The Arity of Disparity: Updating Disparate Impact for Modern Fair Lending

Plaintiffs often cannot prevail under a disparate impact (“DI”) claim of discrimination unless they show the defendant failed to implement a less discriminatory alternative (“LDA”) to a practice that yields a disparate output across protected classes. Traditional LDA analysis focuses on a singular notion of fairness: parity, or the equality of screening decisions across protected groups. However, recent scholarship highlights that parity is only one of numerous competing notions of fairness that may seem just as compelling as, but be mutually exclusive with, parity. The arity of disparity is larger than DI has acknowledged. We propose formalizing LDA analysis as an explicit constraint on choices over screening models and data inputs. The constraint restricts model-induced disparities in both parity and a competing notion of fairness—accuracy—relative to a “budget” that depends straightforwardly on overall model performance. We also show how this trade-off leads to balancing other notions of fairness, as weighted combinations of parity and accuracy span many other fairness notions in a helpful way. For concreteness, our legal argument and our applied examples focus on DI under the Equal Credit Opportunity Act (“ECOA”). We address tension between DI’s traditional focus on parity and ECOA’s statutory emphasis on “credit-worthy” consumers and discuss implications for new frontiers in credit underwriting, including the use of machine learning and alternative data sources.