Facilitating Violence: A Comparison of Gang-Motivated, Gang ...

Journal of Quantitative Criminology, Vol. 15, No. 4, 1999

Facilitating Violence: A Comparison of Gang-Motivated, Gang-Affiliated, and Nongang Youth Homicides Richard Rosenfeld,1 Timothy M. Bray,1 and Arlen Egley1

It is well established that gangs facilitate violent offending by members, but the mechanisms by which that facilitation occurs remain unclear. Gangs may promote violence indirectly by facilitating members’ access to risky situations such as drug markets or directly through gang functions such as turf defense. We explore alternative modes of facilitation in a comparison of gang-affiliated homicides (which involve gang members but do not result from gang activity), gangmotivated homicides (which result from gang activity), and nongang youth homicides in St. Louis. We find important differences as well as similarities in the time trends and event characteristic of the two types of gang homicide; in key respects the gang-affiliated homicides more closely resemble the nongang events. The gang-motivated events exhibit a somewhat distinctive spatial patterning, as might be expected from their connection to turf conflicts. However, all three homicide types are highly concentrated in racially isolated, disadvantaged neighborhoods, which remain the fundamental social facilitators of both gang and nongang violence. KEY WORDS: violence; youth homicide; gangs; facilitation.

1. INTRODUCTION Although the role of youth gangs in violence has been of long-standing interest to criminologists, renewed attention to gangs accompanied the dramatic growth in serious youth violence in U.S. cities over the past decade. Recent research has demonstrated that gang membership facilitates violence at the individual level. The mechanisms by which that facilitation occurs, however, are less clear. In fact, researchers do not agree on the meaning of the expression ‘‘gang-related violence,’’ and even the term ‘‘gang’’ remains in dispute (Howell and Decker, 1998). There are two plausible ways in which gangs might facilitate violence. They may increase individual members’ risk of violence by increasing general exposure to violent and risky situations, as would be the case, for 1

University of Missouri—St. Louis, St. Louis, Missouri 63121. 495 0748-4518y99y1200-0495$16.00y0  1999 Plenum Publishing Corporation

496

Rosenfeld, Bray, and Egley

example, if gang members were more likely than other youth to frequent high-crime areas of the city or carry guns. In such cases, mere affiliation with a gang serves to elevate violent victimization and offending. Alternatively, gangs motivate violence when members attack rival gangs or unaffiliated youth on behalf of the gang itself. What distinguishes such gangmotivated violent acts from those connected with gang affiliation is that they require gang membership. Gang membership is not a necessary condition for, say, robbing a drug dealer, even if gang members are more likely than others to rob drug dealers. In contrast, gang membership is a necessary, if not sufficient, condition for participation in a gang war. The distinction between the two types of relationship between gangs and violence has both theoretical and policy significance. Theories of gang violence should be able to specify with some precision the ways in which gangs promote violent activity by members, specifically whether violence is a by-product of the risky routine activities of gang members (Felson, 1998, pp. 17–18) or is directly promoted by specific gang functions. More general theories of criminal violence, especially those focusing on violent youth, must contend with the group-oriented quality of delinquent activity and, perforce, with the purposes and consequences of youth gangs. Practitioners should benefit from knowledge about how gangs are implicated in violent behavior, specifically whether gangs impel members to engage in violence or merely expose them to violent persons and situations. Presumably, intervention strategies will differ depending on the particular mechanisms by which the facilitation of violence takes place. Whether a significant empirical distinction exists between gang-motivated and gang-affiliated violence remains an open issue, as does the difference between gang and nongang youth violence. In this paper, we explore the utility of these distinctions with data on gang and non-gang homicides occurring in St. Louis, Missouri, between 1985 and 1995. We have three research purposes. First, we briefly document trends in gang homicides over the 10 years, a period during which overall levels of homicide in St. Louis rose sharply (Rosenfeld and Decker, 1996). Second, we compare gangaffiliated, gang-motivated, and nongang youth homicides with respect to characteristics of victims, offenders (when known), and events. We describe the age, sex, and race distributions of victims and offenders, the nature of the victim–offender relationship, the role of third parties, firearm use, drug and alcohol involvement, and the location of the killing. This portion of the analysis replicates Maxson and Klein’s important research comparing gangmotivated, gang-affiliated, and nongang homicides (Maxson and Klein, 1990, 1996; Maxson et al., 1985). Third, we extend this earlier research by examining the spatial distribution of gang and nongang youth homicide in relation to attributes of the neighborhood context in which the incidents

