Persistence Perpetuated: Response to Joseph Francis

In Voigtländer and Voth (2012), we show that German towns and cities that attacked their Jewish communities during the Black Death in 1348/9 were more anti-Semitic in the interwar period. Francis recodes our sources using large language models (LLMs). He argues that his re-examination shows our results to be fragile and driven by doubtful coding choices. Here, we show that a) even when we take Francis' recoded data at face value, none of our conclusions is overturned. Results remain quantitatively similar, and most individual outcomes remain statistically significant. Crucially, the principal component summarizing overall anti-Semitism in the 1920s/30s (which Francis does not report) remains highly significant under his recoding. Moreover, in our preferred geographic matching specifications - comparing nearby towns with and without Black Death pogroms (also not reported by Francis) - most individual outcomes and their principal component continue to display highly significant persistence, yielding near-identical results despite changes to roughly 8% of the codings. b) In addition, Francis' codings are replete with linguistic and historical mistakes and misunderstandings.