Risk and Protective Factors for Gun Violence in Male Juvenile Offenders

Abstract

Objective:

To examine several risk and protective factors as predictors of future gun violence among male juvenile offenders.

Method:

Data came from a longitudinal cohort of 1,170 male juvenile offenders (42.1% Black; 34.0% Latino; 19.2% White) ages 14-19 who were adjudicated for a serious offense. Interviews were conducted with participants every six months for three years and then annually for four years. The outcome was self-reported gun violence assessed at each follow-up. The time-lagged predictors included several self-reported risk factors (i.e., gun carrying, non-gun violence, drug dealing, heavy drinking, poor impulse control, rewards for crime, peer gun and non-gun delinquency, gang membership) and protective factors (i.e., concern for others, expectations and aspirations for work/family, religious beliefs, adult social supports). The data were analyzed using generalized estimating equation models.

Results:

There were 266 participants who reported engaging in gun violence at one or more assessments. Gun carrying was a significant predictor of future gun violence; however, nearly half (49%) of the juveniles who reported gun carrying across the repeated assessments did not report engaging in gun violence. Besides gun carrying, several risk (i.e., drug dealing, heavy drinking, rewards for crime, gang membership, peer gun carrying) and protective (i.e., concern for others, aspirations for work/family, religious beliefs, adult social supports) factors significantly predicted gun violence, after controlling for their co-occurrence (Risk factor odds ratios=1.18-1.50; Protective factor odds ratios=.44-.87; ps<.05).

Conclusions:

Interventions designed to prevent gun violence among juvenile offenders should reduce targeted risk factors, while strengthening protective factors that may offset these risks.

Firearm violence is a longstanding and costly public health problem in the United States that disproportionately affects minority youth living in impoverished urban environments (Oliphant et al., 2019). In recent years, political leaders and international health agencies have called for research studies designed to understand the mechanisms that lead young males to illegally carry guns and engage in gun violence (Oliphant et al., 2019). As a result of these efforts, there has been an acceleration of studies examining the risk factors for gun carrying as a surrogate marker for gun violence. The findings from these studies have not always been as expected. In particular, an implicit assumption underlying this research is most adolescents who carry guns will engage in gun violence (Stinson, Ruan, Pickering, & Grant, 2006), but emerging evidence suggests that a large portion of adolescents will intermittently carry guns without engaging in acts of serious violence (Lizotte, Krohn, Howell, Tobin, & Howard, 2000; Vaughn, Salas-Wright, Boutwell, DeLisi, & Curtis, 2016). As a result, it remains unclear whether the identified risk factors for gun carrying also prospectively predict engagement in gun violence.

Examining the prospective predictors of gun violence longitudinally presents numerous challenges. Most notably, gun violence is rare in community samples (Schmidt et al., 2019). One strategy for overcoming this problem is to conduct research with subgroups of youth who are at high risk for engaging in gun violence, particularly male juvenile offenders (Rowan, Schubert, Loughran, Mulvey, & Pardini, 2019). Longitudinal research of this type also provides an opportunity to test the effects of adverse individual and socio-contextual factors, as well as prosocial protective factors which may help reduce the likelihood that juvenile offenders will engage in acts of gun violence. This empirical information can inform targeted intervention programs designed to prevent gun violence among males involved in the juvenile justice system. Consistent with this approach, this study examined a comprehensive array of risk and protective factors as predictors of future gun violence among a large longitudinal sample of male juvenile offenders who were repeatedly assessed from adolescence into young adulthood.

Gun Carrying as a Prelude to Gun Violence

The prevalence of gun carrying and serious violence among males in the U.S. begins to rise around age 13 and peaks in the late teens to early 20’s (Centers for Disease Control, 2012; Lizotte et al., 2000; Loeber, Farrington, Stouthamer-Loeber, & White, 2008). A recent review of studies found that the lifetime prevalence of gun carrying among adolescents living in urban communities typically ranges from 15%-20% (Oliphant et al., 2019), whereas lifetime gun carrying among male juvenile offenders has been estimated to be as high as 73% (Teplin, 2019). Although a large portion of delinquent youth report carrying a firearm, many of these youth exhibit transient patterns of carrying that do not involve engaging in acts of violence (Oliphant et al., 2019). Despite this fact, lifetime engagement in gun violence among juvenile offenders remains particularly high. One longitudinal investigation of adolescent offenders found that 25% reported a prior history of shooting at someone at a baseline assessment, and 16.3% of the sample reported engaging in subsequent acts of gun violence across a 7-year follow-up period (Gonzales & McNiel, 2020). As a result, there is a need to better understand what additional factors besides gun carrying help delineate delinquent youth at risk for engaging in future gun violence.

Risk Factors for Gun Carrying and Gun Violence

Several longitudinal studies examining risk factors for gun carrying as a prelude to gun violence have focused on factors associated with a pre-existing antisocial propensity (Beardslee, Docherty, Yang, & Pardini, 2019; Lizotte et al., 2000; Spano & Bolland, 2013). Antisocial propensity explanations focus largely on dispositional characteristics and behaviors that have been linked to the initiation and persistence of severe violence among adolescents. Consistent with this model, longitudinal studies have found that adolescents who exhibit impulse control problems (Lu & Temple, 2019), conduct problems (Beardslee, Docherty, Yang, et al., 2019), and positive attitudes toward criminal offending (Goldstick et al., 2019) are more likely to engage in future gun carrying. Drug dealing is also an important risk factor for later gun carrying, as the use of deadly weapons is often necessary to establish, defend, and control illegal drug markets (Docherty, Mulvey, Beardslee, Sweeten, & Pardini, 2019; Lizotte et al., 2000). Relatedly, engagement in frequent and heavy substance use, particularly binge drinking, has been found to increase the likelihood that youth will engage in aggression and risky behaviors like gun carrying (Chen & Wu, 2016).

In addition to an antisocial propensity, longitudinal studies have also examined deviant social influences as risk factors that may motivate adolescents to carry guns (Beardslee, Docherty, Yang, et al., 2019; Oliphant et al., 2019; Spano & Bolland, 2013). Deviant social influence models assert that gun carrying is learned and reinforced through social interactions with antisocial peer networks that provide youth with easy access to firearms (Beardslee, Docherty, Yang, et al., 2019; Roberto, Braga, & Papachristos, 2018). In support of this notion, longitudinal studies have found that gang membership and affiliation with delinquent and gun carrying peers are risk factors for future gun carrying among adolescents and young adults (Beardslee, Docherty, Yang, et al., 2019; Lizotte et al., 2000; Spano & Bolland, 2013). These social networks tend to emphasize the importance of carrying a firearm as a way to exert dominance, increase social status and resolve disputes, as well as provide youth with easy access to illegal firearms (Roberto et al., 2018).

