Delinquency, Substance Use, and Risky Sexual Behaviors among Youth who are involved in the Justice System and Predominantly Reside in Rural Communities: Patterns and Associated Risk Factors

Abstract

There is a significant gap in research examining the prevalence of problem behaviors among youth involved in the juvenile justice system in rural areas. The current study sought to address this gap by exploring the behavioral patterns of 210 youth who were on juvenile probation in predominantly rural counties and who were identified as having a substance use disorder. First, we examined the correlation among 7 problem behaviors representing different forms of substance use, delinquency, and sexual risk-taking and 8 risk factors related to recent service utilization, internalizing and externalizing difficulties, and social support networks. Then, we used latent class analysis (LCA) to identify distinct behavioral profiles based on the observed problem behaviors. LCA identified a 3-class model representing distinct groups labeled Experimenting (70%), Polysubstance Use + Delinquent Behaviors (24%), and Diverse Delinquent Behaviors (6%). Finally, we assessed differences (i.e., ANOVA, χ²) in each risk factor across the behavioral profiles. Important similarities and differences in the association among the problem behaviors, behavioral profiles, and the risk factors were revealed. These findings underscore the need for an interconnected behavioral health model within rural juvenile justice systems that is able to address youths’ multidimensional needs including criminogenic, behavioral, and physical health needs.

Adolescents with a substance use disorder (SUD) are a high-risk population. Youth who meet criteria for a SUD are at a higher risk for co-occurring mental health problems (Schubert et al. 2011; McCance-Katz 2019), engaging in delinquency and other risk-taking behaviors (Baumer et al. 2018), and poor physical health (Schulte and Hser 2013; Winkelman et al. 2017). Even more alarming are the consistent findings that only a small portion of adolescents with a SUD receive treatment, and even fewer participate in evidence-based treatment programs (White et al. 2019). Results from the 2019 National Survey on Drug Use and Health (NSDUH) showed that 4.6% of individuals aged 12–17, or approximately 1.1 million adolescents, needed substance use treatment; however, only 0.7% received substance use treatment (SAMHSA 2020).

Research also provides substantial evidence that youth with a substance use problem are not a homogenous population (Choi et al. 2018; Staras et al. 2011). Significant variability in the types and frequency of substances used has been observed among community-based samples of adolescents (Banks et al. 2017; Su et al. 2018), adolescents with substance use problems (Fallu et al. 2014), and adolescents involved in the juvenile justice (JJ) system (McCuish 2017). Furthermore, adolescents who use substances (e.g., marijuana, alcohol) are more likely to be involved in other problem behaviors that amplify their complex intervention needs and the potential for long-term negative consequences (Childs et al. 2016; Mun et al. 2008). For example, alcohol, tobacco, and marijuana use have been found to increase the likelihood of engaging in risky sexual behavior (Poulin and Graham 2001; Stueve and O’Donnell 2005). Adverse sexual consequences such as unplanned intercourse, multiple partners, and inconsistent condom use are more likely to occur among youth who use substances (e.g., alcohol use, Bonomo et al. 2001; alcohol or other drugs, Poulin and Graham 2001).

Several studies have also found that youth who engage in delinquent behavior report risky sexual practices at a substantially higher rate than youth who do not engage in delinquent behavior (Huizinga and Jakob-Chien 1998; Tolou-Shams et al. 2019), with some studies reporting as high as 4 times greater odds of risky sexual practices among delinquent youth (Belenko et al. 2008; 2009). In an effort to understand the interplay of these risky behaviors in adolescence, several studies have applied mixture modeling techniques (e.g., latent class analysis) to identify distinct patterns of co-occurring problem behaviors. Generally, 3 to 5 groups have been extracted, describing distinct behavioral profiles that range in both the types of behaviors engaged in and the frequency of engagement (Bartlett et al. 2005; Childs 2014; Sullivan et al. 2010).

There is no question that youth with substance use problems are overrepresented in the JJ system (Belenko et al. 2017; Teplin et al. 2002). Using a representative sample of youth, Wasserman et al. (2010) reported that 34% of over 10,000 youth involved in the JJ system across 18 states had a SUD. As these data suggest, the JJ system could represent an efficient service delivery pathway for youth with co-occurring behavioral health problems (e.g., substance use, delinquent behavior, sexual risk-taking) who might otherwise not have access to needed services. The critical role of the JJ system in connecting youth to appropriate intervention services has become more apparent in recent years due to the movement towards a data-driven, decision-making process grounded in the Risk, Needs, Responsivity (RNR) framework. This framework rests on the assumptions that youths’ needs are multi-dimensional, rehabilitation is most likely when a holistic approach targeting the unique array of risk and protective factors presented by each youth is applied, and appropriate services are available and accessible (Andrews et al. 1990). Thus, the reliance on this framework to make intervention decisions offers an opportunity to treat co-occurring problem behaviors (i.e., instead of focusing only on delinquency) and the risk and protective factors related to these different behavioral patterns. Juvenile probation officers (JPOs) are often critical gatekeepers to accessible, community-based intervention services as they are responsible for screening and referring youth to local community-based service providers (Belenko et al. 2017; Sheidow et al. 2020), representing an important opportunity to broaden the scope of services available to youth.

Behavioral Patterns and the Associated Risk Factors among Youth in Rural Communities

To date, very few studies have examined the prevalence of co-occurring substance use, delinquency, and sexual risk-taking among youth living in rural areas. When studying large samples of youth on probation or who have a SUD, any behavioral patterns unique to rural areas likely get lost due to the overrepresentation of youth located in urban or suburban locations. Research on whether specific forms of problem behavior vary across urban and rural locations provides mixed conclusions. While some studies do not find meaningful differences in behavior across urban and rural youth (Staton et al. 2020; Thompson et al. 2018), a substantial number of studies found higher rates of drug use (e.g., alcohol and cocaine, Lambert et al. 2008; tobacco, alcohol, marijuana, cocaine, and heroin, National Center on Addiction and Substance Use 2000), delinquency (Farrell et al. 2005; Wells and Weishelt 2004), juvenile justice contact (Blackmon et al., 2016), and sexual risk-taking (Gale et al. 2019; Milhausen et al. 2003) among adolescents from rural communities. Atav and Spencer (2002) found that rural youth were significantly more likely to use tobacco, alcohol, and other drugs compared to suburban and urban youth, but did not find any significant differences in sexual behavior or carrying a weapon. Studies have also identified differences in some specific risk factors (e.g., peers, family environment, academic achievement) and levels of cumulative risk for these behaviors across rural and urban youth (Connolly et al. 2017; Dembo et al. 2020; Nelson et al. 2010; Thompson et al. 2017).

