BENT — Volume 63, Issue 1

A Nature article criticized scientists' overreliance on p-values in hypothesis testing, pointing out that a p-value of .01 does not mean only a 1% chance of false results, but these statistics deeply influence employment discrimination law. Bent examines how fundamental misunderstandings of p-values and statistical significance have infected antidiscrimination doctrine since the 1970s. The transposition fallacy—misconstruing p-values by ignoring base rate prevalence—affects both lay scientists and legal professionals. Harvard Medical School researchers found 45% of respondents fell victim to this fallacy when evaluating conditional probability: with a 1-in-1000 disease prevalence and a test 100% accurate for positives and 95% accurate for negatives, only 18% correctly calculated that a positive test indicated roughly 2% probability of actual disease. Courts and scholars rarely acknowledge the hidden, unstated assumptions underlying hypothesis tests and p-values in systemic discrimination cases. The Supreme Court's disparate treatment doctrine obscures how statistical significance testing ignores the plausibility of unlawful discrimination as a baseline hypothesis. Bent argues the time has come for a Bayesian revolution in employment discrimination law, bringing prior probabilities into explicit discussion. Part II explores how priors should factor into civil litigation procedure, including dispositive motions and summary judgment, with critical inquiry into legitimate sources of prior assumptions in discrimination cases.

Topics: Labor & Employment · Evidence & Procedure · Legal Theory

Keywords: p-values · statistical significance · employment discrimination · Wal-Mart Stores v. Dukes · Bayesian analysis · transposition fallacy · systemic discrimination