To model the heterogeneity in the longitudinal patterns of depressive symptoms, a set of multiple-indicator (aka second-order) growth models were applied to these data, with each triplet of continuous scales forming the indicators for an underlying depressive symptoms latent trait at that time point (Fig. 1). Comparisons were made between a linear growth model (IS where I represents the intercept and S represents the slope), a quadratic growth model (ISQ where Q represents the quadratic) and a revised quadratic model with no quadratic variance/covariance terms (ISQ*). Summary fit measures were assessed along with residuals for the mean and covariance structures. Age was parameterized such that the intercept corresponded to symptoms at age 13 (hereafter referred to as baseline). Inclusion of actual age at response had no effect on the estimated growth models so models shown have a fixed age of response at each time point. All models were run separately by sex as the development of depressive symptoms through adolescence has previously been shown to differ between males and females [23, 24]. Formal statistical tests were used, where appropriate, to