Facilitating Violence

497

occurred. A reasonable expectation is that the gang-motivated events cluster in and around gang territories and therefore should exhibit a distinctive spatial patterning, while the distribution of the less turf-oriented gangaffiliated events should more closely resemble that of nongang incidents. These analyses are carried out with rich microspatial data on more than 700 homicides coded by the authors from police homicide case files. 2. BACKGROUND The proliferation of gang activity in recent years has sparked renewed scholarly research on urban youth gangs. The sharp escalation in youth homicide rates during the 1980s and early 1990s has led to heightened concern on the part of gang researchers with the role of gangs in fostering serious interpersonal violence (Blumstein, 1995; Maxson, 1999). The study of gang violence has proceeded alongside continuing debate over fundamental conceptual and methodological issues. Essential definitional issues (What constitutes a ‘‘gang’’?) remain unresolved at the same time researchers are pressed to provide policy-makers with information regarding the most effective ways to combat gang violence (Howell and Decker, 1998). Even without agreement on a uniform definition of gangs or precisely how gang offending differs from other group offending by youth, little dispute remains regarding the importance of gangs in facilitating violent and other criminal behavior. Several studies have addressed whether gang membership contributes to offending above and beyond the individual level of propensity. Esbensen and Huizinga (1993) surveyed youth in the Denver area and found that gang members report two to three times as much delinquency as nongang offenders. In a study of offending rates before, during, and after gang membership. Thornberry et al. (1993) confirmed the previous result and concluded that attributes of the gang milieu uniquely contribute to delinquent and criminal behavior. Battin et al. (1998) extended this research further by demonstrating that the influence of gang membership on offending holds even when accounting for the association with delinquent peers. Battin et al. (1998) conclude that their results are consistent with an ‘‘enhancement’’ effect of gang membership on delinquency. They found that prior delinquency was positively associated with gang membership; however, gang membership predicted subsequent delinquency even after controlling for prior delinquency, as well as for delinquent peers. These results do not conform to a pure facilitation model, which implies no difference in pregang delinquency between gang members and nonmembers, but neither do they support a ‘‘birds of a feather’’ model, which implies no relationship between gang membership and offending after controlling for prior delinquency.

498


There is substantial evidence, then, that gangs contribute to delinquent and criminal behavior beyond the propensities of individual members or even the group-oriented nature characteristic of youth offending (see also Huff, 1998). Additional research utilizing official statistics has sought to further explicate the difference between gang and nongang behavior by comparing the characteristics of gang and nongang criminal incidents. Maxson et al. (1985) conducted an early evaluation of such differences in a comparison of gang and nongang homicides in Los Angeles, a so-called ‘‘chronic’’ gang city. They found significant differences between gang and nongang homicides with respect to several characteristics of the setting and participants. For example, gang homicides were more likely than nongang incidents to occur in public areas and involve firearms. Victims and offenders in the gang events were younger and less likely to know one another than in the nongang cases, and the total number of participants was greater in the gang homicides. Given these differences, the authors concluded that special investigation units for gang violence were justified in cities with large gang populations (for similar results see Bailey and Unnithan, 1994). After the publication of the Maxson et al. (1985) study, the debate over the definition of gang-related criminal violence became especially policy relevant, because different definitions of gang-related incidents presumably result in different pictures of the setting and participants and, by extension, different criminal justice responses. Police tend to use one of two working definitions of gang violence for classification purposes. In what has become known as the ‘‘Los Angeles definition,’’ an event is classified as gang-related if the suspect or victim is a known gang member. In contrast, the ‘‘Chicago definition’’ designates an offense as gang-related only if it exhibits qualities of a gang motive, such as retaliation, initiation, or defending gang turf. Under this more restrictive definition, mere gang membership is not sufficient to label a crime as gang-related (for a discussion of the alternative definitions, see Maxson, 1999). Without a uniform classification scheme, meaningful comparisons across cities of the level or characteristics of gang-related crimes cannot be made (Maxson, 1999). Maxson and Klein (1990) considered this obstacle and revisited their Los Angeles gang homicide data. They extracted, based on the Chicago definition, gang-motivated events from the total of gangrelated homicides and compared them with nongang homicides in parallel fashion to their previous comparison of all gang-related and nongang homicides. This ‘‘purification’’ process, according to Maxson and Klein (1990), reduced the number of gang-related homicides by about one-half. However, the gang-motivated homicides (based on the Chicago definition) closely resembled the gang-affiliated homicides (based on the LA definition) in their differences with the nongang incidents, leading the investigators to conclude


499

that ‘‘the choice of motive or member definition would make little difference empirically, conceptually, or in policy relevance’’ (Maxson and Klein, 1990, p. 83). Despite their conclusion of no consequential difference between the two prevailing definitions of gang violence, Maxson and Klein (1990, p. 88) do acknowledge that valuable information is lost ‘‘at the expense of giving up half the cases.’’ Just how much is still debatable. Maxson and Klein (1996) began to address this issue in a study of differences between gang-affiliated and gang-motivated homicides. Their analysis of 1988–1989 Los Angeles gang homicides yielded virtually no difference between the two, prompting them to conclude they are for ‘‘all intents and purposes identical’’ (p. 10). An important remaining research task is to compare both gang-motivated and gang-affiliated homicides with those committed by nongang youth, which should advance understanding of group processes in youth violence generally as well as shed additional light on how gang and nongang events differ. That is our essential objective in this paper. Using Maxson et al. (1985) as a template, we compare participant and setting characteristics of gang and nongang youth homicides. In light of the definitional issue discussed above, we undertake a three-way comparison of gang-affiliated, gang-motivated, and nongang youth homicide cases. In addition, our data are from an ‘‘emerging’’ gang city (Decker and Van Winkle, 1986). Little is known about how different types of gang-related homicides diverge from one another or from nongang incidents in these areas compared to chronic gang cities such as Los Angeles and Chicago. We also augment prior research by exploring the relationship between neighborhood structural characteristics and the three types of homicide. Although high levels of economic disadvantage and ethnic heterogeneity are related to rates of crime and violence in a community (Bursik and Grasmick, 1993; Harries, 1997), the specific features of neighborhoods that promote the social facilitation of gang violence have not been identified. Finally, we explore the spatial distribution of the three types of homicide, focusing particularly on whether the presumed territorial nature of the gang-motivated events induces a distinctive spatial pattern, beyond that attributable to the clustering observed in the neighborhood structural covariates. In each of these respects, our research extends previous work on the presumed distinctiveness of gang violence and the mechanisms facilitating higher levels of violence among gang members. 3. DATA AND METHODS The homicide data used in the analysis were compiled from case files maintained by the St. Louis Metropolitan Police Department. We do not

