First, identify where the representation of minorities in samples matters—for example, where ethnicity may cause different treatment effects. Second, where the representation of a particular group ...