Skip to main content

How Data Can Solve the Elections Resource Allocation Problem

Both the beauty and the challenge of United States elections lies in their versatility. This country is home to thousands of local election jurisdictions, each with distinct electorates and diverse approaches to election administration. A one-thousand-person township in central Maine has the same goal as a multimillion-person county on the coast of California: to provide a free and fair election to every eligible voter. While voters in one jurisdiction may prefer to vote entirely by mail, voters in another could lean heavily towards in-person options. Another jurisdiction might see voters evenly divided between mail, early, and Election Day voting, creating a logistical nightmare for election officials who must balance limited resources between each voting method.

The multitude of approaches to election administration at the local level creates a problem for policymakers striving to set uniform standards for how election resources should be allocated. Is it better to pour resources into mail or in-person voting? If we put more resources towards Election Day voting, can we reduce the number in-person early voting sites in exchange? Should every jurisdiction have the same voting methods available?

In the popular media, questions about voting options have been decoupled from core values and entangled with partisan rhetoric. While one side of the political spectrum seeks to expand all voting options in a kitchen-sink, H.R. 1-style approach, the other seeks to roll back the progress toward mail voting made during the last election. This problem is exacerbated by the sheer difficulty of measuring what qualifies as a sufficient number of voting options. It is easy to assail a reduction in voting sites as an anti-voter attack when it’s not clear how many voting sites are actually needed.

The all-or-nothing dichotomy forgets the most important tenet of good election policy: leading with the core value that elections should be free, fair, and widely accessible to all eligible voters. As a growing number of state legislatures consider new elections legislation, it is critical that we refocus our attention—as policymakers, as citizens, as members of the media—on implementing sustainable voting reforms that both capitalize on the progress already made and provide a meaningful path forward.

This report of the Bipartisan Policy Center strives to do just that. Recognizing the need for a baseline of minimally acceptable voting options, as well as the need to more efficiently distribute limited elections office resources, we propose a versatile Resource Estimator that determines how resources should be allocated based on a jurisdiction’s voter preferences and election laws. Understanding that no single allocation of voting options could suit every jurisdiction, the Resource Estimator acknowledges the need for both standardization and adaptability. It offers a standard formula for allocating resources that applies across the country; this standard formula is then adapted to a jurisdiction’s individual circumstances.

The Resource Estimator serves two purposes. First, it helps policymakers and election officials efficiently allocate resources and manage trade-offs between voting options. Second, it aids the public, academics, and the media in parsing out when an adjustment to the number of available voting options is a detriment to voter access, and when it’s just an administrative shift in resources that rebalances the share of voting options while preserving a voter’s access to the ballot.

We are not so naïve as to believe that all jurisdictions have good intent when rolling back voting options; voting rights must be ardently protected and defended against draconian, incendiary policy moves. But we also acknowledge that election officials need a way to allocate resources (and, occasionally, reallocate resources away from one voting option and towards another) that is efficient and evidence based. Rather than give policymakers cover to reduce voting options, our Resource Estimator actually gives voter advocacy groups a basis to advocate for expansionary yet commonsense reforms that best meet voters’ needs.

Read Next

The Data-Informed Solution to Resource Allocation

An April 2021 Bipartisan Policy Center and Morning Consult survey found that 80% of registered voters—91% of Democrats and 72% of Republicans—were satisfied with their voting options in 2020. Despite all the controversy surrounding the last election, election officials successfully met their voters’ needs, and by an overwhelming majority.

The current distribution of voting options in the U.S. is not in crisis (as popular debate might suggest), but it’s not perfect either. Election officials are too often under-staffed and resource-constrained. More than $400 million in private funding proved critical to making the 2020 election happen without major administrative crisis; countless election officials reported not knowing how they would have managed on government resources alone.

American elections should not be reliant on private philanthropy. The simple solution is for the federal government to enact regular elections funding for state and local jurisdictions. Given that this regularized federal funding has so far been politically infeasible, election officials must figure out ways to more efficiently distribute the limited resources they have now.

Solving this resource problem requires an appreciation for the unique needs of each jurisdiction. States with universal vote-by-mail have different resource needs than states without, as do states that offer early in-person voting compared to those that don’t. Our Resource Estimator not only accounts for this nuance but proposes multiple resource allocation options specific to the policy framework of the jurisdiction in question.

