Taxpayer-funded Automatic Election Recounts:

Conclusions 22

• CL1 Page 19
  There were no tangible benefits from the 2016 recounts because no outcomes were changed. The potential intangible benefits, comfort in not finding evidence of errors, can be far more cost-effectively accomplished by well-known other means.
• CL2 Page 19
  The June 2016 pilot did not provide an adequate basis for the County to extend the recount pilot to the November 2016 General Election. The recount was conducted after certification when there was adequate staff, time and physical space. By being forced to complete any full manual recounts prior to certification, there was insufficient time, staff and physical space to complete any of the manual recounts.
• CL3 Page 20
  The Registrar of Voters should be commended for its extraordinary efforts with the November 2016 recounts. Because of the unprecedented magnitude of the recounts, managers and staff worked excessive overtime hours – nights, weekends and holidays.
• CL4 Page 17
   The BOS, on May 24, 2016, directed the ROV to conduct, on a trial basis, a one-time taxpayer- funded automatic recount pilot for the June 7, 2016, primary election.37  The pilot recount trigger was a margin of 0.5% (one-half of 1 percent) or less of the ballots cast or a 25-vote difference between the winner and loser in any contest.38  The BOS chose to conduct automatic recounts rather than a risk-limiting audit (RLA) citing an ROV comment that RLAs are for audits of an election and recounts are for individual contests.39  The BOS directed that recounts for the June 2016 primary election be manual recounts.40  Of the 19 contests wholly within the county of Santa Clara, there was a single taxpayer-funded automatic recount triggered for the June 2016 primary election. This recount was conducted at a cost of $93,333.19.41 The automatic recount result changed the results by 0.12% (twelve- hundredths of 1 percent) from the original numbers, however, there was no change in the winner of the contest.  On Sept. 13, 2016, the BOS extended the recount pilot to include the Nov. 8, 2016, general election.42  On Sept. 13, 2016, the BOS directed that the pilot automatic recount be conducted manually except in contests for citywide offices in San Jose or countywide offices in Santa Clara County. Those recounts could be conducted by machine.43  On Sept. 13, 2016, the BOS directed that any automatic recount for the November general election be performed prior to certification (Dec. 8, 2016).44  The ROV was unable to complete any of the 10 automatic recounts for the November election before the results were certified.45 The BOS was informed by the ROV on or about Dec. 8, 2016, that it was necessary to continue the automatic recount past the 28-day certification period (Nov. 10 - Dec. 8) for the November 2016 general election because the ROV was unable to complete any of the recounts prior to certification.46  In November 2016, the Santa Clara County ROV conducted automatic recounts of 10 contests out of an eligible 93. According to the accounting calculations reported by the ROV the recounts cost the county $3,288,962 ($2,342,546 for direct and indirect labor; $874,066 for materials and overhead; $72,351 County Employees’ Management Association (CEMA) 37 (BOS, 2016e, p. 55) 38 (BOS, 2016d, p. 7) 39 (ROV, 2016) 40 (BOS, 2016b, p. 3) 41 (BOS, 2016f, pp. 2, 5) 42 (BOS, 2016c) 43 (BOS, 2016c, p. 2) 44 (BOS, 2016b, p. 2) 45 (ROV, 2017, p. 55) 46 (O. o. S. ROV, 2017)
• CL5 Page 18
  agreements costs). The Audit Division of the BOS reported the total cost to be $1,809,18847 by excluding certain costs. The former cost comes to $9.19 per ballot; the latter to $5.06 per ballot.  None of the taxpayer-funded automatic recounts conducted for the November general election resulted in a change of winner.48  In the November 2016 general election automatic recounts, the recount vote tallies differed from the original results by an average of 0.06%.  There has not been a full manual recount in Santa Clara County in at least 15 years.49 47 (BOS, 2017, pp. 2, 3) 48 (ROV, 2017, p. 55) 49 (BOS, 2016f, p. 2)
• CL6 Page 19
