Does it matter if those who matter don’t mind? Effects of gang versus delinquent peer group membership on labeling processes

Abstract

Despite renewed interests in the labeling perspective and the impact of official intervention on individuals’ future outcomes, scant attention has been given to potential conditioning factors for theorized labeling processes. We argue that, when viewed through a symbolic interactionist lens, variations in the nature of primary social groups, through which individuals filter official labels like arrest, may generate patterns for subsequent self-concept and delinquency that are contrary to what labeling theory indicates. To test our rationale, we offer a moderated mediation model in which gang membership is expected to differentially impact the effect of arrest on future delinquency through an intermediary mechanism: self-esteem. We test a gang–nongang dichotomy and then probe further to test whether hypothesized effects are gang specific or occur similarly for nongang youths with highly delinquent peer groups. Analyzed using Rochester Youth Development Study (RYDS) data (N = 961), comparisons between gang members and nonaffiliated youths with similarly highly delinquent peer groups revealed no significant differences in conditional indirect effects of arrest on self-esteem and future delinquency; the two groups were similarly insulated from any negative impact of arrest on self-esteem. For nongang youths with fewer delinquent peers, however, arrest significantly reduced later self-esteem, which in turn increased their future delinquency.

Interest in the labeling perspective peaked in the late 1960s and early 1970s. Subsequently, it was critiqued for several reasons and interest in the perspective waned. To a large degree this was a result of early labeling research findings either failing to support or providing equivocal support for the link between juvenile justice system involvement and self-concept and then self-concept’s impact on delinquency and drug use (Paternoster & Iovanni, 1989).

Renewed concern about the unintended consequences of police contact and arrest for the adolescent life course has emerged alongside a revival of research on how juvenile justice system involvement affects youths’ abilities to succeed educationally, in the workforce, and in social relationships. Researchers have also begun to examine the heterogeneity of labeling processes or, more specifically, why official intervention may not have problematic consequences for all youths (e.g., see Hirschfield, 2008; Morris & Piquero, 2013; Wiley, Carson, & Esbensen, 2016). In some of this research, scholars have accounted for potential factors that may condition the relationship between official intervention and such consequences, yet whether there are factors that condition (moderate) the relationship between official intervention and self-concept has received scant attention. Does official intervention negatively affect self-concept for some individuals but not for others?

Derived from a symbolic interactionist perspective, the labeling perspective is premised on an understanding of social interaction and its effects. It is important to understand the social audience to which an actor may be reacting and the probable impact of said audience on an actor’s subsequent attitude and behavior. For the most part, researchers have assumed that juveniles react to representatives of conventional society such as parents, teachers, and law enforcement agents. For adolescents engaged in delinquent activities, however, such conventional representatives may not be the only, or even the most important, reference groups. Rather, the most influential social audience could very well be members of an actor’s delinquent peer network. If so, the questions arise as to how that social audience might be expected to react to instances of official intervention and how those reactions might then impact the actor’s self-concept and future delinquency.

Predicated on the recognition that youths involved in gangs may view a gang-related social audience as more important than conventional “others,” we provide a theoretical rationale for why members of delinquent gangs may be less likely to follow theoretically posited labeling processes such as altered self-concepts as a result of being arrested. An effect of gang membership that is over and above simply having a large proportion of delinquent friends has been found in prior research (e.g., see Battin, Hill, Abbott, Catalano, & Hawkins, 1998; Thornberry, Krohn, Lizotte, Smith, & Tobin, 2003); thus, we examine whether the impact of gang involvement on labeling processes is distinct from the impact of having a similar or smaller proportion of peers who engage in delinquent behaviors. We do so by contrasting the influence of having a larger or smaller proportion of delinquent friends (but not considering oneself in a gang) with the influence of being gang involved on the mediated effects of arrest on self-esteem and subsequent future delinquency. The resulting predictions are then empirically examined.

1 |. SYMBOLIC INTERACTIONISM AND OFFICIAL INTERVENTION

The labeling perspective is rooted in symbolic interactionism. Although frequently acknowledged on the periphery, this is often ignored in the actual consideration of how official intervention impacts an actor. Such oversight is unfortunate because recognizing the theoretical basis for the perspective may help generate new or different insights on how the labeling process unfolds.

A key premise of symbolic interactionism is that, for an actor, the “self” is a social object; as such, the self develops through social interaction. As Mead (1934, p. 140) stated, “it is impossible to conceive of a self arising outside of social experience.” “Others” react to, define, and label an actor. From these reactions, the actor comes to an understanding about who he or she is. This conception of self becomes a part of the actor’s environment to which he or she continues to react and, through ongoing social interactions, continues to transform. Throughout this dynamic process, the ensuing sense of self, then, is primarily a reflection of the actor’s perception and understanding of how others view the actor (Charon, 2007, pp. 72–73).

Part of the dynamic nature of this process is that an actor typically engages with various groups, leading to multiple reference groups from which the self-concept evolves. The impact of reference groups’ appraisals of an actor depends on 1) the salience of each group to the actor, and 2) the relevance of each group to the particular situation in which an actor finds him- or herself at a given time (Matsueda, 1992, p. 1586). As such, an actor’s conception of self is likely to be most influenced by his or her primary reference group. As the self emerges over time, however, social interactions and reference group primacy may shift and others’ reactions to the actor might also change. As interactions and reactions accumulate, some tend to overlap, allowing an actor to draw conclusions about him- or herself and eventually form a more or less stable self-concept. This self-concept then becomes a social object from which an actor derives social meaning (Mead, 1934) and uses in “forming and guiding” his or her actions, behaviors, or attitudes (Blumer, 1969, p. 62).

The purpose of this brief review of some central tenets of symbolic interactionism is to underscore the importance of 1) the role of the social audience in forming a conception of self, 2) the recognition that there can be multiple social audiences that influence one’s self-concept depending on the situation, and 3) the integral role of primary reference groups in forming one’s self-concept. In addition to these tenets, in the interactionist approach, particularly serious or extreme reactions to an actor may undermine the stability of a self-concept and even produce a change in it (Charon, 2007, p. 82). If we conceptualize official intervention by a law enforcement agent or agency (e.g., arrest) as a particularly serious or extreme reaction to an actor, it follows that such intervention would be predicted to impact the actor’s self-concept. This, of course, is the basic premise of labeling theory.

1.1 |. Defining self-concept

Self-concept is a multidimensional construct referring to the totality of individuals’ thoughts about themselves (Rosenberg, 1979). It includes several subcomponents, including both self-identity and self-esteem (Fitts, 1965). The self-concept subcomponent most consistent with the theoretical logic of the labeling perspective is identity. Reactions of others are thought to elicit the adoption of a deviant identity, thus, fostering continued involvement in problematic behavior. Although deviant identity may be the most appropriate measure of self-concept within a labeling framework, it is not the only aspect of the self that may be altered by the application of a deviant label. Self-esteem, for example, or the overall feeling of self-worth is also expected to be affected by a deviant label.

Stager, Chassin, and Young (1983) suggested that deviant social labeling will result in low selfesteem based on two theoretical principles. First, the principle of reflected appraisals in the context of a deviant label is assumed to be negative, resulting in lower self-esteem. The second principle is that of social comparison (Festinger, 1954; Rosenberg, 1979). Stager and colleagues (1983) argued that there will be invidious social comparisons between labeled and nonlabeled deviants, perpetuating lower selfesteem among the former. The data used in the current study do not include a direct measure of deviant identity, therefore, we explore the impact of official intervention on self-esteem, an operationalization that has been used in other empirical examinations of labeling processes (Al-Talib & Griffin, 1994; Jensen, 1972; Stager et al., 1983; Zhang, 2003).

