Learning and not using? The effect of degree attainment on illicit drug use among at-risk youth

ABSTRACT

This paper examines the causal effect of earning a GED or vocational degree on future illicit drug use, employing random assignment into the United States’ most comprehensive education and vocational training program for at-risk youth – Job Corps – as a source of exogenous variability in degree attainment. Nonparametric bounds under relatively weak monotonicity assumptions are constructed to allow the random assignment to violate the exclusion restriction when used as an instrument. We also use a fixed effect model and propensity score weighting to supplement the results. The results from different methods suggest that degree attainment may have the most significant effect in reducing the illicit drug use of blacks, while the results for whites and Hispanics are less conclusive.

KEYWORDS:

• Job corps
• education
• illicit drug use
• local average treatment effects

JEL CODES:

• I2
• I12
• K42

Supplemental data

Supplemental data for this article can be accessed online at https://doi.org/10.1080/00036846.2022.2137456.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ The most widely used in-school program, Project DARE (Drug Abuse Resistance Education), has had its effectiveness for reducing and preventing drug use called into question. See, for example, Ennett et al. (1994), Rosenbaum et al. (1994), and the references therein.

² Most economic studies posit that early drinking and substance use can lead to students’ poor academic success (see, for example, Cook and Moore 1993; Yamada, Kendix, and Yamada 1996; Bray et al. 2000; Register, Williams, and Grimes 2001; Dee and Evans 2003; Roebuck, French, and Dennis 2004; Chatterji 2006; McCaffrey et al. 2010), rather than the reverse relationship that we study.

³ Flores-Lagunes, Gonzalez, and Neumann (2010) finds that while Hispanics are earning degrees, they do not see increases in their future earnings as do blacks and whites. Amin et al. (2016) finds suggestive evidence that Hispanics do not see as large of a decrease in crime compared to blacks and whites after attaining a degree.

⁴ Nearly $72\mathrm{\%}$ of the control group members participated in some form of education or training program within the 48 months after random assignment.

⁵ The response rates for the interviews were high at 95, 90, 79, and 80% for the baseline and 12, 30, and 48 month follow-ups, respectively (Schochet, Burghardt, and Glazerman 2001). To account for interview non-response items and different sampling into the control and treatment groups, we utilize NJCS supplied probability weights, the construction of which are described in Schochet (2001).

⁶ The sample used in this paper is similar in size to Flores-Lagunes, Gonzalez, and Neumann (2010). The loss of observations due to missing values makes the sample slightly smaller than the one used in Schochet, Burghardt, and McConnell (2008), but larger than the samples in Blanco, Flores, and Flores-Lagunes (2013), Flores and Flores-Lagunes (2013), Bampasidou et al. (2014), and Amin et al. (2016).

⁷ We appreciate the comment from the anonymous referee that using the drug use outcomes at the 48-month follow-up may lead to weakened results by lowered drug use due to employment drug checks.

⁸ We define degree attainment as 60 days before the start of the interview for the 30-month follow-up in September 1997. In this way, educational degree attainment is measured before the drug use outcomes, which is 30 days prior to the 30-month follow-up interview.

⁹ Recall that individuals were not prevented from receiving academic or vocational training outside of JC, even if they were randomized out of JC programs. This allows individuals in the control group to have a potential degree attainment value of $D_i(0)=1$ if they acquire a degree outside of JC and $D_i(0)=0$ otherwise. Likewise, individuals randomly assigned to participate in JC can attain a degree ($D_i(1)=1$) or not ($D_i(1)=0$).

¹⁰ The total average effect of JC randomization, $E[Y(1)-Y(0)]$, is the intention-to-treatment (ITT) effect under non-compliance associated with JC random assignment.

¹¹ Due to the $LAT{E}_c$ bounds being less informative, we present the results in Table B1 of Appendix B and not in the main text.

¹² Data on societal costs come from the United States Bureau of Labor Statistics (2020), Center for Behavioral Health Statistics and Quality (2020), United States National Institute on Drug Abuse (2020), and Recovery Centers of America (2017). The upper bound in potential savings is derived from the “best case scenario” estimate of the ATT for all drug use for the black sample, while the lower bound is derived from the propensity score estimate.

¹³ While the 2019 Fiscal Year operating budget of Job Corps was roughly $\mathrm{\$}1.72$ billion (United States Department of Labor 2020), it is common practice to multiply the cost of a public program by a factor of $1.3$ to account for the deadweight loss of taxation (Finkelstein and Hendren 2020).