The Resource Estimator inputs a jurisdiction’s voter participation rates and mail voting policies to calculate the minimum level of early in-person and Election Day voting sites and poll workers needed. The Bipartisan Task Force on Elections urges all jurisdictions to provide no-excuse absentee voting and at least seven days of early in-person voting during every election. We uphold this recommendation in this report, while also acknowledging the need for efficient resource allocation even in jurisdictions that may fall short of these standards. Our Resource Estimator applies to all jurisdictions regardless of mail or in-person early voting policy. Incorporating the full extent of election administration policies in place into the Resource Estimator promotes optimal accuracy and adaptability.

We propose the Resource Estimator with a brief note of caution. Metrics such as these hold promise to standardize elections and promote voter confidence, but too often with rules-based policy those performing above the minimally-required level fall to the floor instead of continuing to rise above it. The Resource Estimator is intended to help jurisdictions manage hard trade-offs when allocating resources, not give states or localities an excuse to cut back voting options already being provided. Furthermore, we are wary of portraying our findings in a way that promotes false precision, or the tendency of numerical models to produce a level of specificity that is not warranted. Our model produces a point estimate for the number of voting sites and poll workers a jurisdiction needs. While this estimate is backed up by statistical models, it would be impossible to predict exact distributions suitable for every jurisdiction. Election officials should use the Resource Estimator to inform resource allocation questions, but it is critical that it be coupled with their own knowledge of their own jurisdiction’s specific needs and circumstances.


This report uses data from the Election Assistance Commission’s Election Administration and Voting Survey (EAVS). We merged the 2016 and 2018 standard EAVS survey with the 2018 EAVS policy survey to identify key trends in the allocation of voting options. From those trends, we developed a set of formulas that determine the minimum number of early voting and Election Day voting sites and poll workers a jurisdiction needs based on their voters’ preferences and mail voting policies. We applied Ordinary Least Squares (OLS) regression analysis to identify statistically significant relationships between voting options and voter behavior, ultimately proposing an approach that is both reliable and accurate, standardized and adaptable. A detailed description of our methodological approach can be found in the appendix.

The Resource Estimator relies on several inputs to make its calculations. Election officials using the Estimator should gather the following information:

  • Early voting policy: The Resource Estimator has two primary sections: early and Election Day voting. Jurisdictions that do not offer in-person early voting can disregard the section on early voting and proceed to Election Day. We consider in-person early voting to be any traditional, polling-place-style voting that takes place prior to Election Day. In line with the EAVS survey that informs our analysis, we do not consider in-person absentee voting, or any voting transaction involving a ballot envelope, to be in-person early voting. A jurisdiction with in-person absentee voting should keep in mind that the formulas presented in the Resource Estimator assume non-envelope in-person early voting.
  • Mail voting policy: The first question the Resource Estimator asks is whether the jurisdiction proactively sends a ballot (not a ballot application) to every registered voter. We refer to this policy as ‘vote-by-mail’ throughout the report.
  • Expected in-person participation: This input might require some additional work and calculation by the election official. In order to know how many voting sites and poll workers are needed, we must have an estimate for how many voters are expected to participate in early and Election Day voting options. Rather than solely the number of ballots expected to be cast by each method, we rely upon the total number of transactions being made (in other words, “participation”) in each method. For example, if a person tries to vote at a voting site on Election Day but is unable to cast a valid ballot, that still counts as a voting transaction because it requires that the jurisdiction expel resources even if the ballot it not ultimately counted.

To use the Resource Estimator, election officials and policymakers can follow the instructions in the below graphic. Alternatively, an interactive spreadsheet can be downloaded here to produce calculations more easily.

