The purpose of this paper is to clarify the relationship between gang affiliation and criminal thinking.
A sample of 1,354 youth (1,170 males, 184 females) from the Pathways to Desistance Study served as participants in this study, and a causal mediation path analysis was performed on proactive and reactive criminal thinking, gang affiliation and subsequent offending.
Using three waves of data, it was determined that the pathway running from reactive criminal thinking to gang affiliation to proactive criminal thinking was significant, whereas the pathway running from proactive criminal thinking to gang affiliation to reactive criminal thinking was not. A four-wave model, in which violent and income offending were appended to the three-wave model, disclosed similar results.
Two separate targets for intervention with youth at risk for gang involvement: proactive and reactive criminal thinking. The impulsive, irresponsible, reckless and disinhibited nature of reactive criminal thinking may best be managed with a secondary prevention approach and cognitive-behavioral skills training; the planned, cold, calculating and amoral nature of proactive criminal thinking may best be managed with a tertiary prevention approach and moral retraining. Trauma therapy may be of assistance to youth who have been victimized over the course of their gang experience.
These findings reveal evidence of a gang selection effect that is independent of the well-documented peer selection effect, in which reactive criminal thinking led to gang affiliation in youthful offenders, particularly non-White offenders, and a gang influence effect, independent of the frequently observed peer selection effect, in which gang affiliation contributed to a rise in proactive criminal thinking.
Walters, G.D. (2021), "Criminal thinking and gang affiliation: antecedents and consequences", Journal of Criminological Research, Policy and Practice, Vol. 7 No. 2, pp. 150-163. https://doi.org/10.1108/JCRPP-05-2020-0040
Emerald Publishing Limited
Copyright © 2020, Emerald Publishing Limited
The peer influence or socialization effect occurs when a person acquires thoughts and behaviors conducive to crime through their association with those already involved in crime. Peer influence is how the peer–participant delinquency relationship is accounted for by social learning theorists (Akers, 1998; Sutherland, 1947). The individual learns delinquent thoughts and behaviors through their interactions with persons already involved in a delinquent lifestyle. Research has consistently supported the peer influence effect even after controlling for alternate explanations of the peer delinquency–participant delinquency relationship (Brechwald and Prinstein, 2011; Gifford-Smith et al., 2005; Osgood et al., 2015). One such alternate explanation is commonly referred to as the peer selection effect. As the name suggests, the individual selects his or her friends and associates based on commonalities in attitude, behavior and interests. Hence, a child with an antisocial attitude or interest in crime selects a group of individuals with similar attitudes and interests with whom to associate. The peer selection or homophily theory of peer–participant similarity in offending and antisocial behavior is particularly popular with social control theorists (Gottfredson and Hirschi, 1990) and like the peer influence effect, has received its share of empirical support (Jose et al., 2016; Kiesner et al., 2003; Teneyck and Barnes, 2015).
Gang affiliation and criminal thinking
It stands to reason that if peers can produce an influence or selection effect, then so may gangs and various other groups. Gang influence and selection factors may, in fact, parallel peer influence and selection effects. The collective or gang selection effect could explain why some youth join gangs in the first place. Lachman et al. (2013) discerned that the reasons given by youth for joining a gang coalesced into three latent factors: to fill a void (e.g. to be noticed, feel important, avoid home, fill up empty time), instrumental considerations (to get money or other things, for protection, had no choice) and belonging (to be with people like you, for excitement, because the things they do are cool). Although instrumental motives correlated highest with concurrent delinquent behavior, specific motives associated with the void (fill up empty time) and belonging (for excitement) factors were more congruent with Walters’ (2016) observation that reactive criminal thinking (RCT: impulsive, emotional aspects of criminal thinking in which the focus is on immediate gratification) mediates the peer selection effect. Vigil (2003) coined the term multiple marginality to describe how factors such as poverty, oppression and discrimination provoke strain, neutralize protective factors within the family and encourage youth to join gangs and commit crime. Therefore, in addition to establishing RCT as a possible motive for initial gang involvement, strain and family/parenting factors such as parental knowledge need to be controlled in determining the specificity of the effect.
Whereas the gang selection effect explains why youth join gangs, the gang influence effect explains how gangs change youth. Because proactive criminal thinking (PCT: the planned, calculated aspects of criminal thinking that give rise to instrumental criminal actions) has been found to mediate the peer influence effect (Walters, 2015, 2016), there is reason to believe that it may also mediate the gang influence effect. Consequently, one potential effect of gang affiliation on the affiliated individual is to increase PCT. In a recent study using data from the Gang Resistance Education and Training (GREAT) project (Esbensen, 2002), Walters (2019a) ascertained that gang involvement fueled an increase in PCT, consistent with the notion that youth learn the planned, calculated, and instrumental aspects of antisocial cognition through their interactions with other law and rule violators, whether it be individual peers or an entire gang. Research has also uncovered support for the proposition, central to social learning theory, that individuals learn the techniques, attitudes and motives for crime through their association with other antisocial individuals (Brownfield, 2014; Melde and Esbensen, 2011). Brownfield (2014), for instance, determined that gang members were more likely to endorse manipulative and instrumental attitudes toward the law and moral cynicism toward the police than youth who were not gang members. Other gang members are only one interpersonal influence capable of teaching youth the attitudes and techniques for crime; they also learn criminal attitudes and techniques from non-gang affiliated peers and use prior offending as a springboard to future offending (Akers, 1998). Consequently, both prior offending and non-gang peer delinquency were controlled in this study.
