Spillover Effects of Job Skills Training on Substance Misuse Among Low-Income Youths With Employment Barriers: A Longitudinal Cohort Study

Abstract

Objectives. To examine spillover effects of job skills training (vs basic services only [e.g., adult basic education, job readiness training]) on substance misuse among low-income youths with employment barriers.

Methods. Data came from the National Longitudinal Survey of Youth 1997, a longitudinal cohort study of youths born between 1980 and 1984 in the United States. Based on respondents’ reports of substance misuse (past-month binge drinking and past-year marijuana and other illicit drug use) from 2000 to 2016, we estimated substance misuse trajectories of job skills training (n = 317) and basic services (n = 264) groups. We accounted for potential selection bias by using inverse probability of treatment weighting.

Results. Compared with the basic services group, the job skills training group showed notable long-term reductions in its illicit drug misuse trajectory, translating to a 56.9% decrease in prevalence rates from 6.5% in year 0 to 2.8% in year 16.

Conclusions. Job skills training can be an important service component for reducing substance misuse and improving employment outcomes among youths with economic disadvantages and employment barriers.

Today, more than 40 million persons in the United States live in poverty.¹ These individuals are at heightened risk for substance-related problems because they are likely to endure multiple adverse life conditions such as material hardship, unemployment, and physical and mental health problems.^2,3 For example, nearly 20% of male and 10% of female adults in households receiving Temporary Assistance for Needy Families (TANF) benefits because of low income (i.e., < $800 per month for a family of 3 in a majority of states⁴) had a past-year substance use disorder, rates 2 times higher than their nonpoor, nonrecipient counterparts.⁵ In addition, those in poverty often lack access to adequate substance abuse treatment given their increased likelihood of being uninsured and the financial burdens of seeking treatment.^6,7 Their increased risk for substance misuse and barriers to receiving treatment impede their chances of achieving self-sufficiency and raise public health concerns.

To address substance misuse and co-occurring life barriers, enhancing employment among low-income populations is essential. A more stable and better-paying job can help improve physical and behavioral health by addressing chronic and acute life stressors including material hardship, greater job strain, and lack of benefits (e.g., health insurance).⁸ In fact, research shows strong associations between employment status and substance misuse among at-risk populations.^9,10 Better employment can increase substance abuse treatment accessibility through higher wages and insurance benefits. It may also address other barriers to treatment entry and retention⁵ via expanded social networks that can provide information about services and service providers.¹¹

A major challenge to enhancing employment is low educational attainment and limited job skills. Without public support, opportunities to gain job skills that can enhance labor market success will be limited for low-income individuals, especially with the disappearance of jobs in sectors such as manufacturing. To help low-income persons enter or re-enter the workforce, US states have required work-eligible TANF recipients to participate in welfare-to-work programs. For other individuals with employment barriers, including laid-off workers and those with basic skills at or below the eighth-grade level, other programs sponsored under federal law (e.g., the Workforce Innovation and Opportunity Act [Pub. L. No. 113–128]) offer career and training services as do federally sponsored programs specifically serving youths aged 16 to 24 years (e.g., Job Corps).

Although these programs aim to improve participants’ employability, the focus has increasingly been on immediate employment (a “job-first” approach), with job search and basic educational services (e.g., adult basic education and job readiness training, hereinafter called basic services) the main program activities, partly because of cost concerns.^12,13 For instance, the 3-level service structure of the Workforce Investment Act programs, which operated between 1998 and 2014, offered training services only to those who failed to find any type of job after participating in the first 2 levels, which provided basic services only.¹³ However, evidence shows that basic services alone are not effective in improving labor market outcomes.¹⁴ If job skills training is necessary to improve long-term employment and earnings among employment program participants, positive spillover effects into other life domains such as substance misuse are anticipated only for those who received job skills training services.

