Covariance adjustment and other model-based adjustments are used, for different tasks, in both randomized experiments and observational studies. We ask less of model-based adjustments in randomized experiments; so, we have greater reason to expect success in this smaller task. In an observational study, the distribution of observed covariates x may be different in treated and control groups, so in certain regions of x, there may be few observed rT’s, and in other regions of x, there may be few observed rC’s. It can be difficult to recognize that a model does not accurately predict rT in a region of x where there are few rT’s. Of course, the same is true for rC. One of Donald Rubin’s examples is recalled to illustrate this issue.