A third explanatory model of gun carrying that has been the focus of longitudinal studies involves the use of guns for self-protection (Oliphant et al., 2019). According to this model, adolescents who live in impoverished and segregated high crime neighborhoods where they are continually exposed to violence and other criminal activity are prone to carry guns as a means of self-defense (Beardslee et al., 2018). Self-protection explanations have been supported by longitudinal studies indicating that adolescents who witness severe violence, as well as those who become the victims of violence, are at increased risk for future gun carrying (Beardslee et al., 2018; Keil, Beardslee, Schubert, Mulvey, & Pardini, 2019; Spano & Bolland, 2013). In addition, adolescents often report engaging in gun carrying as a way to ward off potential attackers and defensively retaliate when threatened with violence (Wilkinson & Fagan, 2001).

Although longitudinal research has provided evidence that risk factors associated with antisocial propensity, social influence, and self-protection models predict future gun carrying, the utility of these models for predicting gun violence is far from clear. It is possible that these risk factors do significantly predict engagement in later gun violence after statistically controlling for co-occurring gun carrying. Conversely, individual and socio-environmental factors may incrementally increase delinquent youths’ risk for engaging in future acts of gun violence, even after controlling for co-occurring gun carrying. To resolve these issues, longitudinal studies must simultaneously examine gun carrying and other risk factors as predictors of future gun violence during adolescence and young adulthood.

Delineating Risk Factors for Gun Violence

Most existing studies examining risk factors associated with youth engagement in gun violence have focused on cross-sectional associations, making it impossible to establish the fundamental temporal ordering criteria required for causal inference (Schmidt et al., 2019). The cross-sectional studies in this area have generally found concurrent associations between several risk factors outlined in antisocial propensity, social influence, and self-protection explanatory models and engagement in gun violence (Rowan et al., 2019; Schmidt et al., 2019; Tracy, Braga, & Papachristos, 2016). However, only three published longitudinal studies have explicitly examined risk factors that prospectively predict future gun violence among adolescents and young adults (Beardslee, Docherty, Mulvey, & Pardini, 2019; Goldstick et al., 2019; Gonzales & McNiel, 2020). These studies have found evidence linking engagement in non-gun violence, antisocial attitudes, frequent substance use/misuse, and increased affiliation with antisocial peers with future gun violence (Beardslee, Docherty, Mulvey, et al., 2019; Goldstick et al., 2019; Gonzales & McNiel, 2020). However, each of these studies examined a limited set of risk factors and all of them failed to include gun carrying, drug dealing, gang membership, and peer gun carrying as predictors of later gun violence. This omission is a significant limitation as these factors are theorized to be key drivers of youth gun violence.

Need for Research on Protective Factors for Gun Violence

To date, research has focused primarily on identifying negative individual and environmental factors (i.e., risk factors) that increase the likelihood that youth will carry guns and engage in gun violence. Various conceptualizations of protective factors have been used in studies examining predictors of delinquent behavior in community samples (Farrington, Ttofi, & Piquero, 2016; Ttofi, Farrington, Piquero, & DeLisi, 2016). However, the term protective factor in the criminal justice literature commonly refers to positive attributes or strengths that reduce the likelihood that an offender will engage in future criminal acts (Ullrich & Coid, 2011; Viljoen, Bhanwer, Shaffer, & Douglas, 2020). According to this conceptualization, protective factors have a direct cumulative effect on delinquent behavior that can counterbalance the influence of risk factors among a population prone to engage in offending (Ullrich & Coid, 2011). In juvenile offender populations, a protective factor may help to prevent youth from initiating more serious forms of criminal offending (e.g., gun violence), or promote desistance from ongoing patterns of serious offending (e.g., continued gun violence). The identification of protective factors is a critical aspect of strength-based intervention approaches for youthful offenders, which seek to leverage and reinforce positive attributes and resources in adolescents’ lives in order to promote healthy development (Fortune, 2018). This stands in contrast to traditional risk-based intervention approaches which focus primarily on ameliorating individual deficits and reducing exposure to hazardous socio-contextual factors (Fortune, 2018).

As noted in a recent literature review, no published longitudinal studies have examined prosocial protective factors that reduce the probability youth will engage in later gun violence (Schmidt et al., 2019). However, theoretical models delineating prosocial protective factors for youth violence have traditionally emphasized the importance of strong bonds with prosocial adults and institutions, active involvement in prosocial activities, and the adoption of beliefs, expectations and values that emphasize conventional life goals (Fortune, 2018; Pollard, Hawkins, & Arthur, 1999). Consistent with these models, longitudinal studies have found that incarcerated offenders who have close, supportive, and prosocial adult relationships are less likely to recidivate upon release (Intravia, Pelletier, Wolff, & Baglivio, 2017; Ullrich & Coid, 2011). Evidence also suggests that adjudicated youth who engage in helpful behaviors and express concern about the feelings and welfare of others are at decreased risk for engaging in future criminal offending (Keil et al., 2019; Wall Myers et al., 2018). Adolescents’ expectations and aspirations for the future have also been identified as protective factors for future delinquency. Specifically, longitudinal studies have found that youth who are committed to achieving conventional adult goals (i.e., aspirations), and express confidence in their ability to reach these goals (i.e., expectations) are less likely to engage in future criminal acts (Iselin, Mulvey, Loughran, Chung, & Schubert, 2012; Knight, Ellis, Roark, Henry, & Huizinga, 2017). Religious beliefs have also been proposed as a protective factor for criminal offending, particularly feeling connected to a deity that exemplifies moral values antithetical to an antisocial lifestyle (Kelly, Polanin, Jang, & Johnson, 2015). However, longitudinal studies examining religiousness as a factor that promotes desistance from offending have produced inconsistent findings (Giordano, Longmore, Schroeder, & Seffrin, 2008; Ullrich & Coid, 2011).

Limitations of Prior Research

To date, no published longitudinal studies have examined whether risk factors associated with antisocial propensity, social influence and self-protection models uniquely predict future gun violence, after controlling for co-occurring gun carrying (Schmidt et al., 2019). Relatedly, the few existing longitudinal studies in the area have failed to include key risk factors implicated in gun violence, including gang membership, drug dealing, and peer gun carrying. Moreover, no existing longitudinal studies have examined whether prosocial protective factors decrease the likelihood that adolescent offenders will engage in later gun violence (Schmidt et al., 2019). Because some studies have found that protective factors may exhibit modest to no incremental utility in predicting violent recidivism after accounting for known risk factors (Shepherd, Luebbers, & Ogloff, 2016; Ullrich & Coid, 2011; Vincent, Chapman, & Cook, 2010), there is a need to examine potential risk and protective factors for future gun violence within the context of a single study. Empirically delineating these risk and protective factors using longitudinal data is critical for developing evidence-based policies and programs designed to reduce gun violence among male juvenile offenders.

Study Aims

To address these limitations, this study examined a comprehensive set of risk and protective factors as prospective predictors of future gun violence among a large longitudinal sample of male juvenile offenders who were relatedly assessed from adolescence into the mid 20’s. Although adolescent gun carrying is hypothesized to a robust predictors of later gun violence, it is anticipated that several other risk factors will incrementally add to the prediction of future gun violence. These risk factors are posited to span antisocial propensity (e.g., drug dealing, substance misuse, antisocial attitudes), social influence (e.g., gang membership, peer gun carrying), and self-protection (e.g., exposure to violence) models of gun violence. It is also hypothesized that protective factors associated with prosocial attributes (e.g., concern for others) and socio-contextual characteristics (e.g., strong adult social support) will decrease the likelihood that juvenile offenders will engage in future gun violence, thereby counterbalancing the negative influence risk factors have on later gun violence.

Methods

The sample consisted of 1,170 male juvenile offenders (42.1% Black; 34.0% Latino; 19.2% White; 4.6% other) from the Pathways to Desistance Study. Participants were eligible for the study if they had been adjudicated for a serious offense (94% were felonies) in Phoenix, Arizona or Philadelphia, Pennsylvania. The most serious adjudicated charge typically involved acts of violence (41%), theft (27%), drug violations (13%), or weapons violations (10%). At the time of study enrollment, participants were between 14 to 19 years old (M = 16.6, SD = 1.1). Most were living with their biological mother (74%), and approximately one-quarter (23%) had a biological father living in the home. The highest level of education attained by any parent in the home was typically grade school (17%), some high school (39%), or a high school diploma (32%). Nearly half were either living in a home without an employed parent (22%) or a home where the highest level of parental employment involved unskilled/semiskilled manual labor (26%). Most families were living in socio-economically disadvantaged urban neighborhoods (Chung & Steinberg, 2006).

After a baseline interview, study participants were interviewed every 6 months for 3 years and then annually for 4 years from 2000-2010. Youth were approximately 23.0 years old (SD = 1.2) at the final interview. Researchers conducted interviews with the youth in private settings. Prior to conducting the interviews, parental consent and youth assent were obtained for participants under the age of 18, and youth consent was obtained once youth reached age 18. All procedures were approved by the institutional review boards at Arizona State University, Temple University, and the University of Pittsburgh.

The current study used youth-reported data collected across the 10 follow-up assessments after the baseline interview. The baseline assessment was not used because the recall time frame and/or wording of several measures differed from the follow-up assessments. Further details regarding the sample, study methodology, and measures can be found at https://www.pathwaysstudy.pitt.edu. Data from the study can be obtained from the Inter-University Consortium on Political and Social Research (ICPSR) at the University of Michigan (https://www.icpsr.umich.edu/icpsrweb/NAHDAP/studies/29961).

Measures

Gun Violence and Gun Carrying

Gun Violence.

Gun violence was assessed using three items from the Self-Report of Offending scale (SRO) (Huizinga, Esbensen, & Weiher, 1991). Two questions asked participants whether they had shot someone or shot at someone and missed during each recall period. A third question asked whether they had robbed someone with a weapon during the recall period, and whether a gun was used the last time it occurred. For each recall period, a binary variable was created indicating whether participants had engaged in any gun violence (0=No; 1=Yes).

Gun Carrying Without Gun Violence.

Gun carrying was assessed with a single item from the SRO. At each interview, youth reported whether they carried a gun since the previous interview and the number of times they engaged in this behavior. Due to the low base rate of gun carrying, a binary predictor was created indicating whether the participant had engaged in any gun carrying, but not gun violence, during each recall period (0=No; 1=Yes). This was done to delineate gun carrying in the absence of gun violence as an independent predictor. Although not used in the primary analysis, information on the frequency of carrying at each recall period is reported as part of the study descriptive statistics.

Risk Factor Variables

Non-Gun Serious Violence.

Engagement in serious violence not involving a gun was assessed using five items from the SRO (i.e., gang fighting, robbery without a weapon, carjacking, beating someone badly, sexual assault). Due to the low base rate of serious violence, a binary variable representing engagement in any non-gun serious violence was created for each recall period (0=No; 1=Yes).

Drug Dealing.

Two items from the SRO asked participants whether they had sold marijuana or sold other illegal drugs during the recall period. Due to the low base rate of drug dealing, these items were combined into a binary variable indicating whether participants had engaged in any drug dealing (1=Yes; 0=No).

Regular Heavy Drinking.

Two items from the Substance Use/Abuse Inventory (SUAI) (Chassin, Rogosch, & Barrera, 1991) were used to assess regular heavy alcohol use during each recall period. The questions asked participants how often they had consumed five or more drinks at one time and how often they had been drunk, with each item being rated on a 9-point scale (1=“Not at all” to 9=“Everyday”). Regular heavy drinking was defined as engaging in either of the two behaviors at least “2-3 times per month” on average during the recall period (0=No; 1=Yes). Increasing the frequency threshold and/or using only one of the two questions to delineate heavy drinking did not impact the primary study findings.

Regular “hard drug” use (i.e., cocaine, stimulants, opiates) was screened as a potential predictor using three questions regarding the frequency of from SUAI. The use of one or more of these substances was low at each timepoint (8.4%-12.3%), with relatively few participants reporting regular use (i.e., at least “2-3 times per month”) at each assessment (4.4%-6.6%). Due to this low base rate, regular hard drug use was not selected as a predictor in the primary analysis. Marijuana use was not included in the analyses because it has not been shown to consistently predict later violent behavior (McGinty, Choksy, & Wintemute, 2016). Supplemental models also found hard drug use and marijuana use did not predict later gun violence after controlling for other risk factors (see Supplemental Table S1).

Gang Membership.

A single question from a Gang Involvement Questionnaire (Thornberry, Lizotte, Krohn, Farnworth, & Jang, 1994) was used to assess any gang membership at each recall period (0=No; 1=Yes).

Exposure to Violence.

The Exposure to Violence Inventory was used to assess participants’ direct and indirect exposure to serious violence (Selner-O’Hagan, Kindlon, Buka, Raudenbush, & Earls, 1998). Six items assessed direct violence victimization (e.g., been shot, shot at, attacked with a weapon, beaten, raped). Six separate items asked participants whether they had seen these things happen, with one additional item asking whether they had seen someone murdered. A sum of the 13 items was calculated at each recall period. The internal consistency of the scale was adequate across the timepoints (αs=.67-.73).

Peer Gun Carrying and Peer Non-Gun Delinquency.

The self-reported Peer Delinquency Scale was used to index participants’ affiliation with criminal offending peers (Thornberry et al., 1994). For each item, participants’ rated how many of their friends engaged in various criminal acts on a five-point rating scale (1=“None of them” to 5=“All of them”). Peer gun carrying was assessed using a single item asking about gun carrying by friends. Peer non-gun delinquency scale was assessed by averaging 7 items about engagement in other illegal activities (e.g., non-gun violence, theft, drug dealing). The internal consistency of the peer non-gun delinquency scale was high across all timepoints (αs=.88-.90).

Poor Impulse Control.

