Voter Data, Democratic Inequality, and the Risk of Political Violence
Erik Carter

Voter Data, Democratic Inequality, and the Risk of Political Violence

How data is used to marginalize the poor during elections—and what can be done about it

Data and Democracy

A Knight Institute and Law and Political Economy Project essay series considering how big data is changing our system of self-government


Campaigns’ increasing reliance on data‑driven canvassing has coincided with a disquieting trend in American politics: a stark gap in voter turnout between the rich and poor. Turnout among the poor has remained low in modern elections despite legal changes that have dramatically decreased the cost of voting. In this article, we present evidence that the combined availability of voter history data and modern microtargeting strategies have contributed to the rich‑poor turnout gap. That is the case despite the promises of big data to lower the transaction costs of voter outreach, as well as additional reforms that have lowered the barriers to voting in other ways. Because the poor are less likely to have voted in prior elections, they are also less likely to appear in the mobilization models employed by data‑savvy campaigns.

In this article, we draw on a novel data set of voter data laws in every state and show that turnout rates among the poor are lower in states that disclose voter history data to campaigns. We also find that after states change their laws to provide voter history to campaigns, these campaigns are far less likely to contact the poor.

The consequences of this vicious cycle are already known: The unique interests of the poor have been entirely unrepresented in the political process. Such political marginalization and alienation of an entire class from the democratic process is not only a problem for the poor; it poses a systemic threat to political moderation and democratic stability. Politically marginalized and alienated groups may resort to nonpolitical means to effectuate social change and may also become ripe for recruitment by extremist and anti‑democratic elements that are latent in every society. Recent incidents of domestic political violence demonstrate that the United States is no exception.

To address this threat of marginalizing the poor from democratic politics, we advance three sets of proposals. First, we argue that states should regulate the information environment of political campaigns. Prohibiting the collection and distribution of voter history data is not practical, but states should lean into their privacy laws to prohibit the matching of voter files with other administrative data sets and should provide voter history data to campaigns independent of any information about individual political preferences. Second, states should create financial incentives for campaigns to expand their mobilization efforts to include a more representative target population that is more inclusive of the poor. Traditional campaign finance voucher and tax rebate programs are likely inadequate on their own. Instead, we propose a series of novel incentive programs that would provide cash grants to campaigns that report the most donors during each reporting period and to parties that generate more turnout than their historical average. Finally, we advance proposals for social media platforms to self‑regulate “look‑alike” targeting and segmented online political ads that amplify inequalities in mobilization and exacerbate political marginalization.

Political parties and individual campaigns in the United States are currently not mandated by law to promote political equality. The above reforms aim to align the short‑term interests of parties and campaigns (winning the next election) with the long‑term public interest in preserving a healthy democracy. Constructing a more inclusive political system will benefit everyone who seeks to live in a sustainable representative democracy, not just those who are currently marginalized.


Political equality lies at the core of democracy. Equality in participation is a central predicate to majority rule, and majority rule is a critical vehicle through which representational equality can be secured. Unfortunately, the history of the United States is one of failure to achieve the first step: democratic equality in political participation. To explain that failure, scholars have tended to focus on legal prohibitions on voting, which have been applied throughout U.S. history to exclude racial minorities and women. Indeed, until recently, the breadth of these prohibitions made majority rule a convenient fiction to support a myth of American democratic exceptionalism that has been told to its people and the world.

Only in the past half century, after the repeal and replacement of legal voter exclusions with constitutional amendments, laws, and judicial decisions protecting the right to vote, has the United States been able to move closer to its democratic ideals. There has been increasing participatory equality between women and men and citizens of color and whites. As a consequence, more women and racial minorities hold political office than at any time in our history.

And yet, a long road toward democratic equality remains. African American, Asian, and Latinx communities still see their unique preferences and needs being disproportionately unmet. Racial minorities comprise nearly 40%t of the U.S. population and yet only 22% of Congress. Barack Obama has been the only racial minority elected president of the United States. While a woman currently serves as the vice president and another woman won the popular vote in the 2016 presidential election, the fundamental fact remains that no woman has yet been elected president. And despite making up more than half of the total population, women comprise less than a quarter of all members of Congress and approximately 30% of elected members of state legislatures nationwide.

While these forms of enduring democratic inequality continue to be an appropriate focal point for scholars, policymakers, and advocates alike, another striking form of political inequality based on economic class has been somewhat overshadowed: the participatory inequality of the poor. In every federal election over the past half century, there has been a persistent turnout gap between the wealthiest and poorest Americans of approximately 30%. In the 2020 presidential election, 53% of U.S. citizens with family income below $30,000 reported voting, while 83% of U.S. citizens with family income above $100,000 reported voting. Similarly, in the most recent midterm election in 2018, only 40% of U.S. citizens with family incomes below $30,000 reported voting, while 70% of U.S. citizens with family incomes above $100,000 reported voting.

What explains that stark gap in turnout between the rich and the poor? In this article, we argue that newly available voter history data, which campaigns are using for novel microtargeting strategies, have contributed to a growing rich‑poor gap among those who report having contact with a political party or campaign. Because campaigns are a major force for educating and mobilizing the electorate, the large and growing rich‑poor gap in campaign activity has enormously devastating potential for representative government in the United States.

Over the past two decades, scholars have shifted away from a focus on legal barriers that either increase the costs or make it impossible to vote as the primary explanation for the low participation of the poor. That shift in focus arose from the failure of past scholars to explain the paradox of declining legal barriers and unchanged turnout patterns for the poor. Around the same time, social scientists found that the mobilization activities of parties and candidates prior to elections serve as critical vehicles for increasing turnout by lowering the information costs of voting and increasing individuals’ perception of the benefits from voting.

The problem, however, is that campaigns do not contact and mobilize everyone. Instead, campaigns, subject to budget constraints, make strategic choices about who to contact based on their propensity to vote and their likelihood of voting for the candidate.

The availability of voter data from states has been a crucial tool for campaigns’ strategic mobilization. The most important of the state data sources for campaigns has been individuals’ voter history data. Nearly every state has provided for the disclosure of each registered voter’s history of voting. The availability of this data has opened the door to campaigns’ use of more sophisticated microtargeting. This microtargeting has fueled high rich‑poor contact disparities in recent elections and thus exacerbated the persistent political marginalization of the poor.

The political marginalization of the poor is not only a problem for the poor: it poses a systemic threat to political moderation and political stability. The January 2021 American Perspectives Survey found that 69% of Americans believe that American democracy only serves the interests of the wealthy and powerful. Of these respondents, 36% believe that “[they] may have to use force to save” the American way of life. That finding confirms decades of research on political marginalization and violence. And the finding is bipartisan. One month before the November 2020 election, five of the leading American scholars on partisanship and political violence reported survey data that “36 percent of Republicans and 33 percent of Democrats said it is at least ‘a little’ justified for their side ‘to use violence in advancing political goals.’”

Research shows that politically marginalized and alienated groups in any democracy may resort to nonpolitical means to effectuate social change and may also become ripe for recruitment by extremist and anti‑democratic elements that are latent in every society. The United States is no exception. On January 6, 2021, an armed mob violently stormed the U.S. Capitol Building, killing a police officer, seriously injuring more than 130 officers, and audibly chanting for the execution of Vice President Pence. Members of the mob were convinced that America’s political system was not serving their interests and, further, that the system was rigged and would continue to alienate them. That message of marginalization was central to the rise of President Trump in 2016 and continued through the 2020 election. The message was not just that the “silent majority” supported Trump, but that the majority was being silenced by an unfair political system.

On January 6, 2021, this sense of relegation transformed into violence. A majority of those arrested for attacking the Capitol were not affiliated with right‑wing militia groups that have harbored anti‑government resentment for decades; rather they had been manipulated by President Trump and conservative media in just a few short months after the November election to become a new, violent extremist force. According to an analysis of 125 people arrested for breaking into the Capitol, nearly 60% had histories of financial struggles, from unpaid taxes to bankruptcy and foreclosure. According to Professor Don Haider‑Markel, this financial instability fits a pattern of those who feel, “[s]omehow, they’ve been wronged, they’ve developed a grievance, and they tend to connect that to some broader ideology.”

It is thus imperative to redress political marginalization and participatory inequality of all kinds. Our focus in this article is on the marginalization of the poor, which we view as particularly pressing because their exclusion is systematicand has proven durable despite reforms that have improved the plight of other minority populations.

This article proceeds in four parts. In Part I, we trace the history of voting data and voting history data in particular. We show why jurisdictions historically collected this data and examine the role of federal and state law in facilitating campaign access to voter data. In Part II, we examine how campaigns use voter history data in their strategic mobilization of voters and hypothesize about the relationship between voter history data availability and the rich‑poor contact gap. In Part III, we take advantage of state and temporal variations in the provision of voting history data to assess empirically how the availability of this data correlates with the rich‑poor contact gap. In Part IV, we offer a series of policy suggestions to reduce the rich‑poor contact gap. Our proposals include (1) limiting the availability of voting history data, (2) incentivizing campaigns’ use of voter data to mobilize infrequent voters and nonvoters, and (3) curbing microtargeting on social media.

I. Voting History Data

States’ collection and distribution of voting history data—the record showing whether an individual cast a ballot during prior elections—has been left entirely unexamined in the legal literature. One reason is that the history of the state collection and distribution of voting history data has, to this point, been dispersed in studies of English poll books and American registration lists without anyone drawing the connection. That history is important because it sheds light on why states collect and distribute voting history data, while generating insights as to how harmful uses of such voting history data might be remedied.

In this part, we provide the first comprehensive history of voting data as a lead‑in for our empirical examination of the effects of voter history on political inequality and our proposals for regulating the use of voter history data to avoid democratic harms.

The early history of voting history data

In 1696, the English Parliament passed the first known law requiring the collection of individual voting history data in an act requiring that votes in all elections to Parliament be published in a poll book. That Act was passed during the decade following the Glorious Revolution as part of Parliament’s assertion of its newly constitutionalized independence and autonomy from the Crown. As a governing institution now equal to the Crown in the lawmaking process, Parliament sought to defend itself against Crown influence over its membership.

The Glorious Revolution came about through the efforts of King James II to influence the composition of Parliament, as he tried to pack the body with his supporters. James’ attempts led independent‑minded parliamentarians to support the Dutch king and queen’s invasion of England, which led to James’ ultimate abdication of the throne. The Bill of Rights adopted after James’ abdication established that “[the] election of members of Parliament ought to be free,” meaning that the Crown should have no authority to interfere with elections to Parliament.

After the Glorious Revolution, however, the Crown retained the authority to appoint sheriffs who were responsible for electing some of the members of Parliament. The sheriffs returned election results to Parliament from the counties. Through the manipulation of returns, sheriffs, acting pursuant to the authority of the Crown, could determine who was elected to Parliament from the counties, while Parliament possessed limited means to check or contest the sheriffs’ determinations. The 1696 poll book law provided for the publication of voting history data as a means to enforce the Parliament’s right to elections free from Crown interference. The law required clerks appointed by sheriffs “to set[] down[] the [n]ames of each [f]reeholder and the [p]lace of his [f]reehold and for whom[] he[] shall [p]oll.” In addition to the names and individuals’ candidate choice, poll books typically included each voters’ address, parish, qualification to vote, and sometimes a list of qualified electors who did not vote. In disputed or allegedly fraudulent elections, Parliament relied on the voting history data contained in the poll books to decide who had been elected by the freeholders of the counties.

Although there were no provisions in the law for the permanent preservation of poll books, parties and candidates presumably kept poll books from prior elections to canvass in subsequent elections. We have not found anything that describes how campaigns used poll books, but evidence of the apparent value of voting history to campaigns is demonstrated by the continued requirement in the Great Reform Act of 1832 that such evidence be collected, long after the concern about sheriff manipulation of parliamentary contests had faded.

There is no evidence of the use of poll books in the American colonies or in the early history of the American states. Instead, registration laws served as a primary means to protect the integrity of elections against fraud. Massachusetts passed the first registration law in 1800, but it was not until later in the 19th century that there was the more widespread adoption of registration requirements. Initially, government officials registered individuals to vote on the basis of their own knowledge of who was eligible to vote in the jurisdiction. Later, as populations grew and registration on the basis of personal knowledge proved infeasible, cities and other jurisdictions required the personal registration of individuals. As part of their registration, individuals would have to provide their name, address, and relevant information about their eligibility to vote such as their age.

The early registration laws failed as a defense against the fraud that political machines perpetuated in major cities in the late 19th and early 20th centuries. Politicians ran many registration offices. Those officials lacked professional training and, in major urban centers, were either a part of, or subject to the influence of, the corrupt political machines. During that era, repeat voting facilitated by registration officers’ padding of voter rolls served as a primary means of distorting election results. Those blatant forms of corruption led to movements for change to the registration process to secure more honest elections.

Political scientist Joseph P. Harris was an intellectual leader of this movement. During the first half of the 20th century, Harris’ reports on registration systems derived from field work and interviews with registration officials provide the best insights into the development of the registration processes. In fact, Harris is the only expert on registration who has addressed the purpose and value of collecting and distributing voting history data.

In an influential report written in 1929, Harris described the purpose and function of voter registration in the United States. He also offered recommendations for registration reform centered on the goal of preventing fraud and making the process of registration convenient to the voter. To deter fraud, Harris recommended voter identification at the polls through signature verification and a check‑off of the voters. He also recommended better list maintenance to eliminate voter rolls padded with the names of dead persons or persons who had moved away. To support the maintenance of voter rolls, Harris called for a process of cancelling the registration of individuals who failed to vote along with a process for individuals’ reinstatement onto the voter rolls if they responded to a notification of their removal. The process of cancelling registrations required that jurisdictions responsible for registering voters collect individuals’ voting history, but it did not require that such data be shared with anyone.

