Gang Influence: Mediating the Gang–Delinquency Relationship With Proactive Criminal Thinking

Abstract

Controlling for basic demographic variables, parental knowledge, prosocial peer associations, and precursor measures of each outcome, the current study sought to compare two putative intervening mechanisms for the gang affiliation–participant delinquency relationship: a social learning mechanism (proactive criminal thinking) and a self-control mechanism (reactive criminal thinking). The two mechanisms were explored in 3,136 (1,519 male, 1,612 female) early adolescents (mean age = 12.14 years) from the Gang Resistance Education and Training (G.R.E.A.T.) study. A three-wave path analysis of Waves 1 to 3 of the G.R.E.A.T. study revealed a significant social learning pathway (gang affiliation → proactive criminal thinking → delinquency) and a nonsignificant self-control pathway (gang affiliation → reactive criminal thinking → delinquency). These findings were then replicated using data from Waves 4 to 6. From these results, it is concluded that gang affiliation may increase future delinquency by providing youth with increased opportunities to learn proactive criminal thinking.

Keywords

social learning theory proactive criminal thinking reactive criminal thinking youth gangs

Youth gang involvement is a major public safety concern for many law enforcement agencies in the United States. With over one million juvenile gang members in the United States (Pyrooz & Sweeten, 2015), the scope and seriousness of the problem has attracted attention at the local, state, and federal levels. A nationally representative survey of over 2,500 U.S. law enforcement agencies (1996-2012), in fact, revealed a sharp decline in youth gang activity between 1996 and 2001, followed by a gradual increase between 2001 and 2005, and a leveling off at this heightened level between 2006 and 2012 (National Gang Center, 2013). One third of the law enforcement agencies participating in this survey indicated that they had a serious gang problem in their jurisdiction, with larger cities and suburban areas experiencing greater problems than smaller cities and rural communities. The range of criminal activity engaged in by these gangs run the gamut, from minor delinquency, to drug dealing, to serious violence. Some fear that early gang affiliation could pave the way for future individual delinquency and even adult criminality (Krohn & Thornberry, 2008). The purpose of the current investigation was to identify and test a putative mechanism that might explain the temporal relationship between early gang affiliation and subsequent delinquent behavior.

The Gang–Delinquency Relationship

Youth gang affiliations have long been considered cardinal antecedents of delinquent behavior. Although there is still a great deal of uncertainty as to the nature of this relationship, several explanations have been proposed. One of the principal arguments used to explain the effect of gang affiliation on delinquency is that hanging out with gang members means that one’s peer network is saturated with a high degree of delinquency and that one learns from those with whom one associates (Klein & Maxson, 2006; Krohn & Thornberry, 2008). In an early study on this issue, Battin, Hill, Abbott, Catalano, and Hawkins (1998) compared youth gang members with non-gang members, some of whom did and others of whom did not have delinquent friends. The results indicated that gang members engaged in the greatest amount of both official and self-reported offending, whereas non-gang members with delinquent friends engaged in a moderate degree of official and self-reported offending and non-gang members without delinquent friends engaged in the least amount of official and self-reported offending. The authors of this study noted that gang membership seemed to amplify violent offending, but had little effect on minor nonviolent offending.

The results of the Battin et al. (1998) study imply that gang affiliation may lead to delinquency for reasons other than negative peer association, although it could still be argued that gang members had more delinquent peer associations than non-gang youth with delinquent friends. Using 14 waves of data from the Rochester Youth Development Study, Dong and Krohn (2016) sought to tease out the individual contributions of gang affiliation and deviant peer associations by performing a dual trajectory modeling analysis. The results of their analysis revealed that while measures of gang affiliation and perceived peer delinquency were clearly related, they coalesced around distinct constructs and appeared to operate along different lines. Whereas gang affiliation was associated with a rise in violent crime and police arrest extending well into adulthood, peer delinquency was associated with a rise in general delinquency and drug use and a greater likelihood of crime desistance in late adolescence. Dong and Krohn (2016) speculate that the organizational structure, group dynamics, culture, and norms found in youth gangs make them qualitatively distinct from delinquent peer associations and unsupervised routine activities with friends.

Controlling for Peer Delinquency and Routine Activities

In their study on gang affiliation and delinquent peer associations, Dong and Krohn (2016) controlled for the effects of perceived delinquent peer associations and unstructured routine activities on gang affiliation. Controlling for unstructured routine activities, defined as hanging out with friends absent adult supervision, and delinquent peer associations can be important because research indicates that involvement in unstructured and unsupervised routine activities with peers can have a facilitative effect on future delinquency independent of perceived peer delinquency (Hoeben & Weerman, 2016; McGloin & Shermer, 2009). Gangs themselves would appear to have the capacity to increase routine activities for crime. In a study on the routine activities of street gangs in Chicago circa 1959-1962, Hughes and Short (2014) determined that youth gangs contributed to delinquency and violence, in part, by affecting the routine activities of their members and increasing opportunities for unstructured and unsupervised peer socializing. There is also evidence that disengaging from gangs and desisting from crime is associated with decreased involvement in unstructured routine activities with delinquent peers and fellow gang members (Sweeten, Pyrooz, & Piquero, 2013).

Hughes and Short’s (2014) observation that the peer groups of gang members are often composed of other gang members and that gangs increase routine activities for crime implies that gang affiliations and peer associations could be overlapping constructs in gang-affiliated youth. To the extent that gang-affiliated youth consider fellow gang members their principal peer group and arrange their routine activities around the gang, controlling for peer delinquency and peer routine activities could remove meaningful variance from the gang affiliation variable. For this reason, peer delinquency and peer routine activities were not controlled in the current study. Even so, certain facets of peer relations need to be controlled when conducting research on gang affiliations. Accordingly, association with prosocial peers, a peer variable arguably less confounded with gang affiliation than either peer delinquency or routine activities with peers, was included in the present study as a control variable. The results of several investigations indicate that association with prosocial peers can protect youth against negative peer influences, including gang affiliations (Carson, 2013; Katz & Fox, 2010).

Putative Protective Factors

Associating with prosocial peers could be considered a protective factor in the sense that it has the potential to decrease a child’s vulnerability to negative social influences by insulating the child from the nefarious effects of potent risk factors like gang affiliation and delinquent peer associations. Protective factors derive from a resilience or strength perspective and emphasize the positive aspects of a situation rather than the negative aspects. Kupersmidt, Coie, and Howell (2004) discuss a range of school-based interventions designed to bring troubled and gang-related youth into contact with positive role models and conventional peers instead of isolating them from positive interpersonal relations by subjecting them to isolation through the more commonly traveled pathways of detention, classroom segregation, and out-of-school suspension. It should be noted that for the purposes of the current article and investigation, protective factors will be defined in a general way, without reference to their degree of interaction with various risk factors (see Wanamaker, Jones, & Brown, 2018). It should also be noted that because their protective effect has yet to be verified, these variables will be referred to as putative or hypothesized protective factors.

Parental knowledge is another putative protective factor with relevance to research on gang affiliation and delinquency. Research indicates that the more knowledge parents have of their child’s activities and friendship networks, the lower the child’s odds of becoming involved in delinquent activities (Lahey, Van Hulle, D’Onofrio, Rodgers, & Waldman, 2008) and that this effect operates through knowledge of the child’s activities rather than through supervision or monitoring (Stattin & Kerr, 2000). In a recent study examining the effects of both parental knowledge and unsupervised routine activities on crime, Walters (2018c) determined that unsupervised routine activities mediated the relationship between parental knowledge and delinquency. Hence, parental knowledge reduced the child’s involvement in unstructured routine activities, resulting in lower levels of delinquent behavior. In a study examining parental knowledge and gang affiliation in Chinese youth, Pyrooz and Decker (2013) discovered that parental knowledge of a child’s friendship networks was higher in the parents of non-gang members than in the parents of gang members and that such knowledge correlated negatively with self-reported offending on the part of the child.

Social/Self-Control Versus Social Learning

Several researchers in the field of gang studies have sought to frame the relationship between gang affiliation and delinquency within a social or self-control context. Cepeda, Saint Onge, Nowotny, and Valdez (2016), for instance, employed Hirschi’s (1969) original social control model and the four elements of his social bond (attachment, commitment, involvement, and belief) to explain the connection between gang membership and delinquent behavior. Extrapolating from an analysis of 160 Mexican American adolescent gang members, Cepeda et al. concluded that informal control mechanisms like parents, schools, and prosocial peers were largely ineffective in deterring long-term gang involvement in these individuals. It has also been noted that gang membership can weaken the social bond between gang-affiliated youth and conventional institutions and people, while strengthening the social bond with antisocial institutions and people (Weerman, Lovegrove, & Thornberry, 2013). The current study uses another feature of control theory, Gottfredson and Hirschi’s (1990) low self-control construct, which has been linked to gang affiliation (Chapple & Hope, 2003) and has been found to give rise to cognitive impulsivity or reactive criminal thinking (RCT; Walters, 2017c), as one of two possible explanations of the gang–delinquency relationship. The main premise of Gottfredson and Hirschi’s (1990) general theory of crime is that poor parenting gives rise to low self-control, which, in turn, gives rise to delinquency. One question the current study sought to answer was whether a construct related to low self-control, reactive (impulsive, irresponsible) criminal thinking, mediates the gang affiliation–delinquency relationship.

Social learning theory affords an alternative explanation of the gang–delinquency relationship. As Sutherland (1947) and Akers (1998) have argued, individuals not only learn the techniques of crime from those who have previously been involved in crime, but they also learn attitudes conducive to crime, also known as definitions favorable to violation of the law. Adopting a social learning perspective on crime acquisition, Walters (2015) contends that delinquency is learned through association with other law breakers in a process mediated by proactive (planned, calculated) criminal thinking. In a series of longitudinal studies, Walters (2015, 2016, 2017b) has demonstrated that proactive criminal thinking (PCT) mediates the peer influence effect. In one of these studies, RCT was found to mediate the peer selection effect, which runs from participant delinquency to peer delinquency, but not the peer influence effect, which runs from peer delinquency to participant delinquency (Walters, 2016). If PCT is what youth acquire when they interact with delinquent peers then it may also be true that this is what they learn in their interactions with a criminal youth gang. It was consequently reasoned that if gang influence is similar to peer influence then the gang–delinquency relationship should be mediated by PCT or cognitive insensitivity but not by RCT or cognitive impulsivity.

The Current Study

As the preceding literature review suggests, PCT and RCT mediate important criminological relationships (Walters, 2016). Even with a significant amount of overlap between these two dimensions of criminal thought process, proactive and reactive criminal thinking have been found to correlate differentially with various outside criteria (Walters, 2017a). This premise was further tested in the current study using the gang affiliation–delinquency relationship as the outside criterion. It was predicted that PCT would mediate the gang affiliation–delinquency relationship and RCT would not. Hence, in comparing a social learning interpretation of the gang affiliation–delinquency relationship—where PCT mediates the gang affiliation–delinquency relationship—to a self-control interpretation—where RCT serves as the mediator—the social learning interpretation was presumed to be superior to the self-control interpretation.

The research question addressed in the current study asked whether a self-control mechanism (RCT) or a social learning mechanism (PCT) was principally responsible for mediating the relationship between gang affiliation and delinquency. Using the comparison pathways approach (Walters, 2018a) and a causal mediation regression path analysis (Hayes, 2013), it was predicted that the target pathway (the pathway that ran through PCT and was hypothesized to be significant) would achieve significance and the control pathway (the pathway that ran through RCT and was hypothesized to be nonsignificant) would not. Parental knowledge of child friendship networks and prosocial peer associations, putative protective factors that are relevant to the peer influence effect and which may also be relevant to the gang influence effect, were controlled in the current investigation, along with age, sex, and race.

In the current sample of early adolescents, many of whom had no prior criminal involvement, proactive and reactive criminal thinking may seem too strong of terms. As a result, cognitive processes that serve as developmental antecedents to proactive and reactive criminal thinking, cognitive insensitivity (insensitivity to the feelings and rights of others) in the case of PCT and cognitive impulsivity (failure to maintain a consistent train of thought) in the case of RCT (Walters & Espelage, 2018), periodically appear alongside the criminal thinking style terms in the current article. Also, because gang affiliations tend to be intermittent (Pyrooz, Turanovic, Decker, & Wu, 2016) and mediation analysis characteristically produces small effects (Hayes, 2013; Preacher, 2015), the main results obtained using the first three waves of the Gang Resistance Education and Training (G.R.E.A.T.) study were cross-validated using the last three waves of G.R.E.A.T. data.

Method

Participants

The sample for the current study consisted of 3,136 (1,519 male, 1,612 female, 5 unspecified) youth from the longitudinal portion of the G.R.E.A.T (Esbensen, 2002) project. The G.R.E.A.T project is a school-based prevention program taught by law enforcement officers, with an overall objective of immunizing children and adolescents against delinquency, violence, and gang membership. Longitudinal data were collected on convenience samples of early adolescent children present at school on the day the surveys were administered. All respondents were middle school students attending schools in six U.S. cities (Philadelphia, Pennsylvania; Portland, Oregon; Phoenix, Arizona; Omaha, Nebraska; Lincoln, Nebraska; and Las Cruces, New Mexico) between 1995 and 1999. Passive parental consent was obtained for the initial survey and active parental consent was obtained for all subsequent waves of the longitudinal study. The use of these secondary data for research purposes was approved by the Kutztown University Institutional Review Board.

