Table 1 shows the change in model fit (calculated by subtracting the −2 loglikelihood of the full model, with all parameters specified, from the −2 loglikelihood of a model where a parameter is constrained). The table also includes raw (un-standardized and un-squared) parameter estimates – estimates were not bounded and hence, can be positive or negative, however those estimates that include zero within their 95% confidence limits were not statistically significant. In both sexes, lower age at 1st drink was associated with increasing AD symptoms (bmen = −0.0293; bwomen= −0.0412) – this B parameter serves two purposes – it controls for rGE and its magnitude and statistical significance represent the extent to which age at 1st drink is phenotypically associated with AD symptoms. After accounting for rGE, there was statistical evidence for moderation of A and E. As shown in Table 1, when X, the path representing moderation of genetic influences on AD symptoms by age at 1st drink, was set to zero, there was a substantial deterioration of model fit. This statistically significant change in model fit was also