Previous studies have investigated the contributions of job skills training in facilitating the recovery of individuals in substance abuse treatment^15,16 as well as associations between socioeconomic status and substance use over the life course.¹⁷ However, less clear is how job skills training may influence substance misuse among the broader population of youths facing economic disadvantages as they transition into adulthood. If substance misuse does not improve, or if improvements are short-lived, this may imply that spillover effects from job skills training are insufficient for overcoming other adverse life conditions.¹⁸ On the other hand, long-lasting improvements in substance misuse would imply that improved economic conditions aid in maintaining these positive spillover effects from job skills training. Examining substance misuse trajectories among those who participated in job skills training programs versus those who received basic services only can provide insights about the types of services that improve both economic and substance misuse outcomes.

THE PRESENT STUDY

This study addressed 2 research questions: (1) Does job skills training (vs basic services only) lead to less substance misuse among youths participating in government-sponsored employment programs? (2) If so, do these spillover effects persist over a 16-year period into participants’ adulthood? To address these questions, we used data from the National Longitudinal Survey of Youth 1997 (NLSY97) on binge drinking, marijuana and other illicit drug use, and government-sponsored employment program participation. To account for selection into job skills training services, we used inverse probability of treatment (IPT) weighting to balance the distributions of key baseline characteristics between the job skills training and basic services groups.

METHODS

The NLSY97 is an ongoing prospective study using nationally representative samples of birth cohorts born between 1980 and 1984. From the survey’s first wave in 1997 until its latest (17th) wave conducted in 2015 to 2016, the NLSY97 has provided information about study participants’ education and training history, labor market outcomes, and a wide array of physical and behavioral health measures. With use of a stratified multistage area probability sampling design, 9907 eligible individuals were identified from 147 primary sampling units. Each primary sampling unit consisted of a metropolitan area or 1 or more nonmetropolitan counties with a minimum of 2000 housing units selected from the National Opinion Research Center’s 1990 master probability sample. Of 8984 baseline respondents, 7103 completed the latest interview conducted in 2015 to 2016 (a 79% retention rate). The NLSY97’s custom weighting program (https://www.nlsinfo.org/weights/nlsy97) generates longitudinal weights that make the sample representative of the population during the years of follow-up interviews from which the data were drawn while accounting for respondents who did not participate in those years. More details about NLSY97’s study procedures and design are available from the US Department of Labor Bureau of Labor Statistics (https://www.bls.gov/nls/nlsy97.htm).

With data from surveys conducted from 1998 to 2007, we identified 581 individuals who participated in government-sponsored employment programs for low-income youths and adults such as the Job Training Partnership Act Program, Job Corps, and Job Opportunities and Basic Skills Training Program. Of the program participants, 317 (56.4%) reported receiving job skills training services: vocational training for a specific job, on-the-job training, work experience, and other classroom training for a specific job. The remaining 264 (43.7%) participants received only basic services (i.e., not involving job skills development) such as adult basic education, English as a second language, General Educational Development (GED) program, and job search or readiness training services.

Measures

Substance misuse.

We assessed 3 dichotomous measures (0 = no; 1 = yes) of substance misuse: (1) past-month binge drinking (for both men and women, having ≥ 5 drinks on the same occasion [at the same time or within hours of each other] in the past 30 days), (2) any past-year marijuana use, and (3) any past-year other illicit drug use such as cocaine, crack, heroin, or any other substance to get high or achieve an altered state. We retrieved substance misuse data from follow-up surveys conducted between 2000 and 2015 following each program participant’s latest enrollment in a government-sponsored employment program. Although the NLSY survey has been conducted every other year since 2011, information on substance misuse was not collected in 2013; therefore, all substance misuse data are from surveys conducted in 2000, 2002, 2004, 2006, 2008, 2010, and 2015.

Labor market outcomes.

As key labor market outcomes, we examined employment status and annual earnings. We considered respondents reporting any past-year earnings (i.e., wages, salary, commissions, or tips) employed while we considered the rest not employed. We calculated annual earnings by summing the amount of past-year wages, salary, commissions, or tips from all jobs respondents reported, before deductions for taxes or anything else. We retrieved labor market outcome data from follow-up surveys conducted every other year between 2000 and 2015. To account for inflation over the study period, we adjusted earnings to 2018 dollars by using the Consumer Price Index.