Students were included as participants in the current study if they had complete data on at least one of the four principal variables from the main analysis (i.e., Wave 1 gang affiliation, Wave 2 PCT, Wave 2 RCT, and Wave 3 delinquency). This sample represents 87.9% of all students who completed some portion of the G.R.E.A.T. study (Total N = 3,568). All students who participated in the current investigation were enrolled in the sixth or seventh grades at the start of the study (Wave 1) and were in seventh or eighth grade by the end of the study (Wave 3). The average age of participants at Wave 1 was 12.14 years (SD = 0.65, range = 10-15), and the racial distribution was 46.5% White, 20.5% Hispanic, 16.9% African American, 3.7% Asian/Pacific Islander, 3.7% Native American, and 8.6% Other or mixed.

Measures

Independent Variable

Gang affiliation served as the independent variable in the present study. This variable was measured with a single dichotomous item from the G.R.E.A.T. longitudinal pretest (Wave 1) in which participants were asked if they were currently a member of a gang. If the child answered in the affirmative then they were classified as having a gang affiliation (gang = 1). If the child answered in the negative then they were classified as unaffiliated with a gang (gang = 0). A sum total of 65 (4.3% of total) boys and 35 (2.2% of total) girls reported that they were currently involved with a gang at Wave 1.

Dependent Variable

The dependent variable for the current study was delinquent involvement at Wave 3. To appraise this variable, participants were asked to indicate how often they engaged in 14 delinquent behaviors over the past 6 months: destroyed property, carried a weapon, spray painted a building, stole <US$50, stole >US$50, went into building to steal, stole a motor vehicle, hit someone, attacked someone with a weapon, committed armed robbery, involved in a gang fight, shot someone, sold marijuana, and sold other drugs. The frequency counts were then organized into a 7-point scale (0 = no times, 1 = one or two times, 2 = three to five times, 3 = six to ten times, 4 = eleven to fifteen times, 5 = sixteen to twenty times, 6 = more than twenty times), and the scores summed to create a score that could range from 0 to 84. The 1-year test–retest reliability of the delinquency score from Wave 2 to Wave 3 was .38.

Mediator Variables

The current design made allowances for two mediator variables. PCT, the mediator for the target pathway, was measured with nine items designed to assess neutralization techniques, a documented facet of PCT (Walters & Yurvati, 2017). These nine items (e.g., “a small lie is okay if no one is hurt”; “it is okay to steal from the rich who can replace the item”; “it is okay to physically fight to protect your rights”) were each rated by the child on a 5-point Likert-type scale: 1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree. Item ratings were then summed to produce a score that could range from 9 to 45, with higher scores indicating more proactive or neutralizing criminal thinking. An alternate term for PCT, given the relative youthfulness of the current sample, is cognitive insensitivity, the developmental antecedent to PCT (Walters & Espelage, 2018). The internal consistency of this nine-item scale in the current sample of participants was good (α = .88).

RCT served as the mediator variable for the control pathway, which was predicted to be nonsignificant. The RCT scale contained eight items (e.g., “I act on spur of the moment”; “I take risks for fun”; “It is exciting to possibly get in trouble”) each of which was rated on a 5-point Likert-type scale (1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree). Scores on the RCT scale, which was constructed exclusively for the purposes of the current study, ranged from 8 to 40, with higher scores indicating greater RCT. Because most of the items on this scale do not directly reference crime, they may be more appropriately described as indicators of cognitive impulsivity, the developmental antecedent to RCT (Walters & Espelage, 2018). Nonetheless, the terms reactive and proactive criminal thinking were used primarily in this article. The RCT scale achieved good internal consistency in the current sample of participants (α = .81).

Control Variables

Five control variables were included in the current investigation: three demographic measures—age (in years), sex (1 = male, 2 = female), and race (1 = white, 2 = non-white)—and two putative protective factors (parental knowledge and prosocial peer associations). The parental knowledge measure consisted of two items (“parents know where I am”; “parents know who I am with”) rated on a 5-point Likert-type scale (1 = strongly disagree, 2 = disagree, 3 = neither agree nor disagree, 4 = agree, 5 = strongly agree) to create a scale that could range from 2 to 10. Despite the scale’s brevity, it achieved good internal consistency in the current sample of participants (α = .75).

A peer variable believed to be less confounded with gang affiliation than peer delinquency or peer routine activities (i.e., association with prosocial peers) was included in the current study as a control variable. The prosocial peer scale contained eight items (e.g., “how many friends involved in community activities”; “how many friends involved in family activities”; “how many friends obey school rules”) rated on a 5-point Likert-type scale (1 = none of them, 2 = a few of them, 3 = about half of them, 4 = most of them, 5 = all of them), which when summed yielded a scale that could range from 8 to 40. This scale achieved adequate internal consistency (α = .76).

Cole and Maxwell (2003) recommend controlling for prior levels of a predicted variable when conducting a mediation analysis. Because there were three predicted variables (Wave 2 PCT, Wave 2 prosocial, and Wave 3 delinquency) there were three precursor measures, one for each predicted variable. Hence, the Wave 1 PCT score was included as a precursor predictor variable in the equation predicting Wave 2 PCT, the Wave 1 RCT score was included as a precursor in the equation predicting Wave 2 RCT, and Wave 1 delinquency was included as a precursor in the equation predicting Wave 3 delinquency.

Research Design

A three-wave longitudinal fixed panel design was employed as a means of evaluating two alternative mediational models. The mediational models had the same independent and dependent variables, but used different mediator variables (PCT vs. RCT). The first wave of data consisted of the pretest evaluation for G.R.E.A.T., the second wave consisted of the posttest G.R.E.A.T. evaluation, and the third wave was the first G.R.E.A.T. follow-up evaluation. Waves 1 and 2 were separated by 9 to 11 weeks and Waves 2 and 3 were separated by a year. The independent variable (Gang), control variables, and precursor measures were assessed at Wave 1; the mediator variables (PCT, RCT) were assessed at Wave 2; and the dependent variable (Delinquency) was assessed at Wave 3. These results were then cross-validated using the last three waves of G.R.E.A.T data (i.e., Waves 4-6, with each successive wave separated by 1 year) in which the independent, control, and precursor variables were assessed at Wave 4, the mediator variables were assessed at Wave 5, and the dependent variable was assessed at Wave 6. Because data from the G.R.E.A.T. project were nested by classroom, the classroom variable served as a cluster variable in complex model multiple regression analyses of data from Waves 1 to 3 and then again in analyses of data from Waves 4 to 6.

