Adolescent Substance Use Prevention: Long-Term Benefits of School Engagement

Abstract

BACKGROUND:

To determine if school engagement is a viable target for early prevention of adolescent substance use, this study investigated whether school engagement in early adolescence (ages 12–14) is a cause of alcohol and cannabis use during middle to late adolescence (ages 15–19).

METHODS:

To facilitate causal inference, inverse probability of treatment weights (IPTWs), which are based on estimated probabilities of treatment selection (ie, school engagement), were created based on a robust set of potential confounders. Using the IPTWs, a cumulative link mixed model was fit to examine the impact of school engagement on alcohol and cannabis use among an ethnically diverse sample of adolescents (N = 360).

RESULTS:

School engagement was associated with a lower level of alcohol and cannabis use from age 15 to 18. School engagement was not associated with change in alcohol and cannabis use over time, suggesting that school engagement emits its effect early in the developmental course of substance use and offers protection throughout adolescence.

CONCLUSIONS:

This study supports a compensatory role of early school engagement in substance use across middle and late adolescence. School engagement is a malleable factor and thus offers an avenue for prevention efforts.

Alcohol and cannabis use typically begin and escalate during adolescence. According to the 2019 Monitoring the Future national survey, use of these substances increases from 8th to 12th grades (24.5% and 58.5%, respectively have used alcohol; 15.2% and 43.7%, respectively, have use cannabis).¹ Adolescent substance use not only has negative consequences over the course of adolescent development (eg, greater problems at school, involvement in delinquency, health-risking behaviors),² but also may lead to substance abuse and dependence in adulthood.³ Focusing on the etiology of adolescent substance use may be the best way to identify specific targets for intervention to delay the onset and escalation of substance use throughout adolescence. Our orientation is to take a strength-based approach; aiming to identify modifiable promotive factors that may work as a compensatory/cancellation mechanism to protect youth from becoming involved in substance use.⁴

Many cross-sectional studies have identified school engagement as a potential compensatory factor against adolescent substance use,^5–7 and longitudinal studies have largely corroborated these findings.^8–12 However, little work has been conducted to determine if school engagement is simply predictive of subsequent onset and escalation of substance use, or if school engagement plays a causal role in preventing subsequent substance use. In this study, we build on the existing literature to explore whether school engagement from ages 12 to 14 may counter alcohol and cannabis use across middle and late adolescence (ages 15–19). We extend prior research by using a multidimensional construct of school engagement to provide a more comprehensive portrait of school engagement, and we utilize an analytic method capable of evaluating the causal effect of school engagement on subsequent use of alcohol and cannabis.

School engagement is a dynamic motivational process that includes emotional, behavioral, and cognitive dimensions.^13,14 Emotional engagement represents positive reactions to teachers, peers, and schools, which includes a sense of belonging and enjoyment of learning. Social control theory¹⁵ and the social development model¹⁶ explain that emotionally attached students form bonds with their school community and endorse school norms discouraging delinquency and substance use. Behavioral engagement refers to involvement in academic and/or extracurricular activities. Behaviorally engaged students are likely to be busy participating in academic and extracurricular activities; thus, they have no time to explore problematic behaviors. Cognitive engagement refers to a student’s strategic or self-regulated approach to learning to master difficult skills and comprehend complex ideas. Cognitively engaged students are more likely to use positive strategies rather than substances to deal with school difficulties and obstacles.¹⁷ For each of these reasons, school engagement, a modifiable protective behavior, has the potential to prevent the onset and escalation of substance use among adolescents. However, prior research in this area has mainly focused on the emotional dimension of school engagement such as school connectedness or school bonding,^8,10,18 despite the multifaceted nature of school engagement.

