Examining gang membership and victimisation as predictors of gun violence perpetration: testing for interactive effects

ABSTRACT

Research has indicated the relevance of gang membership and victimisation as risk factors for predicting the perpetration of gun violence. However, there is a dearth of research that has examined the potential that gang membership may moderate the impact of victimisation on gun violence perpetration. This study examined this prospect within a sample of justice-involved youth. The Pathways to Desistance dataset was analysed. Generalised estimating equations were used to test for direct and interactive effects of gang membership and victimisation while accounting for nested error corresponding to the repeated measures in this 11-wave longitudinal dataset. Results indicated that gang membership was associated with an increased risk of gun violence perpetration, but not victimisation. The hypothesised interaction effect was also nonsignificant.

KEYWORDS:

• Gangs
• victimization
• gun violence
• justice-involved youth
• generalized estimating equations

Introduction

Gun violence presents a major public health and criminal justice issue facing the present-day United States (Cook, 2018; Johnson et al., 2021). With political division making legislative change regarding restrictions on gun ownership difficult, this puts a large emphasis on identifying additional means of reducing gun violence through targeting risk and protective factors through treatment programming. One robust risk factor for gun violence perpetration that has been identified consistently in the literature is victimisation (McGee et al., 2017; Rowan et al., 2019; Vossekuil, 2002). Individuals who have been exposed to violence demonstrate an elevated risk of perpetrating gun violence, but our understanding of the nuances of this relationship remains limited. For example, there is limited understanding of how certain individuals may be more or less affected by violence exposure that may result in an increased or diminished likelihood of gun violence perpetration when experiencing victimisation. Gang membership has also been identified as a risk factor for gun violence perpetration (Braga, 2003; Pardini et al., 2021), though additional research in this area is needed. It may be that there exists an interactive effect between victimisation and gang membership, as gang membership may moderate the relationship between victimisation and gun violence perpetration. That said, prior research has yet to examine these constructs in this manner. The present study sought to address this gap in the literature by examining the direct effects of victimisation and gang membership on the risk of gun violence perpetration and determine whether or not gang membership moderates the effect of victimisation on gun violence perpetration among a sample of justice-involved youth (JIY) in the United States. This study has the potential to add to the extant literature on the topic by examining how these risk factors may produce an interactive effect on gun violence risk. Significant findings in this area would allow for more effective targeting of treatment resources to try to prevent gun violence, as it would identify priority populations among JIY for doing so.

Direct effects of gang membership and victimization on gun violence perpetration among JIY

As mentioned above, prior research has indicated that gang membership and victimisation are both risk factors for the perpetration of gun violence (Braga, 2003; McGee et al., 2017; Pardini et al., 2021; Rowan et al., 2019; Vossekuil, 2002). In terms of victimisation, there may be several mechanisms that explain this relationship. Prior research has indicated that people report carrying guns following victimisation as a means of self-defence and protection against victimisation and that fear of victimisation drives this increased risk (Fontaine et al., 2018; Logan & Lynch, 2022). In this way, youth may seek to alter the suitability of a target by carrying a weapon in a manner consistent with Cohen and Felson’s (1979) routine activities theory, thus, reducing their risk of experiencing future exposures to violence. However, it should be noted that generalisability of the highlighted studies may be limited here given that they focused on populations that differ substantially from the JIY sample examined here (e.g., adult women studied in the Logan & Lynch, 2022 study). While carrying firearms remains uncommon among adolescents in general, rates of carrying firearms are substantially higher among those who have experienced victimisation vs. those who have not. Research places these estimates of firearm carrying prevalence between 2% and 19% among non-victimised youth and between 13% and 35% for victimised youth (Baiden et al., 2024; Ganson & Nagata, 2021; Lowry et al., 2023). While there exists a wide variance in estimates here, it remains clear that victimised youth tend to carry guns at much higher rates than do non-victimised youth. Second, prior research has indicated that trauma exposure like victimisation may lead to the onset of mental disorder like post-traumatic stress disorder (PTSD) (American Psychiatric Association, 2013). A key symptom of PTSD is hypervigilance, that is, constantly scanning the environment for additional threats and may perceive otherwise neutral stimuli as threatening (Frewen & Lanius, 2006). This may also result in increased sensation-seeking and impulse control (Arnsten et al., 2015; Kerig, 2019; Wojciechowski, 2022); with both being key features of PTSD (American Psychiatric Association, 2013; Brewerton et al., 2018; Sanchez et al., 2020). If survivors of such trauma are more likely to carry guns as a means of self-protection, they may do so to avoid subsequent traumatisation. In this way, individuals who experience victimisation may not only be at greater risk of carrying firearms as a means of self-defence, but hypervigilance may lead them to be more likely to actually use those guns when put in a situation that they find threatening, even if no real threat is present. JIY may represent a priority population in this regard, as these youth present increased risk for experiencing victimisation (Baglivio et al., 2014; Pereda et al., 2017). This indicates the need to examine the effects of this risk factor among this population in order to determine whether this helps to explain gun violence perpetration with a subgroup of youth who, by definition, present an increased propensity for involvement in antisocial behaviour.