Harris also advocated for an even broader state collection of voting history data that served purposes beyond cancelling registration. He considered the voting history of individuals to be “of most value . . . when it is maintained continuously over a long period of time.” Harris ascribed as the primary reason for collecting voting history data the prevention of double voting. But as a political scientist, Harris acknowledged another selfish motive for the state collection of such extensive voting history data. Such data, he explained “would make possible in the future intensive researches in the subject of voting, which at present are greatly handicapped because of lack of adequate records.”

Harris also endorsed the practice of large cities sharing printed registration lists with party workers for purposes of canvassing. He advocated for the arrangement of registration lists by streets and numbers to make them easier for campaigns to use during their canvassing. Notably, Harris omitted any reference to voting history data as part of the registration lists that campaigns could access, presumably because he did not think such information should be relevant for canvassing purposes.

Nearly three decades later, in 1957, Harris wrote a report detailing the progress made in the development of registration processes. In this report summarizing interview responses from registration officials, Harris found that signature verification “acted as an effective deterrent” to election fraud through “impersonation and repeating.” The success of the signature verification rendered less important the continued inadequacies of list maintenance through cancellations of registrations. As list maintenance declined in value as a means of detecting election fraud, the utility of voting history data for the purpose of cancelling registration also diminished in value.

Harris also came to the conclusion that pre‑election fraud detection, primarily through signature verification, was the only way to protect the integrity of elections. Any efforts to redress fraud after the election, Harris discovered, would run into problems because the secrecy of individuals’ vote choice meant that the only remedy for an illegally cast ballot “would be to throw out an entire precinct vote, which would disfranchise qualified voters as well.” The use of voting history data for the purpose of post‑election investigations thus did not appear in the 1957 report.

Harris continued to recommend the collection and sharing of voting history data beyond its use for list maintenance purposes. But the reason for its collection and distribution had been narrowed to that of registration cancellation. Harris still did not consider its potential use for canvassing purposes.

It is unclear how many jurisdictions that had registered voters collected voting history data, but it is likely that many of them did for purposes of list maintenance. It is also unclear how many jurisdictions shared voting history data with campaigns, but the number was likely low. As noted above, Joseph Harris did not advocate for the sharing of voting history data with campaigns. And evidence from mobilization practices of the first half of the twentieth century show that campaigns relied almost exclusively on data about turnout and vote shares at the neighborhood or precinct level to guide their canvassing efforts. With the advent of television in the middle of the twentieth century, campaigns shifted much of their resources away from canvassing to the use of television advertisements to mobilize and persuade voters. Voting history data proved entirely irrelevant to this endeavor, removing any possibility of partisan pressure on cities, counties, and other registration jurisdictions to share individuals’ voting histories with campaigns.

By the end of the twentieth century, however, campaign dynamics began to change once again, just after Congress turned its attention to redressing the problem of perpetually low turnout in American elections. Because social scientists suggested that the process of registration was the problem, Congress had intervened with two laws that ultimately marked the return to relevance of voting history data, the National Voter Registration Act (NVRA) and the Help America Vote Act (HAVA).

Voting history data and federal registration laws

In 1980, Wolfinger and Rosenstone published their seminal empirical study Who Votes? According to the study, the combination of registration rules and practices, including early closing dates for registration, limited opening hours for registration offices, and prohibitions on absentee voting significantly reduced turnout. Wolfinger and Rosenstone found that the liberalization of voting procedures would have increased turnout in the 1972 election by 9.1%, with most of the positive effect on the least educated. Eight years later, political scientists Francis Fox Piven and Richard Cloward, in their influential book Why Americans Don’t Vote, came to a similar conclusion about the harmful effects of registration laws and procedures on turnout and the turnout disparity between the rich and the poor. Piven and Cloward analogized the complicated and varied registration procedures throughout the country to the income restrictions on voting and literacy tests of the past. Thus, the dominant view became that voter registration laws were an important source of low turnout in the United States and the turnout disparities between the rich and the poor.

In 1993, Congress finally broke through two decades of partisan deadlock on the issue of voter registration and passed the National Voter Registration Act (NVRA). Congress followed Joseph Harris’ lead from several decades earlier and identified as the two major goals of registration reform: (1) making it easier or more convenient to register, and (2) protecting the integrity of the electoral process from voter fraud. To make registration easier and more convenient, the NVRA mandated that states provide individuals with the opportunity to register at state public agencies, most notably state departments of motor vehicles. To mitigate election fraud, the law required that states maintain accurate voter rolls through a voter purge process. According to that process, the state may cancel individuals’ registration if they (1) fail to respond to a notice requiring that they confirm a change in address, and (2) do not vote in an election “during the period beginning on the date of the notice and ending on the day after the date of the second general election for [f]ederal office that occurs after the date of the notice.”

The NVRA’s nationalization of list maintenance requirements that depended on information about individuals’ voting history resulted in the collection of voting history data across all jurisdictions that registered voters.

Even though cities, counties, and other localities maintained the voter rolls in most states, the NVRA placed the onus on the states to make registration through public agencies available and ensure the accuracy of the voter rolls. The use of public agencies to register voters made the maintenance of accurate voter rolls particularly challenging in a decentralized registration system. Public agencies in the state not only had to register voters; they were also responsible for transmitting the voter registration applications to the appropriate local registration office so that registrants’ information would be entered into the proper voter roll.

In the decade immediately following the adoption of the NVRA, 15 states set up statewide voter registration systems, joining the five others that had already established such systems prior to the adoption of the NVRA. The states with statewide voter registration systems also maintained statewide voter registration lists with information about all the registered voters in the state, including the voting history data that was used in the registration cancellation process to maintain the accuracy of registration lists. But the centralization and coordination process for registration in some states was not enough to avoid problems at the polls. According to a U.S. census estimate of the 2000 presidential election, “three million registered voters did not vote because of problems with their registrations.”

After the controversy of the 2000 election—which focused public attention on hanging chads and voter technology—election law commissions, registration officers, and other experts focused some of their reform energy on the dysfunction associated with the process of registering voters and maintaining accurate voter rolls. The calls grew loud for centralization of voter registration processes throughout the country. Congress responded with the adoption of the Help America Vote Act (HAVA) in 2002. The two primary provisions of HAVA were (1) a $3 billion appropriation to states that was earmarked for the upgrade of voting technology, and (2) the digitization of voter registration for the purpose of coordinating a statewide recount in the case of close elections. The statute provided that:

each State, acting through the chief State election official, shall implement, in a uniform and nondiscriminatory manner, a single, uniform, official, centralized, interactive computerized statewide voter registration list defined, maintained, and administered at the State level that contains the name and registration information of every legally registered voter in the State and assigns a unique identifier to each legally registered voter in the State ...

The statute also provided that “[a]ny election official in the State, including any local election official, may obtain immediate electronic access to the information contained in the computerized list.” The law’s focus was thus on voter registration and the consistency of voter registration records among election officials. Over the next six years, all 49 states that required voter registration established computerized statewide voter registration systems that were regularly and frequently updated.

HAVA did not address the question of voting history. However, since the NVRA required that states collect voting history data to accurately maintain voter registration lists, the HAVA mandate of statewide voter registration lists resulted in nearly every state collecting voting history data in a centralized database.

Nonetheless, HAVA, like its NVRA predecessor, did not require states to disclose voting history information to the public. But as we show in the next part, after HAVA, states made voting history data increasingly available to campaigns. That data’s availability likely contributed to the rich‑poor turnout gap in direct contradiction to one of the original purposes of the NVRA. As registration has been made easier, we may paradoxically be moving further away from a more representative participatory democracy.

II. Law, Voting History Data, and the Rise of Microtargeting

Campaign mobilization strategies

The NVRA successfully increased the percentage of Americans registered to vote, but contrary to expectations, the addition of newly registered voters neither increased turnout nor reduced the turnout disparity between the rich and the poor. Some of that failure could be attributed to coordination problems and lost registrations of voters who registered at public agencies. Although HAVA did not announce as its goals increasing registration and turnout, an expectation based on the pre‑HAVA studies of registration problems was that the modernization of the election system would increase turnout and possibly contribute to greater participatory equality between the rich and the poor.

There is, however, no evidence that HAVA succeeded in achieving any of these ancillary goals. One key reason for that failure is that the centralization of voter registration systems came at the same time as an important shift in campaign strategy. The NVRA and HAVA inadvertently facilitated the microtargeting of voters through the centralization of voter registration databases. This new tool for political canvassing ultimately strengthened the resilience of rich‑poor turnout disparities, even as states generally made it easier to register and vote.

In the late 1990s and early 2000s, campaigns started to search for new ways to mobilize voters. By the 1990s, social scientists had determined that television advertisements, which had been the dominant tool of voter outreach since the middle of the twentieth century, were a mostly ineffective tool for mobilizing and persuading voters. Labor unions, critical voter mobilizing agents for the Democratic Party, started to test the utility of an old tool of voter mobilization—face‑to‑face canvassing. Prior to the 1998 midterm election, the AFL‑CIO conducted an experiment to test the effect of canvassing on turnout. In that experiment, the labor union canvassed a treatment group and ignored a random control group. The union found that the treatment group turned out significantly more than the control group. Academics then picked up on the question of the effectiveness of canvassing as a mobilizing tool. An influential study by Donald Green and Alan Gerber, using the same experimental design as the unions, found that face‑to‑face canvassing increased turnout by nearly 10%, an effect much greater than that found from the reduction in barriers to registration. Campaigns picked up on these studies and, beginning with the 2004 presidential election, gave canvassing an increasingly prominent role in candidate mobilization strategies.

Initially, campaigns pursued broad geography‑based canvassing in which they contacted everyone in a neighborhood or precinct that had relatively high turnout rates favorable to the party of the candidate. But campaigns began to shift their focus more to mobilization at the household or “micro” level, due in large part to the recommendations of a book published by Green and Gerber in 2004 titled Get Out the Vote! How to Increase Voter Turnout. The political scientists encouraged microtargeting as a more efficient form of voter mobilization. With a microtargeting strategy, campaigns contact individuals who, on the basis of their voting history and attributes indicative of their partisan affinity, are more likely to turn out and vote favorably for the candidate. Following Green and Gerber’s advice, campaigns started to engage in more intensive data collection efforts to assess voter propensities in order to pursue a microtargeting strategy. Perhaps the most important single piece of evidence in this assessment of voter propensities was the data on individuals’ voting histories.

As we have argued elsewhere, microtargeting can have countervailing effects. On the one hand, advances in data and computation empower campaigns to leverage personalized data in order to target likely supporters and then maximize their turnout. In the aggregate, as multiple candidates engage the electorate, microtargeting has the potential to increase turnout significantly. On the other hand, microtargeting gives campaigns the ability to avoid large swaths of the population, whether by choice or unwittingly. One of the key variables in campaign mobilization models is a prediction of whether an individual is likely to turn out, regardless of who they would support. Those data are inherently self‑reinforcing since they discourage candidates from attempting to mobilize individuals who, in their models, are not likely to turn out. Without this contact by campaigns, those individuals are then unlikely to turn out, which makes them a less attractive target for the next candidate, and so on. As we have shown, those in the lowest income brackets are the most susceptible to this vicious cycle, which has contributed to the growing inequality in voter turnout and, as a result, in political representation. That vicious cycle can begin for various reasons. Most explicitly, campaigns throughout America’s history have predominantly featured wealthy candidates who predominantly canvass wealthier neighborhoods. But even in a campaign where the candidates reach out to all voters equally, some individuals may not be able to get off work to vote as they had hoped, or they may fall ill on Election Day, or they may cast a ballot that is rejected for one reason or another. All of those individuals will appear as nonvoters on the voting lists, which means campaigns will be less likely to contact them in the future. That chain reaction is problematic enough. However, because employment status, health outcomes, and electoral clerical errors have a disproportionately negative impact on minorities, including the poor, the result is a slow ratcheting up of participatory inequality.

As it turns out, the highly touted data operations for national campaigns are based on relatively simple models. For all of the attention and commentary on the extraordinary details and “micro‑variables” that campaigns possess about each voter (hundreds of variables), very few variables are probative of the two most important questions that campaigns ask themselves about their mobilization strategy: Will a person support us? And if so, will they show up to the polls? To answer the second question, campaigns rely on predictive models that are driven almost exclusively by an individual’s past voting history. In other words, while national campaigns with vast resources have access to voter files that have been augmented with information from various commercial databases, the raw voter file itself provides one of the most important pieces of information: who has a history of voting. Voter files are provided to campaigns in most states free of charge. Thus, state and local campaigns with fewer resources have access to the same variable as deep‑pocketed national campaigns when devising their mobilization strategies.

Helping (some of) America Vote Act

In the remainder of this part, we explore the relationship between two findings: the link between voter history and campaign mobilization strategies on the one hand and the link between mobilization strategies and political inequality on the other hand. We hypothesize that campaigns with access to voter history files will contact the poor less often than they contact the rich. Our analysis leverages the fact that a number of states changed their laws with respect to the disclosure of voter history files in response to HAVA. Several other states have provided campaigns with voter history files for decades, and 20 states made these data available at the time HAVA was enacted.

HAVA did not require that states disclose voting history data to campaigns or the public. In practice, however, HAVA’s reform has had important implications for the mobilization strategies of parties and campaigns. First, voter registration records often include information about whether voters have turned out in past elections. The purpose of that information is for election officials to monitor potential voter fraud, such as voters voting in different jurisdictions. Second, states have a long history of providing voter registration records to political parties that, in turn, provide that information to individual campaigns. The 21 states that have enacted statutes to comply with HAVA have included provisions that provide for the disclosure of individuals’ voting history to campaigns. The digitization of voter registration records, pursuant to HAVA, thus had the downstream effect of making voter histories more easily available to both state and local party leaders. In many cases, the downstream effects were intentional and explicit. For example, the Connecticut State Legislature passed a law in 2003 requiring both the digitization of voter registration records, including voter history, and their consolidation at the state level. Then, in 2006, the state explicitly provided for the disclosure of these records to state parties. In some states, the response to HAVA was administrative and not legislative. For example, in Florida, the secretary of state published a regulatory rule in 2006 to audit and update the state’s voter registration records and make them publicly available.

