*Note that this methodology includes some sections and footnotes from the above piece for clarity and readability.
As our analysis focuses on four of the five most recent general elections (2016, 2018, 2020, and 2022), we relied on the records of cases from the Heritage Foundation’s database, which “presents a sampling of recent proven instances of election fraud.” Though the “Explore the Data” page indicates categorizations for “Case Type” and “Fraud Category,” we were not able to locate any publicly available information on the methodology used to collect this sampling of cases. The Heritage Foundation’s website states: “To be clear, this database is not an exhaustive or comprehensive list of all election fraud in the states. It does not capture all cases and certainly does not capture reported instances or allegations of election fraud, some of which may be meritorious, some not, that are not investigated or prosecuted.” The database contains 1,600 cases with years listed from 1982 to 2025.
Each case entry we included involved one individual who had been charged with a type of election fraud. Though there were cases listed in 2024 and 2025 in the Heritage Foundation’s database, none of those cases involved fraud that took place during the 2024 election year. We thus focus our analysis on the four prior general elections: 2016, 2018, 2020, and 2022.
For the purpose of our analysis, voting fraud includes but is not limited to altering the vote count, fraudulent use of absentee ballots, buying votes, or ineligible voting, but excludes false registration and ballot petition fraud. To err on the side of inclusion, we define mail voting fraud as any instance of voting fraud involving the use of a mail ballot.
We began our analysis by determining which cases in the database included voting fraud that took place in relation to one or more of the four general election years we examined, as well as which ones could have possibly involved mail voting fraud under our definition. To err on the side of inclusion, we included cases in our database that ended in nolle prosequi notices, plea deals, pre-trial diversion programs, or other alternative sentencing structures. We included cases in which some, but not all, charges were dropped. We also included any cases that took place within 60 days of each year’s general election to provide a maximum possible estimate. While there is variance in state statutes governing the number of days before a general election that mail ballots can be distributed, the earliest specified date is 60 days prior (in Delaware and North Carolina), which serves as a reasonable upper bound for inclusion.
We excluded cases from our database in which the year of offense could not be confirmed, as well as those involving fraudulent ballot petitions. We also excluded any cases that took place over 60 days prior to the election. Any case with a charge of registration fraud was excluded, given that mail voting does not make registration fraud easier or harder to commit. Finally, we excluded any cases in which all legal charges were dropped or defendants were acquitted.
In making the above classifications, we do not intend to assert nor imply any individual determinations of guilt or liability.
For each included case, we independently verified key facts using court filings, administrative records, and news sources. After classification, we used our source files to determine whether each case involved use of mail ballots. Cases with evidence of mail voting were marked as such. Cases with evidence of in-person voting were classified as “no explicit evidence of mail voting.” Cases lacking clear evidence of voting method were also marked “no explicit evidence of mail voting” and excluded in our final count to avoid false positives.
We then reviewed state mail voting laws for each general election year using data from the United States Election Assistance Commission’s Election Administration and Voting Survey (EAVS) to identify which states had excuse absentee, no-excuse absentee, and universal vote-by-mail voting systems in each year.
Some states in each general election year had exceptions to their mail voting policies. For example, in 2016, Minnesota had a no-excuse absentee voting system, with an exception for some non-metropolitan counties or cities with fewer than 400 registered voters. In those locales, there was often the option of using an all-mail voting system. We were unable to confirm whether the cases we analyzed fell under such exceptions given statutory differences across states and years. However, based on exploratory analysis we conducted across a subset of several, randomly selected states, it is likely that very few—if any—of the cases we analyzed fall under these exceptions. We examined all cases of mail voting fraud in Arizona, Idaho, Indiana, Maine, Minnesota, and New Jersey that took place during the 2016 general election and confirmed that no cases fell under the relevant statutory exceptions.