In one of the only studies to examine behavioral profiles among justice-involved youth residing in rural community, Krupa et al. (2020) compared latent class models for urban and rural girls based on several problem behaviors including sexual risk-taking, delinquency, mental health, victimization, and prior criminal history. A 2-class solution, representing a “low risk” group and a “high risk” group was extracted from both urban and rural samples. However, urban girls, compared to their rural counterparts, showed higher rates of mental health and victimization experiences. Also, a greater proportion of rural girls fell into the “high risk” class (22.1% of rural girls versus 2.5% of urban girls). As these findings illustrate, the potential for differences in the manifestation of problem behaviors across youth residing in urban and rural communities highlights the need for additional research that directly focuses on the behavioral profiles of youth who reside in rural areas.

Furthermore, there is a significant gap in research that examines the intervention needs of youth who are involved in the JJ system, have a SUD, and reside in rural communities. This lack of research is troubling because a large body of evidence documents substantial barriers to accessible, evidence-based services for youth with behavioral health problems in rural communities (Fehr et al. 2020; Jensen et al. 2021). Individuals with a SUD who live in a rural community have been described as “disproportionately disadvantaged” due to several factors that are directly related to the availability and quality of intervention services (Pullen and Oser 2014). These include a lack of qualified frontline professionals (Andrilla et al. 2018; Edmond et al. 2015), geographic discordance (Oser and Harp 2015), underutilization of available services (Pringle et al. 2006), geographic dispersion and fewer transportation options (Sung et al. 2011), and a lack of available specialized, evidence-based services (Oser et al. 2011). As such, the role of the JJ system in targeting youths’ multidimensional needs becomes more critical in rural communities (Sheidow et al. 2020).

This study explores the behavioral profiles of youth who are on probation in predominantly rural counties and who have been identified as having a substance use problem. Our goal is to assess the diversity in self-reported engagement in delinquency, substance use, and sexual risk-taking and to compare the recent service utilization, internalizing and externalizing difficulties, and social support risk factors across these patterns of behavior. We specifically focus on these risk factors for three reasons. First, these risk factors represent areas where there is evidence to suggest a potential difference between rural and urban youth on probation, such as prior service utilization and social support networks (Smokowski et al. 2014; Spoth et al. 2001). Second, these risk factors encompass many of the risk factors that are considered during post-adjudication intervention planning. Third, research indicates that risk factors related to youths’ social support networks (i.e., peers, family) and their internalizing and externalizing difficulties vary across subgroups of youth based on engagement in risk-taking behavior (Childs 2014; Modecki 2016).

Our exploratory analyses will add to the scant research that directly assesses the intervention needs of youth who are involved in the JJ system, have been identified as having a substance use problem, and reside in rural communities. Although this sample is unique, there is a paucity of research devoted to understanding the complex and multidimensional needs of this distinctive population. Based on the clear evidence that providing effective intervention to youth in rural areas is challenging (Click et al. 2018) and that youth in rural areas who use drugs and alcohol represent a high-risk population (Monnat and Rigg 2018; Sickmund and Puzzanchera 2014), we seek to add to existing research on the behavioral profiles and intervention needs of youth in rural communities, while considering the critical role that the JJ system can play in ensuring the behavioral and physical health of this population.

Study Objectives

1. Assess the similarities and differences in the relationship between prior service-utilization, family instability, social support, and youth difficulties and a variety of problem behaviors (delinquency, substance use, and risky sexual behavior).

2. Identify patterns or “classes” of youth based on engagement in delinquency, substance use, and risky sexual behavior.

3. Examine whether prior service-utilization, family instability, social support, and youth difficulties vary across groups or “classes” of youth that are based on engagement in delinquency, substance use, and risky sexual behavior.

Materials and Methods

Study Description

The current data are baseline data from an ongoing randomized controlled trial designed to assess the effectiveness of contingency management (CM) delivered by JPOs compared to probation as usual (PAU) delivered by JPOs. CM and PAU are described in other reports on this trial (see Sheidow et al. 2020). Data for the current study were collected prior to youth exposure to CM or PAU. Specifically, the current study utilizes data collected from youth under probation supervision and their parents to better understand patterns of youth engagement in delinquency, substance use, and risky sexual behavior and the associated service utilization, family instability, and behavioral risk factors for these diverse forms of problem behaviors.

Study Sample

The project received referrals for youth under supervision in 13 counties across three states (i.e., Idaho, Oregon, and Nevada). Based on available data from the 2019 American Community Survey (US Census Bureau 2019), the percent of the population under the age of 18 in these counties ranged from 21% to 32% (mean = 26%, SD = 4%). County residents were predominantly White (percent White ranged from 72% to 92%, mean = 80%) and non-Hispanic (percent Hispanic ranged from 8% to 37%, mean = 23%). Furthermore, the percent of the population living below the poverty line ranged from 14% to 30% (mean = 20%, SD = 5%) and the median household income ranged from $47, 204 to $67,043 (mean = $55,393, SD = $5,574). Compared to national estimates, on average, these counties have a lower median household income, lower percentages of Black residents, higher percentages of Hispanic residents, and more individuals living below the poverty line.

Youth were eligible for consideration to participate if they met the following criteria: 1) age 12–18 with a parent or caregiver, 2) recent drug or alcohol use reported, and 3) at least 4 months left on probation supervision. Youth charged with a sexual offense or currently participating in a drug court were excluded from the study to prevent overlapping interventions. Youth who met the eligibility criteria were referred to an assessor from the research center. The research assessor administered the Mini International Neuropsychiatric Interview (MINI Kid; Sheehan et al. 2010) to the youth and separately to the parent to screen for drug and alcohol use disorders. If the youth met criteria for at least one type of SUD based on the results of the youth or parent MINI Kid, they were eligible for the trial. A total of 335 youth were referred for eligibility screening, of which 61 declined screening, 17 were ineligible based on aforementioned criteria, 16 were unable to be contacted for scheduling, 15 declined participation, and 11 were closed on probation before screening. The remaining 215 youth were randomized into the study, however, the current study reports on only those 210 youth who participated in all baseline assessments.² The trial is ongoing, so the current study includes all families recruited as of March 2021.