Example: Applying the Resource Estimator to Two Hypothetical Jurisdictions

Consider two hypothetical jurisdictions, each with 100,000 active and participating voters:

  • Jurisdiction A” is a traditional jurisdiction that relies primarily on in-person Election Day voting. It does not send a mail ballot to all registered voters. Jurisdiction A expects 100,000 voters to participate in its upcoming election, with 30 percent of those voters likely to vote absentee and the remaining 70 percent to vote in-person on Election Day.
  • “Jurisdiction B” recently introduced vote-by-mail and sends a ballot to all registered voters, in addition to offering in-person early and Election Day voting options. For its upcoming election, it expects 40 percent of its voters to vote by mail, 30 percent to vote early, and 30 percent to vote on Election Day.

Based on these estimates for the number of voters participating in in-person voting options, the Resource Estimator produces the following values for the minimum number of voting sites needed for each jurisdiction:

Minimum Number of Voting Sites Needed

This example highlights the different resource needs of jurisdictions, dependent on their mail voting policies. Both jurisdictions expect 100,000 voters to participate in the election, but while most voters in Jurisdiction A vote on Election Day, most voters in Jurisdiction B vote early or by mail. Given this disparity, there is no reason why both jurisdictions should have the same number of voting sites. The Resource Estimator captures this nuance and recommends that Jurisdiction A have more Election Day voting sites to meet voting demand. On the contrary, Jurisdiction B needs about 40 percent fewer Election Day sites because more of its voters are utilizing its early and mail voting options. Should Jurisdiction A implement more expansive mail and early voting policies, it could likely decrease the number of Election Day voting sites in use without negatively influencing voters’ access to the ballot.

The number of poll workers each jurisdiction needs also depends on how many voters are expected to use the different voting options. Jurisdiction B only needs 352 poll workers on Election Day for its estimated 30,000 Election Day voters, while Jurisdiction A needs 566 for its roughly 70,000 Election Day voters. Jurisdiction A could reasonably reduce the number of Election Day poll workers it needs if it were to expand the use of convenience options like mail and early voting.

Minumum Number of Poll Workers Needed

This example highlights the potential of the Resource Estimator to help election officials thoughtfully allocate limited voting resources. No jurisdiction in the United States is the same, thus a one-size-fits-all approach fails to account for the vastly different needs of a diverse voting population. The Resource Estimator is just the beginning of a modernized election administration system that promotes both equitable voter access and administrative efficiency.

Visualizing the Relationship Between Policy, Resource Allocation, and Voter Behavior

The Resource Estimator is the product of multiple linear regression analyses that explored the nuanced relationships between individual voting options and voters’ behavior. Our findings confirm that election administration policies have an undeniable impact on voter behavior. A jurisdiction’s mail and early voting policies heavily influence how voters participate in an election. As such, it is critical that when expanding or contracting one voting option (such as mail voting), policymakers anticipate the downstream impacts on other voting options (such as early voting) and ensure resources are allocated accordingly. For example, a jurisdiction that does not send a mail ballot to all eligible voters could meet demand for in-person voting by opening additional early voting sites.

The Resource Estimator is intended to provide a rough framework for allocating resources, and not an exact solution. Election officials should tailor these recommendations to the unique needs of their voters.

Early In-Person Voting: Voter Participation

There is a clear, positive relationship between the number of early voting sites per jurisdction and the number of voters who participate in early voting. This model suggests that at-large the minimum number of early voting sites any jurisdiction should have is two. Jurisdictions would then add additional locations based on the number of expected early voters (discounting each voter by 0.0001). For example, a jurisdiction expecting 60,000 early voters would have eight early voting sites: floor of two sites plus 60,000 time the discount rate of 0.0001.

Figure 1: Early Voting Sites and Participation

This simple model hints at the huge, transformative potential of data-informed decision-making in the allocation of election resources. The following analyses further flesh out this simple approach by incorporating additional variables that get at the distribution of voting options in a jurisdiction.

Early In-Person Voting: Incorporating Mail Voting Policies

Vote-by-mail jurisdictions have lower rates of in-person early voting than jurisdictions with traditional absentee voting. However, and somewhat deceptively, our analysis also shows that vote-by-mail jurisdictions have more in-person early voting sites available per voter than those without. This is likely caused by two reasons. First, jurisdictions with more expansive mail voting options tend to have lower rates of early voting, and thus there is less demand for each voting site. Second, this reflects a political culture in some areas that prioritizes more voting options as a public good, even if they are not fully utilized.

