2020 General Election: Post-Election Manual Ballot Tabulation Audit
The post-election manual ballot tabulation audit requires an audit of:
1) At least 2% of election day vote centers, including at least 1 randomly selected election day vote center in each county and the remainder of election day vote centers to be chosen by SBE; and
2) A number of votes equal to at least 1% of the statewide total in the previous comparable general election1 of ballots cast during early voting and mail-in and provisional ballots, including a minimum number in each county, as set by SBE. For the 2020 General Election, SBE set 15 votes as the minimum number of mail-in ballots and provisional ballots for each county.
The manual audit must be completed by each local board within 120 days after the general election. If the manual audit shows a discrepancy greater than 0.5%, SBE can expand the manual audit and take any other actions it considers necessary to resolve the discrepancy.1 “Previous comparable general election” is defined as (1) in a presidential election year, the presidential election held four years earlier and (2) in a gubernatorial election year, the gubernatorial election held four years earlier. For the 2020 General Election, the previous comparable general election held was the 2016 General Election.
Selecting the Contest to be Audited
At the November 29, 2018 meeting of the State Board of Elections2, prior to the inaugural manual audit of the 2018 General Election, the State Board decided upon parameters for the contest to be audited. Those parameters were:
• That the same contest would be audited by each local board of elections, and therefore, the contest would be a statewide race.
• That the contest would be a “pick one” contest.
Using these same parameters, the only eligible contest for the 2020 General Election was the contest for President and Vice President.
2 Full minutes of the November 29, 2018 State Board of Elections meeting can be found here (PDF).
Selecting the Ballots to be Audited
Before the election, SBE determined the number of ballots to be audited in each local jurisdiction for ballots cast during early voting and mail-in and provisional ballots. This chart shows the minimum number of ballots from each jurisdiction to be audited based on the requirements set forth in §11-309 of the Election Law Article.
Mail-in and Provisional Ballots
The number of mail-in and provisional ballots in the audit is based on the turnout from the 2016 General Election. The number of mail-in ballots and provisional ballots to be audited was calculated by multiplying the 2016 relevant turnout by 0.01, in order to determine 1% of the total number of ballots cast in each category. If 1% of the 2016 turnout was less than 15 ballots, the local board audited 15 ballots.
Each local board randomly selected the mail-in and provisional ballots to be audited, as they prepared for the respective canvasses. These ballots were then kept separate from the remainder of the mail-in and provisional ballots and marked as “audit ballots.”
At the start of the first mail-in and provisional canvasses, these ballots were presented, reviewed, and tabulated first, and then stored in sealable boxes marked “Mail-in Ballots for Manual Audit” or “Provisional Ballots for Manual Audit.” At the completion of each canvass, the sealed boxed containing the ballots to be audited were stored with the remainder of the mail-in or provisional ballots.
Early Voting Ballots
The formula used to determine the number of mail-in and provisional ballots to be audited was also applied to early voting. Using the early voting turnout from the 2016 General Election, 1% of that turnout was calculated to determine the number of early voting ballots for each local jurisdiction to be audited.
Jurisdictions with only one early voting center audited ballots from that early voting center. For jurisdictions with more than one early voting center, SBE used a random number generator to select the early voting center to be audited. Local boards were notified of the selected early voting centers the day prior to the first day of early voting. A list of the selected early voting centers can be found here.
By 7 pm on the first day of early voting, the chief judges at the selected early voting centers determined which scanner met the minimum number of ballots to be audited. If none of the scanning units met the minimum, two scanners were selected. At the close of the first day of early voting, and after all voters left the building, a local election official ran a results report from the selected scanner(s). The election officials secured the ballots from the selected scanner(s) and the report(s) in a sealed box.
Election Day Ballots
Section 11-309 of the Election Law Article requires that 2% of all statewide precincts must be audited, including at least one precinct from each local jurisdiction.
Because the 2020 General Election was conducted using election day vote centers instead of precinct-based polling places, the members of the State Board of Elections approved at its September 11, 2021 meeting emergency regulations that included “election day voting center” in the definition of “precinct.” This change was necessary to include in the manual audit ballots cast on election day.
For the 2020 General Election, there were 321 election day vote centers, of which 2% equals 6.42. Because auditing just six election day vote centers statewide would not meet the requirement to audit a minimum of one election day vote center in each jurisdiction, each local board audited one election day vote center. All ballots cast at the selected election day vote centers were audited.
To select the required election day vote center from each local jurisdiction, two members of the State Board and SBE staff met at the SBE office to randomly select one election day vote center in each local jurisdiction to audit. A random number generator was used in the same manner that the early voting centers were selected to select the election day vote center. Local boards were notified of the election day vote centers that day. This chart shows which precincts were selected for audit.
Preparing for the Audit
Each local board spent approximately a half day to two days preparing for the manual audit. Smaller local boards generally used all or most staff in audit preparations, while larger local boards used five to six individuals.
Preparation tasks included sorting the ballots into batches, preparing batch tally sheets for each batch, selecting audit tallying teams, setting up the audit room, and preparing audio & video equipment if the audit was livestreamed. Additionally, all local boards provided public notice of the audit, which must be mailed or emailed to candidates for the contest to be audited, as well as posted on the local board’s website and in the front of the office ten days prior to the audit.
