- level causes can exert their effect (Farmer, strate the spatial diffusion of youth homi- 1999; Gostin, Burris, & Lazzarini, 1999,p. cide from one neighborhood to the next 74; Lynch et al., 1998; Tolnay & Beck, 1995; across New York City. By co-varying neigh- Tolnay, Deane, & Beck, 1996; Wilkinson & borhood social and economic characteristics Fagan, 1996). with temporal homicide trends, we are able The spread of ideas, behaviors, and prac- to show that the diffusion of homicide in this tices is contingent on the way in which social era was specific to the most socially isolated structure brings people together in close areas of the city. We isolate gun homicides physical proximity within dense social net- as the contagious agent, showing that it is works (Burt, 1987,p.1288). For example, gun homicides that diffused across New York Rowe and Rodgers (1994) show that that City neighborhoods and that gun homicides an epidemic model combining social con- retreated just as quickly. tagion through social contacts among ado- In the section “Violent Events, Social Net- lescents within a narrow age band explains works, and Social Contagion,” we present the onset and desistance of adolescent sex- data from interviews with young males ual behavior (see also Rodgers & Rowe, active in gun violence during this time. Their 1993). HIV transmission also has been mod- reports of the role of guns in violent events eled as a contagious epidemic (May, Ander- further specify how diffusion may in fact be son, & Blower, 1990). Through a process the result of a dynamic process of social con- of mutual influence involving contact, com- tagion. We conclude the chapter by inte- munication, and competition, adoption of grating these perspectives into a unifying behaviors occurs when information about framework that links elements of models of behaviors is transmitted in a way that com- infectious disease with social interactionist municates the substance of an innovation perspectives to explain the social contagion and the consequences of its adoption. The that contributed to New York City’s homi- consequences can be socially rewarding or cide epidemic, which remains today as the intrinsically pleasurable and may be rein- source of contentious debates in public pol- forced through the benefits of a vicari- icy and social science. ous experience or a trial use. In addition, these behaviors acquire social meaning that is communicated through repeated interac- Violence As Social Contagion tions within social networks (Kahan, 1997; Lessig, 1995,p.1947). Background Contagious epidemics involve the trans- There are many examples of social con- mission of an agent via a host through tagion that inform the development of susceptible organisms whose resilience is P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 691

weakened by other conditions or factors event schema used to organize information (Bailey, 1967). Susceptibility is critical to about how people learn to understand and the ability of an agent to exert its process enact commonplace behavioral patterns. A on a host. This medical rendering of conta- “script” is a cognitive structure or frame- gion can be analogized to social contagion work that organizes a person’s understand- (Jones & Jones, 1994, 1995). Thus, the fun- ing of typical situations, allowing the person damental social causes of disease – primar- to have expectations and to make conclu- ily social structural, or ecological – can be sions about the potential result of a set of seen as pathways along which more micro- events. Script theory has been used widely level causes can exert their effect (Gostin in social psychology to identify patterns of et al., 1999,p.74). According to Gostin and decision making and social interactions that colleagues (1999), these fundamental social persist among persons within social net- causes reflect inequalities that work in two works. Script theory can explain contagion ways. First, these conditions increase expo- in several ways: (1) Scripts are ways of orga- sure to the more proximal causes, whether nizing knowledge and behavioral choices; (2) microbic or behavioral. Second, they com- individuals learn behavioral repertoires for promise the resistance or resilience of social different situations; (3) these repertoires are groups to these proximal causes. That is, stored in memory as scripts and are elicited their exposure and their behavior in those when cues are sensed in the environment; structural circumstances both have social (4) the choice of scripts varies among indi- roots. viduals, and some individuals will have lim- Memetics provides a complementary ited choices; (5) individuals are more likely framework for understanding how beliefs, to repeat scripted behaviors when the pre- ideas, and behaviors spread throughout soci- vious experience was considered successful; ety. Memes are singular ideas that evolve (6) scripted behavior may become “auto- through a process of natural selection not matic” without much thought or weighing unlike the evolution of genes in evolution- of consequences; and (7) scripts are acquired ary biology (Balkin, 1998; Lynch, 1996). The through social interactions among social net- principal law governing the birth and spread work members (Abelson, 1976, 1981). of memes is that of the “fittest ideas,” defined Accordingly, social contagion is the con- as those ideas that are the best at self- vergence of transmission of behaviors and replication rather than those that may be of beliefs that motivate or sustain them. truest or have the greatest utilitarian value Social contagion arises from people in prox- (Lynch, 1996). In the present analysis, vio- imate social structures using one another lence may be the “fittest” behavior, even to manage uncertainty of behavior (Burt, when it contradicts more socially useful nor- 1987; Gostin, 1991; Rodgers & Rowe, 1993; mative values imported from the dominant Rowe & Rodgers 1991, 1994). It requires society. Memes achieve high-level contagion an interaction in which information, behav- through a variety of social interactions across ioral innovation, belief, or meme is trans- social units, such as families and social net- mitted across a social synapse. At its core, works, and each mode increases the “host” contagion occurs when two people inter- population for that meme. The meme is then act where one has adopted a construct and reproduced within networks and transmit- the other has not. Contact, communication, ted across interstitial network boundaries. and imitation are influential processes that Replicated memes become what Balkin make transmission possible (Burt, 1987,pp. (1998,pp.42–57) refers to as “cultural soft- 1288–1289). Synapses themselves are situ- ware” that is expressed in language, behavior, ated within social networks, and the adop- and normative beliefs, creating a set of nor- tion of an innovation or a meme triggers mative expectations or behavioral “scripts.” the adoption by another person. Burt (1987) (Abelson, 1976, 1981; Fagan, 1999). Accord- suggests that adoption has less to do with the ing to Abelson, the script framework is an cohesion of people within social structures P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

692 jeffrey fagan, deanna l. wilkinson, and garth davies

or networks and more to do with the struc- interactions (Wilkinson and Fagan, 1996). tural equivalence – the social homogeneity – Everyday disputes, whether personal insults of the network. That is, transmission is more or retributional violence, in turn are more likely to occur between similarly situated likely to be settled with potentially lethal persons – siblings, fellow graduate students, violence (Fagan & Wilkinson, 1998a; Wilkin- street corner boys – than persons simply son, 2003). because they are closely bonded (Burt, 1987, Whether viewed in social, medical, or p. 1291). memetic frameworks, guns can be con- Within structurally equivalent networks, structed as a primary agent of violence con- similarly situated people are likely to influ- tagion over the most recent epidemic cycle. ence or adopt behaviors from one another Guns are a form of social toxin (Delgado, that can make that person more attractive 1985; Fagan & Wilkinson, 1998b) in every- as a source of further relations. The impor- day social interactions, altering the outcome tance of structural equivalence – or place- of disputes and changing the developmen- ment within a socially homogeneous inter- tal trajectories of young males whose ado- personal network – is that it fosters intercon- lescent development took place in the con- nected patterns of relationships that make texts of high rates of gun use and widely contagion efficient. perceived danger, contributing to an ecol- In the remainder of this section, we ogy of danger that had profound develop- show how transmission of violence occurs mental impacts on adolescents growing up across neighborhoods whose social struc- in these settings (Bingenheimer, Brennan, & tures of densely packed networks are vulner- Earls, 2005; Fagan, 1999). able to rapid contagion. Our previous work The development of such an ecology of on urban youth violence has shown how the danger reflects the confluence and interac- memes of toughness and the valued status tion of several sources of contagion. First is from violence are the object of transmission the contagion of fear. Weapons serve as an and exchange among similarly situated male environmental cue that in turn may increase youth (Fagan & Wilkinson, 1998a,b; Wilkin- aggressiveness (Slaby & Roedell, 1982). Ado- son, 2003). The implications for a social lescents presume that their counterparts are influence model of contagion are discussed armed and, if not, could easily become in the concluding section. armed. They also assume that other adolescents are willing to use guns, often at a low threshold of provocation. Guns and Social Contagion The second source is the contagion of Several processes have contributed to the gun behaviors themselves. The use of guns epidemic of lethal violence. The growth in has instrumental value that is communi- illegal markets heightens the demand for cated not only through urban “myths” but guns as basic tools that are associated with also through the incorporation of gun vio- routine business activity in illegal markets lence into the social discourse of everyday (Blumstein, 1995; Johnson, Williams, Dei, & life among pre-adolescents and adolescents. Sanabria, 1990). In turn, the increased pres- Guns are widely available to adolescents ence of weapons and their diffusion into the (Cook & Ludwig, 2004), and when car- general population change normative per- ried by adolescents, they are frequently dis- ceptions of the danger and lethality associ- played (Harcourt, 2006; Wilkinson, 2003).1 ated with everyday interpersonal disputes, They are salient symbols of power and sta- giving rise to an “ecology of danger” (Fagan tus and strategic means of gaining status, & Wilkinson, 1998a). Thus, we hypothesize domination, or material goods (Wilkinson, that guns were initially an exogenous factor 2003). Wilkinson’s interviews with adoles- in launching an epidemic of gun homicide, cents and young adults in two New York but became endogenous to socially isolated City neighborhoods during the mid-1990s neighborhoods and came to dominate social show that guns are used in a myriad of P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 693

different ways and for different purposes; work to hinder nonviolent conflict resolu- some uses are seemingly more mild and sym- tion. Conflict handling among youth who bolic than others, but may be the first steps are affiliated with other youth at least in or building blocks for later more serious use. part to enhance their personal safety in a For example, very first steps might be sim- dangerous environment acts to increase the ply seeing someone or knowing someone amount of violence that youth experience, who has a gun, then looking for a gun in rather than decrease it. We examine these one’s own house, then maybe trying to get issues empirically at the neighborhood level one, then just flashing one when trying to as well as at the micro-translational level in threaten/scare an opponent, then using the New York City. gun for pistol whipping, then firing the gun to scare someone but not aiming to hit them, The Mirco-Processes of Social Contagion: then actually firing toward someone, then fir- Social Identity and Violence ing to injure but not kill, and then firing to kill. The socialization process into the urban Previously, Wilkinson (2003) and Fagan and youth gun world begins at a young age, with Wilkinson (1998a) identified several pro- influences coming from family and peer net- cesses operating at the event level that illus- works (Wilkinson, 2003; Wilkinson & Fagan, trate the spread of violence within social 2001a). How these processes unfold at the structures and that exert a contracting influ- event level is explored in Part IV of this ence on social networks of adolescents: (1) chapter. Achieving a highly valued social identity Third is the contagion of violent identities occurs through extreme displays of violence, and the consequent eclipsing or devaluation (2) achieving a “safe” social identity may also of other identities in increasingly socially iso- require the use of extreme forms of violence, lated neighborhoods. These identities rein- (3) the ready availability of guns clearly force the dominance hierarchy built on increases the stakes of how one achieves “toughness” and violence, and its salience status, (4) much behavior is motivated by devalues other identities. Those unwilling avoiding being a punk or “herb” (sucker or to adopt at least some dimensions of this weakling), (5) identities can change from identity are vulnerable to physical attack. being a punk or herb into a more posi- Accordingly, violent identities are not sim- tive status such as “hold your own,” (6) ply affective styles and social choices, but guns equalize the odds for some smaller instead are strategic necessities to navi- young men through the process of “show- gate through everyday dangers (Wilkinson, ing nerve,” (7) one can feel like a punk for 2003).2 a specific situation but not take on a punk Finally, when the group nature of youth identity, and (8) one can feel like a “crazy” violence is examined, diffusion and conta- killer in a specific situation but not take gion of attitudes, scripts, and behaviors are on a “crazy” or killer identity. If “compul- clearly visible. The proximal link between sive masculinity” or Anderson’s (1994, 1999) one violent conflict and the next is startling. “street orientation” is dominant in public In addition, the social meanings of violent spaces and personal safety as our data sug- events reach a broader audience than those gest, then those who do not conform will be immediately present in a situation. Each victimized. violent event or potentially violent interac- The maintenance and reinforcement tion provides a lesson for the participants, of identities supportive of violence are first-hand observers, vicarious observers, and made possible by an effective sociocultural others influenced by the communication of dynamic that sets forth an age-grading path- stories about the situation that may fol- way to manhood that includes both behav- low. Expectations, a violent status hierar- iors and the means of resolving violations chy, and norms of interpersonal conduct of respect. Wilkinson (2003) described the among groups of socially isolated young men strong influence of street code, similar to P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

694 jeffrey fagan, deanna l. wilkinson, and garth davies

the codes identiﬁed by Anderson (1999)or 1960 to 31.0 in 1995.By1996, the rate had the code of honor described by Toch (1969), receded to 13.9 per 100,000, a level unseen over the behaviors of young children, adoles- since 1968. Figure 36.1 shows the gun and cents, and young adults. The absence of alter- nongun homicide rates for 1968–2000. native means of attaining valued masculine From 1900 through the beginning of the identities further compounds the problem. run-up in homicide in the mid-1960s, and The transmission of these social processes with one exceptional era following the pas- occurs on both the micro and macro lev- sage of the Volstead Act in 1919, homi- els. Children growing up in this environment cide rates in New York City varied narrowly learn these codes, or behavioral-affective sys- between 3.8 and 5.8 per 100,000 population tems, by navigating their way through inter- (Monkonnen, 2001).3 From 1965 to 1970, personal situations that often involve violent the average annual homicide rate rose from encounters. 7.6 to 12.6 and rose again to 21.7 by 1975. Thus, homicide in New York nearly tripled within a decade. The rates remained elevated The Epidemiology of Youth Homicide above the 1968 rates until 1998. Accord- in New York City ingly, Figure 36.1 suggests that, for three decades, homicides in New York were nor- Historical and Current Homicide Trends malized at an elevated rate and were for a The epidemic of youth violence in New long time characteristic of the city’s social York City since the mid-1980s is best under- landscape. Thus, the escalation in killings stood in a social and historical context that until the 1990s was cumulative, with each spans nearly 35 years. Like the nation’s new peak building on the elevation of the largest cities, New York experienced a sharp base rate established in the previous peak. increase in homicide and other violence rates One interpretation of the recent decline may beginning in the mid-1960s. The homicide simply be the recession of this longer-term rate rose from 4.7 per 100,000 population in social and historical trend.

