Continuity and Change in Gang Membership and Gang Embeddedness

Embeddedness in Deviant Social Networks

In a series of works, Granovetter maintained that action is socially embedded and understood in the context of overlapping social networks. In so doing, he emphasized the “strength of weak ties” for obtaining employment, in that acquaintances play an instrumental role in providing information and connecting people with jobs. The concept of social embeddedness contains both relational and structural components for actors and their social networks. Every social network provides different arrangements of resources and restrictions that influence human behavior. These arrangements vary along levels of social and economic stratification, the extensiveness and distribution of strong and weak network ties, and the cultural or network norms and values (Granovetter 1983; Krohn 1986; Moody and White 2003). For example, among socioeconomically disadvantaged groups, extensive network ties are functionally different from individuals embedded in social networks with extensive acquaintances among more advantaged groups.

Hagan (1993) expanded on Granovetter’s argument to show how criminal embeddedness can lead to long-standing patterns of unemployment in the legitimate labor market. Hagan identified three qualities of criminal embeddedness: ties to criminal others, involvement in criminal acts, and isolation from prosocial networks. These characteristics serve to limit legitimate employment opportunities while presenting further illegitimate opportunities. Later work identified the dynamic nature of criminal embeddedness and its role in the building of criminal capital (McCarthy and Hagan 1995), leading to dense criminal network ties, leadership positions within the criminal network and increased illegal earnings (Uggen and Thompson 2003). Criminal embeddedness is a multidimensional, emergent property encompassing not only conventional network characteristics such as density of network ties or centrality within a deviant network but also the level of involvement in crime, isolation from prosocial networks, positions of leadership within a deviant network, and adoption of deviant values and identities.

Gangs are a type of deviant social network that can be characterized by their street-orientation, youthfulness, durability across time, and group identity that at least partially consists of illegal activity (Klein and Maxson 2006:4). Thornberry et al. (2003:7) referred to gangs as “social networks that embed their members in deviant routines and isolate them from prosocial arenas.” After movement into the gang, group processes and related functions lead members to shed their prior nongang friends and acquaintances (Decker and Van Winkle 1996:141, 187). As a result, gang members accumulate capital in the street, but they do not obtain the social and human capital required to transition successfully from adolescence to adulthood (Anderson 1999; Coleman 1988; Decker and Van Winkle 1996; Hagedorn 1998; Sullivan 1989; Thornberry et al. 2003).

Gang members are not a homogenous group, nor is the gang experience homogenous across individuals. While researchers have long recognized this in both qualitative and quantitative contexts (Klein 1971; Miller 2011; Moore 1991; Thornberry et al. 1993; Vigil 1988), the additional subtleties of gang membership have been underdeveloped. Instead, the gang literature tends to treat gang member heterogeneity as a dual taxonomy. This has been carried out in two ways, the first of which classifies gang members along a core-fringe dichotomy. Klein (1971) did this using street workers' evaluations of dependence on and participation in the gang as the criteria for core gang members. Esbensen et al. (2001) employed a concentric circle to allow gang youth to identify centrality within their gang, and then dichotomized according to inner and outer rings. Gaes et al. (2002) used a correctional administrative approach referring to gang-affiliated inmates as member, suspect, or associate. ¹ The second approach classifies gang members along a stable-transient dichotomy using repeated self-nomination as the key criterion (Battin et al. 1998; Bendixen, Endresen, and Olweus 2006; Craig et al. 2002; Hill, Lui, and Hawkins 2001; Thornberry et al., 2003). The problem, as Papachristos (2006:113) held, is that “relations among [gang] members are far more complex than a simple core-periphery distinction.” Even social network analysis—with methods that tap group centrality and density—has trouble surmounting this limitation because (1) data are limited to select police and historical gang sources (but see Fleisher 2006; McGloin 2005; Morselli 2008; Papachristos 2006; Sarnecki 2001) and (2) qualities of ties such as shared values, attitudes, identification, and relationship strengths are difficult to capture. ²

We extend Hagan’s (1993) concept of criminal embeddedness to gang membership. Criminal embeddedness refers to individual immersion in enduring deviant social networks, restricting involvement in prosocial networks. Gang embeddedness thus captures this immersion, reflecting varying degrees of involvement, identification, and status among gang members—the adhesion of the gang member to the gang. The central implication is that gang embeddedness inversely reduces exposure to and involvement in other networks, reducing information flow, opportunities to fill structural holes, and the ability to accumulate prosocial capital (Burt 1992; Coleman 1988; Granovetter 1983). Accordingly, the consequences of gang membership are unlikely to be experienced uniformly and instead are conditioned by levels of embeddedness.

