Figure 1.

Model of discrete-time survival with lagged BMI as predictor.

Note. Among smoking-naïve, we predict cigarette smoking initiation at three yearly follow-ups (T3, T5, and T6). In discrete-time survival modeling, the hazard of initiation is estimated and predicted, assuming initiation did not occur previously. Therefore, lagged BMI is assessed at the time point before risk of initiation. The prediction of the latent variable lets us control for time-invariant effects. BMI = body mass index percentile based on youth of the same age and sex from self-reported height and weight.