The impulse control subscale from the self-reported Weinberger Adjustment Inventory (Weinberger & Schwartz, 1990) assessed participants’ impulsive decision-making and thrill-seeking behaviors (e.g., “I do things without giving them much thought”). Participants rated how well each item described them using a 5-point scale (1=“False” to 5 =“True”) and a mean of all items was used as an index of poor impulse control. The internal consistency of this scale was good across all timepoints (αs=.79-.84).

Personal Rewards of Crime.

Expectations that engagement in criminal acts would result in pleasurable feelings was measured using seven items from the Social and Personal Costs and Rewards of Crime scale (Nagin & Paternoster, 1994). Each item was rated on an 11-point scale (0=“No fun or kick at all” to 10=“A great deal of fun or kick”). Items were averaged to create a personal rewards of crime score, which exhibited high internal consistency across the timepoints (αs=.89-.91).

Neighborhood Disorder.

An adapted version of the Neighborhood Conditions Measure was used to assess the environment surrounding the home at each recall period (Sampson & Raudenbush, 1999). Twelve items indexed physical disorder (e.g., abandon cars, litter, graffiti) and nine items assessed social disorder (e.g., gangs, adults fighting, people using drugs). The items were rated on 4-point scale (1=“Never” to 4=“Often”). Because the disorder dimensions were highly correlated at each assessment (r=.85-.89), the mean of all items was used to index neighborhood disorder. The internal consistency of the scale was high at each time point (αs>.90).

Protective Factor Variables

Adult Social Supports.

The Contact with Caring Adults inventory assessed the extent to which participants had a close group of supportive adults in their life (Nakkula, Way, Stauber, & London, 1990; Phillips & Springer, 1992). Participants were asked to name an adult they would most likely to turn to for support across eight different domains (e.g., adult who cares about your feelings, you can ask for advice/help). An adult was considered a significant social support if they were named across three or more domains. Using this criterion, the possible range of significant adult social supports varied from 0 to 2 at each assessment.

Expectations and Aspirations for Work/Family.

The adapted Perceptions of Chances of Success scale assessed aspirations and expectations for prosocial future goals (Menard & Elliott, 1996). Four items asked participants to rate the chances they would obtain a good job, earn a good living, create a positive family environment, and establish positive parent-child relationships (1=“Poor” to 5=”Excellent”). The aspirations for work/family scale consisted of the same four items, except participants rated the importance of achieving these goals (1=“Not at all important” to 5=“Very important”). Items were averaged to create separate expectations and aspirations for work/family constructs. The internal consistencies of both scales were good across all assessments (αs = .81-.88).

Religious Beliefs.

The Importance of Spirituality scale was used to assess participants’ personal religious beliefs (Maton, 1989), including whether they had a close relationship with God, felt cared for by God, and turned to their faith to cope with stressors. The three items were rated on a 5-point scale (1=“Not at all true” to 5=“Completely true”), and the mean of the items was used to index religious beliefs. The scale exhibited high internal consistency across assessments (αs = .91-.94).

Concern for Others.

The consideration of others subscale from the WAI was used to assess participants’ levels of concern and caring for others (Weinberger & Schwartz, 1990). The scale consists of 7 items that ask participants whether they are concerned about helping others and try to avoid hurting their feelings (e.g., “I try hard not to hurt others’ feelings”). Participants rated each item on a 5-point scale (1=“False” to 5=“True”), and an average of the items was used to index concern for others. The scale exhibited adequate internal consistency across assessments (αs = .72-.74).

Model Covariates

Demographic Characteristics.

Participants’ age in years at each assessment and self-reported race/ethnicity were included as covariates in all models.

Secure Confinement and Street Months.

The time participants spent in the community during each recall period varied due to placement in secure facilities (e.g., detention, jail, prison) and differences in lengths of time between interviews (i.e., 6 months vs. 12 months). This led to differences in the amount of time participants were able to access and use firearms during each recall period. As a result, a variable indexing the number of months participants spent living outside of secure confinement was created for each recall period (called “street months”). Information on time spent in secure confinement was collected using questions from the Child and Adolescent Services Assessment scale (Ascher, Farmer, Burns, & Angold, 1996). A life events calendar covering each recall period was shown to participants to help facilitate accurate reporting on days of confinement. In addition to street months, a time-varying variable indexing whether or not the participant had spent any time in secure placement during the recall period was included as a covariate (0=No; 1=Yes).

Missing Data

Sample retention for the 10 follow-up assessments in the study was high (M=89%, range 84%-93%), with 730 participants completing all 10 assessments (62%). Missing data was handled using multiple imputation by chained equations (MICE) implemented using Stata 15.1 (StataCorp, 2016). MICE fills in missing observations using multiple values drawn from varying distributions based on the nature of each individual variable to create M complete datasets. Analyses are then run on each dataset and the results are combined using Rubin’s rules (Schafer & Graham, 2002). This MI procedure provides unbiased and efficient estimates under the assumption that the data mechanism is missing at random (MAR). Even when the MAR assumption is violated, MI tends to produce more efficient and less biased estimates than other ad hoc procedures (e.g., listwise or pairwise deletion), particularly when sample retention is high (Schafer & Graham, 2002).

A total of 50 imputed datasets were generated. Consistent with prior research (Beardslee et al., 2018), values on neighborhood disorder, gun carrying and gun violence were imputed when participants were in secure confinement for the entire recall period (i.e., censored observations). Relatedly, 23 participants who were in secure confinement for the entire recall period at every assessment were excluded from the analyses. When the models were re-run using the unimputed data, the findings were largely unchanged (results available upon request).

Data Analysis Plan

Generalized estimating equation (GEE) transitional models with a logit link function were run to examine the associations between the risk and protective factors and future gun violence (Overall & Tonidanal, 2004). The timeseries data were restructured so risk and prosocial protective factors measured at each assessment (T) were predictors of gun violence assessed at the subsequent assessment (T+1). GEE transitional models were used because they provide parameter estimates representing the association between predictors measured at T and future gun violence at T+1 aggregated across all assessment waves. These models are useful for predicting outcomes that are low base rate at any given timepoint (like gun violence), because model parameters are estimated using all events that occur across the time series, thereby increasing statistical power. Robust standard errors were used to account for the non-independence of the repeated assessments within individuals over time (Overall & Tonidanal, 2004). Prior to conducting the analyses, non-categorical risk and protective factors scales were converted to z-scores using the mean and standard deviation of each variable aggregated across the entire time series. As a result, odds ratios for continuous risk and protective factors represent the change in the predicted odds of gun violence per one standard deviation unit change in the predictor.

Three primary GEE models were run to examine the association between the measured risk and protective factors and later gun violence. The first model included only risk factors, the second model included only protective factors, and the third model included both risk and prosocial protective factors. This was done to elucidate the predictive utility of risk factors and prosocial protective factors before and after statistically controlling for their co-occurrence. All models included covariates indexing race and gun violence at time T, as well as participants’ age, months on the street, and placement in a secure facility at time T+1.

