Voter ID and citizenship checks in the US

The count has nearly tripled since 2000, and the chart marks Wyoming's first law in 2021. Most of the change was laws being added or tightened rather than loosened: of 21 states that began with non-photo, non-strict laws, nine had made them stricter by 2021; only Texas and Virginia had eased theirs.

South Carolina first asked voters for a document in 1950; Hawaii, Texas, Florida and Alaska followed by 1980. The states with ID rules by 2000 "had Democratic and Republican majorities". Since then, NCSL says, "virtually all" bills to add or tighten ID have come from Republican-led states.

NCSL dates the acceleration to 2011, 2012 and 2013, when states without ID adopted it and others tightened theirs. On the chart, that is the steep climb from 2010 to 2014; the count has barely moved since.

The count of laws passed has run ahead of the count in force: 34 states had passed ID laws in October 2014 but 32 were in force. The chart marks two court blocks, Wisconsin's in 2014 and North Carolina's in 2020.

In 2009 photo-ID states were a minority, and only Georgia and Indiana made voters without ID return later. By 2026 photo-ID states outnumber non-photo ones two to one. The chart marks the latest moves; Louisiana's, which takes effect in 2027, will make it the 11th strict photo state, NCSL says.

NCSL's strict list included Texas and Arkansas in 2014 but shrank by 2020 as courts and legislatures eased laws, including Texas's and Virginia's. It then grew again: Arkansas and Missouri were back by 2022, North Carolina and Ohio joined by 2024 and New Hampshire by 2026, while Missouri left again.

The two lines cut across each other, so every combination exists, and the states that ask for no document check voters in other ways, most often by signature. The rules keep changing: NCSL's change log lists several changes in 2026 alone, two of them marked on the photo-ID chart above.

Some strict states make room for voters who cannot easily get ID:

• Georgia accepts a "free voter ID card issued by the state or county".
• Wisconsin's law was allowed to stand on the state's pledge "to provide temporary free IDs to those in need".
• Tennessee lets an indigent voter who cannot get ID without paying a fee sign an affidavit instead.

NCSL's tables do not say whether every strict state offers a free ID card.

After five states enacted documentary-proof-of-citizenship laws by 2011, the count did not move until 2024. Eight states then acted in three years.

The chart counts laws enacted, not laws applied to every registrant:

• Not in force: Alabama's and Kansas's laws, Kansas's "due to court decision". Florida's begins in 2027.
• Partly blocked: a federal judge reinstated New Hampshire's affidavit alternative in May 2026.
• Only when records fail: six of the 13 states, including Georgia and Florida, ask for documents only if state records cannot confirm citizenship.
• Federal races only: Utah and South Dakota let voters without proof vote in federal but not state races.

Ohio and Indiana open earliest, about four weeks before Election Day, and Wisconsin latest, two weeks before. Strict ID and early voting already sit together in most strict states, without any national agreement pairing them.

Mississippi and New Hampshire are strict-photo states and, with Alabama, the only three states without early in-person voting for all voters. Even where early voting exists, most strict states start later than the 27-day average across early-voting states.

The report asked states both to require photo ID and to supply it free, with mobile offices registering voters and issuing cards. It also recommended proof of citizenship at registration.

Three commissioners dissented from the ID recommendation. In a September 2005 interview, Carter said Georgia's new law, which charged about $20 for a five-year card or $35 for ten years, "is a poll tax", because the commission's plan required "that the voter ID card be free of charge". Georgia now offers a free card.

Italy's law requires an ID document "purché munito di fotografia" (provided it has a photograph). Britain requires photo ID for UK Parliament elections, Canada requires government photo ID with an address or two other authorised IDs, and France requires ID in communes of 1,000 or more people.

German voters must identify themselves "auf Verlangen" (on request), especially without their polling notice. Japanese voters are checked against the register, and a voter whose identity cannot be confirmed must declare who they are. Two of the seven have no routine document check.

Germany requires every citizen aged 16 or over to hold a valid ID card, and registers eligible voters automatically from residence records. Britain offers a free Voter Authority Certificate.

The Carter–Baker report observed that "most advanced democracies have fraud-proof voting or national ID cards". Whether each comparator's ID card is free or universal was not checked country by country.

Canada lets a voter without ID make a written declaration if another elector at the same polling station vouches for them; Italy lets a poll official or a known local voter attest to identity. Britain has no such fallback on the day: in 2023 its then government rejected vouching, saying it "would undermine the integrity of the check".

Northern Ireland's Chief Electoral Officer reported personation "on a planned and well organised basis" in the early 1980s, and 149 people were arrested for personation at the June 1983 general election. ID followed in 1985 and photo ID in 2003.

A 1998 government review found postal-vote malpractice but said personation at polling stations "has, however, been impossible to quantify", while police experience "certainly suggests that abuse is taking place".

The Electoral Commission later wrote that the extent of fraud "has never been quantified", so the impact of the measures could not be assessed accurately.

The Electoral Commission recommended ID in 2014 because polling-station voting "had few checks in place to detect fraud" once registration and postal voting were tightened. Lord Pickles, recommending pilots, accepted that allegations were low and prosecutions rare, and invoked a "precautionary principle". Since 2016, polling-station personation has produced five convictions and six cautions (as of August 2025).

Canada's 2007 law followed a Commons committee that was "struck by the absence of any requirement for an elector to confirm his or her identity" and saw this as "a significant deficiency … that could encourage fraudulent voting practices". The legislative summary cites no documented case of fraud as the reason.

Both lines fell at each round, and by about two-thirds from the 2023 local elections to the 2024 general election. At the general election, about 16,000 people, one in every 1,200 in-person voters, did not return and vote.

People reminded of the rule by greeters before reaching the ballot desk were not recorded, and some polling stations sent no return; the Commons Library adds that 20 constituencies sent no data in 2024.

About 4% of non-voters in 2024 gave an unprompted reason related to the ID rules, and non-voters in lower social grades were more likely to say they lacked the required ID. The Commission says it "cannot use these figures to accurately quantify the impact of voter ID on turnout".

The share of people who thought voting was safe from fraud rose from 72% at the 2019 general election to 87% in 2024.

The government's evaluation said it "cannot be determined" whether the checks made personation easier to detect. Seven polling-station personation allegations in 2024 led to no prosecution.

