Indirect Associations between Middle-Childhood Externalizing Behaviors and Adolescent Substance Use through Late-Childhood Exposure to Violence

Abstract

Longitudinal research to understand individual risk factors in childhood associated with exposure to violence and substance use is needed to inform prevention efforts. The present study tested indirect associations between age 8.5 externalizing behaviors and age 16 substance use through age 9.5 violence victimization and witnessing. Participants were 650 racially diverse (48.6% European American, 28.1% African American, 13.3% multiracial, and 10.0% other), predominantly socioeconomically disadvantaged youth (49% female). Externalizing behaviors were associated with higher levels of violence victimization and witnessing. The indirect path from externalizing behaviors to substance use was significant through victimization but not witnessing violence. Interventions aimed at reducing early externalizing behaviors may reduce risk for violence victimization, which may, in turn, reduce risk for adolescent substance use.

Introduction

Exposure to violence is a common environmental stressor for many youth (McCrea et al., 2019), and is associated with numerous maladaptive psychosocial outcomes, including substance use (Frank et al., 2011). However, more work is needed to understand individual risk factors (e.g., child gender, child problem behavior) for exposure to violence to inform prevention and intervention (Begle et al., 2011). Using data from a large, prospective, longitudinal study of racially diverse, low-SES children, a population at elevated risk for exposure to violence (Buka et al., 2001), the present study proposes and tests a model where externalizing behaviors in late childhood are related to substance use in adolescence indirectly through victimization and witnessing violence. Additionally, as rates of exposure to violence and developmental sequelae following exposure to violence have been found to be differ by gender (Pinchevsky et al., 2013), the present study tests whether indirect associations between externalizing behaviors and substance use through exposure to violence differ for males and females.

Youth may experience violence directly, as victims, or indirectly by witnessing violence against someone else. Nationally, prevalence rates of direct exposure, referred to as victimization in the present study, are highest among school-age children relative to adolescents or younger children, and primarily encompass physical or verbal violence by a family member or peer (Finkelhor et al., 2015). Rates of indirect exposure to violence, referred to as witnessing violence in the present paper, are highest among adolescents, and are primarily experienced in the community (Finkelhor et al., 2015), which likely reflects the greater autonomy and time spent in the community for adolescents relative to younger children.

Research over the past few decades has shed light on individual and environmental factors that elevate risk for exposure to violence. For example, it is well established in the literature that based on centuries of oppression and systemic societal inequities (McCrea et al. 2019), economically marginalized youth, and especially youth of color (Thomas et al., 2012), are exposed to more violent environments than their more affluent peers. Epidemiological studies suggest that nearly all youth in urban environments are exposed to some form of violence and many experience direct victimization either in the neighborhood, at home via domestic violence or harsh, abusive parenting, or through bullying at school (Finkelhor et al., 2015). In terms of individual factors that elevate the risk for exposure to violence, youth who display higher levels of externalizing problem behaviors (e.g., aggression, impulsivity) are more likely to be victims and/or witness violence (Lambert et al., 2010). Although prior work suggests that rates of exposure to violence may be higher for males than females and that males rather than females may be most at risk for early substance use (Pinchevsky et al., 2013), there are mixed findings in this literature that warrant further investigation.

In addition to increasing the risk for experiencing violence in the environment, externalizing behaviors are an established risk factor for substance use in adolescence (Colder et al., 2013). Youth who display high levels of pure externalizing behaviors, as opposed to youth who display high level of pure internalizing behaviors (e.g., social withdrawal, depressed mood, anxious avoidance) or co-occurring internalizing and externalizing behaviors (e.g., depressive and aggressive behaviors), are at an increased propensity to use tobacco, alcohol, and marijuana in adolescence (Colder et al., 2013). Similarly, physical aggression in early childhood has been shown to be related to tobacco, alcohol, cannabis, and hard drug use at age 18, accounting for sex and other disruptive behaviors (i.e., oppositional behaviors, status violations and property violations) (Timmermans et al., 2008).

Externalizing Behaviors and Exposure to Violence

Elevated externalizing behaviors may increase the likelihood of a child witnessing violence or directly experiencing violence in the home or community (Verhoeven et al., 2010). In a prospective study of children in inner-city Baltimore, both parent- and teacher-reports of aggressive behaviors in first grade were predictive of higher levels of witnessing community violence in middle school (Boyd et al., 2003). However, anxious symptoms in first grade attenuated this association, suggesting that over-inhibition may be protective against exposure to community violence. On the other hand, impulsive behaviors, which are often included in assessments of externalizing behaviors (Achenbach & Rescorla, 2001), appear to heighten the risk for exposure to violence (Verlinden et al., 2014).

Considering Bronfenbrenner’s ecological systems theory (1992) in which individual characteristics (e.g., child gender and behavior) interact with microsystem level factors (e.g., family level variables), which then interact with macrosystem level factors (e.g., neighborhood, societal level variables), it is possible that there is a dynamic interplay between individual risk factors (e.g., youth history of problem behavior), environmental risk factors (e.g., parenting practices, school and neighborhood characteristics), and exposure to violence over time. Prior work highlights how children’s behavior can influence environmental exposures. Elevated disruptive behaviors in early childhood have been found to elicit harsh parenting (Hipwell et al., 2008), increase the likelihood of maltreatment (Stith et al., 2009), and evoke aggressive behaviors from siblings (Williams et al., 2007). In later developmental stages, factors outside the home (i.e., peer relationships and neighborhood characteristics) may become increasingly influential relative to factors within the family context (Côté, 2009).

However, it is also possible this association functions in the reverse direction in which exposure to violence increases the likelihood of children exhibiting problem behaviors. Whether at home or in the neighborhood or school context, exposure to violence has been linked with increased levels of externalizing behaviors and other behavior problems for girls and boys (Sharkey, 2018). Exposure to violence has also been found to account for the association between problem behavior in childhood and risky or delinquent behavior in adolescence. For example, callous-unemotional (CU) traits exhibited in seventh grade have been shown to be related to later delinquency (including drug use) in twelfth grade and the association was mediated by exposure to violence experienced in grades seven to eleven (Oberth et al., 2017). Exposure to violence also mediated the link between CU traits and delinquent behaviors for detained adolescent boys (Howard et al., 2012). Further, CU traits were related to deficits in emotional processing of distressing stimuli in detained adolescent boys with elevated levels of exposure to community violence (Kimonis et al., 2008). Exposure to community violence in particular has also been found to be associated with interpersonal problems and disruptive antisocial behavior for detained adolescent girls and boys (Schraft et al., 2013).