Violent and acquisitive crime
Criminal offenses are often broken down into subcategories of violent and acquisitive or property crime. Whereas acquisitive crime is several times more common than violent crime (Rosenfeld, 2009), there are still many in the field of criminology who believe that the similarities far outweigh the differences (Gottfredson and Hirschi, 1990). By contrast, Zimring and Hawkins (1997) contend that violent and acquisitive crime constitute two distinct patterns of behavior and that they differ from one another in terms of both causes and correlates. In a series of secondary or four-wave analyses, the current investigation sought to determine whether a gang-mediated cognitive pathway would predict both forms of offending. This was done to test the robustness of the pathway. The pathway is believed to begin with a gang selection effect, whereby high levels of RCT increase a youth’s odds of associating with various gang members. These associations then give rise to a gang influence effect, wherein contact with other gang members leads to the acquisition of crime-congruent cognitions like PCT, which, in turn, give rise to heightened opportunities for both violent and acquisitive offending.
Although the present study can be considered an extension or elaboration of prior research on criminal thinking and the peer and gang influence and selection effects (Walters, 2015, 2016, 2019a), there were several unique aspects to the current investigation. These unique aspects are the study’s principal contributions:
understand how RCT and PCT contribute to the gang selection and influence effects, respectively, as part of a larger integrated model;
examine how gang affiliation may co-exist with peer associations to increase liability for offending; and
determine whether the effects are sufficiently robust to influence both violent and acquisitive crime.
Two pathways were tested in this study: a target pathway, which merged the gang selection and influence effects (RCT-0 → Gang-1 → PCT-2), and a control pathway, which reversed the independent and dependent variables (PCT-0 → Gang-1 → RCT-2). It was hypothesized that the target pathway would be significant, the control pathway would be nonsignificant and the two pathways would differ significantly from one another. A supplemental analysis was then performed to determine whether these results held up when violent and acquisitive offending were appended to the target pathway to form a fourth wave.
The sample used in this study came from the Pathways to Desistance study (Mulvey, 2012). All 1,354 members of the Pathways study (1,170 males, 184 females) were included in the current investigation, each of whom had been adjudicated delinquent or convicted of a crime in Philadelphia, PA or Phoenix, AZ. In addition, the offense leading to enrollment in the Pathways study had to have been committed when the participant was between the ages of 14 and 17. The average age of participants at the time of the baseline interview was 16.04 years (SD = 1.14), and the ethnic breakdown of the sample was 20.2% white, 41.4% black, 33.5% Hispanic and 4.8% other. About 80% of the juveniles approached about participating in the Pathways study agreed to do so. Enrollment in the study began in November 2000, the final baseline interviews were held in March 2003, and data collection was complete by March 2010.
Juvenile participants in the Pathways study provided their signed informed assent or consent to participate in the Pathways study. In addition, parental permission was obtained for participants who were younger than 18 years of age at the start of the study. The first three waves of the Pathways study – baseline (Wave 0), the first follow-up six months later (Wave 1), and the second follow-up six months after the first (Wave 2) – were used in the three-wave analysis. The third follow-up (Wave 3), which occurred six months after the second follow-up, was included in a four-wave analysis. Trained interviewers conducted the interviews in person or over the telephone and respondents either provided a verbal response or replied by typing their answers onto a keypad. The baseline interview was administered in two sections, each of which took approximately two hours to complete, and the three follow-up interviews required approximately two hours each to complete.
PCT and RCT served as the independent and dependent variables for this study. PCT was measured with the 32-item moral disengagement scale (MD: Bandura et al., 1996) as was done previously (Walters, 2016). Research indicates that the MD scale loads meaningfully onto a PCT latent factor (Walters and Yurvati, 2017). The MD scale is broken down into eight sections, each of which is designed to assess a different mechanism of MD: moral justification (“It is alright to fight to protect your friends”); euphemistic language (“Talking about people behind their backs is just part of the game”); advantageous comparison (“Stealing some money is not too serious compared to those who steal a lot of money”); displacement of responsibility (“Kids cannot be blamed for using bad words when all of their friends do it”); diffusion of responsibility (“It is unfair to blame a child who had only a small part in the harm caused by a group”); distorting consequences (“It is okay to tell small lies because they don’t really do any harm”); attribution of blame (“If people are careless where they leave their things it is their own fault if they get stolen”); and dehumanization (“Some people deserve to be treated like animals”). Each item on the MD scale is rated by the respondent on a three-point scale (1 = disagree, 3 = agree) and for the purposes of the current study an average score per item was calculated. The MD achieved good to excellent internal consistency during Waves 0 and 2 of this study (α = 0.88–0.91: Mulvey, 2012).
RCT was assessed with eight items from the impulse control (IC) scale of the Weinberger adjustment inventory (WAI: Weinberger and Schwartz, 1990), as was also done in Walters (2016). The WAI-IC was found to load onto a RCT latent factor in the previously mentioned Walters and Yurvati (2017) study. Each item on the WAI-IC (e.g. “I say the first thing that comes into my mind without thinking enough about it;” “I’m the kind of person who will try anything once, even if it’s not that safe”) is rated on a five-point scale (1 = False, 5 = True). As was done with the MD scale, an average score per item was calculated. Items that reflected good IC were recoded so that high scores on each item indicated poor IC. The WAI-IC achieved adequate to good internal consistency during Waves 0 and 2 of the Pathways study (α = 0.76–0.80: Mulvey, 2012).