18-to-24-year-olds and ethnic-minority communities sit well below the Great Britain average; social grades differ little. Northern Ireland, where ID has been required for decades, was about level with Scotland and Wales at their first ID election.

The Electoral Commission attributed the fall mainly to ending the practice of carrying names forward. It rejected the claim that 120,000 people were disenfranchised, but found "some evidence" that eligible people previously registered had not re-enrolled.

The fall measures the registration reform passed with photo ID, not people turned away at the polls.

Researcher Stuart Wilks-Heeg, quoted by the Commons Library, found "no evidence to suggest a fall in turnout". The other measures cited:

Electoral-process reasons, which include being unable to prove identity or address, stayed a small share of non-voters' main reasons at all three elections. The category also includes other problems, so the share blocked by ID alone is smaller and is not published.

A Canadian voter with no ID may vote after a written declaration, if an elector listed at the same polling station shows ID and vouches for them. Canada's rule also accepts two non-photo IDs instead of one photo ID.

In eight of the twenty cases, election officials did the fraud themselves: stuffing the box, forging ballots or leaving returns out of the count.

In Chicago in 1982, election judges cast ballots in the names of voters who had moved or died, more than one-third of the votes in one precinct. In another precinct, a Republican judge voted names a Democratic precinct captain gave her on slips of paper.

The record does contain the fraud a polling-place ID check is aimed at. Tammany Hall's paid "repeaters" voted several times each, changing their facial hair between votes.

In Chicago, the 7th Circuit wrote in 1987, precinct captains "vote their proxies on election day or find some reliable party member to do so", while "cooperative election judges refrain from challenging multiple votes by the same person under different names".

In Washington in 2004, one of 19 suspected "ghost voters" was confirmed to have voted in person in someone else's name.

In fourteen cases nobody came to the polls under a false identity: officials changed or added ballots, ballots went by mail, or real voters sold their votes.

An ID check confirms that the voter is the person registered under that name, which this fraud never relied on. Levitt describes polling-place ID rules as designed for "people showing up at the polls pretending to be somebody else".

In Chicago and Kansas City, votes were cast in other people's names at the polls with election judges taking part or looking away. An ID rule is enforced by those same judges.

Whether a photo requirement would still have made the ghost voting harder (by forcing judges to accept missing ID openly, or by deterring party workers) is not something the records measure.

The 2005 commission found "no evidence of extensive fraud in U.S. elections or of multiple voting, but both occur, and it could affect the outcome of a close election".

It judged that a good ID system "could deter, detect, or eliminate" multiple voting and voting "using the identities of others or those who are deceased", the method of Chicago's ghost voting. It also called absentee ballots "the largest source of potential voter fraud".

Each row of the table gives how the fraud was done, whether a polling-place ID check would reach it, and what happened, as of the date shown.

Kansas City's 1936 fraud led to hundreds of convictions and Chicago's 1982 fraud to dozens (see the table). In Kansas City's ninth trial, the judge noted no jury had yet acquitted a defendant.

In one Kansas City precinct, the Republican congressional candidate received 120 votes and was credited with 46.

A precinct captain testified that he put about 150 "ghosts and sleepers" on the roll. His ward bosses were "not sending repeaters" because "it would be easier to write in the votes", so he stuffed the box himself.

In another precinct about 150 names were written into the poll book after the polls closed, with as many ballots added. In a third, one witness did see a man vote two or three times.

Chicago's 1982 fraud used registrants whom precinct workers knew had died or moved but left on the list; ballots were then cast in their names. Kansas City's fraud began with a padded roll.

A party to a 1987 lawsuit estimated that 9% of Chicago registrants were "tombstones, vacant lots, and people who have moved out of the city".

Chicago's 1982 fraud was exposed by an FBI investigation and a special grand jury, and Kansas City's by a federal grand jury and prosecutor. East Chicago and North Carolina's 9th District were overturned by a state court and a state election board. None of the records read here credits an identity check with catching a case.

Kennedy carried Illinois by about 9,000 votes. Edmund Kallina's study of the recount finds "a considerable number of votes" stolen in paper-ballot precincts from the Republican state's attorney, Benjamin Adamowski, and far fewer from Nixon.

On his reading, Adamowski would have won and Nixon would still have lost. How the votes were stolen is not described in the account read here.

The 1982 governor's race was decided by 5,074 votes statewide. The two Chicago schemes described on appeal cast straight Democratic tickets, the party of the losing candidate, Adlai Stevenson. The Heritage Foundation attributes an estimate of at least 100,000 fraudulent votes citywide to US Attorney Dan Webb, citing press reports from 1983. No primary source read gives a citywide count.

All three modern results that were voided or rerun turned on absentee or mail ballots.

The Supreme Court noted in 2008 that East Chicago's fraud was "perpetrated using absentee ballots and not in-person fraud".

Heritage tags far fewer entries as "impersonation fraud at the polls" than as absentee-ballot fraud, false registration or vote buying. Only "illegal assistance" and "altering the vote count" are smaller.

Fourteen of the 34 impersonation entries are also tagged as absentee-ballot fraud, and two more describe mail ballots, such as a Colorado man who submitted his missing wife's absentee ballot in 2020. At most 18 describe only voting in person under another's name.

When the Supreme Court upheld Indiana's photo-ID law in 2008, the lead opinion said the only fraud the law addresses is in-person impersonation, and "the record contains no evidence of any such fraud actually occurring in Indiana at any time in its history".

It upheld the law anyway, citing "flagrant examples" documented elsewhere in US history and the risk to close elections.

In News21's database of alleged cases since 2000, in-person impersonation is by far the smallest named category.

Justin Levitt, counting every specific, credible allegation from 2000 to 2014 rather than prosecutions, found about 31 incidents among more than 1 billion ballots cast in general and primary elections. Both counts share the detection limit above.

As the Supreme Court recounted in Louisiana v. United States (1965) :

After 1956 a legislative committee and the Citizens Councils challenged Black voters on the rolls, and "thousands of Negroes, but virtually no whites, were purged".

By the Attorney General's estimates to Congress in 1965, Black registration barely moved in the decade before the Act, with white registration "roughly 50 percentage points or more ahead".

The table Congress used in 2006, charted above, shows the gap still there in 1965: Mississippi's white rate was 69.9 percent.

