Keywords: Generalizability, Inverse Probability Weighted Estimators, Randomized Controlled Trial
Because the distribution of treatment modifiers in a population may differ from that observed among randomized trial participants, trial results may not directly reflect the average treatment effect that would follow real world adoption of a new treatment. Much interest has recently arisen around using reweighting methods to more appropriately generalize trial results to real populations. We propose and compare methods for estimating the variance of these generalization estimators, including both closed-form and resampling based approaches. We pay particular attention to how their performance relates to the nature of the target population. We make recommendations for practice based on simulation results and a real data demonstration.