Thus, at the time that the microtargeting strategy was gaining prominence, right around the 2004 presidential election, fewer than 20 states sold or otherwise provided voting history data to candidates and parties. Since 2005, however, the adoption of state laws and regulations providing for the sharing of voting history data to candidates and parties has accelerated. At present, 46 of the 49 states with voter registration requirements either sell or freely distribute to candidates and parties voting history data, along with other information about individuals in their voter registration files.

Our hypothesis in this Article is that the expanded access to voter history files have democracy‑distorting downstream effects on the mobilization strategies of campaigns. More specifically, we hypothesize that political parties and campaigns will contact the poor at a lower rate in states that disclose voter history compared to states that do not disclose voter history. We do not expect the contact rate among the wealthy to vary between disclose and non-disclose states. Thus, we hypothesize that the contact gap between rich and poor will be larger in states that disclose voter history compared to states that do not.

III. Empirical Findings on Voter Data Availability and Microtargeting

To test our hypotheses, we rely on a variety of data sources related to state election rules, campaign contacting behavior, and demographic information.

Cooperative Congressional Election Survey

Our primary dependent variable is a self‑reported measure of contact from the Cooperative Election Survey (CES). The CES is a nationally representative survey conducted every two years by a collaboration of dozens of universities across the United States. The sample size of approximately 55,000 per survey allows researchers to evaluate responses at a sub‑national level. In 2006, 2010, 2012, 2014, and 2016, the CES asked respondents whether they were contacted by a candidate, party organization, or other organization during the 20XX election. From 2010 through 2016, respondents who reported that they had been contacted were asked to identify the mode of that contact, whether in‑person, by phone, by email or text, or by letter or postcard. At the national level, the average self‑reported rate of contact by CES respondents between 2006 and 2016 is 65 %, meaning campaigns and parties made contact with nearly two‑thirds of eligible voters in some way. The one‑third of eligible voters who never heard from a campaign were twice as likely to be in the bottom 20% of the income bracket as the top 20%. To what degree does voter history contribute to this dynamic?

Figure 1 Proportion of survey respondents that report being contacted by a political party or candidate/campaign during an election cycle. Results are displayed by equally-sized income quintiles. The gap between contact rates among the rich (5) and poor (1) are displayed for each year.


As a starting point, we examine the changes in the rich‑poor contact gap over time. We draw on data from the American National Election Study (ANES survey) and the CES. Both surveys asked respondents whether they had been contacted by a political campaign. We stratified responses by equally sized income quintiles across the entire sample and then report the percent of respondents in each quintile who reported that a candidate or political party contacted them during the election cycle. Figure 1 shows that in elections between the late 1970s and early 1990s, a period that clearly preceded the use of microtargeting strategies, the contact gap between the rich and poor, according to the ANES survey, typically hovered around 15%. This result is consistent with the 16% contact gap found in the CES survey for the 2006 election, which like the elections between the late 1970s and early 1990s was contested prior to the extensive use of microtargeting strategies. But by 2010, the next election in which the CES surveyed respondents on campaign contact, the rich‑poor contact gap had jumped to 27%, rising to a high of 28% in the 2014 midterm election. Notably, the contact rate dropped for all income quintiles. In other words, in the aggregate as campaigns have become more reliant on microtargeting, they have contacted fewer people overall (and far fewer poor people). We also observe that the rich‑poor contact gap shrunk to 20% in the 2016 election, likely due to the efforts of the insurgent Trump campaign to mobilize unaffiliated (and less wealthy) voters.

Fifty state survey of election codes

To more specifically address the relationship between the availability of voting history data and the rising rich‑poor contact gap, we read the election codes for all 50 states. These codes comprise of state legislation, administrative handbooks, and regulatory guidelines. For each state, we note the current policy with respect to the disclosure of voter history and also the genealogy of the policies. In all, we have been able to clearly identify the voter history disclosure policies in 43 states as of 2020. In Figure 2, we present a map of states that illustrates the year in which voter history data was digitized at the state level and made available to parties and campaigns.

As we note above, 20 states had disclosure laws on the books by 2004. Seven states adopted laws in 2006, from New Jersey on the East Coast to Tennessee in the South, Illinois and Ohio in the Midwest, and Alaska in the Northwest. In 2010, 16 more states adopted laws in New England, the Deep South, and the West Coast. New Mexico and Texas adopted rules in 2014, and Delaware passed a law in 2016.

The contact gap between the richest 20% (those earning approximately $70,000 or more) and the poorest 20% (those earning approximately $20,000 or less) varies between states in each year in our sample. In order to evaluate the extent to which this variation is the result of voter history disclosure rules, we first look to see whether the average gap among states that disclose voter history in a given year differs from states that do not disclose voter history in that same year.

Figure 2 Map of states color-coded by the year in which the election code was updated to require statewide digitization of voter registration records, including voter history, and their disclosure.


In Figure 3, we plot the differences in both the contact rate among the poor and the contact gap between rich and poor in each year in our sample. We observe that voter history interacts in important ways with the mobilization strategies of campaigns in recent years. The contact rate with the poor was nearly identical between states that disclosed voter history and those that did not in 2006 and 2010.

However, since 2010, campaigns in states that disclose voter history have significantly cut back their outreach to the poor by as much as fi15%. We do not observe a similar decline in contact among the rich, inasmuch as the contact gap has increased significantly in 2010–2016. In fact, in states that disclose voter history, contact rates have increased among the rich while, at the same time, falling among the poor. These findings provide support for our first hypothesis that voter disclosure will decrease contact among the poor and increase the contact gap between rich and poor.

We acknowledge that while differences in the average rates of contact are suggestive, our analysis above does not address the question of causation. Our second hypothesis posits that as states change their disclosure laws, their rates of contact will appear more similar to those in states that disclosed voter history prior to the passage of HAVA, relative to the states that have never adopted disclosure laws.

Figure 3 Dotplot comparing the contact rate with the poor and contact gap between rich and poor in states that disclose vote history (in red) and states that do not (in gray). As the red lines indicate, the contact rate for the poor decreased significantly after 2010 in states that disclose voter history and the contact gap between rich and poor has consistently been large in states that disclose voter history

Unfortunately, the data do not permit us to reliably test this hypothesis. The strongest analysis would focus on presidential elections where microtargeting has been deployed in its fullest form. However, the CES did not ask about campaign contact in 2008, which means there is no pre‑disclosure rates for states that adopted disclosure in 2012. And by 2016, 47 states disclosed voter history to campaigns so a systematic evaluation of changes to disclosure laws is not possible. (Just one state, Delaware, adopted disclosure in 2016).

A second‑best approach would be to evaluate changes to voter history disclosure laws in midterm election years. However, because the CES was created in 2006, there are no pre‑disclosure contact rates to compare against the seven states that adopted disclosure in 2006. That leaves 2010 as the only year where a multi‑state analysis of changes to disclosure laws is possible. There were 16 states that adopted voter history disclosure laws in 2010 that we can compare to the 20 states that disclosed voter history data since 2004.

Table 1 Comparison of contact rates in the 16 states that adopted voter history disclosure laws in 2010 compared to the rates in 20 states that disclosed voter history data since 2004. Source: CES (N=148,016)

In Table 1 we observe support for our primary hypothesis: the contact rate among the poor in the 16 states that adopted disclosure laws in 2010 dropped 7% in the very first election after voter history was disclosed, just months after the law was changed. By the next midterm election in 2014, the rate of contact among the poor dropped another 20%. The contact rate among the rich did not follow a similar decline leading to a 10% increase in the contact gap when disclosure was enacted. This before‑and‑after pattern supports our findings above and supports our first hypothesis.

We also observe some support for our second hypothesis. The 16 states that adopted disclosure in 2010 (“new‑disclosing states”) look nearly identical to the 20 states that have disclosed voter history since at least 2004 (“always‑disclosing states”) on both measures (contact rates among the poor and the contact gap between rich and poor). However, missing from these trends is a comparison to states that did not disclose voter history data during the same time frame. Unfortunately, there are just two states—Nebraska and Hawaii—that did not disclose voter history data during this period. Two states are too few to rely on as a control group for 36 states, meaning further research is necessary to reliably confirm or reject our second hypothesis. Descriptively, the states that disclosed or adopted disclosure exhibit patterns that are more unequal than those of Nebraska and Hawaii: contact rates are consistently about 10% lower among the poor, and the contact gap between rich and poor is larger by about 8%. This is true despite the peculiarities of the 2014 election that saw historically low turnout, particularly in states with relatively high‑income voters. Voter contact and turnout decreased significantly across the board (see Figure 1), yet the rich‑poor gap in those states that adopted disclosure laws in 2010 remained nearly identical to that of states that previously disclosed voter history, and larger than that of Nebraska and Hawaii (though we do not place much weight on our comparisons to Nebraska and Hawaii).

What should we make of these findings? First, we acknowledge that our analysis cannot definitively prove that disclosing voter history causes differential contact rates between rich and poor. Indeed, the fact that new‑disclosing states exhibit contact rates similar to always‑disclosing states before disclosure laws were enacted suggest that the laws themselves may not be the root cause of changing contact rates, but that other characteristics may be responsible for both the contact rates and likelihood of adopting disclosure. For example, it is possible that states with a commitment to the poor have opted to not pass laws permitting the disclosure of voter history. However, our argument is that the direction of the causal arrow is not as important as the strong correlations we identify above. We believe that there are important policy implications due to the link between voter history and contact rates among the poor. Whether the disclosure of voter history causes inequality or represents a proxy for inequality, our findings suggest that states ought to re‑evaluate their policies with respect to the disclosure of voter history to campaigns.

IV. Paths Toward Greater Participatory Equality

The return of intensive canvassing has been widely praised as a critical part of continued democratic development. Knocking on doors brings a personal element to campaigning that television advertisements fail to match. It is effective in not only turning the people contacted out to vote but also in leading others in that person’s network to vote. And when people vote, they are more likely to engage in other forms of democratic activism, including contributing to campaigns, writing letters to elected officials, and serving as canvassers themselves. Politicians seeking re‑election are forced to pay attention to those voters as they construct campaign platforms around categories of voters that they seek to attract, campaign platforms that need to be given policy agenda space and turned into action if those politicians hope to sustain the newly mobilized as supporters in the future. Otherwise, those voters, now in the habit of voting, might turn out for the candidate’s opponent. Canvassing can thus create a virtuous democratic cycle. Participation leads to representation and, as a result, the fundamentals of representative democracy work—but only for those who are canvassed.

For the people that campaigns ignore, a more vicious anti‑democratic cycle arises. The uncanvassed are denied important sources of information relevant to voting and they lose interest in the democratic process. They therefore do not vote, do not contribute to campaigns, do not write letters to elected officials, and do not serve as canvassers themselves. Those nonvoters are ignored by mainstream campaigns. Their needs and preferences go unmet, usually without electoral consequences, as neglect by politicians contributes to nonvoters’ further marginalization and alienation from the political process.

This vicious anti‑democratic cycle of marginalization and alienation does not merely impose democratic harms on nonvoters. It also poses a systemic threat to our democratic system of government. Integrating nonvoters into our representative system is critical to sustaining the viability and legitimacy of American democracy. Those who are not integrated into the democratic system offer tempting targets of recruitment for anti‑democratic forces, including anti‑government militias, conspiracy theorists, and insurgent political candidates with authoritarian tendencies.

American democracy saw a glimpse of the threats from alienation and marginalization when members of militias, white supremacist groups, and conspiracy theorists stormed the Capitol with the intent of overriding one of the central pillars of democracy: the results of a democratically legitimate and fair election. The mob was willing to sacrifice democracy for a man who made them feel as if their voices and viewpoints mattered in ways that prior politicians had not (in particular by preying on their attitudes of ethnic antagonism). Democracy cannot be allowed to fall in the face of extremist violence. But it also will not continue to stand when the conditions are ripe for the rise of extremism. Political alienation and marginalization are central conditions making American democracy ripe for the rise of extremism.

A key question for American democracy is how to break the vicious cycle of political alienation and marginalization. We do not claim that canvassing is the source of all participatory inequality between the rich and the poor in the United States. Disparities in education and resources among other factors contribute to the turnout disparity as well.

Nonetheless, canvassing, especially canvassing tailored toward the political needs of the poor, could be a critical part of the path toward a more representative democracy. Social scientists have raised certain doubts about this proposition. Some experimental studies suggest that canvassing is not an effective tool for mobilizing those who vote infrequently or not at all. The problem with those experimental studies is that they assume the same forms of canvassing practices for frequent voters should work for infrequent voters. But the work of others, including Lisa García Bedolla and Melissa Michelson, suggest that is not the case. Instead, mobilization practices that include canvassers from the communities of the marginalized, who can speak more directly to the needs and concerns of these communities, are quite effective at turning out infrequent voters and nonvoters.

Nonetheless, the conventional wisdom remains that it is too costly for campaigns to invest in mobilizing nonvoters and infrequent voters. Campaigns therefore use voting history data, which is now very accessible, as a key component of their calculus of contact. As a result, they often neglect and ignore marginalized and alienated Americans in their mobilization activities.

Changing campaigns’ calculus of contact is therefore critical to getting the United States on the path toward a more representative democracy. In this part, we describe two potential public policy responses to campaigns’ increasing reliance on voter data: (1) regulating the data, or (2) incentivizing campaigns, and the parties they represent, to use the data in ways that promote political equality. We conclude that the latter approach, which will require a recalibration of the role and responsibilities of parties in our democracy, is the more fruitful path for reform. We also discuss two important reforms to social media microtargeting policies that would stem the tide of inequality, which is being aggravated on these platforms.