The courts found that Louisiana's interpretation test gave registrars "a virtually uncontrolled discretion as to who should vote and who should not", and that registrars in 21 parishes used it against Black applicants.

The other main tool, the poll tax, was a charge for voting; the Supreme Court struck it down for state elections in 1966. Whether a uniform, free photo ID resembles these tools is a judgement the sources do not settle.

By 2004 Black registration was higher than white in Georgia and Mississippi; the largest remaining gap was Virginia's (57.4 percent against 68.2).

The Supreme Court majority wrote that "there is no doubt that these improvements are in large part because of the Voting Rights Act".

The secret ballot spread after 1888; reformers wanted to end vote buying, since a buyer could no longer watch the bribed voter.

The spread of state voter-ID laws since 2000 is covered in the report on ID rules. No source read measures whether any of these rules caused the decline of machine fraud.

In Miami-Dade, over half of the Black voters who appealed were reinstated.

Stuart's analysis put errors at "probably more than 25%" of the list, with Black registrants more often listed in error and Republican county supervisors more likely to use it.

Abigail Thernstrom testified to the Senate in 2001 that the Commission's own data showed whites "erroneously placed on the purge list at almost twice the rate of blacks". The sources agree the list was badly inaccurate; they disagree on whether its errors fell harder on Black voters.

Georgia, Iowa, Louisiana, Ohio and Kansas each found some non-citizens on their rolls, and no state that ran a review reported zero. The chart shows each count per million registrants; the raw counts are in its note.

Even the tallest bar, Louisiana's preliminary count reaching back to records from the 1980s, is about 130 per million registrants. Georgia's confirmed count is the shortest, and Kansas's court finding was 0.002% of its registrants.

The bars span two orders of magnitude partly because the states count different things: Georgia counted only cases it had "conclusively identified", while Ohio and Louisiana counted apparent or suspected cases not tested in court.

Texas ran its entire roll, more than 18 million voters, through SAVE in 2025 and flagged about 0.015% of it. Ohio's office says it now identifies non-citizens "at the time of registration".

Before SAVE was opened to bulk checks in 2025, states such as Georgia and North Carolina relied on licence, jury and individual immigration-record checks.

The longest record comes from the Kansas litigation. The district court found that at most 67 non-citizens registered or tried to register, and 39 got onto the rolls; it found the cases "largely explained by administrative error, confusion, or mistake".

Kansas's proof-of-citizenship law, meanwhile, left 31,089 applicants denied registration. These are applicants who did not supply documents, not people shown to be citizens.

Georgia's first citizenship audit, announced in March 2022, counted applicants held at registration rather than people on the roll:

• 2,258 applications held in "pending citizenship" status; none was added to the rolls and none had voted.
• 1,634 of them could not be verified by SAVE; the office called them non-citizens.
• 624 could not be run through SAVE for lack of information, and were held because licence records listed them as non-citizens.

Because none reached the rolls, they are not in the chart. "Not verified" by SAVE is not the same as confirmed: SAVE can miss citizens whose status is absent from federal records.

The reviews describe the same few routes:

• Clerical error: Texas found people registered by county staff even though they had marked themselves non-citizens on the application.
• Licence-record coding: Arizona's motor vehicle agency coded everyone licensed before 1996 as a citizen, so 218,000 registrants had no proof of citizenship on file. All had sworn they were citizens, and by October 2024 two had been identified as non-citizens.
• Misunderstanding: some North Carolina voters said they did not know they could not vote.
• Administrative anomalies: the Kansas court attributed most of its 39 cases to error and confusion.

Georgia found its 20 because they had sworn they were not citizens to get out of jury duty. Each route runs through a check at the point of registration, which is where a records check or a documentary-proof rule would apply.

Every review that checked ballots found a few non-citizen voters; the chart shows the rate for each single election. Iowa also found five whose ballots were rejected. Counts that span several elections, or that are referrals rather than findings, are below.

Iowa's 2024 rate is the highest in the chart, and North Carolina's 2016 rate is less than half of it.

The lowest bar, the Brennan Center's 2016 survey, comes from interviews with officials in 42 jurisdictions with large non-citizen populations, not a records check; New York City declined to take part. The Brennan Center is an advocacy organisation.

Ohio's suspected voter-fraud referrals of all kinds, sent to county prosecutors:

• 633 sent;
• 12 taken up; prosecutors said the others lacked evidence to pursue indictments;
• 6 indicted after the Attorney General took the cases to grand juries.

After the 2018 referrals, "only a handful were ever prosecuted". Some non-citizens who voted said they did not know they were barred, and North Carolina found some had been misinformed by registration drives or a precinct official.

A case not prosecuted is not thereby shown to involve an eligible voter, and a referral is not thereby shown to involve an ineligible one.

In every program with a published follow-up, a large share of the people flagged turned out to be citizens, from about a sixth in Denton County to nearly all of North Carolina's licence-record flags.

The share depends on the database. Licence records keep showing people as non-citizens after they naturalise, as in Texas in 2019 and North Carolina in 2016. SAVE's misses fall on people whose citizenship is missing from federal records, such as those who acquired it through a parent.

• October 2025: SAVE flagged 2,724 of more than 18 million registrants; counties mailed notices and cancelled those who did not reply in 30 days.
• December 2025: Denton County had 84 flagged voters: 14 proved citizenship, 14 had marked "not a citizen" on their forms and were registered in error (none had voted), and 56 were cancelled for not replying, five of them because their notices were undeliverable.
• July 2026: the state elections director said SAVE had "some inaccurate information" and told counties to reinstate some voters, without yet saying how many.
• September 2026: 578 had demonstrated citizenship, 506 of them because passport data reached SAVE after the state's first run, and 117 cases were referred for investigation of possible non-citizen voting.

No count of confirmed non-citizens among the 2,724 has been published.

In August 2024 Alabama made 3,251 people inactive, using non-citizen numbers recorded by its unemployment and licence agencies. The Secretary of State's chief of staff testified to what followed:

• About 2,000 were citizens: more than 900 proved it themselves, and another 1,000 were cleared when the office checked licence records.
• At least 159 were removed, "a few" of whom said on their forms that they were non-citizens.

A federal court halted the program because it ran inside the 90-day window before a federal election. Lawyers opposing a similar Virginia program said the licence-record check confirmed 93.8% of Alabama's checkbox-based flags as citizens, without confirming the rest as non-citizens.

