Set aside Trump’s claim about the outcome; the Census report fails to provide enough information to evaluate its methods, verify any flagged case individually, or identify enough cases to change any state results in the 2020 presidential election.

To credibly reach the conclusions the report would like to make, it would need to accomplish at least three things: link the right people across enormous databases, verify citizenship and participation in 2020, and investigate the suspect cases individually. The report describes the first two steps without validating either at the individual level, and footnote 1 indicates the team could not have performed the third.

The problem of record linkage.

The research reflected in the Census report began with a commercial voter file. It then used the Bureau’s Person Identification Validation System (PVS) to link those records to federal administrative records. The system verifies an SSN when an incoming record contains one; otherwise, it searches using name, date of birth, sex, and address.

Problem Number One is that the report never says which fields were actually available in the file provided by the company DataClear, how many records were linked using different combinations of fields, or how often those rules linked records belonging to different people. That is important because voter records contain missing, inaccurate, and sometimes placeholder information, such as birth dates of 1/1/1900.

Subscribe now

This is precisely the issue addressed by Stephen Ansolabehere and Eitan Hersh in their article, “ADGN: An Algorithm for Record Linkage Using Address, Date of Birth, Gender, and Name.” The authors developed the algorithm while helping the Department of Justice analyze Texas’s photo-identification law.

The Texas voter file and the state’s identification databases did not always contain a common unique identifier. Ansolabehere and Hersh therefore constructed a series of exact matching rules using different combinations of address, date of birth, gender, and name—thus calling the algorithm “ADGN.”

About 98% of records matched using ADGN also matched in analysis where nine-digit SSNs were available.^[1] Their false-positive rate was 2.2%, and their false-negative rate was 10.1%, which they called upper bounds.

However, Ansolabehere and Hersh went beyond the Census Bureau report by explicitly validating the ADGN algorithm against SSN matches, estimating false-positive and false-negative rates, inspecting anomalous cases, and reporting which combinations of fields identified individuals most uniquely.

The Census Bureau report provides none of these diagnostics, and never defines the “high-confidence” match it says it used.

Ansolabehere and Hersh also found that changing the available fields can considerably change the reliability of record linkage. In particular, aggregating a full date of birth to a birth year sharply reduces the ability to identify people uniquely. This finding is important because publicly available voter files often provide only a birth year, or no birth information, rather than a full date of birth, and the Census report never says what birth information was supplied in the DataClear data, or whether full birth dates were imputed from information outside the voter files.

Subscribe now

Can matching identify noncitizens who voted?

Even if the Census Bureau linked everyone correctly, a second question remains: Did the underlying records correctly establish that these people were noncitizens who voted?

This is where an article by Sharad Goel and coauthors on double voting becomes relevant.^[2] Like the Census Bureau report, they used a commercial national voter file (TargetSmart) to identify records suggesting that the same person voted twice in 2012.

Their initial model estimated approximately 33,000 double voters. But Goel et al. didn’t stop there. They investigated whether the commercial vote histories were accurate. By comparing electronic records with Philadelphia poll books, they found that clerical errors could produce the appearance of double voting. An error rate of only 1.3% would have been sufficient to explain nearly all of their estimated double votes.

The Census report does check the DataClear data against certified turnout reported to the EAC, capturing 91.7% of ballots nationally. But it concedes the comparison is “one of magnitude, not of individuals”—precisely the kind of error Goel et al. found in Philadelphia and that aggregate agreement cannot detect.

Stephen Ansolabehere, Samantha Luks, and Brian Schaffner make the statistical point directly in their response to an earlier estimate of noncitizen voting that used a large academic survey. Although the Census report uses administrative records rather than survey responses, the same rare-event problem applies: citizens vastly outnumber noncitizens, so tiny classification errors can cause false positives.

What states have found when they investigated the matches.

Citizens can land on these lists not because they are noncitizens, but because of the limits of computerized matching. For instance, the report’s principal citizenship evidence consists of a naturalization oath, U.S. birth, or a passport, although it also accepts certain undated citizenship evidence. A citizen who derived citizenship as a child through a naturalized parent may nevertheless lack the records needed to clear the screen. Other citizens may be matched to noncitizens who share a birth year but not a birth date. The Census report does not estimate how often these errors occur.

Stephen Richer has assembled a useful state-by-state compilation of recent investigations that have started with matches similar to those described in the Census report. The investigations vary, but generally produce a common pattern: large preliminary lists of suspected noncitizens become much smaller as officials examine the records and contact the identified voters.

For instance, in 2024, Iowa initially identified around 2,200 registered voters as potential noncitizens using state transportation records. After further investigation, the Secretary of State reported that number was reduced to 277 confirmed noncitizens through analysis using the SAVE system, 35 of whom cast votes in the 2024 general election.

In a 2024 audit of more than eight million registered voters, Georgia identified 20 noncitizens on the rolls and referred another 156 registrations for county investigation.

The pattern is consistent: Matches generate leads. Leads prompt a closer look. Closer looks dismiss the lion’s share of cases and allow the troubling ones to proceed.

I could say more. An Individual Tax Identification Number is treated as affirmative evidence of non-citizenship. Nearly 7,000 of the 24,000 have no country of birth in any linked record. The state counts are unadjusted for the file’s uneven coverage, which runs from 74.5% in Utah to 97.1% in New Jersey. Each deserves its own treatment.

Subscribe now

What do we make of this?

Experienced professionals report uncertainty and carry the investigation to the last mile—from computer matching to individual verification. The Census report describes a linkage exercise without providing enough information to evaluate its work or performing that indispensable final step. The report’s first footnote says the analysts had no access to names “or any other identifying information,” and its two example cases are labeled illustrative composites. The final step wasn’t omitted; it was impossible for this analysis team.

None of this is to say that the Census Bureau report’s 24,000 cases represent nothing. Some may represent noncitizens voting and should be investigated. I would not be surprised to learn that many have been already in states that perform citizenship audits.

Taken together, the scholarship and state investigations indicate that noncitizen voting is a raindrop in the ocean of citizen voting. This doesn’t mean that citizenship audits shouldn’t be conducted. However, politicized efforts poison the well politically and make it difficult for state and local officials to do their own work. Therefore, I’m afraid the main result of the report will be increased resistance to good-faith auditing by the states, when they should be encouraged.

^[1] The Texas voter file didn’t have full SSNs for all registrants, but it had enough to allow Ansolabehere and Hersh to compare ADGN matches with SSN matches.

^[2] Goel et al. appears in the report’s reference list and nowhere in its text.

President Trump has made no secret that passage of the SAVE America Act is his top legislative priority. The Act’s stated goals—^[1]to require identification to vote and prevent non-citizen voting—are popular with the public. Still, the Act is on the path to legislative failure. Why? Although the news and Trump are currently focused on the parliamentary hurdle in the Senate, the deep answer is that while the goals may be popular, the statute as written overshoots even the most conservative state laws.

Subscribe now

The current attention on the roadblocks presented by the Senate attributes the SAVE America Act’s problems to the filibuster, but the more interesting question is whether the Act would pass absent the filibuster. A strong argument can be made that there aren’t 50 votes in the body for the Act. That reality illustrates well the gap between embracing general principles and passing legislation.

The SAVE America Act faces more than a filibuster problem

Focusing on four pivotal Republican senators illustrates the point. Sen. Lisa Murkowski of Alaska has publicly opposed the Act, and three other Republican senators—Collins, McConnell, and Tillis—voted against an effort to attach the SAVE America Act to the reconciliation bill last June. With 53 Republican senators, losing them again would sink the Act.

Like nearly all Republicans in the general public—and even most Democrats—senators Murkowski, Collins, McConnell, and Tillis support requiring voters to show a photo ID when they vote. Republican state legislatures throughout the country have enacted or expanded voter ID requirements over the past decade and have begun requiring documentary proof of citizenship to register, as well. Ballot measures across the country have passed in all but a few of the states where they have been presented to voters.

Murkowski’s expressed opposition to the SAVE America Act, and the likely opposition of some of her colleagues, allows us to consider the gap between broad approval of the idea of voter ID or citizenship verification and the practicalities of writing a bill that would pass Congress. As she wrote in an opinion piece published by the Anchorage Daily News and stated in a Senate speech, the bill would erect significant voting barriers to virtually all Alaskans, with the highest barriers facing women, Native Americans, and rural residents. The problem is not the idea of voter ID or citizenship verification, it’s that the statute as written creates unnecessary barriers to many voters.

The tendency of Congress to defer to the states in governing their elections runs deep, so there is no guarantee that even a bill that addressed Sen. Murkowski’s objections would pass Congress. Still, the failure of SAVE America Act supporters to fine-tune its requirements to win the support of wavering Republicans, and even some Democrats, gives it all the signs of a messaging bill, rather than an effort to achieve policy results.

Republican-controlled legislatures have broadly embraced photo ID laws. We can learn a lot about how these legislatures have crafted bills to implement voter ID requirements and promote proof of citizenship.

Americans want voter ID and non-citizens off the voter rolls, but they also distrust Big Brother

At its core, the states have had to address two major features of American political culture that have, in turn, complicated election administration. These are distrust of government and a fragmented public administration.

First, both voter ID and citizenship verification bump up against deep-seated cultural norms that resist governmental intrusion into Americans’ lives. American political norms and institutions have resisted mandating a national—or even state—registry of all citizens and their addresses. Other robust democratic countries require this, which makes it easier both to keep track of citizens and non-citizens and to issue official identification to everyone.

Subscribe now

The second feature flows from the first. We don’t have a national list of citizens, nor do we have a complete national list of where people live. What we have are data sources for the administration of social programs that also include information about citizenship, including the Social Security Administration. States don’t have a comprehensive list of their residents, only those who have driver’s licenses and perhaps other recipients of public assistance.^[2] The holders of these data can share them, but the systems weren’t designed to interoperate, and there are gaps. Together, all these data sources can account for nearly all citizens, but “nearly all” isn’t “all.”

The enhanced SAVE system illustrates how “nearly all” is not “all”

A good example is the recent effort by the Department of Homeland Security to enhance its SAVE system, which was designed to identify resident aliens and those who have naturalized, by adding the ability to also query Social Security records, which includes a data flag for where a person is a citizen. According to records released in litigation, 3% of SAVE program responses for voter verification as of August 2025 were returned because of a non-definitive response; that is, it was impossible to locate a matching record, or more information was needed to yield a match. Thus, it appears that the federal government can produce the functional equivalent of a list of nearly all adult citizens, but not all of them.

In a country of over 340 million people and 236 million eligible voters, the gap between all and nearly all can amount to millions of people, and the gaps are likely to fall unevenly across age, race, geography, and mobility status. As we will see in later posts, the political consequences, while small, are likely to fall unevenly, and in ways that will surprise many.

Neither resistance to national residential registration nor the lack of a single citizenship registry is a reason alone to oppose requiring identification to vote or ensuring that only citizens are on the voting rolls. (And to be clear, there are many serious arguments against these policies, but those arguments are mostly championed by Democrats, who are currently the minority in Congress.) But factors such as these are why laws and policies need to be carefully designed so that the millions of citizens who exist in the gaps aren’t disfranchised as the price paid to address the tiny amount of voter impersonation fraud or the small number of non-citizens on the voter rolls.

State safeguards

States that have adopted voter ID requirements or have scoured their rolls for non-citizens have addressed these risks—some better than others, and some more or less to the liking of election activists. They have addressed these risks through a portfolio of devices, such as cure procedures, exceptions, free IDs, and administrative flexibility. The SAVE America Act is silent about the risks posed by its provisions and the commonly used techniques to address them. Indeed, by threatening to criminalize election officials who make innocent mistakes, it leans into the risks.

What comes next in this series

These are general thoughts, but each of the two major features of the Act has its own unique features, as well. That will be my focus over the next few weeks. My next post will address the voter ID features of the Act; the following post will take up the citizenship verification requirements.

In those two posts, I focus on the data we can use to understand why large majorities of Americans favor some form of voter ID requirement and attention to citizenship verification, the size of the problem these reforms address, and the size of the gaps that need to be addressed to account for the fact we weren’t all born with a passport in our hands and a locator beacon tied around our necks.

Subscribe now

^[1] All em-dashes in this post were generated by a human.

^[2] One possible exception is my home state of Massachusetts, which conducts an annual census.

Anticipation of President Trump’s address had much of the election-administration community on tenterhooks. My own list of possibilities, ranging from likely to outlandish, ran to forty items. In the end, the speech did not reveal a deeper truth about the integrity of the 2020 election or the administration of the upcoming 2026 election. A few new details appeared, especially in the documents released by the White House. But, as I discuss below, they are concerning in ways different from those the President emphasized.

That does not mean the speech was inconsequential. Its importance lies less in what it proved than in what it legitimized. Trump used a presidential address to elevate the election-skepticism movement and to frame the next 110 days before Election Day and the post-election period that follows as a test of whether the system can be trusted.

Trump spent roughly 25 minutes arguing that the American electoral system is fundamentally unsafe and vulnerable to manipulation. He focused on four subjects: alleged vulnerabilities in voting machines, alleged noncitizens on the voter rolls, a 2020 voter-registration fraud case in Michigan, and China’s acquisition of voter-roll information. It ended with a “Come to Jesus Moment” where he urged listeners to call their representatives and demand they pass the SAVE America Act.1

Subscribe now

Two Data Claims Worth Considering

Because this is a blog about election data, I want to focus on two data-focused claims in the speech: China’s acquisition of voter information and DHS’s alleged matches of noncitizens to state voter rolls.

China Had Voter Files, But That Didn’t Interfere with the Conduct of the Election

First, about the China data. A major part of Trump’s discussion of China and the 2020 election concerned China’s alleged acquisition of voter-registration information. The genuinely new claim was that China had obtained a voter file containing information on more than 200 million American voters; the White House put the figure at roughly 220 million. The supporting materials, however, appear to mix public voter-file data, commercial voter-file data, and possibly compromised datasets, which is not the same as showing a breach of live election systems.

Two points here. First, it was already known that foreign actors, including China, had targeted election-related information and that state and local government networks could be vulnerable. The new White House materials add more detail about PRC acquisition or possession of large voter-data files, but they do not show that China altered voter-registration records, cast ballots, changed vote totals, or affected election results. Second, much of the information in voter files is not secret. States make redacted voter-registration data publicly available, and commercial databases that aggregate all state files are accessible to campaigns, parties, and even researchers.

It is bad that China has been hoovering up voter information. What remains unclear is whether this was specifically about elections or part of China’s broader effort to gather information about Americans. Either way, the election-administration concern would become much more serious if there were evidence that voter-registration records had been changed, or that the information had been used to defraud the election system. So far, that evidence has not been produced. The missing link is evidence that Chinese actors changed voter-registration records, used the data to cast fraudulent ballots, altered vote tabulation, or affected reported results. None of the public record, including the new White House materials, proves that.

Noncitizen Matches Are Not Noncitizen Voters

The second data-related claim concerns noncitizens on the voter rolls. Trump cited a DHS document claiming that roughly 250,000 noncitizens were found on voter rolls in four states. This, too, was new information, but it is important to note that the report is not of independently validated counts of confirmed noncitizens on the rolls.

Anyone with experience in these data-matching exercises should expect those numbers to melt as quickly as ice cream in a Phoenix August. The reason is straightforward. The released DHS summary says the four-state figure was generated from public voter files from states that had not used the enhanced SAVE system, DHS’s citizenship-verification service. That is important because public voter files generally lack the sensitive identifiers needed for reliable first-round citizenship matching. As a result, DHS likely lacked the identifiers needed for high-quality matching. SAVE checks generally require name, date of birth, and a unique immigration or Social Security identifier, and public voter files ordinarily do not contain the most sensitive identifiers.

Stephen Richer recently posted a useful catalog of state efforts to match voter files against citizenship data. The pattern is by now familiar. Initial database matches can produce large lists for investigation, but those lists often shrink sharply after election officials account for false positives, stale records, naturalizations, duplicate records, and data-entry errors. Confirmed cases are usually tiny relative to the size of the voter rolls, and many appear to involve mistakes rather than organized fraud. In that sense, the 278,000 figure is likely a red herring unless and until DHS releases the match methodology, confidence scores, naturalization checks, duplicate handling, state adjudication results, and voting-history information. Its most immediate consequence may be that four states now have to spend valuable time chasing down false positives when they should be preparing to run a high-quality election.

The same DHS document is more revealing in another respect. It reports citizenship matches using the enhanced SAVE system for ten states that provided more complete personally identifiable information. In those states, the number of possible noncitizens on the rolls falls sharply: 10,716 possible matches across ten states with 50.8 million registered voters. That is about 0.02 percent of registered voters. And these are still suspected noncitizens, not confirmed noncitizens. Each state would need to investigate those cases, and, as Richer’s catalog suggests, the number would likely shrink further.

Subscribe now

Why Give the Speech Now?

So what was the speech about? It did not reveal anything earth-shattering, especially to people who have followed election-administration controversies over the past decade from the left, right, or center. Why give it, and why give it now?

I am not inside the President’s head, nor am I privy to the inner deliberations of his administration or its political supporters. But I am a political scientist who has studied these issues for a long time, and I think the speech was consequential for two reasons.

First, the speech will fire up the small but highly active core of election activists who have pursued a post-2020 agenda of questioning election results and challenging mainstream election-administration practices. Second, it lays a predicate for challenging the results of the 2026 election, especially if those results do not go according to Trump’s wishes. It is easy to imagine the claims made last night and the documents released to support them becoming the backbone of challenges that begin the morning after Election Day and continue into the convening of the next Congress.

The Lost Opportunity

I will close with the lost opportunity. Maintaining clean voter rolls is important. Even if the best evidence indicates that noncitizens on the rolls are rare, some people acting in good faith continue to worry about the issue. And when noncitizens are inadvertently added to the rolls, the consequences for them can be serious. There are people across the election-administration world who know how to address this problem responsibly, or at least are ready to roll up their sleeves and work through it. Unfortunately, the current political environment has made that kind of consensus harder to sustain. But that is a post for another day.

Since then, I’ve taken a closer look at Georgia’s voter history file. The individual-level evidence points in the same direction: compared with four years ago, Democratic primary voters appear more engaged than Republican primary voters. Louisiana’s primary last Saturday points in the same direction, too, although the comparison there is less clear because Louisiana has just changed its primary system.

Georgia’s voter history file confirms greater Democratic engagement this time around

Starting with Georgia, I pulled down the voter history file from the 2022 and 2026 primaries and compared how individuals who voted in 2022 have voted thus far in 2026.

First, an explanation. Much to its credit, Georgia posts voter history files online for anyone to download. These files record which registered voters cast ballots in particular elections. Because Georgia includes a unique voter identification number, researchers can track whether the same individual voted across multiple elections.

One important detail is that Georgia is an “open primary” state. Any registered voter can request either party’s ballot—or a nonpartisan ballot—in the primary. This allows independents to participate in partisan primaries and allows voters who usually identify with one party to cross over and vote in the other party’s primary. Crossover voting happens less often than many people think, but even modest crossover can tell a story.

Back to the data. Merging the two voter history files shows, first, that of the 1,030,367 people who cast an advance vote in 2026—either by mail or early in person—471,981 cast an advance vote in 2022, 231,614 cast an Election Day ballot, and 326,772 didn’t vote in the 2022 primary.

Subscribe now

Second, the party-ballot patterns among both repeat primary voters and new 2026 primary voters tilted toward the Democrats. With apologies for the color scheme, the details are shown in the accompanying graph.1 Among the highlights:

• Among 2022 Republican primary voters, 27.6% had already voted in this year’s primary, compared to 54.4% of 2022 Democratic primary voters.

• Among 2026 advance voters who did not vote in the 2022 primary, 70.9% took a Democratic ballot and 29.0% took a Republican ballot.

• Among those who took a Democratic ballot in 2022 and have already voted, only 0.6% took a Republican ballot in 2026, compared to 1.8% who took a Republican ballot in 2022 and then took a Democratic ballot in 2026. (I told you crossover voting was uncommon, but it can still tell a story.)

At the end of the night, we’ll see whether these trends carry over to Election Day voting. My expectation is that the Democratic share of Election Day votes will also be higher than it was in 2022, though probably not by as much as in the advance-voting period.

One possible explanation is that the advance-vote pattern simply reflects the continuing partisan separation in voting modes. I’m skeptical of that explanation, though I could be wrong. Another possibility is that more Republican voters are still weighing their choices and therefore waited until Election Day.

Subscribe now

A word about Louisiana

Louisiana’s primary was on Saturday. Louisiana has long been famous for its “jungle primary,” in which all candidates appear on the same ballot, regardless of party. If no candidate wins a majority, the top two vote-getters advance to the general election. In 2024, the state changed the law so that, starting in 2026, candidates for certain federal and state offices would run in partisan primaries. For that reason, it’s not possible to directly compare 2026 turnout in the partisan primaries to 2022. An alternative is to compare Saturday’s primary turnout with the votes received by Donald Trump and Kamala Harris in Louisiana in 2024.

The unofficial results posted by the Louisiana Secretary of State as of this morning (May 19) show 344,751 votes cast in the Democratic primary for U.S. Senate and 401,119 votes cast in the Republican primary. Compared to the presidential numbers (766,870 for Harris and 1,208,505 for Trump), that works out to a 45.0% rate for the Democratic primary and 33.2% in the Republican primary.

Again, this isn’t an apples-to-apples comparison like I’ve made for other states, but it seems to continue the trend of the most engaged Democratic voters being more eager to participate in primaries than the most engaged Republican voters.

Subscribe now

In the six states that held comparable statewide primaries in both 2022 and 2026—Arkansas, Illinois, Nebraska, North Carolina, Ohio, and Texas—the Republican-Democratic turnout ratio shifted toward Democrats in every case, with Democratic turnout up and Republican turnout down. This pattern is consistent with other evidence suggesting that Democrats currently hold the advantage in political energy, even if it is an imperfect predictor of the November results.

Election substitutes for the anxious

Predicting election outcomes is as old as political punditry. In a way, looking at current indicators of future outcomes is the appetizer for political obsessives: it settles the nerves while they wait for the main course.

There is no shortage of efforts to tell which way the political winds are blowing in 2026. Among these are prediction markets, presidential approval, vote intention, incumbent retirements, fundraising, special election outcomes, and election fundamentals.

All of these signs indicate heavy headwinds for Republicans in 2026. Kalshi investors are signaling a 74% chance that Democrats will take at least one chamber of Congress. President Trump’s net approval rating is the lowest of any president at this point in his term since national polling began and generic congressional vote-intention polls show a +5 Democratic advantage. Republicans are retiring from the House at near-record numbers, Republicans are lagging Democrats in fundraising, and Republicans have been experiencing a significant drop-off in support in special elections in 2025 and 2026.

Some of these signs have a better track record of forecasting than others. Presidential approval has long been central to midterm forecasting, and generic-ballot polling is also informative, although poll-based measures generally become more predictive as Election Day approaches. Fundraising is a noisier signal because money reflects not only a candidate’s strength but also how competitive or threatened a campaign appears to donors.

Despite the glut of auguries out there, I have another one to offer: turnout in the primaries.

Subscribe now

Primary turnout as a sign of things to come

I’ve been following turnout in the current round of primaries, and I’ve noticed something interesting: relative to Republican turnout, Democratic turnout is up compared to 2022.

Before proceeding, I should clarify what I mean by primary turnout. Because few states report the total number of ballots cast in each party’s primary, I instead use the statewide contest in each party that drew the most votes. In other words, for each party in each state, I take the highest-turnout statewide primary as my measure of party turnout. In Ohio in 2026, for example, Democrats cast more votes in the U.S. Senate primary (791,355) than in the gubernatorial primary (762,457), so I use the Senate race to measure Democratic turnout. Republicans, by contrast, cast more votes in the gubernatorial primary (817,159) than in the Senate primary (733,270), so I use the governor’s race to measure Republican turnout.

This measure allows us to compare changes in raw turnout from 2022 to 2026. In each of the six states that have held statewide primaries in 2026 and also held statewide primaries in 2022, Democratic turnout has increased, while Republican turnout has decreased.1

The raw numbers show the same directional story everywhere I look: Democratic turnout is up, and Republican turnout is down. But the cleaner way to compare states is by looking at the Republican-to-Democratic turnout ratio in each cycle. This helps normalize for differences in state population and baseline partisan support.

By that measure, the shift is evident everywhere. In deep-red Arkansas, the Republican turnout advantage fell from nearly 4:1 to just over 2:1, while in deep-blue Illinois, turnout moved from near parity to just over a 3:1 Democratic advantage. In Texas, Ohio, and North Carolina, the Republican advantage visible in 2022 gave way to something much closer to parity in 2026.

These turnout shifts are consistent with other signs, but let’s not get out over our skis quite yet

I can already hear the cries of, “Wait, wait, wait! There’s got to be much more to primary turnout than just partisan enthusiasm.” Those cries are correct, and can be confirmed by looking closely at the various races reflected in the data thus far.

Texas is a good example. In 2022, the top statewide primaries were for governor, and each party had a clear dominant candidate: incumbent Republican Greg Abbott and Democrat Beto O’Rourke, on the rebound from his close 2018 Senate loss to Ted Cruz. In 2026, by contrast, the marquee race was the U.S. Senate primary, and both parties had competitive nomination fights. That likely goosed turnout in 2026, independent of any broader partisan mood.

