In a 2-arm parallel randomized controlled trial (RCT) an outcome of interest is measured at least twice, once before the treatment is administered and once after. However, to measure the stability of an effect over time, additional time points can be added after the first follow-up. In this presentation, I consider the case where an outcome of interest is measured at a baseline, a second follow-up, and a third follow-up. In this design, it might be of interest to know the treatment effect at the third follow-up that is unconditional on the treatment effect at the second follow-up. We derive such an effect and its standard error and apply the theory to a simulated outcome that is correlated over time.

The utility of the method of Average Marginal Effects in many contexts of statistical modeling makes the lack of accessible resources in the literature surrounding them a tragedy for both statisticians and those who consume statistics. In this presentation, I attempt to solve this problem by deriving the estimate of the AME, and its standard error in the context of a common experimental design; namely, the 2-arm parallel randomized controlled trial (RCT) with heterogeneity of effects by site. We follow each section with straight forward programming techniques to apply this method to real data.