Voter fraud is an important problem, and it knows no geographical boundaries. In the last two months alone, allegations of voter fraud made the news. Some 1,000 citizens of Beit Shemish, Israel protested in demand of new elections amid evidence of foul play in an earlier vote. A Florida congressman’s former chief of staff was sentenced to prison for trying to rig an election. Similar stories also came out of Wisconsin, Texas, Michigan, Nebraska, Kansas, North Carolina, New York, Iowa, Egypt, Japan, and New Zealand, to name just a few from just the last few weeks.
Voter fraud is an important problem. When ballot results are viewed as corrupted by some process, elections have the potential to be destabilizing, and in extreme cases, can trigger violence and political regime change. For example, the suspicions surrounding the results of the 2007 Kenyan presidential election triggered large-scale turmoil, leading to more than 1,000 deaths. Observations from international election monitoring in 170 countries indicate 61% of countries experience some degree of cheating, including election fraud, with 27% countries in the sample exhibiting major fraud problems.
But how can we measure it? And how can we give election monitoring agencies the tools they need to pinpoint potential hotspots before an upcoming election? Researchers Walter Mebane and Kirill Kalinin at the University of Michigan have refined a set of tools that capture different aspects of potential election irregularities. Some of these methods are informed in part by techniques developed for detecting financial fraud: numbers changed by humans tend to have patterns that wouldn’t have occurred through the normal process of casting a ballot. Other methods derive from other mathematical regularities or feature simulations of what votes affected by fraud look like. Such methods have revealed fraud in recent elections in Russia, Iran, and Uganda. However, vast amounts of election returns are needed to conduct the data analysis, and many of the tests of election fraud have relied on precinct or polling station results, which can be difficult to get from governments, especially ones under suspicion of engaging in fraudulent behavior.
Enter the Constituency-Level Elections Archive (CLEA). Three of the co-directors of CLEA (Ken Kollman, Allen Hicken, and David Backer) are teaming up with Mebane and Kalinin to provide high quality and detailed constituency-level election results. Together the researchers are testing whether tools developed for detecting fraud using precinct results will also work at a higher level of aggregation, namely the district-level. CLEA provides cleaned and uniformly formatted results for elections around the world. Preliminary results from recent elections in Russia, Uganda, and Mexico suggest that different fraud detection techniques accurately estimate the probability of suspicious behavior at the district-level. For instance, in Russia, serious irregularities were widely alleged by both election monitors and Russian voters. Mebane and Kalinin found significant and substantial evidence of fraud using a range of fraud detection techniques, thus corroborating the firsthand account of monitoring organizations and Russian voters.
Next steps include applying these techniques to the remaining 1200+ elections in the archive with the end goal of providing baseline estimates of likely cases of fraud. Armed with this information, election monitoring organizations will have a sophisticated tool to complement their efforts at promoting free and fair elections. Over the short run, such results can assist monitors in focusing attention on problematic areas in specific countries, but in the long run, these detection techniques have the potential to deter those who wish to engage in election manipulation.