Following the 2024 presidential election in the United States, an organization calling itself the Election Truth Alliance materialized with claims of election fraud concentrated in swing states won by the Republican candidate. Obviously, this created a lot of noise in certain online circles who felt vindicated by the seemingly sophisticated statistical analyses.
Carter Walker and Jessica Huseman investigated these claims for Votebeat, focusing on Pennsylvania. By examining the underlying statistical model, they show that the strength of the claims far outweighs the strength of the evidence. The most damning detail: researchers tested one of the models on simulated election data containing no fraud, and it nonetheless detected roughly the same amount of “fraud” supposedly found in Pennsylvania.
Always run a negative control before making strong claims with serious consequences in the real world!
Hat tip to Andrew Gelman.