  FINDINGS AND RECOMMENDATIONS Finding 1 There were no tangible benefits from the 2016 recounts because no outcomes were changed. The potential intangible benefits, comfort in not finding evidence of errors, can be far more cost-effectively accomplished by well-known other means. Recommendation 1a The County should eliminate its automatic recounts policy and remove Section 3.63 from its policy manual before the November 2018 election. Recommendation 1b If the County rejects Recommendation 1a, then the County should explore whether it can adopt a form of risk-limiting audit for each automatic recount and approve the lease of state certified equipment, physical space, as well as hiring and training of additional staff necessary to complete any recounts prior to certification. Recommendation 1c Pending passage of AB 2125, the County should request authorization from the SOS to adopt a risk-limiting audit in place of the state mandated 1% sample of precincts audit, beginning with the March 3, 2020 statewide primary election. Recommendation 1d Upon implementation of a risk-limiting audit, the automatic recount policy should be ended if it has not been canceled previously. Finding 2 The June 2016 pilot did not provide an adequate basis for the County to extend the recount pilot to the November 2016 General Election. The recount was conducted after certification when there was adequate staff, time and physical space. By being forced to complete any full manual recounts prior to certification, there was insufficient time, staff and physical space to complete any of the manual recounts. Recommendation 2 None
• CL7 Page 20
  Finding 3 The Registrar of Voters should be commended for its extraordinary efforts with the November 2016 recounts. Because of the unprecedented magnitude of the recounts, managers and staff worked excessive overtime hours – nights, weekends and holidays. Recommendation 3 None Finding 4a The County’s use of a 0.5% (one-half of 1 percent) or 25-vote threshold should have been based on empirical evidence or statistical analysis of prior election results. Finding 4b The County’s use of a 0.25% (one-quarter of 1 percent) threshold should have been based on empirical evidence from the 2016 elections. Recommendation 4 If the County rejects Recommendation 1a, then the County should, by June 30, 2019, complete an analysis of thresholds, both percentage and vote count, so that the selection of triggers is based on statistically defensible evidence.
• CL8 Page 22
  The June 2016 Taxpayer-Funded Recount The Recount Results50 As shown in Table 2, although the number of votes changed for each of the candidates, the outcome (winner) was left unchanged. San Jose City Council Manh Percent Percent Lan Diep Difference Nguyen Difference Ballots Difference (votes) (votes) District 4 (votes) (Votes)51 (Ballots) Original 8,687 8,723 36 0.2068% 19,883 0.1811% Recount 8,685 8,697 12 0.0690% 20,116 0.0597% Original-Recount (Difference) 2 26 24 0.1377% 233 0.1214% Table 2 June, 2016 Primary Election Recount Results Analysis The difference in votes between the original tally and the recount for candidate Nguyen was two votes. The difference in votes between the original tally and the recount for candidate Diep was 26 votes. The difference in votes between the two candidates for the original tally was 36 and for the recount 12. Although the vote counts changed by 24 votes between the initial tally and the recount, a change of 0.1377%, there was no change in the winner of the election. Another way of looking at these contests is in ballots counted. There were 233 more ballots counted during the recount than during the initial tally. That amounts to a difference of 0.12%. The difference in numbers of ballots counted between the initial tally and the recount is primarily due to the use of a multi-card ballot. In this election, the San Jose District 4 contest appeared on the second card of the ballot. For purposes of counting the number of ballots cast, the ROV uses the first card of the ballot. It is not unusual for a voter to complete the first card of a ballot but leave the others blank. The ROV does not count blank ballots. The ROV calculates the number of ballots cast during a recount by adding up the votes for each candidate and the number of under votes and over votes. In the District 4 contest, Nguyen 50 The numbers in this table reflect the revised numbers provided to the Grand Jury by the ROV in February 2018 and differ from those in the ROV report (BOS 82471) for the June 2016 primary. The differences are in ballots cast not vote counts (explained below). 51 The percent difference is the difference in votes divided by the total number of votes cast.