The mediating variable in the current study was a simple measure of gang affiliation assessed at Wave 1. Participants were asked at baseline (Wave 0) whether they had ever been involved in a gang. At Wave 1, they were asked if they had been involved in a gang in the past six months. The recall periods for these two measures of gang affiliation did not overlap and for this reason were treated as separate variables. Follow-up questions were asked (e.g. does the gang have a name, does the gang have any associated colors, are there specific rules for socialization among gang members) but gang affiliation was determined solely on the basis of an affirmative answer to the following questions: “were you ever in a gang” (Wave 0); “did you join a gang, or have you been a member of a gang at any time over the past six months?” (Wave 1). Both variables were dichotomously scored (1 = yes, 0 = no).
A fourth wave consisting of violent and acquisitive crime was added to the three-wave model to create a four-wave model (RCT-0 → Gang-1 → PCT-2 → Offending-3). This four-wave target model was then evaluated in a four-wave supplemental analysis. Self-reported offending was assessed in the Pathways study with the self-report of offending (SRO) inventory (Huizinga et al., 1991), a commonly used measure of adult and juvenile offending. The SRO was used to create two Wave 3 outcome measures, a violent (aggressive) offending scale and an acquisitive (income) offending scale. The violent offending scale encompassed 11 offenses (destroyed/damaged property, set fire, forced someone to have sex, killed someone, shot someone, shot at someone, took by force with a weapon, took by force without a weapon, beat up someone with serious injury, in a fight, and beat someone as part of a gang) and the acquisitive scale covered eight offenses (broke in to steal, shoplifted, bought/received/sold stolen property, used check/credit card illegally, stole car or motorcycle, sold marijuana, sold other drugs and been paid by someone for sex).
Variety scores – which have been found to be psychometrically superior to frequency and dichotomous measures in assessing crime (Sweeten, 2012) – were calculated for violent and acquisitive offenses by taking the number of crime categories reportedly engaged in during the recall period and dividing this number by the total number of possible categories (11 for violent offending and 8 for acquisitive offending). These type scores are commonly employed in research in criminology. Because the recall period varied modestly to moderately between participants, time at risk (in months) was included in the four-wave analysis as a control variable. The violent and acquisitive offense measures displayed moderate stability in the 18 months between Waves 0 and 3 (r = 0.26–0.29).
There were eight control variables included in the current investigation. Four of the control variables were demographic in nature: age at baseline (in years), sex (1 = male, 2 = female), race (1 = White, 2 = non-White) and family structure (1 = no biological or adoptive parents in the home, 2 = single-parent home, 3 = two-parent home). Four non-demographic indicators were also included as control variables in this study: non-gang peer delinquency, parental knowledge, academic strain and total offending.
Non-gang peer delinquency was the first non-demographic control variable included in this study. This construct was measured with the 12-item peer delinquency-antisocial behavior scale (Thornberry et al., 1994). In completing the scale, respondents rate the proportion of friends involved in various delinquent acts (e.g. “purposely damaged or destroyed property that did not belong to them”) on a five-point scale (1 = none of them, 2 = very few of them, 3 = some of them, 4 = most of them, 5 = all of them). The mean rating across the 12 individual items was then calculated, after which it was adjusted for the proportion of friends, gang-involved participants estimated were gang members. If most or all of the individual’s friends were gang members, the non-gang peer delinquency score was reset to 1.0 (the low end of the scale); if half of the individual’s friends were gang members, the peer delinquency score was reduced by half or reset to 1.0, whichever was higher; if the individuals’ friends were mostly or all non-gang members, the peer delinquency score was not adjusted. Internal consistency for the peer delinquency scale at Wave 0 of the Pathways study was excellent (α = 0.92: Mulvey, 2012).
Parental knowledge was assessed with the parental knowledge scale of the parental monitoring inventory (PMI: Steinberg et al., 1992). Each of the five items on the PMI (“How much does your primary caregiver know about […] who you spend time with […] how you spend your free time […] how you spend your money […] where you go after school […] where you go at night”) was rated on a four-point scale (1 = doesn’t know at all, 2 = knows a little bit, 3 = knows a lot, 4 = knows everything) and the scores averaged to produce a mean score per item. Internal consistency (α) for the PMI parental knowledge scale ranged from 0.84 to 0.87 in the Pathways study (Mulvey, 2012).
The third non-demographic control variable included in this study was a derived measure of academic strain. Participants in the Pathways study were asked two questions: “How far would you LIKE to go in school?” and “How far do you THINK you will go in school?” These two questions were designed to measure academic aspirations and academic expectations, respectively, and were scored using the same five-point scale (1 = drop out before high school graduation, 2 = graduate from high school, 3 = go to business/technical school or junior college, 4 = graduate from college, 5 = go to graduate or professional school). The expectation rating was then subtracted from the aspiration rating (aspiration – expectation) to create an academic strain index, with higher scores indicating greater strain.
The fourth and final non-demographic control measure was prior offending. Total offending (calculated as the sum of the violent and acquisitive offending variety scores) at Wave 0 served as the control variable in the three-wave analysis. Violent and acquisitive offending at Wave 0 both served as control variables in the four-wave analysis.
Strong causal statements can only be made in the presence of a true experiment. Unfortunately, most questions in criminology cannot be answered with a true experiment. Non-experimental data are therefore required and under certain conditions can support modest causal inferences. The two most important supporting conditions are the temporal order and direction of variables in a study. The use of prospective data from a longitudinal study, a condition satisfied in the current investigation, helps establish the causal order of variables but does not ensure causal direction. To verify the causal direction of variables in a regression analysis, precursor measures of all predicted variables are required (Cole and Maxwell, 2003). In the current study, a prior measure of the mediator (gang affiliation at Wave 0) was included as a predictor of gang affiliation at Wave 1 and prior measures of both dependent variables (PCT and RCT at Wave 0) were included as predictors of PCT and RCT at Wave 2. In truth, the Wave 0 measures of PCT and RCT were already built into the design as independent variables. The use of precursor measures for each of the predicted variables turned the predicted variables into lagged outcome measures.