Changing the information environment of campaigns

Data about individuals’ past voting behavior has emerged as a principal tool by which campaigns selectively mobilize voters, with potentially devastating consequences for American democratic development. One apparently obvious legal solution is to regulate the availability of such data. For example, advocates wanting a more egalitarian democracy might seek to bar states from sharing voting history data with candidates and parties. Such prohibitions on data sharing, however, are unlikely to be effective for several reasons.

First, since many states have been sharing voting history data for the past two decades, a prohibition on such data sharing is unlikely to change campaign behavior for many years to come. Even with a ban on collecting new voting history data, campaigns will still have in their possession older registration lists that contain this information. In the short term, therefore, we might not see much change in campaigns’ mobilization practices. In the long term, as those older voter registration lists become outdated and less useful, campaigns will likely develop predictive algorithms for past turnout behavior based on other publicly available information about individuals. Such predictive algorithms will not be as accurate as voting history data, but to the extent that income and education are included in the calculation, the predictive algorithm could actually exacerbate the rich‑poor turnout disparity beyond where it is right now. In particular, the algorithm might predict on the basis of income and education that certain poorer or less-educated individuals have a low likelihood of voting when they have actually voted in the past, perhaps in part influenced by the information and sense of duty to vote that campaigns have provided through their past contacts. Individuals like these might therefore be pushed out of the virtuous cycle constructed by campaign contact and into the vicious cycle resulting from campaign neglect, leaving even wider swaths of Americans politically marginalized and alienated.

Second, because voting history data is used as a tool to support the integrity of elections through accurate voter rolls, it is not clear how a prohibition on sharing voting history data could be made consistent with current federal law. States use voting history data for purposes of voter list maintenance, which is required under the NVRA. Under the Act’s voter list maintenance requirements, states are responsible for removing individuals from state voter lists when they fail to vote in two consecutive elections and subsequently fail to respond to a state notice with a prepaid return address. For both voting rights and election integrity advocates, access to information about past voting behavior is critical for monitoring the state’s voter list maintenance activities under the NVRA. Although it might be possible in theory to restrict use of voting history data to the oversight of state voter list maintenance, it would be impossible in practice to police campaigns’ use of such data for canvassing purposes.

States should, however, prohibit the matching of voter files to other administrative data sets, particularly those data sets that are disclosed to political campaigns. In particular, voter history files should be free of any information that could reveal individual political preferences. In addition, following a proposal by Professors Chris Elmendorf and Abby Wood, states might provide licenses to access the voter history file with a set of stipulations that prohibit licensees from merging the file with voter registration lists, partisanship estimates, or other commercial data.

Ultimately though, as we outline below, there is more to gain from collecting and sharing voter history data than trying to ban or hide the data altogether.

Incentivizing more equal mobilization

Eighty years ago, political theorist E. E. Schattschneider advanced the thesis that “political parties created democracy and that modern democracy is unthinkable save in terms of the parties.” At the time Schattschneider wrote, his assertion was considered controversial, given the then prevalent, as well as historical, anti‑partyism in American society. James Madison famously repudiated the factionalism that many associated with partisan politics and twentieth century parties had been the source of corruption, bossism, racism, and other politically exclusionary practices. Yet Schattschneider recognized that competition between the parties had catalyzed the expansion of the electorate and that partisan organization and mobilization were key vehicles for democratic participation.

Legal scholars have since embraced political parties as intermediary institutions between the people and government that can enhance governmental accountability and responsiveness. The role of law, according to such accounts, is to regulate impediments to political competition, such as gerrymandered districts and closed political primaries, that undermine the role of parties as democratic intermediaries.

Courts and legal scholars have not, however, thought about political parties in the role that Schattschneider assigned to them, which is as critical agents for more expansive democratic participation. Instead, legal actors seeking to expand participation have focused on the role of tangible barriers to voting in reducing democratic participation. Tangible barriers to voting include voter ID laws, limits on convenience voting such as absentee and early voting, and requirements that individuals register to vote rather than automatically being added to the voter roll upon eligibility. Despite the legal focus on those tangible barriers to voting, political scientists have consistently found that the removal of those barriers only has a small impact on turnout. Political scientists have instead found that the actions of parties and candidates to mobilize individuals to vote have had a much greater impact on turnout. Through their campaign activities, parties are therefore still the key to a more expansive and egalitarian participatory democracy.

From a constitutional perspective, the question is whether parties can be legally obligated to perform the functions critical for a more expansive and egalitarian democracy. In other words, can parties be required to canvass more broadly and contact individuals in a more unbiased way during electoral campaigns? The likely answer is no. If we understand canvassing as a form of speech or association, as courts are likely to do, then it stands to reason from the Supreme Court’s jurisprudence that candidates and parties have a right not to speak or associate with people of their choosing. That raises the following dilemma: If campaign mobilization activities are critical to broader democratic participation and to preventing the vicious cycle of marginalization and alienation, and campaigns cannot constitutionally be required to serve this function, then what can be done?

We argue that government actors should look to incentivizing campaigns to extend the reach of their mobilization activities to nonparticipants in the political process. The effectiveness of these government incentives will depend on the very data about individuals’ voting history that has contributed to the campaign contact gaps between the rich and the poor that we have explored in this Article. The incentives could take on various forms. We provide some general contours of a few possibilities—some currently in use and some more aspirational—that we hope will spark debate and further research. Ultimately, it would be necessary to balance the equality‑promoting features of these incentives with other considerations, such as rising polarization, alternative voting systems (e.g., ranked‑choice voting), and the growing costs of political campaigns.

We note at the outset that the incentive programs we outline below are closely tied to campaign finance law. The most obvious reason for this connection is that any monetary incentives will necessarily implicate campaign finance. But a wealth‑gap disparity in campaign finance is also correlated with disparities in public policy: a rich‑poor gap in donors will be reflected in the candidate selection process and ultimately the policies that are adopted. Incentivizing campaigns to extend their fundraising operations can have important downstream effects since donors are likely to vote, but the money itself has important implications for shaping public policy in a more representative, and thus equal, fashion as well.

Indirect incentives

First, states might offer incentives that indirectly push campaigns to broaden their outreach efforts, for example, with a program where states or cities send a “democracy voucher” to individuals on inactive voter lists. These vouchers, when given to a campaign, would be redeemed for cash. Vouchers remove an important barrier to giving (i.e., the need to open up one’s own wallet) and hence might motivate some individuals to participate who would otherwise prefer not to spend their own money on politics. Perhaps more importantly, vouchers may motivate campaigns to reach out to communities they previously ignored, notably poorer neighborhoods, knowing that “free money” is available and all they have to do is ask. Not only would such outreach increase the size and overall representativeness of the donor pool, but it would hopefully have the downstream effect of turning these new donors into new voters.

Voucher programs have been on the radar of campaign finance reformers since at least the 1990s and are having a moment in contemporary politics. In January 2021, Congress introduced the For the People Act which would provide funds to support voucher pilot programs for congressional elections in three states. In order to be eligible, states would provide a single $25 voucher to any individual who requests one, and Congress would reimburse the states the cost of the voucher plus reasonable administrative costs.

Perhaps more on point, the city of Seattle adopted a voucher program in 2015. According to the rules of Seattle’s Democracy Voucher Program, every registered voter is sent four $25 vouchers that they can give to any candidate running for municipal office who meets certain requirements. The primary motivation of the Seattle voucher system is to increase the number of individuals contributing to campaigns and to reduce the influence of special interest groups. Unfortunately, an evaluation of Seattle’s voucher program finds a very modest impact on participatory equality. According to the study, “[a]lthough participation increased across demographic groups, the largest gains were concentrated among white, higher‑income and older residents, as well as those who vote regularly in elections.” Thus, it seems unlikely that $100 in potential vouchers distributed widely is a sufficient incentive to spur campaigns to engage in systematic and widespread outreach to the poor and other marginalized constituents.

Instead of distributing vouchers to all registered voters, voucher programs are likely to be more successful in promoting equality if they are distributed only to inactive or first‑time voters. Candidates are going to find the money. If everybody possesses a voucher, campaigns have proven that they will tap existing networks to scoop up those free vouchers that are most accessible without the additional expense of expanding their search. On the contrary, if vouchers are only available to those who ask for them (as they would be under the currently proposed, federally‑funded pilot programs), then by definition vouchers will not result in the mobilization of those who are not already engaged in politics. Thus, in order for voucher programs to effectively incentivize a campaign‑led march toward participatory equality in American politics, voter history data must play a central role in the prioritization and distribution of vouchers to inactive or first‑time voters.

One alternative to vouchers would be for government to match small donations to campaigns. The logic behind a matching program is that campaigns that otherwise ignore small donors may find it cost‑effective to engage more people and build a larger donor base of small donors if small donations represented a more significant amount of money. For example, in New York City, any contribution up to $175 to a candidate for city office is matched by the city at a 6‑to‑1 ratio (i.e., a $100 contribution results in a $700 deposit in the campaign’s account). Like voucher programs, matching programs aim to not just increase and diversify the donor pool, but to increase and diversify the voting population as well. Early research suggests that New York City’s matching program has significantly increased the donor pool in poorer neighborhoods by more than 1,000%. Owing to this success, Congress has proposed funding a 6‑to‑1 match for contributions up to $200 for all congressional elections. Despite such gains in diversifying the donor pool, we note that the success of New York City’s small‑dollar matching program may be relatively modest overall. Only a fraction of Americans contribute money to political campaigns, and only a fraction of this fraction are poor or minorities. Thus, a 1,000% increase in the participation of poor, minority, and otherwise marginalized individuals in a small‑donor matching program, while impressive, may translate into an expanded donor pool that represents just one to 2% of Americans. Participatory equality demands more.

As with vouchers, small‑donor matching programs may not result in greater equality without more effort. The devil is in the details. For example, small‑donor matching programs frequently ignore the fact that the majority of individuals who are identified as “small donors” are actually repeat players who spend tens of thousands of dollars in small chunks. Thus, simply matching small donations, as opposed to rewarding small donors, may simply perpetuate existing inequality. The most successful matching program would go further by seeking to subsidize first‑time donors or first‑time or inactive voters. Robust and accurate voter history data would be necessary to calibrate matching programs in this way.

Direct incentives

Whereas voucher programs and small‑dollar matching programs might indirectly incentivize campaigns to expand their voter mobilization activities, such effects may ultimately be inadequate to the task. Achieving full participatory equality is likely to require some out‑of‑the‑box thinking. In this section we introduce a set of novel ideas that take a more direct approach. If participatory equality is the goal, why match small contributions and then hope that candidates will reach out to more donors or voters? Why not simply reward candidates who reach out to the most first‑time donors? Imagine a program where candidates compete for cash grants awarded to the single candidate who reports raising money from the most first‑time donors. For example, consider an average campaign for Congress. The average price of a winning House campaign is just over $2 million. Now imagine a pot of money set aside for “equality grants.” Between January and November of a federal election year, candidates are required to disclose their finances to the Federal Election Commission quarterly in January (for the prior year), March, June, and September. Imagine a program that awards a grant of, say, $100,000 to the candidate who reports the largest number of first‑time donors (regardless of the size of the candidate’s overall war chest). The size of the grant and the frequency of the award would need to be fine‑tuned to entice participation by candidates. For example, candidates may prefer smaller but more frequent grants (requiring a change to current disclosure rules) or graduated grants that provide more money early in the process.

The point is that regardless of the specific details, a direct incentive program is more likely to align the short‑term interest of campaigns (in winning the next election) with the long‑term public interest (in political equality). And voter history data would be crucial for a competition to prospect for the most first‑time donors, as they provide campaigns a sense of the individuals (and, when matched with other campaign data, the supporters) who are most likely to never have given a campaign contribution before.

Both Congress and states have experimented with cash grants in the past, first by providing a spending limit increase to candidates who compete against wealthy, self‑funded opponents (the so‑called “Millionaire’s Amendment” to the Bipartisan Campaign Reform Act of 2002) and later by providing cash grants to candidates who face off against opponents who are privately funded (the so‑called “trigger provision” of the Arizona Citizens Clean Elections Act). The Supreme Court was skeptical of both cash grant programs, striking down the Millionaires’ Amendment in Davis v. FEC and the trigger provision in Arizona Free Enterprise Club’s Freedom Club PAC v. Bennett. In both instances, the cash grants were intended to level the playing field between candidates as a way to bolster public financing of campaigns by avoiding competition that would render public financing either toothless or moot. In both cases, a 5–4 Court interpreted the cash grants as punishments against successful candidates, thus triggering First Amendment scrutiny. And in both cases the Court rejected the argument that providing a level playing field was a compelling state interest.

Cash grants that reward candidates for donor or voter outreach are categorically different than the cash grants invalidated by the Court and would likely survive a legal challenge. First, the motivation for rewarding donor outreach is to magnify the political speech of as many people as possible, not to stifle or mute the relative voice of any one person. Because the Court has analogized campaign contributions to a type of political speech, a grant program that incentivizes the production of this speech could hardly be considered a simultaneous burden on that speech.

Beyond a worry about stifling speech, Chief Justice John Roberts expressed a series of practical concerns about the cash grant program in Arizona. For example, he pointed to the perverse incentive for publicly financed candidates to engage in less spending (and thus political speech) as their opponents get close to the triggering threshold, expecting the trigger to kick in. Roberts also pointed to the “multiplier effect,” where a single candidate funded by private contributions could trigger a match for several opposing candidates at the same time. None of those concerns exist for a donor‑recruitment award program. There are no perverse incentives to reach out to fewer donors or to spend less on donor outreach, and there is no multiplier effect as cash grants would be awarded to the single candidate with the most reported new and unique donors. Similarly, Roberts’s framing of the trigger provision as a “penalty” on private candidates (and Justice Samuel Alito’s similar framing in Davis with respect to wealthy self‑funded candidates) does not apply to a donor‑recruitment award program. In Davis and Arizona Free Enterprise, candidates who raised more funds were not able to reap the rewards of that success. In a donor‑recruitment program, the candidate who is doing better by reaching out to the most donors is the one who is rewarded with the cash grant, not the trailing candidate who might prefer a financial boost.

