Executive Summary:
In 2022, the State of Georgia experienced significant voter wait-time issues throughout the election season. Since then, Georgia has been front and center in the national conversation about election integrity, as legislators try to adapt voting systems to ever-changing technologies and cultural norms. Georgia law generally allocates election resources based on the number of registered voters a precinct serves, but size is only one differentiating factor among precincts. Voters across the state behave differently, so different precincts of similar sizes might require different resources depending on when voters arrive. Using simulation optimization based on voter arrival data obtained from the Georgia Secretary of State, we have found more efficient and equitable allocation methods, compared to simply considering registered voters. By considering the variation between precincts, election administrators can better safeguard the integrity and equality of elections without sacrificing efficiency.
Research Motivation:
Maximizing election integrity and voter enfranchisement, while decreasing voter wait times, is an important goal of election administrators throughout the United States. However, with a limited budget and a scarce number of voting resources, election officials must carefully manage how election materials are allocated and purchased to maintain fair and efficient elections without excessive spending.
In the State of Georgia, the topic of election efficiency has become increasingly important considering record voter turnouts and long waiting times to vote in recent years (Alexander and Fields 2022). Additionally, the topic of equity among Georgia voters has become a critical issue amid discussion of whether long voting lines are disproportionately impacting specific geographic areas or demographic groups in the state. Fowler (2020) argues that voters from marginalized groups in Georgia experienced abnormally long waiting times in the early voting season of 2020, in part due to a scarcity of voting locations which could not accommodate the high number of early voters. To make matters worse, between 2012 and 2020, the average number of voters assigned to a given voting location increased approximately 47%, from 2,046 voters per location to 3,003 voters per location statewide (Fowler 2020). Some counties experienced dramatically greater increases in assigned voters per location than others. Since 2012, Georgia has faced significantly higher voting wait times than the national average (MIT Election Data and Science Lab 2020). In the same time frame, Georgia has also experienced an exceptionally large voter turnout. In 2022, Georgia ranked first in the southeast region and 17th nationally in overall voter turnout for the midterm election (Fabina 2023). With this combination of high voter turnout and long waiting lines in the state, it is important to examine what resources election officials can use to improve the efficiency of the Georgia elections system.
The Georgia Secretary of State’s Office is responsible for purchasing voting machines and providing this equipment to county leaders. Georgia law then requires these county-level election officials to adequately provide voting machines to precincts where voters are assigned (GA Code § 21-2-367 2022). This law provides general guidelines for provisions of voting materials based on the number of registered voters assigned to a precinct, stating that county election officials must provide to each precinct “at least one voting booth or enclosure for each 250 electors therein, or fraction thereof.” However, this law does provide allowances for greater or fewer machines being allocated if election officials determine there to be “relevant factors that inform the appropriate amount of equipment needed.”
The goal of these election officials is to work within existing policy guidelines to properly allocate resources efficiently and equitably. However, to create a system that is both efficient and equitable is incredibly difficult because of the complex and oftentimes competing processes needed to achieve both. Creating an efficient system focuses on maximizing throughput, while equitable systems aim to achieve similar times for all voters within a system (Bertsimas et al. 2012). Thus, our goal in creating a simulation is to balance these factors to create an optimal allocation which is both efficient (with adequate voter throughput) and equitable. Our research sough to explain how a simple voter-population based policy, like the policy dedicating one voting booth per 250 electors, may not necessarily lead to equitable nor efficient allocations because of the variability that exists within voter populations and voter behaviors between voting locations. In our analysis, we will be examining data from the November 2022 midterm election regarding voter arrival times within Fulton County at all 248 voting locations serving a combined total of 749,780 registered voters. From this data, we created a simulation model to optimize the Georgia elections system. We will use this model to determine the extent to which simulation tools could be used to reduce wait times and increase equity throughout the system. This model was created in Simio, an object-based simulation software used to predict the performance of systems by modeling queues and servers using random variables.
Research Objective:
Our objective was to see if there were a way to, without using any additional resources, increase equity without sacrificing efficiency in our elections.
Research Methodology:
To conduct our research we first used census data and election data from the Georgia Secretary of State to characterize polling places in Fulton County. We arrived at four categories based on the concentration of voter arrivals at a precinct and the turnout at the precinct. With this characterization in mind, we were able to select a subset of 20 precincts that showed the broad range of voter behavior in Fulton County. Then, we created a simulation capable of modeling voter experience at each precinct under different equipment allocations. We compared Georgia 1:250 policy to the performance of heuristic we designed and an optimal allocation we found using Opt Quest, Subset Selection and KN Selection.
