Postelection Audits and FAQs

What are Postelection Audits?

Postelection auditing ensures the election outcome is reliable and valid. There are several types of postelection audits including traditional, risk-limiting, and transitive audit.

Traditional audits use a predetermined fixed percentage or number of ballots or ballot types (mail, early, and Election Day), precincts or tabulators where ballots are hand counted and compared to the reported results from those system. In some states, various size discrepancies between the hand count and machine count can lead to additional ballots being selected for auditing, but this is not very common. In addition, requirements can vary as to whether all or only some types of ballots are included in the audit (e.g. provisional, uniform and overseas voters, cured absentee). These audits do not examine the entire ballot, but tend to look at specific contests. Most of these audit procedures are set forth in state statues and define who, how and when the audit is conducted. 

A risk-limiting audit (RLA) is a post-election tabulation auditing procedure that provides statistical assurance that voting hardware and software produced the correct winners. RLAs involve reviewing portions of the audit trail—voter-verified paper records including paper ballots themselves, paper audit trails produced by direct-recording electronic voting machines, or paper ballot records produced by ballot-marking devices —to identify and correct any erroneous electoral outcomes. RLAs first require a ballot counting audit to ensure that the correct number of ballots have been cast based upon the records in the audit trail. Second, because recounting all cast ballots is time-consuming and costly, RLAs assess only a random sample of ballots. The number of recounted or reviewed ballots increases incrementally until a predetermined “risk limit”—the highest probability that the audit will not correct an incorrect outcome—is met or election administrators conduct a full recount.

For example, at a 5% risk limit, there is at most a 5% chance that the audit will not identify an incorrect outcome. The number of ballots that must be hand-counted is a function of risk limit, vote margin, and (weakly) contest size. Lower risk limits entail examining more ballots, as do closer elections. RLAs escalate the number of ballots examined if the risk limit is not met; theoretically this could include a full hand count if necessary.

Transitive audits are re-tabulation audits that digitally rescan ballots on a different voting system or tabulator and the results of the two systems are compared. A transitive election audit, also known as a double-count audit, runs all the ballots through two independent tabulating systems. The ballots are first tabulated on the official tabulator used to record final votes and then run through a second tabulator from a different manufacturer with different hardware and software to confirm the vote totals. If both systems report the same election outcomes it provides evidence that the outcomes are correct, even if it finds some variation across systems. In this case, the post-election audit relies on the independence of the tabulation system’s software and hardware to demonstrate the veracity of the outcomes, not human confirmation.

Compared to other types of post-election audits, transitive audits of this type have the advantage of recounting all the ballots and 100% of the contests.

Frequently Asked Questions

The purpose of postelection auditing is to promote confidence in the counting process by demonstrating that the tabulating machines count votes correctly. Leon County pioneered the transitive audit, which is a 100% retabualtion of all the ballots.

Compared to other types of audits, traditional and risk-limiting, transitive audits have the advantage of counting all the ballots, instead of a much smaller portion, and in every contest, as opposed to just particular contests.

Election auditing is a critical component to election security and integrity. 45 states require some sort of post-election audit and two additional states have post-election audit pilot projects to ensure the accuracy of tabulators and the election outcomes.

Florida election officials are required by law to do a post-election audit (s. 101.591, Florida Statutes).  There are two types of post-election audits that are allowed:

  1. A public manual audit that examines one randomly selected ballot contest for 1%-2% of randomly selected precincts in each jursidiction, or
  2. An automated audit that consists of a tally of all the votes cast in each contest.  This audit includes 20% of precincts in a jurisdiction, chosen randomly.

Election Supervisors in several Florida counties have been experimenting with 100% retabulation audits that use a second tabulator from a different vendor to recount all the ballots. The results from the retabulation audit are then compared to the first tabulation to determine if there is any variation or difference between the two.

When variations in counts could result in a change to the election outcome, differences are adjudicated to determine the correct winner. If the variation in recorded votes does not change the election outcome, then no further action is taken and the 100% retabulation audit results are reported and the official election count remains the same.

The point of a transitive or 100% retabulation audit is to confirm the votes were counted correctly by the first tabulator by counting all the ballots again on a second tabulator. In Leon County, the official vote count comes from two Dominion tabulation systems, one for in-person voting and one for mail ballots. The post-election audit tally comes from the ClearAudit tabulator.