A donor‑recruitment award program could easily translate into an even more direct incentive program for mobilizing voters. Instead of rewarding candidates who raise money from the most first‑time or unique donors, cash grants can be made available as rewards to political parties that increase their rate of turnout. For example, any political party that is able to mobilize turnout higher than its historical average (say, by averaging the most recent three comparable elections) would be eligible for a large cash grant. If the average Republican turnout in a state over the past three midterms is 2.5 million voters, then the Republican Party would earn a substantial cash grant if it were able to attract more than 2.5 million voters to vote for the Republican candidate(s).

Much like vouchers and matching programs, the policy design details of a donor or voter‑recruitment award program will matter a great deal. And competing values will need to be taken into account. For example, a consistent finding in the political science literature is that small donors tend to be more polarized than large donors, and far more polarized than the general population. Thus, reforms that aim to increase the number of small donors may unwittingly contribute to political polarization. This tension dissipates as the number of small donors increases, but only if the increase is substantial. Our primary contention is not that participatory equality should always predominate when in tension with other competing values, only that it be actively factored into all electoral regulatory reform.

Curbing mirotargeting on social media

In addition to legal efforts to mitigate democratic inequality, social media companies have an important role to play as well. The rise of social media has ushered in an era of “cheap speech.” The ability of all citizens, rich or poor, to distribute messages to each other has increased faster than at any time in Earth’s history. As a result, we might have predicted that social media platforms would have a democratizing effect by eliminating the gatekeepers that filter the news we consume and the messages that we receive. On the contrary, cheap speech has contributed both to the polarization of American politics and to the growing inequality in political participation. Political campaigns’ strategies have both responded to, and contributed to, these pathologies. In particular, as we discussed above, the rise of microtargeting strategies to canvass the electorate is correlated with the rich‑poor gap in campaign contributions, the rich‑poor gap in mobilization, and the rich‑poor gap in voting. Although the focus of this Article is largely on the data that make microtargeting possible, particularly voter history data, the role of social media technologies and policies deserves some attention. Whereas the centralization and digitization of voter history and other election data provided campaigns with the ability to identify their likely supporters, social media platforms facilitated the cost‑effective means to actually reach out to these supporters. As it turns out, “cheap speech” applies to campaigns much the same way that it has for individuals. Not only has social media opened the doors for candidates to project their message far and wide, but it has allowed them to precisely target that message to sympathetic and like‑minded people, and only sympathetic and like‑minded people.

For their part, social media companies have largely taken a laissez‑faire approach to political advertising, even in the wake of the 2020 election that featured historic levels of misinformation, anti‑government mobilization, and ultimately an insurrection at the U.S. Capitol. Only after the violent attack on January 6, 2021 did Twitter, Facebook, and Google suspend or ban the accounts of President Trump and hundreds of others. Notably, other social media platforms, such as Parler and Reddit, stepped in to fill the void, at least temporarily. But it was the policies of these platforms before the election that hold broader implications for democratic equality going forward.

In the lead‑up to the 2020 election, the three largest social media companies adopted different political advertising policies. In October 2019, Twitter CEO Jack Dorsey announced that Twitter had “made the decision to stop all political advertising on Twitter globally.” The motivation for this policy was that political internet advertising “brings significant risks to politics.” According to Dorsey, “Internet political ads present entirely new challenges to civic discourse: machine learning‑based optimization of messaging and micro‑targeting, unchecked misleading information, and deep fakes. All at increasing velocity, sophistication, and overwhelming scale.” In addition to banning candidates from buying ads, Twitter also limited independent groups from targeting their ads. Ad buyers could identify the state where an ad should run but were prohibited from targeting users based on zip code or important politically-relevant keywords. In the months that followed, a handful of other digital companies followed suit.

At the other extreme, Facebook declined to ban political ads or to revise its ad‑targeting services. Pointing to the importance of transparency, Facebook lamented the lack of generally applicable regulation related to political ads, arguing in a policy statement, “we don’t think decisions about political ads should be made by private companies.” “In the absence of regulation,” the policy statement continued, “Facebook and other companies are left to design their own policies. We have based ours on the principle that people should be able to hear from those who wish to lead them, warts and all, and that what they say should be scrutinized and debated in public.” Absent from this statement was the acknowledgment that many individuals don’t actually see political ads from prospective leaders, because those messages are targeted to an increasingly manufactured audience. Notably, Facebook updated its policy as the election drew near and misinformation began to spread about the election itself. On September 3, 2020, Facebook announced that it would ban all new political ads on the platform during the final two weeks before the election and label as misleading any premature calls of the election outcome. The company continued to strictly regulate political ads in the weeks following the election, allowing some advertising related to the Georgia Senate runoff race in early January, but otherwise remaining noncommittal on its long‑term plans.

In between these two extremes, Google updated its policies to limit the targeting of political ads to a small set of segments: age, gender, and zip code. Under the new policy, campaigns are no longer able to target based on voting history or political leanings, among other categories. Google based its policy on the approach taken by traditional media (television, radio, and print), which lack the capability to segment their audience into smaller buckets.

Social media giants’ varied approaches to political ads highlight the lack of uniform regulation and the need for monetary incentives that might drive social media platforms to similar policy outcomes. Although the “cheap speech” revolution was supposed to usher in a new age of extreme democratization, it turns out to be quite tricky to leverage the innovative political opportunities of digital media without undermining the stability of organizations—political parties, trusted independent media—that are required for a healthy democracy.

In our estimation, a ban on political advertising on social media platforms is a step too far. The risks associated with microtargeting, misinformation, deep fakes, and manufactured influence (i.e., bots) are serious, and they have already proven disruptive to democratic elections across the world. We are nonetheless optimistic that the promise of democracy‑enhancing, data‑driven political mobilization is still within reach. In addition, government regulation of social media is normatively unappealing. The largest social media platforms are private-sector companies, and the services they provide implicate one of the most fundamental rights enshrined in the Constitution: the right to free speech. As such, current federal law exempts social media platforms from some of the liability regimes that apply to more traditional publishers, which removes one powerful inducement for self‑regulation.

Social media is like any tool: used properly it can improve our lives. Digital advertising is far less expensive and far more likely to reach a broader audience than traditional media alone. As a result, social media has given a voice to those who are otherwise unheard and has facilitated organic dialogue about politics between candidates and constituents and, perhaps more importantly, among constituents themselves. All of these benefits of social media are at risk if more sinister uses are not impeded. And history has provided plenty of evidence to suggest that a completely unregulated environment is not up to the task.

In addition to the incentive programs outlined above, which we believe hold the promise of creating demand for democracy‑enhancing uses of social media, social media platforms bear some responsibility for regulating the supply of these uses as well. Most importantly, social media companies must limit the capability of campaigns to segment their target audience beyond a few general categories. For example, the policy adopted by Google for ads that run on YouTube limits these categories to age, gender, and zip code. These categories are stingy, but they prevent candidates from directly targeting ads based on voter history, partisanship, race, sexual orientation, income, church attendance, or any other number of attributes that a campaign may hope to target.

Social media platforms should also prohibit the use of “custom audiences” for political ads, even if this service is available for commercial advertising. There are two types of custom audiences. The first allows candidates to provide lists of specific user accounts to the platform, which then runs ads only to those users. Because of the big data revolution, campaigns often already have a good idea about the partisanship, race, sexual orientation, etc., of potential voters, and so permitting exact user matches effectively allows targeting on these characteristics. The second type of custom audience is known as a “look‑alike” sample. Instead of matching exact user accounts, a look‑alike sample is algorithmically generated based on the traits of a test group provided by the campaign. While lacking the precision of exact matching, look‑alike sampling still provides campaigns a simple way to circumvent limited audience segments. For example, if a campaign provides a test group of accounts that only includes Hispanic profiles, the campaign will be confident that the look‑alike sample will predominantly feature Hispanic accounts.

Limiting target audience segments cuts against the full potential of social media. Requiring campaigns to run ads to more general audiences may dilute their message and diminish their return on investment, as some users who see their message will be unlikely to support them. But this is exactly the point. By prohibiting campaigns from speaking directly to very narrow segments of society, inappropriate messages (e.g., racial appeals, misinformation) are less likely to remain hidden and thus less likely to be made in the first place. More importantly, making political advertising slightly less efficient introduces a tradeoff: Campaign messages may be slightly less effective, but the public will be slightly more exposed to opinions and political appeals outside of their carefully‑curated social media bubbles. Such exposure, we believe, is crucial for preventing the radicalization of those who feel politically marginalized and an important first step for reclaiming the civil dialogue necessary to save American politics.


Reforms targeted at the current participatory inequalities in our democracy might seem unrealistic on their face. The question that any observer might ask is what would the motivation be for elected officials to pass laws for the benefit of the politically marginalized, when the very fact of their marginalization renders them politically irrelevant for electoral purposes?

Our optimism is rooted in Americans’ selfless commitment to larger social goals and fundamental ideals, values that prompted elected officials to address poverty in the 1960s, homelessness in the 1980s, and the lack of health insurance in the 2010s. But it is also rooted in the more selfish regard Americans must have for the preservation of our democracy.

Including the poor could be critical to all voters’ wellbeing and that of U.S. democracy itself. The political marginalization of the poor introduces broader systemic problems of the sort that have undermined democracies throughout the world and may very well be undermining American democracy today. Constructing a more inclusive political system will therefore benefit everyone who seeks to live in a sustainable representative democracy, not just those who are currently marginalized.


We are extremely grateful to Katy Glenn Bass, Jameel Jaffer, Amy Kapczynski, Mia Speier, Madeline Wood, and participants at the “Data and Democracy” symposium at the Knight First Amendment Institute at Columbia University for helpful comments and feedback. Many thanks to Andrew Gelfand, Mili Pilipiak, Daniel Bialer, and the editorial team at the Cornell Law Review for their diligent work in improving the organization and flow of the article. We are also grateful to Mike Miller for helpful feedback, criticism, and suggestions for presenting this empirical work.


© 2024, Bertrall L. Ross II and Douglas M. Spencer.


Cite as: Bertrall L. Ross II and Douglas M. Spencer, Voter Data, Democratic Inequality, and the Risk of Political Violence, 24-02 Knight First Amend. Inst. (Jan. 8, 2024), [].


See, e.g., Eleanor Flexner, Century of Struggle: The Woman’s Rights Movement in the United States 145–48 (rev. ed. 1975) (describing the women’s suffrage movement); Anna L. Harvey, Votes Without Leverage: Women in American Electoral Politics, 1920–1970, at 46, 9–14 (1998) (describing the policy victories that women achieved after they were granted suffrage); Ari Berman, Give Us the Ballot: The Modern Struggle for Voting Rights in America 39–64 (2015) (describing the implementation of the Voting Rights Act of 1965); Carol Anderson, One Person, No Vote: How Voter Suppression is Destroying Our Democracy 1–43 (2018) (describing the history of voter suppression in the United States).

See Gender Differences in Voter Turnout, Ctr. for Am. Women & Pol., [] (last visited Apr. 27, 2021) (“In every presidential election since 1980, the proportion of eligible female adults who voted has exceeded the proportion of eligible male adults who voted.”); Thom File, U.S. Census Bureau, The Diversifying Electorate— Voting Rates by Race and Hispanic Origin in 2012 (and Other Recent Elections) 3–4 (2013), [] (showing that Black eligible voter rates nearly equaled white eligible voter rates in the 2008 election and Black eligible voter rates exceeded white eligible voter rates in the 2012 election); Bernard L. Fraga, The Turnout Gap: Race, Ethnicity, and Political Inequality in a Diversifying America 39–42 (2018) (showing a decline in presidential elections of the turnout gap between whites and Asian Americans and Latinos mostly due to a decline in white turnout but also a persistently strong turnout gap between whites and Asian Americans and Latinos in midterm elections). But see Kevin Morris & Coryn Grange, Large Racial Turnout Gap Persisted in 2020 Election, Brennan Ctr. for Just. (Aug. 6, 2021), [] (finding the reemergence of a Black‑white turnout gap and a growing minority‑white turnout gap in the 2020 election).

Kristen Bialik, For the Fifth Time in a Row, the New Congress is the Most Racially and Ethnically Diverse Ever, Pew Rsch. Ctr. (Feb. 8, 2019), [] (describing how the 116th Congress that convened in January 2019 was the most racially and ethnically diverse ever, with 116 racial minorities elected to Congress, comprising 22% of that body); Carrie Blazina & Drew DeSilver, A Record Number of Women Are Serving in the 117th Congress, Pew Rsch. Ctr. (Jan. 15, 2021), [] (describing how the 117th Congress that convened in January 2021 is the most gender diverse, with 120 women elected to Congress, comprising 27% of that body).

See John D. Griffin & Brian Newman, Minority Report: Evaluating Political Equality in America 59–115 (2008) (showing through various empirical tests that racial minorities are the beneficiary of disproportionately fewer favorable policies and receive less representation from legislators than whites).

Bialik, supra note 3 .

Gregory Krieg, It’s Official: Clinton Swamps Trump in Popular Vote, CNN (Dec. 22, 2016), [] (reporting that Democratic presidential nominee Hillary Clinton secured 2.9 million more votes nationwide than Republican presidential nominee Donald Trump in the 2016 presidential election). Sarah Palin was the vice presidential candidate to Republican nominee John McCain in 2008 and Kamala Harris was the vice presidential candidate to Democratic nominee Joe Biden in 2020.

Blazina & DeSilver, supra note 3 ; Women in State Legislatures for 2020, Nat’l Conf. State Legislatures (Jan. 7, 2020), [] (“Approximately 2,145 women serve in the 50 state legislatures in 2020, making up 29% of all state legislators nationwide.”).

See Jan E. Leighley & Jonathan Nagler, Who Votes Now? Demographics, Issues, Inequality, and Turnout in the United States 1, 6 (2014) (noting that income bias in voter turnout has been stable since 1972).