School engagement is a long-term developmental process. From the first grade, students are exposed to academic and social experiences that form the basis for a long and gradual development of school engagement or disengagement.¹⁹ The process of school disengagement can predict later outcomes such as dropout and delinquency.^12,20 In addition to this, several studies have shown that early school disengagement is predictive of later substance use. For example, adolescents ages 13 to 14 who felt disconnected from their schools were more likely to use substances 2 years later.⁸ Similarly, adolescents ages 13 to 18 who scored high in school bonding were less likely to use substances at age 18.¹⁸ Also, lower 6th grade GPAs were associated with substance use escalation in junior high school.²¹ As such, early school engagement has been linked with resistance to later substance use, and several substance use prevention/intervention programs focus on school engagement as a major source of protection.^22,23

While there is strong observational and theoretical support for school engagement as an important compensatory behavior to prevent substance use, most studies purporting this claim rely on regression models that do not adjust for confounders (eg, variables that may cause both poor school engagement and subsequent substance use) or utilize adjustment methods that are not capable of uncovering a causal effect. Without a randomized control trial to effectively test a causal effect, other factors such as early behavioral problems and/or poor family environments may confound the effect of early school engagement on substance use.^12,24 In other words, the effect of school engagement on substance use may not in fact be causal, but rather indicative of selection bias in which students who display more positive behavioral patterns in childhood or who have more advanced cognitive skills in childhood are simply more likely to both engage in school and avoid substance use during adolescence. In order to determine if an intervention designed to improve school engagement could result in less substance use, a causal effect of school engagement must be ascertained.

In situations in which a treatment variable (eg, school engagement) cannot be randomly assigned, the use of propensity scores²⁵ and inverse probability of treatment weights (IPTWs²⁶) can be used to mimic a randomized control trial with observational data. This is accomplished by identifying the key confounding variables that can produce selection bias, building a model to predict the treatment (ie, school engagement) from these confounding variables, creating individual weights that reflect the individual probability of having received the treatment based on the confounding variables, and fitting a weighted model for the outcome (ie, substance use) to assess the treatment effect as if all of the confounding variables were unrelated to receipt of the treatment.²⁶

To further elucidate the temporal association and strengthen a causal link between school engagement and substance use; we use IPTWs developed from a robust set of potential confounders including individual characteristics of adolescents and their families. Specifically, Vanderweele and Shpitser²⁷ suggests that a pre-exposure variable should be included as a confounder if it causes either the exposure or outcome or both when examining a causal pathway of interest. Thus, we include confounders that are either a cause of school engagement or of adolescent substance use or both. With regard to adolescent’s characteristics, a lack of cognitive skills and problem behaviors (eg, internalizing and externalizing behaviors) in childhood could lead to both school disengagement²⁸ and substance use in adolescence^29–31 and thus should be controlled. Also, given evidence of intergenerational congruence in substance use^32,33 and a negative association between parent’s age at childbirth and substance use among their children,³⁴ parental substance use history and parent’s age at childbirth are included.

To summarize, our aim for this study is to use IPTWs in a weighted regression to examine the causal effect of a multidimensional construct of school engagement (measured at ages 12–14) on subsequent substance use (measured at ages 15–19) including alcohol and cannabis. We hypothesize that early school engagement will lead to less use of alcohol and cannabis from ages 15 to 19.

METHODS

Data

Data for this project come from the Rochester Intergenerational Study (RIGS), an extension of the Rochester Youth Development (RYDS), a longitudinal study that began in 1988 with 1000 7th and 8th-grade students (referred to as G2) from Rochester, New York. Male adolescents and students from neighborhoods with high resident arrest rates were oversampled to over-represent youth at high risk for antisocial behavior (73% male; 68% Black, 17% Hispanic, 15% White). RYDS participants completed regular interviews every 6 months from 1988 to 1992 (Phase 1), annually from 1994 to 1996 (Phase 2), and biannually from 2003 to 2006 (Phase 3). Beginning in 1999, RIGS selected G2’s oldest biological child (referred to as G3) and added new firstborns to the G3 sample in each subsequent year when the child turned 2years old. The RYDS parent (G2) and other primary caregiver completed annual interviews until the child (G3) turned 18. Children completed annual interviews beginning at age 8. A total of 539 parent-child dyads participated in RIGS. A total of 371 adolescents provided data on school engagement between the ages of 12 and 14 and other control variables. Of these, 11 adolescents were removed because there was no information on substance use between the ages of 15 and 19, yielding the final sample of 360 adolescents. All data collection procedures were approved by the Institutional Review Board of the University at Albany.

