Overview

The ongoing COVID-19 pandemic has ushered in a new reliance on non-traditional voting methods, as over 43 percent of voters utilized vote-by-mail (VBM) in the 2020 election. While this sweeping transition has increased convenience for many voters, it has opened a door for another form of voter suppression: ballot rejection. In the 2020 general election, over 560 thousand vote-by-mail ballots were rejected nationwide. To contextualize this amount, both Georgia and Arizona were decided by less than 12 thousand votes in that same election. Mail ballot rejections are an important component to election outcomes, yet their causes and effects are largely understudied. Are rejected ballots randomly distributed, or are they felt unequally by certain racial or ethnic groups? What can explain patterns of ballot rejection in election administration? What institutional changes can local election administrations make to combat unfair ballot rejection patterns? My book project views these problems through the lens of race and racism and uses a multi-method approach to argue that individual discretion at multiple levels of election administration can explain the severe racial gap in disenfranchisement seen each election. 

Chapter 1: Using Observational Data to Measure Racialized Disenfranchisement and Test Theories of Political Context

Chapter 1 is an investigation into the massive amount of state and county-level variation in who constitutes as an eligible voter and in the case of VBM, what constitutes as an “acceptable” vote. For example, Texas made national headlines in 2022 for rejecting 12% of all mail-in ballots and minority voters getting their ballot rejected 30% more often than white voters. Meanwhile, in other states and counties, almost no ballots get rejected. Chapter 1 utilizes publicly available voter files to create a one-of-a-kind dataset that maps an increasingly relevant variable in election law: the racial gap in ballot rejection, or the numerical difference in the white and non-white ballot rejection across states, counties, precincts, etc. With this data in hand, I then test dominant theories of political context (e.g. racial threat) to explain the observed patterns of ballot rejection across different geographies. 

Also using observational data, I will analyze an increasingly important but largely unscrutinized feature of modern day elections: artificial intelligence in signature verification. North Carolina recently announced a pilot program where 10 counties will utilize some form of automatic signature verification software in the 2024 presidential primary. Because some counties in the state will receive "treatment" and other counties will not, I will conduct a difference-in-differences analysis with an eye towards racial bias in signature rejection. These technologies are becoming more and more common because of their speed and efficiency in signature review, however, just because the process is automated, does not mean there is no discretion. In fact, the Brennan Center has said "it isn’t difficult to imagine, for instance, how attackers could manipulate AI to discriminate in its signature matching." 

Chapter 2: Using an Experimental Approach to Diagnose the Root Cause of Racialized Ballot Rejection

In addition to measuring the racial differences in signature rejection across jurisdictions I also seek out why this trend exists in the first place. At the macro-level, I argue that different racial contexts and their associated political cultures can explain aggregate patterns across counties and states. But at the micro-level, what is motivating a poll worker’s decision to reject a ballot? What is the mechanism explaining patterns of racialized ballot rejection and what can that tell us about modern election administration? In Chapter 2 I describe a principal-agent problem in election administration and report a series of original experiments styled after the Washington survey experiment submitted in Reyes et al v. Chilton et al. With signature reviewers as the “agents” and state/federal officials as the “principals”, I argue that there is both an adverse selection problem as well as a moral hazard problem. Namely, I argue that election administrators face problems in hiring/selecting unbiased reviewers (adverse selection) as well as preventing reviewers from being prejudiced in their decision-making (moral hazard). To prove this, I simulate the working conditions and signature specimens that actual poll-workers face in a controlled laboratory setting and task participants to accept or reject signatures like an election worker would on election night. By manipulating the racial/ethnic connotation of signatures while holding all else constant, I am able to determine how racial attitudes affect the discretionary decision to reject a ballot. 

Chapter 3: Remedying Racialized Ballot Rejection Using Behavioral Public Administration

In Chapter 3 I seek to remedy racialized ballot rejection patterns by implementing practical nudging and debiasing techniques that suppress personal biases and promote controlled, rather than automatic thinking. There are dozens of examples in the public policy literature that demonstrate how a small bureaucratic tweak that is well-informed by behavioral science can drastically improve outcomes. Informing public policy with behavioral insights is so effective that president Obama even issued an executive order mandating that federal agencies “identify policies, programs, and operations for which behavioral insights may yield substantial improvements in social welfare and program outcomes” (Executive Order 13707).  I argue that signature review, like any other bureaucratic process, stands to gain from decades of behavioral science. If implemented correctly, things like accountability, incentives, information, peer pressure, priming, and changes in choice architecture all have the potential to steer signature reviewers towards making unbiased decisions. For example, if the priority in election administration is to reduce the amount of Type 1 error (incorrectly rejecting valid votes) what is the ideal decision making process? Do canvassing review boards minimize the amount of Type 1 error as they get larger as Condorcet Jury Theorem would predict? Similarly, should these committees require a majority rule or unanimous decision in rejecting a signature? By isolating the cognitive effects of each debiasing strategy, I hope to provide a tested-toolkit that can guide election administrators across the country towards a reduced, or even non-existent racial gap in ballot rejection.