Figure 2.

Schematic representation of the multiple mediation models. We fit a separate model for each predictor-outcome pair sharing a significant bivariate correlation. The paths marked a_1–4 represent the effects of each predictor on each of the four mediators. The paths marked b_1–4 represent the effects of each of the four mediators on each outcome variable. The path marked c’ represents the controlled direct effect of each predictor on each outcome. All effects were estimated by conditioning on relevant confounders.