Voting and Registration in the Election of November 2016: Table 7. Reported Voting and Registration of Family Members, by Age and Family Income: November 2020, U.S. Census Bureau (2021), [].

Voting and Registration in the Election of November 2018: Table 7. Reported Voting and Registration of Family Members, by Age and Family Income: November 2018, U.S. Census Bureau (2019), [].

See Bertrall L. Ross II, Addressing Inequality in the Age of Citizens United, 93 N.Y.U. L. Rev. 1120, 1166–69 (2018) (describing the scholarly shift toward a mobilization theory of turnout).


See Bertrall L. Ross II & Douglas M. Spencer, Passive Voter Suppression: Campaign Mobilization and the Effective Disfranchisement of the Poor, 114 Nw. U. L. Rev. 633, 688 (2019) (deriving a calculus of contact from campaign mobilization behavior in which voting history and propensity influence campaign decisions about who to contact); see also Wendy K. Tam Cho & James G. Gimpel, Prospecting for (Campaign) Gold, 51 Am. J. Pol. Sci. 255, 266 (2007) (identifying a similar calculus in the campaign finance context).

Ross II & Spencer, supra note 13 , at 690.

Eitan D. Hersh, Hacking the Electorate: How Campaigns Perceive Voters 64 (2015) (“The state‑by‑state implementation of statewide registration databases, digitally stored and constantly maintained, lays the groundwork for the individual‑level targeting that dominates contemporary campaigns.”).

Daniel A. Cox, After the Ballots are Counted: Conspiracies, Political Violence, and American Exceptionalism: Findings from the January 2021 American Perspectives Survey, Surv. Ctr. on Am. Life (Feb. 11, 2021), [].


See, e.g., Tazreena Sajjad & Anders C. Härdig, Too Many Enemies: Mobilization, Marginalization, and Political Violence, 29 Terrorism & Pol. Violence 1106, 1107 (2017) (The authors “employ a concept [they] call ‘complex hyperpolarization’ . . . to explain a socio‑political context where the space for dialogue is narrow and the risk for violence is high.”); Nicholas Sambanis, Poverty and the Organization of Political Violence, Brookings Trade F. 165, 165 (2004) (“There is an emerging consensus in the literature that a low level of income is a significant or even necessary condition for some forms of political violence, such as civil war or coups.”).

Larry Diamond, Lee Drutman, Tod Lindberg, Nathan P. Kalmoe & Lilliana Mason, Americans Increasingly Believe Violence is Justified if the Other Side Wins, Politico (Oct. 1, 2020), [].

See, e.g., Andrew H. Hales & Kipling D. Williams, Marginalized Individuals and Extremism: The Role of Ostracism in Openness to Extreme Groups, 74 J. Soc. Issues 75, 83–84 (2018) (finding that “ostracism increases interest in extreme groups”).

See Michael S. Schmidt & Luke Broadwater, Officers’ Injuries, Including Concussions, Show Scope of Violence at Capitol Riot, N.Y. Times (July 12, 2021), []; Jesse Rodriguez & Rebecca Shabad, Trump Defends Jan. 6 Rioters’ ‘Hang Mike Pence’ Chant in New Audio, NBC News (Nov. 12, 2021), [].

See Justin Gest, Commentary: Donald Trump Takes Working‑Class Whites Down with Him, Reuters (Nov. 2, 2016), [].

Robert A. Pape & Keven Ruby, The Capitol Rioters Aren’t Like Other Extremists, Atlantic (Feb. 2, 2021), [] (“What’s clear is that the Capitol riot revealed a new force in American politics—not merely a mix of right‑wing organizations, but a broader mass political movement that has violence at its core . . . .”).

Todd C. Frankel, A Majority of the People Arrested for Capitol Riot Had a History of Financial Trouble, Wash. Post (Feb. 10, 2021), [].


An Act for the Further Regulating Elections of Members to Serve in Parliament and for the Preventing Irregular Proceedings of Sheriffs and Other Officers in the Electing and Returning Such Members 1695–96, 7 & 8 Will. III c. 25, [] [hereinafter Elections of Members].

See Tim Harris, Politics Under the Later Stuarts: Party Conflict in a Divided Society 1660–1715, at 162–63 (1993) (describing the series of parliamentary measures in the 1690s aimed at preserving the independence of Parliament from the Crown).

See Bertrall L. Ross II, Challenging the Crown: Legislative Independence and the Origins of the Free Elections Clause, 73 Ala. L. Rev. 221, 281–89 (2021) (describing Parliament’s efforts to defend itself against Crown influence over parliamentary elections after the Glorious Revolution).

See J. R. Jones, The Revolution of 1688 in England 129–30 (1972) (“Of all [the] domestic policies, the campaign to pack Parliament was easily the most important in provoking the Revolution . . . .”).

See David L. Smith, The Stuart Parliaments 1603–1689, at 163 (1999) (providing the chronology of events leading to King James’ abdication of the throne).

The Bill of Rights 1689, 1 W. & M. c. 2.

See Mark A. Kishlansky, Parliamentary Selection: Social and Political Choice in Early Modern England 117 (1986) (explaining the role of the sheriff in the electoral process).

See id.

See John K. Gruenfelder, Influence in Early Stuart Elections 1604–1640, at 14–15 (1981) (describing the sheriff’s power to manipulate the outcome of elections).

See Elections of Members, supra note 26 , at sec. III.

Jeremy Gibson & Colin Rogers, Poll Books c. 1696– 1872: A Directory to Local Holdings in Great Britain 3–4 (2nd ed. 1989) (describing the content of poll books).

See Carol Speirs, Poll Books, U. Reading Special Collection Servs. 1–3 (2010), [] (describing the role of poll books to counter election fraud); Edmund Green, Penelope Corfield & Charles Harvey, Elections in Metropolitan London: 1700-1850: Arguments and Evidence 6 (2013) (identifying the use of poll books as evidence in contested elections).

See England Poll Books 1694 to 1872, supra note 35 .

See Alexander Keyssar, The Right to Vote: The Contested History of Democracy in the United States 67 (rev. ed. 2009); see also Joseph P. Harris, Registration of Voters in the United States 65–88 (1929) (providing an extensive early history of the development of voter registration generally and in specific states).

See Daniel P. Tokaji, Voter Registration and Election Reform, 17 Wm. & Mary Bill Rts. J. 453, 456–57 (2008) (describing the process by which local government officials compiled registration lists in the 19th century).

Id. at 457; see Harris, supra note 40 , at 38–45 (showing that Harris recommended these and other additions to voter registration records in his influential report on voter registration in the United States).

Harris, supra note 40 , at 116.

Id. at 11.

Id. at 11–13.

Id. at viii.

See id. at xi (“This volume brings together the results of 15 months of field study, during which the author visited over 30 states, and practically every large city in this country.”).

Id. at 15, 25–26 (“There is no other single feature of registration systems which is as effective [as signature verification] in preventing voting frauds.”).

Id. at 53–54.

Id. at 43.



Id. at 44.


Joseph P. Harris, Model Voter Registration System (partial rev. 4th ed. 1957).

Id. at 27.

Id. (“Despite the fact that the registers under permanent registration usually include the names of many voters who no longer reside at their registered addresses . . . this has not led to voting frauds.”).

Id. at 14.


Id. at 35–36.

Rasmus Kleis Nielsen, Ground Wars: Personalized Communication in Political Campaigns 144–45 (2012) (describing early campaign mobilization practices).

See Kathleen Hall Jamieson, Packaging the Presidency: A History and Criticism of Presidential Campaign Advertising 34–35 (3d ed. 1996).

See infra Subpart I.B.

Raymond E. Wolfinger & Steven J. Rosenstone, Who Votes? (1980).

Id. at 71–72.

Id. at 78–79.

Frances Fox Piven & Richard A. Cloward, Why Americans Don’t Vote 180 (1988).

Id. (“Contemporary voter registration obstacles . . . function as de facto equivalents of the poll tax, literacy test, and other class‑ and race‑oriented restrictions on the suffrage of an earlier era.”).

See Robert A. Jackson, Robert D. Brown & Gerald C. Wright, Registration, Turnout, and the Electoral Representativeness of U.S. State Electorates, 26 Am. Pol. Q. 259, 269–72, 281 (1998) (finding evidence supportive of the relationship between registration restrictions and low turnout but acknowledging that registration reform is no “panacea”). But see Jonathan Nagler, The Effect of Registration Laws and Education on U.S. Voter Turnout, 85 Am. Pol. Sci. Rev. 1393, 1402–03 (1991) (finding in a replication of the Wolfinger and Rosenstone study that after the addition of an interactive term linking education and registration closing date, the effect of the registration closing date on turnout disappears).

National Voter Registration Act of 1993, 52 U.S.C. §§ 20501–20511.

See Kathleen Hale, Robert Montjoy & Mitchell Brown, Administering Elections: How American Elections Work 78 (2015) (“The NVRA created new opportunities for citizens to register and changed the ways in which states could maintain voter registration files.”).

Tokaji, supra note 41 , at 468.

52 U.S.C. § 20507(d).

See U.S. Election Assistance Comm’n, The Impact of the National Voter Registration Act of 1993 on the Administration of Elections for Federal Office 2011–2012, at 13–29 (2013), [] (documenting state collection of voter registration history that included the number and percentage of active and inactive registered voters).

See Nat’l Comm’n on Fed. Election Reform, To Assure Pride and Confidence in the Electoral Process 28–29 (2001), [] (“The NVRA effectively forced every state to offer voter registration in combination with the single civic act performed almost universally by American adults—obtaining a driver’s license.”); Nat’l Rsch. Council, Improving State Voter Registration Databases: Final Report 9–16 (2010) (describing the list maintenance responsibilities of the states under the NVRA).

Nat’l Rsch. Council, supra note 76 , at 1–2 (describing the process of transmitting voter registration information from public agencies to registration offices).

The 15 states that established voter registration systems between 1993 and 2001 include Alabama, Alaska, Georgia, Indiana, Iowa, Kansas, Kentucky, Louisiana, Massachusetts, Michigan, Missouri, South Carolina, South Dakota, Utah, and Virginia. The five states that established statewide voter registration systems prior to the adoption of the NVRA include Colorado, Hawaii, Minnesota, New Hampshire, and Oregon.

See U.S. Election Assistance Comm’n, supra note 75 , at 13–29 (providing the results of a survey of states regarding the data contained in state voter registration systems).

Caltech/MIT Voting Tech. Project, Voting: What Is, What Could Be 27 (2001), [].

See Charles J. Pults, America’s Data Crisis: How Public Voter Registration Data Has Exposed the American Public to Previously Unforeseen Dangers and How to Fix It, 105 Iowa L. Rev. 1363, 1371–72 (2020) (linking the 2000 election controversy with Congress’ decision to enact the Help America Voter Act).

See H.R. Rep. No. 107‑329, pt. 1, at 35–36 (2001) (citing calls from the Federal Election Commission and the National Commission on Federal Election Reform for the creation of statewide computerized voter registration databases).

52 U.S.C. §§ 21041, 21083; Arthur L. Burris & Eric A. Fischer, Cong. Rsch. Serv., RS20898, The Help America Vote Act and Election Administration: Overview and Selected Issues for the 2016 Election 15 (2016).

52 U.S.C. § 21083(a)(1)(A).

Id. at § 21083(a)(1)(A)(v).

Note that this provision of HAVA does not apply in states with no voter registration requirements. Id. at § 21083(a)(1)(B).

Statewide Voter Registration Systems, U.S. Election Assistance Comm’n (Aug. 31, 2017), [] (describing the different ways in which the states have implemented HAVA’s requirements for statewide voter registration systems).

See Ross & Spencer, supra note 13 , at 690–93 (providing an account of the centralized means by which states collected voting history data).

See Hale, Montjoy & Brown, supra note 72 , at 81 (finding gains in registration after the NVRA did not translate into increases in voter turnout or “significant changes in the socio-economic composition of the voting public”).

See Caltech/MIT Voting Tech. Project, supra note 80 , at 9 (noting “1.5 to 3 million” votes were lost because of registration errors).

See Hale, Montjoy & Brown, supra note 72 , at 81.

See Nielsen, supra note 62 , at 17–19 (describing the shift in thinking about the cost‑effectiveness of campaign television advertisements).

See id. at 41–42 (describing this experiment).

Alan S. Gerber & Donald P. Green, The Effects of Canvassing, Telephone Calls, and Direct Mail on Voter Turnout: A Field Experiment, 94 Am. Pol. Sci. Rev. 653, 660 (2000); see also Hale, Montjoy & Brown, supra note 72 , at 81 (summarizing the finding of no effect on the removal of registration barriers on turnout).

See Joel A. Middleton & Donald P. Green, Do Community‑Based Voter Mobilization Campaigns Work Even in Battleground States? Evaluating the Effectiveness of MoveOn’s 2004 Outreach Campaign, 3 Q.J. Pol. Sci. 63, 64 (2008) (describing the dramatic increase in scale of get‑out‑the‑vote activities during the 2004 presidential election).

See Robert Huckfeldt & John Sprague, Citizens, Politics, and Social Communication: Information and Influence in an Election Campaign 236–37 (1995) (providing evidence of geography-based canvassing employing neighborhood or precinct voting data in the period prior to the study); Nielsen, supra note 62 , at 144–45 (describing the early return to mobilization strategies that relied on precinct‑based targeting of potential voters).

Donald P. Green & Alan S. Gerber, Get Out the Vote: How to Increase Voter Turnout 1–22 (2004). For an account of the influence of this book on campaigns, see Sasha Issenberg, The Victory Lab: The Secret Science of Winning Campaigns 209 (2012) (“The slim paperback became a vade mecum for organizers working in the ranks of groups who had responded to the era of partisan polarization by shifting their resources from persuasion to mobilization.”).

Green & Gerber, supra note 97 , at 26–45 (offering door‑to‑door canvassing strategies and experiment results supporting the cost‑effectiveness of microtargeting).

