An analysis of statewide election recounts, 2000-2023
Executive summary
Conducting election recounts to ensure fair, accurate, and genuine democratic outcomes is a critical component of effective election administration. Trust in elections requires trust in a jurisdiction’s recount process. It is an ongoing learning experience for election administrators to conduct efficient and reputable recounts, as well as to determine when victory margins and data from post-election audits should trigger a recount. In this report, we quantify various aspects of statewide recounts in the United States between 2000 and 2023, including how often they occur, how often they change outcomes, and how much vote totals change. We also use this analysis to make recommendations for policymakers on recount laws. Our main findings are:
- Statewide recounts are rare: Out of the 6,929 statewide general elections between 2000 and 2023, there were 36 statewide recounts. In other words, there was one recount for every 192 statewide elections.
- Outcome reversals are even rarer: Recounts resulted in only three reversals, or one out of every 2,310 statewide elections. All three reversals occurred when the initial margin was less than 0.06% of all votes cast for the top two candidates. The last statewide recount reversal was the 2008 U.S. Senate race in Minnesota.
- Recounts tend to shift only a small number of votes: Statewide recounts resulted in an average margin shift of 551 votes between the frontrunners, representing 0.03% of the vote. Recounts typically widen the gap between the top two candidates instead of decreasing it.
- States should require automatic recounts in close races: When the margin is very close, a recount should occur automatically without a candidate or voters having to petition for one. A threshold of 0.1% to trigger an automatic recount is sufficient to capture all races that are close enough for an outcome reversal to be plausible.
- A threshold for campaign-funded recounts should be established to prevent frivolous recounts: The recent trend of recounts for races with no realistic possibility of an outcome reversal reveals a flaw in the recount statutes in many states. Most states that allow campaign-requested recounts have no upper limit on when these recounts may be requested. Allowing campaign-requested recounts only for close races would prevent frivolous recounts and prevent a decline in voter confidence in our elections.
- Recount laws should go hand-in-hand with rigorous post-election audit procedures: Post-election audits should be tailored to the margins in each race, and the number of audited ballots should increase in relation to the percentage of discrepancies found as the audit progresses.
Recounts in statewide elections, 2000-2023
Between 2000 and 2023, there were 6,929 statewide elections, including for president, U.S. senator, governor, secretary of state, ballot measures, and other offices.
Out of the 6,929 elections, there were a total of 36 recounts, 23 automatic and 13 requested, meaning there were fewer than two recounts per year. A breakdown of the recounts by election type can be seen in the figure below.
2004 had the most recounts (six) of any year in this time period, with recounts occurring in 0.56% of statewide elections. In a close second, 2000 had five recounts, making up 0.94% of its statewide elections, including the closely watched presidential recount in Florida. Recounts were most common for ballot measures and state judicial contests (eight each).
Elections with outcome reversals
Only three out of the 36 recounts from 2000 to 2023 resulted in changed outcomes. These elections all had exceptionally close margins in the initial count.
The Washington gubernatorial election in 2004 was initially declared for Republican Dino Rossi, who was ahead by 261 votes statewide, or just 0.01%. After the recount, Democrat Christine Gregoire won by 129 votes out of 2,746,593 votes cast, or 0.005%. The recount shifted the margin by 390 votes, a change of 0.01 percentage points.
Vermont’s state auditor race in 2006 initially appeared to re-elect Republican incumbent Randy Brook by a margin of 137 votes, a lead of 0.06%. After the recount, Democratic challenger Thomas Salmon won the election by a margin of 102 votes, or 0.05%. The recount changed the margin in Salmon’s favor by 239 votes, a change of 0.11 percentage points. Most of the changes occurred in localities that had inaccurately recorded ballot counts by hand on election night.
Minnesota’s U.S. Senate election in 2008 also included a high-profile recount that reversed the original outcome. Minnesota law provides that original vote tallies within 0.5% automatically trigger a recount. Democratic challenger Al Franken entered the state-mandated recount trailing by only 215 votes, or 0.01%, against Republican incumbent Norm Coleman. After a months-long legal process that hinged largely on questions of voter intent in filling out paper ballots, Franken finished with 225 more votes than Coleman. The recount delivered a 440-vote swing to Franken, just 0.02% of the total votes between the two candidates.
Close margins are needed for reversals to be plausible
In all three reversals, the initial margin between the top two candidates was 0.06% or less. In comparison, the average margin for recounts that did not result in an outcome reversal was 0.21%. The chart below plots the initial margin for each recount from 2000 to 2023; the three red bars represent the three outcome reversals.
