Theoretically driven multiple mediation models (see Fig. 1) were tested using Mplus, Version 6.1 (Muthén and Muthén, 1998–2011) to assess the effects of early childhood temperament on adolescent alcohol use/problems, both directly and indirectly through personality attributes and peer characteristics, in order to test the intermediary role of personality/peers in links between early temperament and later drinking behaviors. Alcohol and temperament factor scores were calculated using the full sample with mediation allowed to vary for males and females. Covariates, as detailed above, were also included. In all models, path coefficients and standard errors were estimated while accounting for nonindependence of observations due to complex sampling (i.e., individuals nested within sibships). A robust maximum likelihood estimator was employed to accommodate missing data. Mplus applies the product of coefficients strategy (Mackinnon et al., 2002; Preacher and Hayes, 2004) in the assessment of indirect effects. In the simple case, partial mediation by a single variable (i.e., a partial indirect effect) is evaluated in relation to the Z-distribution, with the ratio of the product of the component path coefficients (i.e., the a and b