Measures

School engagement.

In yearly interviews, adolescents ages 12 to 14 answered 11 questions germane to three dimensions of school engagement: Emotional, behavioral, and cognitive engagement. Items are presented in Table 1. Where needed, items were reverse coded so that a higher score for each item indicated greater school engagement. The factor structure of these 11 items was explored using a structural equation model in Mplus, Version 8.4. A second-order factor model was fit, which included three first-order factors (emotional, behavioral, and cognitive engagement), each assessed by multiple items. The general construct of school engagement was a second-order factor accounting for commonality among the three specific dimensions. To take into account the nonindependent observations (repeated measures between the ages of 12–14), the Cluster option (ie, Cluster = student ID) in Mplus was used. The model showed acceptable fit (RMSEA = .043; CFI = .964, TLI = .944). The scores of the second-order factor (ie, school engagement) were averaged across ages 12 to 14, and then standardized. The standardized factor score representing overall school engagement was exported from Mplus to be used in subsequent models.

Table 1.

Early School Engagement (ages 12–14)

Items	Range	Mean	SD
Emotional engagement
School is boring to you*	1 = strongly disagree, 4 = strongly agree	2.74	.80
You like school a lot	1 = strongly disagree, 4 = strongly agree	3.02	.74
Behavioral engagement
Homework is a waste of time*	1 = strongly disagree, 4 = strongly agree	3.23	.77
You usually finish your homework	1 = strongly disagree, 4 = strongly agree	3.35	.65
Average grades in English, math, science, and social studies	0 = falling, 3 = above average	2.19	.49
Cognitive engagement
You try hard at school	1 = strongly disagree, 4 = strongly agree	3.47	.61
Getting good grades is very important to you	1 = strongly disagree, 4 = strongly agree	3.67	.52
Someti mes you do extra work to improve your grades	1 = strongly disagree, 4 = strongly agree	3.15	.73

^*Reverse coded item.

Adolescent substance use.

From ages 15 to 19, adolescents self-reported their substance use each year. If they indicated that they used alcohol and/or cannabis since the last interview, they were asked how many times they had used each substance in the past year. Those who indicated monthly use were then asked whether they had experienced any of nine possible consequences including, for example, difficulties with school or work due to substance use and escalated use to achieve satisfaction. Responses were used to create an ordinal measure of substance use at each age from 15 to 19: 0 = no use; 1 = rare use (less than once per month); 2 = regular use (monthly use, but without problems); 3 = problem use (monthly use causing problems).

Internalizing and externalizing behaviors.

In each yearly interview, G2s and the other primary caregiver of male G2s responded to 64 questions from the Child Behavior Checklist.³⁵ We created the average of internalizing and externalizing behaviors spanning ages 7 to 9 to represent overall childhood problem behaviors.

Verbal ability.

Verbal ability was measured using the Peabody Picture Vocabulary Test (PPVT³⁶) when the child was 4 or older. Children older than 4 at the start of the study were administered the test during their first year the family participated in the study. When RIGS began its ninth year, it began administering the PPVT to children at age 8. To account for age heterogeneity, we used a standardized score of verbal ability that can be compared across ages.

Parent’s substance use during adolescence.

In waves 2 through 6 of RYDS (mean age = 14.5–16.5), G2s reported whether they had used alcohol and cannabis since the last interview (approximately 6 months), and if so, how often. Those who indicated monthly use were then asked whether they had experienced any of six possible consequences for each substance use including difficulties with school and escalated use to achieve satisfaction. Based on the responses, we created an ordinal measure of each substance use: 0 = no use; 1 = rare use (less than once a month); 2 = regular use (monthly use, but without problems); 3 = problem use (monthly use causing problems). We retained the maximum of alcohol and cannabis use during adolescence.