Together, the two principles, reflected appraisals and social comparison, are regarded as central to understanding the varying impact of salient reference groups on one’s conception of self (Stryker, 1968). For instance, official intervention does not operate in a vacuum; as others learn of an event like an arrest, they may serve as filters for the impact that an official label has on an actor and his or her sense of self. Moreover, as the subsequent impact of official intervention on an actor’s self-concept is filtered through the reference group most relevant to the situation, it will also be contingent on the actor’s relationship to that group and the nature of the group. Indeed, depending on the nature of the group, others’ reactions to official intervention may vary considerably.

2 |. GANGS AND HIGHLY DELINQUENT PEERS AS REFERENCE GROUPS

A large body of research exists on what it means for an individual to be involved in a gang. Compared with nongang members, gang-involved individuals have elevated rates of violent delinquent activity while they are gang affiliated (Krohn & Thornberry, 2008; Pyrooz, Turanovic, Decker, & Wu, 2016; Thornberry, Krohn, Lizotte, & Chard-Wierschem, 1993). They are more likely to carry a weapon (Lizotte, Krohn, Howell, Tobin, & Howard, 2000) and are more likely to be violently victimized during this time (Pyrooz, Decker, & Webb, 2014; Taylor, Peterson, Esbensen, & Freng, 2007). Even after leaving the gang, former members are more likely to engage in violent criminal behavior than they were before becoming gang involved (Pyrooz et al., 2014). As such, gang members’ increased criminal activity, coupled with increased official scrutiny of gang-related behaviors, leads to higher probabilities of coming into contact with agents of the juvenile or criminal justice system and to greater likelihood of arrest (Krohn, Ward, Thornberry, Lizotte, & Chu, 2011; Tapia, 2011a).

Not only has gang involvement been shown to impact criminal behavior, victimization, and arrest, it has also been found to have problematic consequences for other important life trajectories. Gang-involved individuals are less likely to graduate from high school (Gilman, Hill, Hawkins, Howell, & Kosterman, 2014; Thornberry et al., 2003), less likely to be employed as young adults (Gilman et al., 2014; Melde & Esbensen, 2011), more likely to have children during the teenage years (Krohn et al., 2011; Thornberry, Smith, & Howard, 1997), and more likely to have financial problems in early adulthood (Krohn et al., 2011).

Gang involvement has such a pervasive impact on an individual’s life course, in part, because gang membership represents an alternative to failed family situations (Hill, Howell, Hawkins, & Battin-Pearson, 1999; McDaniel, 2012; Thornberry et al., 2003) and problems with the educational system (Melde & Esbensen, 2011; Taylor et al., 2007). Common risk factors for youth gang joining include originating from families living in poverty, single-parent households, and/or families in which bonds between parent(s) and child are weak; each of these factors is also associated with poorer monitoring and supervision of children (Hill et al., 1999; McDaniel, 2012). Gang members are also more likely to be truant, have low academic aspirations and school achievement, and be less bonded to school (Hill et al., 1999; Taylor et al., 2007; Thornberry et al., 2003). As scholars have emphasized, for youths who are disaffiliated with their family and school, gangs provide an arena for a family-like atmosphere where the members can have, as one gang member stated, “a sense of belonging. I had a family. They supported me in everything I did and I didn’t find that at home.” (Curry, Decker, & Pyrooz, 2014, p. 146).

2.1 |. Relevance to defining “self” and future delinquency

The family and school domains are two key arenas in which youths traditionally form social identities and conceptions of self. According to Woo, Giles, Hogg, and Goldman (2015, p. 136), “individuals have a desire and tendency to define their self-concept in terms of their group memberships.” When youths are disaffiliated with family and/or school, they can turn to gangs to acquire what they ordinarily would have obtained in more conventional social contexts. In influencing a person’s self-identity and self-esteem, the gang, then, replaces or at least provides an alternative to deficient family and/or school environments (Krohn et al., 2011; Sanchez-Jankowski, 2003; Woo et al., 2015).

Second, the gang, by definition, is involved in behaviors that are against the law (Weerman et al., 2009; Wood & Alleyne, 2010), and engaging in unlawful behaviors helps solidify one’s place in the gang (Klein & Maxson, 2006; Woo et al., 2015). A likely consequence of engaging in serious, often violent, behavior is coming into contact with law enforcement (Tapia, 2011a). Given that such activities and their consequences are part and parcel of being gang involved, it seems plausible that, for gang members, being arrested comes with the territory. Furthermore, labels considered stigmatizing in one context may be the norm, or even sources of pride, in other social contexts (Goffman, 1963). Thus, we could expect that, for gang-involved youths, being arrested would have little consequence for their selfesteem because 1) arrests are likely outcomes of the very behaviors promoted within the structure of a gang and 2) gangs as deviant reference groups likely shield members from potential negative reactions by more conventional others (Lemert, 1951; Paternoster & Iovanni, 1989).¹

According to labeling theory, altered self-concepts, like lower self-esteem after official intervention, in turn, increase the probability of future delinquent behavior. We have argued, however, that the expected effect of official intervention on self-esteem theoretically anticipated from a labeling perspective will depend on youths’ gang involvement status. If gang members’ self-esteem is not affected by arrest, then the theoretically predicted indirect effect, in which an official label’s impact on self-esteem would in turn increase subsequent delinquency, is also unlikely to occur. Such logic implies a conditional (moderated) effect of official intervention on the intermediary mechanism: self-esteem; in this case, less of an impact of arrest on self-esteem is expected for gang members compared with nongang members.

Still, it is possible that the conditional processes we have argued are not only gang specific. Our rationale may similarly apply to youths whose peer reference groups are not gang related but simply comprise a greater proportion of delinquent peers than those of other nongang-involved youths. Another aim of this study is to assess such possibilities and contribute to the identification of gang features that are distinct from or similar to other “social collectives … and sources of peer influence” (Pyrooz et al., 2016, p. 384). Is the potential “buffer” that deviant reference groups might provide for the impact of arrest on self-esteem and subsequent delinquency specific only to gangs? Or are youths who are nongang affiliated but who have comparably high delinquent peer association similarly shielded from such labeling effects?

2.2 |. Related research

In previous literature, scholars have demonstrated that gang involvement significantly raises the amount of serious, violent delinquent behavior (Krohn & Thornberry, 2008; Pyrooz et al., 2016; Thornberry et al., 1993) and, more broadly, that having delinquent friends increases the probability of delinquent behavior (Warr, 2002). Therefore, we are further motivated by the question as to whether the “gang effect” is a result of something that is unique to the gang as a social entity or is more simply because gang members likely have large proportions of friends who also engage in delinquent behavior (Battin et al., 1998; Thornberry et al., 2003). Battin and colleagues (1998) partly explored this by dividing a sample into three groups: those who reported being in a gang, those who reported having friends who committed delinquent behavior but were not in a gang, and those who were neither in a gang nor reported having delinquent friends. Comparing the three groups’ delinquent behavior, the authors found that gang members had higher rates of criminal involvement over the past year than did the other two groups. The nongang member group with delinquent friends also had greater criminal involvement compared with the group that did not have delinquent friends. After including all three groups in a structural equation model, the authors found that gang membership independently predicted both self-reported and official delinquency even after the effects of delinquent friends were taken into consideration.

Putting the expected effects of gang membership compared with simply having delinquent friends to a stiffer test, Thornberry and colleagues (2003) divided nongang members into quartiles according to scores on a delinquent peer associations scale. The results showed that gang members had higher rates of delinquency and drug use compared with all other respondents, including the nongang members in the highest quartile peer delinquency group. The authors also compared gang-involved youths with an equal number of nongang members who reported high delinquent peer associations, again finding that gang members had the higher delinquency rates. Finally, they compared the effects of gang membership and associating with delinquent peers in a regression model, finding a significant relationship between gang membership and delinquent behavior, even after the delinquent peer association measure was included in the equation.

