June 5, 2012 - Post Canvass Risk Limiting Audit

Lynda Roberts, Registrar of Voters, Elections


The post canvass risk limiting audit in Marin County of ballots cast in the June 5, 2012 Presidential Primary Election was part of a pilot project on risk limiting audits conducted with the Secretary of State’s Office and UC Berkeley professors Philip Stark, Professor of Statistics and David Wagner, Professor of Computer Science.

Audit Process

After the June 5th election was certified, the Marin County Elections Department scanned 29,121 ballots from Supervisor Districts 2 and 4 with a Department-owned Fujitsu 5950 scanner. The scanner made an optical image of each side of each ballot and imprinted a number one side of each ballot. Supervisor Districts 2 and 4 were chosen because each districts had a district wide race for County Supervisor contained wholly within it.

The Election Department transmitted the ballot images to Professor David Wagner to prepare a transparency count of the votes on the scanned ballots. A comparison of the winners on the county’s Statement of Votes for the districts showed that they were the same as the winners on the transparency count. The remainder of the audit was put on hold until after the November 6, 2012 General Election.

The audit of ballots took place on February 14, 2013. The Elections Office posted a Notice of the audit on its website and front door and sent a copy of the Notice to its Election Advisory Committee. On Feb. 14 the department set up the room for the audit with one laptop computer and printer not connected to the Internet, one desktop computer and printer connected to the Internet to access the audit tools page on the UC Berkeley Dept. of Statistics website, and two projectors and screens for the public to view the ballot images.

The comparison of ballots started at 9:30 AM. There was one public observer present. The Audit Tools report showed that the number of ballots to be audited was 54, based on the number of votes cast in each race and the margin of victory of the candidates. The observer rolled eight 10-sided dice to get a seed number which was then entered into the audit tools to produce the random numbers of the 54 ballots to be compared. Two Elections Dept. staff pulled the ballots in the audit sample from sealed containers while one staff member printed out the results of the transparency count for each ballot. The Observer and Elections Dept. staff compared each ballot with the results from the transparency scan and found that all votes matched. The audit ended at 12:00 PM.

Cost Comparison Between Risk Limiting Audit and 1% Manual Tally

A. Cost of Risk Limiting Audit

Cost of scanning ballots
Cost of transparency audit
Total Cost

B. Cost of 1% Audit

Cost of 1% audit