Results:
Overall, simulation optimization was able to vastly improve upon the allocation currently employed by Georgia law. Average time in system experienced by an arbitrary voter decreased by 18%, from about 50 minutes to 41 minutes. Ten of the twenty precincts experienced a local decrease in average time in system, with Mountain Park Community seeing the most drastic decrease from 2 hours and 47 minutes to just 11 minutes.
Lebanon Baptist and C. H. Gullat recorded a decrease by about a factor of four while average wait time was cut in half at Saint Phillip. Despite the high magnitudes of decrease in average time in system, the highest magnitude of increase experienced by any precinct was an increase by a factor of .678 at Hoyt Smith. For variables designated as response variables in an experiment, Simio generates a SMORE plot (Simio Measure of Risk and Error) showing the 95% confidence interval for that variable. We designated mean average time in system to be a response variable, and from the SMORE plot were able to reject the null hypothesis that the opt quest and current allocation means were equivalent, with 95% confidence. To evaluate the feasibility of a heuristic that models the optimal allocation, linear regression was used to look for correlations between the number of ballot marking devices allocated to a precinct at optimality and other known variables. , the number of groups of 250 registered voters serviced by each precinct, had only a weakly positive correlation to the number of ballot marking devices given to each precinct at optimality. This is particularly interesting, because this is the quantity the state primarily uses to allocate ballot marking devices. On the other hand, , the ratio of arrivals during the peak hour to average arrivals, had a slightly stronger positive correlation to the number of BMDs given to each precinct at optimality than . While neither correlation is particularly strong, this suggests that the arrival pattern does have some impact on the needs of a precinct. Additionally, it supports that there are other variables impacting a precinct’s needs outside of concentration and number of registered voters. One likely suspect would be turnout, as well as the number of election day voters, but because turnout is so volatile it would be a challenge to accurately forecast it at each individual precinct.
To evaluate the feasibility of a heuristic that models the optimal allocation, linear regression was used to look for correlations between the number of ballot marking devices allocated to a precinct at optimality and other known variables. 𝑅𝑖, the number of groups of 250 registered voters serviced by each precinct, had only a weakly positive correlation to the number of ballot marking devices given to each precinct at optimality. This is particularly interesting, because this is the quantity the state primarily uses to allocate ballot marking devices. On the other hand, 𝑃𝑖, the ratio of arrivals during the peak hour to average arrivals, had a slightly stronger positive correlation to the number of BMDs given to each precinct at optimality than 𝑅𝑖. While neither correlation is particularly strong, this suggests that the arrival pattern does have some impact on the needs of a precinct. Additionally, it supports that there are other variables impacting a precinct’s needs outside of concentration and number of registered voters. One likely suspect would be turnout, as well as the number of election day voters, but because turnout is so volatile it would be a challenge to accurately forecast it at each individual precinct.
Of the 5 heuristics simulated, heuristic 4 stood out the most regarding improvement upon the current allocation. Average time in system experienced by an arbitrary voter decreased by 8 percent and once again 10 out of 20 precincts experienced a decrease in average time in system, at much greater magnitudes than increases at any precinct. That this heuristic was able to improve upon the current allocation at all suggests that there is a relationship between the needs of a precinct and the concentration of voter arrivals at said precinct. From Simio’s SMORE plots, the hypothesis that the mean average time in system of heuristic 4 is equivalent to the current scenario can be rejected with 95% confidence.
Conclusions and Future Work:
We found that considering concentration of voter arrivals could greatly reduce inefficiency and inequity in election systems. Additionally, linear regression run on the optimal allocation and known variables demonstrated that concentration and number of registered voters alone are not sufficient to allocate equipment to a precinct. Rather, there are even more variables the state should be considering.
Further experiments examining the efficiency and equity of election resource allocation would ideally focus on broadening the scope of data examined. Currently, our simulation is limited to just 20 precincts in Fulton County which is a mostly urban region. However, the State of Georgia encompasses a variety of regions including suburban, rural, costal, and mountainous. Voters in each of these areas are likely to follow patterns entirely distinct from those in Fulton County, thus expanding the model to cover the entire state could reveal even more about inequity and variability in voting systems. Additionally, examining voter arrival data from a wider range of election years could reveal insights about the repetition and predictability of voting patterns in precincts. By expanding the scope of the simulation, and the data used to build it, it is entirely possible that better heuristics for allocating election equipment could emerge, drawn from patterns of optimal allocations in the state as a whole.