These earlier findings, along with those from more recent research, serve to prompt this study’s examination of anticipated labeling process differences between gang members and nongang youths with delinquent peer networks of varying size. For example, Wiley et al. (2016) incorporated tests of conditional effects (i.e., moderation) along direct pathways from arrest to various outcomes but the authors did not “explicitly examine the labeling process” or include tests of intermediary measures between arrest and said outcomes (Wiley et al., 2016, p. 21). The authors’ goals were to assess whether official intervention was, in fact, an effective deterrent or whether a deviance amplification process, more akin to a labeling approach, was more likely for gang members compared with nongang members. Using propensity score matching combined with moderation techniques, the authors compared outcomes like oppositional attitudes toward conventional norms, future delinquent behavior, future peer delinquency, or likelihood of future gang joining.

The results showed only a marginal increase in gang members’ future delinquency, which, when coupled with patterns for the attitudinal and nonbehavioral outcomes, led the authors to conclude that arrest was “largely irrelevant to gang [members]” (Wiley et al., 2016, p. 21). Gang members were not deterred by arrest but also did not display significant amplification effects after arrest. These results were in stark contrast to the amplification-type patterns found for nongang members whose results, for the same set of outcomes, revealed significant direct effects of arrest.

In other recent studies on the effects of arrest on life-course outcomes for youths more generally (e.g., not necessarily specific to gangs), researchers have employed similar matching techniques, balancing samples on a host of prearrest covariates and confounders including measures of peer deviance (Liberman, Kirk, & Kim, 2014) or youth gang involvement (Widdowson, Siennick, & Hay, 2016). Statistical interactions between arrest and such peer or gang measures, however, were not accounted for in either of these analyses. The results from these studies show that arrest significantly derailed youths’ educational trajectories (Widdowson et al., 2016), increased future self-reported offending, as well as the likelihood of future rearrest (Liberman et al., 2014). The last of these outcomes occurred independent of youths’ rates or seriousness of offending, leading the authors to surmise that the distinct “secondary sanctioning” effect may arise from negative reactions and increased scrutiny by conventional others, like teachers or police (Liberman et al., 2014). Noting law enforcement discretion, these results also highlight the somewhat random nature of arrest; some youths evaded arrest despite rates of self-reported delinquency that were comparable with those of arrested youths (Liberman et al., 2014, p. 363).

Although none of these studies involved tests of moderation along indirect pathways or how labeling-related, mediation processes might differ on account of social audiences’ varying natures, such findings still inform the current project. For instance, Wiley and colleagues (2016) exploration of conditional direct effects across a gang-nongang dichotomy underscores the value of identifying factors that may moderate the impact of official intervention. We build on such work by extending tests of conditional effects along indirect pathways purported as relevant from a labeling perspective. We also extend beyond the gang–nongang dichotomy to assess potential moderating effects for gang members as well as for youths with different levels of delinquent peer association outside the gang context. Dovetailing these notions, we test whether gang membership or delinquent peer groups (for nongang members) differentially impact the effect of official intervention on future delinquency through a self-concept mechanism: self-esteem. We initially examine whether gang membership status moderates the impact of official intervention on self-esteem, and, in turn, affects future delinquent behavior. We then assess whether the moderation effects of having highly delinquent peers for nongang members are similar to the effects of being in a gang, or, if a unique “gang effect” for the impact of arrest on self-esteem and, in turn, future delinquent behavior, is found.

3 |. HYPOTHESES

Based on our rationale and review, we present hypotheses for two sets of comparisons of moderated mediation effects of arrest on self-esteem and future delinquency. For comparisons of gang members and nongang members, we hypothesize that the negative impact of official intervention on self-esteem will be less for gang members than for nongang members (hypothesis 1a). We anticipate that this, in turn, will result in an indirect effect of arrest on delinquent behavior through self-esteem that is also less for gang members than for nongang members (hypothesis 1b).

Next, having compared gang members with nongang members, we extend our model to compare gang members with youths who are not gang affiliated but who either have comparatively high delinquent peer groups or others who do not have highly delinquent peers. The findings from these analyses enable us to assess whether the conditional effect of gang membership is distinct from the effect of having similarly high delinquent peer association, simply outside the context of a gang. For these comparisons, we offer two additional hypotheses. First, that the negative impact of official intervention on self-esteem will be less for gang members than for youths with highly delinquent peer groups, as well as those with less delinquent peer groups (hypothesis 2a). We again anticipate that this, in turn, will result in an indirect effect of arrest on delinquent behavior through self-esteem that is less for gang members than for either group of nongang-affiliated youths, regardless of having highly delinquent or less delinquent peers (hypothesis 2b). If support for these hypothesized relationships is found, such results would be in line with notions of a distinct “gang effect.”

4 |. DATA AND METHOD

Data from the Rochester Youth Development Study (RYDS) are used to examine this study’s hypotheses. The RYDS is a longitudinal panel study of youth at high risk for serious delinquency and drug use that began in 1988. A high-risk sample of 1,000 middle school students was randomly selected from Rochester public schools. The seventh and eighth grade students were stratified by neighborhood tract arrest rates with the highest one third sampled with certainty. Males were also purposefully oversampled at a rate of 3:1 as males are more likely than females to engage in serious delinquent behavior (Blumstein, Cohen, Roth, & Visher, 1986). The initial resultant sample was 73 percent male and disproportionately of minority status with 68 percent African American and 17 percent Hispanic participants.

Three phases of the RYDS followed respondents from the approximate ages of 14 to 31; for the current study, we use only phase 1 data, in which youths participated in nine interviews across 6-month intervals, encompassing, on average, ages 14 to 18. One parent/guardian per youth was also interviewed during the study’s first eight waves and official data were collected from law enforcement, the Rochester City School District, and the Department of Social Services. At the end of phase 1, approximately 88 percent (n = 881) of original respondents remained in the sample. The results of comparisons of retained and nonretained youths show that retained subjects are representative of the original sample (Thornberry, 2013). In our current analyses, we use parent/guardian and youth interview data from waves 2 through 9.

4.1 |. Measures

To ensure appropriate temporal ordering of the processes hypothesized, measures are constructed at one of four time points: at wave 2, across waves 2 through 5, at wave 6, or across waves 7 through 9. Control variables, unless otherwise noted, are measured at wave 2. Measures of perceived official intervention, gang membership, and perceived peer delinquency are all constructed using data across waves 2 through 5, whereas self-esteem is estimated at wave 6, and future delinquency is calculated across waves 7 through 9 (see appendix A in the online supporting information for scale and index items²). Table 1 displays descriptive statistics of all measures for the total sample as well as by gang member and nongang member categories. For categorical variables, table 1 coefficients are interpreted as the proportion of the sample coded as 1.

TABLE 1.

