Thomas C. Buchmueller and Smuel H. Zuvekas studied the relationship between drug use and income and employment in order to determine how drug use affects labor market outcomes. In order to improve past studies on the issue, they distinguished between problematic and non-problematic drug use with specific age groups of men, 18-29-year olds and 30-45-year olds, which proved to make a difference. Their results indicate that, for younger men, income is positively related to non-problematic drug use and negatively related to problematic drug use with no relationship between drug use and employment. For the older age group, problematic drug use is negatively related to both market outcomes tested but not for non-problematic drug use.
In their literature review, the authors point out two common weaknesses of past studies done on this topic. The first is that the data set commonly used, the National Longitudinal Survey of Youth, does not include drug use intensity in the survey—it is important to define what makes someone a drug user. The second weakness, also regarding the NLSY data, is that adverse effects of drug use on labor market outcomes may not show up immediately. They use the National Institute of Mental Health’s Epidemiologic Catchment Area (ECA) Survey, which assesses the prevalence of psychiatric disorders including substance abuse in the general population. One thing they did that I didn’t think to do was to create categories of drug use levels. For example, nonusers and experimental users are in one group if they have used drugs five times or less in their lives.
The authors set up a regression for income using the log of annual income. They also use indicator variables to identify non-problematic and problematic drug users. They use the grouped data model, “which is similar to an ordered probit model but the categorical cut-points are directly observed.” Although, they point out that it is more difficult to interpret the coefficient estimates in this model. To make things easier, they used other researcher’s data and used the midpoints of the income intervals to use in an OLS regression. They also point out that there is potential endogeneity of drug use which could lead to two sources of bias. One could be a direct effect of income on drug consumption and another is unobserved variables. They suggest looking at other studies to help figure out the direction of such a bias, since endogeneity is difficult to rule out. For employment, they use a reduced form probit model where the dependent variable takes on 1 if employed and 0 if unemployed. They state that OLS has understated the negative employment effect of problematic drinking, which could have the same implication for problematic drug use. For their control variables they used: age, ethnicity, marital status, education, ECA site, survey year, and dummy variables for industry and occupation.
They mention that the variables marital status and educational attainment could be influenced by drug use themselves. Past researchers have tried to prevent this from throwing off their results by estimating models with and without controls for the two variables. For the employment regressions, dropping the controls for these two variables increases the negative effect of problematic use. Indicator variables for other types of mental illnesses, such as alcoholism, have been included in past regressions in order to confirm that the drug variables aren’t picking up the negative effects of these illnesses.
Buchmueller, Thomas C., and Samuel H. Zuvekas. “Drug Use, Drug Abuse, And Labour Market Outcomes.” The Economics of Health Behaviours. Volume 2. 468-484. n.p.: Elgar Reference Collection. International Library of Critical Writings in Economics, vol. 223. Cheltenham, U.K. and Northampton, Mass.: Elgar, 2008. EconLit. Web. 24 Oct. 2012.