Two weeks before the 2024 election, Iowa challenged the ballots of 2,176 people who had once told its transportation department they were not citizens.

After it gained access to federal immigration records, it identified 277 non-citizen registrants and 35 voters. The later count is not described as a subset of the earlier list, so it is a smaller final count from a better source, not a measured rate of wrong flags.

Virginia's program removed 1,610 registrants inside the 90-day window in 2024. According to the voters' lawyers, the district court found the state had not shown that any of them were non-citizens, while crediting evidence that citizens had been removed.

The Supreme Court stayed that order on 30 October 2024, with three justices dissenting. No count of how many of them were citizens has been found.

What the Solicitor General told the Supreme Court in September 2026 about the revamped SAVE:

• Checked: more than 65 million voters in 26 states.
• Flagged as potential non-citizens: 28,635. These are flags, not confirmed non-citizens.
• Error rate: not released, as the groups challenging it noted.

USCIS's own fact sheet says SAVE may be unable to confirm people who acquired citizenship through a parent. In 2016, before the overhaul, North Carolina found that about 75% of people who later proved citizenship were still listed as non-citizens in SAVE.

Texas did not check its SAVE list against its own licence records before sending it to counties. In Travis County, 11 of 97 flagged voters had already given the licence agency proof of citizenship.

Across the counties that answered Votebeat, at least 33 citizens had been identified by December 2025. The Texas Secretary of State wrote to USCIS that state licence records sometimes "inaccurately reflect a person's current citizenship status".

In 2019 Texas questioned the citizenship of almost 100,000 registrants, using licence records from people who had once said they were not citizens. The state later said it had overstated the list by about 25,000 names, counties found hundreds of naturalised citizens on it, and the state agreed in a settlement to scrap it.

Each review produces two counts: ineligible registrants found, and eligible citizens wrongly flagged.

In the reviews with published follow-ups, citizens wrongly flagged outnumber the non-citizens confirmed.

Both margins dwarf the non-citizen voters the same states' reviews found, even counting every DHS-flagged Georgian who had ever voted in any election. Iowa's count is statewide, so fewer could have been in the close district.

North Carolina's audit found that even if every ineligible ballot it identified in 2016 had gone to the winner, no race would have had a different outcome.

Louisiana's Secretary of State said fraud was not systemic but noted that some local elections had been decided by a handful of votes; she did not say whether any non-citizen votes affected them.

A race decided by single digits is within the range of counts found statewide. No source here matches non-citizen voters to the precincts or districts where they voted.

Every batch in the chart contains non-citizen registrants, at sizes that vary widely. In October 2025 Ohio forwarded to the Justice Department 1,084 cases going back to 2018 that included 167 people who allegedly voted in a federal election. The batches cover different elections and methods, so they do not show a trend.

Ohio had 8,159,946 registered voters and 5,851,387 ballots cast in November 2024. The alleged voters in the 2025 referral span every federal election from 2018 to 2024; dividing them by one election's ballots would overstate the rate. The Secretary of State, a Republican, called non-citizen registration "exceedingly rare".

Texas's line rises after its ID-number rule and stays well above both its 2020 rate and the national line. Texas also sent 5.8% of its 2024 mail ballots into curing, the process for fixing errors, against 1.5% nationally. The figures are not split by reason, so the share due to the ID rule is unknown.

The March 2022 primary's spike (chart below) was not repeated in November. The state election office credited redesigned envelopes, extra instructions, voter education and voters getting used to the rule. Note too that 2020 was Texas's low point: its 2016 and 2018 rates were about twice as high.

Georgia's line, high under signature matching in 2016 and 2018, has sat close to the national line since SB 202 replaced signatures with an ID number. Its 2020 point is a floor, because the state's count missed some county rejections.

In a survey of county election offices, 32% said fewer mail ballots were rejected in 2022, though a plurality said the new verification had not reduced rejections.

Texas's number had to match whichever one was on the voter's registration record, and rejections rose; Georgia's replaced signature matching, and rejections fell. So an ID requirement's effect on rejections depends on its design and what it replaces. It is not evidence about polling-place photo ID.

In November 2022 the gap between white and nonwhite turnout was the widest since at least 2014.

The white–Black gap also rose in 2016 and 2020 before the law and fell in the 2018 midterm. The Brennan Center, which wrote that SB 202 "erected multiple barriers to voting", says "we cannot say whether the passage of S.B. 202 is the cause" of the 2022 widening. Its 2024 analysis is in the next chart.

In a survey of more than 1,200 Georgians who voted in 2022, few reported problems:

• 1.1% said they had a problem casting a ballot.
• 75% of in-person voters waited ten minutes or less.
• 72% rated their experience excellent, up from 55% after 2020.

The survey reached only people who voted, so it cannot show how many were deterred.

The white–Black gap grew about as much from 2020 to 2024 as from 2016 to 2020, but for a different reason: this time Black turnout fell slightly, where before both groups rose. This analysis uses a different turnout estimate from the 2014–2022 chart, so the two sets of gaps are not comparable.

The bar for no photo ID of any kind, about 1% of adult citizens, is a fraction of the bar for a licence without current name or address, nearly always an old address. Whether that group is burdened depends on the rule: a law accepting any current photo ID does not touch them; one requiring a matching address would.

Over 9% of adult citizens have no birth certificate, passport, naturalization or citizenship certificate, or could not get to one by the next day; far fewer have none at all. The survey did not ask about the marriage or divorce records people who changed their name usually need, so its authors call the estimate for women a lower bound.

Its sponsors were VoteRiders (which helps voters get ID), Public Wise, the Brennan Center and the University of Maryland's Center for Democracy and Civic Engagement; the firm SSRS fielded it. It is self-reported and the only recent national measure; no survey by a government agency or a group favouring ID laws was found to compare it with.

Not driving is by far the most common reason for having no licence; cost and other bureaucratic or economic reasons together account for about a fifth. In 2014 GAO found licence fees in strict-ID states ranged from $14.50 to $58.50. No source measures the time and travel needed to obtain documents.

On the chart's default view, no licence at all, Black, Hispanic and independent citizens are about three times as likely as white citizens or Republicans to have none. For white citizens the more common problem is an outdated licence. A missing licence is not the same as missing photo ID, which the survey does not break down by group.