Continuity and Change in Gang Membership

The movement to explain the longitudinal nature of offending and victimization has grown tremendously over the past two decades (Liberman 2008; Osgood 2005; Piquero et al. 2003; Sampson and Laub 2005). The life-course criminology framework can be constructively applied to gangs because membership follows similar patterns: youth join gangs (onset), persist over a period of time (continuity), and typically leave gangs (desistance). Central to life-course research are the concepts of turning points and trajectories. Laub, Sampson, and Sweeten (2006:314) observed that turning points are important life events that modify pathways in the life-course “in ways that cannot be predicted from earlier events” and Thornberry et al. (2003:7) conceived of gang joining as a turning point because gang processes are capable of “redirecting [a] person’s life.” Melde and Esbensen (2011) demonstrated this on a sample of school-age youth showing not only that entry was associated with elevated levels of delinquency but also harmful shifts in informal social controls, attitudes, and emotions.

A trajectory is a pathway or “line of development over the life span” (Elder 1985; Sampson and Laub 1993:8). Trajectories are enduring life states that are susceptible to change. The state of gang membership can be conceived as a trajectory because it follows a pathway across time and is marked by onset and termination—or identification and deidentification with the gang (Pyrooz and Decker 2011). Findings from longitudinal studies in Denver, Pittsburgh, Rochester, Seattle, and a multisite sample (GREAT) indicate that the trajectories of gang membership are relatively brief. Indeed, the majority of gang youth remained involved with gangs for only one year or less (48 percent to 69 percent). However, many youth reported two (17 percent to 48 percent), three (6 percent to 27 percent), and even four or more (3 percent to 5 percent) years of gang membership (see Appendix A), suggesting considerable variability in trajectories of gang membership.

The problem, however, is that the longitudinal contours of gang membership are largely descriptive. That is, we know much less about the correlates of gang membership patterns. This gap in the literature is likely because researchers have focused overwhelmingly on risk factors for gang joining (Pyrooz, Decker, and Webb 2010). As a result, important questions—for example, what factors are associated with persistence/desistance across time?—remain to be addressed. Given the robust association between gang membership and delinquency (Krohn and Thornberry 2008), hastening periods of gang membership has important criminological implications.

We contend that gang embeddedness is a dynamic component of gang membership, and that it should be associated with continuity and change in the longitudinal contours of gang membership. As noted above, the nature of gang membership and its attendant isolation from broader conventional networks should be experienced more thoroughly among those more deeply embedded, leading individuals to persist along the gang trajectory. As educational researchers have used engagement to school to capture persistence in trajectories of educational attainment (Fredricks, Blumenfeld, and Paris 2004; Libbey 2004), we hold that gang embeddedness is inversely related to the likelihood of desistance from gangs (i.e., those most embedded in a gang should be the individuals least likely to desist). While embeddedness is expected to vary both within- and between-persons over time, between-individual variation in embeddedness should foreshadow both short- and long-term patterns of gang membership. Importantly, although these issues have been discussed (Horowitz 1983; Klein and Maxson 2006; McGloin 2007; Miller 2011; Moore 1991), they have yet to be assessed empirically. This study aims to fill this void.

Method

Independent Variables

Embeddedness

The key independent variable in this study is gang embeddedness. We construct a gang embeddedness scale using a mixed-graded response model (Samejima 1969, 1997) applied to five variables that are asked of gang members only: frequency of contact with the gang (seven categories), position in the gang (three categories), importance of the gang to respondent (five categories), proportion of friends in the gang (five categories), and frequency of gang-involved assaults (four categories) at the baseline interview. These items represent several aspects of the multidimensional, emergent property of gang embeddedness. While these elements have been theoretically linked to criminal embeddedness in prior studies, criminal embeddedness has been operationalized only by criminal involvement (Hagan 1993; Uggen and Thompson 2003), contact with deviant peers (Bernburg et al. 2006; Hagan 1993; McCarthy and Hagan 1995; Uggen and Thompson 2003), proportion of friends involved in crime (McCarthy and Hagan 1995), and gang membership itself (Bernburg et al. 2006). Each of the items included in the present study represents an important dimension of embeddedness and, collectively, an improvement on previous studies. ⁵

