Skip to Main Content

2016 General Election: Post-Election Ballot Tabulation Audit Overview, Reports, and Findings

Introduction

After the 2016 General Election, this office conducted a post-election ballot tabulation audit.  The audit as part of the comprehensive audit conducted after each election.  The purpose of the post-election tabulation audit is to verify and confirm the accuracy of the voting system’s reported results.  A post-election tabulation audit is not a canvass or a recount; it is used to verify that the voting system accurately tallied ballots and that the winners of each contest were called correctly.

In June 2016, this office conducted a pilot program in Carroll and Montgomery Counties where three different post-election tabulation audit methodologies were tested1.A comprehensive report from the pilot program is available on our press room page under “Recent Reports”.  At the conclusion of the pilot program, an independent, automated tabulation audit was recommended for statewide implementation following the 2016 General Election as the most comprehensive, efficient and user-friendly of the three piloted methods.

An independent, automated audit relies solely on the use of independent software to tabulate ballot images. The results from the independent tabulation are then compared to the tabulation results from the voting system. Any variances between the two tabulations can be identified and resolved using the ballot images.  For the 2016 General Election, this office contracted with Clear Ballot, a Boston-based elections technology company which offers the ClearAudit software product. ClearAudit is the only currently available, market-ready software product that can perform an independent automated post-election tabulation audit using ballot images imported from another voting system.

The post-election tabulation audit was conducted using ballot images, as was the pilot program.  The use of ballot images allows elections officials to maximize the technological functions of the new voting system while minimizing human error and eliminating chain of custody issues by using securely stored ballot images, rather than voted paper ballots.  The use of ballot images removes the need for election officials to physically handle or count voted ballots unless a petition for recount or other judicial challenge is granted.

Finally, a discrepancy variance level of 0.5% was established for the post-election tabulation audit2.  If there was a variance greater than 0.5% in any given contest between the voting system results and the ClearAudit results and could not be explained, the adjudication of all ballot images from that contest by both systems could be compared.

Pre-Audit Tasks

To prepare for the post-election tabulation audit, SBE provided Clear Ballot with:

  • PDF files of all ballot styles for all 24 counties;
  • Pre-election reports for all precincts showing no votes cast;
  • Information to assign ballot styles to precinct; and
  • Files needed to process ballots used with the ExpressVote ballot marking device.

Upon receipt of the data, Clear Ballot:

  • Created a Ballot Definition Files for each county;
  • Validated each Ballot Definition File; and
  • Created a ClearAudit database for each county.

Audit Process

Clear Ballot shipped to each local board of elections an external hard drive.  The local boards of elections saved on the hard drives ballot images, in unencrypted form, from election day and early voting and returned the external hard drives to Clear Ballot.
Once Clear Ballot received the external hard drives, Clear Ballot:

  • Copied voted ballot images from each county into its corresponding audit database;
  • Performed an automatic tabulation of voted ballot images from election day and early voting;
  • Resolved unreadable ballots;
  • Performed an audit database review; and
  • Sent a Preliminary Statement of Votes Cast to SBE.

When SBE received Clear Ballot’s Preliminary Statement of Votes Cast, SBE provided Clear Ballot with precinct-level results from early voting and election day.  Clear Ballot used this information to create the Comparison of Votes Cast, a report that compares results from the voting system and Clear Ballot’s tabulation.  This delay in sending the precinct-level results to Clear Ballot was intentional.  This procedure, in effect, resulted in a “blind” audit meaning that Clear Ballot published its tabulation results before knowing the results from the voting system.

Once Clear Ballot finished comparing its results against the results from the voting system, Clear Ballot provided each local board of elections with log-in access to its county audit database, and SBE also had access to all audit databases.

After the absentee and provisional canvass, each local board saved on a hard drive ballot images, in unencrypted form, of all absentee and provisional ballots cast and results from the absentee and provisional canvasses and sent the hard drives to Clear Ballot.  Clear Ballot added these ballot images into the appropriate database, resolved unreadable ballots, performed an automatic tabulation of voted absentee and provisional ballot images, and generated the reports described below.