Exposure to Violence and Substance Use

Previous research has highlighted the association between exposure to violence and increased levels of substance use in adolescence (James et al., 2018) and adulthood (Menard et al., 2015). In two separate prospective studies of children born to substance abusing mothers, researchers demonstrated associations between exposure to violence in late childhood, assessed as a composite of victimization and witnessing, and later endorsement of substance use, accounting for prenatal substance use exposure (Lagasse et al., 2006). Family and parent level factors, like those addressed in the previous study (e.g., parenting practices, substance use), are important variables to account for that may increase the likelihood of child externalizing behaviors, exposure to violence, and adolescent substance use. In a prospective cohort study of inner-city, pre-adolescent and adolescent youth, violence victimization was related to increases in alcohol use at a 2.5 year follow up (Fagan et al., 2014). Among a sample of African American youth (mean age 11.14), alcohol use was concurrently associated with violence victimization, but not witnessing violence (Taylor & Kliewer, 2006). However, using data from the Fragile Families and Child Wellbeing Study, a positive association was found between the amount of violence witnessed in the family between ages 5 and 9 and substance use (tobacco, alcohol, or illicit drug use) at age 15 (James et al., 2018).

Although there is a well-established link between exposure to violence and substance use, there are a few limitations that remain. There have been varying approaches to measuring exposure to violence, including assessing only victimization (Nickerson & Slater, 2009), only witnessing (Joseph et al., 2006), or using a composite of victimization and witnessing (Lambert et al., 2010). However, victimization and witnessing have less frequently been included as separate constructs in the same model (e.g., Oberth et al., 2021; Pinchevsky et al., 2013), which may obscure associations between predictor variables and unique experiences of violence, as well as unique contributions of victimization or witnessing violence to psychosocial development. Examining victimization and witnessing as separate constructs in the same model is particularly important based on previous work suggesting that witnessing violence is more strongly associated with externalizing behaviors (e.g., Boxer et al., 2008), difficulty concentrating and intrusive thoughts (Howard et al., 2002), whereas victimization is more strongly associated with internalizing behaviors (e.g., Fitzpatrick & Boldizar, 1993) and feelings of despondency (Howard et al., 2002). Studies that have included victimization and witnessing violence as separate constructs have focused on exposure to violence in adolescence. However, as exposure to violence is common in earlier developmental stages (Finkelhor et al., 2015), research limited to exposure to violence in adolescence may miss the developmental implications of exposure to violence at earlier ages where experiences can set the stage for successful or maladaptive adolescent development (Olson et al., 2021).

Theories explaining the association between exposure to violence and substance use generally posit that substance use arises as a maladaptive coping strategy to manage psychological symptoms stemming from exposure to violence. For example, the self-medication hypothesis suggests that substance use begins as an attempt to alleviate negative psychological symptoms, and is reinforced when the use of substances temporarily improves the affective state of the user (Shin et al., 2020). Exposure to violence is predictive of increases in depressive and anxious symptoms (Lynch, 2003) and both depressive symptoms (Womack et al., 2016) and anxious symptoms (Haller & Chassin, 2014) have been linked to substance use in patterns that support the self-medication hypothesis. Alternatively, general strain theory focuses on interpersonal relationships and posits that exposure to violence results in fractured interpersonal relationships and subsequently, negative affect which, in the absence of helpful coping resources, may result in deviant behaviors, such as substance use, that provides temporary relief from the psychological and physical pain (Kaufman, 2009).

The Role of Gender

Extant research has demonstrated that males generally experience higher levels of violence both as victims and as witnesses (Buka et al., 2001). Beyond mean level differences, some research also suggests that child gender moderates psychosocial development following exposure to violence (Javdani et al., 2014). For example, violence victimization and witnessing have been shown to uniquely contribute to adolescent alcohol use in females, but neither victimization nor witnessing was related to alcohol use for males, perhaps because exposure to violence is more normative for boys (Pinchevsky et al., 2013). Likewise, witnessing violence has been found to be a stronger predictor of psychological distress for females than males (Javdani et al., 2014). Girls may appraise the stressfulness of a violent event differently than boys (Buckner et al., 2004). For example, girls may be more likely to ruminate about an event and have a more pervasive perception of threat of environmental danger than boys (Simonson et al., 2011). However, in other work, gender did not moderate the association between witnessing violence and substance use (James et al., 2018) or psychological symptoms (Zona & Milan, 2011). Thus, although there appears to be a consensus that males are more likely to experience violence than females, evidence regarding gender differences in psychosocial development following exposure to violence is mixed, with discrepancies pointing to more maladaptive outcomes for females. Based on these mixed findings in the literature, further exploratory investigation of gender in the associations between externalizing behaviors, exposure to violence, and adolescent substance use is necessary.

The Current Study

The current study adds to the existing body of literature by examining associations between behavior dysregulation (e.g., externalizing problem behaviors) in middle-childhood, environmental characteristics (e.g., exposure to violence) in late-childhood, and adolescent substance use using data from a large, prospective, longitudinal study of economically marginalized families from diverse racial and ethnic backgrounds, a population at risk to experience violence. First, it was hypothesized that higher levels of externalizing behaviors in middle childhood would be related to higher levels of violence victimization and witnessing in later middle childhood, which would be related to substance use in adolescence. Previous literature has demonstrated differences in rates of exposure to violence and substance use between males and females, yet mixed findings on the impact of exposure to violence on substance by gender. Thus, as an exploratory analysis, the present study examined gender as a moderator of the indirect paths from middle-childhood externalizing behaviors to adolescent substance use through violence victimization and violence witnessing. Additionally, the current study accounts for relevant child (e.g., gender and race), family (e.g., annual income and parenting practices), and community factors (e.g., site location) that may increase the likelihood of externalizing behaviors and exposure to violence during middle childhood, and adolescent substance use.