Because these items have different numbers of categories, we employ a graded response model, based on item response theory, to construct the embeddedness scale. This model assumes that a normally distributed latent trait, theta (θ), which we call “gang embeddedness,” accounts for the observed response patterns (Osgood, McMorris, and Potenza 2002; Samejima 1969, 1997). The graded response model proceeds in two steps. First, response patterns are analyzed to score questions on two dimensions: the α parameter reflects the strength of the relationship between the question and the latent trait, similar to scores in a factor analysis; and the β parameters reflect the level of θ at which the likelihood of an affirmative response to an ordered response category or higher passes 50 percent. The β estimates reflect the seriousness of each response category, the level of embeddedness at which one would be expected to endorse the response. In the second step, each individual is assigned an estimate of gang embeddedness (θ) which maximizes the probability of the observed response patterns. In cases where an individual responds with seemingly contradictory response patterns (e.g., a gang leader with zero gang friends), the question with higher item discrimination (α) is accorded greater weight. Maximum likelihood is used to obtain estimates in both steps.

A graded response model provides several advantages for our study. First, if the assumptions of the model are met, the model combines information from questions with varying numbers of categories to estimate a latent trait, in our case, gang embeddedness. Other methods of scale construction are not well suited to combining questions of varying numbers of categories. Simply summing the responses arbitrarily accords greater influence to those questions with more categories and therefore more variance. Summed z scores arbitrarily impose equal steps between response categories and also limit the contribution of questions with fewer responses. Second, the assumptions of a graded response model are consistent with our conceptual framework. We are suggesting that previous methods of distinguishing between gang members that only include two or three categories are not sufficient; rather, gang embeddedness is a continuous construct which we estimate using a graded response model.

In Table 1, we present the five questions used for the graded response model, including response categories, response distributions, and model parameters. ⁶ The item discrimination parameters indicate that frequency of contact with the gang and the importance of gang to self are the two most strongly related questions to the latent trait of gang embeddedness. Position in the gang, which in some studies, is used to create categories of gang members (e.g., Venkatesh and Levitt 2000), has the weakest relationship with gang embeddedness among the 5 items. The seriousness parameters provide information about the gang embeddedness continuum. They show, for example, that reporting zero friends who are gang members is associated with extremely low levels of gang embeddedness (−4.13 or lower), while reporting only gang member friends is associated with very high levels of gang embeddedness (2.53 or higher). Participating in any amount of gang violence is associated with high levels of gang embeddedness: any gang violence is associated with gang embeddedness scores of 1.26 or higher, and frequent gang violence (5+ times) is associated with our highest item seriousness parameter (2.90). Remarkably, this frequency of gang violence is endorsed by nearly one quarter of all gang members in our survey. ⁷

OPEN IN VIEWER

Table 1.

Gang Embeddedness Graded Response Model.

	Item Discrimination	Response Categories, Prevalences, and Parameters
Question	α	β1	β2	β3	β4	β5
How often did you have contact with the gang/posse?		less than monthly	two times a week to once a month	three to six times a week	daily
		15%	13%	9%	63%
	1.65 (.11)	−0.95 (.08)	.07 (.05)	.35 (.05)
What is your position in the gang/posse?		member	top person or leader
		72%	28%
	.50 (.10)	2.55 (.52)
How important is the gang/posse to you?		not at all important	a little bit	moderately	quite a bit	extremely
		28%	20%	13%	21%	18%
	1.08 (.09)	−1.22 (.12)	−.07 (.07)	.69 (.09)	1.96 (.16)
How many of your friends are NOT members of the gang/posse?		all	most	half	a few	none
		5%	12%	15%	53%	15%
	.71 (.07)	−4.13 (.48)	−2.25 (.23)	−.86 (.13)	2.53 (.28)
How often have you beaten up, threatened, or physically attacked someone as part of a gang in the past year?		never	once or twice	three or four times	five or more times
		42%	23%	11%	24%
	.77 (.09)	1.26 (.16)	2.15 (.26)	2.90 (.35)

Note: N = 1,155 for model parameters as all gang person-years were utilized, N = 226 for prevalence, baseline gang members used in subsequent analyses.