Descriptive statistics: Means and t tests^a,b,c,d

Variables	Range	Total sample (N = 961)		Gang members (n = 268)		Nongang Member Categories
						All nongang members (n = 693)		High peer delinquency (n = 228)		Low peer delinquency (n = 465)
		Mean	SD	Mean	SE	Mean	SE	Mean	SE	Mean	SE
Self-reported Arrest (yes = l)e	0–1	.14	.35	.28b	—	.09	—	. 14c,d	—	.06	—
Self-esteem Scale Scoref	2.04–1.00	3.18	.43	3.14b	.03	3.21	.02	3.18d	.03	3.22	.02
Delinquency Variety Scoreg	0–18	1.79	2.62	2.92b	.19	1.35	.09	2.33c,d	.19	.87	.08
Moderator-related Measurese
Gang membership (yes = 1)	0–1	.28	.45	—	—	—	—	—	—	—	—
Perceived peer delinquencyi	1.00–3.36	1.31	.38	1.53b	.03	1.22	.01	1.55d	.02	1.06	.00
Control Measures
Prior delinquencyh	0–19	1.70	2.71	3.46b	.23	1.02	.07	2.09c,d	.17	.49	.05
Prior self-esteem h	1.78–4.00	3.08	.41	3.00b	.03	3.11	.02	3.05d	.02	3.13	.02
Ageh	11.9–16.2	14.50	.03	14.60b	.05	14.40	.03	14.50	.05	14.40	.04
Received welfare (yes = l)e	0–1	.65	—	.70b	—	.63	—	.63	—	.63	—
Gender (male = 1)	0–1	.73	—	.71	—	.73	—	.80c,d	—	.69	—
Race (non-White = 1)	0–1	.85	—	.92b	—	.82	—	.86c	—	.80	—

Abbreviations: SD = standard deviation; SE = standard error (omitted for binary variables).

^aMeans for categorical variables can be interpreted as the proportion of the sample coded as 1 for the corresponding measure. Significant t test indicators (shown for ease of presentation) provide comparable inferences as those obtained from chi-square tests of independence and/or analyses of variance that were performed as applicable.

^bStatistically significant (p < .05) t test between gang members and all nongang member categories.

^cStatistically significant (p < .05) t test between gang members and high peer delinquency (nongang) group.

^dStatistically significant (p < .05) t test between high peer delinquency and low peer delinquency (nongang) groups.

^eMeasured at waves 2 through 5.

^fMeasured at wave 6.

^gMeasured at waves 7 through 9.

^hMeasured at wave 2.

ⁱPerceived peer delinquency scores used for categorizing moderator groups are also included as a control in all model analyses.

4.1.1 |. Independent variable

Self-report arrest

Official intervention is operationalized as a binary indicator of any perceived arrest reported before wave 6. In waves 2 and 3, respondents were asked about any arrest since their last interview. Starting in wave 4, arrests were self-reported through a series of follow-up questions offered only if respondents indicated engaging in delinquent or drug-related activity since their last interview, or when provided a more general opportunity to report arrest for any reason other than those already mentioned during the interview.

Nearly 14 percent (n = 134) of the 961 respondents self-reported a perceived arrest before wave 6 when the mean sample age was approximately 16. More than 82 percent (n = 111) of these self-reports were verified in RYDS official report data.³ Among the 134 youths who perceived being arrested, the majority reported one arrest, whereas 43 percent self-reported multiple arrests during this timeframe. Given a priori assumptions about the impact of any perceived arrest experience, as well similar operationalization of official intervention in prior studies where panel data were used [e.g., see Tapia, 2011b, using National Longitudinal Survey of Youth (NLSY97)], all 134 cases of self-reported arrest before wave 6 were retained and coded as 1.

4.1.2 |. Mediating mechanism

Self-esteem

The measure of self-concept for this study is operationalized as respondents’ wave 6 self-esteem scale scores subsequent to self-reported arrest. Although a more direct measure of youths’ deviant selfidentity would be ideal, such a measure is not available in RYDS data; thus, self-esteem is used as a proxy measure of self-concept. The scale comprises nine individual items derived from Rosenberg’s (1965) self-esteem scale and is intended to approximate respondents’ emotional evaluation (or devaluation) of self.⁴ For example, items encompass feelings of satisfaction (or dissatisfaction) with oneself, self-worth (or worthlessness), and positive (or negative) regard for oneself. With higher values indicating higher self-esteem, the combined-item scale score showed strong internal consistency (Cronbach’s alpha coefficient, α = .85). To account for youths’ prior self-esteem, scores from the earliest wave at which this measure is available, wave 2, are included as a control in all analyses.

4.1.3 |. Dependent variable

General delinquency variety score

To maintain the predicted time order between youths’ perceived arrest through wave 5, subsequent self-esteem at wave 6, and future delinquent behavior, a variety score of youths’ general delinquency across waves 7 through 9 is used. Using a variety scale as a measure of future delinquency provides a more reliable and valid indicator of youths’ activities (Sweeten, 2012). Constructing variety scores is less sensitive to less serious, often higher frequency behaviors (Sweeten, 2012) while still capturing variation in delinquency prevalence; greater offense variety is found to correspond with higher offending frequency and severity (Farrington, 1973).

As calculated, the variety scores capture the number of different types of delinquent behaviors youths engaged in across waves 7 through 9 when the mean sample age was 17.5. At each wave, respondents were consistently asked about their involvement in 24 delinquent activities ranging in seriousness from truancy to assault. Each of the 24 behaviors is coded as 1 if self-reported at any time during waves 7 through 9 or as 0 if never reported across these waves. The final delinquency variety score represents a prevalence measure, or count, of youths’ self-reported engagement in up to 24 different behaviors, with higher values corresponding to greater offending variety. The average delinquency variety score for waves 7 through 9 is 1.79. To control for earlier delinquency, youths’ wave 2 variety score is included in all analyses.

4.1.4 |. Moderator measures

Gang membership moderator

A binary indicator of youths’ gang involvement is constructed using any self-nomination before wave 6, coding affirmative responses to the question since we interviewed you last time, were you a member of a street gang or posse? as 1. Approximately 28 percent of respondents (n = 268) reported gang membership during this timeframe. This percentage of self-nominated gang involvement is comparable with other high-risk samples like RYDS or samples obtained from emergent gang cities, like Rochester, NY (e.g., see Decker, Pyrooz, Sweeten, & Moule, 2014; Lahey, Gordon, Loeber, Stouthamer-Loeber, & Farrington, 1999). The use of self-nomination is also a consistent and validated method within the broader gang literature (Esbensen, Winfree, He, & Taylor, 2001; Pyrooz, Sweeten, & Piquero, 2013). The results of prior RYDS analyses have revealed that self-reported gang membership generated a near-identical list of members when compared with lists based on alternative selection criteria such as the name (verified), size of the gang, or role in the gang (Thornberry et al., 1993).

Gang member–high peer–low peer (gang–high–low) moderator

To probe further the anticipated conditional effects of youths’ delinquent peer association beyond the gang–nongang member dichotomy, a gang–high–low moderator was constructed. This moderator accounts for being gang involved or simply having a higher or lower proportion of delinquent peers (but never being gang involved) and permits assessing the conditional effects of being in a gang versus having a comparably high proportion of delinquent peers but not being gang involved. The three-group moderator also permits comparison between all nongang members and conditional effects of reporting either a higher or a lower proportion of delinquent peers. Having already established which youths were gang involved for the binary gang membership moderator, constructing the gang–high–low moderator required estimating scores of all youths’ perceived peer delinquency. The resulting scores were then used to categorize nongang members into “higher peer delinquency” or “lower peer delinquency” groups.

The measure of perceived peer delinquency is estimated across waves 2 through 5 from youths’ responses to seven items asking since the date of last interview, how many of these friends … engaged in various delinquent activities including violent behaviors, such as attacking someone with a weapon, and less serious crimes, such as stealing items worth less than $50.⁵ Response options range from most of them (4) to none of them (1) with higher average-item scores indicating perceptions of having a greater proportion of delinquent peers, or as having a more delinquent peer network. An assessment of internal consistency for the perceived peer delinquency scale revealed high reliability (α = .89).