Results

Descriptive Statistics

Table 1 provides descriptive statistics for all study variables at each timepoint based on the non-imputed data. Nearly half of the study participants reported carrying a gun at one or more assessment points (46%, n=524). Gun carrying was reported by 208 participants at a single timepoint, 120 participants at two timepoints, 82 participants at three timepoints, 52 participants at four timepoints, 33 participants at five timepoints, and 29 participants at six or more timepoints. Nearly one-quarter of participants reported engaging in gun violence at least once across the assessment waves (23%, n=266), which means 51% of all gun carriers engaged in gun violence. Gun violence was reported by 161 participants at a single timepoint, 62 participants at two timepoints, 29 participants at three timepoints, and 14 participants at four or more timepoints.

Table 1.

Descriptive Statistics for Study Variables

	Time 1 (N=1,087)		Time 2 (N=1,083)		Time 3 (N=1,057)		Time 4 (N=1,060)		Time 5 (N=1,061)		Time 6 (N=1,056)		Time 7 (N=1,042)		Time 8 (N=1,031)		Time 9 (N=1,004)		Time10 (N=962)
	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)
Age	17.1	(1.1)	17.6	(1.1)	18.0	(1.1)	18.5	(1.1)	19.0	(1.1)	19.5	(1.1)	20.5	(1.1)	21.5	(1.1)	22.5	(1.1)	23.5	(1.1)
Gun violence	6.0%		5.6%		4.5%		3.5%		3.2%		3.0%		5.1%		4.0%		3.7%		2.7%
Gun carrying only	6.8%		6.9%		8.9%		8.8%		7.2%		7.6%		10.6%		10.7%		8.2%		6.0%
Risk factors
Non-gun serious violence	22.9%		18.6%		15.2%		13.8%		9.8%		10.3%		13.3%		10.1%		9.5%		8.0%
Drug dealing	16.3%		16.8%		16.1%		17.0%		13.6%		15.0%		18.4%		16.9%		17.0%		15.4%
Gang membership	13.9%		12.4%		11.5%		11.0%		10.1%		9.3%		8.7%		7.6%		7.2%		6.6%
Regular heavy drinking	11.9%		13.7%		15.8%		16.1%		15.9%		18.9%		21.6%		22.1%		25.2%		27.6%
Poor impulse control	3.1	(0.9)	3.2	(0.9)	3.2	(1.0)	3.0	(0.9)	3.2	(0.9)	3.2	(1.0)	3.3	(1.0)	3.2	(1.0)	3.3	(1.0)	3.3	(1.0)
Personal rewards of crime	2.4	(2.5)	2.3	(2.4)	2.1	(2.4)	2.1	(2.4)	1.9	(2.3)	1.8	(2.3)	1.6	(2.2)	1.7	(2.3)	1.6	(2.2)	1.6	(2.3)
Peer gun carrying	2.0	(1.3)	1.9	(1.2)	1.9	(1.2)	1.8	(1.2)	1.8	(1.1)	1.7	(1.1)	1.8	(1.1)	1.8	(1.1)	1.8	(1.1)	1.7	(1.0)
Peer non-gun delinquency	1.9	(0.9)	1.8	(0.8)	1.7	(0.8)	1.7	(0.8)	1.6	(0.7)	1.5	(0.7)	1.6	(0.8)	1.6	(0.7)	1.5	(0.7)	1.5	(0.7)
Exposure to violence	1.2	(1.6)	1.1	(1.5)	1.1	(1.5)	0.9	(1.4)	0.9	(1.3)	0.8	(1.3)	1.1	(1.6)	1.0	(1.6)	0.9	(1.4)	0.9	(1.4)
Neighborhood disorder	2.3	(0.9)	2.3	(0.8)	2.4	(0.8)	2.3	(0.8)	2.3	(0.8)	2.3	(0.8)	2.3	(0.8)	2.3	(0.8)	2.2	(0.8)	2.2	(0.8)
Protective factors
Concern for others	3.5	(0.9)	3.5	(0.8)	3.5	(0.9)	3.6	(0.8)	3.6	(0.8)	3.6	(0.8)	3.7	(0.8)	3.8	(0.8)	3.8	(0.8)	3.7	(0.8)
Expectations for work/family	3.8	(0.8)	3.9	(0.8)	3.9	(0.8)	3.9	(0.9)	3.9	(0.9)	3.9	(0.9)	4.0	(0.9)	4.0	(0.9)	4.1	(0.8)	4.0	(0.8)
Aspirations for work/family	4.7	(0.5)	4.7	(0.5)	4.7	(0.5)	4.7	(0.5)	4.7	(0.5)	4.7	(0.5)	4.8	(0.4)	4.8	(0.4)	4.8	(0.4)	4.8	(0.4)
Religious beliefs	3.1	(1.3)	3.0	(1.3)	3.0	(1.3)	3.0	(1.3)	3.0	(1.3)	3.0	(1.3)	3.0	(1.3)	3.0	(1.3)	3.1	(1.3)	3.0	(1.3)
Strong adult social support
No adult supports	19.6%		22.5%		22.0%		25.0%		29.3%		30.0%		29.1%		28.1%		35.4%		38.5%
One adult support	69.8%		69.6%		71.2%		69.1%		63.9%		63.7%		66.4%		66.0%		60.2%		56.4%
Two adult supports	10.6%		7.9%		6.8%		5.9%		6.8%		6.3%		4.5%		5.9%		4.4%		5.1%
Time in Secure Facility
Entire recall period	18.7%		17.0%		13.9%		13.2%		13.8%		14.8%		8.6%		7.7%		6.5%		10.1%
Part of recall period	50.7%		47.6%		40.2%		40.7%		36.4%		36.3%		45.7%		46.0%		43.5%		42.6%
Street months	3.6	(2.5)	3.8	(2.4)	4.3	(2.2)	4.4	(2.3)	4.6	(2.2)	4.7	(2.1)	8.9	(4.4)	8.8	(4.4)	8.8	(4.6)	8.7	(4.5)

Table 2 provides more detailed information about patterns of gun carrying and gun violence after eliminating participants who spent fewer than 30 days (<1 month) outside a secure facility at each timepoint. These participants were excluded because they had little or no time to access firearms. As seen in Table 2, the percentage of youth reporting any gun carrying gradually decreased across the 6-month recall periods (Time 1=18.0% to Time 6=13.6%). Gun carrying prevalence increased at Time 7 (19.2%) before gradually declining through Time 10 (11.0%). The increase at Time 7 was due to a shift from 6-month to 12-month assessments, which expanded the amount of time youth had to carry a gun in a recall period. Consistent with this notion, logistic GEE analysis indicated that age was negatively associated with gun carrying across the time series after controlling for differences in the number of street months within a recall period (OR=.93, p<.001). In contrast, the average rate of gun carrying per recall period month increased across the assessment wave. For example, among those youth who reported any gun carrying, the percent who reported engaging in this behavior once a week or more during the recall period (on average) rose from 40.8% to 64.2% from Time 1 to Time 10.