Socio-demographic variables.

We included a set of sociodemographic variables, including G2 sex (female as the reference group), G2’s age at G3’s birth, G2’s birth year, G2’s highest education level in early adulthood (mean age = 29), arrest rate of G2’s neighborhood during adolescence (a sampling parameter in RYDS), G3’s sex (female as the reference group), and G3’s race/ethnicity using three dummy variables for Black, Hispanic, and non-Hispanic White (other/mixed-race as the reference group).

Analysis Plan

We used inverse probability of treatment weights (IPTWs) to reduce potential bias from confounding.²⁶ To estimate our IPTWs for a continuous treatment (ie, level of school engagement), we fit two linear regression models, both with school engagement as the dependent variable. The first defined the numerator and the second defined the denominator of the equation to construct the IPTWs.²⁶ For the numerator, we fit an unconditional linear regression model. From this model, we computed the value of the probability density function (pdf) for the normal distribution for each individual with mean equal to the intercept and standard deviation equal to the standard deviation of the residuals. For the denominator, we fit a conditional regression model, regressing school engagement on the potential confounders. From this model, we computed the value of the pdf for the normal distribution for each individual with mean equal to the predicted value and standard deviation equal to the standard deviation of the residuals of the model. The IPTWs were formed by dividing the numerator by the denominator and truncating at the 1st and 99th percentiles to eliminate extreme weights. To evaluate the effectiveness of the weights in balancing confounders, we estimated the Spearman correlation coefficient between school engagement and each of the 14 confounders after applying the weights.³⁷ The value of Spearman correlation coefficients indicated minimal confounding as all correlations (range: −.06 to .06) were less than a threshold of 0.10.³⁸

To examine the impact of school engagement on substance use, we used a weighted cumulative link mixed model using the clmm() function from the ordinal package³⁹ in R version 4.0.2. Our model’s intercept was defined at age 15, and we included age and age-squared terms to capture both the instantaneous change in substance use as well as acceleration/deceleration in substance use over time. The need for random effects in these growth parameters was studied using nested models. For both alcohol and cannabis use, the most complex model tested, one with a random intercept, slope, and quadratic term, would not converge—thus, random effects for the quadratic term were excluded. For alcohol use, the addition of a random linear slope did not provide a better fit over the random intercept model (χ² = 5.29, df = 2, p = .07). However, the addition of a random linear slope significantly improved model fit over a random intercept model for cannabis use (χ² = 39.73, df = 2, p < .001). Thus, quadratic growth models were fit for both alcohol and cannabis use—the intercept alone was specified to be random for the alcohol use model, and both the intercept and linear slope were specified to be random for the cannabis use model.

From the fitted models, three predicted cumulative probabilities (ie, any use, monthly use or greater, and problem use) were calculated by using estimated marginal means for 1 SD above and below the mean of school engagement across adolescent age using the emmeans package⁴⁰ in R version 4.0.2.

RESULTS

Table 2 presents the descriptive information for each of the confounders used to form the IPTWs. Table 3 presents descriptive information for the alcohol and cannabis use scores across ages 15 to 19. As is normative, substance use appears to increase with age.

Table 2.

Descriptive Statistics for Covariates (N = 360)

	Range	Mean/Proportion	SD
Adolescent-related variables
Internalizing behaviors during childhood	0–1.12	0.30	0.21
Externalizing behaviors during childhood	0.01–1.00	0.38	0.23
Verbal ability during childhood	45–126	90.69	12.73
Race/ethnicity
Black	0,1	0.68	—
Hispanic	0,1	0.10	—
White	0,1	0.07	—
Male	0,1	0.50	—
Parent-related variables
Alcohol use during adolescence	0–3	1.14	1.12
Cannabis use during adolescence	0–3	0.53	0.94
Highest education	10–18	11.92	1.42
Birth year	1973–1976	1974	0.76
Male	0,1	0.63	—
Age at G3’s birth	15.20–29.90	20.59	2.90
Community arrest rate	0.12–7.87	4.38	1.99

Table 3.

