[Replication] Minds, Not Votes: Reproducing and Stress-Testing Racial Diversity in Jury Decisions

Abstract. Karpowitz, Mendelberg, Elder, and Ribar (2024) report that jury racial composition shifts individual private opinions but barely moves group verdicts, because dissenters of color carry less influence than White dissenters. The published individual standard errors are iid, not jury-clustered—the deposited clustering argument is silently dropped by base R—so under correct clustering the "5 Whites" effect survives (p = .005/.016/.042) while the finer "4 Whites" step does not (p = .121/.172). This is the one published distinction that fails under correct inference. Otherwise the three main-text tables reproduce exactly: 158 of 158 cells are byte-for-byte identical to the authors' outputs, including the ρ = 1.424 two-step Heckman moment, genuine and not an error. The "minds not votes" asymmetry survives a 14-test adversarial battery; the group-verdict null does not depend on the contestable Heckman model; and the dissenter mechanism is corroborated (b = 0.513, permutation p = .0098).