The distribution of estimated embeddedness scores for our 226 baseline gang members is shown in Figure 1. We standardize this construct to have a mean of 0 and a standard deviation of 1 so that the gang embeddedness coefficient is interpreted in terms of standard deviations. Individuals at the low end of the distribution, with scores below -1 typically report infrequent contact with the gang, a noncentral role in gang structure and violence, and have many friends outside the gang. On the other end, those with embeddedness scores above 1 are more likely to be a gang leader, have more frequent contact with the gang, fewer friends who are not gang members, and a higher chance of participating in gang violence.

OPEN IN VIEWER

Figure 1.

Gang embeddedness distribution.

Process variables

Three gang-related measures tap into individual and group characteristics. Expected desistance is from a single dichotomous item asking the subject if they expected to be in gang in the near future (recoded so that 1 = no). ⁸ Expected gang desistance taps into otherwise unobservable individual resolve and preferences. It has long-term predictive validity. Of those expecting to desist, 13 percent were still in the gang five years later, as opposed to 30 percent who did not expect to desist. Gang organization was comprised of five dichotomous items—insignia, sharing money, sharing drugs, rules, and punishments—commonly found in the literature on gang organizational structure (Decker et al. 2008). These items were summated to create an index where higher scores equaled greater organization (Cronbach’s α = .70; mean interitem r = .31). Finally, years in gang was calculated from self-reported gang-joining age relative to the date of the baseline interview. On average, youths in this sample reported having been in the gang for four years, with some indicating that they had essentially been born into the gang.

Because differences in offending and victimization may foreshadow changes in gang membership (Decker and Lauritsen 2002; Krohn and Thornberry 2008), two variety scales of offending and victimization are included containing 21 self-reported offending items and 4 violent victimization items. ⁹ Next, we include a construct that assesses an individual’s social–emotional adjustment to external constraints (Weinberger and Schwartz 1990). A series of 22 items tapping into the constructs of impulse control (comparative fit index [CFI] = .95; root mean square error approximation [RMSEA] = .07), suppression of aggression (CFI = .97; RMSEA = .06), and consideration of others (CFI = .99; RMSEA = .04) were condensed into one measure that, in remaining consistent with the literature, we term self-control. The measure was standardized and scaled so that higher scores indicated higher self-control.

Control variables

Several standard demographic variables were included as controls, including age, male, racially Black, and ethnically Hispanic. Parental educational level was used as a proxy for socioeconomic status. This item was coded so that low scores reflect less education (0 = grade school or less) and high scores reflect more education (4 = college graduate). Next, intact household was a dichotomous item reporting whether a subject’s household contained both biological parents. We also control for site by including whether the subject lived in Phoenix or Philadelphia (1 = Phoenix, 0 = Philadelphia).

Analytic Strategy

We use baseline characteristics of gang members to model patterns of gang membership over five years using a logistic hierarchical linear model (HLM; Raudenbush and Bryk 2002). HLM allows us to parsimoniously model the probability of gang membership over eight follow-up waves. Our model takes on the following form:

L e v e l 1 : η i j = β 0 j + β 1 j T i j + β 2 j T i j 2

L e v e l 2 : β 0 j = y 00 + ∑ k = 1 n γ 0 k P k j + μ 0 j

β 1 j = y 10

β 2 j = y 20

We begin our analysis by estimating a random-intercept HLM model without level-2 predictors. This provides a measure of the magnitude of the intercept term variance. Next, we estimate a model containing all variables to identify the predictors of continuity and change in gang membership. Importantly, this allows us to assess the ability of embeddedness to explain—net of a host of predictors—patterns of gang membership.