Prevalence of Substance Use by Age

	No use (%)	Rare use (%)	Regular use (%)	Problem use (%)
Alcohol use at age 15 (N = 353)	89	7	3	1
Alcohol use at age 16 (N = 341)	83	14	2	1
Alcohol use at age 17 (N = 326)	79	17	2	2
Alcohol use at age 18 (N = 314)	62	28	7	3
Alcohol use at age 19 (N = 295)	50	30	14	6
Cannabis use at age 15 (N = 353)	93	3	2	2
Cannabis use at age 16 (N = 341)	87	6	3	3
Cannabis use at age 17 (N = 326)	81	9	5	6
Cannabis use at age 18 (N = 314)	71	13	8	8
Cannabis use at age 19 (N = 295)	68	7	15	10

Alcohol Use

The left side of Table 4 presents the results of the model for alcohol use. Greater school engagement manifested in less alcohol use at age 15 (b = −.49, SE = .22, p < .05). We did not find evidence that school engagement influenced the instantaneous rate of change (b = −.18, SE = .18) or acceleration (b = .07, SE = .04) in alcohol use. To evaluate whether school engagement had an enduring compensatory effect across the observation period, the centering point for the intercept of the growth model was rotated to consider each subsequent age. More engaged students used less alcohol through age 18 (b = −.60, SE = .16, p < .001 at age 16; b = −.56, SE = .16, p < .001 at age 17; b = −.39, SE = .15, p < .05 at age 18).

Table 4.

Parameter Estimates (standard errors) from Ordinal Cumulative Link Mixed Model

Parameter	Alcohol use	Cannabis use
Fixed effects
Intercept 1	3.24 (.25)	6.98 (.82)
Intercept 2	5.66 (.31)	8.32 (.86)
Intercept 3	7.22 (.36)	9.94 (.90)
Age	.48 (.19)*	2.31 (.43)***
Age2	.08 (.04)	−.26 (.07)***
School engagement	−.49 (.22)*	−1.08 (.42)*
Age × school engagement	−.18 (.18)	.20 (.28)
Age2 × school engagement	.07 (.04)	−.01 (.06)
Random effects
Standard deviation for intercept	2.00	4.44
Standard deviation for slope (centered at age 15)	-	1.03
Correlation between intercept and slope	-	−.73

Note. Age was centered at 15.

^*p < .05,

^***p < .001.

Cannabis Use

The right side of Table 4 presents the results of the model for cannabis use. Greater school engagement manifested in less cannabis use at age 15 (b = −1.08, SE = .42, p < .05). We did not find evidence that school engagement influenced the instantaneous rate of change (b = .20, SE = .28) or acceleration (b = −.01, SE = .06) in cannabis use. Recentering the intercept, we find evidence that more engaged students used less cannabis through age 18 (b = −.89, SE = .29, p < .01 at age 16; b = −.73, SE = .25, p < .01 at age 17; b = −.60, SE = .23, p < .01 at age 18).

Figure 1 depicts the results. Students with high levels of school engagement showed less alcohol and cannabis use through age 18. While students who showed high school engagement reported less alcohol and cannabis use at age 19, the effect was not significant due to a large standard error (b = −.08, SE = .17 for alcohol use; b = −.49, SE = .27 for cannabis use). We do not find evidence that school engagement influenced the rate of change in substance use.

Figure 1.

Cumulative Probabilities of Substance Use as a Function of School Engagement (1 standard deviation (SD) below mean, mean, and 1 SD above mean)

DISCUSSION

The early precursors of substance use are viable targets for prevention and intervention programs aimed at curtailing adolescent substance use uptake and escalation. In this study, we show that early school engagement has long-lasting positive impacts on alcohol and cannabis use throughout adolescence. By using IPTWs and a 5-year follow-up period for substance use, the findings add to longitudinal evidence that early school engagement protects against substance use across middle and late adolescence. Furthermore, we use a multidimensional measure of school engagement to provide a more comprehensive picture of the construct. Overall, we find evidence for a causal mechanism of school engagement that provides protection to young people in delaying onset and reducing the level of substance use.