Rate per 100,000 Population 10

1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000

GUN NONGUN TOTAL Figure 36.1. Gun and nongun homicide rates per 100,000 persons, 1968–2000, New York City. Source: Ofﬁce of Vital Statistics and Epidemiology, New York City Department of Health and Mental Hygiene, various years. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 695

Figure 36.1 also shows that this long- different from the preceding peaks in five term trend involves three sub-epidemics. important ways: (1) Its starting point was The first of these peaked in 1972, the second lower than the starting point for the pre- in 1981, and the third in 1991. Each coincided vious (1981) peak, (2) its peak was about temporally with drug epidemics and the 15% higher than the preceding peak, (3)it growth of retail drug markets: heroin in the had a far greater share of gun killings, (4) early 1970s (Agar & Reisinger, 2002; Egan its decline was far steeper than any previous & Robinson, 1979; Hunt & Chambers, 1976; decline, and (5) homicides have remained at Inciardi, 1979), the emergence of urban their low point far longer than in any of the street drug markets in the late 1970s where previous three epidemics. powdered cocaine was openly sold (John- To illustrate the extent of the differences son et al., 1990; Johnston, 1987; Williams, between the 1985–2000 cycle and its prede- 1989; Zimmer, 1987, 1990), and crack begin- cessors, Figure 36.2 presents the data from ning in 1985 (Johnson et al., 1990). The suc- Figure 36.1 normed to the 1985 base.4 Gun cessive epidemics were cumulative in their killings accounted for all of the increase trends, not distinct. To re-introduce an idea in homicides since 1985 and most of the from Part I, the pattern of killings in particu- decline. Although the declines after 1992 in lar resembles a roller coaster, with an ascent nongun killings are a continuation of the through the late 1970s to a relatively low 8 years of previous decreases, the increase peak, a return to near the previous low point, and decline in gun killings are evidence of a and a sharp increase to a high peak in 1990 homicide spike that is unique from its pre- followed by a sharp drop. decessors. Nongun killings declined steadily Figure 36.1 shows the growing impor- since 1986 and by 1996 were about half the tance of guns in homicides in each of 1985 rate, suggesting a secular decline in the three peaks. Increases in both gun nongun homicide that may be independent and nongun homicides contributed to the of the gun homicide epidemic. tripling of homicide rates through 1972.In The epidemic pattern was well observed 1972, the ratio of gun to nongun homi- through homicide data, but public health cides was 1.23. By the next peak in 1981, data on nonlethal violence in part tell the the 1,187 gun deaths were nearly 1.76 times same story. Data from the City’s Injury greater than the 673 nongun homicides. Surveillance System provide information In 1991, the modern peak, the 1,644 gun on hospitalizations for intentional injuries homicides were 3.16 times greater than (NYC Department of Health and Mental the 519 nongun homicides. In addition to Hygience, 1997, and various years). Figure sharp increases in the number of gun homi- 36.3 shows that the decline in homicides was cides, the gun:nongun ratio also rose sharply accompanied by a general decline in nonfatal because of a long-term decline in the num- assaults. These data were available for anal- ber of nongun homicides. Since 1980, the ysis beginning in 1990, about the same time number and rate of nongun homicides have that the homicide epidemic reached its peak. declined by nearly 50%, from 735 to 335 Figure 36.3 shows that, beginning in 1990, nongun killings in 1996. There are thus two nearly all the decline in nonfatal assaults dynamic and different patterns in the data were declines in gun assaults; assaults by on homicide by weapon. Gun killings follow other means, such as blunt instruments or the roller-coaster pattern of steadily increas- cutting instruments (e.g., knives), declined ing peaks beginning in 1972. Nongun killings at a much lower rate. So, the rise and fall in trend down from 1980 to rates unseen since homicides did not necessarily reflect changes 1960. This long-term secular trend in nongun either in the lethality of gun violence or a killings is substantial, but it has not previ- change in the case-fatality rate. Rather, gun ously been noticed. violence declined generally over time follow- Figure 36.1 also shows that the recent ing increases that also were almost exclu- cycle beginning in 1985 was qualitatively sively the result of gun violence. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

696 jeffrey fagan, deanna l. wilkinson, and garth davies

225

200

175

150

125

100 1985 Base = 100 = Base 1985

8 6 8 0 2 8 0 2 4 6 7 7 8 8 9 9 9 9 974 9 9 9 9 9 9 9 000 1 1970 1972 1 1 1 1 198 1984 1986 1 1 1 1 1996 1998 2

GUN NONGUN TOTAL

Figure 36.2 . Gun and nongun homicide rates, 1968–2002, indexed to 1985 rates, New York City. Source: Ofﬁce of Vital Statistics and Epidemiology, New York City Department of Health and Mental Hygiene, various years.

The importance of the most recent era of titled “The Epidemiology of Youth Homi- homicide rise and decline, not only in its epi- cide in New York City” on the patterns of demiological pattern but also in its inﬂuence homicide and violence in this era, and espe- on law and policy (Feld, 1999; Stuntz, 2001; cially on the period from 1985–1996, when Zimring, 1999), leads us to focus the anal- the rise and fall in homicide were the most ysis in the next part and also in the section dramatic and acute.

16000

14000

12000

10000

8000

6000 Count of Injuries

4000

2000

0 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 Assault Homicides x 5 Gun Assaults Gun Homicides x 5 Total Violence Total Gun Violence

Figure 36.3. Gun and nongun homicide and assault by ﬁrearm, 1985–2000. Source: Ofﬁce of Vital Statistics and Epidemiology, New York City Department of Health and Mental Hygiene, various years; Injury Prevention Bureau, New York City Department of Health and Mental Hygiene, various years. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 697

than the rates of gun homicides. Through- The Social Structure of Homicide out the period, the changes in rates for The demographic patterns of gun and females were quite small, and not far from nongun homicide victimization during this the expected rates historically. The same is period tell a series of interesting stories, true of male nongun killings. Accordingly, some predictable and others surprising. this epidemic is confined to gun killings First, the homicide trends for women differ among males. from the patterns for men. Any benefits of the end of the violence epidemic accrued to age men; women’s risk of violent victimization Much of public and scholarly attention on remained stable at a level far lower than violence in the past decade has focused on that of men. Second, changes in adolescent the increase in gun homicides by adoles- homicide rates were accompanied by paral- cents (Blumstein, 1995; Cook & Laub, 1998). lel but less dramatic changes among older Trends nationwide show that gun homicide populations. This trend varies from the rates for adolescents increased during this national picture of steadily declining rates period while gun homicide rates for persons among older groups. Third, as we saw ear- over 25 years of age were declining. In New lier, the homicide run-up and decline were York City, homicides were not confined to concentrated in gun killings. Fourth, the younger age groups, but were a serious prob- homicide epidemic was concentrated among lem across a wide age range from 15 to 34 non-Whites. We observed these trends for years of age. Table 36.1 shows that gun homi- both homicide victims and offenders, which cide rates were higher than nongun rates for are discussed below. The data are available all age groups. For each year, gun homicide at: http://www2.law.columbia.edu/fagan/ rates were highest for persons aged 20 to 24 researchdata/contagion. in all years. Gun homicides by adolescents aged 15 to 19 rose more sharply over this gender period than other older population groups. Nearly all the increase and decline in killings Nevertheless, although adolescent participa- from 1985–1995 were gun homicides of tion in gun homicide rose sharply from 1985– males. The rate of gun killings among males 1991, rates for other age groups also contin- doubled from 21.8 per 100,000 in 1985 ued to rise during this period. to 44.5 in 1991. Nongun killings of males Gun homicide rates declined sharply for declined steadily throughout this period all three age groups from 1992–1995,to and by 1995 were less than half the 1985 about 50% of their peak rates in 1991, and rate. Killings of males were increasingly gun were about the same as their 1985 rates. events: the ratio of gun to nongun homicide Although the post-1991 decline was precip- victimizations of males increased from about itous for adolescents aged 15 to 19, their 1.5:1 in 1985 to 3.23:1 in 1995. 1995 gun homicide rates remained 25% The temporal patterns were similar for above their 1985 base rate. Table 36.1 also females. The rate of gun homicides of shows that nongun homicide rates declined females peaked in 1991, the same year steadily for all age groups and by 1995 were as males, and sustained their peak rate 50% or more lower than their 1985 rates for approximately 3 years before dropping (NYC Department of Health and Mental sharply in 1994.By1995, gun homicides Hygience, 1997). Similar to gun homicide for females had dropped 5% below their rates, the nongun homicide rates were high- 1985 levels. Nongun homicides of women est for persons aged 20 to 24. declined steadily throughout this period, from 4.7 per 100,000 in 1985 to 3.8 per race 100,000 in 1995. But unlike males, the rates Nationally, virtually all increases in homicide of nongun homicides of females were higher rates from 1985 to 1990 among people 10 to P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

698 jeffrey fagan, deanna l. wilkinson, and garth davies

Table 36.1: Adolescent gun and nongun Table 36.2 : Gun and nongun homicide homicide rates by age, New York City, rates per 100,000 persons by race, 1985–1995 1985–1995

Year <15 15–19 20–24 25–34 ≥35 Gun Nongun Year White Non-White White Non-White Gun Homicides 1985 0.620.834.121.17.1 1985 10.112.88.010.6 1986 1.119.944.125.66.9 1986 10.915.27.311.3 1987 1.233.150.625.38.5 1987 9.820.76.110.1 1988 1.545.556.833.59.2 1988 13.223.66.410.1 1989 1.343.160.037.99.0 1989 15.123.26.78.9 1990 2.250.966.342.210.8 1990 18.825.18.99.0 1991 1.057.466.342.212.6 1991 19.426.66.47.8 1992 1.348.368.142.811.0 1992 19.324.36.06.0 1993 1.648.563.837.511.7 1993 17.224.55.57.2 1994 0.739.647.530.08.3 1994 12.818.95.75.8 1995 0.826.034.718.16.9 1995 9.512.84.55.5 Nongun Homicides 1985 2.910.415.712.29.4 1986 3.48.017.714.18.4 1995 period illustrates the concentration of 1987 1.86.314.112.77.8 the homicide epidemic in gun homicides 1988 3.06.114.612.37.9 among non-Whites. For Whites, the ratio 1989 2.66.512.313.47.0 rises from 1.26:1 in 1985 to a peak of 3.23:1 1990 3.412.815.514.77.4 in 1992, before receding to 2.1:1 in 1995. 1991 3.18.011.110.96.5 For non-Whites, the ratio rises from 1.20:1 1992 2 469867962 . . . . . in 1985 to a peak of 4.05:1 in 1992 and 1993 4 139708965 . . . . . recedes to 2.32:1 in 1995. However, the nar- 1994 3.16.17.08.85.7 row difference between Whites and non- 1995 2.65.26.86.65.1 Whites may reﬂect the inclusion of Hispan- ics among the Whites in the population and 34 years of age were due to deaths of African homicide counts. The extent of this bias can American males. Most of these were ﬁrearm be seen in 1993 data from the New York fatalities that were overwhelmingly concen- City Department of Health injury surveil- trated demographically and spatially among lance system. The mortality and morbidity African American males in urban areas rates of gunshot wounds for Hispanics are (Fingerhut, Ingram, & Feldman, 1992a,b.) 228 per 100,000 persons, compared to 302 Table 36.2 shows that the trends in New for African Americans and 60 for Whites York mirror these national trends. Unfor- in that period (NYC Department of Health tunately, none of the data sources permit- and Mental Hygiene, 1992; New York State ted detailed disaggregation of the homicide Department of Health, 1994). trends by ethnicity over the entire 1985– 1995 period. Detailed data were available victim-offender homogeneity only for African Americans; Whites and His- Most homicides are within-group events, panics were not distinguished in the police especially with respect to gender, race, and or Vital Statistics data until after 1990. ethnicity (Cook & Laub, 1998; Sampson & Therefore, our analysis is limited to com- Lauritsen, 1994).5 We analyzed data from parisons between whites and non-Whites; police reports on the within-age distribu- non-Whites are primarily persons of African tion of homicide events for each year in descent, including some Hispanics. the recent homicide cycle to estimate the The within-race ratio of gun to nongun proportion of homicides in which victims homicide rates for each year in the 1985– and offenders both came from their own P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 699

age group. From 1985–1995, we observed to a process of “status forcing” that promotes within-group homogeneity with respect to cross-age interactions (Wilkinson, 2003). our limited categories of race, a trend evident also in national data (Cook & Laub, contextual effects 1998). Age homogeneity was more varied Both popular and social science explana- and depended on the method of homicide. tions of the homicide epidemic in New Age homogeneity for gun homicides was York and elsewhere have focused on social highest for homicide offenders aged 25 to 34 trends, particularly changes in drug markets and lowest for 20-to24- year-olds. The low (Baumer, 1994; Baumer et al., 1998; Blum- rates for the group aged 20 to 24 reflects their stein, 1995; Cork, 1999; Grogger & Willis, age status between the two other groups and 2000). Fagan et al. (1998) discuss the appeal the higher likelihood of cross-age interac- of these explanations. First, homicide and tions. drug epidemics have been closely phased, Age differences widened during the both temporally and spatially, in New York homicide cycle beginning in 1985.Atthe and nationwide, for nearly 30 years (Fagan, outset of the homicide run-up in 1985, age 1990; Fagan & Wilkinson, 1997). Homi- homogeneity for gun and nongun homicides cide peaks in 1972, 1979, and 1991 mir- was low: about one gun homicide in four ror three drug epidemics: heroin, cocaine involved persons within the same age cate- hydrochloride (powder), and crack cocaine. gories. For adults aged 25 to 34, about 4 in These long-term trends predict that trends 10 gun homicides were within-age killings. A in drug use would occur contemporane- year later, both age homogeneity and homi- ously with trends in homicide. Second, the cide rates increased. Within-age gun homi- emergence of volatile crack markets in 1985 cides for young adults aged 20 to 24 rose is cited as one of the primary contextual from 22.9% of gun homicides in 1985 to factors that have driven up homicide rates 35.8%in1990; for offenders aged 25 to 34 in New York (Bourgois, 1995; Goldstein, years, within-age group homicides rose from Brownstein, Ryan, & Bellucci, 1989; John- 38.5%in1985 to 54.9%in1989. Even with son et al., 1990). Competition between sell- these increases, however, the majority of gun ers, conflicts between buyers and sellers, and killings involved persons from different age intraorganizational conflict were all contrib- groups. During the same period, within-age utors to lethal violence within crack markets nongun homicide rates varied from year to (Fagan & Chin, 1989, 1991; Hamid, 1990). year in an inconsistent pattern. Crack also is implicated in the decline of The results are not surprising: age strati- homicide since 1991 (Curtis, 1998). fication of peer groups has traditionally cre- Figure 36.4 compares trends in gun homi- ated age-specific social networks. Age grad- cides for three age groups with trends in ing is a hallmark of street gangs (Klein, drug overdose deaths. Drug overdose deaths 1995) and adolescent cliques (Schwendinger follow a pattern of short cycles, with rela- & Schwendinger, 1985). These rigid age tively brief periods of increase and decline. boundaries offered few opportunities for The rates increase from 1986 to 1988, decline cross-age social interactions among delin- through 1990, and increase again for 3 quent groups. But contextual changes in years before leveling off. The run-up of gun street corner life in inner cities, where homi- homicide rates in 1985 to 1988 matches an cides were concentrated throughout this increase in drug overdose deaths, but homi- period, contributed to a breakdown of tradi- cides continued to increase even as drug tional age grading. The emergence of street overdose deaths declined. Drug overdose drug markets and dense street corner groups death rates increase from 1992 to 1994, even of males not in the workforce contributed as gun homicide rates decline. Accordingly, to a mixing of the ages on the street. Among there appears to be little mutual influence adolescents and young adults, competition of drug overdose deaths and gun homicide for street status through violence contributes trends for any of the three age groups.6 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

700 jeffrey fagan, deanna l. wilkinson, and garth davies

20 Age-Specific Rate per 100,000 per Rate Age-Specific 10

0 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 Drug Overdose Deaths Gun Homicides 15 – 19 Gun Homicides 20 – 24 Gun Homicides 25 – 34

Figure 36.4. Drug overdose death and age-speciﬁc gun homicide victimization rate, 1985–1995. Source: Ofﬁce of Vital Statistics and Epidemiology, New York City Department of Health and Mental Hygiene, various years.

Changes in drug use patterns may explain arrests, the two drugs that were traded most this disjuncture, with drugs such as heroin actively in street markets. The portion of returning as the favored street drug and felony drug arrests that involved crack or displacing crack and crack markets (Curtis, cocaine rose from 57%in1986 to 64% 1998). These drugs are more likely to cause in 1988 and declined steadily to 48%in overdose deaths. 1995. An alternative indicator of drug market These figures show that neither drug sell- activity is drug arrests. The size, location, ing activity nor increases in problematic drug and intensity of drug markets can be approx- consumption adequately explain the run- imated by drug arrest rates (Cork, 1999; up and decline in gun homicides. Violence Rosenfeld & Decker, 1999). Accordingly, associated with drug use remained relatively drug arrests reflect both strategic decisions infrequent during the onset of the crack cri- by police and drug market characteristics. sis (Fagan, 1992; Fagan & Chin, 1989, 1991). In conjunction with other indicators, arrests Moreover, the share of homicides due to are a useful marker of drug trends. However, drug selling did not rise during the homi- the trend lines in Figure 36.5 for age-specific cide run-up (Goldstein et al., 1989). Drug homicide victimization rates and felony drug selling accounts for an unknown propor- arrest rates show little relationship between tion of homicides, with estimates ranging gun homicides and drug arrests. Both drug from about 10% nationwide in the FBI’s arrests and gun homicide rates increase from Supplemental Homicide Reports (FBI, var- 1986 through 1989, but the trend lines move ious years) to 50% in local studies in New in different directions after that. Homicides York (Goldstein et al., 1989) or Los Ange- increase through 1991 for adolescents and les (Klein, Maxson, & Cunningham, 1991). 1992 for young adults. Drug arrests decline Thus, a decline in street-level drug sell- from 1990 through 1993 and begin to rise ing activity may have reduced, to some again in 1994. Most of these felony drug unknown extent, the types of social inter- arrests were for sale or possession with intent actions that lead to gun killings. But drug to sell, and most were either crack or cocaine selling alone is unlikely to have produced P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 701

250

200

150

100

Age-Specific Rate per 100,000 per Rate Age-Specific 50

0 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 Gun Homicides 15–19 Gun Homicides 20–24 Gun Homicides 25 – 34 Drug Arrests 10–19 Drug Arrests 20–24 Drug Arrests 25 – 34

Figure 36.5. Age-speciﬁc felony drug arrest rates and gun homicide victimization rates, 1985–1995. Source: Ofﬁce of Vital Statistics and Epidemiology, New York City Department of Health and Mental Hygiene, various years; New York City Police Department, various years.

the unprecedented run-up or decline in gun (see, for example, Crane, 1991; Gladwell, killings so consistently across time, social 2000). As the population declined, so too groups, and areas. did the rate of contacts. This is a plausi- Finally, demographic changes also offer ble but unlikely explanation. First, the age limited explanations for either the homicide decline was small: less than 10% in the high- increase or decline. The population for the est risk groups. Second, the breakdown in highest risk groups, non-White males aged age grading of violence during this period 15 to 29, declined by about 10% from 1985– may have mitigated cohort effects and dif- 1995 (Fagan et al., 1998), a far smaller scale fused behaviors broadly across age groups. of change than the change that could pro- Like the effects of declining drug markets, duce the observed declines in gun homicides. the contraction in the highest risk popula- Although it is tempting to dismiss demogra- tion is a potentially important inﬂuence in phy as a correlate of the homicide decline, the decline in ﬁrearm homicides from 1992– the relationship of population to a changing 1996, one that may contain the mechanisms behavioral pattern may be nonlinear (Glad- of decline. We turn to these mechanisms in well, 2000). In other words, did the popula- the next sections. tion decline reach a threshold where it could lead to a decline in the incidence of gun homicides? According to Burt (1987), net- Neighborhood Effects on Social work density promotes social contagion by Contagion of Youth Homicide increasing exponentially the extent of contact between persons within groups is non- We begin with a set of analyses that esti- linear. Perhaps population density among mate the probabilities of the diffusion of the highest risk groups rose during the run- behaviors across social areas or neighbor- up in violence and reached a threshold or hoods. We use census tract as the boundary tipping point at which behavioral change for “neighborhood,” based on the size (area) accelerated and spread through a popula- of tracts in New York and their isomor- tion before beginning its process of decline phism with important social units, such as P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

702 jeffrey fagan, deanna l. wilkinson, and garth davies

public housing developments and feeder tion in speciﬁc areas led to dynamic changes school patterns. Tracts are commonly used in the processes of socialization and social to represent neighborhoods in sociological control in those areas. As middle- and work- research because of their size and robust- ing class African American families moved ness in predicting variation in a variety of away from the inner cities when their jobs social interactions (see, for example, Land, left, there remained behind a disproportion- McCall, & Cohen, 1990). Other studies have ate concentration of the most disadvantaged estimated diffusion across similarly small segments of the urban populations: poor areas comprised of a few tracts that repre- female-headed households with children sent neighborhoods with meaningful social and chronically unemployed males with low boundaries (e.g., Crane, 1991; Fagan, West, & job skills. The secondary effects of this exo- Holland, 2003; Morenoff, Sampson, & Rau- dus created conditions that were conducive denbush, 2001). These analyses set the stage to rising teenage violence: the weakness of for the analysis in Part V of micro-social mediating social institutions (e.g., churches, interactions in which social interactions ani- schools), and the absence of informal social mate dynamics of social contagion that dif- controls to supervise and mentor youths.7 fuse violence across groups and places. Wilson (1987) refers to these conditions of weak social control as social isolation. The concept of social isolation suggests A. Susceptibility: Neighborhood Risk an ecological dynamic in which the com- We draw on the literature of neighborhoods ponents of poverty and structural disadvan- and violence to construct a framework of tage are interconnected with the dynamics structural risk that simultaneously compro- of social control and opportunity structures. mises resilience against transmission while The decline of manufacturing jobs increased increasing susceptibility. Both theory and unemployment among adult males, primar- empirical research suggest that neighbor- ily African Americans, whose lack of tech- hoods are susceptible to the spread of vio- nical skills and deep human capital limited lence when structurally weakened (Massey, them to low-wage and short-term unskilled 1995; Patterson, 1991; Roncek & Maier, 1991; labor positions. Other economic transfor- Rose & McClain, 1990; Taylor & Covington, mations, including the rise of service and 1988). For example, recent studies suggest technical jobs outside central cities, moti- that violence shares several explanatory vated the exodus of middle-class families variables with concentrated poverty (Samp- to the outer rings and suburbs surrounding son & Lauritsen, 1994; Wilson, 1987, 1991), the inner cities. Remaining within the aban- resource deprivation (Land et al., 1990; doned central cities were unskilled males Williams & Flewelling, 1988), and inequal- whose “marriage capital” was low, giving rise ity (Messner & Tardiff, 1986; Morenoff & to an increasing divorce rate and declining Tienda, 1997; Sampson, 1987; Shihadeh & marriage rate. Steffensmeier, 1994). These constructs Changes in the composition of central describe the lack of sufﬁcient means, city neighborhoods also weakened the social including income poverty and inequality, to institutions that were critical to the infor- sustain informal social control (Sampson, mal social control and collective supervision 1993; Sampson & Wilson, 1995). of youths. The weakening of social controls Wilson (1987) argues that there has been had their strongest effects in transactional both an economic and a social transforma- settings of neighborhoods and in places like tion of the inner city, in which the exodus of schools and church where adolescent devel- manufacturing jobs beginning in the 1970’s opment takes place. And, the exodus of has changed the social and economic com- middle-class families from inner cities weak- position of inner cities, leading to a concen- ened the political strength of the remain- tration of resource deprivation. He suggests ing families, leading to physical deterioration that the concentration of resource depriva- (Wallace, 1991), lower housing values, and in P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 703

turn increased residential (spatial) segrega- for the spread of gun violence from one tion (Massey & Denton, 1993). neighborhood to the next. An outward con- In turn, the social isolation of people tagion model posits that adolescent violence and families was extended to institutions spreads out from a central census tract (T) (Wacquant & Wilson, 1989). The rise in or the immediate neighborhood to adjacent poverty and weakening of social institutions census tracts (X, Y, and Z) or the surround- also undermined the presence of and insti- ing community. In this model, the incidence tutional support for conventional behaviors. and prevalence of adolescent homicide or As a result, conventional values and behav- assault violence in a given neighborhood iors were attenuated because they were not exert a significant influence over the inci- salient and had little payoff for one’s sur- dence and prevalence of adolescent violence vival or status (Elliott et al., 1996; Tienda, rates in the adjacent community. 1991; Wilson, 1987). These dynamics in turn This influence is hypothesized to operate attenuated neighborhood social organiza- in at least two different ways. First, a thresh- tion, increasing the likelihood that illegiti- old effect is expected concerning adolescent mate opportunity structures would emerge: homicide counts, such that the presence of illegal economies including drug distribu- at least one adolescent homicide in a given tion or extortion, gangs (Fagan, 1989, 1993; neighborhood will substantially increase the Brotherton et al., 2004), and social net- probability of experiencing at least one ado- works to support them. These structures lescent homicide in the surrounding com- competed with declining legal work oppor- munity. Second, with respect to adolescent tunities both as income sources and as homicide rates more precisely, positive co- sources for social status. As these networks variation is anticipated whereby increases or flourished, the systems of peer and deviant decreases in the adolescent homicide rate social control replaced the controls of social of violence in a neighborhood are reflected institutions and conventional peer networks in concomitant increases or decreases in the (Fagan, 1992, 1993). surrounding community’s adolescent homi- Accordingly, violence and homicide are cide violence rate. more likely to occur in an ecological con- It is also possible that the contagion effect text of weak social control, poorly super- of adolescent violence is reversed. Accord- vised adolescent networks, widespread per- ingly, the inward contagion model asserts ceptions of danger and the demand for lethal that the level of adolescent violence in an weapons, and the attenuation of outlets to immediate neighborhood is at least partially resolve disputes without violence. It is in contingent on the level of adolescent vio- this ecology of danger that violence becomes lence in its broader community. Again, the transmittable through weapons and their two distinct relationship forms (threshold impact on perception and decision making effect and positive co-variation) are possi- in social interactions. ble. By considering the simultaneous influence of adjacent spaces, we address the problem of spatial autocorrelation by effectively Analytic Models: Diffusion controlling for mutual influences within and and Contagion over time. We estimated models of contagion of gun In addition to corresponding adolescent violence (homicide and assault) and its violence rates, both the outward and inward diffusion across New York City neigh- contagion models incorporate relevant struc- borhoods from 1985–2000. Although the tural and demographic features of neighbor- sharpest changes in violence rates occurred hoods and communities as key explanatory from 1991–1995, we took advantage of data constructs. Thus, for the full outward con- through 2000 to chart the continuing pat- tagion model, the presence and rate of the tern of decline over the succeeding 5 years. adolescent homicide rate or assault rate in We tested two distinct conceptual models the surrounding community are a function P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

704 jeffrey fagan, deanna l. wilkinson, and garth davies

of the relevant characteristics of the commu- analytic strategy also assumes that neighbor- nity, as well as the presence or rate of adoles- hoods and communities have varying mean cent homicide violence in the neighborhood. violence levels and that they exhibit dis- In contrast, the full inward contagion model tinct trajectories of violence over time. We suggests that both relevant neighborhood account for the variance in average levels features and the presence or rate of adoles- of violence by specifying random intercepts cent homicide in the community predict the in our models. To estimate changes over presence or rate of adolescent violence in the time, we use a repeated measures design in neighborhood. Although these factors are which year is included as a random effect presumed to play a significant, independent that approximates a developmental growth role in the prediction of adolescent violence curve (Goldstein, 2003). We also estimate rates, it is nonetheless hypothesized that an autoregressive co-variance structure to effects of homicide violence rates as inde- account for the strong correlations of homi- pendent variables will remain substantial, cide and violence rates through time in cen- even once controls for relevant neighbor- sus tracts and their surrounding neighborhood and community characteristics have hoods. been introduced. We use the Poisson form of the model Models were estimated using mixed to estimate counts of both gun and nongun effects regression models. Mixed effects homicides and assaults. Poisson techniques regression models approximate multilevel are appropriate to identify factors that models in which data are hierarchically orga- predict the number of occurrences of an nized. For example, this class of models is event within a specific observation period useful in such cases as estimating the simul- (Gardner, Mulvey, & Shaw, 1995; Land, taneous effects of school climate and individ- McCall, & Nagin, 1996). Such “count” mod- ual student family background on standard- els are appropriate, given the relatively low ized test scores, or neighborhood character- number of homicides in most tracts in most istics and household composition on crime years and the sharp right-hand skew in the rates (Bryk & Raudenbush, 1992; Singer distribution of both homicides and assaults. & Willett, 2003; Snijders & Bosker, 1994). To estimate the adolescent component of Mixed effects models also are useful in esti- the total violence rate, including both homi- mating individual growth curves or within- cides and assaults, we included co-variates subject change over time. In these examples, that estimate the age composition of neigh- we might specify the ecological effects of borhoods including the ratio of youths aged school or neighborhood as fixed effects and 15–24 to persons over age 50 in each tract. individual household or family influences as Because homicide and assault victimization random effects. This specification approx- during this period was disproportionately imates the presumed hierarchy of influ- due to gun violence and also concentrated ences. One may reverse the specification as among African Americans, we estimate sep- well, comparing estimates to assess recipro- arate models for total homicide or assault, cal effects between the two sets of predic- gun homicide or violence, and gun victim- tors. Mixed effects models simulate the hier- ization for African Americans. archy of effects by estimating the differences in error-co-variance matrix structures for Data each set of effects (Singer & Willett, 2003).8 In this analysis, the contagion models dependent variables are specified using both fixed and random The dependent variables include counts of effects. The fixed effects of the neighbor- homicides, gun homicides, injury assaults hood (outward model) and adjacent com- that lead to hospitalization, and gun injury munity (inward model) structural charac- assaults. We also estimate specific counts of teristics and violence rates are interpreted these variables for African American vic- as standard regression coefficients. But the tims. These data were obtained from the P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 705

Table 36.3: Means and standard deviations for census tracts, New York City, 1990

Variable N Mean Standard Deviation

% Households with Public Assistance Income 2157 13.81 13.28 Gini for Total Household Income 2157 0.37 0.09 % Households Under Poverty Level 2157 18.10 15.06 % High-School Graduates – Total – 25+ 2174 66.33 16.34 % Employed in Managerial, Professional, or 2162 29.69 15.17 Technical Jobs Employment Rate 2164 90.27 7.08 Labor Force Participation Rate 2175 60.64 11.55 % Non-White 2175 54.86 36.17 Racial Fragmentation Index 2175 0.38 0.19 % Female Headed Households with Children <18 2157 10.09 10.43 Supervision Ratio (25–64 by 5–24) 2162 2.42 2.08 %Youth Population (5–15) 2175 14.69 6.74 Residential Mobility – Same House as 1985 2175 62.93 11.84 Population – 1990 2216 3304.41 2465.07 Foreign Born 2175 27.32 15.13 Linguistic Isolation 2175 10.95 10.13 Vacancy Rate 2160 5.53 5.93 %Occupied Units That Are Rentals 2157 65.10 25.89 Density – Mean Persons Per Occupied Room 2157 0.67 0.61

Source: U.S. Census Bureau, 1990 Census of Population and Housing, Summary File STF 3 AFile.

Injury Surveillance System of the New York independent variables City Department of Health and Mental Independent variables operationalize the Hygiene. The Injury Surveillance System model of neighborhood risk or suscepti- collates information from the New York City bility described earlier. Data reflecting the Department of Health (DOH) to form a structural and demographic composition of database of injury and fatality locations for neighborhoods and communities are pre- cases involving violence victims. All weapon dictors in the contagion models. Following assault injuries, fatal and nonfatal, are gen- Land et al. (1990), we selected 19 tract- erated through this archive. The data come level variables from the 1990 Census (STF3A from two public health data sources: Vital and 3B files) to characterize social areas. Statistics (mortality records) and the New Means and variances for the 19 initial vari- York State hospital admissions database, ables are presented in Table 36.3. Principal SPARCS, for hospitalized assault injuries. components analysis was used to eliminate The latter data were compiled from the autocorrelation among the 14 variables and hospital patient discharge summaries, with identify 3 orthogonal and conceptually dis- ICD-9 E-Codes (injury codes) for the sup- tinct factors. Table 36.4 shows that neighbor- plementary classification of external causes hoods and communities are characterized of injury. The data are hierarchically orga- along three dimensions: deprivation, pop- nized to avoid duplications for persons who ulation characteristics, and social control. were initially hospitalized but then died.9 Because communities are actually compos- Unfortunately, nonlethal injury data were ites of census tracts, their factor scores are unavailable until 1990, and so the time series actually weighted factor score composites. begins at the point when the trends already Estimates were weighted by the 1990 tract had begun their decline. population. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

706 jeffrey fagan, deanna l. wilkinson, and garth davies

Table 36.4: Factor composition, New York City census tracts, 1990

Rotated Factor % Variance Factor Score Eigenvalue Explained

Poverty/Inequality 2.54 84.73 % Households Under Poverty Level 0.96 % Households with Public Assistance Income 0.93 Gini for Total Household Income 0.87 Labor Market/Human Capital 2.60 65.03 % High-School Graduates – Total – 25+ 0.91 % Managerial, Professional, or Technical Jobs 0.84 Employment Rate 0.74 Labor Force Participation Rate 0.72

Segregation 1.31 65.27 Racial Fragmentation Index 0.81 % Non-White 0.81 Social Control #1 – Supervision 1.97 65.56 %Youth Population (5–15) 0.92 % Female Headed Households w/Children <18 0.80 Supervision Ratio (25–64 by 5–24) −0.70 Social Control #2 – Anonymity 1.02 50.78 Population – 1990 0.71 Residential Mobility – Same House as 1985 0.71 Immigration 1.52 76.25 Linguistic Isolation 0.87 Foreign Born 0.87 Housing Structure 1.24 41.48 %Occupied Units That Are Rentals 0.76 Density – Persons Per Occupied Room 0.55 Vacancy Rate 0.60

contagion effects munity adolescent homicide rate from 1992. The “contagion effect” of adolescent homi- However, models with 2-year time lags pro- cide is not expected to be immediate. Sim- duced results very similar to those reported ilar to the concept of incubation, it is more here. Thus the results as reported do not reasonable to assume that some period of appear to be an artifact of the lag time time must elapse between the occurrence of chosen. an adolescent homicide (threshold effect) or change in the adolescent homicide rate (pos- Results itive co-variation) and the realization of a related occurrence or rate change. For this On a preliminary note, we remind read- study, the time elapsed is estimated to be ers that the critical parameter in this 1 year. For example, with the outward con- class of models is the interaction of the tagion model, the adolescent homicide rate time by contagion measures. The param- in a neighborhood for 1990 is used to pre- eter estimate for the interaction indicates dict the rate in the surrounding community whether the rate of change in the parame- for 1991. Conversely, the adolescent rate for ter is a signiﬁcant predictor of the rate of a neighborhood in 1993 under the inward change of the dependent variable. The main contagion model is estimated using the com- effects can be interpreted as explanations for P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 707

differences in the average rates over the homicides in the surrounded tracts in T1. entire duration of the panel. The results in For the other models in Panel B, the effects Tables 36.5 and 36.6 show parameter esti- also are significant and are larger: an increase mates for the factors of interest in the con- of 1.0% for gun homicides for an increment tagion story, including spatial autocorrela- of one homicide in the surrounding tracts, tion. All models are controlled statistically and 2.0 for gun homicide victimizations of for the neighborhood susceptibility factors African Americans. discussed earlier. We use a quadratic form of The results for contagion of injury assault the time parameter to reflect the nonlinear suggest that there are contagion effects. All distribution of the homicide and injury rates the models in both panels of Table 36.6 over time. are significant, but the effects seem at first There is evidence of both inward and glance to be small. The exponentiated coef- outward contagion for homicide and injury ficients suggest that assaults increase by less assault, but the patterns vary by type of vio- than 1% for each increment in assaults in the lence, means of violence, and race of the vic- surrounding tracts for outward contagion or tims. For total homicide, including gun and in the surrounded tracts for inward conta- nongun homicides, Table 36.5 Panel A shows gion. Contagion is evident both for gun vio- results of three outward contagion mod- lence and nongun violence in these models, els. Neither total homicides nor gun homi- suggesting a more ambiguous role of guns. cides are statistically significant. The param- Although there may be a cross-over conta- eter estimate for the interaction of time by gion effect of gun violence to nongun vio- homicide in the surrounding tracts is sig- lence in the surrounding areas, it is unlikely nificant only in the model for gun homi- conceptually and empirically to be any larger cide victimization of African Americans, and than these observed effects for within-type the direction is negative. The interpretation contagion. for the coefficient is somewhat counterin- Several features of this analysis merit fur- tuitive – it does not imply that there are ther note. First, the coefficients for out- fewer homicides in the surrounding tracts ward homicide contagion parameters are rel- over time. Instead, as the overall homicide atively small and significant only for the trend declines, a negative coefficient suggests model for African American gun homi- that this factor is opposing that trend in the cide victimization. This is not surprising, dependent variable. In this case, then, the given that homicides are comparatively rare negative coefficient for the time by homi- events. But the primary tale in Table 36.5 cide interaction indicates that the rate of is one of inward contagion. Consider first homicide victimization of African Ameri- the sheer magnitude of the estimates for cans in the surrounding neighborhoods is spatial lag in the outward models, and sec- increasing, controlling for the rates in the ond the large and significant coefficients surrounded neighborhood in the previous for the inward contagion parameters. Taken year. The exponentiated coefficient is .989, together, these findings confirm that the suggesting that an increase of one homicide inward contagion of homicide is more potent in a neighborhood in T0 predicts an increase than outward contagion, that a neighbor- of 1.1% in the number of homicides in any of hood is affected more by its adjacent com- the surrounding areas. munity than it affects that same community. Panel B in Table 36.5 presents consis- Though not as dramatic, the same pat- tent evidence of homicide contagion. The tern is evident in Table 36.6. Here, there is estimates for time by contagion are signifi- evidence of both outward and inward con- cant and negative in all three models. The tagion of nonlethal assaults. The coefficients exponentiated coefficient for total homicide for the outward contagion of assault are not is .992, suggesting that an increase of one large, but they are consistently significant. homicide in the surrounding tracts in T0 pre- In relative terms, however, the coefficients dicts an increase of 0.8% in the number of for the inward contagion assault model again P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3 c c c c c c c c c c c c 2000 adults to 11 – 14 32 18 44 30 34 35 73 . 66 77 28 ...... 9 2 5 13 63 58 − − − − 1985 − − 113 114 157 100 918 984 081 10 189 11 980 167 20 041 4 989 . . 030 35 ...... 011 016 173 1 155 1 176 169 020 078 1 030 1 040 1 085 425 4 ...... 2 1 2 − − − − − − c c. c. c c c c. c c. a. c ns 92 02 22 20 97 88 94 00 56 30 88 99 ...... 1 8 19 77 26 − − − − − 110 611 219 117 012 24 014 23 990 160 28 990 906 207 15 . . 069 25 008 2 ...... 012 1 014 1 1888 1 010 149 1 4919 002 066 1 204 169 3 008 1 098 ...... 1 2 − − − − − c. c. c. c c c. c c. c b ns ns . 61 27 52 35 47 84 01 56 88 54 80 73 ...... 3 − 70 20 − − − 175 931 992 7 102 23 942 127 26 168 14 999 429 119 010 22 052 26 003 1 ...... years or less, percent population not born in U.S., percent of households in public housing, and total population. 5 not significant. 003 1 112 1 012 1 156 1 010 1 072 008 050 1 069 0009 888 2 743 . . . . . = 1 − − − − − ;ns 001 . 0 ;c< 01 . 0 ;b< : Poisson regression results for models of inward and outward contagion of homicide, New York City census tracts, 05 5 . . 0 36 B. Inward Contagion Total Homicides Gun Homicides African American Gun Homicides A. Outward Contagion Total Homicides Gun Homicides African American Gun Homicides . All model results were adjusted for effects of tract social and economic factors: percent of persons below poverty, percent in labor market, ratio of . All models were estimated with time as quadratic (nonlinear) predictor. . Outward contagion models include adjustment for spatial autocorrelation of homicides in the surrounding census tracts. . All models were estimated using mixed effects Poisson regression models with autoregressive co-variance structures. Time *Homicide Contagion % African American Population . 1 2 3 juveniles, percent population living in tract 4 Homicide Contagion . Time . Table (coefficient, t statistic, statistical significance of predictor, exponentiated coefficient) PredictorIntercept BPredictor Exp (B)Intercept t p(t)p(t): a < B B Exp (B) Exp (B) t t p(t) p(t) B B Exp (B) Exp (B) t t p(t) p(t) B Exp (B) t p(t) Time *Homicide Contagion Homicide Contagion . Time . Homicide Spatial Lag . % African American Population . Time *Homicide Spatial Lag

708 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3 c c a c c c c c c c c ns adults to 2000 81 19 18 99 98 43 65 70 70 33 87 89 – ...... 2 7 9 12 31 58 − − − − − − 1990 071 992 021 10 999 978 998 032 34 025 30 354 24 004 005 26 713 ...... 021 1 001 031 1 023 0002 026 1 303 1 338 008 004 1 005 1 64 ...... 2 − − − − − − c c c. c c. c c c. c. c. c c. 81 06 88 04 43 91 17 57 33 47 41 27 ...... 9 3 18 46 36 − − − − − 992 328 012 21 999 979 016 24 998 356 37 328 33 020 13 008 2 004 31 ...... 021 012 1 001 113 016 1 283 1 020 1 0002 008 1 008 086 2 004 1 . . . . . 1 − − − − − c c. c c. c c. c c. c. c. c. c 13 59 45 94 31 64 87 45 66 58 37 99 ...... 2 3 11 42 − − − − 263 120 999 996 993 001 44 01 21 999 105 67 002 3 659 22 005 15 008 17 ...... years or less, percent population not born in U.S., percent of households in public housing, and total population. 5 011 1 0001 001 1 0001 100 1 226 9 0002 1 506 1 005 1 004 008 1 007 ...... 2 − − − − . 001 . 0 ;c< 01 . 0 ;b< : Poisson regression results for models of inward and outward contagion of assault, New York City census tracts, 05 6 . . 0 36 B. Inward Contagion Total Assaults Gun Assaults African American Gun Assaults A. Outward Contagion Total Assaults Gun Assaults African American Gun Assaults . All model results were adjusted for effects of tract social and economic factors: percent of persons below poverty, percent in labor market, ratio of . All models were estimated with time as quadratic (nonlinear) predictor. . Outward contagion models include adjustment for spatial autocorrelation of homicides in the surrounding census tracts. . All models were estimated using mixed effects Poisson regression models with autoregressive co-variance structures. (coefficient, t statistic, statistical significance of predictor, exponentiated coefficient) juveniles, percent population living in tract Table PredictorIntercept B Exp (B)PredictorIntercept t p(t) Bp(t): a < B Exp (B) . Exp (B) t t p(t) p(t) B B Exp (B) Exp (B) t t p(t) p(t) B Exp (B) t p(t) 1 2 3 Assault Contagion . Time *Assault Spatial Lag . Time Assault Contagion% African American Population . 4 Time % African American Population Time *Assault Contagion Time *Assault Contagion Assault Spatial Lag .

709 P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

710 jeffrey fagan, deanna l. wilkinson, and garth davies

indicate that neighborhoods exert less of an gion. We identify dynamic social processes influence over their surrounding commu- that fuel the social contagion of youth vio- nities and instead are more susceptible to lence. At the heart of this process are the events in those communities. interactions of individuals within and across The presence of gun contagion of both social networks. Violence plays a central role gun homicides and gun assaults underscores in the maintenance of organizational bound- the importance of guns in the dynamics of aries, norms, and cohesion. Two elements in social contagion at the population level. In the contagion of youth violence are conflict an earlier article, Fagan and Davies (2004) between networks of youths and the role of showed that the neighborhoods with the violence in resolving conflicts. The violent highest homicide, injury, and gun violence events in which these processes unfold and rates are New York City’s poorest neigh- change over time represent opportunities to borhoods. Those finding were confirmed in build or maintain status within networks; these models in the significant contribution in some events, violence is an imperative of the co-variates that express the suscepti- with costs when it is not invoked (Wilkinson, bility of the poorest neighborhoods to conta- 2003). Third parties are especially important gion (data available from authors). Accord- in the spread of violence between networks; ingly, the corollary finding of socioeconomic third parties can animate or intensify vio- risk as a contributor to the spread of violence once a conflict begins, or they can help lence captures both the significance of sus- mediate and suppress it. They also convey ceptibility and the importance of structural the outcomes of violent events to others in equivalence in shaping the trajectory of dif- the larger social worlds that surround these fusion. That is, the adoption at the popu- social networks, helping sustain norms and lation level of gun violence as a means of provide context for the next conflicts that social control and exchange was facilitated may arise. The strategic role of guns in these by the social concentration of poverty and processes intensifies the dynamics that fuel of the close social synapses intrinsic to poor the epidemic. In some instances, the pres- neighborhoods. Accordingly, the social and ence of guns in events links together persons spatial clustering of homicide suggests that and events across time and sustains the pro- it is concentrated within overlapping social cesses of social contagion. networks in small areas (Fagan & Wilkinson, This section unfolds in three parts. First, 1998b). we discuss the mechanisms through which Social contagion theory suggests that indi- violence may spread between social groups viduals are likely to mutually influence the or networks and show how these mecha- behaviors of others with whom they are in nisms are best understood through the anal- frequent and redundant contact (Bovasso, ysis of specific events. Next, the chapter 1996,p.1421; Burt, 1987). The social interac- focuses on the importance of third parties tions underlying assaultive violence suggest in violent events. The third section presents its spread by social contact (Loftin, 1986), three scenarios that illustrate the intersec- and, as we show below and in other arti- tion of these themes that produce violent cles, by specific forms of social interaction events and often set the stage for future ones. (Wilkinson, 2003). We explore these themes next. Violent Events For any attitude, expectation, behavior, or Violent Events, Social Networks, and virus to be “spread” from person to per- Social Contagion son or group to group, interpersonal contact is typically required. Exposure can be We turn next to an analysis of the individual- direct or indirect. Direct exposure to gun and group-level processes of social conta- use and violent behaviors among similarly P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 711

situated networks of young men would likely Oliver examined violent behaviors to iden- increase the risk of transmission. Observa- tify the “rules of engagement” and situations of young men’s decision-making pro- tional causes of violence in the bar setting. cesses in violence use and avoidance pro- He observed a five-stage sequence of events vide a window into the processes that shape similar to Felson and Steadman’s previous interpersonal transmission. Studying vio- classification. His work added insights about lence from an event perspective combines violent events from a sample of African the study of offenders, victims, and social American men, especially with regard to context to yield a more complete picture understanding event closure and the after- of its etiology (Meier, Kennedy, & Sacco, math of violent events. In all of these studies, 2001). The event perspective considers the victim actions, including retaliation, denial co-production by victim(s), offender(s), and of claims, and aggressiveness, were found to others of a violence experience. It empha- be important factors. sizes event precursors; the event as it A focus on violent events demonstrates unfolds; and the aftermath, including report- that most violence is a process of social inter- ing, harm, and redress. The event per- actions with identifiable rules and contin- spective integrates concepts from symbolic gencies (Campbell, 1986; Felson, 1982; Fel- and situational interactionism, routine activ- son & Steadman, 1983; Luckenbill, 1977; ity, and rational choice theories. The social Luckenbill & Doyle, 1989; Oliver, 1994; Polk, geometry of violent conflict provides clues to 1994; Sommers & Baskin, 1993; Wilkinson, understanding what distinguishes one con- 2003). Contrary to common wisdom, vio- flict situation from another or more pre- lent acts can be understood as rational or cisely what distinguishes a nonviolent con- purposive behavior. Most experts agree that flict from a violent conflict (Phillips & rationality is “bounded”; that is, individuals Cooney, 2005). rarely have all of the information necessary Several investigators have found evidence to make a truly “rational” decision. The like- that the interplay between the primary lihood of violence reflects the progression of actors determines, in part, the outcome. For decisions across a series of identifiable stages. example, Felson and Steadman (1983) found Much of this research concludes that there that violent incidents usually began with are contingencies in each stage, shaped by identity attacks, were followed by attempts external influences and social interactions and failures to influence the opponent, then of the actors. Yet the data have generally included verbal threats, and finally, ended in not been available to answer more useful physical attack. In a study of ex-offenders, questions, such as what the contingencies ex-mental patients, and a sample drawn from are, how actors take them into account, the general population, Felson (1982, 1984) and how they vary as an event progresses found a similar pattern. Hughes and Short through these stages. Although prior stud- (2005) confirmed Felson’s earlier findings ies provide generalized classifications of vio- with a sample of gang-involved youth. lent event stages, a finer-tuned assessment of Similarly, Oliver (1994) used detailed nar- the actions and reactions of actors in violent ratives of violent confrontations between encounters than is possible from detailed Black males in bars and bar settings. Oliver event narratives will shed new light on the employed both participant observation and micro-decisions and contextual influences interview methods over a 5-year period across a range of types of violent encoun- (1983–1987) to “systematically examine the ters. Previous studies have generally ignored social functions of the black bar and how information about precipitating actions, as black males interacted with each other and well as the aftermath of violent events. The with females in this setting.” The sample analysis of event stages must also take into consisted of 41 Black men 28 to 45 years account the heterogeneity of violent events old who frequented the research locations. by examining a wide range of violent acts. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

712 jeffrey fagan, deanna l. wilkinson, and garth davies

Black (1993,p.108), “third parties intervene Studies of Third Parties in Violent Events without taking sides.” Third parties witness or somehow become As Black explains, the role of third parties involved in an estimated two thirds of often depends on personal allegiance (or lack acts of interpersonal violence in the United of it) to the main actors. He argues that audi- States (Planty, 2002). The percentage is even ence members allied with either the pro- greater (approximately 73%) for violence tagonist or the antagonist may contribute among young people. Despite the common to the escalation or de-escalation of a dis- nature of third-party presence, researchers pute through verbal statements, body lan- know little about the specific contributions guage, cheering, nonverbal social pressure, that third parties make in promoting or pre- or physical acts of violence. Partisanship and venting the escalation of interpersonal con- solidarity are key features of Black’s thesis. flict to violence. Previous research concludes Cooney (1998) elaborated Black’s theory to that bystanders and third parties contribute include variables on group membership sta- significantly to the outcome of violent tus and articulated hypotheses for four con- encounters (see Black, 1993; Cooney, 1998; figurations of third-party social locations in Decker, 1995; Felson, 1982; Felson, Ribner, & determining their influence over the princi- Siegel, 1984; Oliver, 1994; Phillips & Cooney, pals in a conflict situation. They specified the 2005; Wilkinson, 2003). For example, Fel- predictive power of third parties with close son (1982, 1993) found that, when a dis- and distant ties to individuals, close and dispute occurred between parties of the same tant ties to groups, cross-cutting ties, and no sex, the presence of third parties increased ties. Using interviews with 100 incarcerated the likelihood that a verbal disagreement offenders of assault or homicide, Phillips and would turn into a physical fight. Third par- Cooney (2005) found moderate support for ties may be viewed both as part of the socio- these hypotheses in the first empirical test to cultural context and as participants in the date. co-production of violent events. Largely due Wilkinson (2007) examined the role of to the work of Donald Black and his fol- network peers and third parties as potential lowers, theory in the area of third parties agents of social control in 237 violent events has evolved while empirical studies of third- reported by 159 urban youth. The study clas- party roles in violence remain rare and unfo- sified third parties by their relational ties to cused within criminology. the focal respondent and his opponent(s). Black’s (1993) theoretical work on the The study showed that third parties who social structure of conflict includes a typol- were closely tied to the primary participants ogy of third parties with specification across were mostly likely to join in the violence, two domains: the nature and degree of the rather than doing anything to stop it (55% intervention. He identified 12 third-party of the respondent’s associates actively used roles, “including five support roles (informer, violence, whereas 42% of the opponent’s adviser, advocate, ally, and surrogate) and associates did). Bystanders who were neu- five settlement roles (friendly peacemaker, tral parties toward either side rarely became mediator, arbitrator, judge, and repressive involved in the violence itself, although they peacemaker).” Two other roles that do not actively engaged in some type of social con- fit within either category are the “negotia- trol action in about 20% of events. Bystander tor” whose partisanship cross-cuts both sides actions typically involved yelling to try to and the “healer.” The types are rank ordered stop the violence (8%), actions to break up in terms of degree of intervention, with sup- the conflict (10%), and coming to the aid porting roles organized by the extent of par- of participants nonviolently (3%). From the tisanship and settlement roles by the author- respondents’ perspective, bystanders very itative status of the third parties. Settlement rarely called the police. Third-party pres- roles come into play when, according to ence seems to coincide with police becoming P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 713

involved in both serious and nonserious scenario 1: street corner store and events. Events that occurred at night and residential neighborhood without neutral bystanders present were less In this scenario, let’s call the respondent likely to come to the attention to police (at (Aron) and the opponent (Bruce). Aron least from the respondents’ experience). argues with Bruce at a neighborhood corner store over cutting in line. They step outside and begin to fight. None of the witnesses in Three Scenarios the setting were closely tied to either party, Moving beyond these descriptive findings, in but they did know both youths by face and the remainder of this section we examine reputation. After about 5 minutes of fight- how third-party involvement relates to social ing with fists, Bruce pulls out a razor on contagion in 782 violent events. An event- Aron and uses it to slice him across his arm level analysis is the best way to examine how that Aron had extended to protect his face. violence “flows” as a process from individual Someone in the audience yells at the youth to individual, as well as from group to group. to stop before the police are called. Both By focusing specifically on the dynamics of youth flee. Aron goes back to his block and gun violent events reported by 418 New York recounts the story to his associates. He rallies City male youth aged 16 to 24, we iden- their support for a counterattack by high- tify dynamic social processes that fuel the lighting the ways that his opponent was try- social contagion of youth violence. Peer net- ing to destroy his attractiveness by scarring work involvement in violent acts takes sev- his face and how he disrespected him. After eral forms. First, and most common, is co- a few days pass and the group was fueled participation or co-offending. The decision by visions of revenge, Aron and four of his to co-participate happens at any stage as associates armed themselves with handguns violent conflicts unfold. These processes are and went to Bruce’s block. Aron’s group evident in analyses of the action by actor finds Bruce, two guys, and one girl sitting sequences of violent events. Peer network on the front steps of a neighborhood build- members become actively involved in con- ing. Without verbal warning, Aron pulls his flicts that lead to violence through several 9 mm from his waist and starts shooting in avenues: (1) Their involvement in the vio- the direction of Bruce’s group. Bruce was lent event is strategic and anticipated from caught off guard as he was not expecting the outset; (2) they come to the aid of an conflict that day. One of Bruce’s companions associate who is losing in the battle; (3) pulls out his gun and returns fire. After a few they are threatened/offended/disrespected brief minutes the shooting stops. Two people at some point during the course of observ- are shot–one relatively minor wound to the ing a dispute unfold; (4) they use violence leg on Aron’s side and one serious injury to either in the moment or after the fact to Bruce’s friend. According to the respondent, obtain justice or right some wrong that was the beef or conflict remains ongoing. Aron perpetrated against a group member; or (5) is anticipating a retaliatory attack to avenge they are influenced by gossip about the per- the injury to Bruce’s friend and because he formance and reputation of violence partic- escalated the beef to a “life and death” issue. ipants and take up conflicts to restore the reputation of group members. Peer network scenario 2: a club at 2 a.m. members who are present during disputes Here, Rich, Mike, and four of his associates that escalate into violence play different go to a club to party and socialize with roles depending on the relationship among females. In the club, Mike sees a girl named the combatants, weapon type, and injury Becky with whom he has a causal rela- outcomes. Three scenarios of violence are tionship. She came to the club with six of presented that illustrate the nuances of how her girlfriends to dance and have fun. Rich violence is diffused across peer networks. sees Becky dancing and joins her. Rich rubs P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