Data Analytic Plan

A path analysis of data from the first three waves of the G.R.E.A.T. study was performed using the structural equation modeling program MPlus 8.1 (Muthén & Muthén, 1998-2017). The results of this analysis were then cross-validated on the last three waves of the six-wave G.R.E.A.T. longitudinal study. A maximum likelihood with robust parameters and standard errors (MLR) estimator was used to accommodate the cluster option used in the analyses, and target and control pathways were evaluated using the comparison pathways approach (Walters, 2018a). The target pathway, the one predicted to be significant, ran from Wave 1 gang affiliation, to Wave 2 PCT, to Wave 3 delinquency. The control pathway, the one predicted to be nonsignificant, ran from Wave 1 gang affiliation, to Wave 2 RCT, to Wave 3 delinquency. Because bootstrapping, the preferred method for assessing the significance of indirect effects in a mediation analysis (Preacher, 2015), cannot be used with a cluster variable, the significance of the two indirect effects was evaluated against Preacher and Selig’s (2012) Monte Carlo Method for Assessing Mediation (MCMAM).

Sensitivity testing designed to determine the likelihood that omitted variables could explain the observed effects was performed for all significant indirect effects. This was accomplished with the aid of Kenny’s (2013) “failsafe ef” procedure: (r_my.x) × (sd_m.x) × (sd_y.x) / (sd_m) × (sd_y). The “failsafe ef” produces a coefficient that indicates how strongly an unobserved covariate confounder would need to correlate with both the mediating and dependent variables, after controlling for the mediator and independent variables in the case of the dependent variable, to eliminate the significant coefficient along the b path (from mediator to dependent variable) of the indirect effect. Also, conditioning on the precursor to an outcome variable can bias a path coefficient away from zero by turning the conditioned variable into a collider variable and giving the appearance of a significance effect when, in fact, no such effect exists (Elwert & Winship, 2014). To test whether the precursor measures acted as collider variables (a variable that is caused by two other variables—the independent variable, or a variable correlated with the independent variable, and the outcome measure), a second sensitivity analysis was conducted by removing all precursor measures from the analysis.

Missing Data

In the main analysis (Waves 1-3), less than half the sample had complete data on all 12 study variables (40.3%). Another 26.5% were missing data on one variable, 22.7% were missing data on two or three variables, and 10.5% were missing data on four to 10 variables. For the cross-validation analysis (Waves 4-6), 34.8% of the sample had complete data, 4.2% were missing data on one variable, 8.4% were missing data on two to three variables, 7.7% were missing data on four to eight variables, 40.6% were missing data on nine variables, and 4.4% were missing data on 10 to 12 variables. Missing data in both analyses were handled with full information maximum likelihood (FIML). FIML estimates model parameters and standard errors from calculations performed on nonmissing data and is robust to violations of its basic assumptions (Collins, Schafer, & Kam, 2001; Young & Johnson, 2013). Researchers have found that FIML produces results that are significantly less biased than those produced by traditional missing data methods (Allison, 2012). To further enhance the precision of FIML (Collins et al., 2001), 39 auxiliary variables, measures that are used to calculate parameters and standard errors but are not included in the actual analysis itself, were included in the procedure. The 39 auxiliary variables are listed in the appendix at the end of this article.

Results

Preliminary Analyses

Descriptive statistics and correlations between the 12 study variables can be found in Table 1. As indicated by the results in Table 1, the majority of correlations achieved statistical significance, even after correcting for the large number of individual comparisons. Collinearity diagnostics revealed no evidence of multicollinearity in the regression equations for the main (tolerance = .567-.983; variance inflation factor [VIF] = 1.017-1.763) and cross-validation (tolerance = .561-.995, VIF = 1.005-1.784) analyses.

Table 1:

Descriptive Statistics and Correlations for the 12 Independent, Dependent, Mediator, and Control Variables

Variable	n	M	SD	Range	1	2	3	4	5	6	7	8	9	10	11	12
1. Age	2,870	12.14	0.65	10-15	—	−.10^†	.03	−.07^†	−.10^†	.10^†	.16^†	.15^†	.12^†	.10^†	.12^†	.10^†
2. Sex	3,131	—	—	1-2		—	.06	.17^†	.05	−.06	−.21^†	−.21^†	−.16^†	−.15^†	−.18^†	−.22^†
3. Race	3,109	—	—	1-2			—	−.08^†	−.14^†	.09^†	.19^†	.14^†	.03	.01	.09^†	.03
4. Parental knowledge	2,854	7.39	1.97	2-10				—	.30^†	−.11^†	−.28^†	−.25^†	−.22^†	−.24^†	−.23^†	−.19^†
5. Prosocial peers	2,801	26.07	6.16	8-40					—	−.16^†	−.44^†	−.41^†	−.38^†	−.34^†	−.34^†	−.20^†
6. Gang affiliation	2,686	—	—	0-1						—	.24^†	.23^†	.19^†	.15^†	.45^†	.21^†
7. PCT-1	2,876	24.29	7.03	9-45							—	.70^†	.51^†	.45^†	.47^†	.37^†
8. PCT-2	2,475	24.85	7.52	9-45								—	.43^†	.54^†	.44^†	.36^†
9. RCT-1	2,868	23.33	5.50	8-40									—	.63^†	.36^†	.29^†
10. RCT-2	2,465	22.47	5.92	8-40										—	.30^†	.28^†
11. Delinquency-1	2,788	2.48	5.06	0-64											—	.45^†
12. Delinquency-3	1,734	3.40	6.83	0-59												—

Note. Age = age in years; sex = male (1) versus female (2); race = White (1) versus non-White (2); parental knowledge = parental knowledge of child’s activities and associations at Wave 1; prosocial peers = prosocial peer associations at Wave 1; gang affiliation = self-reported affiliation with a gang at Wave 1, yes (1) or no (0); PCT-1 = proactive criminal thinking at Wave 1; PCT-2 = proactive criminal thinking at Wave 2; RCT-1 = reactive criminal thinking at Wave 1; RCT-2 = reactive criminal thinking at Wave 2; Delinquency-1 = participant delinquency at Wave 1; Delinquency-3 = participant delinquency at Wave 3; n = number of nonmissing cases; range = range of scores in current sample.

†

p < .00076 (Bonferroni-corrected alpha: .05 / 66 correlations).

Main Analysis

Table 2 provides a summary of the results of a path analysis in which data were clustered by classroom and auxiliary variables were included to enhance the precision of FIML. As these results indicate, the a and b paths of the target pathway were statistically significant, whereas only the b path of the control pathway was significant (see Figure 1). Because bootstrapping cannot be used to assess indirect effects in an analysis where clustering or auxiliary variables are employed, confidence intervals were constructed using the MCMAM procedure. The results of this analysis revealed that while the indirect effect of the target pathway was significant, the indirect effect of the control pathway was not (see top portion of Table 3).

