Analyses were conducted using time-varying effect modeling, a type of non-parametric spline regression that estimates the associations between predictors and an outcome as continuous functions of time (Hastie and Tibshirani, 1993; Tan et al., 2012). In these analyses, our time metric was age (coded to the nearest month). By using data from all 4 waves of Add Health, our sample comprehensively spanned the age range from 12 to 31.