714 jeffrey fagan, deanna l. wilkinson, and garth davies

his body up against Becky. Mike observes his need for revenge on Rich. Mike’s boy the violation directly. Mike informs Rich suffered serious damage to his knee, which of his wrongdoing and asks for an account. angered Mike as well. Rumors of revenge for Rich denies wrongdoing, states claim to the the death of Rich’s associate were spreading girl, and places blame on Mike. Mike pulls around. Both sides were on guard and look- the girl away from Rich and returns to his ing for strategic advantage for the next vio- group. Mike’s boys comment on the viola- lent event. Mike anticipated that more vio- tion. Mike watches Rich, thinks about the lence would follow from this event. violation, and is angry. Mike’s boys report back information about Rich and his boys. scenario 3: a drug stash house near Mike returns to his prior activity. Mike’s girl a street drug spot goes back to dancing. Pete and his two associates planned a rob- The girl’s friends compliment her on bery of a drug stash house manned by a being desired by two guys. Rich talks with Dominican crew. Pete got information about his buddies about the girl and Mike’s capa- the best day and time to rob the house, bilities. Both parties wait to see what the what types of weapons would be used to other will do. After some time passes, Rich defend the stash house, and so on. How- begins dancing provocatively with Mike’s ever, his information was incomplete. When ‘girl’ again. Mike is not watching, but hears he and his associates made their armed rob- about it from his man. Rich’s friends watch bery attempt, they were confronted by addi- to see if Rich gets the girl. Mike walks up to tional armed drug dealers who were protect- Rich and punches him in the face. Rich hits ing the drug stash. Pete’s group exchanged Mike back. Both sets of friends watch ini- fire with the drug dealers as they fled the tially. Rich lands some good punches. Mike’s building following a failed robbery attempt. first friend jumps on Rich. Club security Pete’s friend Franky almost got shot. Franky breaks up the fight, issuing warnings. The was recognized by the Dominican drug crew. youths go back to drinking and partying. They came to Pete’s neighborhood to find Mike discusses ways of punishing Rich. and shoot him as revenge for the attempted Both sides watch the other. The status of robbery incident. Pete describes the situa- who “gets” the girl remains open. Both sides tion to our interviewer: plan to attack at the end of the night. Mike believed that Rich must have called some (Pete): They recognized him and shit. of his friends for additional reinforcements So it was some Dominican kids. They and to make sure that when Rich got out- went to our block and we seen like side he would have a gun available. Mike and this blue Lincoln Towne or Escort or his boys essentially make the same type of some shit. We seen this coming around preparations. As soon as Mike moved toward and coming around. And we was like, exiting the club, Rich’s group was preparing ‘yo Franky, I was like you know them for a gun battle. Mike recalls that his side had niggas right there man?’ He was like, three guns that they retrieved from nearby ‘nah, I was like the niggas got some- stashes, whereas it seemed like the other side thing with us, either they scheming at had five or more guns. With more than 20 us or the niggas ready to hit somebody shots fired, injuries were sustained on both else.’ Niggas is sitting up in there either sides. The police came to the scene, but no waiting for somebody or waiting for one was arrested. us to make a move. He gave a peek The injured were transported to the hos- and as soon as he looked, Franky told pital, and one youth from Rich’s side died me, ‘yo Pete that’s one of them nigga at the hospital as a result of this gun event. from that stash house. My hands got Mike heard rumors that Rich had been seen real sweaty, we didn’t have our ghats with Becky following the shoot-out, which on us, My hands got real sweaty. And fueled his anger and, in his mind, justified it’s like I could front it off and I call you P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 715

from across the way, hey ‘yo what’s up what happen why they niggas shot up come here.’ You know what I’m saying Franky and shit. We didn’t want to tell by that time, Franky tells me yo Pete nobody that we went to stash house to just run your way and I ran inside the hit. It was like, ‘nah we had beef with building. And I said alright I’m going these kids and they came around and to run to this Alicia’s house right there shit like.’ where the gate is at. You know what (Interviewer): Why didn’t you want to I’m saying the car all of the sudden just tell? right down the block see but we didn’t (Pete): Because if he tell niggas that we want to run as soon as that shit come. hitting other peoples stash houses they You know what I’m saying as soon as going to probably. God forbid they hit that car come down slowly, it didn’t do up the stash house in our block they nothing. The Dominican kid didn’t do the first thing they going to say is ‘yo nothing, he just looked at us. And went I think it was Pete and Franky and Zee right around the block. We was like ‘cause niggas like hitting stash houses. that ain’t them, that ain’t them. Give Everybody in my block think shysty- it like two to three minutes they came ness, so the first nigga they will prob- walking, walking, it was like a good lit- ably will is to us and they will proba- tle 50 yards. And I said, ‘yo, Franky bly try to smoke us and it ain’t us. You that’s them right there.’ He was like, know what I’m saying. ‘where I don’t see the car.’ I said, ‘nah they on foot right there.’ And Franky The event process can be dissected into said, ‘oh shit, that’s them.’ When me specific stages: anticipatory stage, opening and Franky look at them they looked moves, countermoves and brewing period, at us and I see them real quick running persistence stage, intensification stage, early to pull out they ghat. And . . . .they just violence stage, stewing period, assessment started busting. Blah, blah, blah and I stage, the casting/recasting stage, and the caught my reaction I ran where I said I retaliatory stage. The examples above was going to run. Franky ran in there demonstrate that network peers play impor- but niggas was just kept blah, blah, tant roles at almost every stage of a conflict blah and they went after Franky more that escalates into violence. The commu- than me. They kept just shooting blah, nication of normative expectations, vio- blah, blah and they ran in the build- lence scripts, and violence strategies filters ing. That was like oh shit, I jumped through direct observation, word of mouth over the gate and I was oh shit. I didn’t via rumors, and telling of “war stories.” have my ghat or nothing. I was damn Franky, Franky. Interviewer: What happened? Social Contagion and Social Norms (Pete): Really they didn’t caught Franky, You know what I’m saying they didn’t The dynamics of social contagion can be get him. But the next day they caught, accommodated within concepts of social they shot him up. influence and social norms. The social influ- (Interviewer): So they came back the ence concept of behavior borrows from next day? both economics and sociology (Harcourt, (Pete): Early in the morning cause Franky 1998; Lessig, 1995). Its economic compo- was out there pitching, early in the nent suggests that people will act to max- morning. They came back the next day imize their utility, whereas its sociological and they were shooting at him and he dimension suggests that conduct is shaped didn’t feel it while he was running but through direct and vicarious social interac- it cause the bullet went in and out. I tions. A simple version of this nexus suggests was like you know, everybody was yo that the choice of conduct is influenced by P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

716 jeffrey fagan, deanna l. wilkinson, and garth davies

observation and practice of the most effec- Recent applications of social influence tive options. Choices are contextualized as models to crime control emphasize the sem- well, reflecting both the range of available inal role of the exogenous influence of “disor- options and the specific contingencies in der,” in which minor crimes signal to would- which they are applied (Fagan, 1999). be criminals that crimes in that area will In the contagious dynamics of violence, be tolerated and not reported (Kelling & the social meaning of violence is constructed Cole, 1996; Wilson & Kelling, 1982). At through the interrelationship of its action first glance, “Broken Windows” suggests that and its context. The social meaning in this there is a spread of norms supporting crime case involves actions (violence) that have that overwhelm norms of orderliness. The both returns (identity, status, avoidance of spread comes from the continuing signals attack) and expectations that, within tightly from disorder. Withdrawal of the signs of dis- packed networks, are unquestioned or nor- order will change social norms by allowing mative. Conduct impregnated with social the social influence of orderliness to flour- meaning has influence on the behaviors of ish. Apart from the problematic nature of others in immediate proximity. The social this dichotomous categorization of persons meaning of violence influences the adapta- (Harcourt, 1998), Broken Windows medi- tion of behavioral norms, expected responses calizes the conditions of disorder and crimi- (scripts), and even beliefs (memes) about nality. It assumes that exposure to the disor- systems of behavior. Social norms are the der is a constant and recurring process that product of repeated events that demonstrate signals to the motivated offender that crime the meaning and utility of specific forms of can succeed. Removing the signs of disor- conduct. Social influence thus has a dynamic der will change social norms by allowing the and reciprocal effect on social norms (Har- social influence of orderliness to flourish. But court, 1998; Lessig, 1995). In poor neigh- this theory is limited by focusing only on the borhoods, social interactions are dominated introduction of the original cues or sources by street codes, or local systems of justice, of crime and relying on the causal effects of that reward displays of physical domina- these exogenous factors. This is analogous tion and offer social approval for antisocial to the food poisoning model of epidemics. behavior. Moreover, a literal reading of Broken Win- The endogeneity of social contagion to dows theory would invite problematic legal networks and neighborhoods illustrates the policy responses, such as “social quarantine,” differences in the two types of epidemics. which have limited efficacy and raise moral The origins of a contagious epidemic that quandaries (Markovits, 2005). travels through a population become dis- The dynamics of social contagion instead tal influences on the pathway and dynam- suggest an endogenous process, in which the ics of transmission through populations over spread of social norms occurs through the time. The setting or context of contagion everyday interactions of individuals within reflects the susceptibility of populations to networks that are structurally equivalent the transmission of a socially meaningful and closely packed. Here, the ill grows and behavior, as well as its exposure to the behav- spreads from the inside, often long after the ior that has acquired meaning (Fullilove origins have subsided. This is analogous to et al., 1998). This can be true both for fash- influenza contagion or to the spread of cul- ion and art (Gladwell, 1997; Servin, 1999) tural or political thought (Cavalli-Sproza & and for problematic social behaviors, such Feldman, 1981). as drug use (Rowe & Rodgers, 1991), teenage The concept of contagion neutralizes sexual activity (Rodgers & Rowe, 1993), the categorizations of disorder and order teenage pregnancy (Crane, 1991), child mal- that theoretically inform the new path of treatment (Coultin, Korbin, Su, & Chow, deterrence. A literal translation of conta- 1995), and violence (Anderson, 1999; Cork, gion would emphasize guns as a recurring 1999; Fagan, 1999; Loftin, 1986). source of violence and as an agent in the P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 717

transmission of violence norms. Because the tance. Thanks also to interviewers Richard recent epidemic cycle of violence was in McClain, David Tufino, Davon Battee, Jason reality a gun homicide epidemic, the case Mercado, and Whetsel Wade for creating a for gun-oriented policing strategies (Fagan, unique and rich source of data on street cor- 2002; Fagan, et al., 1998) is much stronger ner life in New York City. than practices based on the more diffuse and unsupported theory of disorder control and order-maintenance strategies. Although dis- Notes order opposes orderliness, cleanliness, and 1 2004 sobriety (Harcourt, 2001), violence appears . Cook and Ludwig ( ) developed estimates of the effects of gun availability, con- to travel on vectors quite unrelated to that trolling for any effects due to reverse causa- particular set of social norms. tion in which the demand for guns among teenagers may affect prevalence. 2. One important development is a breakdown Acknowledgments in the age grading of behaviors, in which the traditional segmentation of younger adoles- An earlier version of this paper was pre- cents from older ones, and behavioral tran- sented at the Urban Seminar Series on Chil- sitions from one developmental stage to the dren’s Health and Safety, John F. Kennedy next, are short-circuited by the strategic pres- School of Government, Harvard University, ence of weapons. Mixed-age interactions play in May 2000. Members of the Kennedy an important role in this process. Older adolescents and young adults provide modeling School Urban Seminar Series, Columbia influences as well as more direct effects. We Law School, and the Columbia Center for found that they exert downward pressure on Youth Violence Prevention provided valu- others their own age and younger through able comments on earlier versions of the identity challenges that, in part, shape the chapter. This research was supported by social identities for both parties. At younger grants from the Centers for Disease Con- ages, boys are pushing upward for status by trol, the National Institute of Justice, and challenging boys a few years older. the Robert Wood Johnson Foundation, in 3. With the exception of the decade influenced addition to support from the Sloan Working by both the passage of the Volstead Act and Group on Social Contagion of Youth Vio- the social and economic instability of the lence. All opinions and errors are those of the Great Depression, homicide rates in New York City varied little. In its 1996 report, the authors. We thank Assistant Commissioner Office of Vital Statistics and Epidemiology Susan A. Wilt, former director of the Injury reports homicide rates prior to 1985 in 5-year Prevention Program of the New York City intervals. Homicide rates rose from an aver- Department of Health and Mental Hygiene, age of 4.9 in 1916–1920 to 7.6 in 1931–1935, for generously sharing data on homicide and and declined to 4.5 by 1936–1940 (NYC injury assault characteristics and locations. Department of Health and Mental Hygiene, Frank Zimring provided excellent comments 1997). during the course of the project. Workshop 4. The 1990 spike for male nongun homicides participants at the Sloan Working Group on most likely reflects the 90 arson homicide Contagion on Social Contagion, Columbia deaths in the Happyland Social Club fire. We Law School, the Mailman School of Public could not adjust the age-, race-, or gender- specific rates for these homicide deaths since Health at Columbia University, the Urban data were not available on their character- Health Seminar at the John F. Kennedy istics. The nongun total for 1990 has been School at Harvard University, and the For- adjusted by deleting 89 of the 90 killings tunoff Colloquium at the New York Univer- from the Happyland Social Club fire, in sity School of Law all provided helpful com- effect counting that episode as one homicide. ments on earlier drafts. Jay Galluzzo, Tamara 5. Exceptions include domestic homicides, and Dumanovsky, Marlene Pantin, and Carolyn homicides that follow rape. See, Dugan, Pinedo provided excellent research assis- Nagin, and Rosenfeld (1999). P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

