An Exploration of the Relationship Between Risky Sexual Behavior and Substance Use by Teenagers and Young Adults
- Presenter:
Chair: Sara Markowitz; Discussant: Sara Markowitz Mon June 5, 2006 13:45-15:15 Room 335
We evaluate the extent to which the relationship between the use of such substances as marijuana and alcohol and various aspects of sexual behavior is causal. That is, does the use of marijuana and alcohol cause young people to initiate sexual intercourse at an earlier age, to be more likely to engage in sexual intercourse in the past month or past year, to be less likely to use condoms or other methods of birth control, and to have had more sexual partners? Establishing a causal effect of substance use on sexual behavior is essential to the design of effective public policies targeted at improving public health by affecting sexual behavior. A simple statistical association between substance use and sexual behavior is not sufficient evidence to determine causality because the observed relationship between substance use and sexual behavior may reflect causality in both directions (i.e., structural endogeneity) and may also reflect the influence of an omitted “third variable” (i.e., statistical endogeneity)—for example, “a thrill-seeking personality.” Thus, to obtain policy relevant causal estimates of the effect of substance use on sexual behavior, it is necessary to use statistical procedures that address these two sources of endogeneity. In previous research economists have used substance use control policies as instruments for use, but these attempts suffer from biases due to weak instruments, especially since the control policies do not vary much within areas over time.
Using panel data from National Longitudinal Survey of Youth 1997 with four observations on each person in the period from 1997 through 2000, we take two new approaches to establish causality. The first is a Granger causality model. The idea here is to see whether past substance use influences current sexual behavior, with past sexual behavior held constant. Conversely, we also regresses current substance use on lagged substance use and lagged sexual behavior. We take first differences to eliminate unobserved individual-specific and time-invariant fixed effects. Since first differences of lagged substance use and lagged risky sex are correlated with the disturbance term, consistent estimates of the first-differenced equation cannot be obtained by ordinary least squares. Therefore, we employ an instrumental variables estimation methodology that has been developed for panel data. In this methodology, second lags and longer lags of levels of substance use and sexual behavior serve as instruments for the lagged first differences. Unlike substance use control policies, these lags have significant explanatory power. Our second approach allows current substance use to have an impact on current sexual behavior. The equations specified in this model omit the lagged dependent variable as a regressor. Once again, first differences are obtained to eliminate fixed effects, and second and longer lags of levels of substance use serve as instruments for the lagged first difference.