Analyses like this can be done for each state. Still, Texas shows why this indicator should not be interpreted mechanically. Competitive primaries can boost turnout independently of broader partisan mood. But they may also be part of the signal, because strong candidates are more likely to run when they sense favorable conditions. In general, when a party sees an opportunity, stronger candidates are more likely to run, more donors contribute, and more voters show up.

Subscribe now

What about the past?

I’ve not found published political science research directly testing whether changes in partisan primary turnout ratios predict party performance in November. To see whether this is a potentially useful indicator, I looked backward. In the same states, I compared the shift from 2018 to 2022. Because 2018 was a much stronger year for Democrats than 2022 was, we would expect primary turnout ratios to have moved in a more Republican direction over that period. In five of the six states, that is exactly what happened.

Georgia preview

I will end the analysis by applying it to the upcoming Georgia primary. Georgia offers a useful test case because its primary is still ahead of us, but the advance-voting data already provide an early read on partisan participation. Once the results are released next Tuesday, we’ll know whether the Peach State saw a turnout shift in Democrats’ favor, but we can already see shifts in mail-ballot requests and in-person early voting.

Mail-ballot requests are down by about 50% compared to 2022; early in-person voting is also down overall, but not by nearly as much. As of yesterday (Thursday, May 14), 45,800 mail ballots have been requested, compared with more than 90,000 at the same point in 2022. In contrast, more than 714,000 early votes have been cast, compared with nearly 800,000 in 2022.2

For the purposes of this post, the key question is not the overall level of advance voting but the partisan balance within it. This year, as in recent Georgia elections, Democrats outpaced Republicans in requesting mail ballots. More interesting is that significantly more Democrats have turned out to vote early in person this year, whereas the opposite was true in 2022.

To make this point more directly, I turn again to the graphs that plot the ratio of Republican-to-Democratic use of mail ballots and early in-person voting. By this measure, Democrats are slightly outpacing Republicans in requesting mail ballots compared to 2022. This is in sharp contrast to in-person early voting, where, in 2022, Republicans outpaced Democrats by 40%; in 2026, Republicans are falling behind Democrats by 20%.

[5/16/26 update: For updates of these two graphs, check out this Twitter post:]

As I mentioned above, lots of things affect primary turnout, and it’s not always predictive of outcomes in the general election. In every comparable statewide primary held so far, the turnout ratio has moved toward Democrats. That doesn’t guarantee anything in November, but it’s another sign that 2026 is shaping up to be a favorable environment for them.

Subscribe now

The order assumes that federal agencies and the Postal Service can assemble and maintain lists accurate enough to determine who may receive mail ballots and where those ballots may be sent before a federal election. That premise may sound obvious. In the real world of absentee voting, it isn’t.

In the real world, absentee voting requires managing late requests, address changes, and voter-registration records that are often updated only as the election approaches. The EO would create a parallel federal system that would have to manage many of the same problems while remaining tightly coordinated with local election offices. And it would have to do so with less staff, less local knowledge, and less capacity to handle complications as they arise.

Subscribe now

The EO is a classic example of what I call how-hard-could-it-be-ism: the habit of underestimating the complexity of running elections. Sometimes that habit merely adds strain to the election administration system. Here, it courts chaos.

To see how this kind of overconfidence operates in practice, start with what the order actually requires.

What the executive order requires

The EO adds a new layer of federal regulation on top of existing state and local systems. It does so mainly in two ways: by requiring mail-ballot applications to be checked against a new, comprehensive list of American citizens age 18 and older, and by barring the U.S. Postal Service from delivering mail ballots except to people authorized to receive them under the new system. It also opens the door to criminal investigations of election officials and election service providers if the new federal rules are not followed.

The executive order has several moving parts, which the accompanying figure helps lay out. Once that machinery is clear, the order’s hidden premise comes into view.

Section 2 directs DHS and the Social Security Administration to create a “State Citizenship List” of “confirmed” U.S. citizens who will be at least 18 by the general election. The system is to be built by June 9, and the list is to be sent to every state by September 4.

Section 3 directs the U.S. Postal Service to write new rules governing how mail ballots are handled, including special envelope requirements, barcode tracking, and state-supplied lists of who should receive them. It also requires USPS, by September 4 [i.e., 65 days before Election Day], to send each state a list of voters “enrolled with the USPS” for mail ballots. During the mail-ballot season, USPS would be barred from delivering ballots to anyone not on that absentee participation list.

Both sections rest on the same premise: that the identities of mail-ballot voters and the addresses to which their ballots should be sent, can be settled well before Election Day and then changed only at the margins.

The EO assumes the lists can be settled early

The core of the EO is the creation of three federal lists meant to determine who may receive mail ballots and where those ballots should be sent. It assumes that those voters and addresses can be settled roughly 60 days before an election, with only minor changes after that. That may be less problematic in vote-by-mail states and for voters already on permanent absentee lists. It is much more problematic for voters who request a mail ballot for a particular election, because those requests often come late. And because many voters move and delay updating their registration records until election season, even the underlying voter file is often less stable than the EO assumes.

Why the lists will not stay fixed

There is no national registry of when mail ballots are requested. But the absentee ballot files from Georgia and North Carolina—two of the few states that make those files readily accessible to the public—show the basic pattern clearly enough. As the accompanying graph shows, requests arrive throughout the year but accelerate sharply in the final two months before the election. This pattern is likely common in other non-VBM states and in states that do not use permanent absentee lists. That means a system built on the assumption that the relevant lists can be settled 60 or 65 days before Election Day is colliding with the actual timing of absentee voting.

But timing is only half the problem. The destination address is often unsettled until late as well.

A significant number of mail ballots are sent to addresses other than the voter’s registration address. According to the 2022 Survey of the Performance of American Elections, 6% of those who voted by mail in 2022 did so because they were away from home, or about 1 million voters nationwide. In the North Carolina absentee file, nearly 20% of mail-ballot applicants asked that the ballot be sent to an address other than the one where they were registered. That suggests the share of ballots sent away from a voter’s registration address may be larger than the SPAE figure alone implies. For a significant number of mail voters, the address on a federal list will not be the address to which the ballot must actually be mailed.

Subscribe now

Another issue is keeping track of the addresses of voters who are already registered. A substantial minority of voters believe that when they move, their registration is automatically updated, which usually is not the case. That likely helps explain why some voters who moved months earlier do not update their registration address, or register anew, until close to the election. Even in a state like Maryland, where many address changes are transmitted through MVA transactions, those updates would still have to be reconciled with the new federal system. Maryland—one of the few states that publishes monthly voter-file activity reports—offers a useful illustration. In 2022, nearly 80,000 address changes were recorded in the Maryland voter file in September and October alone, along with another 55,000 name changes. Those are exactly the kinds of late-cycle changes that would have to be run through the DHS and USPS systems on top of the quality-control steps the state is already performing.

The EO directs the Secretary of DHS to create procedures for citizens to update their records on the State Citizenship List and for states to “propose” amendments to it. But if ballot requests, mailing addresses, and voter-registration records all remain fluid until late in the cycle, many of the resulting problems won’t appear until the final weeks before a federal election. At that point, the federal government would have to process thousands of proposed changes precisely when election offices are busiest and time is shortest.

If local election officials do not know until days before an election who should receive a mail ballot and where it should go, the mail-ballot pipeline begins to break down.

This is not a niche problem

Mail voting is too large and too varied a part of American elections for this to be treated as a niche administrative problem. In the 2022 midterm, 35.2 million ballots—32% of all votes cast—were cast by mail.1 Those ballots were cast under a wide range of state rules, from all-vote-by-mail systems to no-excuse absentee voting to more traditional excuse-based systems. The accompanying map makes the point visually: state rules differ not just across those broad categories, but also in important details such as permanent lists and exceptions for elderly and disabled voters.

The EO does not formally require states to use the citizenship list in any particular way. But it clearly aims to pressure them to reconcile their voter rolls with it in anticipation of the mail-ballot rules in Section 3. It also does so under the shadow of possible prosecution for aiding and abetting noncitizen voting.

A maximal fix for a limited problem

The Executive Order addresses a real but limited administrative problem with a maximal federal solution ill-suited to how elections actually work. There is little credible evidence that citizenship verification failures or mail-ballot abuse are major drivers of voter fraud, but that does not mean these parts of election administration could not be improved. The enhanced SAVE program is potentially a step in the right direction, but only if it is developed into a tool states can use reliably and at scale, without turning over election administration functions to the federal government. States could also take further steps to improve the verification of who votes by mail. But those are incremental reforms that require careful research, careful design, and time, not a rushed effort to impose complicated new procedures before the next election.