Results

Descriptive Statistics

On average, baseline gang members were interviewed 7.4 times across the 8 follow-up waves of the Pathways study, for a total sample size of 1,670. These individuals reported gang membership 38 percent of the time across the follow-up waves. The average prevalence of gang membership dropped quickly from 100 percent at the baseline interview to just over 50 percent only a year later. Five years after baseline, still 22 percent of the subjects remained in a gang. The pattern of gang desistance is not challenged by missing data. Manski (1990) bounds on gang prevalence are no wider than .09 at any given wave. Descriptive statistics are shown in Table 2.

OPEN IN VIEWER

Table 2.

Descriptive Statistics.

	M (SD)	Min	Max
Level 1 (N = 1,670)
Time (years since baseline)	2.47 (1.42)	.41	5.22
Gang membership	.38	0	1
Gang membership, wave 1 (n = 226, T = 0)	1	0	1
Gang membership, wave 2 (n = 207, T ¯ = .5)	.61	0	1
Gang membership, wave 3 (n = 207, T ¯ = 1.0)	.52	0	1
Gang membership, wave 4 (n = 209, T ¯ = 1.5)	.43	0	1
Gang membership, wave 5 (n = 208, T ¯ = 2.0)	.38	0	1
Gang membership, wave 6 (n = 207, T ¯ = 2.5)	.34	0	1
Gang membership, wave 7 (n = 208, T ¯ = 3.0)	.30	0	1
Gang membership, wave 8 (n = 213, T ¯ = 4.0)	.26	0	1
Gang membership, wave 9 (n = 211. T ¯ = 5.0)	.22	0	1
Level 2 (N = 226)
Gang embeddedness	0.00 (1.00)	−2.76	1.91
Expected gang desistance	.45	0	1
Gang organization	0.00 (1.00)	−2.08	1.13
Years in gang	4.04 (2.46)	0.41	15.88
Age at baseline	16.53 (1.08)	14.06	19.00
Male	.90	0	1
Black	.21	0	1
Hispanic	.64	0	1
Other race/ethnicity	.05	0	1
Socioeconomic status (SES)	1.47 (0.99)	0	4
Intact household	.13	0	1
Delinquency variety	5.28 (4.08)	0	17
Victimization variety	.68 (0.97)	0	4
Self-control	0.00 (1.00)	−2.51	2.76
Phoenix	.79	0	1

Youths in our sample were born between 1981 and 1986, with a mean age of 16.5 at the baseline interview and 21.5 at wave 9. The sample is 90 percent male, 21 percent Black, and 64 percent Hispanic. Only 13 percent lived with both biological parents and typically their parents did not complete high school. Not surprisingly, this sample was highly delinquent with over five different delinquency types on average. Likewise, respondents reported a considerable amount of victimization, averaging .68 on a four-point variety scale. Nearly 80 percent of the sample comes from Phoenix. ¹⁰ To determine the relationship between gang embeddedness, other controls, and subsequent gang membership, we turn to the results of our logistic HLM models.

Logistic HLM Predicting Gang Membership

Table 3 presents the results of our logistic HLM models predicting gang membership in the five years following the baseline interview. Model 1 estimates the growth parameters and variance component without any level-2 predictors. Model 2 estimates the effects of all baseline characteristics on the intercept growth term. The variance component is statistically significant in the unconditional model, indicating that there is a considerable amount of variation in gang membership across individuals. Attesting to its explanatory strength, controlling only for embeddedness, we explain 12.6 percent of the variance of the intercept (model not shown). Using all 15 baseline predictors increases the explained variance to 26.6 percent. Reliability of the random effect is fairly strong, decreasing from .79 in the unconditional model to .70 in the final model.

OPEN IN VIEWER

Table 3.

Logistic Hierarchical Linear Modeling (HLM) of Gang Membership (N = 226, NT = 1,670).

	Model 1	Model 2
Intercept (γ00)	.854 (.180)*	1.126 (.202)*
Embeddedness (γ01)		.585 (.134)*
Expected desistance (γ02)		−.581 (.260)*
Gang organization (γ03)		.219 (.135)
Years in gang (γ04)		.009 (.040)
Age at baseline (γ05)		−.008 (.118)
Male (γ06)		1.177 (.418)*
Black (γ07)		1.584 (.463)*
Hispanic (γ08)		1.034 (.446)*
Other race/ethnicity (γ09)		1.188 (.757)
Socioeconomic status (SES; γ010)		−.108 (.125)
Intact household (γ011)		.564 (.340)
Delinquency (γ012)		−.004 (.033)
Victimization (γ013)		−.063 (.122)
Self-control (γ014)		−.304 (.129)*
Phoenix (γ015)		1.653 (.309)*
T (β1)	−.826 (.137)*	−1.134 (.165)*
T2 (β2)	.084 (.023)*	.119 (.028)*
Variance component (μ0j)	5.982*	4.392*
Explained variance		26.60%

Note. Population-average model coefficients with robust standard errors (in parentheses), estimated using restricted maximum likelihood option in HLM 6.0 software.