Individual, family, and school- and peer-related characteristics at baseline.

To equate the distributions between the basic services and job skills training groups, we assessed a total of 28 individual, family, and school- and peer-related covariates at baseline and their missingness patterns that could have led to their selection into job skills training (see Table A, available as a supplement to the online version of this article at http://www.ajph.org, for detailed descriptions). Individual characteristics included basic sociodemographic information (e.g., age, sex, citizenship, highest grade completed), general health condition, behavioral and emotional health, delinquency, criminal justice system involvement, and previous substance use and misuse behaviors. Family characteristics included parental structure, parents’ general health and educational attainment, household income, and family and home risk index. School and peer characteristics included perceived school safety, perceived teacher’s interest in students, and theft experience at school.

Analyses

We conducted statistical analyses in 2 steps. First, we used IPT weighting to adjust for selection bias as individuals with employment barriers (e.g., behavioral health problems, previous criminal justice involvement) may be more likely to receive job skills training. We chose IPT weighting over other propensity score methods for its advantages in generalizing study findings as it allows for incorporating sampling weights in model estimation.¹⁹ To balance covariate distributions between study groups, IPT weighting reweights data based on the inverse of estimated propensity scores. For this study, we used the twang package (https://www.rand.org/statistics/twang/stata-tutorial.html), which uses a generalized boosted model to estimate propensity scores.²⁰ The generalized boosted model is considered an effective machine learning technique especially for models involving nonlinear relationships between treatment condition and a large set of covariates.²⁰ After reweighting data with the IPT weights, we conducted diagnostic assessments on 2 balance metrics, the absolute standardized mean differences and the Kolmogorov–Smirnov statistic, to determine whether satisfactory balance between the basic services and job skills training groups was achieved. Covariate balance is considered achieved if the effect size is less than 0.1 according to Rubin’s²¹ threshold and the Kolmogorov–Smirnov statistic is not significant for any baseline covariate.

Second, we estimated mixed-effects logistic regression models of substance misuse behaviors separately for binge drinking, marijuana use, and other illicit drug use to generate developmental trajectories starting from the year of government-sponsored program participation and up to 16 years after. Consistent with Singer’s sequential model building processes,²² we determined optimal mixed-effects logistic models by adding additional fixed and random effects 1 at a time and examining whether the addition produced a significant increase in the Wald statistic. Also, we initially included higher-ordered (up to cubic) year terms to capture potential curvilinear trajectories and we removed nonsignificant terms.

Lastly, we considered multiple error variance and covariance structures to identify a model with the best fit. Based on the Wald χ² test, we assumed an identity error structure (i.e., equal variances with zero covariances) for the marijuana use models and the independent error structure (i.e., unique variance per random effect with zero covariances) for both binge drinking and other illicit drug use models. The final model equation (with the significant interaction term) is reported here:

where is log odds of substance misuse for individual i at time t; is year(s) since program participation for individual i at time t; and is a job skills training participation indicator for individual i. Parameter estimates included as the intercept, as the slope of the trajectory of substance misuse for the basic service group, as the treatment coefficient (i.e., the job skills training effect when  = 0), as quantification of the degree to which the trajectory of substance misuse is moderated by job skills training, as the random effect for individual i, and as the random effect for the slopes for years since program participation.

RESULTS

Table B (available as a supplement to the online version of this article at http://www.ajph.org) displays baseline sociodemographic characteristics separately for the basic services and job skills training groups. Overall, there were similarities in many individual and family characteristics including age, education, parental structure, mother’s education, and household income–to–poverty level ratio, but the job skills training group was significantly more likely to be male (χ² = 6.03; P = .01) and more likely to be White (χ² = 2.48; P = .06), though the latter did not reach statistical significance.