• CL9 Page 23
  received 8,687 votes, Diep received 8,723 votes; there were 2,424 under votes and 49 over votes for a total of 20,116. The ROV currently uses mark-sense technology (OPTECH) to optically scan paper ballots. In a study using a very similar OPTECH device, the average tabulation error (absolute difference between the initial count, utilizing OPTECH tabulation, and the recount) was 0.55% (approximately one-half of 1 percent).52 The same study found that the average tabulation error for manually counted ballots was 0.87% (approximately nine-tenths of 1 percent).53 What are some takeaways from this recount? The change in percentage of votes and the change in number of ballots counted between the initial tally and the recount numbers are well below the 0.5% (one-half of 1 percent) recount threshold. Both numbers are also well below published tabulation errors for machine and manual tallies.54 As mentioned above, manual counting has been found to be less accurate than machine tabulation in professional studies.55 Given the closeness of both results and the fact that the outcome was left unchanged, what can be said about the value of this recount without any evidence-based statistical analysis? 52 (Ansolabehere & Reeves, 2004, p. 5) 53 (Ansolabehere & Reeves, 2004, p. 6) 54 See the analysis for the November 2016 election below. 55 (Goggin, Byrne, & Gilbert, 2012)
• CL10 Page 24
  November 2016 Taxpayer-Funded Recounts In the context of this analysis, “results” is used to indicate the numbers of votes or ballots. The term “outcome” is used to refer to the winner of the contest. The term “ballot(s)” is used to refer to the physical ballots and voting records counted during either an initial tally or a recount. A ballot may be comprised of multiple physical pages and there may be multiple contests on each page. A ballot may be a paper record completed at a polling place or mailed- in or completed and delivered to a polling place on election day. A ballot may also come from a DRE (Direct-recording electronic) voting machine with a VVPAT (voter verifiable paper audit trail). The term original and initial are used interchangeably and refer to the election. Analysis In reviewing the data from the Nov. 8, 2016, general election, some important facts are worth noting. The largest difference between the vote counts in the original tally and the recount tally of the 10 contests, as a percentage of ballots cast56 was 0.55%. That occurred in the Monte Sereno City Council contest. In the original tally the leading candidate won by a margin of 12 votes. In the recount of that contest the same candidate won by 6 votes. That was the largest percentage change in a result, between the initial count and the recount, of any of the 10 contests recounted. The smallest change in the margin of a contest between the original tally and the recount of that contest was San Jose Unified School District, Measure Y. For that contest the margin of yes votes was calculated against the percentage of yes votes necessary for passage. In this case the measure needed 66.7% yes votes for passage. This contest qualified for a taxpayer- funded recount because the vote margin in the original count was only 0.42%.57 That margin is within the 0.5% margin for triggering a recount. The comparison between the differences in the original results and the recount results is done as percentages because it would make little sense to compare the absolute numbers between contests, the number of ballots cast varying by a wide range. For example, there were 108,757 ballots cast in the San Jose Unified School District Measure Y contest and 2,201 ballots cast in the Monte Sereno City Council contest. The average change in numbers of ballots between the original tally and the recount for the 10 contests is 0.0556%. The range between the highest percentage change in result and the smallest is 0.2711%. 56 Note that this analysis uses absolute vote difference (margin) between original tally and recount in terms of ballots cast. This is the same calculation that was used to determine whether the contest qualified for a taxpayer- funded recount. 57 67.12% - 66.7%
• CL11 Page 25
  Margins and Absolute Differences By Contest Absolute Percentage Original Original Recount Original Vote Difference Difference Margin Count Of Margin Margin Between Between In Votes Ballots In Votes Contest /Ballots Cast Original & Original & Cast Cast Cast Recount Recount 1 San Jose Unified School District 67.12 0.0700% 95,774 67.19 0.07% 0.0000% 2 Los Altos City Council %6 0.0333% 18,028 %6 0 0.0001% 3 SPcahloo SADltDoi Ustnriifcite, dM Secahsouorle D Yi strict 198 0.4874% 40,622 201 3 0.0075% 4 Cupertino Union School District 218 0.3699% 58,942 222 4 0.0084% 5 City of Santa Clara, Chief of Police 105 0.2487% 42,226 110 5 0.0124% 6 Gilroy City Council 95 0.4871% 19,503 100 5 0.0261% 7 San Jose City Council, District 8 97 0.2424% 40,014 74 23 0.0569% 8 Gilroy Unified School District 52 0.2236% 23,259 34 18 0.0772% 9 Los Altos Hills City Council 19 0.3651% 5,204 14 5 0.0959% 10 Monte Sereno City Council 12 0.5452% 2,201 6 6 0.2711% Table 3 Margins and Absolute Differences By Contest This analysis uses the absolute difference in the calculations rather than net change. Net change indicates how many votes a candidate gained or lost (plus or minus) between the original count and the recount. The absolute numbers are the same as the net values. In the absolute numbers, the plus or minus sign has been removed. Whether or not a candidate’s votes increased or decreased is not critical