The share falls steadily with age and with income. For the young the gap is mostly addresses: people who have recently moved. For those earning under $30,000, most of it is having no licence at all.

Every group has a gap, Republicans included; independents and people of color have the largest. Because there are more white than nonwhite citizens, more white citizens in total lack ready access: over 12.9 million against over 8.4 million.

The blocked applications were 12.4% of new registrations from 2013 to 2015; the confirmed non-citizens on the rolls were 0.002% of registered voters.

The trial court noted that even the state's expert's estimate implied more than 99% of applicants held in suspense were citizens. In 2020 the Tenth Circuit upheld its injunction, holding that the requirement unconstitutionally burdened the right to vote and was preempted by federal registration law.

The court did find confirmed non-citizen registrations. Kansas also offered anecdotes, including a clerk's letter about hog-farm employees taken by their employer to register who felt pressured "even though they may not be legal".

In 2016 the court found that evidence "insufficient to show that noncitizens actually voted". After trial it found the confirmed cases "largely explained by administrative error, confusion, or mistake".

Arizona has required proof of citizenship for state elections since 2004, and accepts a licence issued after 1996 as proof. Its motor-vehicle records coded everyone licensed earlier as a citizen without seeing proof, so at least 218,000 voters had full-ballot status without the document on file.

All had sworn to their citizenship, and when this was announced in October 2024 two had been identified as non-citizens. The case shows that long-registered citizens have often never been asked for documents, and that the records used to check citizenship can be wrong.

Only a handful of the referrals proved to be non-citizens who had voted, all licensed before October 1996, the group whose records had wrongly shown citizenship. No statewide outcome for Arizona's 218,000 has been published.

Statewide, 24,636 mail ballots were rejected, and Harris County's rate leapt from almost none in its 2018 primary. Officials said most were rejected for a missing or mismatched ID number: voters missed the field under the envelope flap, or the number had to match one on a registration record that could be decades old.

Very few rejected voters found another way: "roughly 90%" did not take part in the primary at all. They were not occasional voters; about 85% had voted in every general election from 2016 to 2020.

Texas allows mail voting only for certain groups, including voters aged 65 and over and people with disabilities. The study covers one primary; no follow-up exists for the general elections, where rejection rates were lower.

Cantoni and Pons's nationwide panel of 1.6 billion voter-record observations, 2008–2018, found strict ID laws had "no negative effect on registration or turnout, overall or for any group defined by race, gender, age, or party affiliation". It covers ID at the polls, not proof of citizenship at registration or ID numbers on mail ballots.

The panel also finds "strict ID requirements have no effect on fraud – actual or perceived".

GAO compared Kansas and Tennessee, which tightened their ID rules for the 2012 general election, with four states that did not, and found turnout fell more in both. GAO says the estimates do not apply to other states or years; the two states' election offices disputed its comparison states and one dataset.

Of the ten earlier studies GAO reviewed, five found no significant turnout effect, four a decrease and one an increase.

GAO's Kansas and Tennessee estimates lie inside the national study's 95 percent confidence interval for the average effect. The national study rules out large average effects, not effects of that size in particular states or elections.

Neither measures burden directly. The blocked applicants, rejected ballots and people without ID in this report met an obstacle; turnout studies measure whether the total number voting changed.

In Kansas the extra drop was largest for 18-year-olds, then for the newly registered, then for Black registrants; Tennessee showed the same pattern, smaller for age and race. GAO found no larger effect for Hispanic or Asian-American registrants.

The national study's estimate for non-white relative to white turnout is slightly positive and not significantly different from zero. Campaigns became 4.7 points more likely to contact non-white voters, which "might have offset modest effects" of the laws. If so, the lack of a turnout effect partly reflects effort spent overcoming the rule, not a rule with no cost.

Both shares fell at each round; at the July 2024 general election those who did not return came to about 16,000 people. About two-thirds of those first turned away in 2024 came back with ID and voted.

Around 4% of people who did not vote in 2024 gave an unprompted reason related to the ID rules, and non-voters in lower social grades were more likely to say they lacked the required ID. This is the closest published measure of deterrence beyond those turned away at the door; no US equivalent was found.

The chart shows the count and the share of registrants for each cycle. In the latest, most states removed between 3% and 10% of their lists. These are records taken off the lists, not people found to have voted illegally: the same report counts moves, deaths and unanswered notices as the main reasons (see the next chart).

The count climbed through most cycles, while the share of registrants removed stayed in a band, from a low in the 2008 cycle to a high in 2024. Part of the rise in the count is more states reporting: 42 answered in the 2006 cycle, and every state with registration from 2018.

Because the share has moved much less than the count, most of the rest reflects longer lists rather than a higher removal rate. No source links the rise to a change in how often voters move or die.

The Election Assistance Commission (EAC) says that because removal "can take up to two federal general election cycles", "it is inevitable that voter lists will contain some number of voter records for individuals who are no longer eligible to vote."

This is a statement about records, built into the federal process; it is not an estimate of how many there are.

The shares are from the EAC's corrected 2024 figures. The same three were the largest named reasons in every cycle since 2006, although in the 2006 cycle the catch-all "other" category, which then included duplicates, was larger than any of them. In the latest cycle both moves and non-response reached their highest counts in the series.

Non-response has been the single largest reason in most cycles, at between about a third and two-fifths of removals.

It rests on silence rather than on a record that the voter moved: the registrant did not return a card and did not vote for about four years. How many of these people were still living at their address is not measured nationally (see the Ohio chart below and Data still needed).

The chart runs from Utah at the bottom to Puerto Rico at the top, while nearly two-thirds of states removed between 3% and 10%. Some of the spread is definitional: half of Texas's latest removals were duplicate records, a category first counted in 2024; states that merge duplicates rather than delete them report fewer removals.

States sent notices to more of their active voters in the latest cycle than in either of the two before (see the chart note). Most were never returned, and fewer than one in ten came back undeliverable. An unreturned notice is what starts the non-response route in the reasons chart.

Most recipients never replied, and of those who did, about four times as many said they had not moved as said they had.

Justice Breyer read the silence as showing "nothing at all that is statutorily significant"; the Court's majority held that federal law nonetheless makes removal "mandatory" once the notice and two-election conditions are met.

