Vote Buying Detection via Independent Component Analysis

Antonio Neme, Omar Neme

Abstract: Electoral fraud can be committed along several stages. Different tools have been applied to detect the existence of such undesired actions. One particular undesired activity is that of vote-buying. It can be thought of as an economical influence of a candidate over voters that in other circumstances could have decided to vote for a different candidate, or not to vote at all. Instead, under this influence, some citizens cast their votes for the suspicious candidate. We propose in this contribution that intelligent data analysis tools can be of help in the identification of this undesired behavior. We think of the results obtained in the affected ballots as a mixture of two signals. The first signal is the number of votes for the suspicious candidate, which includes his/her actual supporters and the voters affected by an economic influence. The second mixed signal is the number of citizens that did not vote, which is affected also by the bribes or economic incentives. These assumptions allows us to apply an instance of blind source separation, independent component analysis, in order to reconstruct the original signals, namely, the actual number of voters the candidate may have had and the actual number of no voters. As a case of study we applied the proposed methodology to the case of presidential elections in Mexico in 2012, obtained by analyzing public data. Our results are consistent with the findings of inconsistencies through other electoral forensic means.