when determining what the differences are between the counts. In fact, using the gain/loss numbers can be misleading. For example, if one candidate receives five more votes in the recount than they did in the original tally and the other candidate receives five fewer votes in the recount than they received in the initial tally, the numbers cancel each other out. Measuring the absolute difference between the original tally and the recount can enable an evaluation of the accuracy of the initial tally, the tabulation error.58 It should be noted that given the certified results, the Monte Sereno City Council contest would not have qualified for an automatic recount, using the certified result to calculate the percentage margin, because the margin of votes divided by ballots cast (0.5452%) was more than the threshold of 0.50% (one-half of 1 percent) if the basis for the recount was a percent difference threshold. However, that contest would have qualified under the 25-vote or fewer threshold. The difference in count between the original count and the certified results is a good reason to delay recounts until after certification. 58 (Ansolabehere, Burden, Mayer, & Stewart III, 2017)
• CL12 Page 26
  For the San Jose Unified School District Measure Y the percentage difference between the original and recount margins is used. This is to show the margins with respect to each other and with respect to the passing margin of 66.7%. The average percentage difference between the number of ballots cast/counted during the original count and the recounts is 0.0556%. Final Results of The November 8, 2016 Election Vote N of Difference Absolute Number Ballots Candidate Candidate as % of Contest Vote of Ballots Counted A B Total Difference Counted Using an Ballots RLA59 Cast Los Altos City Council 6,355 6,349 6 18,058 18,058 0.0332% City of Santa Clara, Chief of Police 17,618 17,531 87 42,134 7,760 0.2065% Gilroy Unified School District 8,439 8,387 52 23,235 6,915 0.2238% San Jose City Council, District 8 17,258 17,161 97 39,896 3,462 0.2431% Los Altos Hills City Council 1,821 1,802 19 5,201 2,154 0.3653% Cupertino Union School District 19,320 19,102 218 58,688 2,119 0.3715% San Jose Unified School District, 64,280 31,494 399 95,774 40 0.4163% MPaelaos Aulrteo YU nified School District 13,556 13,358 198 40,612 1,502 0.4875% Gilroy City Council 5,471 5,376 95 19,484 1,502 0.4876% Monte Sereno City Council 767 755 12 2,189 1,336 0.5482% Table 4 Final Results Of The November 8, 2016 Election Recounts What are some takeaways from this analysis? Simply conducting a full manual recount of a contest does not ipso facto mean that the recount is more accurate than the original count. It depends on the methods used in conducting a recount. Because there is inherent error in any count, and studies have shown that a hand count is generally less accurate than a machine tabulation, a manual recount cannot be considered any more accurate than the original machine tally. A full manual recount does not provide a cost-effective means of providing a high level of confidence that the original outcomes were either correct or not correct. Simply because the results of the original count and recount are close and did not change does not necessarily 59 These are the approximate number of ballots that would be counted, if no discrepancies are found, in an RLA at a 5% risk limit.
• CL13 Page 27
  mean that the outcome of the contest is correct. Only a statistically valid method of addressing possible errors in the original count and the recount can provide the necessary confidence in the outcome of a contest. In the above analysis the approximate number of ballots that would need to be counted using an RLA with a risk limit of 5% was calculated for each contest. Although this is only an approximation, the numbers of ballots that would have needed to be counted is far less than a full manual recount. The exception is the Los Altos City Council race where the margin of victory was so close that a full manual recount would have been needed. Pertinent California Election Code Information California Elections Code (ELEC) sections §15600 through §15649 govern statewide election recounts. There is no provision in California law for automatic countywide or local election recounts. That is, a county elections official may order a recount (under ELEC §15610) if both of the following apply: “(a) The elections official has reasonable cause to believe the ballots in the precinct have been miscounted.” “(b) The elections official has examined, under oath, the precinct board members or, in the case of ballots counted by a central counting system, the counting board members, and they are unable to explain the returns of their respective precincts.” There is provision in ELEC for voter-requested recounts. It must be requested within five days following the completion of the canvass. Elections official can charge the voter requesting the recount for the costs of that recount. ELEC §15627(a) specifies that if the vote was “cast or tabulated by a voting system,” then the entity making the request for a recount has the right to request that the recount be accomplished either manually or by the same system that was in the original vote. Furthermore, §15627(a) mandates: “Only one method of recount may be used for all ballots cast or tabulated by the same type of voting system.” Under certain specified conditions, the governor may order a state-funded manual recount for a contest involving a statewide office or state ballot initiative, at various vote margins.
