Covariates and outcome adjustments. Prior to model fitting, outcomes were regressed on the following covariates: parental age, child year of birth, number of births, and child sex. Residual scores were then log transformed to correct for positive skew and all variables were standardised prior to model fitting. Transformation of non-normal sum scores to normality improves false positive rates and reduces bias in parameter estimates (Murray et al., 2016). Moderator scores were not transformed as scaling of the predictor has minimal impact on estimates of interactions (Van Hulle & Rathouz, 2015).
