Abstract
Adjustment for covariates has been underused in trials intended to support approval of medical products. This underuse results partly from confusion about whether adjusting changes the thing being estimated or only the estimate of it, and partly from misplaced concerns about assumptions of the analysis. If the thing being estimated is carefully defined, it is seen that adjustments can be used with minimal assumptions. Thus, the benefit of increased efficiency can be realized with essentially no cost in robustness or interpretability.