500


rely on police classification of events as ‘‘gang-related’’ but, rather, employ our own coding scheme based on the distinction between gang-affiliated and gang-motivated homicides discussed above. Any indication of gang affiliation or involvement in the record (e.g., reports of gang name, colors, use of gang signs) led to a classification of the event as one or the other type of gang homicide. A homicide was coded gang-motivated if it resulted from gang behavior or relationships, such as an initiation ritual, the ‘‘throwing’’ of gang signs, or a gang fight. A clear example of a gang-motivated homicide occurred in 1990, when a bystander was killed during a protracted period of conflict between rival gangs over control of an intersection on the north side of St. Louis. A case was coded as gang-affiliated if the homicide involved a suspect or victim identified in the police report as a gang member but did not arise from gang activity. An incident in which a gang member is killed during a robbery would be coded as a gang-affiliated homicide. The gang-affiliated and gang-motivated incidents, then, comprise mutually exclusive categories. Finally, an incident was coded as a nongang youth homicide if no indication of gang activity or affiliation was present in the case file, and the suspect was between 10 and 24 years old.2 It was sometimes difficult to determine from the case files whether a given event was gang-motivated, gang-affiliated, or gang-related at all. Typical of such reports were those containing a witness’s statement that one or more participants was a gang member, but no confirmation or elaboration in the form of self-reports, police documentation, or gang identification. In our initial event codes, we defined such cases as ‘‘possibly’’ gang-motivated or gang-affiliated, resulting in a six-category coding scheme: gang-motivated, possibly gang-motivated, gang-affiliated, possibly gang-affiliated, not gang-related, and insufficient information to identify motive. We applied our event codes to the main reports and all supplemental reports for the 1365 homicides committed in the city of St. Louis between 1985 and 1995. Cases were coded independently by two of the authors. A multiround iterative reliability process was used to achieve an interrater reliability coefficient of 0.70 in two passes (see the Appendix for a description of the coding process). Although we do not rely on police classifications of incidents as gang or nongang related, our coding scheme ultimately depends on the accuracy, completeness, and consistency of the information contained in police records. The police in St. Louis like those elsewhere became more concerned 2

No age restriction was placed on suspects in the incidents coded as gang-affiliated or gangmotivated. However, over 90% of the suspects in those incidents fall within the age range used for the nongang cases.


501

about gang-related violence when rates of youth violence were escalating sharply during the late 1980s and early 1990s, and so it is possible that upward trends in gang homicides are inflated by a greater willingness by the police over time to record evidence of gang affiliation or function. Although we cannot discount that possibility, the St. Louis police report no change in the kind of information they include in homicide case files. That would mean, for example, if a witness reported that the victim in a homicide was a gang member, that information is as likely to appear in a 1985 case file as in one for 1995. Further, were such changes to have occurred they are unlikely to affect our comparison of gang and nongang homicides by victim, offender, and event characteristics, which is limited to the years 1990–1995, when the police were highly concerned about gang violence. Again, however, the extent to which our trend estimates may be influenced by changes in police classification is unknown. The incident reports were also coded for several other attributes investigated in previous research on gang homicide, including the number, sex, age, and race of participants; victim–offender relationship; firearm use; drug and alcohol involvement; and incident location. We have extended that research by characterizing the community contexts in which the incidents occurred. The homicides were aggregated to the 588 census block groups comprising the city of St. Louis in 1990, with an average population of 675 residents per block group. Prior research has shown that aggregate levels of criminal violence are associated with rates of economic disadvantage, population heterogeneity, and residential instability in communities (Bursik and Grasmick, 1993; Land et al., 1990). We used the following measures to tap those dimensions of community at the block group level: family poverty rate, percentage males unemployed and jobless (includes persons not in the labor force), percentage female-headed households with children under 18 years old, vacancy rate, owner occupancy rate, percentage of the population in the same household 5 years ago, percentage of families receiving public assistance income, and percentage of the population between 15 and 24 years of age.3 A principal-component analysis of these nine items yielded two factors with eigenvalues above the conventional threshold of 1.00. The first factor, which we labeled Neighborhood Disadvantage, exhibits high loadings for the poverty, public assistance income, and female-headed household measures. Percentage owner-occupied housing units and population residing in the same household 5 years ago are the only measures with appreciable loadings on the second factor, which we labeled Neighborhood Instability. The resulting 3

All measures are from the 1990 decennial census STF-3A tapes compiled by the data resources center of the National Consortium on Violence Research.