For coding the 693 nongang members into one of two groups: high, for higher peer delinquency, or low for lower peer delinquency, the aim was to identify a high group of nongang youths who reported proportions of delinquent peers that were similar to levels of perceived peer delinquency reported by gang members. Several cutoff values were explored, and a perceived peer delinquency score of 1.25 permitted the most balanced comparison to gang member reports.⁶ The resulting gang–high–low moderator includes three categories: 268 gang members, a high group of 228 nongang members with higher peer delinquency, and a low group of 465 nongang members who reported significantly lower proportions of delinquent peers (table 1).

4.1.5 |. Control measures

The perceived peer delinquency estimates are also included as controls in all analyses, in addition to youths’ prior delinquency and self-esteem at wave 2, the earliest available measure for each. Given the measurement of self-reported arrest across waves 2 through 5, we consider wave 2 as the optimal point for preserving temporal ordering while controlling for prearrest delinquency and self-esteem covariates. Still, given the potential sensitivity of models to controls measured at waves more proximal to the mediator or dependent variables, we performed supplemental analyses using delinquency and self-esteem controls across waves 2 through 5, the findings from which are noted in the Results section.⁷ All analytic models are also conditioned on demographic measures, which include two dummy indicators to control for gender (1 = male) and race/ethnicity (1 = non-White), as well as a control for respondents’ age. For the race/ethnicity control, African American and Hispanic respondents were combined as the reference category coded as 1. The last control is a proxy measure for respondents’ socioeconomic status constructed from parent/guardian responses to questions about receiving welfare (e.g., public assistance).

4.2 |. Analytic sample

The final sample (N = 961) is approximately 72 percent male, 85 percent non-White, and nearly 65 percent reside in a household receiving public assistance. From the initial 1,000 cases in RYDS phase 1, 29 observations were removed as a result of complete nonresponse patterns and entirely unavailable data. Another 10 observations were dropped as a result of time-ordering concerns related to arrests reported at waves prior to (or contemporaneous with) self-reported gang membership.

Despite 89 percent of the sample having complete data for this study’s measures, the remaining observations presented sporadic item-level nonresponse to questions related to the control, moderator, mediator, and/or outcome measures. To retain all possible remaining observations for model estimation, multiple imputation techniques were employed and five imputed data files were created using the STATA mi impute chained procedure (StataCorp, 2013). Responses missing at random for individual items and/or composite measures were replaced for the following variables: perceived peer delinquency at waves 2 through 5, self-esteem at waves 2 through 6, and delinquency waves 2 through 9.⁸ Descriptive statistics for postimputation estimates revealed means and standard errors that were consistent with preimputation values, confirming that variable estimates were not sensitive to imputation. Descriptive and inferential analyses were performed across the five data files with resulting parameter estimates combined to allow for valid statistical inference (Rubin, 1974, 1987).⁹

4.3 |. Analytic plan

To test this study’s labeling-premised hypotheses about the how and for whom official intervention affects later self-esteem and future delinquency requires methodologies that permit investigating conditional processes, specifically differences in indirect effects as functions of a focal moderator. We use the PROCESS macro (Hayes, 2018) to analyze such conditional process models; this tool was chosen as a result of its flexibility in user-coded, regression-based modeling approaches that 1) permit assessing moderation without the traditional reliance on formal significance tests of interactions along pertinent paths and 2) allow models to be assessed as a whole as opposed to a more piecemeal or individual pathway approach (Hayes, 2018, p. 425). For instance, inferential tests of statistical significance have historically guided modeling decisions to retain (or omit) interaction components as well as interpretation of whether indirect effects are moderated (Hayes, 2015). The presence of a statistically significant interaction, however, should not lead to automatic inference of a moderated indirect effect (Fairchild & Mackinnon, 2009). On the other hand, when statistical interactions are unsubstantiated by traditional significance tests, the presence of conditional effects should not automatically be dismissed; the indirect effects still may very well be conditional by moderator values or categories (Hayes, 2015, p. 3). To this latter point, recent findings and methodological advancements across disciplines demonstrate that “capturing the dynamics of mediation” may hinge on including statistical interactions regardless of significance (VanderWeele, 2015, p. 46). The practice of including or retaining interactions regardless of statistical significance can provide useful information about observed effect heterogeneity, whereas ignoring or omitting such interactions may keep valuable information about processes of interest buried (VanderWeele, 2015).

4.3.1 |. Statistical inference and interpretation using PROCESS

An attractive feature of the PROCESS macro is that it does not require combinations of fragmented significance tests of separate moderation or mediation effects. Instead, when investigating a relationship between a moderator and an indirect (mediation) effect, PROCESS provides an “index of moderated mediation” (IMM) estimate, in which the statistical model as a whole is considered. Compared with alternative moderated mediation methods in which inferences may require interpretations of multiple tests of individual paths or model components, inferences in PROCESS are determined from a single test based on the size of the IMM estimate (Hayes, 2015, 2018).¹⁰ As such, the IMM is a “direct

quantification of the linear association” between an indirect effect and the hypothesized moderator of that effect (Hayes, 2015, p. 3); it is the formal test of hypotheses that indirect effects for any two values of a moderator significantly differ.

For an IMM estimate that is nonzero, corresponding 95th percentile bootstrapped confidence intervals provide inference into whether the conditional indirect effects for moderator categories significantly differ from one another (Hayes, 2015, 2018). Bootstrapped confidence intervals are preferred for estimating and comparing indirect effects because 1) they account for the irregular sampling distribution of the products of two regression coefficients, 2) they have been shown to offer greater statistical power for detecting indirect effects compared with alternative methods like Sobel tests, and 3) they have been shown to provide more accurate estimates, overall (Hayes, 2018, pp. 98 and 107).^11,12

4.3.2 |. Current study’s application of PROCESS

Figure 1 illustrates generalized conceptual and statistical path models that apply to this study’s analyses. Looking at figure 1a, the direct effect of arrest (X) on delinquency (Y) is interpreted as the effect of arrest on future delinquency that is independent of any intermediary impact arrest may have on self-esteem (M). The parameter estimate of this effect is marked c′ in figure 1b.

FIGURE 1.

Conceptual (a) and statistical (b) path models applicable to this study^a

^aFigure 1b displays subscript j to convey any number of values or categories (i.e., binary, multicategorical, or continuous) for a moderator (W_j) and its corresponding interaction terms (XW_j) in a given model. The k subscript encompasses all control measures (U_k) used in estimation of a mediator (M) and an outcome (Y). Further explication of parameters and notation is provided in appendix B in the online supporting information.

Next, the indirect effect is modeled as the effect of arrest (X) on delinquency (Y) that is transmitted through self-esteem (M), with corresponding parameter estimates a_X and b_M shown in figure 1b. Further distinguishing this effect as conditional is the path from moderator (W), in which the effect of arrest (X) on self-esteem (M) is modeled as varying linearly with values of a moderator (W). The parameter estimates $a_{W_j}$ and $a_{{\mathit{XW}}_j}$ in figure 1b permit application of this generalized model to either the binary and multicategorical moderators tested in this study.

In our first analysis, we test the binary gang membership moderator (W_j), initially coded for all gang members (1) and all nongang members (0). Next, we analyze a model using the three-category gang–high–low moderator (Wj), initially coded for all gang members (1) and nongang youths with highly delinquent peers (0) or less delinquent peers (−1). All analyses are further conditioned on the set of seven control variables shown as U_k in figure 1a.