Table 2.

Gun Carrying and Gun Violence for Participants who Spent at Least 30 Days Outside a Secure Facility

	Time 1 (N=650)		Time 2 (N=702)		Time 3 (N=778)		Time 4 (N=780)		Time 5 (N=786)		Time 6 (N=797)		Time 7 (N=846)		Time 8 (N=836)		Time 9 (N=790)		Time10 (N=736)
	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)	M/%	(SD)
Age	17.0	(1.1)	17.5	(1.1)	18.1	(1.1)	18.5	(1.1)	19.0	(1.1)	19.5	(1.1)	20.5	(1.1)	21.5	(1.1)	22.5	(1.1)	23.5	(1.1)
Gun violence
Any gun violence	8.9%		7.7%		5.7%		4.4%		4.1%		3.9%		6.2%		4.9%		4.6%		3.3%
Robbed with a gun	4.3%		3.7%		2.6%		3.0%		2.8%		2.0%		3.2%		2.3%		2.5%		0.8%
Shot at but missed	7.1%		6.3%		5.0%		3.1%		2.4%		3.0%		4.6%		3.6%		3.5%		2.6%
Shot and hit	1.9%		2.6%		1.0%		0.4%		0.8%		0.6%		1.2%		1.4%		0.6%		0.8%
Gun carrying
Any gun carrying	18.0%		16.2%		17.0%		15.6%		12.6%		13.6%		19.2%		17.8%		14.5%		11.0%
Frequency of carrying
< 1 time a month	40.0%		35.0%		29.9%		35.7%		26.3%		19.4%		25.0%		23.5%		19.3%		18.5%
1-3 times per month	19.2%		18.0%		14.2%		17.4%		5.1%		13.0%		12.5%		13.4%		16.0%		17.3%
1-4 times per week	13.3%		19.7%		23.6%		19.1%		22.2%		23.2%		23.1%		22.8%		25.2%		24.7%
> 4 times per week	27.5%		27.4%		32.3%		27.8%		46.5%		44.4%		39.4%		40.3%		39.5%		39.5%

Note. Numbers based on subsample of youth who spent at least 30 days outside a secure facility at each timepoint. Recall periods are six months for Time 1-6 and twelve months for Time 7-10. Average frequency of gun carrying was calculated by dividing the number of times a participant carried by the number of street months in a recall period.

As seen in Table 2, the prevalence of gun violence also decreased from Time 1 (8.9%) to Time 10 (3.3%). The most common form of gun violence reported at each timepoint tended to be shooting at someone without hitting them (2.6%-7.1%). A smaller proportion of youth who carried guns reported engaging in gun violence from Time 1 (18% carried a gun, 9% committed gun violence) to Time 10 (11% carried a gun, 3% committed gun violence).

Risk Factors Predicting Gun Violence

An initial baseline model was run examining the extent to which gun carrying/use status at T was associated with gun violence at T+1 across the time series, controlling for age, race, street time, and secure confinement. As expected, both gun carrying without gun violence (OR=4.59, 95% CI=3.45-6.08) and gun violence (OR=10.68, 95% CI=7.54-15.12) were robust predictors gun violence at the next timepoint. However, most youth who carried guns did not engage in later gun violence. Specifically, post hoc crosstab analysis indicated the predicted probability that youth would engage in gun violence at the subsequent wave was 2.5% (95% CI=2.1%-2.9%) for non-gun carriers, 11.5% (95% CI= 9.1%-13.8%) for gun carriers without gun violence, and 22.9% (95% CI=18.1%- 27.8%) for youth who engaged in gun violence. Crosstab analysis also indicated that 50.6% (95% CI = 44.9%-56.3%) of the acts of gun violence reported across the time series were committed by youth who did not report carrying a gun at the prior wave.

Table 3 shows results for an analysis using only risk factors (T) to predict future gun violence (T+1), after controlling for prior gun violence (T), gun carrying without gun violence (T), and other model covariates. The effects associated with the covariates of age, race, street months, and placement in a secure facility for all models are presented in supplemental Table S2. Six of the ten risk factors significantly predicted future gun violence: Regular heavy drinking, drug dealing, personal rewards of crime, gang membership, peer gun carrying, and exposure to violence. Although gun violence (OR=3.08, p<.001) and gun carrying (OR=2.08, p<.001) remained robust predictors of future gun violence, the magnitude of the OR for these predictors was significantly attenuated relative to the baseline model that did not include other risk factors.

Table 3.

Generalized Estimating Equation Models Examining Risk and Protective Factors as Predictors of Later Gun Violence

	Gun Violence (T+1)
	Model 1: Risk Only			Model 2: Protective Only			Model 3: Full Model
Predictors (T)	OR	95% C.I.	p	OR	95% C.I.	p	OR	95% C.I.	p
Gun violence	3.08	[1.94, 4.88]	<.001	9.27	[6.53, 13.16]	<.001	2.75	[1.74, 4.35]	<.001
Gun carrying only	2.08	[1.46, 2.95]	<.001	4.26	[3.19, 5.67]	<.001	1.99	[1.39, 2.83]	<.001
Risk Factors
Non-gun serious violence	1.24	[0.90, 1.70]	0.191				1.25	[0.91, 1.72]	0.166
Regular heavy drinking	1.46	[1.06, 1.99]	0.019				1.43	[1.05, 1.95]	0.025
Drug dealing	1.50	[1.13, 1.98]	0.005				1.47	[1.11, 1.94]	0.006
Poor impulse control	0.92	[0.80, 1.06]	0.246				0.94	[0.82, 1.08]	0.380
Personal rewards of crime	1.24	[1.10, 1.40]	<.001				1.20	[1.06, 1.35]	0.004
Gang membership	1.55	[1.12, 2.14]	0.009				1.50	[1.08, 2.08]	0.016
Peer gun carrying	1.27	[1.10, 1.47]	0.001				1.30	[1.12, 1.50]	<.001
Peer non-gun delinquency	0.91	[0.78, 1.05]	0.192				0.88	[0.76, 1.03]	0.107
Exposure to violence	1.14	[1.03, 1.26]	0.009				1.18	[1.06, 1.30]	0.002
Neighborhood disorder	1.07	[0.94, 1.22]	0.318				1.08	[0.95, 1.24]	0.253
Protective Factors
Concern for others				0.84	[0.75, 0.94]	0.003	0.87	[0.77, 0.98]	0.025
Expectations for work/family				1.00	[0.88, 1.14]	0.988	1.05	[0.92, 1.21]	0.480
Aspirations for work/family				0.84	[0.76, 0.92]	<.001	0.82	[0.74, 0.91]	<.001
Religious beliefs				0.86	[0.76, 0.98]	0.021	0.87	[0.76, 0.99]	0.029
Strong adult social supports
One adult support				1.02	[0.80, 1.30]	0.887	1.00	[0.78, 1.28]	0.987
Two adult supports				0.46	[0.23, 0.91]	0.026	0.44	[0.21, 0.89]	0.022

Note. OR = odds ratio; C.I. = confidence interval. Sample size is 1,147 participants with 10,323 observations. Covariates indexing street months (T+1), age (T+1), placement in a secure facility (T+1), and race were included in all models.