The original victory margin in many races with a recount was not within the narrow margin that makes an outcome reversal plausible. While all three reversals occurred when the initial margin was less than 0.06%, more than half of the 36 recounts fell outside FairVote’s recommended threshold of a 0.1% initial margin.
Partial recounts tend not to change the initial outcome
In some cases, election administrators have performed partial recounts, and these recounts are even less likely to reverse the outcome than statewide recounts. Partial recounts are not included in our totals, but are addressed in this section. Partial recounts are typically ordered in certain counties, but not the whole state. Several examples of this method include the 1988 U.S. Senate race in Florida, the 1995 Maine referendum on seat belts, the 1998 attorney general race in New York, the 1998 U.S. Senate race in Nevada, the 2002 gubernatorial race in Alabama, and the presidential race in Arizona in 2020.
Partial statewide recounts also occurred in Ohio during the 2004 presidential race and in a U.S. Senate race in Indiana in 2006. These recounts were both requested by petitioners seeking to draw attention to problems in how ballots were counted, not to overturn outcomes.
Recounts leads to negligible vote gains for both sides
Recounts typically lead to a slightly higher vote total, rather than a lower total. Much of this hinges on determinations of voter intent, often on ballots that were unreadable by tabulation machines.
Of all the candidates and ballot measure positions considered in the 36 recounts, the vote totals of 78% of candidates and positions increased during recount. In two-thirds (65%) of recounts, the number of votes increased for both sides. On average, winning candidates increased their vote totals by 0.08 percentage points and runners-up increased their vote totals by 0.07 percentage points. This almost negligible change in votes reinforces the argument that extremely close margins are a necessary condition for a recount to possibly overturn an outcome.
Overview of existing state recount laws
Twenty-three states and the District of Columbia have laws in place that require automatic recounts for state and federal elections when certain conditions are met. The requirements to trigger an automatic recount vary by state.
Twenty-seven states do not have automatic recount laws, but 26 of those have laws permitting a request for a recount. Only Mississippi does not have any recount provisions in place in the event of a close race.
Numerous states empower voters and political parties to request recounts. Some states only allow candidates to petition if the results are within a particular margin, while some states charge candidates a fee to petition. The fees are typically returned to the petitioner only if the recount changes the outcome of the election in the petitioner’s favor.
Since 2020, six states have changed their recount laws: Arizona, Hawaii, Michigan, New York, Oklahoma, and Utah. Most notably, Hawaii added an automatic recount provision for elections within a margin of 0.25%. Arizona increased its threshold for an automatic state-funded recount from 0.1% to 0.5%. Arizona conducted recounts in two statewide races in 2022; neither would have occurred automatically had lawmakers not raised the threshold. Neither recount led to a meaningful change in votes or an outcome reversal. Also of note, Michigan added new provisions in 2024 that prohibit recounts based solely on allegations of fraud, and place a larger cost burden on the recount petitioner for recounts in races with very large margins. That law will take effect in 2025.
Allowing campaign-funded recounts with no threshold can lead to frivolous recounts
There have been a large number of campaign-funded recounts in recent years, many of which do not meet our criteria for full statewide recounts and are not included in our analysis. However, this section will briefly consider the recent trend of recounts that are funded by campaigns or private organizations and where the margins fall outside of the range where they could be considered consequential.
On the state level, there have been several noteworthy campaign-funded recounts for which an outcome reversal was not plausible. These recounts produced negligible results. In Nevada’s 2022 Republican gubernatorial primary, Joey Gilbert’s campaign paid $190,000 to recount votes in all 17 of the state’s counties after the original tally showed he trailed his opponent Joe Lombardo by 11 percentage points. Tina Peters, a Republican candidate for secretary of state in Colorado, paid $256,000 for a statewide recount although she trailed by 14 percentage points. This landed her only 13 additional votes. In Kansas, anti-abortion activists spent $120,000 for a frivolous recount of the highly contested No State Constitutional Right to Abortion ballot measure that lost 59-41%.
Partial recounts in the 2020 presidential election occurred in several states where the margins of victory were outside the range where an outcome reversal was plausible. In 2020, five pro-Donald Trump organizations gave $5.6 million to fund a recount in Maricopa County, the most populous county in Arizona. The original tally showed Joe Biden winning Arizona by more than 10,400 votes, or 0.3%, which is larger than the margin that would have triggered an automatic recount under the law at the time. The recount yielded 99 additional votes for Biden and 261 fewer votes for Trump. Similarly, in Wisconsin, the Trump campaign paid approximately $3 million to fund a partial recount in Milwaukee County and Dane County after Biden held a 22,500 vote lead (or 0.6%) over Trump. This recount resulted in a net vote gain for Biden.