• CL14 Page 28
  Legal authority for the Santa Clara County automatic recount pilots Recounts are governed under: California Elections Code - ELEC DIVISION 15. SEMIFINAL OFFICIAL CANVASS, OFFICIAL CANVASS, RECOUNT, AND TIE VOTE PROCEDURES [15000 - 15702] The state of California does not have a taxpayer-funded automatic recount provision. The California State Election Code (CSEC) does permit the County Registrar of Voters to perform a recount if they believe that an error has been made in the earlier count (CA Elections Code Div. 15 Ch. 9 Article 2 §15610). Under that authority, Santa Clara County adopted the automatic recount. The idea is that if an error were to occur, an automatic recount would determine if in fact that error was large enough to change the outcome of a contest. The County determined that its automatic recount pilot was permitted under the CSEC so long as it was performed during the canvass prior to the required certification of election results to the secretary of state. This was determined even though state-funded recounts must be performed after certification of the election in question.
• CL15 Page 29
  Further Discussion About Automatic Recounts Election law covering recounts varies from state to state. FairVote, a nonpartisan group that advocates for election reform, conducted a survey of state recount policies. In that survey FairVote lists 16 states plus the District of Columbia as having provisions for automatic recounts, while 34 states did not. Forty states and the District of Columbia allow for petitioning for a recount by a candidate.60 The laws governing recounts and the statistics stated here are for statewide elections but have general applicability to local elections. The FairVote survey found that statewide election recounts are rare. Of 4,687 elections in FairVote’s survey, there were 27 recounts, 15 of which it cited as being “consequential”. That is, the margin between the winning candidate and the losing candidate was 0.15% (fifteen- hundredths of 1 percent) or less. That study found that of those 27 recounts, only three resulted in a change in outcome (winner). Of relevance to local elections is the fact that recounts for contests in which there were relatively more voters resulted in a lower percentage change in the vote margin than in contests with relatively fewer voters. For example, in contests with greater than 2 million votes cast, the margin difference was an average of 0.016% (sixteen-thousandths of 1 percent). In contests where the total votes cast were less than one million, the change in the vote margin was an average of 0.039% (thirty-nine thousandths of 1 percent).61 The FairVote report does try to extrapolate from their statewide findings to local elections, however, although it does not provide any data to support its advice, the report recommends that a 0.5% (one-half of 1 percent) margin may be appropriate for small local electorates. The report recommends that for smaller states a 0.2% (two-tenths of 1 percent) trigger is appropriate and it recommends a 0.1% (one-tenth of 1 percent) threshold for larger states.62 There were 875,176 registered voters in Santa Clara County at the time of the Nov. 8, 2016, General Election.63 Santa Clara County has more registered voters than 13 states. That puts Santa Clara County into the category of a smaller state and consequently a threshold of 0.2% (two-tenths of 1 percent) for triggering an automatic recount could be justified based on the FairVote study recommendation. 60 (Ritchie & Smith, 2016, p. 11) 61 (Ritchie & Smith, 2016, p. 3) 62 (Ritchie & Smith, 2016, p. 14) 63 (ROV, 2017, p. 8)
• CL16 Page 30
  Risk-limiting Audits Explained Risk-limiting audits can be much more efficient, timely, and cost-effective than full recounts because an RLA enables election officials to count a sample of ballots until the auditors are convinced that there won’t be any change in the original outcome. The sample of ballots counted is generally far fewer than in a full hand recount (unless the margin is extremely small or the contest outcome is in fact incorrect) and thus can be completed faster and with fewer resources. “The risk component in RLAs is the maximum risk that elections officials are willing to take that the audit will not result in a full hand recount when a full hand recount would show that the apparent outcome is wrong.” 64 The basis for RLAs lies in sound statistical practice. The legislature typically decides how small a risk/chance they are willing to take, say 1% or 0.1%, that the audit will not change the outcome of an election when the original outcome is wrong. The RLA process dictates the size of the random sample that needs to be counted to reach the selected level of confidence, in light of what the audit finds as it progresses. The audit stops when the evidence that the outcome is correct is strong enough to meet the risk limit, or when a complete hand count has been conducted. The size of the sample is chosen using easily understood mathematical formulas, instead of a fixed percentage of precincts. Open source software for use by election officials to implement an RLA is readily available that requires minimal training.65 The desired confidence level is selected by the legislature; the mathematical mechanisms are well understood within the world of professional statisticians. This helps make the audit process transparent. The