With the parameters shown in figure 1b, IMM estimates are calculated as the product of two pathway coefficients: the interaction term $a_{{XW}_j}$ along the path from arrest to self-esteem multiplied by the b_M coefficient linking self-esteem to future delinquency. The resulting product represents the weight of a focal moderator in the function responsible for allocating the size of an indirect effect (Hayes, 2015, 2018, p. 425; Morgan-Lopez & MacKinnon, 2006) or, said differently, how much the moderator affects the impact of arrest on delinquency through self-esteem. Thus, the IMM quantifies the difference in indirect effects between moderator values (or categories). For IMM estimates that are nonzero with corresponding confidence intervals that also do not include zero, inference can be more confidently made that an indirect effect is “systematically larger or smaller [depending on the sign of the coefficient] for some moderator [categories] than others” (Hayes, 2015, p. 4; Hayes, 2018). Analytic results, including conditional indirect effect estimates, corresponding inferential statistics, and interpretations of the findings are reported in the following section.¹³

5 |. RESULTS

The results from bivariate tests of all model variables consistently revealed anticipated relationships that were statistically significant and in expected directions (see table B-1 in the online supporting information for a correlation matrix). For several measures, basic inferential analyses (e.g., t tests, chisquare tests, analyses of variance) showed significant differences in means or proportions (for categorical variables) between gang members and nongang members, and between nongang youths grouped by peer delinquency levels (e.g., high vs. low nongang groups). For instance, the observed likelihood of arrest significantly differed from what was expected across gang and nongang groups, with a significantly higher percentage of gang members reporting arrest (28 percent, n = 74) compared with all other moderator categories (see table 1).

5.1 |. Results for gang membership moderator analyses

The findings from our first set of analyses offered support for both hypotheses pertaining to the gang membership moderator; the effects of arrest on self-esteem were anticipated as being dependent on youths’ gang membership status, with gang members’ self-esteem being less negatively impacted than that of nongang members (hypothesis 1a). The results showed support for this hypothesis; arrest significantly reduced nongang members’ self-esteem by .15 units while having no significant impact on gang members’ self-esteem (−.01) (table 2, panel A). Figure 2 illustrates the significant differences in effects of arrest on self-esteem for gang versus nongang members.

TABLE 2.

Conditional indirect effects and unconditional direct effect of arrest on self-esteem and delinquency by gang membership moderator^a,b

Panel A (a paths): Conditional effects of arrest on self-esteem mechanismc
		Moderator Reference Categoryd
		Gang members			Nongang members
Paths	Variables	b	SE	t	b	SE	t
ax	Arrest	−.01	.05	−.14	−.15**	.05	−2.93
$a_{W_j}$	Gang membership moderator	−.04	.03	−1.07	.04	.03	1.07
$a_{{XW}_j}$	Arrest X gang membership	−.15*	.07	−2.03	.15*	.07	2.03

Panel B (b path): Unconditional effect of self-esteem mechanism on delinquencye
Path	Variable	b	SE	t
bm	Self-esteem	−.66**	.19	−3.42

Panel C (aXb paths): Conditional indirect effects of arrest on delinquencye
	Moderator Reference Categoryd
	Gang members				Nongang members
			95% CI				95% CI
Variable paths	CIE	SE	LLCI	ULCI	CIE	SE	LLCI	ULCI
Arrest > Self-esteem > Delinquency	.01	.04	−.07	.08	.10*	.05	.02	.22

Panel D (axwb paths): Index of moderated mediation for conditional indirect effects
	Gang membersd				Nongang membersd
			95% CI				95% CI
Difference in CIE by Category	IMM	SE	LLCI	ULCI	IMM	SE	LLCI	ULCI
Gang vs. nongang member		−.10*	.06	−.24	−.01	.10*	.06	.01	.24

Panel E (c′ path): Unconditional direct effect of arrest on delinquencye
Path	Variable	b	SE	t
C′	Arrest	.86***	.23	3.78

Note: Coefficients are unstandardized; Bootstrap (n = 5,000).

Abbreviations: SE = standard error; CIE = conditional indirect effect; CI = confidence interval; LLCI = lower level 95% confidence interval threshold; ULCI = upper level 95% = confidence interval threshold; IMM = index of moderated mediation.

^aTotal sample (N = 961), gang members (n = 268), nongang members (n = 693).

^bAll models conditional on all demographic, prior self-esteem, prior delinquency, and perceived peer delinquency controls (coefficients shown in appendix B, table B-2, in the online supporting information).

^cPanel A model fit for self-esteem: R² = .24; Model significance: F(10,950) = 30.37^***.

^dModerator reference category coded as lower value (0) for repeat analyses performed for gang members = 0 and nongang members = 0. Conditional indirect effect where W = 0 calculated as ((a_xb) + ((a_xwb)W)) or (a_xb). IMM estimate a_xwb calculated when W = 1

^ePanels B through E model fit for delinquency: R² = .29; Model significance: F(9,951) = 42.90^***.

^†p< .10;

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed).

FIGURE 2.

Conditional effects of arrest on self-esteem by gang membership status (N = 961)

We also tested the related hypothesis of predicted differences in indirect effects of arrest on delinquency through the self-esteem mechanism by gang membership. These processes were again expected to impact gang members less than nongang members, meaning a significantly smaller indirect effect for gang members (hypothesis 1b). Parameter coefficients pertinent to estimating indirect effect differences show that, when holding constant the effect of prior arrest, wave 6 self-esteem had a significant inverse effect on youths’ delinquency at waves 7 through 9 (table 2, panel B). For every one unit reduction in self-esteem, youths’ delinquent activities increased by .66 units. Upon including the impact of prior arrest on self-esteem by gang membership status, the conditional indirect effect estimates for gang members and nongang members diverged as hypothesized (panels C and D). For nongang members, the significant negative impact of arrest on self-esteem accounted for an indirect increase in future delinquency (CIE = .10*, whereas for these same labeling processes, gang members’ estimate was far smaller and statistically nonsignificant (CIE = .01).

Historically, the statistically significant interaction terms found in this model indicate that the effect of arrest on self-esteem varies linearly by gang membership status (e.g., panel A: $a_{{XW}_j}$ = .15*); however, when using PROCESS, the corresponding index of moderated mediation (IMM) is the formal test for differences in indirect effect estimates between moderator groups. The resulting IMMs and 95 percent bootstrapped confidence intervals support the interpretation that the conditional indirect effect estimate of arrest on delinquency through self-esteem significantly differed for gang members versus nongang members and was significantly less for gang members (IMM = −.10*, panel D). The unconditional direct effect of arrest on delinquency was also statistically significant (panel E).¹⁴ Holding wave 6 self-esteem constant, youths who were arrested, compared with those who were not, reported an .86 unit increase in future delinquent activities. Since conditional effects for the direct relationship of arrest on delinquency were not hypothesized, direct effect results were identical in analyses designed to test the gang–high–low moderator; however, indirect effect comparisons between these three moderator categories revealed interesting conditional patterns that were not entirely as hypothesized.

5.2 |. Results for gang-high-low moderator analyses

The findings from our second set of analyses offered partial support for the two hypotheses pertaining to tests of the effects of arrest on self-esteem and delinquency that compared gang members with two groups of youths who either reported having more delinquent peers (high category) or fewer delinquent peers (low category) simply outside the context of gangs. Although no hypotheses were specified for anticipated differences between the high and low nongang member groups, estimates related to such comparisons are also shown in table 3.

TABLE 3.