Gang membership also represents a risk factor that may precipitate the perpetration of gun violence. As noted, statistics indicate that gangs and gang members are disproportionately responsible for gun violence (Braga, 2003; Pardini et al., 2021). This may be due to the need to maintain territorial control and dominate rival gangs (Stretesky & Pogrebin, 2007). Gangs may socialise youth into rationalising and justifying gun violence as a necessary condition for serving the higher loyalty to the gang in a manner consistent with Akers’s (1973) social learning theory, with research indicating the importance of social learning theory processes as it pertains to socialisation in gangs (Winfree & Freng, 2015). In these ways, gun violence may serve a distinct purpose in preserving the interests of gangs, with members carrying out this violence on behalf of the collective. Another important consideration here is that gang members are also more likely to experience victimisation than non-gang members (Peterson et al., 2004; Taylor et al., 2007; Webb et al., 2011), though there have been methodological concerns regarding evidence of this relationship (Gibson et al., 2012). This highlights the particular importance of this other risk factor for gun violence for understanding the effects of gang membership. Again, JIY may present a priority population for understanding the impact of this construct on gun violence risk. The elevated propensity for JIY to be involved in antisocial behaviour make them prime candidates for the types of behaviours carried out by gangs and research has indicated that they report gang involvement at high rates (Hill et al., 1999; Robertson et al., 2004). For both of these identified risk factors, JIY then present an important population worthy of examination in order to understand the antecedents of gun violence perpetration. The lack of research on the effects of victimisation on gun violence within the context of gangs highlights the need for continued study on this topic in order to better understand the elevated risk of gun violence among gang members.

Gang membership as a moderator of the relationship between victimization and gun violence perpetration

The fact that rates of victimisation are elevated among gang members highlights the potential that both factors should be considered for predicting gun violence perpetration (Braga, 2003; McGee et al., 2017; Pardini et al., 2021; Rowan et al., 2019; Vossekuil, 2002). While this may provide a causal mechanism through which gang membership affects the risk of perpetrating gun violence, it may also be that gang membership may moderate the effects of victimisation on the risk of gun violence perpetration. Gang membership may act to amplify the effects of victimisation in this regard. It may be that exposure to both of these risk factors results in a synergistic effect in which both elevate the risk for gun violence. Thornberry et al. (1993) provide a theoretical model that identifies the potential factors by which gangs lead to antisocial behaviour that may be pertinent here, with selection, facilitation, and enhancement components. Selection relates to the idea that certain pre-existing characteristics may predispose antisocial individuals to gang membership, facilitation relates to the idea of the gang context providing an environment that facilitates involvement in antisocial behaviour, and enhancement combines both of these previous models. There has been evidence observed by prior research that indicates that the enhancement model may be relevant (DeLisi et al., 2009; Wu & Pyrooz, 2016), as pre-existing characteristics may be amplified by the gang context to facilitate greater antisocial behaviour.

It may be that gang members who have been exposed to violence seek to further protect themselves from further traumatisation by using a gun as protection, as this has been highlighted as a reason for carrying firearms by survivors of violence (Fontaine et al., 2018; Logan & Lynch, 2022). This type of hypervigilant behaviour coupled with the increased risk of gun violence perpetration due to general gang activities may result in more instances during which the individual feels threatened and may use a gun. The cognitive impacts of victimisation on sensation-seeking and impulse control should also be considered here (Arnsten et al., 2015; Kerig, 2019; Wojciechowski, 2022). If victimisation results in lower impulse control and heightened sensation-seeking, then the gang context may exacerbate these effects. The activities of gang members related to territorial protection and general protection of their interests may place individuals like this in contexts in which gun violence may be more likely more often. If these individuals are also highly impulsive and/or demonstrate high levels of sensation-seeking due to previous victimisation (Arnsten et al., 2015; Kerig, 2019; Wojciechowski, 2022), then they may be more likely to react in these situations with gun violence when provoked. In these ways, gang membership should amplify the effects of victimisation that may manifest in gun violence.