Reports Produced by Clear Ballot
ClearAudit technology provides visual reporting tools that allow election officials to generate sortable contest, ballot and precinct reports that create a visual connection to each ballot image and provide detailed information about how it was adjudicated. Clear Ballot produced the following audit reports for each county:

  • Comparison of Cards Cast for each canvass:  This report compares the number of ballots counted during early voting, on election day, during both absentee canvasses, and during the provisional canvass against the number of ballots tabulated by ClearAudit.  This ensures that ClearAudit counted the same number of ballots as the voting system.
  • Comparison of Ballots Cast by Precinct: This report compares the number of ballots cast in each precinct3  against the number of ballots tabulated by ClearAudit.  This is another way to ensure that ClearAudit counted the same number of ballots as the voting system. 
  • Comparison of Votes Cast: This report compares the results from the voting system against the results tabulated by ClearAudit and identifies possible discrepancies by candidate or choice. 
  • Contest Vote Discrepancy Threshold Report: This report shows - by contest - the number of vote differences between the two systems, the total votes cast by the voting system, and the vote difference as a percentage. Before the audit was performed, SBE determined that a percentage of 0.5% or higher would trigger an additional review, which could include a manual review of voted paper ballots.

The report for each county and an explanation of any variance greater than 0.5% are provided.

2016 General Election: Post-Election Tabulation Audit Reports

Allegany County

Anne Arundel County

Note: In precinct 1-19, there is a difference in the number of ballot images for President/Vice President and U.S. Senator. Clear Ballot's Comparison of Votes Cast report show 834 ballots, while the voting system report shows 833 for President/Vice President and 835 for U.S. Senator. This difference is the result of two errors during the review of write-in votes for candidates who did not officially file as a write-in candidate. These errors do not impact the verification of the voting system's results.

Baltimore City

Baltimore County

Note: For precinct ​015-026, the voting system transferred election results but not the ballot images. This issue has been reported to the vendor and is the reason for the difference in Baltimore County's reports. This explains why the difference in the results is greater than 0.5% in some contests.

Calvert County

Caroline County

Carroll County

Cecil County

Charles County

Dorchester County

Frederick County

Garrett County

Note: Election results from Garrett County's 2 ballot questions had to be reported by precinct. To do this, election officials created for each precinct a unique contest for each ballot question. This enabled the voting system to report the results for these ballot questions by precinct. When Clear Ballot tabulated the ballot images from Garrett County, they tabulated and totaled the results by ballot question (not by precinct). This explains why the difference in the ballot questions is greater than 0.5%.

Harford County

Note: For the second absentee canvass, there is a difference of 10 ballot images in Clear Ballot's Comparison reports. This difference is the result of Harford County receiving and tabulating 10 timely absentee ballots after all other ballot images were sent to Clear Ballot. These errors do not impact the verification of the voting system's results.

Howard County

Kent County

Montgomery County

Prince George's County

Queen Anne's County

St. Mary's County

Somerset County

Talbot County

Washington County

Wicmoico County

Note: Voters in Wicomico County had a 2-page ballot for this election. During the second absentee canvass, election officials divided the ballots into two groups - page one and page two. They then fed into one tabulator all of the page one sheets and fed into a different tabulator all of the page two sheets. Because the voting system looks for a page one sheet, one blank page one sheet was scanned into the tabulator for the "page two" sheets. This single blank page one caused the 1 ballot difference in Wicomico County's reports.

Worcester County


Audit Finding

Using the Clear Audit system, we validated the results of the 2016 General Election. 

An independent, automated post-election tabulation audit method maximizes the use of technology in election administration, minimizes human error and handling of official election materials, including voted ballots, and provides the fastest, most comprehensive and most visual and transparent presentation of audit results.  

In response to a request from the Maryland General Assembly, SBE submitted a report on its post-election tabulation audit of the 2016 General Election (PDF) 1  The post-election audit methodologies tested during the pilot were 1) an independent, automated audit; 2) a ballot level audit applying Risk Limiting Audit principles; and 3) a fixed percentage audit.
2  This is calculated as the percentage of:  Absolute sum of all discrepancies in a contest 
Total number of votes in a contest
3  For canvassing purposes, each early voting center is a separate “precinct.”   Each absentee canvass and provisional canvass is a separate precinct.