The order is less a reform than a wish dream of control. It assumes that a decentralized democratic process can be brought to heel from Washington through matching, enrollment, and prohibition. But election administration does not work that way, and neither does mail voting. Serious reform would begin with modesty rather than how-hard-can-it-be-ism: better tools where needed, better data where possible, and respect for the fact that states and localities are dealing with actual voters, shifting addresses, and late-breaking complications. This order takes the opposite approach. That is why it is so unlikely to improve verification, and so likely to significantly disrupt voting by mail.

Subscribe now

Background: What is the EPI?

The EPI was launched over a decade ago by the Pew Charitable Trusts as part of a project aimed at improving the effectiveness of elections for Americans. The overall effort came at the end of a decade that had seen the quality of American elections called into question by the failings revealed in the 2000 Florida recount. Among these failings were voting machines that didn’t work, poor ballot design, antiquated information systems, and inaccurate voter lists.

The decade of the aughts was marked by election reform efforts, the most consequential of which was the passage of the Help America Vote Act in 2002. In the midst of these reform efforts, many voices began asking, “How will we know if elections have improved?” One of these voices was Heather Gerken, then a law professor at Yale and now the president of the Ford Foundation. In an influential book, The Democracy Index, Gerken proposed creating a state-by-state rating based on how well they conduct elections, akin to U.S. News & World Report’s college rankings.

Inspired by Gerken’s idea, Pew assembled a large advisory group of election officials and academics who spent three years hammering on the question of how to assess the convenience and security of elections at the state level. The result was the Elections Performance Index. Pew handed over management of the EPI to MEDSL in 2017, and we have updated it biennially ever since.

At the core of the EPI are 19 indicators1 that measure a variety of factors involved in the administration of elections. Most of these measures gauge policy outcomes, such as mail ballot rejection rates, wait times to vote, and voter turnout. Other indicators measure the presence of state policies meant to enhance security and the integrity of outcomes (post-election audits, ERIC membership, etc.).

Subscribe now

The EPI is the Elections Performance Index. While the EPI website allows users to explore state performance for each indicator, the bottom line is a single index score on a 0–100% scale. The index itself is calculated by averaging scores across the 19 indicators, each of which has also been converted to a 0–100% scale.

While the EPI website ranks the states and D.C. from top to bottom based on the index score, we at MEDSL do not emphasize this ranking for two reasons. First, we want to emphasize that every state is capable of improving, and that it is primarily important to know that a state is improving, and only secondarily to know if it is improving compared to other states. Second, over time, as states overall have improved, their index scores have begun to bunch in the 80s, to the point where moving from 81.7% to 80.4% means dropping from number 17 to number 27 in the rankings, which doesn’t seem especially informative.

There’s much more to be said about the EPI background, much of which can be read in the hundred-page EPI methodology document.

The Big Story: Improvement Since 2008

Across the five presidential elections rated by the EPI, average scores have risen markedly. The most rapid period of improvement occurred from 2008 to 2020, when average index values rose from 62.3% to 80.4%. In addition, with one exception, every state’s score has improved, in some cases substantially. (The one exception is North Dakota, which started at the top with a score of 83.9% in 2008, and then retreated to 82.6% in 2024.)

The average index score slipped a bit in 2024, to 79.8%, but I interpret that as a flattening of the improvement curve rather than nationwide backsliding.

The Greatest Improvements Have Occurred in Online Presence and Auditing

To understand where the EPI has improved over the past five presidential elections, we need to dig a bit deeper into the indicators themselves. One way to classify them is functionally—absentee ballot management, voter registration management, post-election auditing, etc. Another is to classify them as either measured on a binary yes/no scale or on a continuous scale. Binary indicators include things like whether a state allows online voter registration or requires post-election tabulation audits. Continuous scales include measures such as turnout rate and average wait times.

If we examine the change in the index components from 2008 to 2024, we see that, by far, the biggest improvements have been in three binary indicators: allowing online registration, requiring post-election tabulation audits, and being a member of the Electronic Registration Information Center (ERIC). The fourth-biggest improver behaves almost like a binary measure and captures how many lookup tools a state has on its website, out of five.2

The continuous indicator that has improved the most measures disability access, followed by one that captures the degree to which people cite voter registration problems as a reason for not voting.

These remaining indicators all address difficult-to-change practices that must generally be implemented at the precinct level, whereas the first four can usually be changed with a single decision by the state’s chief election official or the legislature.3

Take the provisional ballot rejection rate indicator, for example. This indicator measures the number of provisional ballots rejected in a state, divided by the total number of provisional ballots cast. For a state to move from the bottom to the top of this indicator, it would have to significantly revamp its voter registration system, voter registration requirements, and precinct check-in practices, and then carry those practice changes to every precinct in the state. That’s a far cry from changing just one website, which can be challenging enough.

Indeed, I know of one state, which will remain anonymous, that contacted me early in the life of the EPI to tell me that their low performance on the provisional ballot cast indicator had prompted them to change how questioned voter registrations were handled statewide. They worked hard to change state regulations and then to train local officials on the new system. That state now issues about 25% fewer provisional ballots than in 2008. T hat’s not nothing, but all that effort only increased their index score by a fraction of a point. On the other hand, adding online registration and website capabilities increased its score by almost ten points.

Subscribe now

All this means that actually improving the practice of election administration at the precinct level is orders of magnitude harder than improving services centrally. Given those differences, I still think it’s an accomplishment that almost every indicator, whether binary or continuous, has improved steadily since 2008. There remain a few states where significant operational gains could be made—for instance, those with longer average wait times than in the best-performing states—but these changes will require prioritizing these areas of practice statewide over multiple election cycles.

Whither the EPI?

The EPI was developed at the end of a decade that was kicked off by the most consequential election administration failure in American history. Constrained by data availability and methodological considerations, the EPI reflected many of the concerns of the time and was infused with the belief that shining a light on performance could spur improvements. An honest assessment of the EPI would conclude that it has had an important supporting role in spurring the high-priority reforms that seemed most important a decade ago.

The question to consider at this juncture is whether the EPI should change and, if so, how?

The simplest sub-question to consider is whether new indicators should be added and others retired. New issues have come into view over the past decade, such as cybersecurity and list maintenance, two aspects of election administration that are not reflected in the EPI. Indeed, as we wrote in the original methodology document, measures of security and integrity are generally underrepresented, despite our intention to treat them as equally important, whereas voter convenience remains relatively overrepresented in practice.

A harder question is how to make the EPI more central to policy debates about election administration. A close look at the EPI shows that many indicators reflect policies beyond the control of election administrators. (One indicator, turnout, is mostly a consequence of demographics and political campaigns.) And yet, over the years, I’ve noticed that it is mostly election administrators who pay attention to the EPI and feel aggrieved when their state performs poorly—often not because of something they’ve done, but because of their adherence to the law, which is their first duty.

The EPI is a policy wonk’s index. It exists alongside other efforts to rate state elections, such as the Cost of Voting Index and the Election Integrity Scorecard, which emphasize state laws rather than a comprehensive suite of output measures. (This is not to cast shade on these other indices, which do a good job of communicating their values and rating states accordingly.)

I continue to believe that voters are best served when they can cast a ballot conveniently and accurately, and can be assured that their ballot will count honestly. This relies on the basics of election administration being well executed—the blocking and tackling of elections, as I often call it. At the end of the day, that is what the EPI is all about.

Subscribe now

Share

Others have done well by handicapping the likely success of the threatened EO, criticizing his up-is-down reading of the Constitution, and fact-checking the claim that the U.S. is alone in allowing mail voting.

I want to focus on the least empirically challenged part of the post: that the president is “going to lead a movement to get rid of MAIL-IN BALLOTS, and also, while we’re at it, Highly “Inaccurate,” Very Expensive, and Seriously Controversial VOTING MACHINES. . .” (Emphasis added)

The only problem with this sentence is that he is already leading that movement—the push to delegitimize easily used mail-in ballots—and has been for a while. He is also claiming leadership of the anti-voting-machine movement, which will be the subject of another post.

Universal vote by mail is different from absentee ballots, sometimes in Donald Trump’s mind, and often in the minds of voters.

One of the things that confuses many people is that Donald Trump vehemently opposes voting by mail while still voting by mail himself. In the past, he has tried to make it clear that he distinguishes universal VBM and more traditional by-mail absentee voting. The clearest expression of this came in mid-2020, when he tweeted:

However, Trump has made this distinction less and less over time, to the point that his August 18, 2025 post seems to lump universal VBM in with traditional absentee mail ballots.