While gang membership may indeed moderate the effects of victimisation by amplifying the effects of victimisation on gun violence perpetration, the moderation effect may also operate in the opposite direction. This is to say that gang membership may result in a diminished effect of victimisation on gun violence perpetration. Consider the fact that prior research has indicated that protective gun ownership/gun carrying is related to experiencing of victimisation (Spano & Bolland, 2013; Warner & Thrash, 2020). Also, consider that individuals have reported seeking safety and protection as a reason for joining a gang (Melde et al., 2009; Miller, 2001; Padilla, 1992); an especially curious relationship given the increased risk of victimisation among gang members (Peterson et al., 2004; Taylor et al., 2007; Webb et al., 2011). It may be that gang membership functions as a means of providing this sense of safety and protection even in the absence of carrying a gun. If this is the case, then fear and anxiety around future victimisation may be assuaged by feelings of protection by the gang and may mitigate feelings of need to carry a gun for self-defence, thus, reducing the probability that the individual will perpetrate gun violence. This would then result in a negative interaction in which gang membership still moderates the effect of victimisation on gun violence perpetration, but in the opposite direction as predicted above.

As noted above, victimisation and gang membership are both risk factors for the perpetration of gun violence (Braga, 2003; McGee et al., 2017; Pardini et al., 2021; Rowan et al., 2019; Vossekuil, 2002). Despite this, there remains a dearth of research which has examined the interaction between these constructs for predicting the behaviour. Given that both of these constructs may increase the risk for perpetration of gun violence, there exists the potential that there may be a positive interaction between the two, as gang membership moderates the impact of victimisation and amplifies the effects of such exposures on the risk of gun violence perpetration. Identification of a significant interaction effect here could facilitate more effective targeting of interventions for JIY who are gang members and who have or have not experienced victimisation. This could help to facilitate reductions in gun violence moving forward. However, there remains a dearth of research testing these propositions. The present study sought to address this gap in the extant literature by testing the following hypotheses:

Hypothesis 1:

Being a gang member and being exposed to violence will be associated with increased odds of gun violence perpetration.

Hypothesis 2:

Gang membership will amplify the impact of victimisation on gun violence perpetration, with gang members demonstrating a significantly stronger relationship between victimisation and gun violence perpetration relative to non-gang members.

Methods

Data

This study used data from all 11 waves of the Pathways to Desistance study. This dataset is comprised of responses from 1,354 JIY who were adjudicated for a serious offence in the United States just prior to baseline measurements. These JIY were then followed across the subsequent seven years, with observation periods spaced six months apart for the first 36 months that participants were involved with the study and every 12 months for the remainder of their time in the study period thereafter. Recruitment for the study occurred from 2000–2003, with the entire study period lasting from 2000–2010. Recruitment took place at study sites located in Philadelphia, Pennsylvania and Maricopa County, Arizona in the United States. Serious offences which resulted in adjudications that qualified participants for inclusion in the study consisted of all felony offences, as well as misdemeanour sexual assault and weapons charges. It should also be noted that a cap was applied to the total number of male drug offenders who were included in the study at 15% of the baseline sample. This was done in order to ensure heterogeneity in these constructs at baseline. Attrition peaked at the final wave of data collection, with 16.8% of the original sample no longer providing data to the study team at that time. Of all qualified JIY who were approached regarding their interest in the study, 20% declined the opportunity to participate.

All data used in this study were collected through self-report interviews with participants. These interviews were held in locations that were convenient for participants (e.g., criminal justice facilities, libraries, participants’ homes, etc.). The research team provided each participant with a laptop computer to use during interview sessions and manually input responses to the survey questions. Informed consent was obtained from all adult participants and assent and parent/guardian consent was obtained from all minor participants. Ethical review for the present study was unnecessary because it entailed analysis of secondary data. The original Pathways to Desistance data collection received institutional review board approval from the University of Pittsburgh.