Table 2:

Robust Maximum Likelihood Path Analysis of the Gang Affiliation–Delinquency Relationship: Waves 1 to 3 (N = 3,136)

Predictor	b [95% CI]	β	z	p
PCT-2 (Outcome)
Gang affiliation	2.034 [0.636, 3.431]	0.052	2.85	.004
Age	0.381 [−0.053, 0.815]	0.033	1.72	.085
Sex	−1.287 [−1.752, −0.821]	−0.086	−5.42	<.001
Race	0.056 [−0.489, 0.601]	0.004	0.20	.841
Parental knowledge	−0.126 [−0.255, 0.003]	−0.033	−1.91	.056
Prosocial peers	−0.160 [−0.204, −0.116]	−0.132	−7.13	<.001
PCT-1	0.620 [0.579, 0.660]	0.583	29.68	<.001
RCT-2 (Outcome)
Gang affiliation	0.408 [−0.724, 1.539]	0.013	0.71	.480
Age	0.163 [−0.173, 0.498]	0.018	0.95	.342
Sex	−0.640 [−1.027, −0.252]	−0.054	−3.23	.001
Race	−0.325 [−0.709, 0.059]	−0.028	−1.66	.097
Parental knowledge	−0.229 [−0.339, −0.118]	−0.077	−4.05	<.001
Prosocial peers	−0.115 [−0.150, −0.080]	−0.121	−6.47	<.001
RCT-1	0.588 [0.555, 0.621]	0.549	34.87	<.001
Delinquency-3 (Outcome)
PCT-2	0.128 [0.055, 0.201]	0.137	3.44	<.001
RCT-2	0.094 [0.034, 0.154]	0.079	3.06	.002
Age	0.454 [−0.050, 0.958]	0.042	1.76	.078
Sex	−1.491 [−2.104, −0.879]	−0.107	−4.77	<.001
Race	−0.124 [−0.759, 0.512]	−0.009	−0.38	.703
Parent knowledge	−0.170 [−0.328, −0.013]	−0.048	−2.12	.034
Prosocial peers	−0.005 [−0.080, 0.071]	−0.004	−0.12	.901
Gang affiliation	1.448 [−2.720, 5.616]	0.040	0.68	.496
Delinquency-1	0.457 [0.260, 0.654]	0.332	4.55	<.001
PCT-2 with RCT-2	7.848 [6.601, 9.096]	0.333	12.33	<.001

Note. PCT-2 (Outcome) = regression equation with proactive criminal thinking at Wave 2 as the outcome measure; RCT-2 (Outcome) = regression equation with reactive criminal thinking at Wave 2 as the outcome measure; Delinquency-3 (Outcome) = regression equation with participant delinquency at Wave 3 as the outcome measure; gang affiliation = self-reported affiliation with a gang at Wave 1, yes (1) or no (0); age = age in years; sex = male (1) versus female (2); race = White (1) versus non-White (2); parental knowledge = parents’ knowledge of child’s activities and associations at Wave 1; prosocial peers = prosocial peer associations at Wave 1; PCT-1 = proactive criminal thinking at Wave 1; PCT-2 = proactive criminal thinking at Wave 2; RCT-1 = reactive criminal thinking at Wave 1; RCT-2 = reactive criminal thinking at Wave 2; Delinquency-1 = participant delinquency at Wave 1; PCT-2 with RCT-2 = covariance between proactive and reactive criminal thinking at Wave 2; b (95% CI) = unstandardized coefficient and the lower and upper limits of the 95% CI for the unstandardized coefficient (in brackets); β = standardized coefficient; z = Wald Z test; p = significance level of the Wald Z test.

Figure 1:

MLR Path Analysis of the Mediating Effects of Proactive and Reactive Criminal Thinking on the Gang Affiliation–Delinquency Relationship Between Waves 1 and 3 of the G.R.E.A.T. Study (N = 3,136)

Table 3:

Total, Direct, and Indirect Effects for Pathways Running From Gang Affiliation at Waves 1 and 4 to Delinquency at Wave 3 and 6 Using MCMAM Confidence Intervals (N = 3,136)

	MCMAM
Pathways	Lower	Upper
Main analysis
Specific indirect effects
Gang-1 → PCT-2 → Delinquency-3	0.0664	0.5094
Gang-1 → RCT-2 → Delinquency-3	−0.0767	0.1592
Cross-validation analysis
Specific indirect effects
Gang-4 → PCT-5 → Delinquency-6	0.0022	0.9115
Gang-4 → RCT-5 → Delinquency-6	−0.1412	0.4244

Note. Gang-1 = self-reported affiliation with a gang at Wave 1; PCT-2 = proactive criminal thinking at Wave 2; RCT-2 = reactive criminal thinking at Wave 2; Delinquency-3 = participant delinquency at Wave 3; Gang-4 = self-reported affiliation with a gang at Wave 4; PCT-5 = proactive criminal thinking at Wave 5; RCT-5 = reactive criminal thinking at Wave 5; Delinquency-6 = participant delinquency at Wave 6; MCMAM = Monte Carlo Method for Assessing Mediation; lower = lower boundary of the 95% confidence interval; upper = upper boundary of the 95% confidence interval.

Sensitivity Analyses

Implementation of the “failsafe ef” procedure revealed that the current results were highly robust to the effects of omitted variable bias in that an unobserved covariate confounder would need to correlate .33 with PCT-2 and .33 with Delinquency-3, controlling for Gang-1 and PCT-2 in the case of Delinquency-3, to lower the significant b path of the target indirect effect to zero. When all three precursor measures from the three regression equations included in this study were removed from the analysis, the path coefficients increased rather than decreased. This suggests that the precursor measures did not function as collider variables in this study.

Cross-Validation Analysis

Evaluating the independent, control, mediating, and dependent variables from the main analysis in Waves 4 through 6 of the G.R.E.A.T., a cross-validation analysis was performed. Only 6% of the youth who indicated that they were currently involved in a gang at Wave 1 reported being currently involved in a gang at Wave 4. Forty other youth who did not report gang involvement at Wave 1 were gang affiliated at Wave 4, for a total of 46 (27 boys, 19 girls) gang-affiliated youth at Wave 4. Congruent with the results from Waves 1 through 3, the a and b paths of the target pathway were statistically significant, whereas only the b path of the control pathway was significant (see Figure 2). In addition, the MCMAM confidence intervals were significant for the target pathway but not for the control pathway (see bottom portion of Table 3). Sensitivity testing revealed that an unobserved covariate confounder would need to correlate .27 with PCT-5 and .27 with Delinquency-6, controlling for Gang-4 and PCT-5 in the case of Delinquency-6, to eliminate the significant coefficient on the b path of the significant indirect effect. There was no evidence of a collider effect in this cross-validation analysis.