Conducting these recounts when there were not close victory margins in either state not only produced arbitrary vote gains, but also undermined voter faith in election integrity. The lack of confidence in the country’s ability to conduct an honest election puts the overall democratic process into jeopardy. While recounts are crucial in very close races, frivolous recounts in races that are not close do more harm than good.
Counting procedures for recounts
Just as recounts can be triggered differently in different states, recounts are often conducted differently due to different election equipment and recount requirements. Moreover, the circumstances leading to a recount can affect what kind of recount is done. In the 2004 U.S. Senate race in Alaska, for example, all ballots were re-scanned and there was a manual hand count of a sample of ballots in order to evaluate concerns that the machines were not tallying all ballots accurately. In Minnesota’s U.S. Senate race in 2008, a statewide manual hand count was conducted because Minnesota law seeks to verify voter intent. Full manual hand recounts decided the 2006 state auditor race in Vermont, the 2004 constitutional amendment referendum in Alabama, and the 2004 gubernatorial race in Washington. All recounts prior to the introduction of voting machines involved manual hand counting, while the 2000 State Education Board election in Colorado was an example of an automatic machine recount.
The margin shifts tend to reflect the process by which the ballots were recounted, although not dramatically so. Manual hand recounts are more costly and time-intensive than machine counts and require careful procedures to minimize human error, but such manual recounts also have resulted in larger margin swings, presumably because humans evaluated voter intent differently (and ideally more accurately) than the machines in the original count.
Implications for policymakers
Our findings indicate that the overwhelming majority of elections have outcomes that are not realistically disputable, absent genuine indications of systematic fraud or error. In the relatively few statewide races with recounts in the last 23 years, the original outcome was rarely overturned, and initial victory margins only changed slightly. Yet recounts should be conducted in exceptionally close races even if costly to taxpayers – and procedures should be in place to identify fraud or error.
Given data on margin shifts in statewide recounts with modern voting machines, we recommend a 0.1% initial margin as a threshold for automatic recounts. Our research indicates that only elections within that margin are close enough that a recount could plausibly alter the outcome, although reversals are rare even within that margin.
The 27 states without any automatic recount provisions should establish them. At the same time, the states that conduct automatic recounts based on higher thresholds could safely lower their thresholds while still ensuring secure elections. This would reduce the number of inconsequential recounts and save taxpayer money.
Automatic recounts within the 0.1% margin should be state-funded, because it is a key function of state governments to ensure accurate election outcomes. Requested recounts outside of that margin are not considered plausible and should be funded by the petitioner to avoid extra taxpayer burden.
Additionally, we recommend that recount statutes be updated to place an upper limit on which races permit a campaign-requested recount. We recommend allowing candidate-requested recounts in races with a margin of up to 5%, and ensuring that those requested recounts are funded by the individual or entity that requests them. In races with larger margins of victory, there is no realistic chance that a recount will change the outcome, absent highly credible evidence of error or fraud, meaning the recount serves no genuine purpose except to raise doubt about the election.
Considering the possibility of recounts in national popular vote elections
Since 2006, every state legislature in the nation has debated legislation to enact the National Popular Vote plan for president, which comes in the form of identical statutes entering participating states into an interstate compact to guarantee election of the presidential candidate who earns the most votes across all 50 states and the District of Columbia. Seventeen states and the District of Columbia have signed the National Popular Vote plan into law, representing 209 electoral votes. The agreement would only become active once the participating states collectively have more than half of the nation’s electoral votes, or 270 votes.
The National Popular Vote plan could trigger the need for a national recount, but the odds are lower than in our current method of electing the president. Currently, any one of 50 states could have a close race where an outcome reversal could swing the national outcome.
Consider that only one out of every 192 statewide general elections over the past two decades triggered any kind of recount. Less than half (15 out of 36) were within the margin where we consider a recount necessary and consequential, for one necessary recount for every 460 elections. Only one out of every 2,310 statewide general elections led to the outcome being reversed by a recount.