The current study used a non-experimental fixed-sample longitudinal panel design to explore the relationship between criminal thinking and gang affiliation. In this design, the independent and control variables were assessed at Wave 0, the mediator variable was assessed at Wave 1, and the two dependent variables were assessed at Wave 2. Four-wave models were also tested in which Wave 3 offending data were appended to each of the three-wave models. There was a six-month gap between successive waves of data across the first four waves of the pathways study, and there was no overlap between waves. This means that the current study qualified as prospective in nature and was capable of establishing the temporal order of variables. Implementing this design, path analyses composed solely of manifest variables were conducted. The data used in this study were originally collected between 2000 and 2010 and permission to use them in a secondary data analysis was granted by the Kutztown University Institutional Review Board.
A path analysis was performed with a maximum likelihood (ML) estimator. ML uses a logistic model to assess categorical outcome variables like the gang affiliation mediator included in the present study. Indirect effects were evaluated against bias-corrected bootstrapped confidence intervals (5,000 bootstrapped replications). A 95% confidence interval that did not include zero was used to determine significance. Research indicates that non-parametric bootstrapped confidence intervals are superior to normal theory procedures in evaluating indirect effects and controlling for non-normality in the dependent variable (Hayes, 2018). One theoretically derived target pathway (RCT-0 → Gang-1 → PCT-2) and one control pathway constructed by reversing the independent and dependent variables (PCT-0 → Gang-1 → RCT-2) were investigated and compared, with comparisons being made using Preacher and Hayes (2008) contrast test. The path analyses were performed with MPlus 8.3 (Muthén and Muthén, 1997-2017).
Two sensitivity tests were performed in an effort to rule out omitted variable and endogenous selection bias as alternate explanations for the current results. Omitted variable bias was assessed using Kenny’s (2013) “failsafe ef” procedure: (rmy.x) × (sdm.x) × (sdy.x)/(sdm) × (sdy). The coefficient produced by this procedure specifies the degree to which an unmeasured covariate confounder would need to correlate with both the mediator and dependent variables, controlling for the independent and mediating variables in the case of the latter, to completely eliminate the coefficient along the b path (from mediator to dependent variable) of the significant indirect effect. Endogenous selection bias, also known as a collider effect, was tested based on the realization that conditioning on a precursor to a predicted variable, as was done in the current study, can lead to overestimated path coefficients (Elwert and Winship, 2014). This sensitivity test was carried out by recalculating the analyses without the precursor measures to determine whether the path coefficients decreased once the precursors were removed.
A good portion of the sample had complete data on all 14 variables (n = 1,075, 79.4%). Another 12.7% of participants were missing data on one variable, 3.7% were missing data on two variables, 4.2% were missing data on three to five variables, and 0.1% were missing data on eight variables. There were four variables that had more than 5% missing data: parental knowledge (5.2%), Wave 1 gang affiliation (6.9%), Wave 2 PCT (6.9%) and Wave 2 RCT (6.9%). Missing data were handled with full information maximum likelihood (FIML). The FIML procedure uses all available information to estimate standard errors and population parameters for the entire sample. Researchers have found that FIML produces significantly more accurate results than traditional missing value procedures like listwise deletion and simple imputation (Allison, 2002).
Descriptive statistics and intercorrelations for the 14 variables included in the three-wave analysis are listed in Table 1. Approximately two-fifths of the intercorrelations were significant using a Bonferroni-corrected alpha level (p = 0.00064). Multicollinearity diagnostics detected no evidence of excessive inter-correlation between predictor variables in any of the regression equations (tolerance = 0.764–0.982, variance inflation factor = 1.019–1.309).
Table 2 summarizes the results of a three-regression equation path analysis of the three-wave target and control models. Bias-corrected bootstrapped confidence intervals for the total, direct and indirect effects are then listed in Table 3. As predicted, the indirect effect for the target pathway was significant (i.e. the 95% confidence interval did not include zero), whereas the indirect effect for the control pathway was not (i.e. the 95% confidence interval included zero). Contrary to predictions, the difference in indirect effects between the target and control pathways was nonsignificant. Interestingly, the peer influence and gang influence effects were both observed in the regression equation predicting Wave 2 PCT.
A re-analysis was conducted using the unadjusted peer delinquency measure (gang and non-gang peer delinquency combined). The results indicated that neither the a (Z = 1.46, p = 0.14) nor b (Z = 1.84, p = 0.07) paths of the target pathway were significant and the total indirect effect was nonsignificant as well (Estimate = 0.0007, 95% BCBCI = −0.0001, 0.0028). These results suggest that the gang influence effect, like the peer influence effect, operates interpersonally, either through individual contacts (non-gang peer influence effect) or collective associations (gang influence effect). Removing peer delinquency from the analysis altogether produced results similar to those obtained with the adjusted peer delinquency measure.
Sensitivity testing designed to assess the likelihood of omitted variable bias produced a “failsafe ef” of 0.16. This indicates that a confounding covariate would need to correlate 0.16 with Gang-1 and 0.16 with PCT-2, controlling for RCT-0 and Gang-1 in the case of PCT-2, to completely eliminate the significant b path of the target pathway. Sensitivity testing designed to assess the likelihood of endogenous selection bias revealed that the path coefficients increased rather than decreased when the precursor measures were removed from the equation, a finding inconsistent with the presence of a collider variable or endogenous selection bias.
