The San Mateo County Registration & Elections Division will be conducting a pilot risk-limiting audit for the June 23, 2020 County Service Area No. 1 Special Mail Ballot Election

What is a risk-limiting audit?

A risk-limiting audit (RLA) is a newly approved post-election audit procedure meant to guarantee the accuracy of the election outcome, and is a necessary step in the canvassing of an election, a requirement to certify election results. Currently, most post-election audits in California have been conducted using the one percent manual tally, which involves choosing a random sampling of one percent of all precincts from Vote Centers and one percent of batches of Vote by Mail ballots.

The risk-limiting audit differs in that its randomly selected ballots come from all ballots cast in the election regardless of precinct or batch. The physical ballots are then retrieved from their secure storage containers and inspected by an independent audit board, and their results are compared to the voting system’s cast vote records.

The San Mateo County Registration & Elections Division will be conducting a pilot risk-limiting audit for the June 23, 2020 County Service Area No. 1 Special Mail Ballot Election. The public is welcome to attend both the rolling of the dice to generate the random seed as well as attend to observe the audit itself.

The June 23 RLA Pilot Press Release was posted June 19 and can be read here.

Random Seed

This 20-digit number is generated through a dice roll using three ten-sided dice and matching the resulting three numbers to a corresponding number in the sequence, for the purposes of this 20-digit number, that will be Red-White-Blue-R-W-B-R-W-B-R-W-B-R-W-B-R-W-B-R-W, or 6 rolls of three dice, and the final roll consisting of two dice.

At 10:00 a.m. on Friday, June 26, the random seed was generated at the offices of the San Mateo County Registration & Elections through a dice roll. The event was livestreamed on our Twitter page @smcvote and can be viewed here. The resulting random seed was 66901406763576268685.

Audit Boards

Audit boards will consist of elections staff who must sign an oath of letter of intent. “At all times, at least one member shall serve as an observer of the audit and may not make determinations of voter choices. At least two members, excluding the observer, shall make determinations of voter choices for each ballot card examined by the audit board. Audit board members may rotate roles.”

This process differs from the one percent manual tally in that the goal is not to match the number of votes for each candidate in the entire batch, rather, the goal is to interpret the votes on each ballot and compare this to the cast vote record for that particular ballot. In any instance wherein a ballot has been duplicated, such as an overseas or military voter’s ballot or ballot marked with ink not recognized by the ballot scanner, the original ballot will be inspected by the audit board rather than the duplicated ballot.

If there is a discrepancy between the CVR and the selected ballot, meaning that the voting system’s tabulation of the ballot differs from the actual markings on the ballot (Elections Code section 15154), audit boards will determine whether the difference is neutral, an understatement, or an overstatement, depending on the effect of changing the voting system interpretation of the ballot to match the hand interpretation. If changing the interpretation of a ballot according to the voting system to make it match the human interpretation of the ballot would widen every pairwise margin in every contest under audit, that ballot has an understatement. Understatements do not call the outcome into question. If changing the interpretation according to the voting system to match the human interpretation would narrow any pairwise margin in any contest under audit, the ballot has an overstatement. If enough ballots have overstatements, the outcome could be wrong.

To determine an overstatement or understatement, the audit board needs to know the final outcome of the contest within the county, meaning that, if Proposition 2 is being audited, the audit board needs to know the outcome of the contest.

Here is an example of how the audit boards shall proceed:

The results of Proposition 2 were:

Yes 57.0% 570

No 43.0% 430

The audit board is given these results, as well as the CVR and the ballots that they are required to audit.

Neutral

Their first ballot CVR states that the ballot should read Yes: 1 No: 0.

They interpret this ballot, and the markings match the CVR. This is neutral, meaning that vote totals and margin do not change. There is neither an overstatement nor an understatement.

Overstatement

If the paper ballot doesn't match the CVR and the change in vote narrows the margin of victory, it is an overstatement of the winner of that contest.

On the next ballot, the CVR claims that this should be Yes: 1 No: 0.

However, when interpreting this ballot, the audit board finds that the ballot is in fact a “no” vote instead of a “yes” vote. The ballot was incorrectly tallied as a “yes” vote, so a yes vote should be taken away from the totals leaving Yes (56.9% 569) and No (43.1% 431). It is an overstatement because it was incorrectly tabulated in favor of the winner of the contest and as a result was an overstatement of that outcome.

Understatement

When a discrepancy occurs between the CVR and the actual ballot that widens the margin of victory, that is an understatement.

On the next ballot, the CVR claims that this should be Yes: 0 No: 1.

But the audit board finds that the ballot does not match the CVR, and in fact the ballot is a “yes” vote rather than a “no” vote, bringing the new total to be Yes (57% 570) and No (43% 430).

Overvotes

Overstatement and understatement should not be confused with an undervote and overvote, although there might be overlaps with these terms. When a ballot is determined to be an overvote and it was incorrectly tabulated, that ballot can no longer be counted, which reduces the denominator, or total number of ballots cast in the contest.

The CVR claims that this should be Yes: 1 No: 0.

But the audit board finds that the ballot does not match the CVR, and in fact the ballot has clear markings for both “yes” and “no”, bringing the new total to be Yes (56.96% 569) and No (43.04% 430). In this instance, the overvote is both an overstatement as well as an overvote.

Observers

Members of the public are welcome and encouraged to attend both the dice roll and the audit. The risk-limiting audit and the dice roll will be held at the Registration & Elections Division at 40 Tower Road in San Mateo. Please call 650.312.5222 or email registrar@smcacre.org if you wish to observe the audit or be a part of the process of generating the random seed. Follow @smcvote on Twitter for updates.

Ballot Manifest

The ballot manifest is a log kept independently of the voting system that tracks the storage containers of all ballots cast in an election. This log is manually kept by Elections staff in keeping with California Elections Code.

202-1,4
101-1,51
101-2,40
101-3,90
101-4,58
101-5,65
101-6,59
101-7,70
101-8,71
101-9,80
101-10,77
102-1,44
102-2,72
102-3,47
102-4,74
101-11,148
101-12,40
102-5,54
102-6,71
102-7,74
202-2,1
102-8,20
301-1,34

RLA Tool

An online software utility used by the Secretary of State and local Election Officials to conduct the risk-limiting audit. This was developed by Philip Stark and can be found here.

RLA Results

A ballot level comparison risk-limiting audit was set with a 5% risk limit on July 7, 2020 at 9 am. The ballots selected for the RLA were:

1, 83, 101-2, 28
2, 96, 101-3, 1
3, 147, 101-3, 52
4, 158, 101-3, 63
5, 193, 101-4, 8
6, 213, 101-4, 28
7, 247, 101-5, 4
8, 372, 101-7, 5
9, 438, 101-8, 1
10, 477, 101-8, 40
11, 541, 101-9, 33
12, 586, 101-9, 78
13, 586, 101-9, 78
14, 700, 102-1, 35
15, 726, 102-2, 17
16, 778, 102-2, 69
17, 823, 102-3, 42
18, 837, 102-4, 9
19, 960, 101-11, 58
20, 964, 101-11, 62
21, 993, 101-11, 91
22, 1077, 101-12, 27
23, 1133, 102-5, 43
24, 1169, 102-6, 25
25, 1177, 102-6, 33
26, 1244, 102-7, 29

There was one round of the RLA. There were no discrepancies in the first round of the RLA and the risk limit was met at 5%.