Abstract
In this paper, I develop and estimate a dynamic structural model of education, employment, and substance use decisions of young men in order to determine the causal effects of substance use on educational attainment and career paths. Heavy substance use is correlated with lower school attainment and labor market outcomes; however, it is unclear if heavy substance use causes these worse outcomes. One concern is that those who are more likely to use marijuana or alcohol frequently are those for whom the labor market would offer lower wages regardless of their substance use. I utilize variation in the prices of substances, the price of college, local law enforcement characteristics, and unemployment rates to help identify the channels through which current substance use and school decisions affect future substance use, employment decisions, and wages. Current research generally treats the substance use decision as a binary choice, making it difficult to distinguish the effects of moderate versus heavy use. I allow individuals to make choices about their levels of alcohol, cigarette, and marijuana use in order to capture the full relationship between substance use and outcomes.
I estimate my model using restricted-access data from the 1997 National Longitudinal Survey of Youth and Bayesian Markov Chain Monte Carlo (MCMC) methods. My model includes extensive time-persistent unobserved heterogeneity, which helps me to estimate the causal effects of substance use on outcomes. I use a modified version of the estimator proposed in Imai, Jain, and Ching (2009) in order to make estimation feasible by easing the computational burden of evaluating my likelihood function and value functions.
I find that heterogeneity in preferences plays an important role in an individual's choice to use cigarettes and marijuana. That is, without heterogeneity, the proportion of individuals using cigarettes and marijuana would be much less. I find that cigarette and alcohol use have causal effects that decrease the wages of white and Hispanic males, but have no statistically significant effects on the wages of black males. I also find that past and present alcohol and cigarette use affect an individual's choice to use marijuana, with alcohol having a larger effect. Additionally, I find that marijuana use leads individuals into heavy cigarette use, supporting the reverse gateway theory.
Lastly, I find that white males have a lower probability of arrest than black and Hispanic males, conditional on age, previous arrests, and substance use. In addition, the use of heavy marijuana increases the probability of arrest more for black males than it does for white or Hispanic males. White males are also more likely to graduate from high school and to be working full-time at the age of 24, even though their substance use is comparable to that of black and Hispanic males. This suggests that arrests may be contributing to the education gap between white and minority males. I run two policy simulations to see if the outcomes of minority males can be improved by decreasing the probability of arrest. In the first, I set the coefficients in the probability of arrest equations of Hispanic and black males equal to those in the equation of white males. I find that high school graduation rates increase by 3.8 percent and 6.7 percent for Hispanic and black males, respectively. I also find that the proportion of black males using heavy amounts of substances decreases substantially. In the second policy simulation, I consider the effects of decreasing the marginal effect of marijuana use on the probability of arrest of black males, through, for example, legalizing marijuana. This policy change has no effect on high school graduation rates, but I similarly find that decreasing the probability of arrest decreases the proportion of black males using heavy amounts of cigarettes, alcohol, or marijuana.
