3.3. Analytic approach
I use multilevel growth models (xtmixed in Stata 10) to estimate healthy behavior trajectories between ages 13 and 24. This approach is consistent with the life course perspective, which emphasizes the importance of modeling intra-individual and cohort trajectories over time
(Elder, 1998). Multilevel models adjust for the clustering of individuals in schools (at level 3), across-person differences attributable to time-invariant traits such as raceethnicity
and gender (at level 2), and individual attributes that vary across repeated observations (at level 1). The unconditional three-level model is as follows: