Chunk #51 — Reasons to be Concerned about the Published cGxE Literature — Problems with the Recipe: Statistical Concerns in cGxE Research — Power to detect and characterize different types of interactions
Recently, techniques for estimating the standard error of the cross-over point have been proposed which could, in principle, allow stronger inferences about the actual form of the interaction (Widaman et al., 2012). Alternatively, establishing regions of significance around each regression line using standing approaches (e.g., Johnson & Neyman, 1936) to characterize where two slopes overlap and where they do not could also be used to increase confidence that an ostensible cross-over shows a desired degree of statistical differentiation from an ordinal interaction. Such approaches, in principle, could provide greater confidence in believing a true “cross-over” has been detected. Consistent with the other points made above, such approaches are dependent upon being confident that the interaction is not an artifact of scaling, is not caused by (unmodeled) nonlinearity, and is not a Type 1 error.
