Determining resource requirements for elections using indifference-zone generalized binary search

Abstract

Election lines are more than a nuisance. In recent elections, needing to wait in lines deterred hundreds of thousands from voting and likely changed the winner in multiple cases. Part of the challenge is that even after the voter reaches the front of the line in some locations, it can require more than ten or twenty minutes to cast a ballot. Moreover, the ballot in one precinct might be twice the length of another precinct's in the same county because of additional referenda and levies. We consider the decision problem faced by election officials and other leaders: How many resources (poll books, workers, booths, or voting machines) should be allocated to each location so that lines can reasonably be expected to be minimal? We formulate a simulation optimization problem to identify the combinations that minimize resource requirements while guaranteeing acceptable user-defined service levels. We propose an Indifference Zone Generalized Binary Search (IZGBS) method with rigorous assurances on solution quality, and demonstrate it using the 2016 presidential election. In that election, we describe how the methods helped to reduce waiting times by more than three hours in at least one location affecting thousands of voters and likely increasing turnout.