Fair Election as Probatio Diabolica?

• In Event: Poster Session: Political Methodology

Abstract

Election forensics is important since fair elections are fundamental to democracy. Digit-based election forensics (DBEF) employs a chi-square test to estimate whether the distribution of digits calculated from election returns is significantly different from the theoretical distribution of digits (Mebane 2006; Beber and Scacco 2012). Although election forensics is to detect only the “signal” of whether a given election is fraud, several studies assume the significance of test statistics as “evidences” for the existence of fraud (Weidmann and Callen 2013). Moreover, some studies alert us to uncritically use the DBEF because it erroneously signals fraud even though a target election is alleged to be fraud-free (Shikano and Mack 2011). This paper proposes using “divergence” instead of utilizing the significance of test statistics. Divergence employed in information geometry is to measure the difference between two probability distributions. From the Bayesian-like perspective, we perceive that a given election is more likely to be fraud when divergence of this election is larger than before. On the other hand, we can judge elections as non-fraud when we believe that the elections are stably fair, even though divergence is large and stable. I mathematically and empirically discuss this geometric interpretation of the DBEF using divergence.

Author

• Yuta Kamahara, Yokohama National University