502


factor scores are used to capture the two dimensions of community context.4 The racial composition of an area, which figures prominently in prior homicide research, is conspicuous by its omission from our factor analysis of neighborhood disadvantage and instability. It has become common practice to include a measure of racial composition, typically the fraction of the population African-American, along with other measures of community ‘‘disadvantage’’ in such data reduction exercises. Given the widely observed correlation between racial composition and aggregate measures of economic deprivation, it is not surprising that race usually loads heavily on factors reflecting high levels of deprivation. The interpretation of those factors is ambiguous, however, both conceptually and empirically. Race is conceptually distinct from ‘‘disadvantage,’’ and treating them as attributes of the same dimension confounds attempts at untangling their distinct influences on levels of violence in a community (see Massey, 1998). Therefore, we have retained the racial composition of the block group as a separate indicator in our analysis.

4. RESULTS 4.1. Trends in Gang-Related Homicide Figure 1 displays the trends in gang-related homicides in St. Louis from 1985 to 1995. The bars represent the number of gang-affiliated and gangmotivated homicides ocurring in each year and the lines designate each type’s percentage of all homicides. Four observations can be made of these trend data. First, we observe little gang involvement of either type in homicide during the 1980s; like other so-called emergent gang cities, gang-related homicides did not begin to escalate in St. Louis until the end of the decade. Second, gang homicides take off in 1989–1990, with an especially steep rise in gang-motivated homicides. Third, gang-motivated homicides peak in 1993 and decline over the next 2 years, while gang-affiliated homicides continue to rise through 1995. Fourth, since the early 1990s, roughly a quarter to a third of all homicides in St. Louis have been gang-related. The trend data provide useful background for the more detailed comparisons of gang and nongang homicides to follow. The rise in St. Louis gang-related homicides followed the increase in youth homicides, which began in the mid-1980s (Rosenfeld and Decker, 1996). That sequencing suggests that gangs, or more precisely gang violence, emerged in response to increased levels of risk for youth, which, in turn, may have been triggered 4

The factor scores for the second factor were multiplied by −1 to convert them to measures of residential instability.


503

Fig. 1. Trends in St. Louis gang-affiliated and gang-motivated homicides, 1985–1995. * Homicides with known suspects. Source: St. Louis Homicide Project.

by the firearm violence associated with the marketing of illicit drugs during the 1980s (Decker and Van Winkle, 1996; Blumstein and Rosenfeld, 1998). As in other large cities, overall levels of homicide in St. Louis have declined in recent years. The St. Louis police recorded 100 fewer homicides in 1996 (167) than at the peak in 1993 (267).5 That turnaround is reflected in the trend for gang-motivated homicides but not for gang-affiliated incidents, which continued to grow in number and as a fraction of all homicides through 1995. It is not clear why the gang-affiliated incidents have continued to increase, but one possibility is that gang members are more likely than other youth to persist in high-risk criminal activities such as drug dealing, even as overall levels of violence are on the decline. We find some evidence consistent with this interpretation in our comparison of the event circumstances of gang and nongang homicides, described in the next section. 4.2. Participant and Event Characteristics Table I compares gang-affiliated, gang-motivated and non-gang youth homicides by characteristics of participants, events, and neighborhood context. The comparisons are limited to homicides that occurred between 1990 5

Personal communication with the St. Louis Metropolitan Police Department.

504


Table I. Characteristics of Gang-Affiliated, Gang-Motivated, and Nongang Youth Homicides in St. Louis (1990–1995)

Victim Male Black Mean age (N ) Suspectc Male Black Mean age (N ) Incident 2Csuspects Stranger d Witness Firearm Alcohol Drug Location Street Other Public Residence (N ) Neighborhood Disadvantagee Instabilitye % black (N )

Gangaffiliated

Gangmotivated

Nongang youtha

Total

Significanceb

0.91 0.94 26.4 (119)

0.91 0.99 20.0 (145)

0.84 0.85 26.9 (443)

0.87 0.89 25.4 (707)

NS GAyN, GMyN GAyM, GMyN

0.99 1.00 20.5 (115)

1.00 0.99 18.7 (131)

0.95 0.92 19.7 (443)

0.97 0.95 19.7 (689)

GAyN, GMyN GAyN, GMyN GAyM, GMyN

0.41 0.37 0.60 0.94 0.28 0.46

0.52 0.35 0.79 0.99 0.23 0.19

0.30 0.25 0.58 0.82 0.28 0.38

0.36 0.29 0.63 0.88 0.27 0.35

GMyN NS GAyM, GMyN GAyN, GMyN NS GAyM, GMyN

0.34 0.34 0.27 (119)

0.50 0.37 0.10 (145)

0.38 0.24 0.32 (443)

0.40 0.28 0.26 (707)

GAyM GMyN GAyM, GMyN

0.68 0.08 0.84 (119)

0.69 0.13 0.85 (145)

0.66 0.17 0.74 (443)

0.67 0.14 0.78 (707)

NS NS GAyN, GMyN

a

Suspect 10–24 years-old. GAyM, significant difference between proportion gang-affiliated and gang motivated; GAy N, significant difference between proportion gang-affiliated and nongang youth; GMyN, significant difference between proportion gang-motivated and nongang youth. NS, no significant differences, α G0.05. Pairwise differences evaluated by Tamhane’s T2 test. c First named suspect. d Cases with unknown victim–offender relationship excluded (NG621). e Factor score from principal–components analysis; see text for description of variables. b

and 1995. With respect to attributes of partipants, we observe no meaningful difference in the sex or race of either offenders or victims in the gangrelated homicides. Participants in gang homicides tend overwhelmingly to be black males. Even though somewhat higher proportions of the victims in nongang youth homicide are female (0.16) and white (0.15), the nongang events, too, are dominated by black males. With respect to the age of


