Founded Generalizations: Sowell, Disparate Impact, and Statistical Proof

When does a fact about a group support a conclusion about a particular case? This paper answers that a generalization supports such a conclusion only when it is founded, that is, grounded in features that produce the outcome. Thomas Sowell's typology frames the question: his Discrimination IB, reliance on group regularities when individual information is costly, is its evidentiary form. On this standard, Wards Cove's demand for a specific practice and Inclusive Communities' requirement of "robust causality" are foundation requirements. In the vocabulary of the problem of universals that Paper 1 applies to the RICO pattern, a bare disparity is a pattern post rem, made by the comparison rather than found in the practice; Teamsters and Hazelwood show what founded statistical proof looks like. The standard is symmetrical. A defendant's actuarial generalizations are legitimate when founded and not otherwise, and a plaintiff's disparity, however large, cannot by itself show the intent that Feeney requires: action "because of," not merely "in spite of," its effects. The standard also explains why courts refuse liability on naked statistics, as in the Blue Bus case and Cohen's gatecrasher paradox: a bare statistic is founded in nothing that produced this particular outcome. Thomson's account of individualized evidence anticipates that explanation, so the paper's contribution lies chiefly in the symmetry claim and in Sowell's frame. Drafted with the assistance of Claude (Anthropic); the author is responsible for the arguments and citations.