Figure 2:

MLR Path Analysis of the Mediating Effects of Proactive and Reactive Criminal Thinking on the Gang Affiliation–Delinquency Relationship Between Waves 4 and 6 of the G.R.E.A.T. Study (N = 3,136)

Discussion

Findings obtained from a path analysis of the first three waves of the G.R.E.A.T. project confirmed the presence of a directional relationship between gang affiliation and subsequent delinquency that was mediated by the same variable known to mediate the peer delinquency–participant delinquency relationship (Walters, 2015, 2016, 2017b)—namely, PCT or cognitive insensitivity. Using the comparison pathways approach to contrast social learning (PCT-mediated) and self-control (RCT-mediated) explanations of the gang–delinquency relationship, support was obtained for the social learning model only, suggesting that one way gang affiliation facilitates delinquency is by teaching youth attitudes and techniques conducive to crime. Sensitivity testing results from the “failsafe ef” procedure indicated that the current results were moderately to highly robust to the effects of omitted variable bias. In addition, there was no evidence that any of the precursor measures served as collider variables. It should also be pointed out that results obtained with data from Waves 1-3 of the G.R.E.A.T. project were successfully replicated on data from Waves 4-6 of the same sample. This was particularly important given that only 6% of the youth who indicated that they were involved with a gang at intake (Wave 1) were still involved with a gang 2 years later, at Wave 4.

Theoretical Implications

Whereas the indirect effect of gang affiliation on participant delinquency via PCT was significant in both analyses, the direct effect of gang affiliation was significant in one of the analyses (i.e., cross-validation). This suggests that variables other than those included in the present study may have also been mediating the gang–delinquency relationship. Co-offending could be one such variable. Research indicates that co-offending accounts for a significant portion of youth crime (Felson, 2009; Zimring & Laqueur, 2015) and that gang associations are a principal means by which opportunities for co-offending are realized (Grund & Densley, 2015; McCuish, Bouchard, & Corrado, 2015). Co-offending, in fact, may be particularly responsive to current gang status. Prior research has shown that as youth drop out of gangs their propensity for crime also drops (Gordon et al., 2004; Sweeten et al., 2013). This could partly be the result of a sudden drop in opportunities for co-offending as youth leave the gang. Self-selection would also appear to play an important role in promoting gang affiliations. Selection effects, in fact, are more congruent with the self-control model that provided the control pathway for the current investigation than they are with the social learning model. Gottfredson and Hirschi (1990) argued that delinquents select one another in a process of homophily (“birds of a feather”). It is not hard to see how that same process could be operating in gangs. While it is likely that a sizable portion of criminally oriented youth select themselves into gangs, evidence from the present as well as several prior (Bendixen, Endresen, & Olweus, 2006; Pyrooz et al., 2016) studies indicates that a gang can have a facilitative effect on future delinquent and criminal behavior.

Another theoretical implication of the current results is that collective gang associations may be just as vital in the formation of delinquency-promoting cognitive patterns like PCT as individual delinquent peer associations. Prior research denotes that peer influence encourages delinquency by facilitating PCT (Walters, 2015, 2016, 2017b). Associating with delinquent peers exposes people to PCT, which then increases the probability of them applying these planned, calculated, and scheming aspects of antisocial cognition to situations in their own lives. Even after controlling for prosocial peer associations and parental knowledge, gang affiliation continued to predict a change in PCT. In other words, gang affiliation predicted an increase in neutralization techniques and moral disengagement strategies from baseline (Wave 1) to follow-up (Wave 2). Additional study is required to answer several follow-up questions: (a) What aspects or features of gang structure and/or function facilitate the acquisition of PCT? and (b) Is it loyalty to the gang, a sense of acceptance from other gang members, or shared norms that is driving this effect? It will also be necessary to determine whether this effect applies to other social institutions, such as families, schools, and neighborhoods.

Practical Implications

There are several practical implications to the current results. Given a reasonable degree of stability in criminal thinking (Walters, 2015, 2016, 2017b), it is probably unrealistic to expect that simply removing a child from the original learning situation (i.e., gang) will necessarily eliminate that which has already been learned; in this case, PCT. Instead, the child must acquire counter-PCT attitudes and beliefs. In a recent study on this issue, Walters (2018b) discovered that peer influence and peer resistance affected future delinquency via the same mediator—namely, PCT. Hence, once PCT is learned, counter-PCT thought patterns and cognitive skills like empathy must be acquired to suppress, override, and replace the PCT. In one study, empathy skills improved as juvenile offenders progressed through a 10-week empathy training program as a result of increased moral maturity and decreased neutralization (Barriga, Sullivan-Cosetti, & Gibbs, 2009). In another study, a group of juvenile offenders enrolled in animal therapy displayed significant gains in empathy relative to a group of untreated controls (Dawson, 2016). The next step in this area of research is determining whether these changes in empathy lead to significant reductions in long-term risk of recidivism.

As important as developing counter-PCT attitudes and beliefs is to breaking the bond between gang affiliation and delinquency, it is just as important to prevent gang involvement before it occurs given its ability to foster PCT and other criminogenic factors not included in the present study but potentially contained in the significant direct effect observed in the cross-validation analysis. Effective youth gang prevention and intervention is conducted at three levels: at the individual-level with at-risk children, at the family-level with parents, and at the school- and community-level. Individual-level programs that hold the child accountable for his or her actions, reinforce clear expectations for appropriate behavior, and provide the child with the opportunity to engage in socially rewarding interactions with prosocial peers that divert the child’s attention away from gangs have all been found effective in reducing future gang involvement (Wyrick, 2006). Family-level prevention should probably target parental knowledge, given its borderline significant relationship to PCT and significant association with delinquency in the present study. School- and community-level prevention strategies that have been found effective in reducing risk are those that provide children with healthy outlets for their social and activity needs, under proper adult supervision (Farrington & Welsh, 2007).

Limitations

In the current study the independent variable, gang affiliation, was measured with a single dichotomous item. Dichotomous assessment of gang affiliation is a fairly common practice in gang research, but if the field is to move forward more detailed assessments are required. In addition, the gang measure, as is true of all variables in the present study, came from a single source, “participant self-report.” Youth may be less than fully forthcoming when questioned about gang affiliations or delinquent peer associations. What is more, the exclusive use of a single method, as was the case here, can lead to mono-operational bias and inflated path coefficients (Shadish, Cook, & Campbell, 2002). In a study of gang involvement in 421 youth referred by the justice system for intensive mental health treatment, Boxer, Veysey, Ostermann, and Kubik (2015) employed a definition of gang affiliation that went beyond simply asking youth if they were involved in a gang. This definition was based, in part, on problems related to being a member of a gang (e.g., gang fights) and, in part, on a review of official and clinical records and other corroborating information.