Applied to our four-year presidential election cycle, a necessary nationwide recount might occur once every 1,840 years, with an overturned election once every 9,200 years. In contrast, in our current system based on 51 separate elections individually determining allocation of electoral votes, any closely contested presidential election where the outcome could hinge on one state’s outcome is far more likely to trigger a necessary recount even with such seemingly long odds – perhaps once every nine presidential elections (because nine presidential elections represents 459 separate statewide races, and only in very close races would the Electoral College result potentially hinge on the state being recounted).
In examining what might be a necessary recount in a national election, consider that the median change in victory margin in consequential recounts was 0.02%. In a race with 130 million votes, it seemingly would take a margin of under 26,000 votes to trigger the need for a recount where a change in outcome is plausible, unless there was clear evidence of fraud or error affecting an inordinate number of votes.
Former University of Pennsylvania Professor Jack Nagel’s independent analysis of the same questions also concluded that the odds of a recount are significantly less for a single nationwide vote pool than for the current Electoral College system in which each state’s votes are counted separately. He writes: “Defenders of the Electoral College often attempt to turn the Florida 2000 fiasco into a reason for rejecting the direct vote alternative. They ignore the obvious answer: The national vote in 2000 was not close enough to dispute, nor has the popular vote been that close in any recent election. Using any reasonable assumption about how close an election must be for recount demands to arise, the likelihood of disputes is greater under the current Electoral College system than it would be in an election decided by the national popular vote.”
Methodology
Definition of “recount”
When performing a recount of election results, ballots are retabulated to confirm the original tally identified the correct winner. Depending on the state, a recount could be performed with different methods, such as electronic review of voting machines or physically counting ballots by hand. Recounts often occur because of a close victory margin, concerns of election fraud, or worry over administrative error when initially processing ballots. For our study, we define an election recount as the process of tabulating the ballots in each precinct in a statewide election a second time.
Definition of “statewide election”
To narrow down the number of elections and recounts between 2000 and 2023, we examine only statewide general elections. These elections provide an opportunity for all citizens, no matter where they live in the state, to vote for the same candidates or on the same ballot question. A majority of these elections occur in November, while others can take place earlier in the year, such as a special general election or when a ballot question is up for a vote during primary elections. We do not include primary elections because those precede general elections, winners need to win a general election to take office, and primary elections often do not permit every voter to participate. We also do not include most elections for the U.S. House of Representatives because candidates run in districts, and only voters who live in those districts can vote for the respective candidates. However, we include House elections in the states with only a single House district, meaning all voters in the state may participate.
Data collection
We gathered data on statewide general elections between 2000 and 2023 by visiting secretary of state and Election Board websites. Contact was also made by phone or email with 46 out of 50 state offices. For states with unresponsive offices, data was collected through Google and Lexis-Nexis searches using the term “recount.” Additionally, Ballotpedia and reports from local media channels were utilized for any data collection not available on state websites.
An election where more than one seat was open was counted as one statewide election because only one recount was needed to confirm the results.
For the elections that had recounts, we collected vote totals for only the top two candidates. Therefore, the vote totals included in the report are a combination of votes for only the top two finishers in a given race.
Data on recount laws was also gathered for all 50 states.
Definition of “victory margin”
The victory margin between the top two candidates was recorded, as well as the size and direction of any shifts in votes. The victory margin was determined by taking the vote difference between the top two candidates and dividing that by the total number of votes cast for those two candidates.
Explore the data
The full dataset, including a list of all recounts and a tally of all statewide elections, is available here.
Data sources and acknowledgements
The data used in this report was obtained from election results archived on secretary of state and Election Board websites, by calling or emailing their offices, and from the Westlaw and Lexis-Nexis news databases. Recent updates were done with the help of Ballotpedia, the National Conference of State Legislatures, the State Recount Laws Searchable Database by Citizens for Election Integrity Minnesota (CEIMN), and Thomson Reuters Westlaw.
Appendix: List of statewide recounts, 2000-2023
About the authors
Deb Otis is the Director of Research and Policy at FairVote. With a decade of experience in research and analytics, Deb is passionate about sharing the data-driven case for why our country needs election reform. Deb’s areas of research include election recounts, ranked choice voting, proportional representation, comparative electoral systems, representation for women and people of color, and the electoral college.
Bryan Huang is the Software Engineer and Research Analyst at FairVote, where he manages and develops software tools to analyze how ranked choice voting (RCV) and proportional RCV ensure elections have representative outcomes. Leveraging his expertise in data science and quantitative research, he is committed to using data-driven research to support voting rights and electoral reform.