Methods

Participants

Participants were 731 parent-child dyads recruited as a part of the Early Steps Multisite Study, a longitudinal randomized controlled trial studying the efficacy of the Family Check-Up parenting intervention (Dishion et al., 2008). Families were recruited from Women, Infant, and Children (WIC) Nutritional Supplementation centers in and around the metropolitan areas of Pittsburgh, PA, Eugene, OR, and Charlottesville, VA. At recruitment, target youth (49% female) were between 2 years 0 months and 2 years 11 months of age. Families were screened to be at risk based on sociodemographic risk (i.e., primary caregivers had no more than 2 years post-high-school education and low income), child behavior problems (e.g., conduct problems and other externalizing problem behaviors, high-conflict relationships with adults), and family factors (e.g., maternal depression, substance-use problems, teen-parent status). To meet eligibility, families had to meet criteria for the child behavior problems risk factor and at least one of the other criteria (i.e., sociodemographic risk or family problems).

Of the 1,666 families initially approached at WIC centers across the three study locations, 879 met the eligibility requirements, and 731 consented to participate in the study. Of the 731 families recruited, 272 (37%) were recruited from Pittsburgh, 271 (37%) were recruited from Eugene, and 188 (26%) were recruited in Charlottesville. At the baseline assessment, primary caregivers reported their children belonging to the following racial groups: 27.9% African American, 50.1%, Caucasian, 13.0% biracial, and 8.9% other races (e.g., American Indian, Native Hawaiian). In terms of ethnicity, 13.4% of the sample identified as Hispanic American.

The study analytic sample included 642 families (88.9% of the original 731 families) that completed at least one of the age 8.5, 9.5, or 16 assessments, as the independent, mediator, and dependent variables were measured at those assessments. Attrition analyses were conducted on key demographic variables, covariates, and baseline externalizing behaviors (the primary independent variable). Retained families had higher incomes (t_118.92 = 2.60, p = .01, Cohen’s d = 0.28). Primary caregivers in retained families also smoked fewer cigarettes (t_89.53 = −2.51, p = .01, Cohen’s d = 0.32) than in attritted families. However, retained and attritted families did not significantly differ on child gender, child race, site location, intervention status, age 2 child externalizing behaviors, parent marijuana use, parent antisocial behavior, or harsh parenting.

Procedures

Data for the present study were collected during in-home assessments with families at child ages 2, 3, 4, 5, 7.5, 8.5, 9.5, and 16. Of note, primary caregivers in the study sample were the caregivers for only one target child in the sample. Assessments were conducted between 2002 and 2018 and involved a variety of tasks including questionnaires, demographic interviews, recorded parent-child and child-examiner interactions, and examiner impressions. Custodial parents provided written consent prior to the administration of any measures at each assessment. A Certificate of Confidentiality was obtained from the National Institute of Health to offer further protection of participants’ confidentiality and encourage honest reporting. Institutional review board approval was obtained at each site for all screening and assessment procedures.

Measures

Target youth externalizing behaviors.

Primary caregivers completed the Child Behavior Checklist 6–18 (CBCL), a parent-report inventory of emotional and behavioral problems in children ages 6–18 years old (Achenbach & Rescorla, 2001) at the age 8.5 assessment. The CBCL consists of 73 items asking parents to rate the presence of child behaviors over the past 6 months on a scale from 0 (not true), 1 (somewhat or sometimes true), or 2 (very true or often true). For purposes of the present study, we employed the broadband externalizing factor, which includes the 35 items on the rule-breaking and aggressive behaviors subscales. A sample item on the rule-breaking subscale is “lying and cheating” and a sample item on the aggressive behavior subscale is “gets in many fights.” The Cronbach alpha for externalizing behaviors was 0.93.

Exposure to Violence.

At the age 9.5 assessment, target youth completed the Exposure to Violence Scale, which assesses the frequency of witnessing and victimization of violent acts in the home, school, or neighborhood context within the past year (Finkelhor et al., 2009). Target youth responded on a 4-point Likert scale from 0 (never), 1 (sometimes), 2 (often), 3 (almost every day). A Victimization score was calculated by summing scores from thirteen items, which asked about instances in which the target youth was a victim of violent acts. An example victimization items is, “How often over the past year have you been beaten up in school?”. A Witnessing score was created by adding the scores from nine items related to instances in which the target youth witnessed a violent act. An example witnessing item is, “How often over the past year have you seen someone else getting beaten up in school?”. The additional four items included in the Victimization score pertained to bullying (i.e., getting picked on or chased, being called names). Cronbach alphas for the Victimization and Witnessing scores were 0.74 and 0.83, respectively.

Adolescent substance use.

At the age 16 assessment, target youth were asked to report if they have ever tried the following substances: cigarettes/smokeless tobacco, electronic nicotine products, beer, wine, hard liquor, marijuana, synthetic marijuana, psychedelics, prescription drugs (without a prescription or at higher levels than prescribed), steroids (without a prescription), inhalants, heroin, or other hard drugs (i.e., cocaine, crack, crystal meth). Rates of endorsement of synthetic marijuana, psychedelics, prescription drugs, steroids, inhalants, and other hard drugs were low (less than 4%), and frequency and quantity of electronic nicotine use was not assessed. Therefore, the present study focused on cigarette, beer, wine, liquor, and marijuana use because use of these substances is relatively normative in adolescence (Johnston et al., 2019), and rates of endorsement were relatively high in the current sample (i.e., ranging from 16.3% for cigarettes to 29.1% for marijuana).

Positive endorsements of trying a substance were followed up with questions about frequency and quantity of typical use over the past three months. For each substance, frequency was coded on an 8-point Likert scale from Never to 2–3 times per day. For alcohol and tobacco use, quantity was coded on a 7-point Likert scale from less than one drink/cigarette to 6 or more drinks/cigarettes. Marijuana quantity was assessed on an 8-point Likert scale from 1–2 hits to 6 or more bowls joints, or blunts. Individuals who denied using a substance received frequency and quantity scores of 0. To assess the most problematic patterns of use, frequency scores were multiplied by quantity scores.

Demographic characteristics.

Demographic covariates included target youth gender and target youth race, as reported by the primary caregiver at the initial assessment. Additionally, site location and intervention status were accounted for in the mediation models. Finally, household income, averaged across ages 2 to 8.5, was accounted for in all analyses.

Harsh parenting.