Figure 1 presents the prevalence of respondents’ baseline binge drinking and past-year marijuana and other illicit drug use by job skills training service receipt status. For comparison purposes, the substance misuse of NLSY97 participants who did not participate in government-sponsored employment programs is also reported. Overall, the job skills training group showed the highest involvement in substance misuse at baseline, followed by nonparticipants, and then basic services recipients. For instance, among older adolescents (aged 15−16 years at baseline), about 21% of job skills training participants engaged in binge drinking in the past month, and 41% used marijuana and 14% used other illicit drugs in the past year. The group differences in some sociodemographic characteristics and substance misuse behaviors between the job skills training and basic services groups require accounting for treatment selection bias to correctly estimate substance misuse trajectories after program participation.

FIGURE 1—

Prevalence Estimates of Baseline Substance Misuse by Government-Sponsored Employment Program Participation and Job Skill Training Receipt Status for (a) ≥ 5 Drinks on Same Occasion in Past Month, (b) Marijuana Use in Past Year, and (c) Other Illicit Drug Use in Past Year: National Longitudinal Survey of Youth 1997 (Years 1997–1998), United States

Note. BS = basic services; JST = job skills training; NS = no service. Questions about binge drinking and marijuana use were first asked in the 1997 survey. Questions about other illicit drug use were first asked in the 1998 survey. For this reason, the ages for other illicit drug use at baseline were 13 to 15 years for younger cohorts and 16 to 17 years for older cohorts. Sample size for each group was as follows. For ages 12–14 years: no service (n = 4736), basic services (n = 149), and job skills training (n = 182); for ages 15–16 years: no service (n = 2902), basic services (n = 115), and job skills training (n = 135).

Inverse Probability of Treatment Weighting and Covariate Balance

IPT weighting resulted in satisfactory covariate balance on the baseline covariates and their missingness indicators between the basic services and job skills training groups based on both absolute standardized bias and Kolmogorov–Smirnov test measures. Thus, no further refinements in propensity score models were necessary. Also, the estimated IPT weights ranged from 1.05 to 4.14, indicating no evidence of extreme weights.

Estimated Labor Market Outcome Trajectories

Before estimating substance misuse over time, we examined employment status and annual earnings trajectories up to 16 years after program participation separately for the job skills training and basic services groups. For employment status, nearly 90% of the job skills training group was employed throughout the study period, which was about 10 percentage points higher than the basic services group. In terms of earnings trajectories (in the natural logarithm to the base of the mathematical constant e), the job skills training group reported significantly higher average annual earnings beginning in year 3 following program participation (Figure 2).

FIGURE 2—

Estimated Log Earnings Trajectories Among Government-Sponsored Employment Program Participants for Basic Services and Job Skills Training: National Longitudinal Survey of Youth 1997 (Years 2000–2016), United States

Note. BS = basic services only; JST = job skills training. The estimated inverse probability of treatment weights were used to reweight the sample to adjust for each individual’s conditional probability of assignment to the basic services or job skills training group. Earnings were transformed by natural logarithm to the base of the mathematical constant e.

Substance Misuse Trajectories

Table 1 presents findings from the mixed-effects logistic models of substance misuse. Binge drinking decreased significantly ( = −0.06; P < .001) among both the basic services and job skills training groups with no group differences found in the trends. Slightly more than 40% of both groups engaged in binge drinking initially, and these rates gradually decreased to approximately 30% in year 16 (Figure A, available as a supplement to the online version of this article at http://www.ajph.org). For past-year marijuana use, the basic services group reduced use until year 2, but we observed no other notable reductions in the trajectories nor differences in prevalence for the 2 groups over time (Figure B, available as a supplement to the online version of this article at http://www.ajph.org). Between 11% and 16% of the groups engaged in past-year marijuana use throughout the study period since year 2 following program participation.