Characteristics of this sample are presented in Table 1. Sixty-seven percent of the youth in the sample were on probation in a zip code defined as rural. This designation is based on the definitions set forth by the Federal Office of Rural Health Policy (i.e., includes non-metro areas, see FORHP, 2020). The sample was 44% Hispanic, 91% White, 67% male, and 89% heterosexual, with an average age of 15 years (SD = 1.45). Approximately 46% of families received public assistance at the time of the interview.

Table 1.

Description of the Sample (N = 210)

	N (%) / Mean (SD, Range)
Sample Characteristics:
Age	15.38 (1.45, 11–18)
Sex
Male	141 (67%)
Female	69 (33%)
Race
White	191 (91%)
Nonwhite	19 (9%)
Ethnicity*
Non-Hispanic	118 (56%)
Hispanic	91 (44%)
Orientation
Heterosexual	186 (89%)
Non-Heterosexual	24 (11%)
Public Assistancec
No	114 (54%)
Yes	96 (46%)
Independent variables:
Externalizing behaviora	10.01 (2.58, 6–17)
Internalizing behaviora	11.31 (3.44, 7–21)
Living Risk Index (LRI)a	5.41 (1.82, 6–21)
Social Risk Index (SRI)a,*	7.72 (3.52, 6–30)
Substance Problem Scale (SPS)a	3.55 (4.48, 0–17)
Primary caregiversc	1.94 (1.05, 1–8)
Outpatient services for MH/SAb	0.45 (0.50, 0–1)
Inpatient services for MH/SAb	0.08 (0.27, 0–1)
Dependent variables:
Public disorderb	0.58 (0.82, 0–3)
Property/theftb	0.93 (1.43, 0–8)
Against-personb	0.12 (0.32, 0–1)
Five or more drinks (drunk)a	0.94 (3.11, 0–28)
Marijuanaa	4.73 (8.08, 0–30)
Other drugsa	0.21 (0.41, 0–1)
Risky sexual behaviorb	0.66 (0.90, 0–3)

^apast 30 days

^bpast 3 months

^cpast 12 months

^*One case is missing information.

Data Collection

All procedures were approved by the Institutional Review Board. Data were collected in-person or by video conference (e.g., Zoom; per COVID-19 pandemic response protocols). An assessor from the Center obtained parent consent and youth assent. The assessor completed the Structured Adolescent Interview (see below) with the parent and youth together, and then met with the youth individually to complete remaining measures. Families were compensated $10 for participation in screening and $20 for participation in the baseline interview. Randomization occurred after the baseline interview, so participants were unaware of their assignment to CM or PAU.

Dependent Variables

This study examined seven diverse forms of self-reported delinquency, substance use, and risky sexual behavior, as well as distinct groups (i.e., classes) of youth based on engagement in these problem behaviors.

Delinquent Behaviors

Delinquency was assessed through the Self-Report Delinquency Scale (Huizinga and Elliot 1986; Thornberry and Krohn 2000) administered to youth, considered the gold standard for youth self-reported delinquent behavior (Pechorro et al. 2019). Three scales of self-reported delinquency in the past 90 days were calculated to summarize the variety of crimes committed: public disorder offenses, property offenses, and against person offenses. All individual items were dichotomized (0 = none; 1 = 1 or more times) and were summed to create three “diversity” indices (see Sweeten 2012). Public disorder offenses included whether each participant reported being drunk, loud, rowdy, or unruly, or carried a hidden weapon in a public place. The range for public disorder offenses was 0 – 3 with an average of 0.58 (SD = 0.82); 60% of the sample did not report engaging in a public disorder offense in the past 90 days. Property offenses included engagement in 12 offenses that cover arson, damaging property, burglary, theft, and possession of stolen property. The range of responses for property offenses was 0 – 8, the average was 0.93 (SD = 1.43), and 57% of the sample reported no property offenses in the past 90 days. Against person offenses included three items: using a weapon or force to get money or things from people, attacking or attempting to attack someone with the idea of seriously hurting or killing him or her, and throwing objects at people or cars. Due to the low number of participants who reported engaging in more than one against-person offense, this item was dichotomized; 12% of the sample reported engaging in at least one against-person offense in the past 90 days.

Substance Use

Substance use was measured using the Global Appraisal of Individual Needs (GAIN), which is a commonly used measure of substance use among adolescents (Dennis et al. 2002; 2003). The GAIN contains 20 items to measure youth self-reported use of alcohol and drugs (not prescribed to the youth) during the past 30 days. Three measures of substance use were calculated: alcohol use, marijuana use, and other drug use in the last 30 days. For alcohol use, youth were asked “During the past 30 days, on how many days have you gotten drunk or had 5 or more drinks?” The average days reported was 0.94 (SD = 3.11) and 76% of the sample reported 0 days. Youth also reported the number of days they used marijuana, blunts, vaping of THC, edibles, or other forms of THC in the past 30 days. The average number of days reported was 4.73 (SD = 8.08), and 46% of the sample reported 0 days. Other drug use was measured by totaling the number of days in the past 30 days each participant reported using non-prescription drugs including, crack, cocaine, inhalants, heroin, methadone, opiates, PCP, LSD, tranquilizers/benzodiazepines, methamphetamines, ecstasy, stimulants, and downers/barbiturates. Due to the small number of youth reporting use of these substances, we dichotomized this variable to reflect using “other” drugs one or more times in the past 30 days; 21% of the sample reported using “other” drugs at least one day.

Risky Sexual Behavior

The Sexual Risk Behavior Scale (SRBS) was used to measure risky sexual practices (Fino et al. 2021; Jemmott et al. 1999). This measure was calculated by computing the total score for three dichotomous indicators of risky sexual behavior in the past 90 days: 1) age of first sexual encounter; 2) number of partners; and 3) frequency of unprotected sex. Following previous research that suggests that engaging in sexual intercourse prior to age 14 represents an important risk factor for a range of behavioral health problems (Kaestle et al. 2005; Lindberg et al. 2019; Siebenbruner et al. 2007), age of first sexual encounter was dichotomized into 13 and younger (representing early onset =1) and 14 and older (=0). Total number of partners distinguishes between 2 partners or more partners in the past 3 months (=1) or less than 2 partners in the past 3 months (= 0). This decision was also informed by previous research on youth who have been arrested (e.g., Krupa et al., 2021), adjudicated (e.g., Kan et al. 2010; Morris et al. 1995), or identified as having a SUD (e.g., Carey et al. 2004; Staras et al. 2011). Across these studies, average rates of sexual partners in the past year rarely exceeded 4 partners. Thus, 2 or more partners in the past 3 months (instead of past year) corresponds with these estimates. Frequency of unprotected sex reflects whether the youth had unprotected sex at least once in the past 90 days (0 = no reported occurrences of unprotected sex). The final risky sexual behavior variable was calculated by summing the scores of the three indicators with 0 reflecting no risky sexual behavior and/or not sexually active in the past 90 days and 3 indicating engagement in all three types of risky sexual behavior at least once. Across the sample, the average score was 0.77 (SD = 0.86); 47% reported no risky sexual behavior.