Structured interactions between caregivers and their children (e.g., clean-up and teaching tasks) were video-recorded at each assessment between ages 2 and 5. Structured interaction tasks (e.g., clean-up, teaching) were not completed after age 5. A composite index of observed harsh parenting was created by summing three items from the Relationship Affect Coding System (RACS; Peterson et al., 2008) and five items from a coder impressions inventory (see Moilanen et al., 2010). The three items from the RACS captured negative verbal, directive, and physical behavior by the parents, and items from the coder impressions inventory captured age-inappropriate reasons for behavior change, displays of anger, frustration, or annoyance toward the target child, criticism of the target child for family problems, use of physical discipline, actively ignoring or rejecting the target child, and making statements that the target child is worthless. The eight individual items were standardized and summed at each study wave to create a harsh parenting score. Cronbach alphas ranged from .74 to .79 between ages 2 and 5 (observed harsh parenting was not assessed at age 8.5). 20% of video recordings were double coded to establish inter-rater reliability (kappa = .93, indicating adequate consistency between coders). Scores from age 2–5 were averaged to create an overall harsh parenting score.

Primary caregiver substance use.

At the age 2–8.5 assessments, primary caregivers reported their frequency of alcohol and marijuana use on a 9-point Likert-scale from never to two or more times a day. Primary caregivers reported on the number of cigarettes smoked per day. Reports of substance use were averaged across the age 2–8.5 assessments to provide mean primary caregiver alcohol use, marijuana use, and tobacco use scores.

Primary caregiver antisocial behaviors.

At the age 4–7.5 assessments, primary caregivers completed a 17-item abbreviated version of the Elliot Self-Report of Delinquency scale (Elliot et al., 1985). To encompass lifetime antisocial behaviors, questions at the age 4 assessment were adapted to ask how frequently the primary caregiver had performed specific antisocial activities since they were 12. Items at subsequent waves (i.e., ages 5 and 7.5) asked about antisocial behavior over the past year. Items were coded as 0 (never), 1 (once or twice), 2 (2–3 times), 3 (4–10 times), or 4 (more than 10 times). All items began with “Since the age of 12–13, (or In the past year for ages 5 and 7.5) have you ever…” and sample items included, “Purposely damaged or destroyed property that did not belong to you?” or “Been drunk in a public place?”. Items were averaged within waves to create an antisocial behavior score, and scores were averaged across waves to calculate a total lifetime antisocial behavior score. Cronbach alphas for caregiver antisocial behavior scores were .87, .60, and .67 at ages 4, 5, and 7.5, respectively (caregiver antisocial behaviors were not assessed at age 8.5).

Data Analysis

Descriptive statistics were calculated using the base package in R version 4.1.2 (R Core Team, 2020). A multiple mediation model was fit using the lavaan package version 0.6–9 (Rosseel, 2012) to test the primary study hypotheses. A multiple mediation model allowed us to test indirect associations between externalizing behaviors and substance use through violence victimization and witnessing violence simultaneously in a single model. Adolescent substance use was measured as a latent construct encompassing the frequency by quantity score for beer, wine, liquor, tobacco, and marijuana. Direct effects of child externalizing behaviors, witnessing violence and violence victimization on adolescent substance use were assessed. Additionally, two indirect paths were assessed: (1) externalizing behaviors to witnessing violence to substance use and (2) externalizing behaviors to violence victimization to substance use. Confidence intervals for indirect effects were calculated using a Monte Carlo simulation with 20,000 draws (MacKinnon et al., 2004). Because the substance use indicators were highly skewed, frequency by quantity scores were log transformed before model fitting. The multiple mediation model was estimated using maximum likelihood estimation with robust standard errors. Age 8.5 externalizing behaviors, violence witnessing, violence victimization, and substance use were regressed on study covariates.

Gender was explored as a moderator of the a-paths (externalizing behavior to violence witnessing and victimization) and b-paths (violence witnessing to substance use and victimization to substance use) by adding three moderator terms to the mediation model. Specifically, violence victimization and witnessing violence were regressed onto the gender by externalizing behavior interaction term. Additionally, target child substance use was regressed onto witnessing by gender and victimization by gender interaction terms. All moderation terms were included in the model simultaneously along with study covariates. Gender was grand mean centered and externalizing behaviors, violence witnessing, and violence victimization were standardized prior to creating the interaction terms to reduce multicollinearity and improve interpretability.

Missing data.

Based on the longitudinal nature of the study, there were missing data on the independent (13.6%), mediating (18.4%), and dependent variables (8%) in the study sample (i.e., subset of 642 families). Little’s missing completely at random (MCAR) test yielded a statistically significant result (X² = 1,577.8, df = 1,225, p < .001), indicating that missing data did not meet strict MCAR criteria. There is no statistical test to determine if data are missing at random (MAR), which arises when missingness is related to observed variables (e.g., demographic variables), but not unobserved variables (e.g., the value of the missing data point) (Rubin, 1976). A series of logistic regression models revealed that observed covariates significantly predicted missingness on the independent, mediator, and dependent variables. Externalizing behaviors at age 8.5 were more likely to be missing for White than nonwhite youth (B = 0.71, SE = 0.31, p = .020) and were more likely to be missing for youth from Pittsburgh (B = 1.13, SE = 0.37, p = .003) and Virginia (B = 0.87, SE = 0.42, p = .038) than Oregon. Males were more likely than females to have missing data on exposure to violence (B = −0.74, SE = 0.24, p = .002). Greater primary caregiver marijuana use was associated with missingness on target youth beer (0.09, SE = 0.03, p = .011), wine (0.71, SE = 0.03, p = .007), and marijuana use (0.07, SE = 0.03, p = .027). Higher levels of primary caregiver antisocial behavior were associated with a greater likelihood of missingness on target youth liquor use (0.64, SE = 0.22, p = .004), nicotine use (0.57, SE = 0.23, p = .011), and marijuana use (0.46, SE = 0.22, p = .035). Thus, missingness on all independent, mediator, and outcome variables was associated with at least one covariate and data were assumed to be MAR. Missing data were handled using full information maximum likelihood estimation in lavaan, which assumes that data are MAR.