Wave 3 violent and acquisitive offending were appended to the three-wave models in an attempt to create a complete gang influence effect (RCT-0 → Gang-1 → PCT-2 → Offending-3). The consequent results were similar to those obtained with the three-wave models. The target pathway successfully predicted both violent (Estimate = 0.00010, 95% BCBCI = 0.00002, 0.00027) and acquisitive (Estimate = 0.00012, 95% BCBCI = 0.00003, 0.00034) offending, whereas the control pathway failed to predict either violent (Estimate = 0.00005, 95% BCBCI = −0.00003, 0.00027) or acquisitive (Estimate = 0.00006, 95% BCBCI = −0.00003, 0.00036) crime. The path diagram containing the target pathway is reproduced in Figure 1. Not only were each of the path coefficients in the target pathways significant, but the coefficients for the paths running from Wave 1 gang affiliation to Wave 3 violent and acquisitive offending were also significant (see Figure 1).
The hypothesis tested in this study held that RCT would lead to gang affiliation and that gang affiliation would lead to PCT, but that the reverse pattern (i.e. PCT leading to gang affiliation and gang affiliation leading to RCT) would prove nonsignificant. Both individual path coefficients and the total indirect effects in a causal mediation analysis confirmed these predictions, although the two pathways did not differ significantly from one another. These results did not change when offending was added as a fourth wave to the standard three-wave model and sensitivity testing revealed that the b path (from mediator to dependent variable) of the significant target pathway was modestly robust to omitted variable bias and that endogenous selection bias could not explain the results. Taken as a whole, these findings suggest that a gang selection effect, in which the impulsive and thrill-seeking features of RCT encourage youth to affiliate in gangs, and a gang influence effect, in which the instrumental aspects of PCT are learned in association with other gang members, could operate simultaneously, even after controlling for academic strain, parental knowledge, prior offending and non-gang peer delinquency. Whereas the gang selection and influence effects may run parallel to the peer selection and influence effects, their ability to shape and inform the cognitive skills that support crime may be strengthened further by the group process.
Several factors limit the conclusions that can be drawn from this study. One such limitation is that the sample was composed of seriously delinquent youth. Because of this, it is uncertain how well the results generalize to less serious delinquents and at-risk non-delinquents. This study was also limited by the fact that two key variables, academic strain and gang affiliation, were measured with single, dichotomous indicators. It would have been better to assess these constructs with multi-item scales, but such measures were unavailable in the Pathways database. Also, gang selection and influence effects likely operate at least partially through group process, but it was not possible to examine these processes using the data available in the Pathways study. A fourth limitation of this study is the small size of the path coefficients and indirect effects. What needs to be understood is that mediation effects are nearly always small (Kenny and Judd, 2014; Walters, 2019b). The diminutiveness of mediation effects can be ascribed, in part, to the use of prospective data and lagged predicted outcome measures designed to establish the temporal order and temporal direction, respectively, of the variables in a study, and, in part, to the fact that indirect effects are a product of two or three individual effects. This was why the total indirect effect estimates for the four-wave model were smaller than the total indirect effect estimates for the three-wave model, even though the b path (from second mediator, PCT-2, to dependent variables, violent and acquisitive offending) coefficients in the four-wave model were nearly twice the size of the path coefficients from the three-factor model.
There are several potentially important implications to the current results that deserve mention. First, there are the implications these results have for criminal thinking’s role in future research on delinquency and crime. Criminal thinking, as is true of many social cognitive variables, is normally treated as a mediating variable. In research on the peer and gang selection and influence effects, for instance, RCT has been identified as a consistent mediator of the peer/gang selection effect, and PCT has been identified as a potent mediator of the peer/gang influence effect (Walters, 2015, 2016, 2019a). From a theoretical standpoint, the current results indicate that while formal strain may not be a powerful motive for youth gang affiliation, related processes, such as RCT, may well be. “To be with other people like you” was one reason youth gave for joining a gang in the Lachman et al. (2013) study. A gang selection effect, driven, in part, by RCT, may therefore be an important motivating factor for those who eventually enter into gang activities. Melde and Esbensen (2011) note that joining a gang leads to associated changes in attitudes, routine activities and crime propensity, some of which does not change once the individual leaves the group. In the current study, gang affiliation gave rise to a gang influence effect that led to heightened violent and acquisitive offending. Thus, while the motive to join a gang is frequently positive, as indicated by the results of studies using both quantitative (Lachman et al., 2013) and qualitative (Sanders, 1994) methodologies, the outcome is often negative (i.e. increased violent and acquisitive offending) and spurred by a rise in PCT.
A practical implication of the current results is that they identify two separate targets for intervention. The first target is the presumed mediator of the gang selection effect – i.e. RCT. To the extent that RCT functions as a risk factor for future gang involvement, managing this dimension of criminal thought process could help divert some youth away from a gang trajectory. The impulsive, irresponsible, reckless and disinhibited nature of RCT may best be managed with cognitive-behavioral skills training. Research, in fact, indicates that skill-based programming can protect at-risk youth by increasing social, problem solving and anger management skills (Eddy et al., 2003; van der Stouwe et al., 2016). Assessing youth on a measure of RCT and providing those who produce elevated scores with cognitive-behavioral skills training would appear to be an effective means of secondary prevention. The other target for intervention identified in this study is the presumed mediator of the gang influence effect – i.e. PCT. Treating PCT will likely require a tertiary prevention approach, greater resources and less favorable outcomes than the cognitive-behavioral skills training that is recommended for RCT. Callous-unemotional traits, such as poor empathy, are difficult, although not necessarily impossible, to change. Just the same, these traits can be used to identify associated problems like trauma exposure that may lend themselves better to intervention and change than the more stable callous-unemotional traits (Mendez et al., 2020).