Measures

Gun violence perpetration

The main dependent variable examined in this study was gun violence perpetration measured at each wave. Two measures were combined to create the measure used in analyses. These measures asked participants whether they had engaged in two forms of gun violence during the prior observation period: shot someone with a gun or shot at someone with a gun and missed. A binary measure was computed from these combined measures that delineated participants who reported either shooting someone or shooting at someone during the prior observation period from those who did not (0 = No; 1 = Yes).

Gang membership

One of the key independent variables examined in this study as a predictor of gun violence perpetration is gang membership at each wave. This construct was measured at each wave using a binary variable which delineated participants who reported being part of a gang during the prior observation period from those who did not (0 = No; 1 = Yes). A one-observation period lag was assigned to this variable in order to establish temporal ordering between this as an independent variable and the gun violence-dependent variable of interest.¹

Victimization

The other key independent variable examined in this study was victimisation during each wave. This construct was measured using the Exposure to Violence Inventory (Selner O’Hagan et al., 1998). Distinct subtypes of victimisation were assessed. A binary measure was used that delineated participants who reported directly experiencing victimisation during a given observation period from those who did not (0 = No; 1 = Yes). A one-observation period lag was assigned to this independent variable. The following items were used to create the binary measure used in analyses: have you been attacked with a weapon, like a knife, box cutter, or bat?; have you been shot at?; have you been shot?; have you been chased where you thought you might be seriously hurt?; have you been beaten up, mugged, or seriously threatened by another person?; have you been raped, had someone attempt to rape you or been sexually attacked in some other way?.

Control variables

Several control variables were also included in analyses in order to mitigate the risk of bias in the estimation of effects of interest. The first of these variables was gender, as prior research has indicated that men present a greater risk of gun violence perpetration than women and that this may be related to masculinity norms (Levant, 2022). Gender was measured using a binary variable at baseline, which delineated male and female participants into two distinct categories (0 = Male; 1 = Female).

Race was also controlled for in analyses because prior research has indicated differential risk of gang membership based on racial identification (Tapia, 2011). Race was measured using a four-category nominal variable at baseline with the following response options: Black, Hispanic, White, and Other Race. Four dummy variables were then computed, which distinguished participants in a given racial category from all other participants (e.g., 1 = Black; 0 = All other participants). The dummy variable corresponding to White participants was then excluded from analyses in order to provide an omitted reference group that race coefficients could be interpreted in relation to.

Socioeconomic status (SES) was also controlled for in analyses because past research has indicated that low SES is associated with increased risk for gun violence (Sanchez et al., 2020). SES was measured at baseline using Hollingshead’s (1957) two-factor index of social position. This instrument probed the educational attainment and occupational prestige of participants’ parents’ and computed a compositive score comprising both of these components of SES. If both parents provided data, then a single mean score was computed so that all participants had one SES score at baseline for analyses.

Hostility was also controlled for at each wave, as prior research has indicated that hostility is a risk factor for perpetration of violent behaviour (Walters, 2020). Hostility was measured at each wave using the Brief Symptom Inventory (Derogatis & Melisaratos, 1983). This instrument was comprised of a series of ordinal items that asked participants to rate the degree to which they had been bothered by various hostility symptoms during the prior week (e.g., “Having urges to break or smash things”). A mean score was then calculated from the individual ordinal scores so that every participant had a single hostility score at each wave. A one-observation period lag was also applied to this time-varying covariate.

Deviant peer association was also controlled for in analyses because prior research has indicated that it is a potential risk factor for perpetration of gun violence (Beardslee et al., 2021). A scale that was adapted from the Rochester Youth Development study was used to assess this construct at each wave (Thornberry et al., 1994). This scale used a series of ordinal items to assess the general number of peers who participants reported influenced them to engage in seven different antisocial behaviours (e.g., “During the last six months how many of your friends have suggested that you should sell drugs?”). A mean score was then computed so that every participant had a single score pertaining to this construct at each wave. A one-observation period lag was applied to this time-varying covariate as well.

Sensation-seeking and impulse control were also controlled for in analyses given the aforementioned link between victimisation and these constructs in the literature review. Impulse control was measured at each wave using the Weinberger Adjustment Inventory and sensation-seeking was measured using the Youth Psychopathic Traits Inventory (Andershed et al., 2002; Weinberger & Schwartz, 1990). Both instruments used a series of ordinal items which asked participants to rate their degree of agreement with various statements about themselves pertaining to each construct (Impulse control example: I say the first thing that comes into my mind without thinking enough about it; Sensation-seeking example: I like to be where exciting things happen). A mean score was computed for the impulse control items and an index score was computed for the sensation-seeking items at each wave. A one-observation period lag was applied to both variables in analyses.