American Statistical Association (ASA), an organization of U.S. professional statisticians, has endorsed RLAs as a best practice for post-election audits.66 Colorado has had mandatory RLAs since November 2017 and Rhode Island requires RLAs as of 2018. In 2007, the California Secretary of State issued the Top-To-Bottom Review of California’s Voting Systems.67 An outgrowth of the Review and the Evaluation of Audit Sampling Models and Options for Strengthening California’s Manual Count68 were two pilot programs implementing RLAs in actual elections in the state of California. The initial pilot RLA program was conducted in 2008 and involved four elections in three counties: The first was conducted during the February 2008 primary election in Marin County; Marin, Santa Cruz and Yolo counties used RLAs in the November 2008 general elections.69 The RLAs were conducted in parallel with the state-mandated 1% of precincts post-election audit. 64 (Philip B Stark, 2009) 65 (Philip B. Stark, 2012) 66 (American Statistical Association Statement on Risk-Limiting Post-Election Audits, 4/17/10, 2010; Lindeman et al., 2008) 67 (California Secretary of State, 2007) 68 (Jefferson et al., 2007) 69 (Hall, 2009, p. 1)
• CL17 Page 31
  In 2010, AB 2023 became law, authorizing the California secretary of state to conduct the Post-Election Risk-Limiting Audit Pilot Program that eventually involved 11 counties. The 11 counties “… successfully completed their audits and confirmed the official election results by reviewing a relatively small number of individual ballots (e.g., a few dozen to a few hundred ballots). By contrast, the statutorily-mandated 1% manual tally conducted in the same elections provided little statistical evidence that the election outcomes were correctly tallied by the voting system, despite requiring substantially more ballots to be hand-counted and examined.”70 There are various ways to conduct RLAs. The 2011-2013 California pilot program used batch- level comparison audits, ballot-level comparison audits, and a ballot-polling audit. A ballot- polling audit examines randomly selected ballots until auditors have achieved the statistical confidence chosen by the election official. No special ballot counting equipment is necessary. A second type of RLA is a comparison audit. Here, outcomes are examined by comparing the original count of clusters of ballots to a hand count of the same clusters. For example, a precinct could be a cluster. In a ballot-level comparison audit, a cluster is composed of a single ballot. To implement ballot-level comparison audits, officials must have a means of comparing a manual interpretation of a ballot with the machine interpretation of that ballot. Ballot-level comparison audits require a way to match a human interpretation of a given paper ballot with how the machine interpreted that same ballot. The County does not now have this technology, but the ROV plans to release an RFP for new equipment in 2018 that appears to have the requisite functionality to enable the ROV to implement comparison RLAs. This new equipment is slated to be available for use in 2019.71 RLAs eliminate the need for automatic or other types of recounts because an RLA provides a desired level of confidence that the outcomes of contests are correct, generally with far less work than a full manual recount. Further, RLAs are far more efficient and cost-effective than full manual recounts (except in cases where the reported results are incorrect and the audit leads to a full manual recount). As is stated in the SOS report on the Pilot Program: “The adoption of laws and regulations permitting or requiring risk-limiting post-election audits will allow elections officials to use the new audit methods to confirm – or correct – official election results, which will help build public confidence in elections and may reduce the need for voter-requested manual recounts.”72 70 (California Secretary of State, 2014a, p. Exec. Summary) 71 (CACE, 2017) 72 (California Secretary of State, 2014a)
• CL18 Page 32
  Assessing Voter-intent Counting Mismarked Ballots What is a mismarked ballot? During the original ballot count by vote tabulating machines, certain ballots are rejected by the tabulating machines because a voter, who failed to follow the printed instructions, improperly marked a ballot, or the ballot card was badly damaged. For example, a voter who did not draw the required solid black line between the end-points next to a candidate’s name or ballot measure choice. Other mistakes include incomplete erasures, cross outs, hand- drawn lines, and arrows to indicate a choice. What happens to mismarked ballots? In the case of ballots rejected by the tabulating machine, a team of ROV workers, attempts to visually assess the voter’s intent. The result is the creation of a duplicate ballot that is properly marked, using the team's determination of voter intent. If voter intent is not unanimously agreed to by the team, the vote(s) for the questionable contest(s) are not counted. The duplicate ballot is then tabulated by machine. Once a mismarked ballot is replaced or discarded during the original count, it is not re- examined during a recount. The ROV estimates that less than 0.1% of ballots in Santa Clara County are replaced by team-decided voter intent in the experience of current ROV staff. The number of such replacement ballots are not recorded.
