We used age as the metric of time, and polynomial age terms (ie, linear, quadratic) were included to describe the linear and quadratic change in heavy episodic drinking over time. The categories at the tail ends of the age spectrum were collapsed because of small cell sizes, such that they ranged from 12 (and below) to 32 (and above). Heavy episodic drinking was log transformed to adjust for skewness. Mixed effects models were conducted using the lme4 package38 for R.39 Because COGA included related individuals, we fit 2-level models (observations nested within individuals over time) and robust standard errors to adjust for familial clustering.