Protective Factors Predicting Gun Violence

Table 3 also shows results for an analysis using only protective factors (T) to predict future gun violence (T+1). Four of the five protective factors were significantly associated with a decreased risk of engaging in future gun violence: Concern for others, aspirations for work/family, religious beliefs, and having at least two adult social supports. In the protective factor only model, the strength of the association between prior gun violence (OR=9.27, p<.001) and gun carrying without gun violence (OR=4.26, p<.001) on later gun violence remained largely unchanged relative to the baseline model. This clearly demonstrates that co-occurring risk factors, but not protective factors, dramatically attenuated the association between prior gun carry and future gun violence.

Risk and Protective Factors Combined Predicting Gun Violence

Results for the final model that included all risk and protective factors as predictors of future gun violence are presented in Table 3. All statistically significant risk factors and protective factors previously identified continued to significantly predict future gun violence when included together as predictors in this final model. Prior engagement in gun violence (OR=2.75, p<.001) and gun carrying without engaging in gun violence (OR=1.99, p<.001) continued to significantly predict future gun violence in the final model. Post hoc analyses indicated the magnitude of the odds ratios for these two predictors were not significantly different from one another (t=1.46, p=.15).

Discussion

The current study found that gun carrying was a robust predictor of future gun violence, even after controlling for prior gun violence and several other risk and protective factors. This is consistent with evidence indicating that gun carrying emboldens adolescents to engage in serious violence and other criminal acts (Emmert, Hall, & Lizotte, 2017). However, results indicated that most juveniles who carried guns did not engage in subsequent gun violence. Moreover, several risk factors besides gun carrying incrementally contributed to the prediction of future gun violence and significantly attenuated the strength of the association between prior carrying and later gun violence. These risk factors spanned models involving antisocial propensity, social influence, and self-protection explanations of adolescent gun carrying and gun violence. Although prior longitudinal studies have found that these risk factors predict future gun carrying (Oliphant et al., 2019), this is the first longitudinal study to demonstrate that these factors incrementally contribute to the prediction of future gun violence after controlling for co-occurring gun carrying. In addition, this is the first longitudinal study to delineate several protective factors that identified youth at decreased risk for future gun violence, even after controlling for the negative influence of risk factors. Taken together, these findings suggest that targeted interventions designed to prevent gun violence among male juvenile offenders should focus on reducing several aversive risk factors besides gun carrying, as well as strengthening protective factors that may offset these risks.

Patterns of Gun Carrying and Gun Violence

Consistent with studies based on community samples (Docherty, Beardslee, Grimm, & Pardini, 2019; Vaughn et al., 2016), gun carrying was a relatively transient behavior for a significant proportion of juvenile offenders. Specifically, among those youth who reported carrying a gun at least once across the 10 assessments, 40% reported carrying at a single timepoint and 23% reporting carrying at two timepoints. Although there was a steady decrease in the prevalence of gun carrying among juvenile offenders from the late teens to early 20s (after adjusting for difference in street time), the frequency of gun carrying per street month among active carriers tended to increase over this developmental period. Despite this increase in carrying frequency, the proportion of active gun carriers who also reported engaging in acts of gun violence at each assessment declined over time. This suggests that juvenile offenders who possess guns may be more likely to use them during commission of a crime than adult offenders who own guns. Taken together, these findings indicate that there is considerable heterogeneity in juvenile offenders’ patterns of gun carrying across development, making it important for future studies to examine the processes underlying transient versus persistent gun carriers (Vaughn et al., 2016).

Risk Factors for Gun Violence

Several risk factors associated with an antisocial propensity predicted engagement in future gun violence, including viewing crime as personally rewarding and engaging in drug dealing. Consistent with these results, longitudinal evidence indicates that adolescents who have tolerant attitudes toward antisocial behavior are at risk for engaging in later gun violence (Goldstick et al., 2019). Longitudinal studies have also found that adolescents who deal drugs in urban environments are more likely to engage in serious forms of violence, including homicide (Loeber et al., 2005). Drug dealers’ engagement in gun violence can serve several purposes, including resolving disputes over drug sales, managing conflicts with rival dealers, eliminating informants, punishing customers for non-payment, and/or retaliating against armed robbers (Jacques & Wright, 2008). For this reason, effective strategies aimed at reducing gun violence will likely require concerted efforts to combat the proliferation and maintenance of illegal drug markets.

The current findings also indicated that juvenile offenders who engaged in regular heavy drinking were more likely to commit acts of gun violence. This association is presumably due to the acute effects alcohol intoxication has on cognitive and behavioral functioning (Kuypers, Verkes, van den Brink, van Amsterdam, & Ramaekers, 2020). Intoxicated individuals are more likely to misperceive social cues as threatening, escalate conflicts, and engage in risky behaviors that have the potential for adverse consequences (Kuypers et al., 2020). As a result, heavy alcohol consumption may be a “tipping point” that facilitates acts of gun violence during interpersonal conflicts among juvenile offenders. In contrast, there was no evidence that hard drug or marijuana use was associated with future gun violence in the current sample (see Table S1). Other studies have found evidence indicating that illicit drug users are not the primary drivers of gun violence within illegal drug markets (McGinty & Webster, 2017), though they may be prone to engage non-violent property crimes (White, Loeber, & Farrington, 2008).

In support of social influence models, both gang membership and affiliation with gun-carrying peers independently predicted engagement in gun violence. Importantly, the effect of gang membership on later gun violence was not explained by co-occurring engagement in drug dealing. In addition to gang membership, greater affiliation with peers who carry guns exerted an independent effect on later gun violence. Associating with gun carrying peers provides youth with easy access to firearms and can place them in risky situations where gun violence is more likely to occur (Roberto et al., 2018). These peers may also reinforce the notion that guns are an effective mechanism for resolving disputes, exerting power, and protecting oneself. In contrast, there was no evidence that affiliation with peers engaged in non-gun related offending contributed to the prediction of future gun violence. This suggests that future studies should focus more specifically on gang membership and peer gun carrying as risk factors for gun violence, as opposed to general peer delinquency.

Juvenile offenders who reported higher indirect and direct exposure to violence were also more likely to engage in gun violence, consistent with self-protection models. Prior longitudinal studies have found that adolescents who witness and/or experience serious violence are more likely to engage in future gun carrying (Beardslee et al., 2018; Spano & Bolland, 2013) and commit acts of serious violence (Bingenheimer, Brennan, & Earls, 2005). Although the current study extends these findings by linking victimization to later gun violence, the mechanisms that may account for this association remain unclear. According to self-protection theory, youth who are exposed to violence are more likely to carry and use guns to ward off potential attackers in self-defense. However, it is also possible that juvenile offenders who are violently attacked, or witness loved ones being victimized, are more likely to use gun violence as a means for vengeful retaliation (Goldstick et al., 2019). To further clarify this issue, future studies should collect additional details from adolescents about the situational and motivational factors that precipitated reported acts of gun violence.

Three risk factors did not uniquely contribute to the prediction of future gun violence: impulse control problems, engagement in non-gun-related violence, and neighborhood disorganization. Prior longitudinal studies have found that poor impulse control is inconsistently associated with later serious violence, particularly after controlling for co-occurring conduct problems (Moffitt et al., 2008; Pardini, Byrd, Hawes, & Docherty, 2018). Evidence suggests impulse control problems may underlie the emergence of conduct problems in children, but have less utility in delineating conduct problem youth at risk for serious violence (Pardini et al., 2018). Relatedly, non-gun-related serious violence did not incrementally contribute to the prediction of future gun violence in the current study. This suggests that violent acts involving the use of guns are not simply the result of a general propensity toward aggression.

Although neighborhood disorder was not significantly associated with later gun violence, the current study focused on serious adolescent offenders who predominately came from disadvantaged and segregated neighborhoods. Evidence from community samples suggests that the influence of familial and neighborhood disadvantage on adolescent gun violence is largely mediated through risk factors that emerge earlier in development, such as increased affiliation with delinquent peers and higher levels of conduct problems in childhood (Beardslee, Docherty, Mulvey, et al., 2019). Similarly, prior cross-sectional analyses with the current sample has found that the association between neighborhood disorder and delinquent behavior is mediated through its association with peer deviance (Chung & Steinberg, 2006). Together, these findings suggest that neighborhood disorder may facilitate an accumulation of risk factors across childhood and early adolescence, which in turn increase the likelihood that youth will engage in serious delinquency, including gun violence.

Protective Factors for Gun Violence

Aspects of juvenile offenders’ interpersonal relationships emerged as important protective factors for gun violence, even after controlling for multiple risk factors. Specifically, adolescents who reported having substantive bonds with at least two supportive adults were at lower risk for engaging in gun violence. This is consistent with studies indicating that juvenile and adult offenders who have positive adult social support are at reduced risk for exhibiting aggressive behaviors and are less likely to recidivate upon release (Intravia et al., 2017; Ullrich & Coid, 2011). Interestingly, youth who reported having only one supportive adult were not less likely to engage in gun violence compared to those who reported having no adult supports. Previous studies have found that most adjudicated adolescents report having at least one supportive adult, whereas the mean number of supportive adults in community samples of adolescents is just over two (Zwecker, Harrison, Welty, Teplin, & Abram, 2018). As a result, the current findings suggest that the protective effective of strong adult supports occurs when the size of juvenile offender’s support network is within a normative range.

Prosocial characteristics, beliefs and aspirations also emerged as important protective factors in the current study. For example, juvenile offenders who reported higher sensitivity to others’ feelings and a desire to help others were at decreased risk for engaging in gun violence. Other longitudinal studies have found that adolescents who report empathic concern and caring for others are less likely to engage in more minor forms of violence (Keil et al., 2019; Wall Myers et al., 2018). Juvenile offenders who endorsed religious beliefs as a source of comfort and support were also less likely to engage in future gun violence. Prior longitudinal studies have found mixed evidence for religiosity as a protective factor against recidivism (Giordano et al., 2008; Ullrich & Coid, 2011). It is possible that religious beliefs primarily serve as a protective factor for violent acts with a high potential for serious injury or death, as these acts are antithetical to fundamental religious codes regarding moral behavior. Lastly, juvenile offenders’ personal aspirations for achieving conventional adult goals (e.g., getting a good job, creating a positive family environment) were also protective against future gun violence. Prosocial aspirations may have emerged as a more robust protective factor than expectations for obtaining prosocial goals in the current study because they more closely represent adolescents’ personal commitment to conformity.

Limitations

The findings from this study need to be interpreted in the context of several limitations. As with all longitudinal studies, the associations observed may be accounted for by unmeasured confounding factors (i.e., non-causal). In addition, the predictors and outcomes in this study were based solely on self-report, which may have resulted in biased associations due to shared method variance. The findings were also based on male juvenile offenders from two urban cities who were at high risk for engaging in gun violence. Although focusing on this population helped to address the low base rate problem associated with predicting gun violence, the results may not generalize to females, community-based samples, or juvenile offenders from different geographic locations. Given that the prevalence of gun violence was relatively low even in this high-risk sample, community-based longitudinal studies will likely have to oversample antisocial youth in order to examine gun violence as an outcome. Relatedly, future studies should include questions regarding engagement in less serious forms of gun violence, such as threatening others with a gun (Goldstick et al., 2019). It will also be important for future investigations to assess the precipitating factors that lead to the initial onset of gun violence, as well as the situational and motivational factors that underlie specific acts of gun violence. Moreover, the current study focused on proximal predictors of gun violence, but did not examine potential mediating mechanisms. For example, evidence suggests that factors such as neighborhood disorder and socioeconomic disadvantage may be related to firearm violence (at least in part) through their association with early conduct problems and exposure to deviant peers in childhood (Beardslee, Docherty, Mulvey, et al., 2019).

A final limitation worth noting involves the conceptualization of protective factors used in this study. Consistent with prior work, a protective factor was defined as a prosocial aspect of the individual and/or socio-contextual environment associated with a reduced probability of gun violence among a high-risk sample (Ttofi et al., 2016). However, alternative conceptualizations of protective factors have been proposed in the literature (Lösel & Farrington, 2012; Ttofi et al., 2016). For example, a “buffering protective” factor has been conceptualized as a variable that nullifies the negative effect of a risk factor (i.e., an interaction effect), but is not directly associated with the outcome (Lösel & Farrington, 2012). Other studies have conceptualized risk and “direct protective” factors as opposite poles of the same construct, rather than distinct variables (Lösel & Farrington, 2012). Consequently, future investigations are needed to delineate protective factors for gun violence using these alternative conceptualizations.

Implications

This study clearly identified several risk and protective factors besides gun carrying that prospectively predicted later gun violence among male juvenile offenders. These findings provide empirical evidence supporting potential targets for interventions designed to prevent gun violence among males in the juvenile justice system. Although some of these factors are likely more difficult to modify (e.g., getting youth to care about others) than others (e.g., reducing regular heavy drinking), interventions that reinforce prosocial strengths and reduce aversive risks may be most effective. On a more macro-level, comprehensive approaches to gun violence prevention should also include coordinated community-based strategies (e.g., targeted policing, gang suppression) designed to reduce adolescents’ access to firearms, drug dealing, gang involvement, and violent victimization.