Everybody lies? Using Benford's Law to Detect Fraud in Campaign Finance

Mr. Antônio Fernandes
Mr. Dalson Figueiredo Filho

The increasing availability of campaign finance records provides an unprecedented opportunity to social science empirical research. Yet the data is often examined at candidate aggregate level which reduces both the range of statistical tools that can be implemented and the type of research questions likely to be explored. Using one of the largest collections of campaign finance data ever compiled – 234,467,987 records from 543,654 candidates from 2002 to 2018 – we apply Benford´s Law to detect suspicious activities in campaign finance records in Brazil. Following Cho and Gaines (2007, we disaggregated campaign contributions data by state, political party and type of donor which allows a more complete understanding of fraudulent behavior patterns. We show candidates from left parties are less likely to deviate from expected Benford´s distribution. Unlike U.S data, campaign finance records quality is increasing over time. Our two supplementary analyses suggest that data quality are equally suspicious across states and individual level contributions are closer to expected distribution under Benford´s Law compared to contributions from corporations. By demonstrating the utility of Benford´s Law, our work highlight that audit institutions should implement similar procedures to evaluate campaign finance data quality.