Background A large literature exists suggesting that amongst adolescents a greater number of transitions through housing and neighbourhood environments (residential mobility) leads to increased participation in risky health behaviours such as cannabis use. However, previous research has failed to adequately account for underlying selection differences between mobile and non-mobile adolescents. Given that these groups differ across a range of characteristics these findings may therefore be subject to considerable selection bias. The aim of this study was to determine if the relationship between residential mobility and cannabis use in adolescence are spuriously driven by selection effects and therefore if conventional models produce biased results.

Methods In this study we utilise data from 4767 adolescents in a UK cohort, The Avon Longitudinal Study of Parents and Children. First we employ a multilevel version of a conventional logistic regression analysis to determine the extent to which adolescent residential mobility is associated with cannabis use. Secondly, we utilise a between-within multilevel model to decompose this relationship into between-child and within-child effects in order to determine the extent to which cannabis use in adolescence is driven by underlying selection differences between mobile and non-mobile adolescent groups (the between-child part) or by changes in residential mobility for individuals (the within-child part). All analyses are conducted in Stata and MLwiN.

Results Consistent with previous research we find a positive association between cumulative residential mobility and cannabis use with conventional logistic regression models (Odds Ratio: 2.56, 95% Confidence Interval: 1.20–5.78), implying that adolescents who experience a residential change are more likely to use cannabis than those who remain residentially stable. However, decomposing this relationship into its between- and within-child components reveals that a conventional model is underspecified and misleading; we find that differences in cannabis use between mobile and non-mobile adolescents are due to underlying selection differences between these groups (between-child log odds: 3.56, standard error: 1.22), not by a change in status of residential mobility (within-child log odds: 1.33, standard error: 1.02).

Conclusion Our findings suggest that transitions through housing and neighbourhood environments in adolescence do not have a causal effect on cannabis use, but that the two are spuriously driven by selection effects. A number of limitations must be acknowledged; only a small number of variables could be decomposed into between- and within-effects; school moves were unavailable; and reason for moving was unknown. Future research into differences in health behaviour participation between groups should take account of selection bias in order to determine accurate effect estimation.