Prior research has also indicated the importance of controlling for age in analyses, as prior research has indicated that offending risk is age-graded (Moffitt, 1993). Age was measured at each wave in single-year intervals and each of these measures at each wave were used in analyses.

Observation period length was also controlled for in analyses because longer observation periods necessarily meant additional time for participants to engage in gun violence. While observation periods were generally of the same length for all participants at each wave, there was some between-participant variance in the exact length of observation periods. As such, a variable that measured the exact number of days in each observation period for each participant at all waves was included in analyses as a control variable.

It was also relevant to control for the amount of street time that participants had during each observation period, as being confined to secured facilities with no community access would necessarily restrict their capacity to obtain and use guns (e.g., prison, jail, psychiatric facilities). As such, a variable ranging from 0–1 which estimated the proportion of time that participants reported spending in secured facilities during each observation period was included in analyses, with higher scores corresponding to more time spent in secured facilities during each observation period (e.g., .67 = 67% of observation period spent in secured facilities with no community access).

Prior research has also indicated that constructs of interest are correlated with the amount of missing data reported for participants in this study (Kijowski & Wilson, 2022). As such, the number of waves which participants provided valid data to the Pathways to Desistance research team was controlled for in analyses. A variable was included in analyses which provided a count of the total number of waves that participants provided valid data to the study team in order to control for differential attrition across the study period.

General offending was also controlled for in analyses because the dependent variable was derived from the general offending items, so this was controlled for in order to rule out a more general effect of independent variables on gun violence. A binary variable which delineated participants who reported engaging in any offending during the prior observation period from those who did not was used here (0 = No; 1 = Yes). A one-observation period lag was applied to this variable.

The final control variable included in analyses was a lagged measure of the binary gun violence-dependent variable. By including a measure of this construct lagged by one observation period, this then facilitated a better understanding of how continuity of involvement in gun violence impacted later gun violence perpetration and how victimisation and gang membership resulted in a change in risk for gun violence at each wave.

Analytic strategy

The present study utilised a series of generalised estimating equations (GEE) in order to better understand the direct and interactive effects of gang membership and victimisation on gun violence perpetration risk. GEE is used within longitudinal data to simplify the estimation procedure while also accounting for correlation inherent to within-person repeated measures. This is done by including a correlation matrix that functions as a weight in analyses that applies a correlation structure to the repeated measures, with the researcher choosing the nature of this correlation structure. Stata/MP 16.1 was used to conduct all analyses and this software facilitates the use of three options for correlation structures. The independent structure assumes no correlation across the repeated measures, the exchangeable structure assumes equal correlation across the repeated measures, and the unstructured structure makes no assumptions regarding the nature of the correlation structure and allows it to vary freely. Preliminary analyses tested associations using all three correlation structure options to determine robustness of the results. The unstructured and independent correlation structures yielded consistent findings, whereas the model using unstructured correlation structure failed to estimate when the interaction term of interest was included. As such, results using the exchangeable correlation structure are reported here, as this option entailed a more realistic understanding of the correlated nature of errors across repeated measures. Logistic regression was used within the GEE framework to estimate relationships of interest because of the binary nature of the dependent variable of interest. Coefficients are interpreted as the expected difference in the log-odds of reporting engagement in gun violence during a given observation period based on a one-unit increase in the independent variable of interest, net of all covariates in the model. Multiple imputation was used to manage missing data in analyses, as variables with missing data were regressed on gender, race, and number of waves with valid data variables. Two models were estimated. Model 1 estimated the direct effects of victimisation and gang membership on gun violence perpetration risk net of all control variables. Model 2 included an interaction term modelling the hypothesised moderating effect of gang membership on the relationship between victimisation and gun violence perpetration.

Results

Table 1 provides descriptive statistics for all variables that were included in analyses. In the pooled sample, 9.63% of the participants reported gang membership at any given time on average. On average, 18.41% of the participants reported experiencing victimisation during any given observation period. Across all waves, 4.14% of the participants reported engaging in gang violence on average during a given observation period. The sample was mainly comprised of male JIY (Male = 86.41%; Female = 13.59%). The plurality of participants in the sample were Black (41.43%), followed by Hispanic participants (33.53%), then White participants (20.24%), and finally participants identified as Other Race (4.80%). Chi-squared tests indicated significant gender differences in gang membership (Male = 10.68%; Female = 3.29%; χ² = 103.062; p < .001) and gun violence perpetration (Male = 4.64%; Female = 1.15%; χ² = 50.189; p < .001) in the pooled longitudinal data. Table 2 provides Model 1 and Model 2 GEE regression results.