Amici cited by Justice Sotomayor reported that African-American-majority neighbourhoods in downtown Cincinnati had 10% of voters removed for inactivity, against 4% in a majority-white suburb.

The Electronic Registration Information Center (ERIC), a members' compact of state election offices, matches voter files against motor-vehicle and Social Security death data and sends members leads to check. Its totals to the end of September 2026:

• In-state address updates: 31,170,181
• Cross-state movers: 14,600,683
• In-state duplicates: 1,459,110
• Deceased registrants: 681,231

These are leads, not removals: members must still contact movers under federal and state law. ERIC itself says "Illegal voting is exceedingly rare".

Cross-state mover reports peaked in 2021 and fell steeply as members left; the members reading shows the drop.

A cross-state match needs both states in the compact, so fewer members means fewer possible matches. ERIC does not say how much of the fall is due to departures rather than fewer moves.

NPR reported in October 2023 that "Nine states — all Republican-led — have now withdrawn from ERIC" without a replacement. In order of departure they were Louisiana, Alabama, Missouri, Florida, West Virginia, Iowa, Ohio, Virginia and Texas. The chart marks the report, Hawaii's joining and Virginia's order to rejoin.

Iowa objected to ERIC's bylaws, including its requirement that members contact eligible but unregistered people; Virginia cited data security. ERIC's current list has 27 states and the District of Columbia.

Pew's study, using data from 2004 to March 2011 and published in February 2012, put inaccurate or no-longer-valid records at nearly 13 percent of the national total; the chart breaks that down by type. Pew said the analysis "likely underestimates" the problem.

The Supreme Court's majority in Husted (2018) opened with this estimate, and ERIC's website still cites the same 2012 study for the scale of the problem. No later national estimate of the same kind was found.

The figure predates ERIC, which began in 2012, and the growth in removals shown above, so it cannot be read as the size of the problem today.

Pew counted as "inaccurate" an eligible voter whose file has wrong details, such as an old address, alongside "no longer valid" records of people who moved or died. NPR's report on the study said "There's little evidence that this has led to widespread voter fraud". The study measured records, not ballots cast in those names.

The Eleventh Circuit held in 2014 that Florida's 2012 non-citizen program, first using motor-vehicle records and then the federal SAVE database, broke the 90-day limit; two named plaintiffs flagged by it were citizens.

The court said eligible voters "removed days or weeks before Election Day will likely not be able to correct the State's errors in time to vote." In October 2024 a federal judge blocked Alabama's program on the same ground, finding it began "eighty-four days before" the election.

The limit does not cover individual removals for death, request or conviction.

In Husted v. A. Philip Randolph Institute (2018), the Court ruled 5–4 that Ohio's process, which uses failure to vote as the reason to send a confirmation notice and removes only after the notice and two-election steps, "does not violate the Failure-to-Vote Clause or any other part of the NVRA". Four justices dissented.

On 30 October 2024 the Supreme Court stayed a district-court order that had required Virginia to restore registrants removed under its non-citizen program; three justices would have denied the stay.

The district court had found, according to the challengers' filing, that the state "failed to show that any of the 1,610 people removed" were non-citizens. No count of citizens among them has been published.

Every programme in the table flagged people who were eligible. The Brennan Center describes the two usual causes as bad source data (Arkansas's felony list) and loose matching rules (Texas's "weak" death matches, Crosscheck's name-and-birth-date match).

A 2017 Stanford-led study estimated that Crosscheck applied nationally would "impede 300 legal votes for every double vote prevented." In North Carolina, the challengers' brief quotes the Fourth Circuit as finding "97.6% of persons identified by the DMV as noncitizens, in fact were citizens".

A Harvard Kennedy School analysis estimated that "probably more than 25%" of names on the list were there in error, that more Black registrants were wrongly listed than white or Latino ones, and that Republican county supervisors were more likely to use it. In Miami-Dade, over half of the African Americans who appealed were reinstated.

Abigail Thernstrom, a dissenting member of the US Commission on Civil Rights, testified that the commission's own data showed "whites were erroneously placed on the purge list at almost twice the rate of blacks."

The federal SAVE database was overhauled in 2025 to check citizenship in bulk. The Solicitor General's figures, which are not audited:

• Checked: more than 65 million registrants in 26 states.
• Flagged as potential non-citizens: 28,635.

In Texas:

• Flagged: 2,724, about 0.015% of its registrants.
• July 2026: the elections director said SAVE had "some inaccurate information" while "most of the data" matched the licence agency's.
• September 2026: 578 of the 2,724 (21.2%) had demonstrated citizenship, 506 of them because passport information was added to SAVE after the first run.

Flagged Texans who do not reply within 30 days are cancelled but can be reinstated at the polls with proof of citizenship. The non-citizen findings themselves are in Non-Citizens on US Voter Rolls.

The Brennan Center's "almost 16 million voters" purged between 2014 and 2016 is the EAC's total of removals for every reason, more than half of them moves and deaths (see the first two charts).

The Brennan Center itself says that "When done correctly, purges ensure the voter rolls are accurate and up-to-date." The errors in the table are real but are thousands to hundreds of thousands of records against removals in the tens of millions; no national count of wrongful removals exists.

Judicial Watch counts more than five million names removed since 2017 after its NVRA suits, the largest in Los Angeles County, New York City and Kentucky (see the table). These were registrations removed through the notice-and-wait process; the group describes them as ineligible, and they are part of the national totals in the first chart.

The county claims compare registrations, including inactive registrants awaiting the two-election clock, with sample-based Census estimates of eligible citizens. Judicial Watch's release says inactive voters "are still registered voters"; the statistician quoted by Snopes said "If you haven't voted, you haven't voted fraudulently."

A registration rate above 100% therefore measures stale records the process has not yet removed, which the EAC says lists will always carry, rather than ballots cast.

Every Pew reading since 2018 shows large majorities of both parties in favour of photo ID (chart). Gallup's October 2024 poll agrees, at 84% of adults.

MIT's post-election survey of voters finds Republican support "nearly unanimous" (92.4% in 2024) and Democratic support rebounding after a dip.

In Pew's series, Democratic support was flat at 61% from 2021 to 2023 and then rose in 2024 and 2025 (chart). Gallup's Democrats dipped in 2022 and then rose 14 points by 2024.

Republican support stayed in the 90s throughout. The rise came while photo ID was being argued over in Congress and the states; the surveys do not establish why it happened.