Directions for future research
The current results indicate that the gang selection and influence effects operate in a manner similar to that of the peer selection and influence effects to the extent that both selection effects are driven by RCT and both influence effects are mediated by PCT. Moreover, the gang selection and influence effects persisted in the current study even after non-gang peer delinquency was controlled, indicating that the two sets of effects were not redundant to one another. Additional research is required to identify the group processes responsible for the gang selection and influence effects and whether gang and peer affiliations produce an additive or interactive effect when combined.
Path analysis of the relationship between reactive criminal thinking at Wave 0 and violent and acquisitive offending at Wave 3 via gang affiliation at Wave 1 and proactive criminal thinking at Wave 2, along with the accompanying peer influence effect (non-gang peer delinquency to proactive criminal thinking to offending)
Descriptive statistics and correlations for the 14 variables included in the present study
|4. Family structure||1354||2.17||0.49||1–3||0.04||0.05||0.02||−0.01||0.03||0.06||−0.01||−0.01||0.04||0.01|
|5. Peer delinquency||1322||2.06||0.91||1–5||−0.15†||−0.03||0.26†||0.18†||0.14†||−0.18†||−0.21†||0.13†||0.10|
|6. Parental knowledge||1284||2.70||0.81||1–4||−0.05||−0.26†||−0.21†||−0.17†||−0.15†||−0.13†||−0.13†||−0.11†|
|7. Academic strain||1339||0.34||0.80||−3–4||0.08||0.08||0.09||0.04||0.04||0.07||0.07|
|8. Total offending-0||1351||0.35||0.32||0.00–1.79||0.34†||0.36†||0.34†||0.29†||0.20†||0.23†|
Variable = variable name, Age = chronological age in years, Sex = male (1) vs female (2), Race = white (1) vs non-White (2), Family Structure = 3-level scale of family structure (1 = no parenting figure in the home, 2 = one-parent home, 3 = two-parent home) at baseline (Wave 0), Peer Delinquency = non-gang peer delinquency at Wave 0, Parental Knowledge = parental knowledge of child’s friends and whereabouts as reported by child at Wave 0, Academic Strain = difference between academic aspirations and academic expectations, Total Offending-0 = sum of violent and acquisitive crime at Wave 0, PCT-0 = proactive criminal thinking at Wave 0, RCT-0 = reactive criminal thinking at Wave 0, Gang-0 = gang affiliation at Wave 0, Gang-1 = gang affiliation at Wave 1, PCT-2 = proactive criminal thinking at Wave 2, RCT-2 = reactive criminal thinking at Wave 2, M = mean, SD = standard deviation, Range = range of scores in current sample.
p < 0.00055 (Bonferroni-corrected alpha: 0.05/91 correlations)
Three-equation path analysis of the proactive–reactive and reactive–proactive criminal thinking cross-lags as mediated by gang affiliation
|Variables||b (95% BCBCI)||β||Z||P|
|Outcome = Gang-1|
|Age||0.001 (−0.013, 0.014)||0.002||0.08||0.937|
|Sex||−0.022 (−0.058, 0.016)||−0.022||−1.14||0.256|
|Race||0.079 (0.046, 0.112)||0.095||4.72||<0.001|
|Family structure||0.008 (−0.025, 0.042)||0.012||0.50||0.619|
|Peer delinquency||−0.073 (−0.096, −0.049)||−0.198||−6.01||<0.001|
|Parental knowledge||−0.012 (−0.035, 0.009)||−0.030||−1.10||0.273|
|Academic strain||−0.005 (−0.026, 0.016)||−0.010||−0.42||0.670|
|Total offending-0||0.152 (0.081, 0.216)||0.145||4.48||<0.001|
|PCT-0||0.536 (−0.019, 1.085)||0.056||1.92||0.054|
|RCT-0||0.225 (0.060, 0.394)||0.064||2.63||0.009|
|Gang-0||0.340 (0.286, 0.400)||0.428||11.57||<0.001|
|Outcome = PCT-2|
|Age||0.002 (−0.014, 0.017)||0.007||0.26||0.792|
|Sex||−0.101 (−0.145, −0.057)||−0.097||−4.51||<0.001|
|Race||0.023 (−0.018, 0.064)||0.026||1.12||0.262|
|Family structure||0.014 (−0.019, 0.049)||0.020||0.83||0.407|
|Peer delinquency||0.025 (0.001, 0.048)||0.063||2.08||0.037|
|Parental knowledge||−0.004 (−0.026, 0.019)||−0.010||−0.37||0.711|
|Academic strain||0.013 (−0.010, 0.037)||0.027||1.08||0.282|
|Total offending-0||−0.027 (−0.091, 0.040)||−0.024||−0.79||0.428|
|Gang-1||0.089 (0.025, 0.152)||0.084||2.73||0.006|
|RCT-0||0.193 (−0.004, 0.403)||0.051||1.86||0.062|
|PCT-0||4.682 (4.035, 5.361)||0.461||13.74||<0.001|
|Outcome = RCT-2|
|Age||−0.001 (−0.042, 0.038)||−0.002||−0.07||0.944|
|Sex||−0.053 (−0.206, 0.086)||−0.019||−0.72||0.469|
|Race||−0.290 (−0.401, −0.179)||−0.122||−5.09||<0.001|
|Family structure||−0.066 (−0.150, 0.025)||−0.034||−1.44||0.149|
|Peer Delinquency||0.017 (−0.043, 0.076)||0.016||0.57||0.568|
|Parental knowledge||−0.009 (−0.071, 0.051)||−0.007||−0.28||0.782|
|Academic strain||0.029 (−0.032, 0.086)||0.023||0.95||0.341|
|Total offending-0||0.005 (−0.158, 0.161)||0.002||0.06||0.949|
|Gang-1||0.066 (−0.097, 0.220)||0.023||0.82||0.414|