718 jeffrey fagan, deanna l. wilkinson, and garth davies

6. Other indicators, such as drug use among References arrestees recorded in the Drug Use Fore- casting System (DUF), also show little rela- Abelson, R. P. (1976). Script processing in atti- tionship with trends in gun homicide rates. tude formation and decision-making. In J. S. Fagan et al. (1998) show that the incidence of Carroll & J. W. Payne (Eds.), Cognition and drug-positive arrestees remained unchanged social behavior. Hillsdale, NJ: Erlbaum. throughout the period, and was unrelated Abelson, R. P. (1981). Psychological status of the to both firearm and non-firearm homicide script concept. American Psychologist, 36, 715– trends. 729. 7. The male divorce rate also is a consistent Agar, M., & Reisinger, H. S. (2002). A heroin predictor of violence and homicide rates, epidemic at the intersection of histories: The and effects are greater for juveniles than 1960s epidemic among African Americans in for adults. For example, Messner and Samp- Baltimore. Medical Anthropology, 21(2), 115– son (1991) showed that Black family dis- 156. ruption was substantially related to rates of Anderson, E. (1994, May). The code of the murder and robbery involving Blacks. These streets. Atlantic Monthly, 81–94. findings are consistent with the consistent Anderson, E. (1999). Code of the street: Decency, findings in the delinquency literature on violence, and the moral life of the inner city. New the effects of broken homes on social con- York: W. W. Norton. trol and guardianship. The effects of male Bailey, N. T. (1967). Mathematical approach to divorce can be interpreted either as a conse- biology and medicine. New York: Wiley. quence and correlate of the rise of female- Bailey, N. T. (1975). The mathematical theory of headed households, or as an indicator of infectious diseases and its applications. London: weak social control of children who then Griffin. are raised primarily by women. Whatever Balkin, J. M. (1998). Cultural software: A theory of its meaning, the male divorce rate has pos- ideology. Danbury, CT: Yale University Press. itive, clear cut effects on robbery, assault, Baumer, E. (1994). Poverty, crack, and crime: A rape and homicide (Sampson and Lauritsen, cross-city analysis. Journal of Research in Crime 1994 ). and Delinquency, 31(3), 311–327. 8. A second advantage of the mixed models Baumer, E., Lauritsen, J. L., Rosenfeld, R., & approach is that it allows for greater flexi- Wright, R. (1998). The influence of crack bility in specifying the covariance structure cocaine on robbery, burglary, and homicide of the data. Specifically, mixed models allow rates: A cross-city, longitudinal analysis. Jour- for the analysis of data where the requi- nal of Research in Crime and Delinquency, site assumptions of Ordinary Least Squares 35(3), 316–340. regression concerning error term indepen- Berman, A. L. (1995). Suicide prevention: Clus- dence are violated. This is particularly impor- ters and contagion. In A. L. Berman (Ed.), Sui- tant in research involving aggregate units. cide prevention: Case consultations (pp. 25–55). Because the neighborhoods and communi- New York: Springer. ties are comprised of geographically contigu- Bingenheimer, J. B., Brennan, R. T., & Earls, ous census tracts, autocorrelation is inherent F. J. (2005). Firearm violence exposure and in the data structure and it would be inappro- serious violent behavior. Science, 308, 1323– priate to assume a simple covariance struc- 1326. ture for these analyses. All of the models Black, D. (1993). The social structure of right are instead analyzed with an autoregressive and wrong. New York: Cambridge University covariance structure. Press. 9. Additional records are available on deaths Blumstein, A. (1995). Youth violence, guns, and from other means (accidents, disease classi- the illicit-drug industry. Journal of Criminal fications, self-inflicted violence) to estimate Law and Criminology, 86(1), 10–36. overall mortality rates by area and by cause. Blumstein, A., & Beck, A. J. (1999). Popula- Accordingly, the database has the capacity tion growth in U.S. prisons, 1980–1996.InM. for spatial, temporal, and demographic disag- Tonry & J. Petersilia (Eds.), Prisons (Vol. 26, gregation and analysis of several dimensions pp. 17–62). Chicago: University of Chicago of mortality. Press. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 719

Bourgois, P. (1995). In search of respect: Selling and teenage childbearing. American Journal of crack in El Barrio. New York: Cambridge Uni- Sociology, 96(5), 1226–1259. versity Press. Curtis, R. (1998). The improbable transformation Bovasso, G. (1996). A network analysis of of inner-city neighborhoods: Crime, violence, social contagion processes in an organizational drugs, and youth in the 1990s. Journal of Crim- intervention. Human Relations, 49(11), 1419– inal Law and Criminology, 88, 1233–1276. 1435. Decker, S. H. (1995). Reconstructing homicide Brotherton, D., & Barrios, L. (2004). The almighty events: The role of witnesses in fatal encoun- Latin king and queen nation: Street politics and ters. Journal of Criminal Justice, 23(5), 439– the transformation of a New York City gang. 450. New York: Columbia University Press. Delgado, R. (1985). ‘Rotten social background’: Bryk, A. S., & Raudenbush, S. W. (1992). Hier- Should the criminal law recognize a defense archical linear models: Applications and data of severe environmental deprivation? Law and analysis methods. Newbury Park, CA: Sage. Inequality, 3, 9–90. Burt, R. S. (1987). Social contagion and Eckberg, D. L. (1995). Estimates of early innovation: Cohesion versus structural equiv- twentieth-century U.S. homicide rates: An alence. American Journal of Sociology, 92(6), econometric forecasting approach. Demogra- 1287–1335. phy, 32(1), 1–16. Burt, R. S. (1992). Structural holes: The social struc- Egan, D. J., & Robinson, D. O. (1979). Models of ture of competition. Cambridge, MA: Harvard a heroin epidemic. American Journal of Psychi- University Press. atry, 136, 1162–1167. Campbell, A. (1986). The streets and violence. Elliott, D. S., Wilson, W. J., Huizinga, D., Samp- In A. Campbell & J. Gibbs (Eds.), Violent son, R. J., Elliott, A., & Rankin, B. (1996). transactions: The limits of personality. New The effects of neighborhood disadvantage on York: Blackwell. adolescent development. Journal of Crime and Cavalli-Sproza, L. L., & Feldman, M. W. (1981). Delinquency, 33, 389–426. Cultural transmission and evolution: A quanti- Fagan, J. (1989). The social organization of drug tative approach. Princeton NJ: Princeton Uni- use and drug dealing among urban gangs. versity Press. Criminology, 27(4), 501–536. Chin, K. (1996). Chinatown gangs: Extortion, Fagan, J. (1990). Intoxication and aggression. enterprise, and ethnicity. New York: Oxford Drugs and Crime – Crime and Justice: An University Press. Annual Review of Research, 13, 241–320. Coleman, J. S., Katz, E., & Menzel, H. Fagan, J. (1992). Drug selling and licit income in (1966). Medical innovation; A diffusion study. distressed neighborhoods: The economic lives Indianapolis: Bobbs-Merrill Co. of street level drug users and dealers. In A. Cook, P. J., & Laub, J. H. (1998). The unprece- Harrel & G. Peterson (Eds.), Drugs, crime and dented epidemic in youth violence. Crime and social isolation (pp. 99–146). Washington, DC: Justice, 24, 27–64. Urban Institute Press. Cook, P. J., & Ludwig, J. (2004). Does gun preva- Fagan, J. (1993). The political economy of drug lence affect teen gun carrying after all? Crim- dealing among urban gangs. In R. Davis, A. inology, 42, 27–54. Lurigio, & D. P. Rosenbaum (Eds.), Drugs and Cooney, M. (1998). Warriors and peacemakers: community. Springﬁeld, IL: Charles Thomas. How third parties shape violence. New York. Fagan, J. (1999). Context and culpability of ado- New York University Press. lescent violence. Virginia Review of Social Pol- Cork, D. (1999). Examining space-time interac- icy and Law, 6(3), 101–174. tion in city-level homicide data: Crack markets Fagan, J. (2002). Policing guns and youth violence. and the diffusion of guns among youth. Journal Future of Children, 12(2), 133–151. of Quantitative Criminology, 15(4), 379–406. Fagan, J., & Chin, K. (1989). Initiation into crack Coulton, C. J., Korbin, J. E., Su, M., & Chow, and powdered cocaine: A tale of two epi- J. (1995). Community-level factors and child demics. Contemporary Drug Problems, 16(4), maltreatment rates. Child Development, 66, 579–617. 1262–1276. Fagan, J., & Chin, K. (1991). Social processes of Crane, J. (1991). The epidemic theory of ghet- initiation into crack cocaine. Journal of Drug tos and neighborhood effects on dropping out Issues, 21, 432–466. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

720 jeffrey fagan, deanna l. wilkinson, and garth davies

Fagan, J., & Davies, G. (2004). Natural history of Felson, R. B., & Steadman, H. J. (1983). Situa- neighborhood violence. Journal of Contempo- tional factors in disputes leading to criminal rary Criminal Justice, 20(2), 127–147. violence. Criminology, 21(1), 59–74. Fagan, West, & Holland 2003 Fingerhut, L. A., Ingram, D. D., & Feldman, J. Fagan, J., & Wilkinson, D. L. (1997). Firearms and J. (1992a). Firearm and nonﬁrearm homicide youth violence. In D. Stoff, J. Brieling, & J. D. among persons 15 through 19 years of age. Maser (Eds.), Handbook of antisocial behavior. Differences by level of urbanization, United New York: Wiley. States, 1979 through 1989. Journal of the Fagan, J., & Wilkinson, D. L. (1998a). Guns, youth American Medical Association, 267(22), 3048– violence, and social identity in inner cities. 3053. Crime and Justice, 24, 105–188. Fingerhut, L. A., Ingram, D. D., & Feldman, Fagan, J., & Wilkinson, D. L. (1998b). Situa- J. J. (1992b). Firearm homicide among black tional contexts of adolescent violence. Revue teenage males in metropolitan counties. Com- Europen´eenne des Migrations Internationales, parison of death rates in two periods, 1983 14, 63–76. through 1985 and 1987 through 1989. Journal Fagan, J., Zimring, F. E., & Kim, J. (1998). Declin- of the American Medical Association, 267(22), ing homicide in New York City: A tale of two 3054–3058. trends. Journal of Criminal Law and Criminol- Fryer Jr., R. G., Heaton, P., Levitt, S. D., & Mur- ogy, 88(4), 1277–1324. phy,K.M.(2005). Measuring the impact of Fagan, J., West, V., and Holland, J. (2003) “Recip- crack cocaine (No. 11318). Washington, DC: rocal Effects of Crime and Incarceration in NBER Working Papers. New York City Neighborhoods.” Fordham Fullilove, M. T., Heon, V., Jimenez, W., Parsons, Urban Law Journal (30), 1551- 1602. C., Green, L. L., & Fullilove, R. E. (1998). Farmer, P. (1999). Infections and inequalities: The Injury and anomie: Effects of violence on modern plagues. Berkeley, CA: University of an inner-city community. American Journal of California Press. Public Health, 88 (6), 924–927. Federal Bureau of Investigation. (various years). Gardner, W., Mulvey, E. P., & Shaw, E. C. (1995). Uniform crime reports [United States]: Sup- Regression analyses of counts and rates: Pois- plementary homicide reports, 1976–2003 son, overdispersed poisson, and negative bino- (ICPSR Study No. 4351, 2005). Retrieved mial models. Psychological Bulletin, 118(3), from http://webapp.icpsr.umich.edu/cocoon/ 392–404. ICPSRSTUDY/04351.xml Gladwell, M. (1997, March 17). Annals of style: Feld, B. C. (1999). Bad kids: Race and the transfor- The coolhunt. The New Yorker. mation of the juvenile court. New York: Oxford Gladwell, M. (2000). The tipping point: How little University Press. things can make a big difference. Boston: Little Felson, R. B. (1982). Impression management Brown. and the escalation of aggression and vio- Goldstein, H. (2003). Multilevel statistical models lence. Social Psychology Quarterly, 45(4), 245– (Vol. 3). London: Edward Arnold. 254. Goldstein, P., Brownstein, H. H., Ryan, P. J., & Felson, R. B. (1993). Predatory and dispute- Bellucci, P. A. (1989). Crack and homicide in related violence: A social interactionist New York City, 1988: A conceptually based approach. In R. V. Clarke & M. Felson event analysis. Contemporary Drug Problems, (Eds.), Routine activity and rational choice: 16, 651–687. Advances in criminological theory (Vol. 5,pp. Gostin, L. O. (1991). The interconnected epi- 103–126). New Brunswick, NJ: Transaction demics of drug dependency and AIDS. Har- Press. vard Civil Liberties-Civil Rights Law Review, Felson, R. B. (1984). Patterns of aggressive social 26, 114–184. interaction. In A. Mummendey (Ed.), Social Gostin, L. O., Burris, S., & Lazzarini, Z. (1999). psychology of aggression: From individual behav- The law and the public’s health: A study of ior to social interaction (pp. 107–126). New infectious disease law in the United States. York: Springer-Verlag. Columbia Law Review, 99(1), 74. Felson, R. B., Ribner, S. A., & Siegel, M. S. (1984). Gould, M. S., Wallenstein, S., & Kleinman, M. Age and the effect of third parties during crim- (1990). Time-space clustering of teenage sui- inal violence. Sociology and Social Research, cide. American Journal of Epidemiology, 131(1), 68(4), 452–462. 71–78. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 721