The current study was further limited by nonuniform periods of time between waves. The lag between Waves 1 and 2 was on the order of 9 to 11 weeks, whereas the lag between Waves 2 and 3 was 1 year. These concerns are alleviated somewhat by the fact that these results were successfully cross-validated on Waves 4 to 6 of the G.R.E.A.T. study, where there was a consistent 1-year gap between waves. It should also be noted that gang affiliation was low and delinquency scores small in the present investigation. Both findings are clearly understandable given the general population nature of the sample, yet it likely placed restraints on the study’s power to detect group differences. Even with this loss of power, however, the research hypotheses received support in both the main and cross-validation analyses. Finally, there was a significant amount of missing data in the present study, particularly in the cross-validation analysis. The use of FIML and 39 auxiliary variables, some of which correlated highly with the variables analyzed in the current investigation, provides some reassurance as to the validity of the results.

Concluding Remarks

Gangs would appear to have an effect on the future delinquent activities of its members above and beyond the simple fact that gangs are composed of individuals with histories of delinquency (Battin et al., 1998). Findings from the current investigation highlight the potential significance of collective peer influence as a complement to the individual peer influences that have traditionally been studied in criminology. Moreover, collective peer influence holds promise of informing both policy and practice. We might reasonably anticipate that many of the present-day policies designed to identify and prevent youth gang delinquency (e.g., early childhood interventions, school programs, after-school programs, community-based initiatives, and formal treatment) will continue to produce modest to moderate effects (Lafontaine, Ferguson, & Wormith, 2005). What the current study suggests is that to fully appreciate and prevent youth gang delinquency we must understand what the gang means to the individual. In other words, what is the source of the collective peer influence effect? Once the source of the effect has been identified we will be in a position to develop programs and policies that directly target the criminogenic influence that youth gangs have on their members. The notion that groups, institutions, and culture affect human behavior in ways that supersede the influence of the individual members of these groups, institutions, and cultures is certainly not new (Durkheim, 1893/1997). The present study’s contribution to the literature is that it brings into clearer focus the necessity of identifying how the collective peer effect shapes the thinking of youthful gang members and facilitates their future involvement in delinquency and crime.

Footnotes

Appendix

ORCID iD

Glenn D. Walters

Glenn D. Walters is a professor of criminal justice at Kutztown University (Pennsylvania) where he teaches classes in offender management, research methods, criminology, and substance abuse and crime. He has published nearly 400 articles and book chapters and authored or coauthored 19 books and monographs. His current research interests include criminal thinking, mediation analysis, peer effects on delinquency development, and integrated theories of criminology.

References

Akers

R. L.

(1998). Social learning and social structure: A general theory of crime and deviance. Boston, MA: Northeastern University Press.

Allison

P. D.

(2012). Handling missing data by maximum likelihood (SAS Global Forum 2012, Paper 312-2012). Cary, NC: SAS Institute.

Barriga

A. Q.

Sullivan-Cosetti

Gibbs

J. C.

(2009). Moral cognitive correlates of empathy in juvenile delinquents. Criminal Behaviour and Mental Health, 19, 253-264.

Battin

S. R.

Hill

K. G.

Abbott

R. D.

Catalano

R. F.

Hawkins

J. D.

(1998). The contribution of gang membership to delinquency beyond delinquent friends. Criminology, 36, 93-115.

Bendixen

Endresen

I. M.

Olweus

(2006). Joining and leaving gangs: Selection and facilitation effects on self-reported antisocial behaviour in early adolescence. European Journal of Criminology, 3, 85-114.

Boxer

Veysey

Ostermann

Kubik

(2015). Measuring gang involvement in a justice-referred sample of youth in treatment. Youth Violence and Juvenile Justice, 13, 41-59.

Carson

D. C.

(2013). Perceptions of prosocial and delinquent peer behavior and the effect on delinquent attitudes: A longitudinal study. Journal of Criminal Justice, 41, 151-161.

Cepeda

Saint Onge

J. M.

Nowotny

K. M.

Valdez

(2016). Associations between long-term gang membership and informal social control processes, drug use, and delinquent behavior among Mexican American youth. International Journal of Offender Therapy and Comparative Criminology, 60, 1532-1548.

Chapple

C. L.

Hope

T. L.

(2003). An analysis of the self-control and criminal versatility of gang and dating violence offenders. Violence and Victims, 18, 671-690.

10.

Cole

D. A.

Maxwell

S. E.

(2003). Testing mediational models with longitudinal data: Questions and tips in the use of structural equation modeling. Journal of Abnormal Psychology, 112, 558-577.

11.

Collins

L. M.

Schafer

J. L.

Kam

C. M.

(2001). A comparison of inclusive and restrictive strategies in modern missing data procedures. Psychological Methods, 6, 330-351.

12.

Dawson

(2016). Empathy through animals: Generating evidence-based outcomes for empathy development. Juvenile and Family Court Journal, 67(4), 43-54.

13.

Dong

Krohn

M. D.

(2016). Dual trajectories of gang affiliation and delinquent peer association during adolescence: An examination of long-term offending outcomes. Journal of Youth and Adolescence, 45, 746-762.

14.

Durkheim

(1997). The division of labor in society. New York, NY: Simon & Schuster. (Original work published 1893)

15.

Elwert

Winship

(2014). Endogenous selection bias: The problem of conditioning on a collider variable. Annual Review of Sociology, 40, 31-53.

16.

Esbensen

F.-A.

(2002, September). National evaluation of the Gang Resistance Education and Training (G.R.E.A.T.) program: Final report. Washington, DC: U.S. Department of Justice.

17.

Farrington

D. P.

Welsh

B. C.

(2007). Saving children from a life of crime: Early risk factors and effective interventions. New York, NY: Oxford University Press.

18.

Felson

(2009). The natural history of extended co-offending. Trends in Organized Crime, 12, 159-165.

19.

Gordon

R. A.

Lahey

B. B.

Kawai

Loeber

Stouthamer-Loeber

Farrington

D. P.

(2004). Antisocial behavior and youth gang membership: Selection and socialization. Criminology, 42, 55-87.

20.

Gottfredson

Hirschi

(1990). A general theory of crime. Stanford, CA: Stanford University Press.

21.

Grund

T. U.

Densley

J. A.

(2015). Ethnic homophily and triad closure: Mapping internal gang structure using exponential random graph models. Journal of Contemporary Criminal Justice, 31, 354-370.

22.

Hayes

A. F.

(2013). Introduction to mediation, moderation, and conditional process analysis: A regression-based approach. New York, NY: Guilford.

23.

Hirschi

(1969). Causes of delinquency. Berkeley: University of California Press.

24.

Hoeben

E. M.

Weerman

F. M.

(2016). Why is involvement in unstructured socializing related to adolescent delinquency? Criminology, 54, 242-281.

25.

Hughes

L. A.

Short

J. F.

(2014). Partying, cruising, and hanging in the streets: Gangs, routine activities, and delinquency and violence in Chicago, 1959-1962. Journal of Quantitative Criminology, 30, 415-451.