However, voters seem able to make this distinction. To see this, let’s start with their attitudes about universal VBM policies. Ever since it launched in 2008, the Survey of the Performance of American Elections has asked respondents whether they favored “run[ning] all elections by mail.”

Throughout this time, support for universal VBM has divided the two parties nationwide. Democrats’ support has grown steadily (with a 2020 spike), while Republicans’ support has held flat or declined, expanding the partisan gap from 13 points in 2008 to 53 points in 2020. The gap was cut nearly in half to 28 points in 2024, but is still twice that of the pre-pandemic period.

Donald Trump is not the only person to oppose universal VBM and still vote by mail. In 2024, 22% of Republicans and 28% of Democrats who opposed universal VBM nonetheless voted by mail in the election. If we exclude respondents from the eight universal VBM states in 2024, those numbers are 17% and 23%, respectively.

To be sure, these rates of mail-ballot usage are lower than among supporters of universal VBM. In the non-universal VBM states, mail ballots were used by 32% of Republicans who would support universal VBM and 41% of Democrats. So, opposition to universal VBM is correlated with a drop in the tendency to vote by mail, but it doesn’t drop it anywhere near to zero.

Clearly, voters—Republicans and Democrats alike—can distinguish between a policy that automatically mails every voter a ballot and requiring voters to ask for one. So, too, can party leaders. In the state legislative sessions held immediately after the 2020 election, as many Republican states were passing laws to tighten up voting procedures, no state that had had no-excuse absentee voting prior to 2020 eliminated it.

The future success of Trump’s anti-universal VBM movement faces political hurdles, not just legal ones.

If Trump wants his movement to succeed, he has about 90% of his supporters on his side on the issue of universal VBM, although a sizeable minority of his voters still rely on mail to vote. Not wanting to lose these discretionary by-mail voters, Republican state legislatures will likely be reluctant to roll back no-excuse absentee laws dramatically.

Of course, if the goal of the movement is really to roll back universal VBM, it has one important barrier: the state legislatures where VBM is currently the law.

It’s hard to believe that any federal court will agree with the proposition that

the States are merely an “agent” for the Federal Government in counting and tabulating the votes. They must do what the Federal Government, as represented by the President of the United States, tells them. . . .

Thus, to succeed in eliminating universal VBM, Trump will have to take his movement to the state legislative arena. Except for Utah, where universal VBM is scheduled for phase-out in 2029, universal VBM appears secure in the other states.

The most promising route to rolling back universal VBM in the remaining states would seem to be the citizen initiative. As it stands right now, support for universal VBM hovers around 50% in all states that have it. (See the figure below.) Five of the states where Democratic legislatures would block a universal VBM rollback (Oregon, Washington, Colorado, California, and Nevada) have the citizen initiative.1

In none of these states would an anti-VBM citizen initiative be a slam dunk. Yet, as the current initiative effort in California to require photo ID to vote suggests, it may be possible to harness public skepticism of some election administration practices to put conservative election policies in front of voters and then see what happens.

Trump’s movement may succeed as a political force when legal tactics fail.

I take President Trump at his word that he intends to use the full power of the federal government to try and force states with universal VBM to end the practice. As a legal and constitutional matter, I don’t think he’ll succeed. However, if it is a movement he wishes to lead, he may be more successful, though not universally so.

Trump’s opposition to universal VBM will undoubtedly spur conservative election activists in liberal states to ramp up efforts to gum up the works of mail ballots in these states. In this regard, they may learn from their efforts at pushing voting-machine skepticism.

At the same time, Trump will have to manage the enthusiasm of his supporters who would take his opposition to VBM as a signal that they should push hard to roll back no-excuse absentee voting in the many Republican states that have it. This raises the real possibility that, in the end, Republican voters and elected officials will end up bearing many of the costs of the movement’s efforts.

Subscribe now

Rockland County, New York has found itself facing a lawsuit over the results of the 2024 general election. Two voters, a U.S. Senate candidate, and an election advocacy group alleged irregularities in the election results that, they claim, call into question not only Rockland County’s outcome but results across the country.

In response, local political observers have offered a straightforward explanation: the political preferences of certain religious communities in Ramapo, a town in the county, where voters declined to support Harris despite backing Democratic Senate candidate Kirsten Gillibrand.

Election results in Palm Tree, New York, in neighboring Orange County, support the idea that some Haredi voters supported Gillibrand while largely withholding support from Harris. Palm Tree, which is coterminous with the village of Kiryas Joel, shows a voting pattern nearly identical to that in the predominantly Haredi areas of Ramapo.

Before proceeding to analyze Palm Tree’s results, let’s return to voting patterns in Rockland County. According to the complaint, Rockland County exhibited two significant anomalies. First, Diane Sare, a U.S. Senate candidate running under the LaRouche Party label, claimed that more people voted for her than the official election returns reflected. Her allegation of stolen votes relied on affidavits from a few Rockland County voters, including Sare herself. (The official returns reported 397 votes for Sare in Rockland County, compared to 72,003 for Gillibrand and 64,082 for the Republican, Michael D. Sapraicone. Sare received 39,421 votes statewide, out of nearly 8.4 million cast.)

Second, based on an analysis performed by the advocacy group SMART Elections, the lawsuit claims significant discrepancies between the number of votes cast for Gillibrand and those cast for Kamala Harris. (A parallel claim was made regarding the discrepancy between the vote totals for Sapraicone and Donald Trump.)

I will set aside the claims about Sare’s allegedly missing Senate votes—the number is small, and such affidavits are inherently unverifiable. Instead, I focus on Harris’s underperformance relative to Gillibrand, as it involves more votes.

It is true that Harris significantly underperformed Gillibrand in Rockland County when compared to the rest of the state. Statewide, Gillibrand received 59% of the vote, while Harris received 56%, a three-point gap. In Rockland County, Gillibrand received 53% and Harris 44%, producing a nine-point difference.

Drilling down further in Rockland County, the town of Ramapo emerges as the center of this discrepancy. Gillibrand earned 50% of the Senate vote in Ramapo, compared to Harris’s 32%—an 18-point gap.1 In the rest of the county, the margin was just three points—53% for Gillibrand vs. 50% for Harris.

Examining Ramapo’s election returns more closely, the most dramatic differences between Gillibrand and Harris are concentrated in a small number of precincts. A typical difference between the two candidates at the precinct level ranges from 3 to 10 points. However, in 10 precincts, the gap exceeds 40 points: precincts 21, 35, 55, 95, 97, 98, 109, 111, 117, and 118.

Where are these precincts? See the highlighted sections on the Ramapo electoral district map below. I have also marked the locations of two villages in Ramapo—Kaser and New Square. According to the 2015 American Community Survey, over 89% of households in these two villages spoke Yiddish at home. This is in contrast with 16% of the rest of the town.

This voting pattern made me wonder whether it occurred elsewhere in New York. It did.

This takes us to Orange County, New York, home to Kiryas Joel, a Haredi community involved in longstanding disputes over public services with neighboring jurisdictions. In Palm Tree, the town coterminous with Kiryas Joel, Gillibrand received 86% of the Senate vote, while Harris received just 2%. In the rest of Orange County, Gillibrand received 51% vs. 48% for Harris.

There’s one last issue to address in the Rockland County controversy. Some have linked the vote discrepancies to a claim that a de minimis pre-election update to ES&S voting machines contributed to tabulation errors. Rockland County uses ES&S equipment; Orange County doesn’t. The claim that this minor software patch to voting machines could have facilitated a targeted manipulation of votes in New York (and elsewhere?) is implausible on its face. The fact that Orange County uses a different voting equipment manufacturer should put to rest the idea that these discrepancies were due to software upgrades.

If I were one to editorialize, I would launch into a number of topics—starting with how the social media posterati glom onto fringe theories about election improprieties, and how legal proceedings in out-of-the-way jurisdictions can be amplified into national causes célèbres. I won’t say any more than this: the Rockland County election results in this case are a nothingburger.

(One last thing: for more information on the Rockland County controversy, check out the Snopes webpage on the topic.)

(A second last thing: Christopher Kenny has quite a detailed analysis of Rockland County here.) [Updated 6/23/25 at 11:03 a.m. EDT.]

I have thoughts, but I also have data. Let me address one part of the feasibility question on the proof of citizenship, to the degree that passports would be allowable as proof.

I am currently writing a report on the 2024 Survey of the Performance of American Elections, a survey I have conducted every presidential election (and some midterm elections) since 2008. One battery of questions we have asked every presidential election except 2020 is about the possession of various forms of identification, including driver’s licenses, birth certificates, and passports. For this post, let me focus on the possession of passports.