As the below figure depicts, there is a clear relationship between the number of early voting sites and a jurisdiction’s mail voting policy. As jurisdictions progress from the absentee voting to vote-by-mail (in which a mail ballot is sent to all eligible voters), the number of early voting sites offered per early voter increases. Jurisdictions with more mail voting availability tend to offer more early voting sites per early voter, even if their total number of early voting sites is smaller.

Vote-by-mail jurisdictions require more than four times as many early voting sites per early voter as absentee areas. This unique contrast suggests that while vote-by-mail jurisdictions have lower rates of early voting, they need more sites per voter to meet demand. This makes sense when considering that each early voting site can only cover so large of a geographic area, capping the potential number of voters served by each site. While jurisdictions with higher early voting populations can serve more voters per site, jurisdictions with fewer early voters need more sites per voter (but less sites in total).

Figure 2: Early Voting Sites and Mail Voting Policy

Early In-Person Voting: Poll Workers

The growing use of vote center-style voting options makes estimating the number of poll workers needed per voting site challenging. A jurisdiction with five vote centers could require the same number of poll workers as a jurisdiction with 20 traditional voting sites. That said, our analysis is intended to capture general trends and propose a floor for the number of poll workers needed—not to provide a definitive solution. We found that, across the board, jurisdictions need about nine poll workers per early voting site. Election officials utilizing vote centers should adapt this estimate to their jurisdiction by increasing the ratio of poll workers needed per vote center.

Because our Resource Estimator incorporates multiple inputs to determine the ideal number of voting sites, identifying the ideal number of poll workers requires only a simple model based on voting sites alone (which, after following the Resource Estimator, will have already accounted for both mail voting policy and voter participation rates). This approach minimizes collinearity and promotes efficient resource allocation.

Figure 3: Early Voting Sites and Poll Workers

Election Day Voting: Voter Participation

The number of Election Day voting sites a jurisdiction needs is highly dependent on the number of voters who choose to vote in-person on Election Day.

Figure 4: Election Day Voting Sites and Participation

However, participation alone does not paint the whole picture. Incorporating the other voting options that a jurisdiction has available—namely, their mail voting policies—is crucial to identifying the minimum number of Election Day voting sites needed.

Figure 5 shows how a jurisdiction’s mail voting policy impacts the number of voters served per Election Day voting site. Similar to the trends we saw with mail voting policies and early voting rates, vote-by-mail jurisdictions tend to serve about 30 percent more voters per Election Day voting site than absentee-only jurisdictions. When voters have more options to vote before Election Day, fewer of the jurisdiction’s voters will vote in-person on Election Day. Yet because a physical voting site can only serve so large a geographic area, early voting jurisdictions must offer more sites per voter in order to serve all voters in an equitable manner. Critically, the Resource Estimator allocates voting sites based on the number of voters participating in a specific voting option. Therefore, even though vote-by-mail jurisdictions need more sites per participating voter, because the number of voters participating in Election Day voting is lower they are likely need fewer sites in total than absentee jurisdictions.

Figure 5: Election Day Voting Sites and Mail Voting Policy

Election Day Poll Workers

There is a clear, linear relationship between Election Day voting sites and poll workers. While vote-center-style voting is less common on Election Day than in early voting, election officials should adjust this formula to voters’ in-person voting preferences. There is a positive relationship between Election Day voting sites and poll workers (more voting sites yields more poll workers), but a jurisdiction could reasonably consolidate the number of voting sites while expanding the number of poll workers and still meet voter needs.

Figure 6: Election Day Poll Workers and Voting Sites


No political party has a monopoly on how to run good elections. Yet over the past decade, election policy has become overrun with partisan strife; Democratic states are assumed to be havens for expansive voter access, while Republican states are seen as inherently regressive on pro-voter reforms. Reality is far more complicated. Expanding all voting options unilaterally is inefficient and, when mandated without sufficient resources, risks overwhelming election jurisdictions. Paring back convenience voting options is bad for voters of all parties.

Building a functional elections system—the very bedrock of any democracy—requires buy-in from the whole of the American electorate. Productive election policy must begin as a bipartisan endeavor that neither tolerates anti-voter rhetoric nor unilaterally excludes any one party.

