Research suggests intentional voter fraud, for example through false or duplicate registration, is very rare in the U.S. today.
But unintentional administrative errors can also threaten an accurate election and undermine public confidence in the voting system. One way election officials weed out errors and identify potential cases of registration fraud is through advanced data analysis using machine learning.
Detecting fraud in voter rolls
As illustrated in the video above, the Caltech/MIT Voting Technology Project (VTP) has developed computer algorithms to monitor for potential fraud or error on a daily basis. These algorithms are currently being employed in California's Los Angeles and Orange counties.
Post-election, the VTP analyzes and graphs voter turnout across precincts and compares the results to typical turnout patterns. In a 2018 analysis of Orange County turnout data, patterns differed from what was expected in a small number of cases. These cases were then examined and determined to be the result of reporting or administrative errors, not fraud.
Caltech created a new data-driven method to test the integrity of a voter-registration system. Using algorithms, researchers can determine whether registration data is complete and accurate, whether voter rolls include large numbers of duplicate records, and whether voter data has been tampered with or manipulated.
Elections for all public offices are administered by local election officials, which means procedures vary between states and can even vary between counties.
Our vision is to have all states upload voter data on a daily basis and to have algorithms monitor their integrity.
Professor of Political and Computational Social Science