|PCT-0||3.406 (1.980, 4.862)||0.126||4.54||<0.001|
|RCT-0||4.459 (3.873, 4.996)||0.444||15.47||<0.001|
|RCT-2 with PCT-2||0.063 (0.049, 0.079)||0.260||8.45||<0.001|
Outcome = outcome measure, Age = chronological age in years, Sex = male (1) vs female (2), Race = White (1) vs non-White (0), Family Structure = 3-level scale of family structure (1 = no parenting figure in the home, 2 = one-parent home, 3 = two-parent home) at baseline (Wave 0), Peer Delinquency = non-gang peer delinquency at Wave 0, Parental Knowledge = parental knowledge of child’s friends and whereabouts as reported by child at Wave 0, Academic Strain = difference between academic aspirations and academic expectations at Wave 0, Total Offending-0 = sum of violent and acquisitive crime at Wave 0, PCT-0 = proactive criminal thinking at Wave 0, RCT-0 = reactive criminal thinking at Wave 0, Gang-0 = gang affiliation at Wave 0, Gang-1 = gang affiliation at Wave 1, PCT-2 = proactive criminal thinking at Wave 2, RCT-2 = reactive criminal thinking at Wave 2, with = covariance, b (95% BCBCI) = unstandardized coefficient and 95% biased-corrected bootstrapped confidence interval (in parentheses), β = standardized coefficient, Z = Wald Z test statistic, p = significance level of the Wald Z test statistic, N = 1,354
Total, direct and indirect effects of gang affiliation on the reactive criminal thinking–proactive criminal thinking and proactive criminal thinking–reactive criminal thinking relationships
|Effects from RCT-0 to PCT-2|
|Total indirect effect||0.0020||0.0004||0.0060|
|Specific indirect effect|
|RCT-0 → Gang-1 → PCT-2||0.0020||0.0004||0.0060|
|Effects from PCT-0 to RCT-2|
|Total indirect effect||0.0036||−0.0037||0.0223|
|Specific indirect effect|
|PCT-0 → Gang-1 → RCT-2||0.0036||−0.0037||0.0223|
|Preacher–Hayes contrast test||0.0042||−0.0020||0.0112|
RCT-0 = reactive criminal thinking measured at baseline (Wave 0), PCT-0 = proactive criminal thinking measured at Wave 0; Gang-1 = gang affiliation at Wave 1; PCT-2 = proactive criminal thinking measured at Wave 2; RCT-2 = reactive criminal thinking measured at Wave 2; Preacher–Hayes Contrast Test = comparison between the two pathways using Preacher and Hayes’ (2008) contrast test with standardized outcome measures; BCBCI = bias-corrected bootstrapped 95% confidence interval (b = 5000); Estimate = point estimate; Lower = lower boundary of the 95% confidence interval; Upper = upper boundary of the 95% confidence interval; N = 1,354
Akers, R.L. (1998), Social Learning and Social Structure: A General Theory of Crime and Deviance, Northeastern University Press, Boston.
Bandura, A., Barbaranelli, C., Caprara, G.V. and Pastorelli, C. (1996), “Mechanisms of moral disengagement in the exercise of moral agency”, Journal of Personality and Social Psychology, Vol. 71 No. 2, pp. 364-374.
Brechwald, W.A. and Prinstein, M.J. (2011), “Beyond homophily: a decade of advances in understanding peer influence processes”, Journal of Research on Adolescence, Vol. 21 No. 1, pp. 166-179.
Brownfield, D. (2014), “Testing a subcultural theory of crime and delinquency in a gang context”, Journal of Gang Research, Vol. 21 No. 4, pp. 1-9.
Cole, D.A. and Maxwell, S.E. (2003), “Testing mediational models with longitudinal data: questions and tips in the use of structural equation modeling”, Journal of Abnormal Psychology, Vol. 112 No. 4, pp. 558-577.
Eddy, J.M., Reid, J.B., Stoolmiller, M. and Fetrow, R.A. (2003), “Outcomes during middle school for an elementary school-based prevention intervention for conduct problems: follow-up results from a randomized trial”, Behavior Therapy, Vol. 34 No. 4, pp. 535-552.
Elwert, F. and Winship, C. (2014), “Endogenous selection bias: the problem of conditioning on a collider variable”, Annual Review of Sociology, Vol. 40 No. 1, pp. 31-53.
Esbensen, F.-A. (2002), National Evaluation of the Gang Resistance Education and Training (G.R.E.A.T.) Program: Final Report (94-IJ-CX-0058), U.S. Department of Justice, Washington, DC.
Gifford-Smith, M., Dodge, K.A., Dishion, T.J. and McCord, J. (2005), “Peer influence in children and adolescents: crossing the bridge from developmental to intervention science”, Journal of Abnormal Child Psychology, Vol. 33 No. 3, pp. 255-265.
Gottfredson, M.R. and Hirschi, T. (1990), A General Theory of Crime, Stanford University Press, Stanford, CA.
Hayes, A.F. (2018), Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach, 2nd ed., Guilford, New York, NY.
Huizinga, D., Esbensen, F. and Weihar, A. (1991), “Are there multiple paths to delinquency?”, The Journal of Criminal Law and Criminology, Vol. 82 No. 1, pp. 83-118.