505

participants, the gang-motivated incidents stand apart from both the gangaffiliated and the nongang youth homicides. Victims and suspects in gangmotivated homicides are very close in age (19–20 years old on average), compared to a 6-year difference in the ages of victims and offenders for the other homicide types. The similarity in the ages of the victims and suspects in the gang-motivated homicides reflects their corporate character (many result from gang fights with same-aged opponents) and is consistent with several other characteristics distinguishing them from the gang-affiliated and nongang events. Gang-motivated homicides are more likely to involve multiple suspects, occur more often in the presence of witnesses, and are more public in nature (nearly 90% ocur on the street or other public places). In each of these details, the gang-affiliated homicides look as much or more like the nongang events as they do the gang-motivated homicides. In only one respect, the use of firearms, do the gang-related events more closely resemble one another than the nongang youth homicides. But, as with the race and sex composition of offenders and victims, perhaps the best characterization of gun involvement in youth homicides in St. Louis is that the gang events resemble the nongang events, only more so. The gang-motivated incidents are significantly less likely than the other types of homicide to involve illicit drug use or transactions. That does not mean, however, that homicides involving gang members are unrelated to drugs. Nearly half of the gang-affiliated homicides are drug-related and, although the difference is not statistically significant, they are more likely to involve a drug component than other youth homicides. In results not shown, we observe a growth overtime in the proportion of gang-affiliated homicides occurring in connection with drug marketing. If gang members are more willing or able than other youth to face the dangers in and around urban drug markets, especially as demand falls and markets shrink, that may help to explain the persistence of gang-affiliated homicides during a period of declining gang-motivated and other youth homicide. Finally, we observe little difference in the neighborhood contexts of the three homicide types. Table I presents the mean values (in z-scores) for our composite measures of neighborhood disadvantage and instability. Based on these measures and using the census block group as the unit of analysis, both gang and nongang youth homicides are concentrated in disadvantaged areas with moderate levels of instability. Gang homicides are also concentrated in predominantly black neighborhoods, but so are the nongang youth homicides. Although statistically significant, the difference in the racial composition of the neighborhoods with gang and nongang homicides is small and parallels the difference between gang and nongang homicides in race of victim and offender.

506


Fig. 2a. Gang-affiliated and gang-motivated homicides by neighborhood disadvantage St. Louis block groups, 1990–1995.

The pronounced concentration of all three types of homicide in disadvantaged, predominantly black St. Louis neighborhoods is immediately evident from the maps presented in Figs. 2a–3b. In Figs. 2a and 2b we have shaded the block groups by their level of disadvantage. Areas depicted as ‘‘very disadvantaged,’’ with the dark shading, have factor scores above the


507

Fig. 2b. Nongang youth homicides by neighborhood disadvantage St. Louis block groups, 1990–1995.

value of one; those shown as ‘‘disadvantaged,’’ with the light shading, have scores between zero and one; and those shown as ‘‘not disadvantaged,’’ with no shading, have scores below zero. Figure 2a shows the locations of the gang-related homicides and Fig. 2b shows the locations of the nongang incidents by level of neighborhood disadvantage. With few exceptions

508


Fig. 3a. Gang-affiliated and gang-motivated homicides by racial composition St. Louis block groups, 1990–1995.

(nearly all of which are in the nongang category), the homicides cluster in the more disadvantaged areas of the city. Figures 3a and b display the distribution of the gang and nongang homicides by neighborhood racial composition. Block groups with populations more than 75% black are shown with darker shading, those between


509

Fig. 3b. Nongang youth homicides by racial composition St. Louis block groups, 1990–1995.

25 and 75% black are shown with lighter shading, and those with populations less than 25% black are shown with no shading. We observe a marked concentration of homicides in predominantly black neighborhoods, again, with a small number of exceptions in the nongang category. Comparing across the two sets of maps, we also observe a striking degree of overlap between the disadvantaged and predominantly black areas. Almost without

510


exception, the more disadvantaged neighborhoods are predominantly black, and vice versa. The correlation between the disadvantage score and the percentage African-American for the 588 St. Louis block groups is rG0.701 (pF0.001). These observations regarding the distribution of homicides by the racial composition and level of disadvantage in St. Louis neighborhoods raise two questions for further investigation: (1) To what extent is the observed clustering in the homicide events a function of the equally evident clustering in attributes of neighborhood context? and (2) Is the effect on the different types of homicide of neighborhood socio-economic disadvantage separable from that of racial composition? We address each of these isues in the following section. 4.3. Spatial Distribution The spatial distribution of gang and nongang youth homicides in St. Louis is clearly not random by neighborhood characteristics. As we have seen, both types of homicide tend to cluster in areas with high levels of disadvantage and large concentrations of African-American residents. However, less obvious from purely visual inspection is whether the spatial clustering in the homicides is explained by the spatial proximity of such areas or whether the clustering results from ‘‘contagious’’ influences in the homicides themselves, such as the presence of retaliatory motives or turf battles (Decker and Van Winkle, 1996, pp. 22–23; Loftin, 1986). And what is not at all clear from the maps is whether neighborhood disadvantage has an effect on the distribution of homicides that is independent of racial composition. The nonrandom nature of the distribution of the homicides is confirmed by two common tests for spatial autocorrelation, Moran’s I and Geary’s c (see Anselin, 1995; Cliff and Ord, 1981). Both tests show significant autocorrelation in all three homicide types.6 To assess the presence of spatial order in the homicide events while controlling for the effects of disadvantage and race (and therefore their induced spatial clustering), we employed a maximum-likelihood function suggested by Anselin (1995) in separate equations for the gang-motivated, gang-affiliated, and nongang youth homicides. The maximum-likelihood estimation of the spatial lag model finds the value of the spatial autocorrelation coefficient which 6