Independent Variables

The current study examined the association between the dependent variables and several independent variables: recent treatment intervention, family living situation, social support, youth behavioral health problems, and socio-demographic characteristics.

Demographics and Recent Treatment Intervention

The Structured Adolescent Interview (SAI) (Brown 1989) collected background information on the youth and family, as well as service utilization. To assess outpatient and inpatient treatment, four questions were used. Two separate questions asked whether, in the past 3 months, the youth received outpatient or office-based treatment for a substance abuse problem or a mental health problem. Two additional questions asked whether, in the past three months, the youth spent the night in a residential facility for a substance use or a mental health problem. Both outpatient and inpatient items were dichotomized to represent any outpatient treatment and any inpatient treatment (0 = no; 1 = yes). Approximately 44% of youth engaged in outpatient treatment in the last 90 days and 7% in inpatient treatment.

Family Living Situation

Family living situation was measured using data from the SAI (completed by parent and youth together) and the GAIN Living Risk Index (LRI; completed by the youth). The total number of primary caregivers the youth had in the past 12 months was reported in the SAI. On average, youth had 1.9 primary caregivers in the past 12 months (SD = 1.04). The LRI asked youth to report on the people they have regularly lived with in the past 30 days for 4 items measured on a 5-point Likert-type scale (none, a few, some, most, or all of them): were involved in illegal activity; weekly got drunk or had 5 or more drinks in a day; used any drugs; shout, argue, and fight most weeks. These items were summed to measure risk with the average score 5.41 across youth (SD = 1.81; α = 0.54).

Social Support

Social support was calculated using the GAIN Social Risk Index (SRI), which included the same 4 items as the LRI, but youth were reporting on the people they have regularly socialized with in the past 30 days. On average, youth scored 7.72 on the SRI (SD = 3.52; α = 0.79).

Youth Difficulties

Youth difficulties were measured using items from the Brief Problem Checklist (BPC) (Chorpita et al. 2010). The BPC includes 12 items measured on a 3-point scale (0 = not true; 1 = somewhat true; 2 = very true). The BPC has two subscales to capture internalizing behaviors and externalizing behaviors, each measured with 6 items. Internalizing difficulties included: I worry a lot; I am unhappy, sad, or depressed; I feel worthless or inferior; I feel too guilty; I am self-conscious or easily embarrassed; and I am too fearful or anxious. These items were summed; the average score was 11.29 (SD = 3.43; α = 0.83). Externalizing difficulties included: I argue a lot; I destroy things belonging to others; I disobey my parents or people at school; I am stubborn; I have a hot temper; and I threaten to hurt people. The average score was 9.99 (SD = 2.58; α = 0.74)

Additionally, the GAIN Substance Use Problem Scale (SPS) was calculated. This scale includes 17 dichotomous items (0 = no; 1 = yes) designed to measure whether youth substance use in the past 30 days has resulted in a range of problems, including complaints from parents, friends, or family members and failure to meet responsibilities at work, school, or home. These items were summed, with an average score of 3.55 (SD = 4.48; α = 0.92).

Socio-Demographic Characteristics

We also examined variation in each dependent variable across important socio-demographic characteristics. These variables included ethnicity (0 = Non-Hispanic, 1 = Hispanic), race (0 = Non-White, 1 = White [Note that Hispanic is included as ethnicity and not as race]), age (continuous), sex (1 = male, 2 = female), sexual orientation (0 = non-heterosexual, 1 = heterosexual), and receipt of public assistance (0 = no, 1 = yes).

Analysis Plan

The analyses for this study incorporated bivariate analyses and mixture modeling techniques. We did not perform multivariate regression analyses for three reasons. Due to the cross-sectional nature of our study, we were unable to account for time order in a valid manner so conclusions about causal relationships are not possible. Also, some important behavioral indicators have relatively low prevalence rates (see Table 1) which increases likelihood of inflated standard errors and type 1 errors (Allison 1999). Most important, our objective was to obtain a comprehensive understanding of the array of risk factors that are present at a critical stage of intervention planning (i.e., start of probation), regardless of their causal pathway.

To address the first research question, Kendall’s tau-b was used to examine the correlation among each of the risk factors and each of the seven problems behaviors. We used Kendall’s tau-b estimates and confidence intervals to account for the ordinal measurement and the modest sample size (Agresti 2010; Walker 2016). We also examined whether there were differences in engagement in each problem behavior across demographic characteristics. These analyses relied on analysis of variance (ANOVA, eta-squared) and chi-square tests of differences (χ², Cramer’s V) to identify any meaningful variations across subgroups of youth based on race, ethnicity, age, sex, sexual orientation, and family receipt of public assistance.

To address the second research question, a series of latent class models (LCA) based on the seven observed behaviors, were performed in Mplus 8.0 (Muthèn & Muthèn 1998–2017).³ LCA estimates a model that identifies latent “classes” or categories that are based on the covariation among observed indicators. These patterns of covariation are posited to be related to an underlying and unobserved factor (i.e., class). We relied on several fit and classification measures. The Bayesian Information Criterion (BIC), which is based on the log-likelihood value of the fitted model and entropy were considered. Lower BIC values suggest a better fitting model (Nylund et al. 2007). Entropy represents the quality of classification and ranges from 0 to 1; values closest to “1” suggest good classification (Vermunt and Magidson 2002). The classification table was also considered. High diagonal values, typically greater than 0.90, indicate good classification. Finally, the Bootstrapped Likelihood Ratio (BLRT) test compares the specified “k” class model to a “k-1” class model. Lower p-values suggest that the model with the smaller number of classes can be rejected in favor of the one with an additional class (Nylund et al. 2007). To ensure our LCA results have sufficient power to detect the correct number of classes, we then estimated Cohen’s d for each observed item included in the LCA. When relying on mixture modeling techniques, class separation is a crucial indicator of statistical power (Lubke & Muthen 2007; Meehl & Yonce 1996). Following the work of Tein et al. (2013), Cohen’s d above .80 was considered a valid indicator of class separation.