COVID-19 precautions were taken during each step of audit preparations, including the wearing of facemasks and social distancing for staff, and setting up hand sanitizer stations and spacing the audit tallying teams so that tables were six feet apart during the audit. Most local boards reported that adding COVID-19 precautions added time during audit preparations and the audit, but the time added was not substantial.
Conducting the Audit
Each local board conducted a manual audit of the required ballots within 120 days of the 2020 General Election, starting with Anne Arundel County on January 15, 2020, and concluding with Washington County on February 26, 2021.
Time to complete the audit: A survey of local boards of elections revealed that the time to complete the manual audit ranged from two hours to two days. The chart below breaks down the time needed for each local board to complete the manual audit.
Two local boards completed the audit over two days. These local boards, Frederick County and Montgomery County, both initially had a discrepancy between the audit results and results from the voting system and certain ballots needed to be retabulated. Frederick County was able to reconcile the discrepancy, while Montgomery County was not, which is explained in the following section.
Number of individuals involved in the audit: The most common number of audit tallying teams was five, which seven local boards needed to complete the audit. Six local boards used between six and ten audit tallying teams, while the remaining local boards required only two to four audit tallying teams. Each audit tallying team is composed of two individuals. The total number of individuals required to complete the audit, including the audit tallying teams and any other required staff members, ranged from five to 27 individuals. As expected, larger local boards required more individuals to complete the audit.
COVID Precautions: All local boards followed state and local COVID precautions when conducting the audit. While some precautions varied, all local boards required the individuals (staff, audit tallying teams, and public viewers) to wear a mask and practice social distancing. Other common precautions included use of hand sanitizer, wiping of all surfaces with sanitizing wipes, temperature checks at the building entrance, and availability of gloves. Finally, due to the additional space needed to accommodate social distancing, not all local boards were able to accommodate in-person viewing of the audit by the general public, or were only able to accommodate a certain number of public viewers at a time. Fifteen local boards offered limited in-person viewing, while three local boards livestreamed the audit online in addition to limited in-person viewing. Six local boards offered livestreaming of the audit only.
Results of the Audit
This chart shows the numerical results of the post-election manual tabulation audit. As the chart shows, there were no vote differences between the manual audit and the voting system results in 21 of the 24 local jurisdictions. Two discrepancies were determined to be human error, and the other discrepancy was a one vote discrepancy and well below the 0.5% threshold to trigger more auditing. Overall, 62,084 ballots from all 24 local jurisdictions were audited, resulting in a statewide audit contest vote discrepancy of 0.0064%. Total, there were 3,037,026 ballots cast for President, which results in a statewide contest vote discrepancy of 0.00013%.
Harford County: There was a one vote difference in the votes for President cast on election day. The voting system reported four overvoted ballots, and the manual audit revealed five. While attempting to reconcile the discrepancy, the local board spoke with the chief judges at the selected election day vote center, who reported that a voter placed an overvoted ballot in the scanner and left before the screen alerting a voter to an overvoted ballot appeared on the scanner. The judges unlocked the front flap of the scanner and dropped the untabulated ballot into the ballot box with the tabulated ballots. This ballot was included in the manual audit and is responsible for the one vote difference in the manual audit.
There were 1,517 ballots counted in the audit resulting in a difference or "total vote discrepancy” of 0.065%, which is well below the 0.5% threshold to trigger additional auditing. If the untabulated ballot had not been counted in the audit, there would not be a discrepancy.
Kent County: There was a one vote difference in the votes for President cast on election day. The voting system reported one additional blank vote than the manual audit, and the manual audit had an additional vote for Trump/Pence than the voting system. After an in-depth review, it appears that one ballot was responsible for the discrepancy. The voting system did not identify an acceptable mark for President and therefore did not count a vote for President, while the audit team saw a mark and counted it as a vote for Trump/Pence. Because this led to a discrepancy in both votes cast for Trump/Pence and undervotes, the result was an absolute sum of vote differences of two.
The audit teams retabulated the election day ballots multiple times and came up with the same results. There were 660 total votes cast for President/Vice-President in Kent County, and an absolute sum of vote differences of two. This difference or “total vote discrepancy” is 0.002%, which is well below the 0.5% threshold to trigger additional auditing.
Montgomery County: There was a one vote difference in the votes for President cast on mail-in ballots. The voting system counted 20 votes cast for write-in candidates, and the audit counted 19 votes cast for write-in candidates. After reviewing the ballots and retabulating the ballots multiple times, the local board determined that a ballot was scanned twice during the mail-in canvass. The specific ballot that was scanned twice was identified in the audit, as it was the only ballot with a write-in vote in a batch of 101 scanned ballots that, upon multiple manual retabulations, was determined to contain only 100 ballots.
Counting the identified ballot only once, there were 6,100 mail-in ballots audited for Montgomery County, resulting in a total vote discrepancy of 0.016%. If the identified ballot had not been tabulated twice during the canvass, there would not be a discrepancy.
These audits showed that the voting system accurately counted 62,084 votes cast for President and Vice President. Another audit – the automated ballot tabulation audit – confirmed that the voting system accurately counted over 3 million ballots in the 2020 General Election.