26.

Katz

C. M.

Fox

A. M.

(2010). Risk and protective factors associated with gang-involved youth in Trinidad and Tobago. Pan American Journal of Public Health, 27, 187-202.

27.

Kenny

D. A.

(2013). Mediation: Sensitivity analysis [Webinar]. Retrieved from http://davidakenny.net/webinars/Mediation/Sensitivity/Sensitivity.html

28.

Klein

M. W.

Maxson

C. L.

(2006). Street gang patterns and policies. New York, NY: Oxford University Press.

29.

Krohn

M. D.

Thornberry

(2008). Longitudinal perspectives on adolescent street gangs. In Liberman

(Ed.), The long view of crime: A synthesis of longitudinal research (pp. 128-160). New York, NY: Springer.

30.

Kupersmidt

J. B.

Coie

J. D.

Howell

J. C.

(2004). Resilience in children exposed to negative peer influences. In Maton

K. I.

Schellenbach

C. J.

Leadbeater

B. J.

Solarz

A. L.

(Eds.), Investing in children, youth families, and communities: Strength-based research and policy (pp. 251-268). Washington, DC: American Psychological Association.

31.

Lafontaine

Ferguson

Wormith

J. S.

(2005). Street gangs: A review of the empirical literature on community and corrections-based prevention, intervention and suppression strategies. Regina, Canada: Saskatchewan Corrections and Public Safety.

32.

Lahey

B. B.

Van Hulle

C. A.

D’Onofrio

B. M.

Rodgers

J. L.

Waldman

I. D.

(2008). Is parental knowledge of their adolescent offspring’s whereabouts and peer associations spuriously associated with offspring delinquency? Journal of Abnormal Child Psychology, 36, 807-823.

33.

McCuish

E. C.

Bouchard

Corrado

R. R.

(2015). The search for suitable homicide co-offenders among gang members. Journal of Contemporary Criminal Justice, 31, 319-336.

34.

McGloin

J. M.

Shermer

L. O.

(2009). Self-control and deviant peer network structure. Journal of Research in Crime and Delinquency, 46, 35-72.

35.

Muthén

(1998-2017). Mplus user’s guide (8th ed.). Los Angeles, CA: Author.

36.

National Gang Center. (2013). National Youth Gang Survey analysis. Retrieved from http://www.nationalgangcenter.gov/Survey-Analysis

37.

Preacher

K. J.

(2015). Advances in mediation analysis: A survey and synthesis of new developments. Annual Review of Psychology, 66, 825-852.

38.

Preacher

K. J.

Selig

J. P.

(2012). Advantages of Monte Carlo confidence intervals for indirect effects. Communication Methods and Measures, 6, 77-98.

39.

Pyrooz

D. C.

Decker

S. H.

(2013). Delinquent behavior, violence, and gang involvement in China. Journal of Quantitative Criminology, 29, 251-272.

40.

Pyrooz

D. C.

Sweeten

(2015). Gang membership between ages 5 and 17 in the United States. Journal of Adolescent Health, 56, 414-419.

41.

Pyrooz

D. C.

Turanovic

J. J.

Decker

S. H.

(2016). Taking stock of the relationship between gang membership and offending: A meta-analysis. Criminal Justice and Behavior, 43, 365-397.

42.

Shadish

W. R.

Cook

T. D.

Campbell

D. T.

(2002). Experimental and quasi-experimental designs for generalized causal inference. Boston, MA: Mifflin.

43.

Stattin

Kerr

(2000). Parental monitoring: A reinterpretation. Child Development, 71, 1072-1085.

44.

Sutherland

E. H.

(1947). Principles of criminology (4th ed.). Philadelphia, PA: J. B. Lippincott.

45.

Sweeten

Pyrooz

D. C.

Piquero

A. R.

(2013). Disengaging from gangs and desistance from crime. Justice Quarterly, 30, 469-500.

46.

Walters

G. D.

(2015). Proactive criminal thinking and the transmission of differential association: A cross-lagged multi-wave path analysis. Criminal Justice and Behavior, 42, 1128-1144.

47.

Walters

G. D.

(2016). Friends, cognition, and delinquency: Proactive and reactive criminal thinking as mediators of the peer influence and peer selection effects. Justice Quarterly, 33, 1055-1079.

48.

Walters

G. D.

(2017a). Modelling the criminal lifestyle: Theorizing at the edge of chaos. Cham, Switzerland: Palgrave Macmillan.

49.

Walters

G. D.

(2017b). Proactive criminal thinking and deviant identity as mediators of the peer influence effect. Youth Violence and Juvenile Justice, 15, 281-298.

50.

Walters

G. D.

(2017c). Reactive criminal thinking as a consequence of low self-control and prior offending. Deviant Behavior, 38, 119-129.

51.

Walters

G. D.

(2018a). PM effect size estimation for mediation analysis: A cautionary note, alternate strategy, and real data illustration. International Journal of Social Research Methodology, 21, 25-33.

52.

Walters

G. D.

(2018b). Resistance to peer influence and crime desistance in emerging adulthood: A moderated mediation analysis. Law and Human Behavior, 42, 520-530. doi:10.1037/lhb000029

53.

Walters

G. D.

(2018c). Unsupervised routine activities as a mediator of the parental knowledge-delinquency relationship. Justice Quarterly, 35, 871-894.

54.

Walters

G. D.

Espelage

D. L.

(2018). Cognitive insensitivity and cognitive impulsivity as mediators of bullying continuity: Extending the psychological inertia construct to bullying behavior. School Psychology Quarterly, 33, 527-536. doi:10.1037/spq0000240

55.

Walters

G. D.

Yurvati

(2017). Testing the construct validity of the PICTS proactive and reactive scores against six putative measures of proactive and reactive criminal thinking. Psychology, Crime & Law, 23, 1-14.

56.

Wanamaker

K. A.

Jones

N. J.

Brown

S. L.

(2018). Strengths-based assessments for use with forensic populations: A critical review. International Journal of Forensic Mental Health, 17, 202-221.

57.

Weerman

F. M.

Lovegrove

P. J.

Thornberry

(2013). Gang membership transitions and its consequences: Exploring changes related to joining and leaving gangs in two countries. European Journal of Criminology, 12, 70-91.

58.

Wyrick

P. A.

(2006). Gang prevention: How to make the “front end” of your anti-gang effort work. United States Attorneys’ Bulletin, 54, 52-60.

59.

Young

Johnson

D. R.

(2013). Methods for handling missing secondary respondent data. Journal of Marriage and Family, 75, 221-234.

60.

Zimring

F. E.

Laqueur

(2015). Kids, groups, and crime: In defense of conventional wisdom. Journal of Research in Crime and Delinquency, 52, 403-413.