Once the best LCA model was identified, the posterior probabilities for most likely class membership were saved for use as the dependent variable. We then performed an additional set of bivariate analyses (using SPSS 28.0) exploring the association among prior service-utilization, family instability, social support, and youth difficulties across the identified classes. We also examined differences in demographic characteristics across the classes. Based on the categorical nature of the latent classes, analysis of variance (ANOVA) was used, along with the appropriate effect size measures (i.e., Bonferroni, eta-squared). For all bivariate analyses, we report our results using the traditional p < .05 threshold as our research questions are informed by theoretical and empirical work and therefore produce some general expectations about the relationships between the problem behaviors and the associated risk factors. We also believe this is appropriate since the modest sample size also guards against Type 1 errors. We also report the more conservative p < .001 threshold (which achieves Bonferroni’s correction method for multiple comparisons; 8 dependent variables/standard p < .05 threshold = .006) in the tables.

Results⁴

We performed a series of bivariate correlations to answer the first research question, which sought to examine differences across key risk factors (i.e., prior service-utilization, family instability, social support, youth difficulties) and engagement in each problem behavior. The results are presented in Table 2. A number of important similarities and differences in risk factors were found across the problem behaviors. For instance, externalizing difficulties and the SPS were significantly and positively correlated with all of the substance use (externalizing behavior: τ_b range = .13 - .22; SPS: τ_b range = .42 - .59) and delinquency (externalizing behavior: τ_b range = .18 - .33; SPS: τ_b range = .18 - .26) measures, but only externalizing difficulties was significantly correlated with risky sexual behavior (τ_b = .20). Statistically significant correlations were also observed among internalizing difficulties and the use of other drugs (τ_b = .20), all three delinquency measures (τ_b range = .14 - .28), and risky sexual behavior (τ_b = .24) but internalizing problems were not significantly related to getting drunk or using marijuana. These same findings were also found for prior outpatient mental health or substance use services. Youth who reported prior outpatient treatment had significantly higher scores for the use of other (more serious) drugs (τ_b = .21) and the three delinquency measures (τ_b range = .18 - .21).

Table 2.

Bivariate Associations among Risk Factors and Problem Behaviors (N = 210)

	Drunk	Marijuana	Other drugs	Public disorder	Property	Against-person	RSB
Risk Factors	τb (CI)	τb (CI)	τb (CI)	τb (CI)	τb (CI)	τb (CI)	τb (CI)
Externalizing problems	.19** (.13 −.25)	.13* (.04 – .22)	.22** (.16 – .28)	.25** (.18 – .32)	.33** (.26 – .45)	.18** (.13 – .24)	.20** (.12 – .28)
Internalizing problems	.06 (−.05 – .13)	. 09 (.07 – .18)	.20** (.14 – .27)	.20** (.12 – .28)	.28** (.20 – .36)	.14* (.08 – .19)	.24** (.16 – .32)
Primary caregivers	−.05 (−.11 – .01)	.06 (−.15 – .03)	.08 (.02 – .13)	.05 (−.03 – .13)	.09 (.02 – .17)	.06 (.01 – .11)	.12 (.04 – .19)
Living Risk index (LRI)	.12 (.05 – .19)	.08 (.00 – .17)	.12 (.06 – .18)	.17** (.09 – .24)	.16* (.07 – .24)	.06 (.00 – .11)	.02 (−.07 – .10)
Social Risk index (SRI)	.15* (.07 – .22)	.19** (.10 – .27)	.21** (.14 – .28)	.26** (.18 – .34)	.25** (.17 – .33)	.09 (.03 – .15)	.15** (.07 – .24)
Substance Problem Scale (SPS)	.42** (.36 – .48)	.59** (.53 – .64)	.48** (.42 – .54)	.26** (.19 – .33)	.25** (.17 – 33)	.18** (.13 – .24)	.01 (−.07 – .09)
Outpatient MH/SA treatment	.10 (.04 – .14)	−.07 (−.13 – −.05)	.21** (.16 – .27)	.18* (.11 – .27)	.21** (.14 – .28)	.20** (.15 – .24)	.11 (.04 – .18)
Inpatient MH/S treatment	−.07 (−.10 – −.04)	−.09 (−.14 – .06)	.07 (.04 – .11)	.04 (.00 – .08)	.10 (.06 – .14)	.06 (.03 – .09)	.08 (.04 – .13)

^*p < .05

^**p < .001 (achieves Bonferroni correction method threshold)

RSB = risky sexual behavior index

CI = confidence interval; RSB = risky sexual behavior

A significant and positive correlation was observed among the LRI and two delinquent behaviors: public disorder offending (τ_b =.17) and property offending (τ_b = .16). The SRI, which measures the characteristics of the participant’s peer network, was significantly correlated with all of the problem behaviors examined (τ_b range: .15 - .26), except against-person offending. Finally, the number of primary caregivers in the past year and prior inpatient mental health or substance use treatment were not significantly correlated with any of the problem behaviors.

Bivariate analyses of (ANOVA, χ²) were also performed to identify any differences in behaviors across youth characteristics. No meaningful differences were observed across sex, ethnicity, or age. A greater proportion of participants who identified as nonwhite reported the use of other, more serious forms of drugs (White youth: 18%; Nonwhite youth: 47%, χ² [1] = 8.80, p < .006, Cramer’s V = .21) and a greater proportion of participants who reported receiving public assistance reported engaging in at least one form of against-person felony (public assistance: 19%, no public assistance: 6 %, χ² [1] = 7.09, p < .006, Cramer’s V = .19). Also, a larger percentage of participants who did not identify as heterosexual reported the use of other more serous forms of drugs (non-heterosexual: 42%, heterosexual:18 %, χ² [1] = 7.02, p < .05, Cramer’s V = .18) and engaging in at least one type of against-person offense (non-heterosexual: 25%, heterosexual: 10 %, χ² [1] = 4.43, p < .05, Cramer’s V = .15).