Table 1. Descriptive statistics.

	Coefficient	Standard deviation	Minimum	Maximum
Gun Violence (Pooled) No = 95.86% Yes = 4.14%
Gang Membership (Pooled) No = 90.37% Yes = 9.63%
Victimization (Pooled) No = 81.59% Yes = 18.41%	1.347	.753	0	3
Gender Male = 86.41% (N = 1170) Female = 13.59% (N = 184)
Race Black = 41.43% (N = 561) Hispanic = 33.53% (N = 454) White = 20.24% (N = 274) Other Race = 4.80% (N = 65)
Socioeconomic Status at Baseline	51.409	12.299	11	77
Deviant Peer Association (Pooled)	.784	.822	0	4
Hostility (Pooled)	.561	.684	0	4
Impulse Control (Pooled)	3.197	.969	1	5
Sensation-Seeking (Pooled)	12.734	3.421	4	20
Number of Observation Periods with Valid Data	9.001	1.967	0	10
Average Time Spent in Secured Facilities at Each Wave (Pooled)	.329	.416	0	1
Number of Days in Observation Period (Pooled)	248.399	93.832	61	428
Age (Pooled)	18.898	2.453	14	26
Offending (Pooled) No = 50.62% Yes = 49.38%

Table 2. Generalised estimating equation logistic regression modeling of covariate effects on log-odds of gun violence (pooled N = 13,540).

	Model 1			Model 2
	Coefficient	p-Value	95% Confidence interval	Coefficient	p-Value	95% Confidence interval
Gang Membership (0 = No; 1 = Yes) (Lagged)	.546	.001	.238 .854	.746	<.001	.366 1.127
Victimization (0 = No; 1 = Yes) (Lagged)	.185	.176	−.083 .453	.277	.071	−.024 .578
Victimization X Gang Membership (Lagged)	–	–	–	.425	.115	−.954 .103
Gender (0 = Male; 1 = Female)	−1.575	<.001	−2.314–.836	−1.561	<.001	−.298–.825
Race (reference = White) Black Hispanic Other Race	.555 .412 .549	.005 .058 .089	.167 .942 −.013 .838 −.084 1.183	.556 .410 .547	.005 .059 .090	.170 .942 −.015 .835 −.085 1.179
Socioeconomic Status at Baseline	−.002	.681	−.013 .008	−.002	.650	−.013 .008
Deviant Peer Association (Lagged)	.273	<.001	.133 .414	.276	<.001	.136 .416
Sensation-Seeking (Lagged)	.001	.959	−.039 .041	.001	.976	−.040 .041
Impulse Control (Lagged)	−.108	.171	−.263 .047	−.110	.165	−.265 .045
Hostility (Lagged)	.022	.801	−.153 .197	.024	.790	−.151 .198
Age	−.102	.009	−.178–.026	−.104	.007	−.180–.028
Days in Wave Observation Period	.002	.012	.001 .004	.002	.011	.001 .004
Proportion of Observation Period Spent in Secure Facilities	−.589	<.001	−.887–.291	−.589	<.001	−.885–.292
Total Number of Waves with Valid Data	−.190	<.001	−.252–.128	−.188	<.001	−.250–.126
General Offending (0 = No; 1 = Yes) (Lagged)	.301	.042	.012 .590	.292	.048	.003 .581
Gun Violence (0 = No; 1 = Yes) (Lagged)	.581	.001	.252 .911	.612	<.001	.282 .942
Constant	.859	.354	−.961 2.679	.845	.361	−.972 2.663

Model 1 results indicated that belonging to a gang was associated with an increased risk of gun violence perpetration at follow-up, though victimisation was not a significant predictor of gun violence in this model (Gang membership coefficient = .546, p < .001). Additionally, being male, greater deviant peer association, being younger, longer observation periods, spending less time in secured facilities, providing fewer waves of valid data, having offended at all during the previous observation period, and having perpetrated gun violence during the previous observation period were all associated with increased risk of perpetrating gun violence. Black participants reported greater risk for perpetrating gun violence compared to White participants in this model also. Model 2 results indicated that gang membership continued to exert a significant direct effect consistent with those observed in Model 1 and that the effect of victimisation remained nonsignificant (Gang membership coefficient = .746, p < .001). The interaction between these two constructs was nonsignificant. Being male, having offended at all during the prior observation period, greater deviant peer association, being younger, longer observation periods, providing fewer waves of valid data, spending less time in secured facilities, and perpetrating gun violence during the previous observation period were also all associated with increased risk of gun violence perpetration in this model. Black participants also reported a greater risk for gun violence perpetration in this model compared to White participants.