Results

Descriptive Statistics and Intercorrelations

Means and standard deviations of the study variables can be found in Table 1 and intercorrelations between independent, mediator, and outcome variables can be found in Table 2. Exposure to violence was common in the study sample. Over three quarters of the youth (76.5%) endorsed at least one item related to victimization and over half (58.2%) endorsed at least one item related to witnessing violence in the past year. Half of the youth (51.1%) endorsed at least one victimization and witnessing item in the past year. For context, national prevalence rates of past year exposure to violence for children ages 6–9 are 47.6% for victimization and 15.1% for witnessing (Finkelhor et al., 2015). There were no significant differences by gender in violence victimization or witnessing scores.

Table 1.

Descriptive Statistics of Study Variables

Descriptive Statistics Continuous
	Full Sample (n = 642)			Males (n = 326)	Females (n = 316)
Variable Name	Mean (SD)	Range	Number Missing (%)	Mean (SD)	Mean (SD)	t-test
Externalizing Behaviors (Age 8)	10.91 (9.35)	0–50	87 (13.6%)	11.84 (9.79)	9.91 (8.77)	t = 2.44df=552*
Violence Victimization (Age 9)	3.02 (3.41)	0–23	118 (18.4%)	3.14 (3.62)	2.90 (3.21)	t = 0.81df=498ns
Violence Witnessing (Age 9)	2.53 (3.57)	0–22	118 (18.4%)	2.54 (3.87)	2.52 (3.27)	t = 0.07df=488ns
Beer Frequency*Quantity (Age 16)	0.59 (2.68)	0–25	141 (22.0%)	0.66 (2.79)	0.51 (2.56)	t = 0.06df=498ns
Wine Frequency*Quantity (Age 16)	0.22 (0.85)	0–8	151 (23.5%)	0.14 (0.49)	0.30 (1.10)	t = 2.00df=330*
Liquor Frequency*Quantity (Age 16)	0.63 (2.52)	0–24	141 (22.0%)	0.48 (1.80)	0.78 (3.06)	t = 1.38df=409ns
Tobacco Frequency*Quantity (Age 16)	0.85 (4.04)	0–35	116 (18.1%)	0.84 (3.93)	0.85 (4.16)	t = 0.05df=523ns
Marijuana Frequency*Quantity (Age 16)	2.91 (9.39)	0–56	140 (21.8%)	3.03 (9.77)	2.80 (9.02)	t = 0.28df=495ns

Note. For t-tests and X²-tests, ^ns refers to a non-significant difference,

^*p < .05.

Table 2.

Intercorrelations between Independent, Mediator, and Dependent Variables

Variable	1. Externalizing	2. Victimization	3. Witnessing	4. Beer	5. Wine	6. Liquor	7. Tobacco	8. Marijuana
1. Externalizing Behaviors	1
2. Violence Victimization	.132*	1
3. Violence Witnessing	.179*	.502*	1
4. Beer Frequency*Quantity	.031	.117*	−.014	1
5. Wine Frequency*Quantity	.056	.074	.060	.615*	1
6. Liquor Frequency*Quantity	.021	.081	.011	.794*	.685*	1
7. Tobacco Frequency*Quantity	.051	.218*	.066	.603*	.392*	.598*	1
8. Marijuana Frequency*Quantity	.102*	.172*	.115*	.623*	.541*	.676*	.586*	1

Note.

^*p < .05

Violence victimization and witnessing scores were moderately correlated, providing support for assessing victimization and witnessing as separate constructs that capture unique experiences in a child’s environment. Externalizing behaviors were significantly correlated with violence victimization and witnessing. Violence victimization was significantly correlated with beer, tobacco, and marijuana use, whereas witnessing violence was only significantly correlated with marijuana use. The substance use variables were highly intercorrelated. Intercorrelations between age 8.5 externalizing behaviors, exposure to violence, and all study covariates are presented in Supplementary Table 1. Study covariates were weakly to moderately intercorrelated (r’s = −.19 to .31), suggesting that multicollinearity between study covariates is not a major concern.

Mediation Model

A measurement model was fit to examine the loadings of beer, wine, liquor, tobacco, and marijuana use on a substance use latent variable. The measurement model fit the data well (X² = 11.42_df=5, p = .044; RMSEA = .05; CFI = 0.98, TLI = 0.96). Standardized factor loadings of the substance use indicators ranged from 0.70 to 0.90 (p-values < .001). The substance use factor was then related to age 8.5 externalizing behaviors. Contrary to expectations and previous research, parent-reported externalizing behaviors were not significantly associated with substance use in adolescence (r = .04, p = .352). Despite a non-significant association between externalizing behaviors and substance use, indirect effects were tested in line with Zhao and colleagues’ (2010) indirect-only mediation where significant indirect effects can exist outside the context of a significant direct effect.

The mediation model without covariates had an acceptable fit to the data (X² = 36.95_df=17, p = .003; RMSEA = .04; CFI = 0.97; TLI = 0.95). As expected, externalizing behaviors at age 8.5 were significantly related to violence victimization (B = 0.05, SE = 0.02, p = .007) and witnessing violence (B = 0.07, SE = 0.02, p < .001) at age 9.5. Additionally, in line with hypotheses, violence victimization was positively associated with substance use in adolescence (B = 0.13, SE = 0.05, p = .006), but witnessing violence was not associated with adolescent substance use (B = −0.05, SE = 0.03, p = .106). The indirect path from age 8.5 externalizing to age 9.5 violence victimization to adolescent substance use was significant (B = 0.006, 95% confidence interval = 0.001, 0.013).

The full mediation model including covariates fit the data acceptably (X² = 165.49_df=85, p < .001; RMSEA = .04; CFI = 0.96; TLI = 0.91; see Figure 1). As expected, externalizing behaviors at age 8.5 were positively associated with violence victimization (B = 0.04, SE = 0.02, p = .019) and violence witnessing (B = 0.06, SE = 0.02, p = .003) at age 9.5. Additionally, in line with hypotheses, violence victimization was positively associated with adolescent substance use (B = 0.10, SE = 0.04, p = .014); however, witnessing violence was not significantly associated with substance use (B = −0.04, SE = 0.03, p = .280). The indirect path from age 8.5 externalizing to age 9.5 violence victimization to adolescent substance use was significant (B = 0.004, 95% confidence interval = 0.0002, 0.0096). The indirect path from externalizing to substance use through violence witnessing was non-significant (unstandardized 95% confidence interval = −0.007, 0.002). The pattern of mediation observed is congruent with Zhao et al. (2010) indirect-only mediation, which suggests that indirect effects (i.e., the a*b path) can exist outside the context of significant direct effects (i.e., the c path). See Supplementary Table 2 for all coefficients from the mediation model.

