Visualizing effect modification on contrasts

Abstract

A recurring problem in statistics is estimating and visualizing nonlinear dependency between an effect and an effect modifier. One approach to handle this is polynomial regressions of some order. However, polynomials are known for fitting well only in limited ranges. In this article, I present a simple approach for estimating the effect as a contrast at selected values of the effect modifier. I implement this approach using the flexible restricted cubic splines for the point estimation in a new simple command, emc. I compare the approach with other classical approaches addressing the problem.