TABLE 1—

Mixed-Effects Logistic Regression Analysis of Substance Misuse by Job Skills Training Receipt Status Among Government-Sponsored Employment Program Participants (n = 581): National Longitudinal Survey of Youth 1997 (Years 2000–2016), United States

	≥ 5 Drinks on Same Occasion (Past Month)		Marijuana Use (Past Year)		Other Illicit Drug Use (Past-Year)
	b (SE)	Exp(b)	b (SE)	Exp(b)	b (SE)	Exp(b)
Fixed effects
Job skills training ()	0.15 (0.22)	1.16	−0.34 (0.26)	0.71	1.55** (0.49)	4.71
Year (	−0.06*** (0.02)	0.95	−0.62*** (0.10)	0.54	−0.05 (0.10)	0.95
Job skills training * year (	. . .	. . .	. . .	. . .	−0.17* (0.07)	0.85
Intercept ()	−0.57** (0.19)	0.56	−1.19*** (0.21)	0.30	−5.33*** (0.59)	0.01
Variance of random effects
Intercept ()	3.68 (0.52)	. . .	0.45 (0.13)	. . .	3.04 (1.08)	. . .
Year slope ()	. . .	. . .	0.45 (0.13)	. . .	0.03 (0.02)	. . .
Intraclass correlation	0.53		0.12		0.48
Wald χ2 statistic	13.81**		38.05***		15.11**

Note. The estimated inverse probability of treatment weights were used to reweight the sample to adjust for each individual’s conditional probability of assignment to the basic services or job skills training group. The Wald statistic shows whether all fixed effects parameters in each model are simultaneously zero. For both binge drinking and marijuana use models, the interaction term between job skill training status and year was not included because of insignificance. The binge drinking model did not include random effects for year because of insignificant increases in the Wald χ² statistic compared with a model without the random effects for year. For the marijuana use model, equal variance for the random effects was assumed in addition to a random effect covariance of zero because the error covariance structure led to a significantly higher Wald χ² statistic compared with the default assumption (i.e., unique variance parameters for each random effect and a covariance of zero between the random intercept and slope). *P  < .05; **P  < .01; ***P  < .001.

For past-year other illicit drug use, the job skills training group reported significantly higher involvement (6.5%) in its initial year of program participation, although this overall rate is much lower than the baseline rate among those aged 16 to 17 years (14%). The marginal effects of job skills training on other illicit drug use (i.e., the differences in predicted probabilities between the 2 groups) decreased over time from 4.6% in year 0 to −2.5% in year 16 (Figure C, available as a supplement to the online version of this article at http://www.ajph.org). This resulted in a 2.8% illicit drug use rate among the job skills training group in year 16, sharply contrasting with the increasing trajectory among the basic services group from 1.8% in year 0 to 5.2% in year 16 (Figure 3). In addition, supplementary analyses (Figure D, available as a supplement to the online version of this article at http://www.ajph.org) indicated that once earnings were controlled, group differences in the prevalence estimates for illicit drug use immediately following program participation and in the trajectories disappeared.

FIGURE 3—

Predicted Probabilities of Past-Year Other Illicit Drug Use Among Government-Sponsored Employment Program Participants for Basic Services and Job Skills Training: National Longitudinal Survey of Youth 1997 (Years 2000–2016), United States

Note. BS = basic services only; JST = job skills training. The estimated inverse probability of treatment weights were used to reweight the sample to adjust for each individual’s conditional probability of assignment to the basic services or job skills training group. The predicted probabilities were derived from the mixed-effects logistic model of past-year illicit drug use other than marijuana.

DISCUSSION

We examined substance misuse among individuals with economic disadvantages over a 16-year period following participation in US government–sponsored employment programs. Over time, both groups were less likely to binge drink. In the initial years of program participation, both groups were also less likely to use marijuana and other illicit drugs compared with baseline; however, only the job skills training group continued to experience further reductions in other illicit drug use. Given that reductions in other illicit drug use were no longer significant once earnings were controlled, and that other studies show associations between socioeconomic status and substance misuse,¹⁷ examining the roles of improved employment and earnings on reduced substance misuse among job skills training participants is also needed.