Jose, R., Hipp, J.R., Butts, C.T., Wang, C. and Lakon, C.M. (2016), “Network structure, influence, selection, and adolescent delinquent behavior”, Criminal Justice and Behavior, Vol. 43 No. 2, pp. 264-284.
Kenny, D.A. and Judd, C.M. (2014), “Power anomalies in testing mediation”, Psychological Science, Vol. 25 No. 2, pp. 334-339.
Kenny, D.A. (2013), “Mediation: sensitivity analysis [webinar]”, available at: http://davidakenny.net/webinars/Mediation/Sensitivity/Sensitivity.html.
Kiesner, J., Poulin, F. and Nicotra, E. (2003), “Peer relations across contexts: individual-network homophily and network inclusion in and after school”, Child Development, Vol. 74 No. 5, pp. 1328-1343.
Lachman, P., Roman, C.G. and Cahill, M. (2013), “Assessing youth motivations for joining a peer group as risk factors for delinquent and gang behavior”, Youth Violence and Juvenile Justice, Vol. 11 No. 3, pp. 212-229.
Melde, C. and Esbensen, F.-A. (2011), “Gang membership as a turning point in the life course”, Criminology, Vol. 49 No. 2, pp. 513-552.
Mendez, L., Mozley, M.M. and Kerig, P.K. (2020), “Associations among trauma exposure, callous-unemotionality, race or ethnicity, and gang involvement in justice-involved youth”, Criminal Justice and Behavior, Vol. 47 No. 4, pp. 457-469.
Mulvey, E.P. (2012), “The pathways to desistance study: design and methods”, Paper presented at the American Society of Criminology Annual Meeting, Chicago, IL, available at: http://citation.allacademic.com/meta/p574246_index.html (accessed 3 June 2013).
Muthén, B. and Muthén, L. (1997-2017), Mplus User’s Guide, 8th ed., Muthén and Muthén, Los Angeles, CA.
Osgood, D.W., Feinberg, M.E. and Ragan, D.T. (2015), “Social networks and the diffusion of adolescent problem behavior; reliable estimates or selection and influence from sixth through ninth grade”, Prevention Science, Vol. 16 No. 6, pp. 832-843.
Preacher, K.J. and Hayes, A.F. (2008), “Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models”, Behavior Research Methods, Vol. 40 No. 3, pp. 879-891.
Rosenfeld, R. (2009), “Crime is the problem: homicide, acquisitive crime, and economic conditions”, Journal of Quantitative Criminology, Vol. 25 No. 3, pp. 287-306.
Sanders, W.B. (1994), Gangbangs and Drive-Bys: Grounded Culture and Juvenile Gang Violence, Routledge, New York, NY.
Steinberg, L., Dornbusch, S. and Darling, N. (1992), “Impact of parenting practices on adolescent achievement. Authoritative parenting, school involvement, and encouragement to succeed”, Child Development, Vol. 63 No. 5, pp. 1266-1281.
Sutherland, E.H. (1947), Principles of Criminology, 4th ed., J.B. Lippincott, Philadelphia.
Sweeten, G. (2012), “Scaling criminal offending”, Journal of Quantitative Criminology, Vol. 28 No. 3, pp. 533-557.
Teneyck, M. and Barnes, J.C. (2015), “Examining the impact of peer group selection on self-reported delinquency: a consideration of active gene–environment correlation”, Criminal Justice and Behavior, Vol. 42 No. 7, pp. 741-762.
Van der Stouwe, T., Asscher, J.J., Hoeve, M., van der Laan, P.H. and Stams, G.J.J.M. (2016), “Social skills training for juvenile delinquents: post-treatment changes”, Journal of Experimental Criminology, Vol. 12 No. 4, pp. 515-536.
Vigil, J.D. (2003), “Urban violence and street gangs”, Annual Review of Anthropology, Vol. 32 No. 1, pp. 225-242.
Walters, G.D. (2015), “Proactive criminal thinking and the transmission of differential association: a cross-lagged multi-wave path analysis”, Criminal Justice and Behavior, Vol. 42 No. 11, pp. 1128-1144.
Walters, G.D. (2016), “Friends, cognition, and delinquency: proactive and reactive criminal thinking as mediators of the peer influence and peer selection effects among male delinquents”, Justice Quarterly, Vol. 33 No. 6, pp. 1055-1079.
Walters, G.D. (2019a), “Gang influence: mediating the gang–delinquency relationship with proactive criminal thinking”, Criminal Justice and Behavior, Vol. 46 No. 7, pp. 1044-1062.
Walters, G.D. (2019b), “Why are mediation effects so small?”, International Journal of Social Research Methodology, Vol. 22 No. 2, pp. 219-232.
Walters, G.D. and Yurvati, E. (2017), “Testing the construct validity of the PICTS proactive and reactive scores against six putative measure of proactive and reactive criminal thinking”, Psychology, Crime and Law, Vol. 23 No. 1, pp. 1-14.
Weinberger, D.A. and Schwartz, G.E. (1990), “Distress and restraint as superordinate dimensions of self-reported adjustment: a typological perspective”, Journal of Personality, Vol. 58 No. 2, pp. 381-417.
Zimring, F.E. and Hawkins, G. (1997), Crime is Not the Problem: lethal Violence in America, Oxford University Press, New York, NY.
About the author
Glenn D. Walters is based at the Department of Criminal Justice, Kutztown University of Pennsylvania, Kutztown, Pennsylvania, USA. He, PhD, is a Professor in the Department of Criminal Justice at Kutztown University where he teaches classes in corrections, criminology, research methods and substance misuse and crime. His principal research interests include bullying, mediation analysis and the development of an overarching psychological theory of crime.