Hidden in plain sight? Irregularities on statutory forms and electoral fraud

The most common method of tabulating election results around the world is manually compiling paper forms at the local level. Recent election disputes in developing democracies, particularly in Africa, have centered on irregularities observed on these forms. However, scholars do not yet have a good understanding of the distribution of these irregularities, nor of their relationship to systematic fraud. In this paper, we theorize a catalog of irregularities that goes beyond simple vote tally editing. We use deep neural networks to identify these irregularities on forms from about 30,000 polling stations in Kenya’s 2013 presidential election. We find that although irregularities manifest differently in government and opposition strongholds, they do not correlate with election outcomes, and they are unaffected by the presence of electoral observers. Taken together, our findings suggest scholars of election integrity should pay greater attention to problems of benign human error and overtaxed bureaucrats.