Gould, M. S., Wallenstein, S., Kleinman, M., Klein, M. W. (1995). The American street gang: O’Carroll, P., & Mercy, J. (1990). Sui- Its nature, prevalence, and control. New York: cide clusters: An examination of age-specific Oxford University Press. effects. American Journal of Public Health, Klein, M. W., Maxson, C., & Cunningham, L. 80(2), 211–212. (1991). Crack, street gangs, and violence. Crim- Grogger, J., & Willis, M. (2000). The emergence inology, 29(4), 623–650. of crack cocaine and the rise in urban crime Krivo, L. J., & Peterson, R. D. (2000). The struc- rates. Review of Economics and Statistics, 82(4), tural context of homicide: Accounting for 519–529. racial differences in process. American Socio- Hamid, A. (1990). The political economy of logical Review, 65(4), 547–559. crack-related violence. Contemporary Drug Land, K., McCall, P., & Cohen, L. (1990). Struc- Problems, 17, 31–78. tural covariates of homicide rates: Are there Harcourt, B. E. (1998). Reflecting on the subject: any invariances across time and space? Ameri- A critique of the social influence conception can Journal of Sociology, 95(4), 922–963. of deterrence, the broken windows theory, and Land, K., McCall, P., & Nagin, D. S. (1996). order-maintenance policing New York style. A comparison of poisson, negative binomial Michigan Law Review, 97(2), 291. and semiparametric mixed poisson regression Harcourt, B. E. (2001). Illusion of order: The false models with empirical applications to crim- promise of broken windows policing. Cambridge, inal careers data. Sociological Methods and MA: Harvard University Press. Research, 24(4), 387–442. Harcourt, B. E. (2006). Language of the gun: Youth, Lessig, L. (1995). The regulation of social mean- crime, and public policy. Chicago: University of ing. University of Chicago Law Review, 62, Chicago Press. 943–1046. Hughes, L., & Short, J. (2005). Disputes involv- Loftin, C. (1986). Assaultive violence as a con- ing youth street gang members: Micro-social tagious process. Bulletin of the New York contexts. Criminology, 43(1), 43–76. Academy of Medicine, 62(5), 550–555. Hunt, L. G., & Chambers, C. D. (1976). The heroin Luckenbill, D. F. (1977). Homicide as a situated epidemics. New York: Spectrum Publications. transaction. Social Problems, 25, 176–186. Inciardi, J. A. (1979). Heroin use and street crime. Luckenbill, D. F., & Doyle, D. P. (1989). Struc- Crime and Delinquency, 25, 335–346. tural position and violence: Developing a cul- Johnson, B. D., Williams, T., Dei, K. A., & tural explanation. Criminology, 27(3), 419– Sanabria, H. (1990). Drug use in the inner 436. city: Impact on hard-drug users and the com- Lynch, A. (1996). Thought contagion: How belief munity. Crime and Justice: An Annual Review spreads through society. New York: Basic Books. (Drugs and Crime), 13, 9–67. Lynch, J. W., Kaplan, G. A., Pamuk, E. R., Cohen, Johnston, L. D. (1987). National trends in drug R. D., Heck, K. E., Balfour, J. H., et al. (1998). use and related factors among American high Income inequality and mortality in metropoli- school students and young adults, 1975–1986 tan areas of the United States. American Jour- (No. ED288142, DHS ADM 87–1535). Ann nal of Public Health, 88(7), 1074–1080. Arbor, MI: University of Michigan. Markovits, D. (2005). Quarantines and distribu- Jones, D. R., & Jones, M. B. (1994). Testing tive justice. Journal of Law, Medicine and for behavioral contagion in a case-control Ethics, 33(2), 323–344. design. Journal of Psychiatric Research, 28, 149– Massey, D. (1995). The new immigration and 164. ethnicity in the United States. Population and Jones, D. R., & Jones, M. B. (1995). Preferred Development Review, 21(3), 631–652. pathways of behavioral contagion. Journal of Massey, D., & Denton, N. (1993). American Psychiatric Research, 29, 193–209. apartheid. Cambridge, MA: Harvard Univer- Jones, M. B. (1997). Behavioral contagion and sity Press. official delinquency: Epidemic course in ado- May, R. M., Anderson, R. M., & Blower, S. M. lescence. Social Biology, 45(1–2), 134–142. (1990). The epidemiology and transmission Kahan, D. M. (1997). Social influence, social dynamics of HIV-AIDS. In S. R. Graubard meaning, and deterrence. Virginia Law Review, (Ed.), Living with AIDS. Cambridge, MA: MIT 83, 367–373. Press. Kelling, G., & Cole, C. (1996). Fixing broken win- Meier, R. F., Kennedy, L. M., & Sacco, V.F. (Eds.). dows. New York: Free Press. (2001). Advances in criminological theory. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

722 jeffrey fagan, deanna l. wilkinson, and garth davies

Vol. 9: The process and structure of crime: Crim- routine activities to the empiricism of hot inal events and crime analysis. New Brunswick, spots. Criminology, 29, 725–754. NJ: Transaction. Rose, H., & McClain, P. (1990). Race, place, and Messner, S., & Tardiff, K. (1986). Economic risk: Black homicide in urban America. Albany inequality and levels of homicide: An analy- NY: SUNY Press. sis of urban neighborhoods. Criminology, 24, Rosenfeld, R., & Decker, S. H. (1999). Are arrest 297–318. statistics a valid measure of illicit drug use? Monkkonen, E. H. (2001). Murder in New York The relationship between criminal justice and City. Berkeley, CA: University of California public health indicators of cocaine, heroin, and Press. marijuana use. Justice Quarterly, 16, 685. Morenoff, J., Sampson, R. J., & Raudenbush, S. Rothman, K. (1986). Modern epidemiology. 2001. Neighborhood inequality, collective effi- Boston: Little Brown. cacy, and the spatial dynamics of urban vio- Rowe, D. C., & Rodgers, J. (1991). Adolescent lence. Criminology, 39, 517–560. smoking and drinking: Are they “epidemics”? Morenoff, J. D., & Tienda. M. (1997). Underclass Journal of Studies on Alcohol, 52, 110–117. neighborhoods in temporal and ecological per- Rowe, D. C., & Rodgers, J. (1994). A social con- spective. Annals of the American Association of tagion model of adolescent sexual behavior: Political and Social Science, 551, 59–72. Explaining race differences. Social Biology, 41, New York City Department of Health and Mental 1–18. Hygiene. (1992). Injury mortality in New York Sampson, R. J. (1987). Urban Black violence: The City, 1980–1990. New York: Office of Vital effect of male joblessness and family disrup- Statistics and Epidemiology. tion. American Journal of Sociology, 93, 348– New York City Department of Health and Men- 382. tal Hygiene. (1997). Summary of vital statistics, Sampson, R. J. (1993). Linking time and place: 1996. New York: Office of Vital Statistics and Dynamic contextualism and the future of Epidemiology. criminological inquiry. Journal of Research in New York State Department of Health. (1994). Crime and Delinquency, 30(4), 426–444. Injury facts for New York State. Albany, NY: Sampson, R. J., & Lauritsen, J. L. (1994). Violent Author. victimization and offending: Individual, situa- Oliver, W. (1994). The violent social world of black tional, and community-level risk factors. In A. men. New York: Lexington Books. J. Reiss Jr. & J. Roth (Eds.), Understanding and Patterson, E. B. (1991). Poverty, income inequal- preventing violence: Social influences (Vol. 3,pp. ity, and community crime rates. Criminology, 1–114). Washington, DC: National Academy 29(4), 755–776. Press. Peterson, R. D., & Krivo, L. J. (1993). Racial segre- Sampson, R. J., Raudenbush, S. W., & Earls, F. J. gation and urban Black homicide. Social Forces, (1997). Neighborhoods and violent crime: A 71, 1001–1026. multilevel study of collective efficacy. Science, Phillips, S., & Cooney, M. (2005). Aiding peace, 277, 918–924. abetting violence: Third parties and the man- Sampson, R. J., & Wilson, W. J. (1995). Toward agement of conflict. American Sociological a theory of race, crime, and urban inequality. Review, 70(2), 334–354. In J. Hagan & R. D. Peterson (Eds.), Crime Planty, M. (2002). Third-party involvement in and inequality (pp. 37–54). Stanford: Stanford violent crime, 1993–1999. Washington, DC: University Press. Bureau of Justice Statistics Special Report. Schwendinger, H., & Schwendinger, J. S. (1985). Polk, K. (1994). When men kill: Scenarios of mascu- Adolescent subcultures and delinquency. New line violence. New York: Cambridge University York: Praeger. Press. Servin, J. (1999, February). Cool hunting with Robertson, L. S. (1992). Injury epidemiology. New Jane. Harpers Bazaar, 90. York: Oxford University Press. Shihadeh, E. S., & Steffensmeier, D. J. (1994). Rodgers, J., & Rowe, D. C. (1993). Social con- Economic inequality, family disruption, and tagion and sexual behavior: A developmen- urban Black violence: Cities as units of strat- tal EMOSA model. Psychological Review, 100, ification and social control. Social Forces, 73, 479–510. 729–751. Roncek, D., & Maier, P. A. (1991). Bars, blocks, Singer, J., & Willett, J. B. (2003). Applied lon- and crimes revisited: Linking the theory of gitudinal data analysis: Modeling change and P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

social contagion of violence 723

event occurrence. New York: Oxford University violent urban youth of neighborhood social Press. control dynamics and situations. Justice Quar- Slaby, R. G., & Roedell, W. C. (1982). Develop- terly (forthcoming). ment and regulation of aggression. In J. Worell Wilkinson, D. L., & Fagan, J. (1996). Understand- (Ed.), Psychological development in the elemen- ing the role of firearms in violence ‘scripts’: tary years (pp. 97–149). New York: Academic The dynamics of gun events among adoles- Press. cent males. Law and Contemporary Problems, Snijders, T., & Bosker, R. (1994). Modeled vari- 59 (1), 55–90. ance in two-level models. Sociological Methods Wilkinson, D. L., & Fagan, J. (2001a). A the- and Research, 22, 342–363. ory of violent events. In R. F. Meier, L. W. Sommers, I., & Baskin, D. (1993). The situational Kennedy, & V. F. Sacco (Eds.), Advances in context of violent female offending. Journal of criminological theory. Vol. 9: The process and Research in Crime and Delinquency, 30, 136– structure of crime: Criminal events and crime 162. analysis (pp. 169–196). New Brunswick, NJ: Stuntz, W. J. (2001). The pathological politics of Transaction. criminal law. Michigan Law Review, 100(3), Wilkinson, D. L., & Fagan, J. (2001b). What we 505–600. know about gun use among adolescents. Clin- Sullivan, M. (1989). Getting paid: Youth crime and ical Child and Family Psychology Review, 4(2), work in the inner city. Ithaca, NY: Cornell Uni- 109–132. versity Press. Wilkinson, R. G. (1996). Unhealthy societies: The Taylor, R. B., & Covington, J. (1988). Neighbor- afflictions of inequalities. London: Routledge. hood changes in ecology and violence. Crimi- Williams, K. R., & Flewelling, R. L. (1988). The nology, 26(4), 553–590. social production of criminal homicide: A Tienda, M. (1991). Poor people in poor places: comparative study of disaggregated rates in Deciphering neighborhood effects on behav- American cities. American Sociological Review, ioral outcomes. In J. Huber (Ed.), Macro-micro 53, 421–431. linkages in sociology (pp. 244–262). Newbury Williams, T. (1989). The cocaine kids: The inside Park, CA: Sage. story of a teenage drug ring. Reading, MA: Toch, H. (1969). Violent men: An inquiry into the Addison-Wesley. psychology of violence. Chicago: Aldine. Wilson, J. Q., & Kelling, G. L. (1982). Broken Tolnay, S. E., & Beck, E. M. (1995) A festival of windows. Atlantic Monthly, 249(3), 29–36, violence: An analysis of lynchings, 1882–1930. 38. Urbana, IL: University of Illinois Press. Wilson, W. J. (1987). The truly disadvantaged. Tolnay, S. E., Deane, G., & Beck, E. M. (1996). Chicago: University of Chicago Press. Vicarious violence: Spatial effects on South- Wilson, W. J. (1991). Studying inner-city social ern lynchings, 1890–1919. American Journal of dislocations: The challenge of public agenda Sociology, 12, 788–815. research. American Sociological Review, 56, 1– Wacquant, L. D., & Wilson, W. J. (1989). The 14. costs of racial and class exclusion in the inner Zimmer, L. (1987). Operation pressure point: city. Annals of the American Association of Polit- The disruption of street-level trade on New ical and Social Sciences, 501, 8–25. York’s Lower East Side. New York: New York Wallace, R. (1991). Expanding coupled shock Research Center in Crime and Justice, New fronts of urban decay and criminal behavior: York University School of Law. How U.S. cities are becoming “hollowed out.” Zimmer, L. (1990). Proactive policing against Journal of Quantitative Criminology, 7, 333– street-level drug trafficking. American Journal 356. of Police, 9, 43–74. Wilkinson, D. L. (2003). Guns, violence and iden- Zimring, F. E. (1999). American youth violence. tity among African-American And Latino youth. New York: Oxford University Press. New York: LFB Scholarly Publishing. Zimring, F. E. (2006). The great American Wilkinson, D. L. (2007). Local social ties and will- crime decline. New York: Oxford University ingness to intervene: textured views among Press. P1: KAE 052184567Xc36 0 521 84567 X February 27, 2007 1:3

724