Results available on request. Both Moran’s I and Geary’s c were computed using a first-power inverse distance weights matrix based on the distance between block group centroids. The matrix was row standardized and no limits were imposed in its construction, so that homicides occurring in each block group were allowed to affect every other block group.


511

Table II. Maximum-Likelihood Regression Results for Gang-Affiliated, Gang-Motivated, and Nongang Youth Homicide (NG588)a Gang-affiliated Neighborhood Instability Neighborhood Disadvantage Percentage population African-American Total population Spatial autocorrelation coefficient Intercept Log likelihood a

−0.0022 (0.0157) 0.0203(ns) (0.0149) 0.0028** (0.0008) 0.0002** (0.0000) 0.4320(ns) (0.2798) −0.1235* (0.0622)

(ns)

−419.431

Gang motivated

−0.0063 (0.0157) 0.0542** (0.0108) —

(ns)

0.0002** (0.0000) 0.8512** (0.1005) −0.0947* (0.0448) −425.402

(ns)

0.0185 (0.0191) 0.0199(ns) (0.0181) 0.0034** (0.0009) 0.0002** (0.0000) 0.5449* (0.2447) −0.0256** (0.0725) −536.352

Nongang youth (ns)

0.0130 (0.0192) 0.0629** (0.0131) —

0.0669* (0.0339) 0.1618** (0.0319) 0.0045* (0.0018) 0.0009** (0.0001) 0.3842(ns) (0.2721) −0.3646(ns) (0.1949)

0.0002** (0.0000) 0.8821** (0.0808) −0.1505** (0.0535) −542.562

−868.511

St. Louis block groups in 1990; standard errors in parentheses. **pF0.01; *pF0.05.

(ns)

0.0532(ns) (0.0336) 0.2082** (0.0239) — 0.0010** (0.0001) 0.7765** (0.1359) −0.4808** (0.1379) −871.484

pH0.05.

maximizes the following likelihood function L: ¯ i )ANy2 ln(2π )ANy2 ln(σ 2 ) lG∑ ln(1Aρω i

A( yAρWyAXβ )′(yAρWyAXβ )y2σ 2 where ρ is the spatial autocorrelation coefficient, ω is the Eigenvalue of the weights matrix. W represents the matrix of spatial weights, and X represents the matrix of covariates. Each equation regresses the number of homicides in a block group on the instability and disadvantage measures, and percent of the population African-Amerian. To capture remaining spatial effects, the equations also contain a spatial lag term, computed from the same spatial weights matrix discussed above. The regressions are based on event counts rather than population-standardized rates because homicide victims need not reside in the block group in which they are killed. To account for differences across neighborhoods in aggregate risk, block group population size is included on the right-hand side. Table II presents the regression results for the three homicide types. Given the large correlation between racial composition and the disadvantage measure, we report results from specifications with and without racial composition. In the equations containing the racial composition indicator, the measure of neighborhood instability has a significant effect on homicide only in the nongang category. In the equations from which racial composition has been omitted, neighborhood instability exerts no significant influence within any of the homicide types. In contrast, the level of neighborhood disadvantage significantly increases the frequency of nongang homicide in

512


both specifications. In the gang homicide categories, however, the measure of disadvantage is significant only in the equations from which racial composition has been omitted. Based on these results, we cannot separate the influence on gang homicides of a neighborhood’s level of socioeconomic disadvantage from the influence of racial composition. Although interpretation is confounded by the strong relationship between the two measures, it appears that when racial composition is controlled, neighborhood disadvantage does not significantly increase the frequency of gang homicide.7 The number of nongang homicides is significantly higher in more disadvantaged neighborhoods, however, and that effect appears to be independent of neighborhood racial composition. The remaining issue is whether spatial autocorrelation in the homicide distributions remains after accounting for neighborhood context. In the equations containing racial composition along with the other contextual measures, the coefficient for the spatial lag term is significant only in the gang-motivated category. When racial composition is omitted, we observe significant spatial autocorrelation within all three homicide types. These results suggest that the observed distribution of gang-affiliated and nongang youth homicides is a function of neighborhood context—the clustering observed in the homicides reflects the clustering in community disadvantage and racial composition. The spatial dependence observed in gang-motivated homicides, however, is attributable only in part to the clustering in the covariates. Although other unmeasured neighborhood characteristics may account for the residual spatial dependence in gang-motivated homicides, our results are consistent with a ‘‘contagion’’ hypothesis: the spatial distribution of gang-motivated homicides may reflect intrinsic features of the phenomenon and not simply the presence of facilitating neighborhood characteristics. 5. DISCUSSION Our results replicate and extend those of previous research on gangrelated homicide. Gang homicides emerged in St. Louis in the late 1980s, considerably later than in ‘‘chronic’’ gang cities such as Chicago and Los Angeles (Curry et al., 1996). Gang-motivated homicides peaked and began to decline in the early 1990s, whereas the gang-affiliated homicides continued to increase through 1995. Contrary to the results of prior research (Maxson and Klein, 1990, 1996), that is one of several distinctions we find 7

When disadvantage is omitted from the model and racial composition is retained, in contrast, racial composition continues to exert a significant influence on gang homicides. Results available on request.


