Chunk #44 — Reasons to be Concerned about the Published cGxE Literature — Problems with the Recipe: Statistical Concerns in cGxE Research — The use of cross-product terms
Statistical tests of cGxE effects often rely on the modeling of a cross-product term in a regression-type model. Valid detection of true interactions in these models requires that factors that could produce spurious interactions be ruled out. For example, when predictors are correlated and quadratic terms are not modeled, the cross-product term can carry the variance of the unmodeled quadratic term and generate spurious interactions (Lubinski & Humphreys, 1990). Moreover, failure to include quadratics can also result in false negative findings of interactions or the reversal of sign of true interactions (Ganzach, 1997). More generally, if the underlying relationship between G and/or E and an outcome is nonlinear (e.g., a spline or higher-order polynomial), misspecification of the analysis by failing to include a term to model the nonlinearity can generate a significant interaction term in the absence of a true interaction.
