[Replication] Robust to Reclassification: Replicating the Partisan Turnout Reversal in Ahmed, Hodler and Islam (2024)

Abstract. Every headline cell of Ahmed, Hodler and Islam (2024) reproduces exactly, and the partisan turnout reversal survives an adversarial robustness battery. The original reports that non-partisan get-out-the-vote campaigns in the 2018 Bangladeshi election raised turnout in government strongholds and lowered it in opposition strongholds, tilting the electorate toward the incumbent: the legitimacy treatment moved ink-finger turnout +15.45pp in government villages and −20.68pp in opposition villages, a pooled sign-flip interaction of −36.13 (SE 1.44). The interaction survives randomization inference on the cross-panel difference itself, cluster-robust resampling, leave-one-out checks, multiplicity correction, and three relabeling perturbations: dropping the 12 villages discordant with their 2001/2008 vote-gap sign (−35.8), reclassifying panels by historical vote-gap (−33.8), and strong-versus-marginal splits. The reproduction is exact and the measured interaction is robust; reading the opposition drop as sophisticated abstention is partially identified, because measurement reactivity, preference falsification, and documented booth-level suppression are not separable within the design.