*p < .05.

The three growth parameters in model 1 predict the log odds of gang membership. At the first follow-up interview, where T = .5, the log odds of gang membership is estimated to be .46 (.854 + .5 × (−.826) + .5 × .5 ×.084), corresponding to a probability of gang membership of .61, exactly equivalent to the actual value, shown in Table 2. Four years postbaseline, the log odds of gang membership is calculated as .854 + 4 × (−.826) + 16 × .084 = −1.106, corresponding to a probability of .25, again very close to the actual value of .26 at wave 8. In model 2, the growth parameters define the expected growth curve at the sample average for all baseline characteristics.

Model 2 confirms our theoretical prediction that greater embeddedness in a delinquent network such as a gang corresponds to longer lengths of association. A 1 SD increase in gang embeddedness at the baseline interview is associated with .585 higher log odds of gang membership throughout the five-year follow-up period:

η i j = 1.126 + .585 × E m b − 1.134 × T i j + .119 × T i j 2 .

Expectations for continued involvement with the gang also have a statistically significant impact on gang membership. Youths' expectations about departing gangs tend to be correct but not overwhelmingly so. Expected gang desistance is associated with −.58 lower log odds of gang membership after one year, relative to no expectation of desistance. Neither gang organization nor years in the gang have any impact on gang membership net of the other predictors.

Of the other control variables, male, Black, Hispanic, and Phoenix site are all associated with lower rates of desistance from gangs. Remarkably, these variables appear to have a much larger effect on the log odds of gang membership than expectations of desistance from the gang. Black youths, for example, have 1.58 higher log odds of gang membership relative to non-Black non-Hispanic youths. Of particular interest for theory, higher self-control predicts lower rates of continued gang membership, with a 1 SD difference in self-control associated with a .30 difference in the log odds of gang membership.

In Figure 2, we illustrate the relative impact of our predictors of gang desistance. The dotted line represents the pattern of gang desistance holding all covariates at their sample averages. We also present three growth curves that vary only levels of gang embeddedness, showing the predicted probability of gang membership over the five-year follow-up period for those with 1 SD above the mean, 1 SD below the mean, and 2 SD below the mean on gang embeddedness. We then show the added effect of gang expectations, self-control, Hispanic, and Phoenix.

OPEN IN VIEWER

Figure 2.

Predicted gang membership from six months to five years after baseline by gang embeddedness, ethnicity, self-control, and expectations.

Figure 2 shows that gang embeddedness is associated with large differences in the probability of gang membership over the entire follow-up period. At the one-year follow-up, 67 percent of those who were embedded at 1 SD above average are expected to remain in the gang versus 38 percent of those embedded at 1 SD below average. At the two-year follow-up, and thereafter, the former group is more than twice as likely to remain in the gang. As more risk factors are added, the expected rate of desistance declines even further. Strongly embedded Hispanic gang members in Phoenix who do not expect to desist from the gang and have poor self-control are nearly universally expected to remain in the gang at the first follow-up, and nearly 60 percent are expected to be gang members after five years.

Since our sample averaged 16.5 years old at the start of the survey, it is clear that a sizable proportion of the sample persists in gang membership into early adulthood, well past typical ages for gang desistance (Klein and Maxson 2006; Krohn and Thornberry 2008). Importantly, gang embeddedness has long-term implications for continuity and change in gang membership.