ClearAudit is a browser based central count tabulation system. ClearAudit can tabulate ballots created by all major voting systems certified in the state of Florida offering a truly independent retabulation. By independent we mean that there is no hardware or software in-common between the first machines used to count at the precinct, early vote center, or election center, in the case of the vote by mail, and the second count done on the ClearAudit tabulator at the election center.

The Dominion and ClearAudit tabulators use different methods to identify votes. Dominion software searches for a vote by starting at the center of an oval and moving outward to its ends. ClearAudit software uses a larger zone encompassing each contest to search for and identify a vote for each candidate. By looking at a larger area around the oval, the ClearAudit system can identify stray marks and marks outside of the oval that could be a vote. For example, if someone circles the oval instead of filling it in, the ClearAudit system will see those marks and can better identify whether those are likely votes or not. ClearAudit software also includes a vote visualization tool that potentially allows auditors to find uncounted or incorrectly counted ballots.

Florida’s primary elections are “closed” meaning that only registered members of a political party are able to cast ballots for that party’s nominees. In Florida primary elections there are separate ballots given to members of each political party. In other words, the ballot that a registered Democrat receives will include primary contests between Democratic candidates, but would not include primary contests between Republican candidates. This way, only Democrats are voting for which nominee they wish to represent their party in the general election, and respectively, only Republicans are voting for their party’s nominee.

In Tallahassee in the primary election, those voters with No Party Affiliation (NPA) or a non major party affiliation are not allowed to vote in either party’s primary elections but they are allowed to vote in the nonpartisan races or contests in the city and the county. For example, all of the city contests as well as the county commission and county judge races are nonpartisan.

In the primary dashboard the Party of Contest button is in the top middle and refers to the partisan ballot that was issued to the voter and has four possible selections: 1) All, 2) Republican (REP), 3) Democrat (DEM), 4) Nonpartisan Race (NPR). If you select the filter for “Democrat (DEM)” both dashboard data tables will filter Democratic ballot contests.  Similarly, if you filter “Republican (REP)” both dashboard data tables will filter Republican ballot contests and if you filter “Nonpartisan race” both dashboard data tables will filter nonpartisan contests in the city and county.  Of course, both Republican and Democratic ballots also include the nonpartisan contests. Therefore, if you select the filter for “Democrat Party” you will see ballots completed by registered Democrats, which will include partisan primary contests, as well as non-partisan contests where candidates have no declared political affiliation (such as those running for city commission).

In the August 2022 Florida primary election, some contests had no Republican primary at all. For example, while multiple Democratic candidates ran against each other to become the Democratic nominee for Attorney General, Republican incumbent Ashley Moody was uncontested for her party’s nomination, therefore she does not appear on the ballot. The same is true for the U.S. Senate election, where incumbent Republican Marco Rubio was unchallenged, but multiple Democrats ran against one another to secure their party’s nomination. In Florida, when a race is uncontested it does not appear on the ballot.

The contest for Florida Commissioner of Agriculture was the only statewide race in the August 2022 primary election where both Republican and Democratic voters cast ballots for nominees.

An undervote is when a voter does not cast a vote for any candidate in a given contest. In other words, if voters leave a particular contest blank and do not “bubble in” or “mark” any candidate’s name, they have “undervoted” in that contest.

Undervoting can be purposeful or accidental. A voter may intentionally avoid voting in a particular contest or may miss it. Undervotes are more common in “down-ballot” contests, such as retention votes for judges, and are less common for statewide races, such as for the Florida Governor.

An overvote is when a person casts votes for more than one candidate in a given contest. In other words, if a person “bubbles in” more than one person for the same contest, they have overvoted. Overvotes are not counted. In other words, if a person voted for competitors Charlie Crist and Nikki Fried for Florida Governor, neither candidate would receive a vote for that ballot. While overvoting on a particular contest does not count for that contest, valid votes cast for all other contests on that ballot will continue to be counted.

Bubbled in ovals or choices are rated from low to high confidence based upon density marks. Theoretically, the lightest marks within an oval are the least confident and may need further examination or adjudication, especially if the election is close. Ovals completely filled in are likely to score high in confidence and have a low chance of a vote count error, while an oval with only a dot or tiny ink mark or marks outside of the oval are likely to score lower in confidence.