See Hersh, supra note 15 at 94.

See Costas Panagopoulos & Peter W. Wielhouwer, Polls and Elections: The Ground War 2000–2004: Strategic Targeting in Grassroots Campaigns, 38 Pres. Stud. Q. 347, 358–59 (2008) (describing how voting history data influenced campaigns’ choices about who to target during the 2000 and 2004 presidential campaigns).

See Ross II & Spencer, supra note 13 , at 682–84; see also Ross II, supra note 11 , at 1177.

See Ross II & Spencer, supra note 13 , at 684.

Id. at 679–81.



Id.; see also Bertrall L. Ross II & Su Li, Measuring Political Power: Suspect Class Determinations and the Poor, 104 Cal. L. Rev. 323, 345–50 (2016); Bertrall L. Ross II & Terry Smith, Minimum Responsiveness and the Political Exclusion of the Poor, 72 L. & Contemp. Probs. 197, 203–04 (2009).

For a discussion of disparate rates of public health outcomes and voting see Angela P. Harris & Aysha Pamukcu, The Civil Rights of Health: A New Approach to Challenging Structural Inequality, 67 UCLA L. Rev. 758, 780 (2020); Douglas M. Spencer, Lisa Grow Sun, Brigham Daniels, Chantel Sloan & Natalie Blades, Viral Voter Suppression (BYU Law Research Paper, Paper No. 21-13, 2021) (manuscript at 3–4), []. See also Sophie Chou & Tyler Dukes, In North Carolina, Black Voters’ Mail‑In Ballots Much More Likely to be Rejected Than Those From Any Other Race, ProPublica (Sept. 23, 2020) [] (reporting that “ballots mailed by Black voters during the midterms were more than twice as likely as those sent in by white voters to be rejected”); Anna Baringer, Michael C. Herron & Daniel A. Smith, Voting by Mail and Ballot Rejection: Lessons from Florida for Elections in the Age of the Coronavirus, 19 Election L.J. 289, 298–300 (2020) (showing that ballot rejection rates were 2–3 times higher for racial minority voters and young, firsttime voters).

Hersh, supra note 15 at 94.


See Ethan Roeder, Opinion, I Am Not Big Brother, N.Y. Times (Dec. 6, 2012), []. Mr. Roeder, the data director for Obama for America, wrote “How do we predict whether people are going to vote or not? We look at the voter file.” Id.

Importantly, because ballots are secret, there is no way to record the candidate of choice for any individual. The only record is whether the voter casts a ballot or not.

See Sharad Goel, Marc Meredith, Michael Morse, David Rothschild & Houshmand Shirani-Mehr, One Person, One Vote: Estimating the Prevalence of Double Voting in U.S. Presidential Elections, 114 Am. Pol. Sci. Rev. 456, 458–59 (2020).

See Susan Bysiewicz, State of Connecticut State Plan as Required by the Help America Vote Act, Pub. L. No. 107–252 § 253(b), at 2–3 (2003), [].

FVRS Voter Registration Procedures, Fla. Admin. Code Ann. r. 1S-2.039 (2012).

North Dakota does not require that individuals register in order to vote.

See generally Stephen Ansolabehere & Douglas Rivers, Cooperative Survey Research, 16 Ann. Rev. Pol. Sci. 307, 308–10 (2013) (describing the CES).

Id. at 309. We rely on variable v4065 in the 2006 survey and variable CC425a in the 2010–2016 surveys. See Stephen Ansolabehere, Brian Schaffner & Sam Luks, Guide to the 2016 Cooperative Congressional Election Survey 114 (Aug. 2017) [hereinafter 2016 Guide]; Stephen Ansolabehere, Guide to the 2006 Cooperative Congressional Election Survey 81 (Aug. 20, 2010) [hereinafter 2006 Guide].

This information is captured in the series of variables CC425b_1 through CC425b_4. 2016 Guide, supra note 117 , at 114–15.

Ross II & Spencer, supra note 13 , at 678.

Id. at 678 n.211.

2016 Guide, supra note 117 , at 114; National Election Studies 2000 Pilot: Codebook Variable Documentation, 271 (2000), []. In the first period, we rely on survey data from the American National Election Study (ANES) that preceded the CES. Questions about whether a respondent was contacted by a candidate campaign were only included on the ANES between 1978–1994. There are challenges with comparing two different sources of data, conducted decades apart. However, we note that the sampling frame was similar for both surveys (adult U.S. population) and the two surveys asked nearly identical questions to respondents about whether they had been contacted by a party or candidate campaign.

We first normalize income data on a 12‑point scale, from less than $10,000 per year to $120,000 or more. We then pool all survey respondents and cut respondents into five equally sized bins. While there is some variation in respondents’ income across years, the top and bottom income quintiles are very stable (i.e., our findings are identical when we cut respondents into five equally sized bins in each year). Furthermore, a 2012 Pew report found that the difference in non‑response rate between those in the top 20% by income and the lowest 20% by income was just 1%. See Assessing the Representativeness of Public Opinion Surveys, Pew Rsch. Ctr. (May 15, 2012), [].

Our sample includes 315,827 observations, which represents the total number individual survey respondents to the CES across all years, excluding Alaska, Louisiana, and North Dakota, which we exclude because there are no CES respondents in either the highest or lowest income quintile in at least one year in our sample. For all of the analysis presented below, we apply the common content weights provided with the CES to ensure that the sample properly reflects the sampling frame.

See Stephen Ansolabehere, Guide to the 2008 Cooperative Congressional Election Survey 19–93 (July 15, 2011).

The 16 states are Alabama, Arkansas, California, Connecticut, Florida, Maine, Maryland, Mississippi, Montana, Nevada, New York, North Dakota, Oklahoma, Oregon, Rhode Island, and Wyoming.

The 20 states are Arizona, Colorado, Georgia, Indiana, Iowa, Kentucky, Louisiana, Massachusetts, Michigan, Minnesota, Missouri, New Hampshire, North Carolina, Pennsylvania, South Carolina, South Dakota, Utah, Virginia, Washington, and Wisconsin.

Domenico Montanaro, Rachel Wellford & Simone Pathe, 2014 Midterm Election Turnout Lowest in 70 Years, PBS NewsHour (Nov. 10, 2014), [] (reporting a 27% decrease in turnout in Washington state, 25% in California, and 17% in both New York and New Jersey). New Jersey, California, and Washington all rank among the top ten states by median household income. See 2019 Median Household Income in the United States, U.S. Census Bureau (Sept. 17, 2020), [].

See, e.g., David Niven, The Mobilization Solution? Face‑to‑Face Contact and Voter Turnout in a Municipal Election, 66 J. Pol. 868, 869 (2004) (describing the promise of mobilization).

See David W. Nickerson, Is Voting Contagious? Evidence from Two Field Experiments, 102 Am. Pol. Sci. Rev. 49, 54–55 (2008) (finding individuals contacted by campaigns are both more likely to vote and also to pass on about 60% of their propensity to vote onto other persons in the household).

See Sidney Verba, Kay Lehman Schlozman, Henry Brady & Norman H. Nie, Citizen Activity: Who Participates? What Do They Say?, 87 Am. Pol. Sci. Rev. 303, 308–10 (1993) (describing the volume and form of citizen political activity).

See David E. Broockman, Distorted Communication, Unequal Representation: Constituents Communicate Less to Representatives Not of Their Race, 58 Am. J. Pol. Sci. 307, 307 (2014) (describing how politicians seek to be responsive to the constituents that they serve).

See Alan S. Gerber, Donald P. Green & Ron Shachar, Voting May Be Habit‑Forming: Evidence from a Randomized Field Experiment, 47 Am. J. Pol. Sci. 540, 548 (2003) (finding that “voting and nonvoting . . . appear to create behavioral patterns that persist over time”).

Henry E. Brady, Sidney Verba & Kay Lehman Schlozman, Beyond SES: A Resource Model of Political Participation, 89 Am. Pol. Sci. Rev. 271, 283 (1995) (finding a positive relationship between political interest and decision to vote); Sidney Verba, Would the Dream of Political Equality Turn out To Be a Nightmare?, 1 Persps. on Pol. 663, 675–76 (2003) (connecting disadvantaged citizens’ lack of interest in politics to economically stratified political mobilization).

See Leighley & Nagler, supra note 8 , at 137–38 (finding a significantly higher degree of political alienation for individuals in low‑income classes than in high‑income classes).

See Pape & Ruby, supra note 23 .

See Larry M. Bartels, Ethnic Antagonism Erodes Republicans’ Commitment to Democracy, 117 Proc. Nat’l Acad. Scis. 22752, 22756 (2020).

See Frankel, supra note 24 (“A majority of the people arrested for [the] Capitol riot had a history of financial trouble.”).

See Wolfinger & Rosenstone, supra note 65 , at 34–36 (finding a positive relationship between education, income, and voting).

See Ross II & Spencer, supra note 13 , at 685–86.

See, e.g., David Niven, The Limits of Mobilization: Turnout Evidence from State House Primaries, 23 Pol. Behav. 335, 343 (2001) (finding a small effect of door‑to‑door canvassing on the turnout of infrequent voters).

See Lisa García Bedolla & Melissa R. Michelson, Mobilizing Inclusion: Transforming the Electorate Through Get‑Out‑the‑Vote Campaigns 92–126 (2012) (describing the unique challenges associated with door‑to‑door canvassing in low‑income communities).

Id. at 125 (“[C]anvassing that takes advantage of existing social networks can be more effective than campaigns using non‑local walkers. It is th[e] sociocultural interaction on the doorstep . . . that makes this strategy so powerful and effective.”); see also Stacey Abrams, I Know Voting Feels Inadequate Right Now, N.Y. Times (June 4, 2020), [] (describing the canvassing strategy among marginalized communities by her gubernatorial campaign that “increased turnout to record numbers and engaged voters who never wanted to participate before”).

See Christopher S. Elmendorf & Abby K. Wood, Elite Political Ignorance: Law, Data, and the Representation of (Mis)Perceived Electorates, 52 U.C. Davis L. Rev. 571, 620 (2018).

One could imagine an algorithm that develops a likelihood of voting score based on characteristics found relevant to validated voting behavior from older registration lists.

National Voter Registration Act of 1993, 52 U.S.C. § 20507(b).


As election law scholars we acknowledge—and lament—that such a move might limit the ability to conduct research on the impacts of various public policies on turnout. We think it is important for policy analysts and scholars to understand the impact of new laws and regulations on the shape of the electorate. We believe that such scholarship could still be performed under limited conditions, but that any voter file matching may need to be anonymized before it is shared with researchers or the public.

See Elmendorf & Wood, supra note 143 , at 613 (“Asserting ownership of its voter‑file data, a state might ban licensees from ‘linking’ the voter file to estimates of voters’ political preferences generated from commercially acquired data.”).

E. E. Schattschneider, Party Government 1 (Holt, Rinehart and Winston 1960) (1942).

See, e.g., Nancy L. Rosenblum, On the Side of the Angels: An Appreciation of Parties and Partisanship 165–209 (2008) (describing the extensive anti‑partyism in the United States that began during the Progressive Era and extended through the first half of the twentieth century).

See The Federalist No. 10 (James Madison) (identifying factions as dangerous to popular government); see also Washington’s Farewell Address 1796, The Avalon Project: Documents in L., History & Dipl., [] (“[T]he common and continual mischiefs of the spirit of party are sufficient to make it the interest and duty of a wise people to discourage and restrain it. It serves always to distract the public councils and enfeeble the public administration. It agitates the community with ill‑founded jealousies and false alarms, kindles the animosity of one part against another, foments occasionally riot and insurrection. It opens the door to foreign influence and corruption, which finds a facilitated access to the government itself through the channels of party passions.”).

See Rosenblum, supra note 150, at 175–80 (describing the historical corruption and bossism of parties); Samuel Issacharoff & Richard H. Pildes, Politics as Markets: Partisan Lockups of the Democratic Process, 50 Stan. L. Rev. 643, 664–68 (1998) (describing the racism and exclusionary practices of political parties in the middle of the twentieth century).

See E. E. Schattschneider, The Semisovereign People: A Realist’s View of Democracy in America 100–01 (1960) (“The expansion of the electorate was largely a by‑product of the system of party conflict. The rise of the party system led to a competitive expansion of the market for politics.”).

See, e.g., Samuel Issacharoff, Private Parties with Public Purposes: Political Parties, Associational Freedoms, and Partisan Competition, 101 Colum. L. Rev. 274, 299 (2001) (“[V]ibrant political parties are indispensable intermediary organizations that organize citizens so as to bring competitive vitality to the electoral arena.”); Samuel Issacharoff & Daniel R. Ortiz, Governing Through Intermediaries, 85 Va. L. Rev. 1627, 1630 (1999) (identifying parties as “political intermediaries [that] intervene in order to make representation more direct and transparent”).

See, e.g., Issacharoff & Pildes, supra note 152, at 648 (arguing that the courts should be less focused on controlling “politics directly through the centralized enforcement of individual rights” and more focused on maintaining “the background structure of partisan competition” to protect representative democracy); Guy‑Uriel E. Charles, Racial Identity, Electoral Structures, and the First Amendment Right of Association, 91 Cal. L. Rev. 1209, 1254 (2003) (identifying from the Supreme Court’s political association cases a theory that “political winners and losers . . . ought to be the product of the political marketplace and political competition” and not predetermined by the state). But see Nathaniel Persily, In Defense of Foxes Guarding Henhouses: The Case for Judicial Acquiescence to Incumbent‑Protecting Gerrymanders, 116 Harv. L. Rev. 649, 668–69 (2002) (offering a defense of safe electoral districts through bipartisan gerrymanders as a means of securing the broader representation of interests that diverge from those of the median voter).

See Ross II & Spencer, supra note 13 , at 672.

See id. at 654–55 (summarizing the social science studies of the relationship between tangible barriers to voting and turnout).