513

between the two types of gang homicide. Although offenders and victims in both types of gang homicide are overwhelmingly black males, compared to the gang-affiliated events, the gang-motivated homicides involve more participants who are closer to one another in age, are more likely to take place in public, exhibit a somewhat distinctive spatial distribution, and are less likely to involve drugs. Along these dimensions, the gang-affiliated homicides more closely resemble the nongang youth homicides. These results are generally consistent with so-called ‘‘threat’’ explanations of gang violence (Decker and Van Winkle, 1996). Gangs emerged, by this account, as protective mechanisms in increasingly threatening urban environments during the 1980s. If the account is correct, then during the upswing in overall levels of youth violence in the late 1980s and early 1990s, the frequency of gang homicide should have increased; as the youth violence epidemic began to wane in the 1990s, gang homicides should have declined with diminishing threat. Those are the patterns we observe in the trend data for St. Louis, but only for the gang-motivated homicides. The continued rise in gang-affiliated homicides, we suggest, results from increased involvement of gang members—but not gangs—in the drug trade. Our results offer powerful evidence of the clustering of both gang and nongang youth homicides in areas characterized by high levels of socioeconomic disadvantage and racial isolation. Although it is difficult to isolate their separate influences, neighborhood disadvantage appears to have a more important effect on the frequency of the nongang homicides, whereas the effect of neighborhood racial composition is greater for the gang homicides. American youth gangs traditionally have been organized along ethnic and racial lines, and prior research has shown levels of gang homicide to vary with the ethnic composition of communities (Curry, 1994; Curry and Spergel, 1988). The reasons why gang formation and gang violence remain so closely bound to patterns of racial and ethnic segregation merit additional research. One promising avenue for that research is to explore the connection between prison gangs and urban youth gangs. The emergence and spread of contemporary urban gangs coincided with the sharp escalation of U.S. incarceration rates, especially among African-American males, during the 1980s. The dress and demeanor of street gang members in many respects seem modeled on the prison inmate. The prison could well function as an important source of ‘‘hierarchical diffusion’’ of street gangs, a possibility that deserves closer attention from both gang and prison researchers (an important early source is Moore, 1978).8 8

Unlike classic ‘‘contagious’’ spatial diffusion, hierarchical diffusion refers to the spread of phenomena from larger, dominant centers or nodes to smaller units (see Morrill et al., 1988).

514


In terms of the dynamics of the facilitation process, we find evidence for both ‘‘routine activities’’ and ‘‘gang function’’ mechanisms. Gang membership may facilitate access to risky situations such as drug markets that are not themselves the product of gang activity. Gangs may directly facilitate violence by virtue of the public and participatory nature of gang conflicts. Although we found evidence of a distinctive spatial patterning in gang-motivated homicides, however, the most important form of facilitation is contextual. Gang homicides are highly concentrated in disadvantaged, predominantly African-American communities in St. Louis. So, too, are homicides involving nongang youth. Although the accumulated evidence for gang facilitation of violence is quite compelling, our results serve as a reminder that concentrated disadvantage and racial isolation remain the fundamental sources of lethal violence in urban areas. APPENDIX The coding process consisted of two stages, a first-stage dichotomous classification of the homicides as gang-related or not gang-related and a second-stage classification of the gang-related events as either gang-motivated or gang-affiliated. Two of the authors participated in the coding. In the first stage, one of the authors classified each of the 1365 events as gangrelated or not gang-related, and the second author (blind to the results) applied the same code to a 10% sample of randomly drawn cases (nG133). The two assessments were highly correlated (rG0.898, pF0.01). In the second stage, one author coded each of the gang-related events in one of four categories: gang-motivated, gang-affiliated, possibly gangmotivated, and possibly gang-affiliated. The second author applied the same codes (blind) to a 25% random sample of those cases (nG66). The two assessments exhibited modest agreement after this round (κ G0.51, pF0.01). The coders revisited each discrepancy, discussed clarifications to the category definitions, and agreed on recodes for each of the events. A second 25% sample was then randomly drawn from the remaining 198 gang-related homicides which the second coder had not yet seen. The resulting 49 homicides were classified by the second coder in the manner described above. This second pass produced an acceptable level of intercoder agreement (κ G 0.70, pF0.01). Again, each discrepancy was revisited and mutually recoded. The coders then reviewed the principal areas of disagreement and, not surprisingly, found that the coding discrepancies were concentrated in the two ‘‘possibly’’ categories. After mutual recoding, those categories comprised 37% of the gang-affiliated and 38% of the gang-motivated cases, respectively. (Just 68, or 5%, of the 1365 total homicides could not be classified