For the cases we have determined involved mail voting fraud, we cross-referenced the state in which the case took place to determine what type of ballot was used: an excuse absentee ballot, a no-excuse absentee ballot, or a universal vote-by mail ballot. By default, we kept the state listed in the Heritage Foundation database, as that was the jurisdiction in which the individual was charged. For duplicate voting cases in which an individual voted by mail in one state and in person in another, we assigned the state to that in which mail voting took place, to err on the side of inclusion. When duplicate voting involved mail ballots in two different states, we assigned the state that was not the individual’s primary residence, as that is the state in which they cast a duplicate, fraudulent vote. In states with universal vote-by-mail, if an individual voted on behalf of two people and was not listed in the Heritage Foundation’s database as committing impersonation fraud, we again erred on the side of inclusion and assumed use of a mail ballot.
Using EAVS data, we calculated the total number of mail ballots transmitted to voters in each general election by summing the number of ballots transmitted to domestic absentee voters and the number of ballots transmitted to voters under the Uniformed And Overseas Citizens Absentee Voting Act (UOCAVA). We also used EAVS data to calculate the total number of mail ballots cast nationally and in each state for each general election year. Because no cases of fraud in the Heritage Foundation’s database occurred in American Samoa, Guam, the Northern Mariana Islands, Puerto Rico, or the U.S. Virgin Islands, we excluded these jurisdictions and focused our analysis on the 50 states and the District of Columbia.
Using our compiled dataset, we calculated the percentage of mail voting fraud per total mail ballots cast for each general election by dividing the total number of mail voting fraud cases by the total number of mail votes cast and multiplying by 100. We then took the average percentage across all four general elections.
For our counterfactual analysis, we used the News21 database of election fraud cases. According to News21, “a national investigative reporting project” their team sent out over “2,000 public-records requests and spent nearly $1,800 on fees for records searches and copies of documents. The team also reviewed nearly 5,000 court documents, official records and media reports. The result is the most extensive collection of U.S. election fraud cases ever compiled.” The database contains over 2,000 cases and lays out its methodology and limitations. We scraped the data from the News21 website and, consistent with our own analysis, filtered out cases related to registration fraud, petition fraud, campaign fraud, procedural error, and intimidation, leaving the 1,605 cases dealing with voting fraud specifically. To err on the side of inclusion, we retained cases that had not yet reached a conviction. As with our primary dataset, we do not intend to assert nor imply any individual determinations of guilt or liability in making these determinations.
We divided these 1,605 voting fraud cases by the 12 years covered by the database to estimate the average number of voting fraud cases per year, yielding an average of 134 cases. We then divided 134 by the total number of mail votes cast for each general election and multiplied by 100 to calculate the counterfactual percentage of mail voting fraud for each general election. The 134 cases on average per year do not in reality each represent a case of mail voting fraud. Finally, we averaged these percentages across all four general elections.
Next, again using our modified Heritage Foundation database, we calculated the share of total mail voting fraud percentage represented by each of the three mail ballot types (excuse absentee, no-excuse absentee, and universal vote-by mail).
We calculated the total number of mail ballots cast nationally for each general election year by aggregating state-level data. Using our prior classification of state mail voting systems, we sorted the states into three categories for each general election: those that used excuse absentee voting; those that used no-excuse absentee voting; and those that used universal vote-by-mail. For each of these three categories, we summed the total votes cast in relevant states to determine how many votes of that type were cast across the nation for each general election year from 2016 to 2022. Using the cases we determined to be related to mail voting fraud, we followed a similar method to calculate the total number of mail voting fraud cases of each type for the same time span.
Finally, we calculated a normalized rate of mail voting fraud by dividing the total number of mail voting fraud cases per type by the total number of mail ballots cast of that type in each general election (e.g. the number of excuse absentee fraud cases in 2016 divided by total mail ballots across states that used an excuse absentee voting system in 2016). We normalized to account for the fact that the number of mail ballots cast in each state and number of states of each type varied widely. We then summed the normalized rates of mail voting fraud across the three systems—excuse absentee, no-excuse absentee, and universal vote-by-mail—and divided each system’s normalized rate by this combined total, multiplying by 100 to express the result as a percentage. This final calculation allowed us to identify the percentage of the overall (and already exceedingly small) mail voting fraud percentage attributable to each mail voting type.
For further questions or additional information on our methodology, please contact [email protected].