The initial LCA entailed the specification of three separate models ranging from two to four classes. We compared the BIC, entropy, BLRT, and the classification table values as well as the substantive meaning of the identified classes. Compared to the 2-class model, the 3-class model had a lower BIC value (2-class BIC = 3967.02; 3-class BIC = 3878.88), both models showed good classification quality (2-class entropy = 0.99; 3-class entropy = 0.98), and high average latent class probabilities (range .93 – 1.00). The BLRT indicated that the two-class model should be rejected in favor of the three-class model (p < .01). Additionally, the characteristics of the additional (third) class were distinct and meaningful. Furthermore, the 4-class model produced a nonsignificant BLRT (p > .05), indicating that the three-class model should not be rejected.

We then computed Cohen’s d for each of the seven observed indicators in the LCA model (i.e., standardized difference between the mean values across classes). Cohen’s d was above 0.80 for all of the observed indicators (d ranged from 0.81 to 3.95) except risky sexual behavior (d = 0.36). Due to poor separation across classes (d < .80, see Tein et al., 2013), we removed risky sexual behavior and performed another set of LCA models (i.e., 2, 3, and 4 classes). The results were consistent across models with and without the risky sexual behavior measure. The 3-class solution showed better model fit indices (i.e., BIC, BLRT) compared to the 2-class model (BIC: 2-class = 3395.03, 3-class = 3302.95; entropy: 2-class = 0.99, 3-class = 0.98; BLRT suggested rejecting the 2-class model in favor of the 3-class model, p < .001). Additionally, class membership and the value of the class estimates in the 3-class model remained the same after removing risky sexual behavior. As a result, we report the results of the 3-class LCA model without the risky sexual behavior measure. A description of the three classes is presented in Table 3.

Table 3.

Latent Class Analysis

	C1: Experimenting	C2: Polysubstance Use + Delinquent Behavior	C3: Diverse Delinquent Behavior
	n = 147	n = 50	n = 13
Observed Indicators:	Mean/SE	Mean/SE	Mean/SE
Drunk	0.01 (0.01)	1.33 (0.07)**	0.00 (0.00)
Marijuana	3.37 (0.57)**	8.92 (1.34)**	2.85 (1.45)**
Public disorder	0.34 (0.05)**	0.96 (0.12)**	1.75 (0.33)**
Property/theft	0.47 (0.07)**	1.27 (0.21)**	4.47 (0.43)**
	Conditional Probability/SE	Conditional Probability/SE	Conditional Probability/SE
Against-person (yes)	0.03 (0.02)*	0.23 (0.06)**	0.58 (0.13)**
Other drugs (yes)	0.09 (0.02)**	0.48 (0.07)**	0.37 (0.14)*

^*p < .05

^**p < .001 (achieves Bonferroni correction method threshold)

The three classes are characterized by: 1) experimentation, 2) polysubstance use and delinquent behavior, and 3) diverse delinquent behavior. The first class, referred to as the “Experimenting” group comprised 70% of the sam ple. The youth in this class reported the lowest rates of offending behavior, getting drunk, and using serious drugs. On average, this class reported smoking marijuana on three of the past 30 days. The second class, which comprised 24% of the sample, was labeled the “Polysubstance Use + Delinquent Behavior” group. The major distinction between the behavioral profiles of this class and the “Experimenting” class is the higher rates of getting drunk, using marijuana, and the use of “other” drugs. The third class, which comprised 6% of the sample, was characterized by “Diverse Delinquent Behavior.” This group revealed the lowest rates of getting drunk and using marijuana, and the highest rates of all three delinquency measures. Of note, the near zero standard errors for “drunk” reflect the low rates of engagement in this behavior among youth in the “Experimentation” and “Diverse Delinquent Behavior” classes. Most notably, the youth in this group reported substantially higher rates of against person offenses and property offenses.

The final set of analyses involved examining differences in prior service-utilization, family instability, social support, and youth difficulties across the latent classes. This is the focus of the third research question and the findings are displayed in Table 4. Four risk factors were found to be statistically significant across the classes: externalizing problems (F [2, 207] = 9.27, η² = .08), substance problems (F [2, 207] = 31.18, η² = .23), LRI (F [2, 207] = 4.43, η² = .04), and SRI (F [2, 207] = 9.01, η² = .08). Pairwise comparisons indicated that the “Experimenting” class scored significantly lower than the other two groups on externalizing behavior and the SRI (Cohen’s d ranged from 0.58 – 0.76). For the LRI and SPS, pairwise comparisons showed that the “Experimenting” group scored significantly lower than the “Polysubstance + Delinquent Behavior” class (Cohen’s d: LRI = .43, SPS = 1.23) but there were no significant differences in LRI or SPS scores among the “Experimenting” and the “Diverse Delinquent Behavior” groups or the “Polysubstance + Delinquent Behavior” and the “Diverse Delinquent Behavior” groups.

Table 4.

ANOVAs Examining Differences in Risk Factors across Latent Classes

	C1: Experimenting	C2: Polysubstance Use + Delinquent Behavior	C3: Diverse Delinquent Behavior
	n = 147	n = 50	n = 13
Risk factors:	Mean (SD)	Mean (SD)	Mean (SD)
Externalizing problems**	9.54 (2.44)b,c	11.02 (2.53)a	11.54 (2.82)a
Internalizing problems	11.02 (3.33)	11.68 (3.44)	13.23 (4.21)
Primary caregivers	1.97 (1.08)	1.78 (0.71)	2.31 (1.60)
Living risk index (LRI)*	5.18 (1.48)c	6.02 (2.43)a	5.77 (2.09)
Social risk index (SRI)**	7.07 (2.96)bc	9.22 (4.34) a	9.31 (3.79)a
Substance Problem Scale (SPS)**	2.19 (3.48) c	7.24 (4.65) a	4.77 (5.75)
Outpatient MH/SA treatment	0.41 (0.49)	0.54 (0.50)	0.62 (0.51)
Inpatient MH/S treatment	0.07 (0.26)	0.04 (0.20)	0.23 (0.44)

^*p < .05

^**p < .001 (achieves Bonferroni correction method threshold)

^asignificantly different from C1

^bsignificantly different from C2

^csignificant different from C3

We also explored whether youth demographic characteristics varied across latent classes. No statistically significant differences were observed across race/ethnicity, sex, age, sexual orientation, or receipt of public assistance (results are available upon request) and all effect size estimates (eta-squared or Cramer’s V) were below .10. These findings suggest that the observed covariation among the problem behaviors was relatively universal among the youth in our study.