• CL19 Page 33
  November 2016 Recount Data Los Altos City Council Lynette Neysa Absolute Percent Ballots Percent Lee Eng Fligor Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 6355 6349 6 0.0472% 18028 0.0333% Recount 6369 6363 6 0.0471% 18058 0.0332% Original-Recount 14 14 0 0.0001% 30 0.0001% Difference Palo Alto Unified School Melissa Heidi Absolute Percent Ballots Percent District, Governing Board Batan Emberling Vote Difference Cast/ Difference Caswell Difference Votes Cast Counted Ballots Cast Original 13556 13358 198 0.7357% 40622 0.4874% Recount 13580 13379 201 0.7456% 40612 0.4949% Original-Recount 24 21 3 0.0099% 10 0.0075% Difference Cupertino Union School Phyllis Gregory Absolute Percent Ballots Percent District Governing Board Vogel Anderson Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 19320 19102 218 0.5674% 58942 0.3699% Recount 19267 19045 222 0.5795% 58688 0.3783% Original-Recount Change 53 57 4 0.0121% 254 0.0084% City of Santa Clara, Chief Michael J. Pat Nikolai Absolute Percent Ballots Percent of Police Sellers Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 17618 17513 105 0.2989% 42226 0.2487% Recount 17625 17515 110 0.3130% 42134 0.2611% Original-Recount Change 7 2 5 0.0142% 92 0.0124% Gilroy City Council Paul V. Tom Fischer Absolute Percent Ballots Percent Kloecker Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 5471 5376 95 0.8758% 19503 0.4871% Recount 5490 5390 100 0.9191% 19484 0.5132% Original-Recount 19 14 5 0.0433% 19 0.0261% Difference
• CL20 Page 34
  San Jose City Council Sylvia Jimmy Absolute Percent Ballots Percent District 8 Arenas Nguyen Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 17258 17161 97 0.2818% 40014 0.2424% Recount 17254 17180 74 0.2149% 39896 0.1855% Original-Recount 4 19 23 0.0669% 118 0.0569% Change Gilroy Unified School BC Doyle Paul Nadeau Absolute Percent Ballots Percent District, Governing Vote Difference Cast/ Difference Board Difference Votes Cast Counted Ballots Cast Original 8439 8387 52 0.3090% 23259 0.2236% Recount 8428 8394 34 0.2021% 23235 0.1463% Original-Recount 11 7 18 0.1069% 24 0.0772% Difference Los Altos Hills City Roger Garo K. Absolute Percent Ballots Percent Council Spreen Kiremidjan Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 1821 1802 19 0.5244% 5204 0.3651% Recount 1819 1805 14 0.3863% 5201 0.2692% Original-Recount 2 3 5 0.1381% 3 0.0959% Difference San Jose Unified School Yes No Percent Minus Ballots Difference District, Yes Needed To Cast/ Yes Org to Measure Y Pass Counted Recnt 66.7% Required To Pass Original 64280 31494 59.10% 7.5958% 108757 0.2600% Recount 64347 31531 59.37% 7.3333% 108389 0.2600% Original-Recount 67 37 0.26% 0.2625% 368 0.0000% Difference Monte Sereno City Curtis Rowena Absolute Percent Ballots Percent Council Rogers Turner Vote Difference Cast/ Difference Difference Votes Cast Counted Ballots Cast Original 767 755 12 0.7884% 2201 0.5452% Recount 765 759 6 0.3937% 2189 0.2741% Original-Recount 2 4 6 0.3947% 12 0.2711% Difference
• CL21 Page 35
