Three Cholesky decomposition models (one for each environmental measure) were employed to assess whether the heritability of each of our environmental measures was accounted for via their association with the behavioural phenotypes (see Fig. 1). Environmental measures were entered as the final variable in the model (the variable to the far right in Fig. 1). This meant that variance in the environmental measure that is shared with each of the behavioural phenotypes would be accounted for. As such the final A (or C or E) factor in the model would comprise variance unique to the environmental measure. If the path estimate for this factor was greater than zero then this would indicate that our behavioural phenotypes had not accounted for all of the variance in our environmental measure. We were also interested in the nature of bivariate associations within the multivariate model (e.g. whether life events had a stronger genetic correlation with depression or delinquency). We therefore transformed the Cholesky decompositions into the more easily interpretable correlated factors solution (Loehlin 1996) and report bivariate genetic correlations and factor loadings. We report estimates of residual genetic variance from the Cholesky decompositions in the text.
