Fraudulent Democracy? An Analysis of Argentina's Infamous Decade using Supervised Machine Learning

In this paper we introduce an innovative method to diagnose electoral fraud using vote counts. First, to circumvent data availability problems and to study a particular type of fraud, we create synthetic data using Monte Carlo methods. Next, we build a supervised machine learning tool and use a Naive Bayes classier to distinguish between Benford and Benford-deviant data sets. To illustrate our technique, we examine elections in the province of Buenos Aires (Argentina) between 1931 and 1941, a period with a checkered history of fraud. Using a novel dataset of district-level vote counts, our results corroborate the validity of the conventional wisdom: Conservative manipulation of the electoral process, rather than changes in voters' preferences, led to the dramatic electoral shifts during this period. More generally, our endings indicate that Benford's Law can be an exective tool for identifying fraud, even when minimal information (i.e. electoral returns) is available