Duke 2008 | ASHE | Equity and Efficiency in Health and Healthcare

Although peer effects may be important determinants of adolescent health outcomes and a useful policy target, identifying causal relationships is especially difficult. We examine a novel approach exploiting variations in grade configurations of schools, with the aim of inferring the effect of greater exposure to older schoolmates (who have higher levels of risky behaviors) on an individual's risky behavior.

Schools attended by adolescents in the United States vary in their grade configurations across districts and over time. Most notably, during the last 40 years, junior highs (grades 7-9) have been replaced by middle schools (grades 6-8) in most American school districts. In addition, a nontrivial proportion (over 10%) of public school districts include combined schools with grades 7-12. The grade composition of one's school has significant implications for the peer environment, because risky behaviors – such as smoking, drinking, marijuana use, and sexual intercourse -- increase steeply by grade. In addition, research in social psychology and sociology suggests that older peers are especially influential for risky behaviors among adolescents.

We use two nationally representative, longitudinal data sets: the National Educational Longitudinal Survey (NELS), which follows a cohort, beginning in eighth grade, from 1988 to 1992; and the first two waves of Add Health, which follow several grade cohorts (in 7th-12th grade) between 1994 and 1996.

In contrast to most existing work on peer effects, the proposed natural experiments have a transparent source of variation in peer environments. The interpretation of our results as causal estimates depends on potential differences across school types in student-level and school-level characteristics that may influence risky behaviors. We find that middle schools and junior highs are nearly identical in a range of observable characteristics, except that middle schools are somewhat more likely to be suburban. Combined schools (grades 7-12), by contrast, make for a less clean comparison with middle schools, as the former type has a higher proportion of white students, lower enrollment per grade, and higher likelihood of being rural. To help account for these differences, we include a rich set of student-level and school-level covariates in our analysis, and we examine difference-in-differences across school types and grade levels. In addition, we discuss carefully a range of potential unobserved factors, based on a historical examination of how and why grade span configurations have changed over time.

Our preliminary results indicate that attending a school with older students is associated with a modest increase in smoking, drinking, and marijuana use, but not sexual behavior, for female students, but not males. For example, between 7th and 8th grade, females in combined (grades 7-12) schools are more likely to start smoking, drinking, and using marijuana than females in standard grade configurations (grades 6-8, then 9-12), whereas between 8th and 9th grade the reverse is true. The higher apparent influence on females, as compared to males, is consistent with the fact that females are more likely to report having close friends in older grades. We are currently investigating the robustness of these associations to alternative specifications