There were concerns regarding the use of binary independent variables and a binary dependent variable in these analyses that may have led to the lack of variation resulting in the nonsignificant interaction effect. Analyses were re-estimated using a continuous victimisation variety variable as a means of checking the robustness of these findings. The interaction effect of interest remained non-significant in these sensitivity analyses, indicating the robustness of this effect. Additional sensitivity analyses were estimated with mean-centred gang membership and victimisation variables for the interaction effect. This interaction effect was similarly nonsignificant.

Discussion

This study provided insight into the relevance of gang membership and victimisation for predicting risk of gun violence perpetration among a sample of JIY. Findings indicated that gang membership was only associated with increased log-odds of carrying out gun violence during the subsequent observation period, but not victimisation. Further, the hypothesised interaction effect was not significant. There are a number of implications of these findings for reducing the prevalence of gun violence that may be acted upon by criminal justice professionals and clinicians specialising in trauma exposure in the United States

As noted above, gang membership significantly predicted increased risk for perpetration of gun violence in the subsequent observation period, findings that were consistent with prior research on the topic (Braga, 2003; Pardini et al., 2021). These results provide further indication that treatment programming to try to reduce gun violence should prioritise gang members if they are to be the most effective, though as will be noted below, victimisation may still be relevant in this regard for non-gang members. Further, prevention programming may also aid in reducing this risk as well, particularly as it pertains to gang membership. Gang Resistance Education and Training (GREAT) programming is one evidence-based modality that has been demonstrated for being effective at preventing gang membership by empirical research (Esbensen et al., 2012; Papachristos, 2013). The programme functions to educate youth about gangs and provide communication and decision-making skills to facilitate successful avoidance of having youth join gangs during adolescence. This may pair well with gun violence specifically because of the additional focus of the programme on violence prevention also.

Victimisation was not found to exert a significant direct effect on gun violence in this study, running counter to prior research on the topic (McGee et al., 2017; Rowan et al., 2019; Vossekuil, 2002). This is particularly surprising since victimisation has been found to increase gun violence risk in prior research among JIY specifically (Rowan et al., 2019). However, the highlighted study examined victimisation experiences that didn’t specifically involve a gun as a predictor of one’s own perpetration of gun violence; whereas the measure used in the present study was inclusive of this form of gun-related victimisation. It may be that this helps to explain the unexpected results observed here. Individuals who have experienced gun-related victimisation may have particular sensitivity to carrying/using firearms themselves. Given the nature of triggering stimuli being a particularly salient symptom of post-traumatic stress disorder that can lead to further distress when encountered (American Psychiatric Association, 2013), it may be that carrying/using guns serves as a trigger in this regard. If this is the case, then including those individuals who have experienced gun-related victimisation in the measure employed in this study may have deflated the direct effect of victimisation, since these individuals would be less likely to use a gun. This, however, remains speculative. Future research should seek to explore this possibility to a greater extent to determine if gun-related victimisation and non-gun related victimisation have distinct effects on gun violence perpetration.

The hypothesised interaction between victimisation and gang membership for predicting gun violence perpetration was found to be nonsignificant. Despite the potential mechanisms that would explain amplification of dulling of direct effects via this moderation, these findings did not support the corresponding hypothesis of the study. Based on this and other findings of the study, JIY who are gang members should be prioritised for treatment to prevent gun violence regardless of their victimisation history. It should be noted that the present study was unable to disentangle temporal ordering between onset of gang membership and victimisation in the sense that victimisation may have happened before or after these JIY became gang members. Considering that this temporal ordering may have had implications for feelings of safety that were predicted to have implications for gun carrying behaviours, this may aid in explaining the null findings here. This indicates the need for additional research on the topic using data that can disentangle the temporal ordering issue identified here in order to determine the robustness of these findings.