Students who showed high school engagement in early adolescence reported less alcohol and cannabis use from ages 15 to 18. The findings align with recommendations for early interventions that encourage school engagement.^12,19 Indeed, several programs designed to increase school engagement in early adolescence have effectively reduced the risks that students will drop out of school, abuse substances, and suffer from mental health problems across adolescence.^22,41,42 The early middle school years are likely to be a key time for altering trajectories of substance use.

While school engagement predicted both less alcohol and cannabis use at each age, stronger countereffects are apparent for cannabis use. According to the 2019 Monitoring the Future study, problematic alcohol use (ie, binge drinking) has declined in the last decade among adolescents,¹ but cannabis use has increased rapidly and surpassed tobacco use,⁴³ perhaps because cannabis has been legalized in many states. Although legalization is currently limited to adults, it has been shown to increase adolescent cannabis use. For example, after recreational cannabis was legalized in Washington, 8th and 10th graders reported increased use and decreased perceptions that it is harmful.⁴⁴ Given the potential harmful impact of cannabis use on adolescents, such as poor mental health, comorbid substance use, and substance use disorder,^45,46 and the impact of cannabis legalization on adolescents, our findings have implications for intervention and prevention planning.

Strengths and Limitations

The findings are strengthened in that the sample is racially/ethnically diverse, a notable strength given that the majority of work to ascertain risk and protective factors for substance use has been conducted with majority White samples.^11,47 Efforts to understand risk and protective factors in non-White populations are critical, and can offer insight for addressing health and academic disparities among young people of color.^48,49 Considering that school engagement is positively correlated with academic achievement,⁵⁰ early prevention and intervention programs aimed at promoting school engagement may prevent adolescent substance use and make progress toward reducing achievement gaps. Schools could strongly justify interventions that promote school engagement because they may both prevent substance use and enhance academic outcomes.

Although we add to the literature by examining the long-term effect of early school engagement on adolescent substance use, our study is not without limitations. First, the results are based on findings from a community-based sample in one urban jurisdiction in New York. Thus, generalizability may be compromised. Second, all measures were self-reported or parent-reported, and are prone to reporting error. Third, although the study included multiple aspects of school engagement, it ignored facets such as peer and teacher relationships and school contexts such as school climate that might influence engagement and substance use.

Conclusions

We have shown that students who become engaged with their schools are more likely to avoid problem substance use across adolescence. We strongly stress the need for preventive interventions in early adolescence, before high school, to encourage students to establish and maintain school engagement and thus decrease both short- and long-term substance use and related problems.

IMPLICATIONS FOR SCHOOL HEALTH

The use of a multidimensional construct of school engagement has varying implications for preventive interventions that might increase school engagement across multiple school settings, such as increasing participation in classroom and school-based activities, building relationships, and teaching problem-solving skills. Also, considering that school engagement has long-lasting impacts on substance use across adolescence, early intervention may have a substantial impact on problem behaviors across adolescence. Implications for improving and maintaining school engagement to prevent adolescent substance use are outlined below:

• Prevention efforts to enhance school engagement in early adolescence may be more effective when they integrate students’ behavioral, emotional, and cognitive engagement.⁵⁰

• Schools should identify students who are at risk for disengagement as early as first grade by creating an early warning measure such as standardized test scores and attendance. Implementing early warning systems and interventions of at-risk students in school settings are essential.¹²

• Support for enhancing school-family relationship can be beneficial. For example, family-focused interventions designed to enhance school engagement among middle school students through supportive parenting skills (eg, parent-child communication to help parents positively reinforce and monitor school work) showed long-term effects on substance use in late adolescence.⁴¹ Thus, increasing promotive or compensatory factors in school and family settings may help prevent substance use across adolescence.

• After a year of remote learning due to the COVID-19 pandemic, students may have lower levels of school engagement. Thus, allocating resources to support school-based interventions to improve school engagement in a post COVID-19 era is needed.