Pew finds at least seven in ten adults in each racial and ethnic group it reports favour photo ID requirements (2025). Among Democrats in 2021, White Democrats were the least supportive.

Automatic registration splits the parties far more than photo ID, and Republican support for it fell after 2018 and has not recovered (chart).

No-excuse mail voting splits them further: 83% of Democrats and 32% of Republicans favoured it in 2025. Of the access measures Pew asks about, only early in-person voting has majority support in both parties.

In Pew's series, Republican support for two weeks of early in-person voting was nearly flat from 2021 to January 2024, then rose in August 2025 (chart).

Gallup's differently worded item runs the other way over a longer period: Republican support fell between 2016 and 2024 while Democrats' stayed high (Gallup chart below). Read together, the two series show Republican support falling after 2016 and partly recovering by 2025.

Photo ID (83%) and two weeks of early voting (80%) were asked of the same half-sample in August 2025. Pew does not publish how many respondents favoured both. The two shares alone guarantee that a large majority backed both, but not the exact size of that group; see Data still needed.

In Gallup's three readings, photo ID and early voting each have the support of at least three in four adults, and automatic registration less (chart). The partisan gap on photo ID widened from 32 to 44 points between 2016 and 2022 before narrowing again.

Support for requiring proof of citizenship at first registration matches support for photo ID, and divides by party and race in the same way (chart).

Gallup asked the question for the first time in 2024, so there is no trend. The question does not say what documents would count, which is where the dispute over such rules lies (see the burdens of ID and citizenship rules).

Republicans have grown more likely to expect ineligible voting to be a major problem, and Democrats less likely (chart). Concern that eligible voters will be blocked is similar in both parties, apart from a Democratic spike in 2020 (next chart).

In open-ended answers in October 2026, 19% of Republican voters raised illegitimate or fraudulent votes.

Democratic voters more often raised legitimate voters being denied access (10%) or interference by the Trump administration (11%).

90% of adults said in April 2026 that it is important that "no eligible voters are prevented from voting" and 85% that "no ineligible voters are permitted to vote". The parties disagree about which goal current elections fall short of, not about whether both matter.

Each side is least confident in the safeguard the other side worries about less (chart). Republican confidence that noncitizens will be stopped is up from 35% among Trump voters in 2024.

Republicans were more confident than Democrats under George W. Bush, the two converged in 2008, and Democrats were more confident under Barack Obama and Joe Biden (changes of president are marked under the chart). The gap reached a record 56 points in 2024.

A study pooling about 30 national surveys from 2000 to 2012 finds "a consistent relationship between voting for the winner and confidence". It attributes the decade's fall in national confidence almost entirely to "a forty-point decline among Republicans".

Gallup names two exceptions. In 2008 both parties stood at 57%, and in 2020 Republican confidence fell under a Republican president "amid then-President Donald Trump's public questioning of the validity of mail voting" (marked on the chart).

A study of 2016 finds the winner effect outweighed pre-election "rigged" messaging from elites. In 2018 there was essentially no winner effect at all.

More than half of Americans have been confident in every Gallup reading since 2004 (chart), most of all in 2004 and 2006. But the share "not at all confident" rose from 6% in 2004 to 19% in 2024.

Confidence in one's own polling place is higher: 76% in 2024.

MIT's post-election survey of voters shows the swing directly: each party's confidence in the nationwide count rises after its candidate wins the presidency and falls after a loss (chart).

Earlier work on 2000–2012 put the average rise among winners at about 20 points for confidence in one's own vote, against a fall of about 2 points among losers.

In 2016 Democrats went from 15 points more confident than Republicans before the election to 13 points less confident after it.

In July 2026 Republicans and Democrats were about equally confident that the November elections would be conducted fairly and accurately (chart).

Republican confidence rose and Democratic confidence fell after the change of administration (marked under the chart). The convergence came from both sides moving, not from either party's confidence reaching earlier highs.

Voters are less confident in the mail count than the in-person count, and Democratic confidence in mail ballots fell sharply in 2026 while Republicans' edged up (chart).

Confidence follows voters' own method: 76% of mail and absentee voters are confident in the mail count, against 47% of in-person voters. Mail-ballot rules are covered in a separate topic.

US confidence in the honesty of elections has averaged 45% since 2006, with its low in 2016 (chart). This is a different question from Gallup's domestic vote-count item, and gives lower figures.

Party was recorded only from 2024. That year the Democratic–Republican gap was 55 points; in 2025 it closed to 4 points.

In 2026 it reopened: 63% of Democrats were confident against 37% of Republicans.

Americans who approved of national leadership were the more confident group under Bush, Obama and Trump's first term; under Biden the gap widened, and since 2025 it has vanished (chart).

Gallup attributes the collapse partly to approvers changing from mostly Democrats to mostly Republicans, and partly to Democratic confidence falling and Republican confidence rising under Trump.

Across more than 100 countries, approvers of their national leadership are far more confident in election honesty than disapprovers (chart). In 2026 the US gap ran the other way, which Gallup says no other country has shown in 15 years.

The Electoral Integrity Project's expert survey found US integrity scores fell in 2024, but rated "the counting, results and procedural aspects of the election" strong; the weak areas were "electoral boundaries, media and finance". Expert scores measure the process, not public belief, and do not address whether ineligible people voted.

Republican confidence in Georgia's statewide count rebounded in 2022, the first general election under SB 202 (marked under the chart), while Democratic confidence dipped.

The authors caution that the rebound may reflect "the noted winner's effect", since the Georgia GOP won every statewide office except a Senate seat in 2022. Voting itself went smoothly: 1.1% of respondents reported a problem casting a ballot.

Most Republicans said SB 202 had increased their confidence; a plurality of Democrats said it had decreased theirs (chart). A rule passed by one party raised confidence on that party's side and lowered it on the other.

Milyo finds that adopting strict ID raised the probability of high confidence by 14 percentage points, "a 25% increase relative to the mean", using a direct question about the honesty and integrity of state elections and controlling for party.

He notes as "an important limitation" that few states adopted or removed strict ID in the period.

Stewart, Ansolabehere and Persily find that people in states with ID laws "do not have greater confidence in elections or perceive lower rates of voter impersonation fraud", and that many people do not know whether their state has such a law.