The debate on voting reforms must move away from partisan extremes—more of everything or a wholescale reduction in convenience voting options—and meet voters where they are. The electorate will continue to change: voters in 20 years may have very different preferences about how to vote than they do today, just as today’s voters expect a different slate of options than voters 20 years ago. Election administration should not be static, and much work is needed to bring our electoral system into the modern age. Innovation is only possible if reallocation and evidence-based policymaking can flourish.


The authors would like to thank Michael Hanmer, Research Director of the Center for Democracy and Civic Engagement and Professor at the Department of Government and Politics at the University of Maryland, for his thoughtful feedback on the methodological approach of this paper.

Appendix: Methodology

This report relies upon data from the Election Assistance Commission’s Election Administration and Voting Survey (EAVS). We merged the 2016 and 2018 standard EAVS survey with the 2018 EAVS policy survey to identify key trends in the allocation of voting options. From those trends, we performed regression analysis to devise a set of formulas that determine the minimum number of early voting and Election Day voting sites and poll workers a jurisdiction needs based on their voters’ preferences and mail voting policies. Given that 80 percent of voters were satisfied with their voting options in 2020 (as found by an April 2020 Bipartisan Policy Center/Morning Consult survey), it can be assumed that in general the current supply of voting options is sufficient to meet voters’ needs. As such, we operationalize the minimally acceptable level of voting options as that which best fits what is currently offered across the United States.

The below table includes the questions that were used from the EAVS survey:

The below table includes the questions that were utilized from the EAVS Policy survey:

Data from both 2016 and 2018 was joined into a single dataset. Jurisdiction-level data from the EAVS survey was paired with state-level data from the EAVS Policy survey to get a comprehensive picture of voting options and voter behavior across the United States. From there, the combined dataset was analyzed in a Python Jupyter Notebook. Scatter plots were used to identify general trends and correlations between variables. Regression analysis (conducted via the Python scipy.stats library) was then used to confirm the statistical significance of any hypothesized relationships.

“Participation” in each voting option was used as opposed to votes counted, because there is still a strain on resources to conduct any voting transaction, even if the transaction results in a rejected ballot.

Data Cleaning

Many jurisdictions erroneously submitted early or Election Day vote totals in place of the number of voting sites. Survey responses for jurisdictions in which the number of voting sites was greater than or equal to the number of votes cast were removed from the dataset. Additionally, responses for jurisdictions in which the number of voting sites was greater than or equal to the number of poll workers were removed from the dataset, but only for the section of the analysis directly related to poll workers.

Early analysis revealed the presence of outliers that drastically impacted the model’s results. Because our goal was to identify general trends, we ultimately removed all outliers more than three standard deviations from the mean.

OLS versus Negative Binomial Regression Model

Our analysis relied upon OLS regression analysis. There are several limitations with this approach—OLS assumes that input data is continuous and capable of being negative. Neither is true for our data, which is comprised of positive integers, such as the number of voting sites. To combat this limitation, we tested a negative binomial regression, ideal for discrete count data. However, because our coefficients were extremely small (>0.001), the negative binomial model predicted inappropriately high values for larger jurisdictions. This is a known limitation of negative binomial models. Our goal was to capture general trends, and not to produce a highly accurate predictive model. As such, we ultimately choose an OLS rather than negative binomial regression approach.

Feature Selection and Limitations

The lack of data on the number of early voting days available at the jurisdiction level is a key limitation of our report. However, preliminary analysis of data from five jurisdictions (Pasco County, Wilson County, Escambia County, Ramsey County, and Maricopa County) from 2016, 2018, and 2020 indicates that when controlling for the number of early votes cast, the number of early voting days is highly correlated with the number of early voting sites (R2 = 0.73). This suggests that including the number of early voting days offered would not significantly alter results. When controlling for number of votes cast early, 73% of variation in days of early voting can be explained by the number of early voting sites.

Figure 7: Number of Early Voting Days has Minimal Impact on Number of Early Votes Cast

In addition, our report does not control for the type of voter registration available. Jurisdictions with same-day voter registration (a very resource-intensive policy) likely need more voting sites and poll workers than jurisdictions without.

Downloads and Resources