Examination of Alternative Models of GxE Moderation

The Cholesky parameterization has been used to examine the potential shared latent structure of two variables. Several modifications have been implemented to examine gene x environment interactions with the aim of reducing... [ view full abstract ]

The Cholesky parameterization has been used to examine the potential shared latent structure of two variables. Several modifications have been implemented to examine gene x environment interactions with the aim of reducing overall error rates, especially in the case of moderators correlated between twins. We propose an alternative model for consideration when examining the moderating effect of one definition variable on the relationship between two other variables. A model allowing for correlations between the latent genetic and environmental factors of two variables would be appropriate to use when there is no a priori justification for testing one direction of influence over another. Additionally, moderation between the two variables by a third can be examined on each of the variables, as well as on the correlations between their latent variables. This ‘correlated factors’ (CF; bivariate) moderation model does not lend itself easily to comparison with a classic Cholesky (trivariate) decomposition to examine the relationships between 3 variables. Rather, the correlated factors model was compared to the ‘extended bivariate’ (EB) model proposed by van der Sluis et al. (2012); with modification accounting for two variables rather than the original univariate designs. Since it is often the case that moderators of interest are correlated between twins, the alternative model is applied to all models tested during the simulations. Using simulated data (500 simulations; 1000 mono/dizygotic pairs), model fits were compared between CF and EB models. Ability to accurately detect simulated moderation on covariances, means and correlations was tested across all models. Likelihood ratio tests were performed to compare goodness of fit. Overall, the CF model provides a more readily interpreted framework for the study of multivariate GxE interactions.

van der Sluis et al., 2012. A Note on False Positives and Power in G x E Modelling of Twin Data. Behavior Genetics. 42:170–186.