A final consideration should be made for the nonsignificant interaction effect observed here. Upon a stepwise approach to inclusion of control variables, it appears that inclusion of the cognitive variables (sensation-seeking, impulse control) resulted in attenuating the interaction effect to non-significance. Without these control variables in the model, this was a significant and negative interaction. This may be indicative of the disruptions to cognition that may occur following victimisation (Arnsten et al., 2015; Kerig, 2019; Wojciechowski, 2022). It may be that controlling for these factors helped to explain the relationship between victimisation and gun violence and thus mediated this interaction effect to the point of non-significance. If this is the case, then it would make sense to potentially study these relationships further, as moderated mediation that may have been observed here may also have treatment implications. For example, it may be that these cognitive constructs only matter for explaining the relationship between victimisation and gun violence risk for gang member or non-gang member JIY. That said, a full examination of these processes was beyond the scope of this study and future research should focus on examining these relationships further.

There are also several noteworthy limitations of this study. The first of these limitations pertains to the potential limited generalisability of findings. The Pathways to Desistance sample was comprised entirely of JIY who were recently adjudicated for a serious offence just prior to baseline measurements. The high-risk nature of this sample then calls into question whether these results are relevant for general population youth. Further compounding this issue is the fact that this sample was collected purposively. The nonprobability sampling method used here then may have further hampered capacity for generalisability beyond this sample. These issues indicate the need for continued study on these topics in order to better understand whether or not gang membership and victimisation are relevant for youth in the general population. That said, JIY present a salient population for studying these processes, as they present elevated risk for experiencing victimisation, gang membership, and involvement in antisocial behaviour (Baglivio et al., 2014; Hill et al., 1999; Robertson et al., 2004). As such, these results still provide a great deal of utility for potentially targeting resources to prevent gun violence among this population with potentially strong impact given the increased propensity for involvement in antisocial behaviour among this population of youth. A related limitation here is that the Pathways to Desistance data were collected from 2000 to 2010. These data then are becoming somewhat dated and gangs may function differently today than they did in the past. As such, it is unclear whether these findings are relevant for these processes in the present day. Future research should seek to explore these processes again using data that were collected more recently in order to determine the robustness of these findings. Another limitation of this study pertains to the inability to disentangle different forms of victimisation. Examinations of more distinct forms of victimisation (like sexual victimisation vs. physical abuse) could not be examined in the publicly available data. It may be that there is further moderation here in terms of the specific type of victimisation that an individual is exposed to. Unfortunately, the Pathways to Desistance measure of direct victimisation does not allow for decomposition of these individual types of victimisation along these lines. This indicates the need for continued study of these processes of interest in order to determine whether the type of victimisation (e.g., physical vs. sexual) and/or the context in which violence exposure occurs (e.g., community, school, family) matter for understanding the effects of gang membership on gun violence perpetration risk. Another limitation here relates to the inability of these analyses to parse temporal ordering between gang membership and victimisation. This is to say that victimisation that occurred prior to joining a gang may have a different effect on gun violence perpetration than victimisation that occurred after joining the gang. Unfortunately, a more nuanced investigation of these relationships was beyond the scope of this study, as this would likely entail mediated moderation analyses which would necessitate analysis of distinct waves of data, rather than the data as a whole. This would have significant drawbacks of reduced power for analyses, with only marginal capacity to actually parse temporal ordering. Related to this is the fact that there was no way to identify whether the victimisation occurred within the gang itself. This could help to explain some of the unexpected findings of the study if the victimisation occurred within the gang and this impacted temporal ordering of these risk factors for some participants, but not others. This indicates the need for additional research on this topic in order to test the robustness of these findings while addressing this temporal ordering and source of victimisation. A final limitation pertains to the binary coding of the independent and dependent variables. This may have led to issues with a lack of variation that led to the null interaction result. However, sensitivity analyses re-examined the hypothesised interaction using a continuous measure of the and the victimisation independent variable and the findings were robust. As such, this should increase the confidence in the final findings of no significant interaction observed here.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Thomas Wojciechowski

Thomas Wojciechowski is an assistant professor with the Michigan State University School of Criminal Justice. His research mainly focuses on mental health, substance use, violence, and human development.

Notes

1. Two sets of analyses were estimated here. The main analyses, as mentioned above, utilised a lagged measure of gang membership as the independent variable, whereas the sensitivity analyses utilised a concurrent measure of gang membership for predicting gun violence perpetrated during the same observation period. The results from both sets of analyses were analogous to one another, indicating the robustness of the findings reported as the main set of analyses.