515

due to missing information.) Analysis revealed that the cases coded as ‘‘possibly’’ gang-motivated or affiliated did not differ appreciably from their more definitively classified counterparts with respect to key situation and participant variables. We therefore collapsed those cases into the two broader gang-moivated and gang-affiliated categories. ACKNOWLEDGMENTS The authors wish to thank Carolyn Phillips and Luc Anselin for valuable comments during the preparation of this paper and the anonymous reviewers for their comments on an early version. We thank the St. Louis Metropolitan Police Department for providing access to the homicide data used in the analysis. The research reported in the paper was supported by a grant from the National Consortium on Violence Research (NCOVR). The points of view and conclusions expressed in the paper do not necessarily represent those of the NCOVR or the National Science Foundation. REFERENCES Anselin, L. (1995). SpaceStat, A Software Program for the Analysis of Spatial Data, Version 1.80, Regional Research Institute, West Virginia University, Morgantown. Bailey, G. W., and Unnithan, N. P. (1994). Gang homicides in California: A discriminant analysis. J. Crim. Just. 22: 267–275. Battin, S. R., Hill, K. G., Abbott, R. D., Catalano, R. F., and Hawkins, J. D. (1998). The contribution of gang membership to delinquency beyond delinquent friends. Criminology 36: 93–115. Blumstein, A. (1995). Youth violence, guns and the illicit drug industry. J. Crim. Law Criminol. 86: 10–36. Blumstein, A., and Rosenfeld, R. (1998). Explaining recent trends in US homicide rates. J. Crim. Law Criminol. 88: 1175—1216. Bursik, R. J., and Grasmick, H. G. (1993). Neighborhoods and Crime: The Dimensions of Effective Community Control, Lexington Books, New York. Cliff, A. D. and Ord, J. K. (1981). Spatial Processes: Models and Applications, Pion, London. Curry, G. D. (1994). Gang-related violence. Clearinghouse Rev. 28: 443–451. Curry, G. D., and Spergel, I. A. (1988). Gang homicide, delinquency, and community. Criminology 26: 381–405. Curry, G. D., Ball, R. A., and Decker, S. H. (1996). Estimating the national scope of gang crime from law enforcement data. In Huff, C. R. (ed.) Gangs in America, 2nd ed., Sage, Thousand Oaks, CA, pp. 21–36. Decker, S. H., and Van Winkle, B. (1996). Life in the Gang: Family, Friends, and Violence, Cambridge University Press, New York. Esbensen, F., and Huizinga, D. (1993). Gangs, drugs, and delinquency in a survey of urban youth. Criminology 31: 565–586. Felson, M. (1998). Crime & Everyday Life, 2nd ed., Pine Forge, Thousand Oaks, CA. Harries, K. D. (1997). Serious Violence: Patterns of Homicide and Assault in America, 2nd ed., Charles C Thomas, Springfield, IL.

516


Howell, J., and Decker, S. H. (1988). The Youth Gangs, Drugs, and Violence Connection, U.S. Department of Justice, Office of Juvenile Justice and Delinquency Prevention, Washington, DC. Huff, C. R. (1998). Comparing the criminal behavior of youth gangs and at-risk youths. Research in Brief, U.S. Department of Justice, National Institute of Justice, Washington, DC. Land, K. C., McCall, P. L., and Cohen, L. E. (1990). Structural covariates of homicide rates: Are there any invariances across time and space? Am. J. Sociol. 95: 922–963. Loftin, C. (1986). Assaultive violence as a contagious social process. Bull. N.Y. Acad. Med. 62: 550–555. Massey, D. S. (1998). Back to the future: The rediscovery of neighborhood context. Contemp. Sociol. 27: 570–572. Maxson, C. L. (1999). Gang homicide: A review and extension of the literature. In Smith, M. D., and Zahn, M. A. (eds.), Homicide: A Sourcebook of Social Research, Sage, Thousand Oaks, CA, pp. 239–254. Maxson, C. L., Gordon, M. A., and Klein, M. W. (1985). Differences between gang and nongang homicides. Criminology 23: 209–222. Maxson, C. L., and Klein, M. W. (1990). Street gang violence: Twice as great, or half as great? In Huff, C. R. (ed.), Gangs in America, 1st ed., Sage, Newbury Park, CA, pp. 71–100. Maxson, C. L., and Klein, M. W. (1996). Defining gang homicide: An updated look at member and motive approaches. In Huff, C. R. (ed.), Gangs in America, 2nd ed., Sage, Thousand Oaks, CA, pp. 3–20. Moore, J. W. (1978). Homeboys: Gangs, Drugs, and Prison in the Barrios of Los Angeles, Temple University Press, Philadelphia. Morrill, R. L., Gaile, G. L., and Thrall, G. I. (1988). Spatial Diffusion, Sage, Newbury Park, CA. Rosenfeld, R., and Decker, S. H. (1996). Consent to search and seize: Evaluating an innovative youth firearm suppression program. Law Contemp. Problems 59: 197–220. Thornberry, T. P., Krohn, M. D., Lizotte, A. J., and Chard-Wierschem, D. (1993). The role of juvenile gangs in facilitating delinquent behavior. J. Res. Crime Delinq. 30: 55–87.