Conditional indirect effects and unconditional direct effect of arrest on self-esteem and delinquency by gang–high–low moderator^a,b

Panel A (a paths): Conditional effects of arrest on self-esteem mechanismc
		Moderator Reference Categoryd
		Gang			High			Low
Paths	Variables	b	SE	t	b	SE	t	b	SE	t
ax	Arrest	−.01	.05	−.08	−.04	.07	−.63	−.26***	.07	−3.60
$a_{W_j}$	Moderator Comparison Category
	Gang members	—	—	—	−.02	.04	−.64	−.04	.04	−1.04
	High delinquency peers (nongang)	.02	.04	.64	—	—	—	−.01	.03	−.36
	Low delinquency peers (nongang)	.04	.04	1.04	.01	.03	.36	—	—	—
$a_{{XW}_j}$	Arrest × Moderator Category
	Arrest × gang members	—	—	—	.04	.09	.47	.26***	.09	2.91
	Arrest × high delinquency peers	−.04	.09	−.47	—	—	—	.22*	.10	2.14
	Arrest × low delinquency peers	−.26***	.09	−2.91	−.22*	.10	−2.14	—	—	—

Panel B (b path): Unconditional effect of self-esteem mechanism on delinquencye
Path	Variable	b	SE	t
bM	Self-esteem	−.66**	.19	−3.42

Panel C (aXb paths): Conditional indirect effects of arrest on delinquencye
	Moderator Reference Categoryd
	Gang				High				Low
			95% CI				95% CI				95% CI
Variable paths	CIE	SE	LLCI	ULCI	CIE	SE	LLCI	ULCI	CIE	SE	LLCI	ULCI
Arrest > Self-esteem > Delinquency	.00	.04	−.07	.08	.03	.06	−.09	.15	.17*	.08	.05	.35

Panel D (axwb paths): Index of moderated mediation for conditional indirect effectse
	Moderator Reference Categoryd
	Gang				High				Low
			95% CI				95% CI				95% CI
Difference in CIE by category	IMM	SE	LLCI	ULCI	IMM	SE	LLCI	ULCI	IMM	SE	LLCI	ULCI
Gang members	—	—	—	—	−.03	.07	−.17	.11	−.17*	.08	−.36	−.04
High delinquency peers	.03	.07	−.11	.17	—	—	—	—	−.15*	.09	−.36	−.01
Low delinquency peers	.17*	.08	.04	.36	.15*	.09	.01	.36	—	—	—	—

Panel E (c’ path): Unconditional direct effect of arrest on delinquencye
Path	Variable	b	SE	t
c’	Arrest	.86***	.23	3.78

Note: Coefficients are unstandardized; Bootstrap (n = 5,000).

Abbreviations: SE = standard error; CIE = conditional indirect effect; CI = confidence interval; LLCI = lower level 95% confidence interval threshold; ULCI = upper level 95% confidence interval threshold; IMM = index of moderated mediation.

^aTotal sample (N = 961), gang members (n = 268); high peer delinquency (nongang) (n = 228); low peer delinquency (nongang) (n = 465).

^bAll models conditional on all demographic, prior self-esteem, prior delinquency, and perceived peer delinquency controls (coefficients shown in appendixB,tableB-2, in the online supporting information).

^cPanel A model fit for self-esteem: R² = .25; Model significance: F(12,948) = 25.78^***.

^dThree repeat analyses performed using indicator coding for moderator reference category coded as lowest value (W_j = −1). Conditional indirect effect where W_j =0 or1 otherwise calculated as $((a_X{b}_M)+((a_X{}_{W_j}{b}_M)W_j)).$

^ePanels B through E model fit for delinquency: R² = .29; Model significance: F(9,951) = 42.90^***.

^†p< .10;

^*p < .05;

^**p < .01;

^***p < .001 (two-tailed).

The impact of arrest on self-esteem was again anticipated to vary significantly across moderator categories; the negative impact of arrest on self-esteem was expected to be less for gang members than for either the high or the low group (hypothesis 2a). Only partial support for this hypothesis was found, however, as patterns from comparisons of the gang–high groups were inconsistent with patterns found in comparisons of the gang–low groups. Significant moderation was found only in comparisons of the gang–low groups (table 3, panel A), in which arrest significantly reduced the low group’s self-esteem by .26 units but again had no significant impact on gang members’ self-esteem (−.01).

Unanticipated was the null finding of any significant impact of arrest on self-esteem for the high peer delinquency group of nongang members (−.04), a result that entirely diverged from the pattern observed for the low group. Instead, the high group’s patterns primarily mirrored those found for the gang member category; these patterns are illustrated in figure 3.

FIGURE 3.

Conditional effects of arrest on self-esteem by gang–high–low moderator (N = 961)

Taking the labeling process comparisons one step further, we tested our final hypothesis for differences in the indirect effects of arrest on delinquency through the self-esteem mechanism across the three moderator groups. The indirect effect for gang members was again expected to be significantly smaller than indirect effect estimates for both the high and the low groups (hypothesis 2b). When the pathways from arrest to self-esteem and self-esteem to delinquency were taken together, however, our hypothesized differences in conditional indirect effects were, again, only partially supported.

For the high group of nongang members, no significant conditional indirect effect was found (CIE = .04); a corresponding nonsignificant index of moderated mediation (IMM = .03) confirmed statistically nondistinct indirect effect estimates between the high group and the gang group (table 3, panels C and D). These results were not as hypothesized. Conversely, and aligning with our hypotheses, results for the low group showed a significant negative impact of arrest on self-esteem that, in turn, accounted for an indirect increase in future delinquency (CIE = .17*); for gang members, much like for the high group of nongang members, this effect was entirely nonsignificant (CIE = .00).

Altogether, the only significant differences across the three conditional indirect effect estimates were found in comparisons with the low group (table 3, panel D). For this group of nongang youths who reported fewer delinquent peers, the indirect effect of arrest on delinquency through self-esteem was significantly greater than the effects for gang members (IMM = .17*) and the effects for youths with highly delinquent peer groups outside of the gang context (IMM = .15*). In summary, our findings do not offer support for a distinct “gang effect” on conditional labeling processes; our results do indicate, however, a more general “delinquent peer effect”. These findings serve to demonstrate the relevance of assessing the potential impact of moderating factors on the mechanisms and processes posited by labeling theory.

6 |. DISCUSSION

At the outset of this article, we provided a theoretical rationale for why observed effects of arrest on self-esteem and future delinquency may differ by youths’ gang involvement status or, for nongang members, by level of peer delinquency. Our rationale and subsequent hypotheses were informed by the symbolic interactionist perspective, in which reactions from a primary social audience serve as important references from which an actor conceptualizes and even sometimes alters his or her sense of self. As actors come to view themselves as reflections of others’ appraisals of them, the nature of their primary social audience also matters, as does the audience’s relevance to the situation. Audiences’ reactions to an arrest are likely to influence the actor’s self-concept further, but these reactions and their influence may significantly vacillate between stigmatizing or supportive. For this study, we hypothesized that such differences would depend on youths’ gang involvement or, for nongang members, on having higher or lower proportions of delinquent peers.

Our findings fully supported hypotheses 1a and 1b. The effects of official intervention on self-esteem were not only significantly less for gang members compared with nongang members but arrest was also found to have no significant influence whatsoever on gang members’ self-esteem. For nongang members, however, results revealed that arrest was in fact detrimental to self-concept, significantly reducing self-esteem, which in turn increased nongang members’ future delinquency.

We then wanted to examine whether the impact of gang membership was simply a function of gang members having mostly delinquent friends. To do so, we used a multicategorical moderator for gang members and two groups of nongang members with more or fewer delinquent peers. We could then compare the moderating effects of gang membership with the effects of peer delinquency across the two groups of nongang members. The results from these analyses provided partial support for our latter two hypotheses.

For both hypotheses 2a and 2b, we anticipated that being gang involved would provide a greater “buffer” to the impact of arrest on self-esteem and delinquency, with stronger negative effects expected for the two nongang-involved groups. For nongang members with more delinquent peers, however, the indirect patterns entirely mimicked those observed for gang members and diverged entirely from patterns found for youths with fewer delinquent peers. Instead, only nongang members with lower risk peers followed the anticipated indirect pathways posited by labeling theory; arrest negatively impacted their self-esteem mechanism, which in turn led to increased future delinquency. The self-esteem of nongang members with higher risk peer groups, though, was entirely unaffected, following patterns observed for gang members.

