Constraining Localized Vote Tampering in the 2020 US Presidential Election

TL;DR

This paper develops a statistical regression-based method to detect localized vote tampering in elections, applied to the 2020 US presidential election, effectively ruling out some fraud scenarios and constraining others.

Contribution

It introduces a novel regression analysis approach to identify and limit localized vote tampering, enhancing election security analysis.

Findings

Successfully detected artificial fraud signals in some cases

Rigorously constrained certain localized fraud scenarios

Achieved detection uncertainties at the few-percent level

Abstract

Voter fraud in the United States is rare and the vote-counting system is robust against tampering, but there remains widespread distrust in the security of election infrastructure among the public. We consider statistical means of detecting anomalous election results that would be indicative of large-scale fraud, focusing on scenarios in which votes are modified in in a localized setting. The technique we develop, based on standard regression analysis, makes use of the fact that vote share is correlated with demographics. We apply our method to the results of the 2020 US presidential election as a proof-of-concept, resulting in uncertainties at the few-percent level. We are able to readily detect an artificial signal of such fraud in some cases, ruling out some scenarios of localized fraud and placing constraints on other scenarios.