Despite gradual improvements in substance misuse, the high prevalence of binge drinking and illicit drug use in the first year of program participation points to the need for early screening and treatment services. Of those receiving job skills training, more than 40% at program entry (and nearly 30% at the end of the study’s 16-year follow-up period) engaged in binge drinking, higher than the average rate (25%) among the general adult population aged 18 to 34 years in 2015.²³ Although other illicit drug use prevalence among the job skills training group (6.5%) was more than 3 times that of the basic services group (1.8%) at program entry, the prevalence for the job skills training group declined substantially over the study period, whereas it increased for the basic services group. Public assistance offices and government employment agencies may be important intervention settings for addressing substance misuse among program participants as other studies indicate.^24,25

However, little evidence exists to guide effective substance use screening and intervention practices at public assistance offices and government employment agencies. Stigma and fear of not meeting eligibility requirements or losing benefits because of substance misuse (under 21 USC 862b) have been some of the major challenges in early detection among TANF-eligible individuals.²⁶ Morgenstern et al.²⁵ found that screening by an addiction counselor increased identification rates among public assistance recipients with a substance use disorder by between 10% and 49% relative to generic screening by caseworkers using a short paper-and-pencil self-report survey such as the CAGE Questions Adapted to Include Drugs.²⁷ Such promising approaches should be assessed further to help more youths and young adults receiving public assistance obtain needed substance abuse education and treatment services and as alternatives to drug toxicology testing, which is seen as punitive.

Moreover, the effectiveness of substance abuse interventions in public assistance programs may be enhanced through comprehensive, coordinated interventions that incorporate job skills training.^28,29 For instance, the Therapeutic Workplace, which involves job skills development services (phase I) and employment assistance (phase II), not only increased employment but also increased drug abstinence and medication adherence.³⁰ On the other hand, substance abuse treatment approaches focusing on job readiness training (e.g., workshops on resume building and job interviews) such as the Job Seekers’ Workshop³¹ did not lead to significant employment gains over time relative to standard care.³² Furthermore, strategies such as intensive case management^24,28 can help individuals receiving public assistance meet needs such as childcare, transportation, and housing so they can participate in and successfully complete training programs. Operant conditioning using monetary incentives may also help those with substance use disorders achieve their employment goals.¹⁶ To better understand the effectiveness of these innovative approaches, more research is needed with low-income youths at risk for substance misuse.

Limitations

The current study had several limitations. First, we used a case–control design, limiting ability to make causal assertions. The effects of job skills training may have been confounded if group differences that were not accounted for by observed baseline characteristics affected substance misuse differentially by treatment status. Second, social desirability and recall biases as well as subjective assessment of binge drinking (e.g., what constitutes an alcoholic drink) may have affected response accuracy. In addition, attrition might have caused bias, although NLSY97’s longitudinal weights were constructed to make each follow-up survey’s respondents representative of the baseline sample to minimize bias.

Third, the NLSY97 does not include information on severity or clinical assessment of substance misuse and related problems that could enhance understanding of the problems individuals with economic disadvantages face. Fourth, although substance misuse tends to differ across population subgroups (e.g., by gender, race/ethnicity),³³ substance misuse trajectories for subgroups could not be estimated, partly because of the small sample size. Given women’s unique life experiences such as pregnancy and primary childcare responsibilities,^34,35 investigation of gender-specific substance misuse trajectories is needed to better serve women who misuse substances and need employment and training services.

Lastly, lack of contextual factors for marijuana use (e.g., changes in perceptions and accessibility) and information about types of illicit drugs used limited further investigation. Examining recent policy and perception changes regarding marijuana use may aid in understanding marijuana’s distinctive use trajectories compared with other illicit drugs following job skills training. Measures of prescription drug misuse would also be helpful in clarifying whether employment in certain occupations results in greater drug use in the initial years of program participation.

Public Health Implications

Our findings suggest that job skills training can be an important strategy for reducing illicit drug involvement over the life course for youths facing socioeconomic disadvantages and employment barriers. However, continued high rates of binge drinking and marijuana use indicate the need to test early screening and intervention measures at public assistance offices and government employment agencies to reduce these forms of substance misuse and to determine how job training and employment programs can assist in these efforts.