Our results align with those of prior research and with our theoretical rationale in several regards. First, for gang members, and even nongang members with higher risk peer groups, the reduced impact of arrest on self-esteem is what we anticipated, given the salience of primary reference groups that are highly delinquent or gang specific. For example, from qualitative interviews of 20 previously arrested young adults from high-risk communities, Hirschfield (2008) concluded that higher risk, socially disadvantaged youths, perhaps as a result of arrest being part of their daily lives and social contexts, were often insulated from negative labeling effects on self-concept. With few respondents reporting loss of self-esteem or “enduring shame” from their arrests, arrest simply did not carry the same stigmatizing effects that it might in other, perhaps more conventional, social contexts (Hirschfield, 2008).

Upon disaggregating the singular nongang member group into separate categories by level of delinquent peer association, the only distinct gang effect was found in comparisons between gang members and lower risk nongang members. For the higher risk (in terms of delinquent peers) nongang members, the effect sizes and patterns were statistically indistinguishable from those of gang members. Therefore, in terms of the impact of arrest on self-esteem and subsequent delinquency, being in a gang is not more insulating than being in a peer group that is highly delinquent. We anticipated that it would be. Findings from prior research, for example, point to an effect of gang membership on delinquent behavior that is over and above an effect of simply being a member of a highly delinquent peer group (e.g., see Battin et al., 1998; Thornberry et al., 2003). The findings from this study, however, show that individuals with delinquent social networks, gang related or otherwise, are similarly insulated from the indirect effects of arrest on self-esteem and delinquent behavior. Recognizing the importance of interactions with “others,” along with the significant role that social audiences play in actors’ views of themselves and their subsequent behaviors are essential for understanding the relative effects of official intervention.

On a broader theoretical level, our results underscore the call for an examination of conditions under which the label will have adverse effects (Paternoster & Iovanni, 1989). We should not expect everyone to react to an arrest or any other label similarly. Reactions to the label may vary by demographic factors (Ageton & Elliott, 1974; Bernburg & Krohn, 2003), the presence of socially supportive networks (Dong & Krohn, 2016), and other potential conditioning variables. When conditioning factors are not taken into account, it is not surprising that support for the central hypotheses of labeling theory is not found.

6.1 |. Limitations

The data used in this study are from Rochester, NY, a city that has a high crime rate and, at the time of data collection, was considered an “emergent gang city.” Rochester gangs were less likely to be well organized in comparison with more traditional gangs in larger cities with longer histories of gang presence. Thus, our results may not be generalizable to more traditional gangs. For that reason, we consider ours a conservative examination of the impact of gang membership on the relationship between arrest and self-esteem and delinquency. If membership in emergent gangs served to moderate the impact of arrest on self-esteem, then membership in more traditional gangs, with the expectation that members would be more embedded in such networks, would be even more likely to protect against the negative impact of official intervention.

Another limitation of this study is that we could not examine the intersectionality of gender and/or race and ethnicity with our hypothesized processes, or whether the observed relationships operated differently for males and females or across varying racial and ethnic groups. Although sex and race were controlled for, separate analyses requiring additional interaction effects for males and females or by racial/ethnic groups could not be reliably examined to determine whether different patterns would emerge. Our data had an insufficient number of female respondents who self-reported arrest and were also categorized as gang involved or as having highly delinquent (nongang) peer groups; females comprised nearly 20 percent of the higher peer delinquency group but only accounted for one self-reported arrest in this category. The results of supplementary analyses involving a male-only sample were substantively similar to those presented but did reveal some reduced statistical power (e.g., females comprised nearly 30 percent of gang members), informing our decision to retain the entire analytic sample of 961 in our final models. In future studies with data that permit such tests, researchers should aim to assess the intersectionality of gender or racial/ethnic differences with similar conditional labeling processes.

Furthermore, our decision to use a self-report measure of arrest rather than official reports was twofold. First, across the wave 2 through 5 timeframe, there were more arrests perceived and selfreported than officially documented. Second, not only are the two arrest measures highly correlated (Krohn, Lizotte, Phillips, Thornberry, & Bell, 2013), but our choice is further justified by the consideration that, from a labeling perspective, whether actors thought they had been arrested is the more pertinent factor for influencing self-esteem. To assess whether our choice of the self-report measure influenced our observed patterns, analyses were replicated using official reports and similar patterns were found, although statistical power was a concern. Future studies in which data allow for additional model specifications, for instance, by official intervention types (e.g., conviction or police contact) or by offense types (e.g., violent versus nonviolent) could further contribute to understanding the conditional processes tested in this study.

Finally, as mentioned in the measures section, another potential limitation is our use of a global measure of self-esteem as a proxy for self-concept. Some scholars have suggested that results may differ if one was to break the concept down into different dimensions, like actively aware, mindful perceptions of “self” versus less voluntary, more automatic views of “self” (Ostrowsky, 2010). The current data did not permit testing such dimensional self-esteem measures, so we could not explore this possibility. The data also did not include measures that allowed for direct operationalization of symbolic interactionst constructs like reflected appraisals or reference groups, thus, requiring grounded assumptions about the labeling processes and about the peer context as a relevant social audience for our sample of mid-to-late adolescents. Without direct measures of the social visibility of official intervention, or measures that ensured an actor’s peer audience learned about an arrest, we must assume that the anticipation of “others” learning about an official reaction to delinquency (e.g., arrest) is enough to induce the expected labeling processes (Winnick & Bodkin, 2008). In these regards, again, ours is a more conservative test of such processes. Future studies comprising data where measurements of different dimensions of self-esteem, alternative measures of self-concept or deviant identity, and/or additional symbolic interaction-related constructs are available could shed more light on the processes tested herein.

7 |. SUMMARY AND CONCLUSION

In this study, we set out to offer and test a rationale for how and whom labeling processes might differ, specifically focusing on the moderation of these processes by youths’ gang involvement or highly delinquent (nongang-related) peer groups. Overall, the results aligned with our theoretical rationale, in which we argued that the social audiences most relevant to gang members, and even youths with highly delinquent peer groups but not gang involved, would likely shield them from the stigma theoretically anticipated as arising from arrest and negatively impacting subsequent self-esteem. Any postarrest reduction in self-esteem was only found for youths with lower delinquency peer groups, or, by the logic of our rationale, youths whose relevant social audience did not buffer such negative effects. By further probing the patterns found in comparisons of gang members with nongang members and parsing out differences across nongang members’ levels of peer delinquency, through this study, we provide a more illustrative examination of the effects of arrest on self-esteem and delinquency. We also contribute to research aimed at identifying features of gangs that may further distinguish them from, or show they are more akin to, other peer contexts. Finally, in conjunction with our findings of moderated mediation processes by which arrest impacted future delinquency through self-esteem, the direct effect of arrest was also found to have a significant increase on future delinquency. Altogether, such results can inform prevention and intervention efforts, in which broad scale arrest tactics may be counterintuitive to crime reduction and ensnare youths at lower risk for future delinquency, along with gang members and youths with similarly high risk for future delinquent behavior.

AUTHOR BIOGRAPHIES

Molly Buchanan is an assistant professor at Marist College. She earned her Ph.D. in criminology, law, and society from the University of Florida. Her primary research interests are in developmental and life-course criminology, sanction effects, influences of risk and protective factors, and evidence-based practices.

Marvin D. Krohn is a professor in the Department of Sociology and Criminology & Law at the University of Florida. He is primarily interested in developmental approaches to the explanation of delinquency, drug use, and crime.