Learning from Recounts

We compare the results of two recent statewide recounts in Wisconsin — the 2011 Supreme Court election and the 2016 presidential election. We argue that recounts provide a valuable window into the accuracy of initial vote counts. Recount accuracy is best measured in terms of the absolute deviation between initial vote counts and recounted vote counts. We show that the accuracy of this assessment declines as the level of aggregation in comparing vote counts increases. In the particular case of Wisconsin in 2011 and 2016, we find that: (1) simply comparing the recounted victory margin with the election night victory margin significantly understates the degree to which counting errors occurred in the original vote count, (2) aggregated reporting vote totals understate the number of errors made in the original count of ballots, (3) recounts reveal far more errors in counting ballots of minor-party and write-in candidates than for major-party candidates, and (4) ballots originally counted by computerized means appear to be at least as accurate as ballots originally counted by hand.