Discussion

Only recently has the field of criminology moved beyond treating deviant network or group participation as an in/out dichotomy. Indeed, the introduction of social network theory and methodology has spearheaded this movement forward. However, deviant network participation involves more than just the structural properties—density, centrality, reciprocity—of the network. Network ties include “the amount of time, emotional intensity, the intimacy (mutual confiding), and the reciprocal services” input into relationships (Granovetter 1973:1361). In extending Hagan’s notion of criminal embeddedness to gangs, we captured deviant network immersion in terms of routine contact, identification of importance, participation in acute activities, the proportion of out-group friendships, and individual social status. Based on our findings presented above, we focus on three main points in this discussion.

First, the results demonstrate a robust relationship between embeddedness and continuity in gang membership. That is, individuals weakly embedded in gangs desist at a faster rate than those more deeply embedded in gangs. Individuals are limited in time and energy to invest in social arenas, such that a greater investment in Social Network A reduces the opportunity for investment in Social Network B (Burt 1992; Granovetter 1983; Moody and White 2003), especially in the context of deviant social networks (Hagan 1993; Schroeder et al. 2007; Uggen and Thompson 2003). In particular, gangs constrain rather than facilitate connections to other (pro)social networks. As embeddedness increases so too will the constraining forces of the gang. Efforts devoted to maintaining a social connection to the gang will preclude growth in social and human capital in other important social realms, such as education and employment. In this sense, embeddedness acts as an evolving, cumulative disadvantage making the successful transition to adulthood problematic due to being undereducated, undertrained, and socially isolated. As we have demonstrated, one consequence of embeddedness is the continued involvement with the gang.

Gang membership contains added dimensions of formal and informal controls that are likely conditioned by gang embeddedness. Formally, deeply embedded gang members are prone to be the targets of suppression strategies such as problem-oriented policing and civil gang injunctions, as well as subject to the penalty enhancements associated with gang-related criminal behaviors. Formal social control could result in placement in detention facilities which are hotbeds for gang activity or result in the “mark of a criminal record” that further restricts chances for employment (Griffin 2007; Pager 2003). Consistent with this thesis, supplementary analysis (not shown) revealed that “street time” was inversely related to continued gang membership. Informally, the folklore of gang culture and myths associated with gang leaving—such as getting “beaten out” or having to assault a relative (Decker and Van Winkle 1996; Vigil 1988)—are likely to permeate more thoroughly among those more embedded in a gang. Against this backdrop, it is not unrealistic to find that embeddedness slows the rate of desisting from gang membership.

Second, static or time-stable factors were associated with longitudinal patterns of gang membership. Males, Blacks, and Hispanics remained in gangs over longer time periods than their female and White counterparts. While it is well known that males and racial/ethnic minorities are overrepresented in gangs (Klein and Maxson 2006) and that this relationship tends to strengthen as the seriousness of the sample increases (e.g., school vs. detention samples), this is among the first systematic documentation that males and minorities persist over long periods. Why might this be the case? Ethnographic studies provide insight into some of these trends. Sullivan’s (1989) study of three different racial/ethnical neighborhoods in New York and Suttle’s (1968) research on various ethnic gangs in a Chicago neighborhood indicate that White youth, compared to Black and Hispanic youth, were more integrated into parochial networks connected to legitimate enterprises. This allowed them to obtain employment through extended contacts thus cutting ties with the gang. Minorities had a more difficult time because their larger social networks were less integrated into, for example, business and city government communities (see also Hagedorn 1998; Vigil 2002). These potential explanations appear relevant for both Black and Hispanic trends.

We also observed that increased self-control was associated with shorter durations of gang membership. The fact that those with poorer self-control remained in gangs for longer periods poses important questions for theory and gang research: to what extent do Gottfredson and Hirschi’s (1990) hypotheses extend into the context of gangs? It could be that those with poorer self-control are not equipped with the requisite skills to transition from gangs into other life domains. Gangs, instead, provide continued opportunities to generate income. Despite the direct implications for one another (Kissner and Pyrooz 2009), these literatures have proceeded without crossing paths in meaningful ways, especially in the longitudinal context.

We also found no relationship between the organizational structure of the gang and continuity and change in gang membership. In other words, individual immersion in gangs (i.e., embeddedness) “mattered” more for continuity/change than the structural properties of the group. It should be considered, however, that the measure of embeddedness was drawn from five graded items, as opposed to five binary items for organizational structure. Also, embeddedness likely draws from aspects of gang organizational structure and related group processes that we are unable to tap. We thus hesitate to assert that gang member embeddedness is a factor “capable of overcoming the group processes that reinforce gang membership” (Klein and Maxson 2006:234). Instead, we hope this finding spurs additional inquiry into the intersection between individuals and groups and the “level of explanation” problem in criminology (Short 1998).