Discussion

The purpose of this study was to add to the limited research on the behavioral profiles and associated risk factors for youth who are involved in the JJ system, have been identified as having a substance use problem, and reside in mostly rural communities. The first objective was to explore the risk factors for engagement in seven different forms of problem behavior that encompass diversity in substance use, delinquency, and sexual risk-taking. Higher externalizing problems were correlated with every problem behavior measured and the SPS scale was correlated with all of the problem behaviors except risky sexual practices. These findings provide clear evidence that engaging in problems behaviors is associated with difficulties in other domains of life functioning that may serve as effective targets of intervention, regardless of the youths’ specific behavioral profile. The only other risk factor found to be significantly and positively correlated with getting drunk and using marijuana was the SRI. Since the use of alcohol and marijuana are the most common substances used by adolescents (CDC 2019; SAMHSA 2020), it could be that use of these more common substances is largely impacted by one’s social setting. Research suggests that peer substance use is a strong and consistent predictor of both getting drunk and using marijuana among adolescents residing in rural communities (Farrell et al. 1992; Zhen-Duan and Taylor 2014). Thus, youth who primarily use alcohol and marijuana, and use them frequently, may represent a unique group of youth with a SUD in terms of their intervention needs. Focusing on the social environment, peer influences, and redirecting the use of free time may be beneficial.

Furthermore, we found the same patterns for internalizing difficulties and prior outpatient treatment. Higher scores for each risk factor were correlated with the three delinquency measures, use of serious forms of drugs, and risky sexual practices, but not getting drunk or using marijuana. This finding appears to be somewhat counterintuitive. Youth who recently completed mental health or substance use treatment should report lower rates of serious drug use, delinquency, and risky sexual practices - but only if the intervention was effective. The similar patterns found suggest a possible mismatch of intervention services and treatment needs. Given the robust findings that behavioral health services for adolescents in rural areas are lacking quality, specificity, and effectiveness (Andrilla et al. 2018; Gamm et al. 2010; Hardin et al. 2018), this is a likely scenario.

Several scholars have provided both theoretical and empirical evidence that the use of serious drugs, as well as engaging in delinquent and risky sexual behavior, serve as a mechanism to cope with internalizing problems (Agnew 1992, 2017; Loeber et al. 1999). Research also suggests that youth residing in rural communities may experience unique components of negative affect (e.g., anger, alienation, depression) due to the higher rates of poverty, geographic isolation, and community instability in rural communities (Swaim et al. 2001; Taylor et al. 2013). Thus, the distinctive circumstances of youth who reside in rural communities may lead to internal coping mechanisms that are unique to rural youth, and therefore, require specialized intervention services that effectively target their internalizing difficulties and facilitate the development of healthy coping strategies in the context of a rural community.

The second objective was to identify whether distinct subgroups of youth could be identified based on the covariation among the seven observed behaviors. Three classes were identified, each representing a distinct behavioral profile that varied in both the frequency and types of problem behaviors (see Table 3). The characteristics of the LCA classes generally aligned with prior research, especially the extraction of a small group of youth who engage in high frequency and diverse delinquent behaviors. Among samples of youth who are involved in the JJS, the presence of a small group of youth who report high rates of diverse and delinquent behavior has been established for decades (e.g., see for example seminal works such as Wolfgang et al. 1987, Farrington & West 1993 or LCA studies such as Jennings & Reigle 2012). However, the removal of risky sexual behavior from the LCA is a unique finding, especially since most prior LCA studies that include measures of sexual behavior do not assess Cohen’s d across classes (e.g., Childs et al. 2016; Krupa et al. 2021; Sullivan et al. 2010, Mun et al., 200). Confirmation of the (lack of) covariation among sexual risk-taking and other problem behaviors through replication with other samples of adolescents is necessary, including whether this finding is unique to predominantly rural samples, justice-involved youth, or youth with a SUD. Given the relatively recent push to introduce prevention and intervention efforts targeting the sexual health of youth involved in the JJS (Elkington 2020; Belenko et al. 2008), an important next step for researchers is investigating how best to intervene in the lives of youth who engage in multiple problem behaviors and whether their sexual health should be a priority within JJ settings.

The third objective was to examine whether youth difficulties varied across the identified groups. Four risk factors significantly differed across the classes: living risk, substance use problems, social risk, and externalizing behaviors. These types of risk (e.g., home environment, social support, and externalizing difficulties) appear to be important risk factors that distinguish youth who infrequently engaged in diverse forms of problem behavior (i.e., experimentation) from youth who are currently engaging (or have recently engaged) in higher frequency and more diverse forms of substance use or delinquent behavior. Interestingly, however, no significant differences in risk factors were observed across the “Diverse Delinquent Behavior” class and the “Polysubstance Use/Delinquent Behavior” class. This finding could signal similar intervention needs among these groups (based on the risk factors included in our analyses) or that important distinctions in needs or risk factors across these unique classes are driven by factors not included in the current study (see limitation sections below). It is important that future studies address this issue by including a more comprehensive set of risk factors that align with current research on youths’ risk and need factors (e.g., Bonta & Andrews 2016; Brogan et al. 2015).

In general, our descriptive analyses suggest that there are critical nuances in the behavioral patterns of youth that must be considered during intervention planning - even in a sample that could be judged as homogenous (i.e., youth involved in the justice system who meet criteria for a SUD and reside in mostly rural areas). These important distinctions in co-occurring problem behaviors and the associated risk factors are not easily extracted from a universal threshold of “substance use problems” or “substance use disorder.” According to our findings, even among samples of youth who meet SUD criteria and are involved in the justice system, recent engagement in problem behaviors varied by frequency and type of behavior. For example, a very small group reported high rates and diverse forms of delinquent behavior, while a somewhat larger group engaged in higher rates and diverse forms of substance use. Accordingly, in-depth assessments that can account for these distinct behavioral profiles are necessary. One recent study of JJ agencies found that rural agencies (compared to urban agencies) had higher rates of substance use among youth under community supervision but were less likely to use substance use screening or assessment tools (Marks et al. 2019). Without proper screening and assessment, it is difficult for frontline personnel to match intervention services to the multidimensional and unique needs of youth involved in the justice system. From an organizational standpoint, the lack of aggregate-level screening and assessment data also prevents data-driven decisions about needed provider contracts, program selection, and treatment modalities. Thus, our findings provide further evidence that comprehensive screening and assessment practices across key intervention points of JJ systems in rural communities are necessary to ensure proper services are available, accessible, and delivered effectively to youth who have diverse substance use and delinquency needs.