Cantoni and Pons find strict ID has "no effect on fraud – actual or perceived", alongside no effect on turnout. Milyo attributes his different result to the direct confidence measure and longer period, not to his party controls.

Clinton finds voters "less likely to express confidence in election results conducted using procedures they oppose", beyond the effects of partisan control and outcomes, with "particularly large" effects for voter ID laws and among Republicans and independents.

This fits the Georgia pattern. It suggests a rule raises confidence among its supporters and lowers it among its opponents, so its net effect on confidence depends on how widely it is supported.

The clause says the times, places and manner of congressional elections "shall be prescribed in each state by the legislature thereof; but the Congress may at any time by law make or alter such regulations".

The President's part is the ordinary one for any law: every bill is presented to him to sign or veto. The clause gives him no power to set election rules by order.

In Arizona v. Inter Tribal Council (2013) the Court quoted its 1880 ruling: Congress's power over the times, places and manner of congressional elections "is paramount, and may be exercised at any time, and to any extent which it deems expedient".

On that basis it held that the NVRA overrode Arizona's demand for documentary proof of citizenship from people using the federal form. A national ID or proof-of-citizenship law for federal elections would rest on this power.

The Court said the Elections Clause "empowers Congress to regulate how federal elections are held, but not who may vote in them". It added that Arizona could ask the Election Assistance Commission to add its requirement to the form's state-specific instructions, and go to court if refused.

Voter qualifications such as citizenship are set by the states; checking them at registration sits where the two powers meet.

Justice Stevens's lead opinion accepted Indiana's interests in deterring fraud and in public confidence, which "has independent significance, because it encourages citizen participation in the democratic process". It held those interests "both neutral and sufficiently strong" to defeat the facial challenge.

Three more justices concurred, finding the burden of a free photo ID "simply not severe". The vote was 6–3, with no opinion joined by a majority.

The lead opinion said the only fraud the law addresses is in-person impersonation, and "the record contains no evidence of any such fraud actually occurring in Indiana at any time in its history". It relied instead on documented cases elsewhere.

It also recorded the trial court's estimate that around 43,000 Indiana residents lacked a state licence or ID card. The dissent said the law "threatens to impose nontrivial burdens on the voting right of tens of thousands of the State's citizens".

Moore v. Harper (2023) held that the Elections Clause "does not insulate state legislatures from the ordinary exercise of state judicial review". "The states" in this argument means each state's whole constitutional system, including its courts and constitution, not only its legislature.

Since 2002, HAVA has required a first-time voter who registered by mail to show one of these:

• "a current and valid photo identification"
• a copy of a utility bill, bank statement, government check, paycheck or other government document with their name and address

A voter without one may cast a provisional ballot. HAVA also bars states from processing a federal registration without a driver's licence number or the last four digits of a Social Security number.

The rule accepts non-photo documents and covers only first-time voters who registered by mail; it does not ask for proof of citizenship.

The Voting Rights Act, the NVRA and HAVA each imposed rules on how states run federal elections: how people register, how rolls are cleaned and what new mail registrants must show.

They leave each state to choose what ID to ask for at the polls and whether to demand documents proving citizenship, within the limits ITCA set for the federal form.

The NVRA's federal form, written by the Election Assistance Commission, "requires only that an applicant aver, under penalty of perjury, that he is a citizen". Both the SAVE America Act and the 2025 executive order would require documents proving citizenship instead.

The bill would have required early voting, no-excuse absentee voting and same-day registration, and required states with ID laws "to accept specified information or documentation to satisfy ID requirements".

The two parties' main bills of the past five years therefore both sought national standards, in opposite directions: one mandating ID and proof of citizenship, the other limiting what ID states could demand.

Both bills in the chart failed to end debate: the Democrats' in January 2022 and S. 5271, a Republican photo-ID-only bill, in August 2026. The Freedom to Vote Act had already failed cloture once, in October 2021. Neither became law.

• 10 April 2025: the SAVE Act (proof of citizenship to register) passed 220–208, with four Democrats in favour.
• 11 February 2026: the SAVE America Act, which combined it with a photo-ID requirement, passed 218–213.
• 22 July 2026: the House passed it again, packaged with the National Defense Authorization Act to make it harder to vote down in the Senate.

After the 8 August vote, Republicans looked to the budget reconciliation process, which cannot be filibustered, as a route for the bill. A budget plan was paused when President Trump intervened. No later Senate vote was found in reporting checked on 23 October 2026.

Executive Order 14248 told the Election Assistance Commission to require "documentary proof of United States citizenship" on the national mail registration form.

The federal court in Washington, D.C. permanently barred the commission from implementing that section in October 2025, on the plaintiffs' separation-of-powers claims. The federal court in Massachusetts followed in June 2026, holding that the challenged sections "exceeded the President's constitutional authority" and conflicted with the NVRA, UOCAVA and HAVA. The government has appealed.

Executive Order 14399 directs DHS to send each state a list of adult citizens built from federal citizenship, Social Security and SAVE records.

On 24 August 2026 the Supreme Court stayed the injunction against it, calling the order "an internal directive" that "neither requires nor forbids anything of anyone outside the Executive Branch". It held the states were unlikely to have standing, since they need not use the lists.

The majority wrote that its ruling "does not mean that any measure taken by the Government to implement the Order will necessarily be lawful. On that score, time will tell." The dissent said the decision "merely postpones adjudication" of whether the President's actions are lawful.

The only merits rulings on presidential power so far are the two district-court rulings on the 2025 order.

A district judge blocked the revamped SAVE system in June 2026 under the Social Security Act, the Privacy Act and the Administrative Procedure Act. In September the Supreme Court stayed that order, saying it stopped "individualized inquiries, which are permitted under federal law" during the 90 days before an election.

The checks and flags in the table are the government's own figures, and its opponents said no error rate had been released. What those flags turned out to be is in the report on non-citizens on voter rolls.

The number of states asking for ID at the polls has nearly tripled since 2000, reaching 36 in September 2026. Thirteen states had passed proof-of-citizenship laws by July 2026; not all are in effect because of court rulings.

The Supreme Court let New Hampshire enforce a new law dropping student IDs on 22 October 2026, saying lower federal courts "should ordinarily not alter the election rules on the eve of an election". The state-by-state detail and its trend are in the report on voter ID rules.