See, e.g., Gerber & Green, supra note Error! Bookmark not defined. , at 661 (“Face‑to‑face interaction dramatically increases the chance that voters will go to the polls.”).

See Cal. Democratic Party v. Jones, 530 U.S. 567, 577–82 (2000) (invalidating California’s blanket primary because it violated the Democratic Party’s freedom of political association, which encompasses the right to exclude non‑members from their party functions including primary elections).

See, e.g., Abhay P. Aneja, Jacob M. Grumbach & Abby K. Wood, Financial Inclusion in Politics (2021) (manuscript at 3-4), []) (arguing that because 91% of all individual campaign contributions come from white people, there is an impact on who is elected and whose preferred policies are enacted, which expands the racial wealth gap).

We note that several states offer tax credits or refunds for small political contributions. Donors in Arkansas, Ohio, Oregon, and Minnesota may be reimbursed by the state up to $50. In Virginia, donors may earn a tax credit up to 50% of their contribution, up to a maximum value of $25. In Montana, donors may earn up to $100 tax deduction. Michael J. Malbin, Campaign Fin. Inst., Citizen Funding for Elections: What Do We Know? What Are the Effects? What Are the Options? 9 (2015), []. These refunds and tax incentives may provide similar incentives for campaigns to expand their mobilization activities but are far more onerous for individual donors than vouchers.

See, e.g., Richard L. Hasen, Clipping Coupons for Democracy: An Egalitarian/Public Choice Defense of Campaign Finance Vouchers, 84 Cal. L. Rev. 1, 5 (1996) (proposing voucher system as replacement for current campaign finance system); Bruce Ackerman, Crediting the Voters: A New Beginning for Campaign Finance, Am. Prospect (Dec. 19, 2001), [] (proposing similar voucher-based reform).

For the People Act of 2021, H.R. 1, 117th Cong. § 5101 (2021).

Id. at §§ 5101–02.

Democracy Voucher Program, City of Seattle, [] (last visited Apr. 23, 2021); Brian J. McCabe & Jennifer A. Heerwig, Diversifying the Donor Poll: How Did Seattle’s Democracy Voucher Program Reshape Participation in Municipal Campaign Finance?, 18 Election L. J. 323, 324 (2019).

McCabe & Heerwig, supra note 165, at 326 (describing the goals of the Seattle voucher program).

Jennifer A. Heerwig & Brian J. McCabe, Building a More Diverse Donor Coalition: An Analysis of the Seattle Democracy Voucher Program in the 2019 Election Cycle 1 (2020), [].

Id. (emphasis added).

There are other design details that may explain the ineffectiveness of Seattle’s voucher program to induce broader outreach by candidates. For example, the voucher program establishes a cap on participation due to the program’s $3 million budget. See Abby K. Wood, Christopher S. Elmendorf & Douglas M. Spencer, Mind the (Participation) Gap: Vouchers, Voting and Visibility 1, 4 n.2 (2019). Thus, at most 16% of Seattle residents could use a voucher (if each person only used one of their four vouchers). In reality, approximately 5% of registered voters used a voucher. See id. at 10.

For the People Act of 2021, H.R. 1, 117th Cong. §§ 5101–02 (2021).

Malbin, supra note 161, at 25 (“Twenty‑four times as many small donors from the poor and predominately black Bedford‑Stuyvesant neighborhood and the surrounding communities gave money to candidates for the City Council than for the State Assembly. For Chinatown the advantage was 23 to 1. In the heavily Latino neighborhoods of Eastern Manhattan and South Bronx, it was 12 to 1. The data support the claim that small donor matching funds help bring participants into the political process who traditionally are less likely to be active.”).

For the People Act of 2021, H.R. 1, 117th Cong. § 5111 (2021).

Adam Hughes, 5 Facts About U.S. Political Donations, Pew Rsch. Ctr. (May 17, 2017), [] (reporting that 12% of survey respondents report making a contribution to a political candidate in 2016, including federal, state, and local candidates); see also Donor Demographics, OpenSecrets, [] (last visited May 7, 2021) (providing reports on the percentage of the U.S. population that contributes at least $200, ranging from 0.5% in 2016 to 1.5% in 2020).

See Aneja, Grumbach & Wood supra note 160, at 4 (reporting that only 9% of donors to political campaigns are non‑white and outlining the ways in which campaign contributions from primarily white donors are correlated to racial wealth inequality).

Andrew Mayersohn, Those Prized Small Donors? They May Not Be as Small as You Think, OpenSecrets (Apr. 5, 2017), [] (“[T]he average household that made a $250 contribution to one committee also gave $2,102 to other committees . . . . Even some true megadonors, including five of the 2016 cycle’s top 100 contributors, found at least one campaign or committee that merited only a $250 contribution.”).

See, e.g., Adam Bonica, Small Donors and Polarization, Bos. Rev. (June 26, 2012), [] (“A seldom acknowledged feature of small donations is that much of it originates from individuals who spread thousands of dollars over dozens of candidates.”).

Id. Additional calibration will also be required to identify the most effective means of incentivizing campaigns. For example, research shows that a smaller match (4‑to‑1) on slightly larger contributions (up to $250 instead of $175) had a larger impact on the size of the donor pool in New York City than the 2009 transition to a 6:1 ratio. See Michael J. Malbin, Peter W. Brusoe & Brendan Glavin, Small Donors, Big Democracy: New York City’s Matching Funds as a Model for the Nation and States, 11 Election L. J. 3, 8–9 (2012). New York City’s current ratio is 8:1. Ese Olumhense, How Small-Dollar Public Financing Helped NYC Elect Its Most Diverse City Council Ever, City Limits (Nov. 16 2021), [].

Karl Evers‑Hillstrom, State of Money in Politics: The Price of Victory Is Steep, OpenSecrets (Feb. 19, 2019), [] (reporting the average cost of a winning non‑incumbent Democratic House campaign at $4.4 million and the average cost of a winning non‑incumbent Republican House campaign at $1.6 million).

52 U.S.C. § 30104(a).

2 U.S.C. § 441a-1 (2006).

Ariz. Rev. Stat. Ann. § 16-952 (2007).

Davis v. FEC, 554 U.S. 724, 743–44 (2008).

Ariz. Free Enter. Club’s Freedom Club PAC v. Bennett, 564 U.S. 721, 728 (2011).

See id. at 748; Davis, 554 U.S. at 741.

Davis, 554 U.S. at 738–39 (“While BCRA does not impose a cap on a candidate’s expenditure of personal funds, it imposes an unprecedented penalty on any candidate who robustly exercises that First Amendment right. [The legislation] requires a candidate to choose between the right to engage in unfettered political speech and subjection to discriminatory fundraising limitations.”) (emphasis added); Ariz. Free Enter., 564 U.S. at 736 (“Much like the burden placed on speech in Davis, the matching funds provision ‘imposes an unprecedented penalty on any candidate who robustly exercises his First Amendment rights.’”) (emphasis added).

Davis, 554 U.S. at 741 (“The Government maintains that [the Millionaires’ Amendment]’s asymmetrical limits are justified because they ‘level electoral opportunities for candidates of different personal wealth.’ . . . Our prior decisions, however, provide no support for the proposition that this is a legitimate government objective.”); Ariz. Free Enter, 564 U.S. at 749 (“We have repeatedly rejected the argument that the government has a compelling state interest in ‘leveling the playing field’ that can justify undue burdens on political speech.”) (internal citations omitted).

Ariz. Free Enter, 564 U.S. at 741 n.7 (“If the matching funds provision achieves its professed goal and causes candidates to switch to public financing . . . there will be less speech: no spending above the initial state‑set amount by formerly privately financed candidates, and no associated matching funds for anyone.”) (internal citations omitted).

Id. at 737.

Germany offers a model for this type of innovation, as the public funding of parties in that country is linked to the proportion of voters that turn out in support of the party during the general election. See Edith Palmer, Germany: Campaign Finance: An Overview, in Campaign Finance: An Overview: Australia, France, Germany, Israel, United Kingdom (2009), [].

Note that in a cash‑for‑turnout scheme, turnout would need to be measured in the number of voters, not the percentage of eligible voters. Otherwise, parties would have countervailing incentives to restrict voting to a very small group in order to maximize the percentage turnout while driving down overall participation rates and likely increasing democratic inequality.

For a summary of the literature, see Richard H. Pildes, Small‑Donor‑Based Campaign‑Finance Reform and Political Polarization, 129 Yale L. J. F. 149, 156–61 (2019); Richard H. Pildes, Participation and Polarization, 22 U. Pa. J. Const. L. 341, 371–76 (2020); Raymond J. La Raja, Campaign Finance and Partisan Polarization in the United States Congress, 9 Duke J. Const. L. & Pub. Pol’y 223, 229 (2014).

See, e.g., Eugene Volokh, Cheap Speech and What it Will Do, 104 Yale L. J. 1805, 1806–07 (1995) (explaining how new media technologies reduce the costs of distributing speech).

Id. at 1807 (arguing that social media technologies will “be much more democratic and diverse than the environment we see [in 1995]”).

See, e.g., Richard L. Hasen, Cheap Speech and What it Has Done (to American Democracy), 16 First Amend. L. Rev. 200, 201 (2018) (“The demise of local newspapers sets the stage for an increase in corruption among state and local officials. Rather than democratizing our politics, cheap speech appears to be hastening the irrelevancy of political parties by facilitating the ability of demagogues to secure support from voters by appealing directly to them, sometimes with incendiary appeals. Social media also can both increase intolerance and overcome collective action problems, both allowing for peaceful protest but also supercharging polarization and raising the dangers of violence in the United States.”).

See supra Part II.

See Hasen, supra note 194 , at 212 (illustrating the ability to use effectively free social media platforms to directly appeal to voters).

See Jeff Jones, In Election 2020, How Did the Media, Electoral Process Fare? Republicans, Democrats Disagree, Knight Found. (Dec. 7, 2020), [] (reporting that “more than four in five U.S. adults believe they were exposed to misinformation during the election campaign” and that “[s]ix in [ten] Americans, including a broad majority of Republicans, think misinformation swayed the outcome of the election”); see also Soroush Vosoughi, Deb Roy & Sinan Aral, The Spread of True and False News Online, 359 Science 1146, 1148 (2018) (finding that “false political news routinely reached the most unique users” and diffused faster than truth).

Laurel Wamsley, On Far‑Right Websites, Plans to Storm Capitol Were Made in Plain Sight, NPR (Jan. 7, 2021), [] (“After Trump promoted a Jan. 6 protest in D.C., a lot of his extremist supporters interpreted this as a call to action for them.”) (internal quotations omitted).

Atlantic Council’s DFRLab, #StopTheSteal: Timeline of Social Media and Extremist Activities Leading to 1/6 Insurrection, Just Sec. (Feb. 10, 2021), [].

Hannah Denham, These Are the Platforms that Have Banned Trump and His Allies, Wash. Post (Jan. 14, 2021), [].

Id. (noting that while Twitter, Facebook, and Google suspended right‑wing social media accounts on Jan. 6–7, “Amazon Web Services cut off the social media site Parler [the following] weekend”); Siladitya Ray, The Far‑Right Is Flocking to These Alternate Social Media Apps—Not All of Them Are ThrilledForbes (Jan. 14, 2021), [] (“Traffic has surged on Gab, a right‑wing alternative to Twitter, mostly from new members who are Trump supporters, believers of the QAnon conspiracy theory and other right‑wing extremists. The platform, which has been joined by several prominent right‑wing extremist figures, said [one week after Jan. 6] it has had 1.7 million signups and clocked 52 million page views in the past week.”).

Jack Dorsey (@jack), Twitter (Oct. 30, 2019), [].




Emily Stewart, Why Everybody is Freaking Out About Political Ads on Facebook and Google, Vox (Nov. 27, 2019), [].

See Kim Lyons, Spotify Will ‘Pause’ Political Ads in Early 2020, Verge (Dec. 27, 2019), []; Nico Grant, Adobe to Ban Political Ads Ahead of Presidential Election, Bloomberg (July 31, 2020), [].

Rob Leathern, Expanded Transparency and More Controls for Political Ads, Meta (Jan. 9, 2020), [].


Jeff Horwitz, Facebook to Limit Political Ads Week Before Election, Label Premature Calls, Wall St. J. (Sept. 3, 2020), [].

An Update on the Georgia Runoff Elections for Advertisers, Meta (Dec. 15, 2020), []; Elena Schneider, Facebook to Restart Political Ad Ban After Georgia Senate Runoffs, Politico (Jan. 5, 2021), [].

Scott Spencer, An Update on Our Political Ads Policy, Google (Nov. 20, 2019), [].

Id. (noting that Google had previously offered “basic political targeting capabilities . . . based on public voter records and general political affiliations”).

Id. (“This will align our approach to election ads with long‑established practices in media such as TV, radio, and print . . . .”).

James Doubek, How Disinformation and Distortions on Social Media Affected Elections Worldwide, NPR (Nov. 16, 2017), [] (describing the disruptive force of social media on elections in Kenya, Gambia, Venezuela, Zambia, France, the Philippines, and the United States).

For a discussion on the relationship between social media self‑regulation and government intervention, see Abby K. Wood, Facilitating Accountability for Online Political Advertisements, 16 Ohio State Tech. L.J. 520, 556 (2020); Abby K. Wood & Ann M. Ravel, Fool Me Once: Regulating “Fake News” and Other Online Advertising, 91 S. Cal. L. Rev. 1223, 1244–48 (2018).

Richard L. Hasen, Deep Fakes, Bots, and Siloed Justices: American Election Law in a “Post‑Truth” World, 64 St. Louis U.L.J. 535, 545–54 (2020).

See 47 U.S.C. § 230(c).

See Wood & Ravel, supra note 217, at 1231.



Bertrall Ross is the Chancellor’s Professor of Law at U.C. Berkeley School of Law.

Douglas Spencer is Professor of Law and Public Policy at the University of Connecticut.