  References American Statistical Association Statement on Risk-Limiting Post-Election Audits, 4/17/10. (2010). Ansolabehere, S., Burden, B. C., Mayer, K., R., & Stewart III, C. (2017). Learning from Recounts (2017-12). Ansolabehere, S., & Reeves, A. (2004). Using Recounts to Measure the Accuracy of Vote Tabulations: Evidence from New Hampshire Elections 1946-2002. BOS. (2016a). Board Referral 79950. BOS. (2016b). Board Referral 79950 Request for CACE Policy Recommendation On Recounts Minutes. Board Referral. Board Referral 82973 B Automatic Recounts for the November 8, 2016 General Election September 13, 2016, 82973 B C.F.R. (2016c). BOS. (2016d). Minutes May 24, 2016 9:00 AM Regular Meeting. BOS. (2016e). Report 81537 Consider recommendations from the Office of the County Executive relating to automatic recounts. . BOS. (2016f, August 11, 2016). Report 82471. BOS. (2017). Board Referral 89314 Automatic Recount Policy 2017DEC5. BOS (2018a). [Interview with BOS staff member(s)]. BOS. (2018b). Summary of Proceedings February 27, 2018 9:30 AM. CACE. (2015). Report 75194. CACE. (2017). Minutes Citizens’ Advisory Commission on Elections Meeting7 November 2017. California Secretary of State, C. (2007). Top-To-Bottom Review of California’s Voting Systems. California Secretary of State. Aug. 2007. California Secretary of State, C. (2014a). Post-Election Risk-Limiting Audit Pilot Program 2011-2013 Appendices. (073014). California Secretary of State, C. (2014b). Post-Election Risk-Limiting Audit Pilot Program 2011-2013 Final Report to the United States Election Commission. (073014). California Secretary of State.
• CL22 Page 36
  Goggin, S. N., Byrne, M. D., & Gilbert, J. E. (2012). Post-Election Auditing: Effects of Procedure and Ballot Type on Manual Counting Accuracy, Efficiency, and Auditor Satisfaction and Confidence. Election Law Journal, 11(1). Hall, J. L., Luke W. Miratrix, Philip B. Stark, Melvin Briones, Elaine Ginnold, Freddie Oakley, Martin Peaden, Gail Pellerin, Tom Stanionis, and Tricia Webber. (2009). Implementing Risk-Limiting Post-Election Audits in California. Paper presented at the Proceedings of the 2009 Electronic Voting Technology Workshop/Workshop on Trustworthy Elections (EVT/WOTE ’09). Interviewee (2018). Jefferson, D., Ginnold, E., Midstokke, K., Alexander, K., Stark, P. B., & Lehmkuhl, A. (2007). Evaluation of Audit Sampling Models and Options for Strengthening California’s Manual Count. California Secretary of State. July 2007. Lindeman, M., Halvorson, M., Smith, P., Garland, L., Addona, V., & McCrea, D. (2008). Principles and Best Practices for Post-Election Audits. Lindeman, M., & Stark, P. B. (2012). A Gentle Introduction to Risk-limiting Audits. IEEE SECURITY AND PRIVACY(SPECIAL ISSUE ON ELECTRONIC VOTING, 2012). NCSL. (2017, 2017 October 10). Post-Election Audits. Ritchie, R., & Smith, H. (2016). A Survey and Analysis of Statewide Election Recounts 2000- 2015. ROV. (2014). Official Final Results November 4, 2014 Gubernatorial General Election. ROV (2016, 2015-10-27). [October 25 2016 Communication ROV to BOS on RLAs]. ROV. (2017). November 8, 2016 Presidential General Election Post-Election Report Rev. Oct. 2017. ROV (2018). [Communication from the office of the Santa Clara County Registrar of Voters]. ROV, O. o. S. (2017) Interview Office of the Santa Clara County Registrar of Voters. Stark, P. B. (2009). Risk-Limiting Audits. Paper presented at the Workshop on Post-Election Audits - American Statistical Association, Alexandria, VA. Stark, P. B. (2012). Tools for Comparison Risk-Limiting Audits.

No Responses Found 1

Government entities assigned to respond to this report. No response documents have been linked in our database.