Passport Possession

Among respondents to the 2024 SPAE—a representative sample of registered voters—56% reported possessing a passport, which is up from 44% when the question was last asked in 2016. Of those who reported having a passport in 2024, 13% stated that it was expired, and 2% indicated that it did not have their legal name. All told, 48% reported they had an unexpired passport with the holder’s legal name.

One important feature of the SPAE is that the sampling unit is the state, which means we can get a pretty good estimate not only of who passport holders are demographically, but also where they live.

Below, I have plotted the percentage of respondents who reported having a non-expired passport in their own name in 2024 against the Center for American Progress’s estimate of the percentage of the whole population in a state that holds passports. Even though the CAP’s estimates are just that, estimates, and the SPAE’s estimates are based on self-reports, the two numbers line up pretty well, excepting that a few states at the high end of passport possession (Massachusetts, New York, New Jersey, and California) show much lower rates using the SPAE measure than the CAP measure.1

It has been well established that passport possession is correlated with predictable demographic characteristics, particularly income and education. Those patterns are clear in the SPAE data. For instance,

• 56% of respondents with at least some college education reported having a passport compared to 30% of those with a high school diploma or less, and

• 68% of respondents with a household income greater than $70k reported having a passport compared to 31% of respondents with a household income less than $70k.

Nearly equal fractions of Whites (46.8%) and Blacks (47.2%) reported holding a passport. Hispanic (58.0%) and Asian (64.8%) respondents were more likely to report holding a passport than either Whites or Blacks.

I tend to focus on administrative issues when considering things like proof of citizenship, but we cannot ignore the political dimensions, especially since we have the data. Among Democratic respondents, 54% reported having a passport, compared to 44% of Republicans and 38% of independents and third-party respondents.

This partisan difference is not only because blue state residents are more likely to hold passports than their counterparts in red states. If we control for the “state effect” using a technique called fixed-effects regression, Democrats are still nine percentage points more likely to hold a passport than Republicans. This difference remains at five percentage points even after further controlling for education, income, age, and the respondent's race.

In other words, if passport possession were required to register to vote—and clearly, there would be other ways to establish citizenship—Democratic voters would be advantaged over Republicans.

What about Birth Certificates?

Another way to establish citizenship could be a birth certificate. This is a document many more people have than passports. The 2024 SPAE reports a much higher rate of birth certificate possession at 89%. Unfortunately, the SPAE did not ask respondents whether the name on their birth certificate was their current legal name. This is particularly important in light of recent news reports that have highlighted the difficulties married women, in particular, face when using their birth certificates to establish citizenship during registration. Additionally, a 2023 report by the Pew Research Center found that 79% of married women changed their names upon marriage, with an additional 5% hyphenating their names.

If there were not a huge disparity between men and women in whether the name on the birth certificate matched the legal name, the passport-possession disparity might not be so important. Leaving aside these important gender differences, state statistics on birth certificate possession are more equitable across states than those on passport possession, ranging from 80% in the District of Columbia to 96% in West Virginia. Of particular interest is the fact that states with higher birth certificate possession are generally those with lower passport possession, although this correlation is not especially strong (r = -.28) (See the figure below).

Additionally, possession of a birth certificate is not as demographically stratified as possession of a passport. For instance, 88% of respondents who did not attend college reported having a birth certificate, compared to 89% of college attendees. The same comparison holds for households with annual incomes below and above $70k. Birth certificate possession is nearly equal by race, with the notable exception of Asian Americans, 70% of whom report having a birth certificate compared to 89% of the remaining respondents.

Thus, if birth certificates could be counted on to have the correct name of all eligible voters, the problem of inequitable possession of passports might be solved, or at least close to being solved. Yet, that seems to be a big “if.”

How Hard Is Establishing Citizenship for Voting? Harder than You Think

There is much more to be said about President Trump’s executive order on voting. It focuses on a policy outcome that is universally accepted: only citizens should vote in federal elections, and only eligible voters should be registered.

There is no evidence that non-citizen voting, let alone registration, is widespread in America. Yet, many Americans are worried about the problem. We know from extensive research that Americans most concerned about this issue will not be convinced that the problem is minor by simply repeating statistics about how rare non-citizen voting is and why it would be perilous to non-citizens themselves.

American citizens are also not required to possess citizenship proof and to check in with authorities when they move. This is one area where Americans’ fierce civil libertarian streak, which is unique in the world, confronts the difficult problem of ensuring that not only are non-citizens kept off the voting rolls but also that all Americans can be assured of this fact.

Because of this, even if the executive order survives legal challenge, of which I have my doubts, we will have dealt with one problem by creating others. There has to be a better way—but that’s a topic for another day.

We’ve come to the end of the time in the election cycle when the politically engaged, tired of obsessively going to FiveThirtyEight or the Silver Bulletin, look to other tea leaves. A few of you may still wonder what the mail and early in-person voting trends portend for the results tonight. Here are some thoughts on that.

(Before proceeding, a little nomenclature: Some people—journalists, election officials, and just normal folks—mean different things when they say “early voting.” When some refer to early voting, they mean any votes cast before Election Day, whether cast in person or at home. Others only mean votes cast in person. To try to avoid that confusion here, I will refer to the in-person version as “early voting,” and combined early in-person and by-mail voting as “advance voting.” I will refer to ballots cast after being mailed to a voter as “by-mail ballots.” )

My simple answer to the “what does this all mean” question is that things seem to be returning to normal in two related ways. First, voters are mostly returning to pre-pandemic voting patterns (with important exceptions). Second, campaigns are recognizing they need all the turnout weapons in their arsenal to win.

2024 represents the return to a long-term trend of increased advance voting

First, the patterns of the use of vote modes. As illustrated in the graph below, prior to 2020, the fraction of people voting on Election Day had been declining linearly for a quarter century, replaced in equal measure by people voting by mail and voting early. This pattern was interrupted by the pandemic, which shifted large numbers of voters out of Election Day voting booths into other modes, especially voting by mail. In 2022, mode usage either returned to the pre-pandemic trend or at least was much closer to the trend. (By-mail voting was about 5 points above trend in 2022, while early voting would have been right on trend if we account for the obvious seesaw pattern in the data between on- and off-year elections.)

The dashed lines in this graph are based on simple linear trends in the use of the three voting modes from 1996 to 2018, estimated using linear regression. Although the trends are based on the data from 1996 to 2018, they have been extended to 2024 to see how the pandemic and the recovery from it affected the long-term trends.

Not surprisingly, in 2020, Election Day voting was significantly below trend, while by-mail voting significantly exceeded it. Early voting was roughly what we would have expected without the pandemic. In 2022, Election Day voting returned us to where the long-term trend predicted; early voting was slightly below the trend, and mail voting slightly exceeded it.

The ends of the trend lines provide the predictions of mode usage in 2024 if we just allow the pre-pandemic trends to continue. This is obviously a mindless prediction, if you want to call out-of-sample prediction using regression mindless, but it does help define the expectations ballpark for this year.

Keep in mind that most states made changes to their laws that either facilitated or reined in advance voting before 2022, so state-law-induced factors weighing on 2024 are already baked into the trends for the most part. The decision by Republican strategists to validate advance voting in 2024 may have played a role in the growth of voting in the pre-Election Day period this year. Still, the shift could also have reflected voters deciding how they prefer to vote, given what’s going on in their lives.

As voters shift to early voting in 2024, they are also shifting away from by-mail voting, compared to 2020. Based on Michael McDonald’s indispensable early voting website and some back-of-the-envelop calculations of my own, it appears we are on target for over 40 million early votes in 2024, up from 36 million in 2020, and 48 million by-mail votes, down from 66 million in 2024. Both of these numbers are right where the trend lines in the above graph would predict if turnout ends up at around 160 million. This also means that 2024 may be the second election in a row in which most votes are cast before Election Day. (It will at least be close.)

The closing partisan gap

Of some note, in 2024, there has been a shift in thinking among Republicans about the value of voting by mail. Certainly, the GOP and its candidates are giving mixed signals about advance voting. Still, just as Donald Trump’s attack on by-mail voting helped create a partisan chasm in voting by mail in 2020, his campaign’s change of attitude in 2024 has to make a difference, too.

For the decade before 2020, Democrats were only slightly more likely to vote by mail than Republicans, and almost all that difference was accounted for by the greater availability of no-excuse absentee voting and universal vote-by-mail in predominantly Democratic states. As the graph below shows, in 2020, a gap of nearly 30 points opened up between Democrats and Republicans in usage rates of by-mail ballots. That gap was then cut down to 12 points in the 2022 midterm.