Figure 1.

Indirect effects of externalizing behaviors on substance use through exposure to violence. Solid lines represent significant paths and dashed lines represent non-significant paths. For clarity, paths from individual covariates are not included in the figure. Beta values presented are unstandardized.

Moderation by Gender

Gender did not moderate the path from externalizing behaviors to violence victimization (B = −0.06, SE = 0.10, p = .547) or witnessing violence (B = −0.04, SE = 0.10, p = .666). The association between externalizing behaviors and violence victimization (B = 0.11, SE = 0.05, p = .020) and witnessing (B = 0.15, SE = 0.05, p = .003) was significant. However, the association between gender and violence victimization (B = −0.02, SE = 0.09, p = .805) and witnessing (B = −0.01, SE = 0.08, p = .919) was nonsignificant. Additionally, gender did not moderate the association between violence victimization and substance use (B = −0.01, SE = 0.24, p = .968) or witnessing violence and substance use (B = −0.07, SE = 0.20, p = .705). The association between violence victimization and substance use was significant (B = 0.33, SE = 0.14, p = .015), whereas the association between witnessing violence and substance use was nonsignificant (B = −0.12, SE = 0.12, p = .290). Additionally, the association between gender and substance use was significant (B = 0.34, SE = 0.17, p = .046), indicating that males were more likely to use substances than females.

Sensitivity Analyses: Exposure to Violence to Substance Use through Externalizing Behavior

As some previous research suggests that exposure to violence may increase externalizing behaviors (Mrug et al., 2016), sensitivity analyses were conducted to explore the directionality of the present study’s findings. Specifically, a mediation model was fit to test whether victimization or witnessing violence at age 9.5 was associated with substance use at age 16 mediated by externalizing behaviors at age 10.5. As externalizing behaviors were assessed at multiple timepoints, externalizing behaviors at age 9.5 were controlled for in the model. The alternative mediation model fit the data well (X² = 202.70_df=94, p < .001; RMSEA = .04; CFI = 0.95; TLI = 0.91). Externalizing behaviors at age 9.5 were strongly associated with externalizing behaviors at age 10.5 (B = 0.81, SE = 0.04, p < .001). Neither violence victimization (B = 0.13, SE = 0.13, p = .342) nor witnessing violence (B = 0.01, SE = 0.11, p = .940) were significantly associated with externalizing behaviors at age 10.5. Moreover, externalizing behaviors at age 10.5 were not significantly associated with substance use at age 16 (B = 0.02, SE = 0.01, p = .089). The indirect association between victimization and substance use through externalizing behaviors was nonsignificant (B = 0.002, 95% confidence interval = −0.003, 0.011) as was the indirect association between witnessing violence and substance use through externalizing behaviors (B = 0.000, 95% confidence interval = −0.005, 0.005). The full results from the alternate mediation model can be found in Supplementary Table 3.

Post-Hoc Analysis: Predicting Marijuana Use

As violence witnessing was most strongly correlated with marijuana use, but quite weakly correlated with other substances, an additional multiple mediation model was fit post hoc to examine endorsement of solely marijuana use as the outcome. Consistent with the results of the initial mediation model, the path through witnessing violence was not significant (B = 0.006, 95% confidence interval = −0.008, 0.010).

Post Hoc Analysis: Including Time-Varying Covariates

To account for family factors that may be variable over time in a more complex model, a mediation model was fit with family income and primary caregiver substance use included as time varying covariates at ages 8.5, 9.5, and 16 (observational measures of harsh parenting were not collected at all study timepoints). When the model was made more complex (e.g., by including time-varying covariates), the magnitude of the associations remains the same, but the indirect effect is no longer significant. Specifically, consistent with results from the mediation model with time-invariant covariates, externalizing behaviors at age 8.5 were significantly associated with violence victimization and witnessing violence at age 9.5. Additionally, violence victimization was significantly associated with youth substance use at age 16. However, the indirect association between externalizing behaviors and substance use through violence victimization was not statistically significant (B = 0.004, 95% confidence interval = −0.000, 0.009). Including sociodemographic covariates as time-varying accounts for marginally more variance in the independent, mediator, and outcome variables, reducing the statistical significance of the indirect association between externalizing behaviors and substance use through violence victimization. Importantly, including income and caregiver substance use as time-varying covariates does not attenuate the associations between the constructs in the model. All coefficients from the time-varying covariate mediation model can be found in Supplementary Table 4.

Discussion

There is a well-established body of evidence demonstrating associations between exposure to violence and maladaptive psychosocial development (Shin et al., 2020), and an expanding body of research on individual characteristics that heighten the risk for exposure to violence (Buka et al., 2001), yet gaps in the literature persist. Specifically, two forms of exposure to violence (i.e., victimization and witnessing) have rarely been included as separate constructs in the same model, which may obscure associations between predictor variables of interest and unique experiences of violence. Further, previous work has seldom included additional variables that may also increase risk for exposure to violence (e.g., externalizing behaviors) while also accounting for child, family, and contextual factors associated with both externalizing problem behavior and adolescent substance use. Additionally, little attention in the previous literature has been devoted to experiences during the transition to adolescence, which can set the stage for adaptive or maladaptive adolescent development (Olson et al., 2021). The current study addressed these gaps in the literature by examining associations between externalizing problem behaviors during middle childhood, violence victimization and witnessing in later middle childhood, and adolescent substance use using data from a large, prospective, longitudinal study of economically marginalized families from diverse racial and ethnic backgrounds, a population at risk to experience violence.

The current study adds to the growing literature on individual risk factors for exposure to violence by examining late childhood violence victimization and witnessing as mediators of the association between middle-childhood externalizing behaviors and adolescent substance use. An indirect path between externalizing behaviors and substance use was observed through violence victimization, but not witnessing. While higher rates of violence victimization and witnessing were observed for males, gender did not moderate the indirect associations between externalizing behaviors and substance use through violence victimization or witnessing. Moreover, gender did not moderate any of the paths between externalizing behaviors and exposure to violence or exposure to violence and substance use.

