Estimating Peer Influences in Teenage Substance Use when Friendship Links are Unobserved

This paper estimates peer influences on the tobacco, alcohol and cannabis consumption of a school based sample of UK 15 year olds. As is the case for many existing studies of peer effects, friendship links are not explicitly observed in the data; instead each individual reports the extent to which their friends, and they themselves, engage in the different behaviours, but without identifying who their friends are. Estimating peer effects with such data involves a trade-off between peer group relevance and selection bias. For these behaviours, a child’s group of friends is likely to be the most relevant but also the most endogenously selected peer group, whereas a child’s school class is likely to be a less relevant but more exogenously determined peer group. Our data suggest significant peer effects in tobacco, alcohol and cannabis use between friends, using both single equation and instrumental variables approaches. We also find evidence of peer effects in alcohol use between classmates, at least for girls, but no evidence of such peer effects for tobacco and cannabis use.