It appears that the gap has been diminished even further in 2024, although we cannot be 100% certain about how much. The party statistics reported by the states and communicated by Michael McDonald reflect the party registration of those returning ballots in states that record party registration. We won’t know for sure about the size of the partisan gap until surveys that track such things begin to report their results.

An increase in Republican by-mail voting in 2024 could just represent the demise of the stigma associated with mail voting. A decline in the use of by-mail voting by Democrats could also just represent a weakening of the sense that voting by mail is part of one’s Democratic duty.

Of course, it could be that a decline in advance voting by Democrats in comparison to Republicans could reflect a better mobilization effort by the Trump campaign and a lagging effort by Republicans. I’m doubting it, though. If I were a betting person, my money would be on voters and campaigns getting closer to a pre-pandemic normal. By Tuesday night, we’ll see if I’m right.

As Election Day approaches, it’s time to dust off what we learned about the blue shift/red mirage in 2020 and apply it to 2024.

Vote share trends in 2020

To jump to the chase, the following two graphs summarize partisan patterns to the reported vote in each state in 2020. Let’s start there.

The first graph shows Biden's two-party vote share from the close of polls in every state through midnight on Wednesday following Election Day. (The subgraphs are admittedly tiny, so click on this link and zoom in if you’d like to take a closer look.) The blue circles show each vote report during this period. The gray line shows the percentage of the vote reported and when it was reported.

For an election data nerd like me, I can spend hours studying this graph, but a couple of things stand out. First, although the phenomenon is called the blue shift, suggesting a steady evolution of vote share from a Trump majority to a Biden majority, in fact, for many states, the pattern was nonlinear. Take Georgia, for example. The very first vote report, with 4% of the eventual vote reported, showed Biden with 55% of the two-party vote. However, his share quickly dropped until it bottomed out at 42% just an hour and a half later. At that point, Biden’s share of the reported vote started a slow and steady rise. (He did not overtake Trump until Thursday, which is not on these charts.) Pennsylvania showed a similar pattern that was even more exaggerated.

Second, most states actually saw a red shift during this period. Setting aside the earliest vote reports (which I define as all reports before 10% of all ballots were reported), only 13 states saw Biden’s vote share grow during this time.

A broader view of the vote share shifts can be shown if we plot the two-party vote share for Biden against the percentage of all votes reported in a state. Those plots are shown in the next graph. (Click here if you want a bigger graph you can zoom in on.)

This graph perhaps provides a clearer view of the two patterns I discussed above when I confined myself to the first day and a half after the polls closed. Most states—31—saw Biden’s reported vote share start above his final vote in the state before dipping below his final share. Georgia and Pennsylvania again provide good examples of this.

These graphs help make the point that vote reports did not all follow the same path in 2020. They also show that the idea of the blue shift is more complicated than often understood. What most people are thinking of when they say that Georgia and Pennsylvania had blue shifts is what they experienced toward the end of the vote-counting period, when the nation was on tenterhooks about the outcome of the election. And what they fail to appreciate is that many states, in addition to Arizona, also had pronounced red shifts late into the count.

Some thoughts for 2024

These are the patterns from 2020. What do they tell us for 2024?

For most states, we can probably expect similar patterns. The patterns observed in the graphs above reflect the results of an interaction of state election laws, local practices, and unique urban/rural differences. Most of these things haven’t changed much in most states.

However, they have changed in some states. For instance, Georgia now requires its counties to report its early in-person and mail votes within an hour of polls closing. This could easily put 75% to 80% of all the state’s votes out at once, leaving little room for any movement afterward. It will also reduce, but not eliminate, the pattern of rural counties reporting first and urban counties coming in later. Similarly, Michigan’s allowing for pre-processing of mail ballots should dampen any blue or red shift and shorten the period over which it occurs.

Finally, reports of early in-person and mail voting suggest that Republicans are more active in turning out before Election Day than in 2020. In those states where the blue/red shift is caused by differentials in how the two parties’ supporters use different voting modes, differences in vote reports over time will be muted.

The simple answer to all these questions is that we’re likely to see similar patterns to 2020 and 2022, although some states will be more similar than others. The reason for the similarities is simple: for many states, the rules governing the counting and reporting of votes will be quite similar to 2020.

I can hear you saying, “Wait a minute. Didn’t 2020 see a surge of voting by mail that won’t repeat in 2024? Surely, that will affect what the vote reports look like.” In some states, that will be undoubtedly true. However, for most states, the laws in 2020 that sped up the counting of mail ballots, notably allowing for mail ballot pre-processing, will remain in place. To the degree that mail ballots may slow down the count, the presence of fewer mail ballots in 2024 may speed things up a bit. In addition, in many states, the mail ballots are the first to be reported, not the last, because they’re available to be tallied centrally either during Election Day or as soon as the polls close.

The pace of counting in 2020

In 2020, the team at the MIT Election Data and Science Lab (MEDSL) scraped the election results reported by the New York Times, which, in turn, were provided by Edison Research. This gives us a picture of how quickly states reported results in 2020 as a baseline for what to expect in 2024.

Also in 2020, MEDSL published a report that analyzed these data. One of the analyses we performed was on the velocity of vote reporting, measured in several ways. One simple way to illustrate cross-state differences was to show the percentage of votes reported 4, 8, 24, and 48 hours after polls closed in the state. A figure showing those numbers appears below, with the fastest-reporting state after four hours (Iowa) at the bottom and the slowest state (D.C.—I’ll call it a state for these purposes) at the top.

This figure helps to illustrate the frustration many voters and politicians felt on election night and beyond. While some states could get almost all their votes reported by the wee hours of Wednesday morning, only about half had reported 90% of their votes eight hours after polls closed. Among the seven battleground states this year, North Carolina was, by far, the fastest out of the shoot, trailed a long distance by Wisconsin, which was about in the middle of the pack. Michigan and Pennsylvania were among the ten slowest.

I invite the reader to study this figure (or look at the report) because it reveals all sorts of nuances about how states report their vote totals. Georgia, for instance, was a relatively slow-reporting state after four hours but quickly caught up after midnight of election night. After 48 hours, it was in the top half of states in terms of votes reported. On the other hand, Alaska reported about half its votes by Wednesday morning and then stopped for several days owing to the challenges in aggregating vote totals in that far-flung and road-poor state.

The role of pre-processing

One detail about vote reporting not evidenced in this figure is what the pre-processing of mail ballots does to the speed of voting counting. In the statistical analysis performed in the MEDSL report, the failure to pre-process did slow down the count, on average, but by a small amount. Within four hours of the polls closing, the average state with no pre-processing had reported 63% of its votes, compared to 78% that didn’t restrict pre-processing. At the twenty-four-hour point, the gap between states had essentially disappeared.

In a later academic paper written by me and two MEDSL colleagues, we also found that states with more mail ballots reported at a slower pace. However, the effect of mail-ballot load also declined over time.

Three thoughts for 2024

What does all this mean for 2024? I would highlight three things.

First, pre-processing make a difference, although it makes the biggest difference in the early hours of counting. In 2020, seventeen states prohibited processing of mail ballots before Election Day. In 2024, that number declined to seven. Unfortunately, three are battleground states: New Hampshire, Pennsylvania, and Wisconsin. (Michigan will allow pre-processing in 2024.)

Second, there will be fewer mail ballots in 2024 than in 2020 to be processed on Election Day in these states. At this point before Election Day in 2020, nearly 3 million mail ballots had been sent out in Pennsylvania; as of today, just over 2 million have been sent out for 2024. In Wisconsin, nearly 1.5 million mail ballots had been issued a week before the 2020 election; this year, the number is about 600,000. Thus, even though there will be a lot of mail ballots to process on Election Day, that number will be significantly less than in 2020.

Third, perceptions of vote-reporting speed in 2020 were heavily influenced by the closeness of the count. Georgia, for instance, got a reputation as a slow-counting state even though it was in the middle of the pack after midnight of election night. The long wait until Thursday for Georgia to be “called” for Biden wasn’t because Georgia was slow, but because of a small number of ballots—such as straggling UOCAVA ballots, provisional ballots, and damaged ballots—that needed to be processed before the count was all finished.

There are other things to say about the speed of vote reporting, especially about reporting in non-battleground states, but I will leave that for another time.

One last thing

I will end with this, however. It is important to remember that the reports of votes on election night are just that, reports. Once the initial unofficial reports are issued, localities and states begin the canvass, which confirms the results, resolves discrepancies, and moves toward official results. It is important to get the results reported quickly, but it is even more important for them to be accurate. In the days following the election, I hope the public will watch the canvassing process carefully, but also give election officials the room they need to make sure the results are correct.

Subscribe now