Consistent with prior work (Salzinger et al., 2006), the present study’s findings indicate that higher levels of externalizing behaviors at age 8.5 are associated with higher levels of both victimization and witnessing violence at age 9.5. Children with a more impulsive or aggressive temperament may elicit aggressive reactions from their peers (Verlinden et al., 2014) or parents and siblings (Williams et al., 2007). However, it is also possible that children exposed to violence may develop more aggressive response styles in an effort to protect themselves from acts of violence or psychological aggression against them, presumably experienced beginning in early childhood (Schraft et al., 2013). As exposure to violence was measured at only one wave of the study, the possibility cannot be ruled out that exposure to violence in the home, neighborhood, or school contributed to associations with school-age externalizing behaviors. To attempt to account for some of these environmental contributions, harsh parenting, parent antisocial behavior, and parent substance use were included as covariates.

Further, in the present study, only direct exposure to violence was associated with adolescent substance use. The pattern of findings is consistent with some previous research (e.g., Taylor & Kliewer, 2006), but differs from studies that have found both victimization and witnessing violence to be associated with substance use when included as separate constructs in the same model (Pinchevsky et al., 2013). The most commonly endorsed exposures to violence were threats, being picked on, and being hit, slapped, or punched, with extremely low base rates of exposure to more extreme violence (i.e., have been/seen someone get shot at). Therefore, exposure to violence in late childhood may reflect more of a history of bullying at school and in the neighborhood than exposure to more severe violence. Bullying victimization has been found to be a stronger predictor of later internalizing problems than witnessing bullying (Werth et al., 2015). Thus victimization, but not witnessing, may contribute sufficiently to heightened psychological distress that individuals seek out substances to alleviate (Janosz et al., 2008). Alternatively, children who exhibit higher levels of reactive aggression and are rejected by their peers and/or are victims of bullying in late childhood are at greater risk of developing substance use through affiliating with more deviant peers (Fite et al., 2007). Based on the nuanced findings in relation to exposure type (i.e., witnessed or victimized), it is important for future researchers to continue to disentangle the influences of unique experiences of violence to ensure associations between predictor variables and types of exposure to violence are not obscured.

Direct exposure to violence was most strongly associated with marijuana use, which is consistent with findings from a sample of 1,655 adolescents (ages 11–22) living in an urban area where direct exposure to community violence and child abuse were significant predictors of frequency of marijuana use but not alcohol use (Wright et al., 2013). In prior work, it has been speculated that marijuana use may serve as more of an “escape” than alcohol use. Further research is necessary to elucidate the associations between direct exposure to violence and the use of specific substances. Qualitative research may be particularly well suited for investigating such nuances and participants’ self-reported reasons behind them that quantitative measures typically fall short in examining.

The indirect association between externalizing behaviors at age 8.5 and substance use at age 16 through violence victimization needs to be interpreted based on the null association between externalizing problems and substance use, which stands in contrast to previous work (Colder et al., 2013). Externalizing behaviors assessed at age 8.5 generally reflect more overt aggressive and oppositional behaviors relative to externalizing behaviors assessed at later developmental stages, which reflect more covert and often more serious antisocial behaviors (Olson et al., 2013). Within a sample recruited to be at elevated risk for externalizing pathology and substance use (Dishion et al., 2008), variability in externalizing problems in middle-childhood may explain relatively little variance in substance use outcomes in mid-adolescence, when substance use is strongly linked to delinquency. However, children that are more aggressive or oppositional may find themselves in risky or violent environments, or alternatively, they may elicit violent responses from their environment (Salmivalli & Nieminen, 2002). Findings from the present study provide insight into a potential path by which externalizing problems in middle childhood may indirectly lead to substance use in adolescence.

Contrary to study hypotheses and prior work (Buka et al., 2001), there were no mean differences in levels of victimization and witnessing violence for girls and boys. The absence of mean differences in exposure to violence between girls and boys may reflect the risk status of girls in the Early Steps study relative to community samples. Specifically, youth were recruited to be at risk for conduct problems and later substance use, and recruitment criteria included child conduct problems, maternal depression, maternal substance use, family conflict, and low family income (Dishion et al., 2008). Such factors may sufficiently increase the risk for exposure to violence in both males and females that factors contributing to gender differences in community samples are subsumed. Additionally, as 45.9% of the female youth scored greater than one standard deviation above the population mean on externalizing behaviors at the baseline assessment, females in the present sample may have been at a higher likelihood to elicit more aggressive responses from parents, siblings, or peers than females in community samples. Thus, the individual and environmental risk of youth in the present study may have contributed to similar patterns of exposure to violence for males and females.

Adding to previous research that has found no gender differences in substance use following exposure to violence (James et al., 2018), in the current study, gender did not moderate the association between externalizing behaviors, exposure to violence and adolescent substance use. As the present sample was recruited to be at risk for later substance use, females in the study sample may have been more likely to respond to adverse environmental factors by using substances than females in community samples. Alternatively, exposure to violence may contribute equally to substance use for males and females as evinced by previous work which found moderating effects of gender on the association between exposure to violence on some substance use outcomes (e.g., binge drinking), but not others (e.g., marijuana use) (Pinchevsky et al., 2013). The extant body of research on psychosocial development after exposure to violence is relatively small, and it is recommended that future studies continue to examine moderators of the association between exposure to violence and psychosocial outcomes. For example, future research should examine the influence of peer and family relationships as a possible moderator in such associations.

Strengths and Limitations

The current study makes a unique contribution to the literature by connecting bodies of research that have found associations between behavior dysregulation and exposure to violence (Verlinden et al., 2014), and exposure to violence and later substance use (James et al., 2018). The results from the present study suggest a unique path from externalizing behaviors in middle childhood to adolescent substance use through violence victimization, but not witnessing violence in late childhood. A particular strength of the present study was the ability to account for several factors that may influence child development and/or their likelihood of experiencing violence in their environments. Importantly, the associations between externalizing behaviors, victimization and witnessing violence, and substance use remained after accounting for primary caregiver substance use and antisocial behavior, as well as observations of harsh parenting. While accounting for parent’s self-reported behaviors and substance use does not completely rule out the possibility that the observed results are driven by a passive gene-environment correlation, it does increase confidence in the observed associations between youth behaviors and environmental stressors.