Understanding the multidimensional needs of youth who are involved in the justice system is an essential first step to ensuring that the right interventions are available within the JJ system. Since the primary focus of the JJ system is public safety, the term “effective” in justice settings is routinely defined through recidivism or reductions in delinquent behavior and rarely considers other forms of problem behavior. Most JJ systems are not oriented toward increasing access to healthcare, providing intensive and targeted substance use treatment, or targeting multiple co-occurring behaviors simultaneously. Recently, Belenko et al. (2017) have developed a framework for linking the JJ system with behavioral and primary healthcare agencies through screening at key “transition points” in the system, where missed opportunities for screening, assessment, and intervention often occur. The behavioral health cascade model calls for an informed, service delivery system that provides both public health and public safety benefits (Knight et al. 2015). As described by Marks et al. (2019, 40) “…decreasing substance use among justice-involved youth is contingent on an interconnected behavioral health service cascade including screening, assessment, service referral, and sustained delivery of treatment and recovery support services.” This framework is particularly relevant considering the characteristics of the “Polysubstance Use + Delinquency” class found in our study. The need for an interconnected behavioral health service model becomes even more obvious in rural areas where effective services are limited (Oser et al. 2011), children have poorer health profiles (Curtis et al. 2011), and rates of substance use are found to be high (Marks et al. 2019).

Study Limitations

There are important limitations to the current study that should be acknowledged. This study was based on a relatively small and unique sample. As a result, generalizability of our findings is limited. It is important to note, however, that a great deal of research on the behavioral health of adolescents from rural areas is also based on smaller samples of adolescents, given the difficulties in conducting research in small, hard to reach communities. The sample size also led to a very small latent class representing youth who engaged in high frequency and diverse delinquent behavior (i.e., Class 3). This size of this class prevented us from performing multivariate analyses assessing differences across latent classes. There are also additional risk factors that have been found to be important predictors of adolescent problem behavior that were not included in our study. These factors broadly fall into categories that include puberty and development, cognitive functioning, school difficulties, current and prior offense severity, and parental attachment and monitoring (Cottle et al. 2001; Smith et al. 2013; Willoughby et al. 2014).

Additionally, given the focus of the broader study from which this sample was drawn, our service history measure only included prior mental health or substance use service utilization. Reification must also be addressed. LCA provides a valuable tool for identifying heterogeneity in the covariation among behavioral items within samples (Lanza and Rhoades 2013), these groups should not be considered concrete groups because the patterns observed could be unique to the specific sample, measurement, and study context. For example, the youth in our sample reported relatively low frequencies of engagement in several of the dependent variables (e.g., getting drunk, delinquency measures, risky sexual behavior). Therefore, LCA results based on samples of youth who report higher rates of these behaviors may yield difference results. Several scholars have noted the limitations of using mixture modeling techniques (e.g., LCA) to draw firm conclusions about policy or practice (Lanza & Rhoades 2013; Nagin 2004).

Conclusion

This study represents our effort to better understand the intervention needs of youth who are involved in the JJ system, have been identified as having a substance use problem, and reside in largely rural communities. Our findings provide clear evidence that the behavioral profiles and associated risk factors of this population are multidimensional. Future research should continue to understand the complex intervention needs of rural youth who have substance use problems, identify which interventions are most effective for youth with different constellations of needs, and consider how JJ agencies can address missed opportunities for intervention in the community. Specific to intervening with rural youth, prior results from the current trial have proven that JPOs in these rural communities can deliver CM with a high degree of adherence (Sheidow et al., 2020), which may provide a viable intervention for youth in the JJ system who have substance use problems. The effectiveness of CM as a means for JPOs to reduce substance use and other problem behaviors, will be examined in future reports from the trial.

Biographies

Author Bios

Kristina Childs, Ph.D. is an Associate Professor in the Department of Criminal Justice at the University of Central Florida. Her research interests include juvenile risk and need assessment practices, evaluation of prevention and intervention programs for at-risk youth, and the effectiveness of mental health and de-escalation training and education for front-line juvenile justice decision-makers.

Jill Viglione, Ph.D. is an Assistant Professor in the Department of Criminal Justice at the University of Central Florida. Her research focuses on the implementation of evidence-based practices, individual and organizational responses to policy reform, and decision making within correctional agencies.

Jason E. Chapman, Ph.D., is a Senior Research Scientist at the Oregon Social Learning Center who specializes in research design, measurement development and evaluation, and advanced statistical methods. His research includes studies evaluating the efficacy, effectiveness, dissemination, and implementation of evidence-based practices across mental health, child welfare, and justice settings. Additionally, Dr. Chapman focuses on IRT-based development and evaluation of instruments for measuring intervention and implementation fidelity.

Tess K. Drazdowski, Ph.D., is a Research Scientist and licensed clinical psychologist at the Oregon Social Learning Center. Her research concentrates on the prevention and intervention for the misuse of prescription drugs, cannabis use, and polysubstance use in primarily young adults. Dr. Drazdowski has published peer-review articles on anxiety, sleep, substance use, young people impacted by the justice system, people of color, and the LGBTQ community.

Michael R. McCart, Ph.D., is a licensed clinical psychologist with specialized training in cognitive, behavioral, and family systems approaches to treating serious problems in adolescents and emerging adults. Dr. McCart’s research centers on enhancing behavioral health services for two high-risk populations: (1) adolescents and emerging adults with substance use and co-occurring behavior problems and (2) victims of interpersonal violence. His work spans all phases of intervention research, including initial development and pilot testing of treatment protocols, randomized efficacy studies, and multi-site effectiveness trials.

Ashli J. Sheidow, Ph.D., is a Senior Research Scientist and Science Director at the Oregon Social Learning Center (OSLC). Dr. Sheidow researches treatments for mental health and substance use problems in adolescents and emerging adults, particularly those who have co-occurring problems or are justice-involved. She is also focused on effective dissemination of evidence-based practices, in particular through improving training and support for community-based providers. Her research interests have focused broadly on the development, prevention, and treatment of adolescent and young adult psychopathology and delinquency from an ecological perspective, with concentrations in co-occurring disorders, effective implementation of evidence-based practices, and advanced quantitative methods.