Third, important insights into the nature of gang membership are drawn from life-course and social network frameworks. Criminal careers research has shed light onto the nature, volume, and extent of criminal involvement across the life-course within diverse populations (Piquero et al. 2007). Much like criminal careers research, the contours of gang “careers” contain important conceptual and methodological implications. We contend that gang membership is a dynamic process that can be understood by changes in within-individual gang embeddedness over time. As such, embeddedness plays a critical role in understanding key parameters—intermittency and duration—in the evolution of gang involvement. Intermittency, or the rejoining of a gang, was reported by 49 respondents, or 57 percent of multiwave gang members. ¹¹ Thornberry et al. (2003; Lovegrove and Thornberry 2008) reported an intermittency rate between 57 percent and 66 percent. It is unknown whether this finding is an empirical reality or an artifact of the operationalization of gang membership; criminal careers research has recognized that intermittency is fraught with conceptual and analytic issues (Piquero, 2004). However, it is likely that the levels of embeddedness remain elevated during periods of temporary deidentification, providing a thread of continuity for intermittent trajectories of gang membership. Duration, or the period between onset and termination of gang membership, tended to be longer in the current study than previous research (Appendix A). Indeed, our most and least conservative estimates indicate that 50 percent to 62 percent remain in gangs for one year, 21 percent to 22 percent for two to three years, and 16 percent to 28 percent for over three years. Roughly one of every nine respondents reported continuous gang membership throughout the entire study period. If we included their self-reported years of membership prior to the baseline interview, this would increase to an average of nine years of gang membership. Given what is known with respect to the age-crime curve, the bulk of offending within criminal careers will likely be found within gang careers.

In light of these findings, the contours of gang membership highlight two important study limitations. First, the high-risk nature of our sample is a likely source of divergence for the duration findings, which may not represent typical gang careers. Among adjudicated samples stochastic selectivity processes are in place, thereby possibly capturing more serious gang members. There are, however, enough longitudinal projects with self-reported gang membership items to embark on comparative research in order to organize knowledge about these patterns. Second, we only focused on baseline levels of gang embeddedness, with self-reported gang membership serving as the gate item triggering embeddedness questions. As an emergent property, gang embeddedness should amplify as individuals trend toward identification with the gang and subside as individuals trend away from deidentification—this may help explain the continued consequences of gangs despite deidentifying as a gang member (Melde and Esbensen 2011; Pyrooz et al. 2010).

We conclude by offering a modest agenda forward. First, related to the operationalization of embeddedness, it would be useful for future surveys to use similarly graded (Likert) scales to simplify for factor analytic construction. Second, it would be equally useful to examine the similarities and differences between existing operationalizations of what we refer to as embeddedness, such as the concentric circle (Esbensen et al. 2001), administrative or practitioner (Gaes et al. 2002), repeated self-nomination (Thornberry et al. 1993), or social network analytic approaches. In addition, researchers could build on this measure of embeddedness by examining other dimensions of overlapping role and network relations (Krohn 1986). Third, studies should examine the degree to which factors such as race/ethnicity, gender, and neighborhoods moderate embeddedness across various outcomes such as desistance, offending, victimization, or other life-course transitions. Fourth, it is imperative to examine the correlates of gang embeddedness and the time-varying nature of embeddedness. This area of research will benefit our knowledge of the turning point significance of gang membership and assist in identifying intervention points that would be valuable to policymakers and social service agencies.

Finally, at the outset of this article, we posited that embeddedness is relevant for various deviant social networks, such as race supremacy groups, religious cults, motorcycle clubs, organized crime groups, and terrorist networks. “Gang” can be substituted for any of the aforementioned networks and embeddedness can be examined. There are many similarities across deviant social networks, especially with regard to the desistance process (Bjorgo and Horgan 2009; Ebaugh 1988). If embeddedness extends across contexts, this will aid in developing a larger, more comparative knowledge base of deviant social networks to the benefit of gang research in particular and criminological research in general.