Modeling Voting Service Times with Machine Logs

Abstract

We propose a new method for estimating voting service times using machine log files from electronic voting machines, particularly the iVotronic DREs used in South Carolina. Estimating voting service times requires both parsing the relevant information from the files and estimating when machines are being used to capacity in the precincts to which they are deployed. After estimating average voting service times for each precinct in South Carolina in the 2016 presidential election, we conduct statistical analysis to explore correlates of service times with ballot features (number of candidates and offices), ballot behavior (straight-ticket voting and ballot roll-off), and demographics (age and race). This analysis is potentially helpful for both reaching a more nuanced understanding of why voting times vary, and for providing election administrators with a method for estimating ballot-marking times more comprehensively than previously possible.

Keywords: voting, wait times, voting machines, ballots, ballot marking