In some cases we can control for confounding even when we do not have data on a sufficient set of deconfounders. In these cases we can use different adjustment formulas—not the conventional one, which is only appropriate for use with the back-door criterion—and still eradicate all confounding.
