A note on false positives and power in G × E modelling of twin data

Area of Science:

• Behavioral Genetics
• Quantitative Genetics
• Statistical Modeling

Background:

• Gene-environment interaction (GxE) models are crucial for understanding trait etiology.
• Purcell's (2002) variance components models are widely applied for GxE analysis.
• Both bivariate and univariate parameterizations exist, with variance decomposition depending on the moderator (M) and trait (T).

Purpose of the Study:

• To investigate the impact of correlations between moderators (M) and traits (T) on GxE models.
• To identify potential inflation of false positive moderation effects in univariate GxE models.
• To propose and evaluate extensions to univariate GxE models for improved accuracy.

Main Methods:

• Analysis of variance components models for GxE.
• Comparison of bivariate and univariate parameterizations under correlated M and T.
• Development and testing of an extended univariate moderation model.

Main Results:

• The standard univariate parameterization significantly increases false positive moderation effects when M and T are correlated, and M is correlated between twins.
• A proposed extension to the univariate model effectively prevents this false positive inflation, assuming the M-T covariance is not moderated.
• If M-T covariance is moderated by M, univariate results require careful interpretation due to potential conflation of effects.

Conclusions:

• Researchers should exercise caution when interpreting univariate GxE moderation effects, especially with correlated moderators and traits.
• The full bivariate moderation model is recommended for studying moderation on the M-T covariance.
• The extended univariate model offers increased power over the bivariate model when M-T covariance moderation is absent or can be ruled out.