Despite the strengths of the present study, there are a few limitations that should be acknowledged. First, exposure to violence was assessed at one timepoint, which limited the ability to examine reciprocal relationships between externalizing behaviors, exposure to violence, and substance use across multiple time periods. It is important to note that there may be a bidirectional relationship between exposure to violence and externalizing behaviors longitudinally (Farrell et al., 2014). The present study made assumptions about the temporal relationships between externalizing behaviors, exposure to violence, and substance use. As externalizing behaviors and exposure to violence were measured at ages 8.5 and 9.5 respectively, it is unlikely that youth were using substances prior to the measurement of exposure to violence. However, it is possible that the temporal relationship between exposure to violence is reversed and exposure to violence is related to increases in externalizing behavior (Weaver et al., 2008). Of note, sensitivity analyses to explore the directionality of the present study’s findings did not reveal a significant association between higher levels of exposure to violence at age 9.5 and elevated externalizing behaviors at age 10.5. The pattern of findings provides support for the need to conduct additional research exploring individual risk factors for exposure to violence to inform prevention efforts. It may be that children with more aggressive behaviors that find themselves in more violent environments may be more often utilizing reactionary aggression to protect themselves from such violence, leading to greater exposure to violent behavior. Future research would benefit from more nuanced measures of externalizing problem behaviors or observed measures to ascertain whether aggressive behaviors are reactionary in nature to violent environments.

Additionally, the measure of exposure to violence did not thoroughly assess violence in home and family context, instead only assessing violence from siblings. The assessment of harsh parenting was another attempt to account for violence in the home. However, these strategies did not fully capture measurement of violence in the home or family context. Future research would benefit from a more thorough measurement of exposure to violence in the home in addition to experiences at school and in the neighborhood. Additionally, future research should assess both exposure to violence (including violence in the home) and externalizing behaviors at multiple ages, which would permit greater understanding of the temporal and potential transactional associations between child behavior and exposure to violence, including a more refined estimation of indirect effects (Cole & Maxwell, 2003).

A second limitation is that the study sample is comprised of target children at greater risk for externalizing behaviors, exposure to violence, and substance use than a more nationally representative sample (Dishion et al., 2008); thus, generalizability is somewhat limited. However, given the higher likelihood for youth in this sample to experience externalizing behaviors, exposure to violence and substance use, it is a particularly well-suited analytic sample to investigate such associations. Further, the strength of such associations and pattern of findings would likely not differ even in samples where risk for such experiences is relatively low. Additionally, the analytic subsample is drawn from the Early Steps Multisite study in which a randomized controlled trial of the Family Check-Up was implemented, and recent work suggests that this sample’s response to resources is congruous with findings from other family-based interventions (Dishion et al., 2014). Nevertheless, understanding the individual-environment interactions among the highest-risk populations is important in informing interventions and decisions to disperse community resources.

Implications

The findings of the present study are important for clinicians, researchers, and social policy makers. Specifically, the findings highlight the need for clinicians to be mindful of and account for possible exposure to violence in their case conceptualizations of children with conduct problems, especially when considering risk factors for the sensitive developmental period of the transition to adolescence. Further, understanding the pathways by which childhood risk factors are related to exposure to violence and later substance use in present and future research is necessary to inform effective, well-timed prevention and intervention techniques, especially longitudinal work focusing on the transition to adolescence (Haller & Chassin, 2014). More work is critical as recent research has highlighted the profound impact of exposure to violence in childhood and adolescence on long-term adult outcomes (Oberth et al., 2021). Interventions aimed at reducing early externalizing behaviors may reduce the risk for violence victimization, which may in turn reduce the risk for adolescent substance use. Several family-based interventions have been shown to reduce children’s externalizing behavior through increases in positive parenting techniques (e.g., monitoring, limit setting, and positive behavior support; Antshel & Barkley, 2008). Prevention and intervention efforts are especially critical for economically marginalized communities, who are at greater risk of experiencing violence in adolescence (McCrea et al., 2019). Public policy and community work designed to dismantle systems of inequity that create disproportionate risk for exposure to violence for marginalized children and youth should be pursued first and foremost; however, research that highlights the role of childhood exposure to violence in risk for adolescent substance use is critical for calling attention to this necessary work.

Conclusion

Prospective, longitudinal research is needed to understand individual risk factors in childhood associated with exposure to violence and subsequent adolescent substance use to inform prevention efforts. Rarely have victimization and witnessing been included as separate constructs in the same model that also assesses variables that may increase risk for exposure to violence (e.g., externalizing behaviors), as well as unique contributions of victimization or witnessing violence to risk for later substance use. Overall, the findings from the present study contribute to the growing literature on individual pathways to risk for exposure to violence in late childhood and later substance use in adolescence. Elevated externalizing problem behaviors at age 8.5 were related to higher levels of victimization and witnessing violence at age 9.5 and higher levels of violence victimization was related to greater adolescent substance use. Research to understand the nuances of what might reduce risk for exposure to violence (i.e., through investigating pathways to exposure) and what might buffer the negative effects of exposure to violence on child behavior (i.e., through developing and testing the effectiveness of interventions) is critical to the science of fostering healthy child development.

Biography

Shannon Savell is a graduate student of clinical psychology at the University of Virginia. Her research interests include investigating how positive family processes can ameliorate the negative impact of a variety of adverse events for children.

Sean Womack is a graduate student of clinical psychology at the University of Virginia. His research interests include disentangling genetic and environmental contributions to behavioral, emotional, and cognitive development and longitudinal data analysis.

Melvin Wilson is a professor of Psychology at the University of Virginia. His research interests include understanding contextual processes and outcomes and conducting parental interventions in low-income, ethnic minority families.

Kathryn Lemery-Chalfant is a professor of Psychology at Arizona State University. Her research focuses on risk and resilience processes that impact children’s mental and physical health by using genetically-informative study designs, such as twin studies and studies that include genotyping.

Daniel Shaw is a professor of Psychology at the University of Pittsburg. His research interests include development and prevention of early child conduct and emotional problems, family-centered interventions for treating conduct problems in early childhood and adolescence, use of novel community platforms for implementing preventative interventions in early childhood, identification of gene x environment interactions in relation to brain function and child psychopathology.