In the dashboard, ballots can be filtered by two oval confidence categories, either 1-20 or all the ballots. A low confidence oval has a low ballot confidence score; the lowest score is 1. Using the 1-20 filter you can look at the 20 ovals with the lowest confidence for a candidate or contest. Once you get past the first few low confidence ovals it is often hard to see any ballot differences with the human eye. This is why we only provide the filter for the lowest 20 ovals. However, you can click the column heading tab for oval confidence and see confidence rankings larger than 20.

To view ballots cast within a specific precinct, you can either click on that area on the map in the dashboard (which filters the data presented), or you can select the precinct(s) from the drop-down menu above the map.

State law requires that, “When one or more ballot types, also known as counting groups, in a race or an issue have fewer than 30 voters voting on the ballot, the ballot type must be reported as zero except for the total votes counting group for that precinct.” This is necessary to ensure voter privacy. Precincts with a small number of voters or precincts with a small number of voters within vote mode types risk the possible secrecy of a voter’s ballot.

In the primary, we found that when there were a very small number of nonpartisan and Republican voters participating, there was a high risk of linking ballots to voters. This could be done by triangulating aggregate vote information from the county and state websites and individual level data from the voter file. Therefore, for the primary we had to take additional precautions to ensure that voter ballot choices could not be determined from the records we present, including removing the precinct information from all the ballots and removing the tick marks that indicate the ballot style (tick marks identify the precinct, party and language of the ballot).

There are 43 ballots that we excluded from the dashboard because the unique combination of contests on the ballots, together with the voter files, would reveal individual vote choices and violate voter privacy.

First, confidence data was truncated to only include confidence levels for most of the ovals because the supervisor’s office was unable to capture it all. ClearBallot did not produce the confidence rankings for the 90 most confident ovals for the "Vote for Candidate" category for the Democratic Party gubernatorial primary candidate Nicole "Nikki" Fried, the 2,536 most confident ovals for the "Vote for Candidate" category for the general election candidate for County Commissioner (District 2) Hannah Crow, and the most confident ovals for the "Vote for Other Candidate" for all candidates across different contests.

Second, ClearBallot could not determine the confidence levels for a small number of ovals across different categories and candidates.

These ballots are nonetheless available to view on our dashboard.

Sometimes mail ballots arrive damaged in such a way that they are not readable by the tabulator. This happens because sometimes a ballot is torn or otherwise damaged. For example, sometimes folding the ballot causes the tick marks to be damaged making a mail ballot unreadable. Sometimes the ballot will have coffee or other drinks or food spilled on it that can prevent it from being read. Sometimes the humidity can create problems and ballots stick together. In these cases, county staff duplicate the ballot. This process is completed by one staff person and audited by another. The original ballot is also tagged and is kept for review if there were any litigation around the contest.

A vote center is a voting location at which any registered voter in the county can vote. During early voting Leon County has many vote centers conveniently located throughout the county. A voting precinct is one of several areas or districts into which a county is split for voting purposes. On election day voters are required to vote at their precinct polling location unless they vote at the election center.

We included Dropbox links to folders of ballot images we thought voters might find interesting. The topics include:

  • Primary Election ballots with signatures/General Election ballots with signatures – to protect voter privacy, we searched through ballots looking for initials or signatures that could possibly identify a voter. In a few cases, we found voters names or signatures. We replaced the signature with a John Hancock to show where on the ballot and what quantity of space the signature took up.
  • Primary ballots with no votes – some voters go to the polls and vote for no candidates. In the primary, we found 48 ballots with no votes. In many cases, voters wrote interesting marks on their ballots, such as, “no thank you”
  • Primary Election/General Election ballots with stray marks or other interesting features – these ballots include written comments, drawings, and stray marks.
  • Primary overvoted governor ballots – these are ballots where voters marked two candidates for governor. Looking at the ballots it seems clear that a few voters were confused by the instructions, “Governor and Lieutenant Governor Vote for One”

To calculate the difference in vote counts between the Dominion tabulator and the ClearBallot tabulator, we took the sum of the number of times each contest appeared on a ballot. We then divided that number by the sum of the absolute differences in votes counted for each candidate between the two systems. Finally, we subtracted that number from one to get the match rate.

This project was funded by the MIT Data Election Science Lab and the Florida State University College of Law Election Law Program.

For full experience visit from a desktop computer