This is a section of the Cops Canvass Report
. See also San Diego County Canvass Procedure
Ballot Control and Scanning
The first step of our review is to determine if all the ballots were correctly processed.
Acceptable Error Rate
Voluntary Voting Systems Guidelines
were authored by the Technical Guidelines Development Committee (TGDC), a committee authorized under the HELP America Vote Act (HAVA) of 2002, and researchers at the National Institute of Standards and Technology (NIST) for the Election Assistance Commission (EAC). These guidelines were authored primarily to guide voting machine manufacturers in their design of voting systems.
They were recently modified to require a vote-capture device that uses independent voter-verifiable records (IVVR). For example, durable paper ballots that are scanned represents a record that can be easily read by the voter and interpreted by machines, the "Cast Vote Record" (CVR).
This document also describes acceptable error rates. In particular, it defines "disenfranchisement" (Section 22.214.171.124 "Manageable failures per election") as follows:
Any failure that results in all CVRs pertaining to a given ballot becoming unusable or that makes it impossible to determine whether or not a ballot was cast is called disenfranchisement. It is unacceptable for even one ballot to become unrecoverable or to end up in an unknown state. For example, an optical scanner that shreds a paper ballot, rendering it unreadable by human or machine, is assessed a disenfranchisement type failure; so is a DRE that is observed to "freeze," providing no evidence one way or the other whether the ballot was cast, when the voter attempts to cast the ballot.
The guidelines suggest that no error in terms of disenfranchisement is acceptable (error rate=0), and that the central tabulator would exhibit an error rate of 1.237×10-6
, that is, one lost ballot in 808,407 ballots cast. In San Diego, with 1,245,947 ballots cast in November 2008, we would expect no more than 1.54 ballots to be lost due to central tabulation error. This can be rounded up to 2, i.e. the election system in San Diego County, according to the federal guidelines, should lose no more than 2 ballots in the November, 2008, election.
Although the Election Assistance Commission states that there should be zero error rate with regard to disenfranchisement
, this requirement, although desired, is not feasible in any real system. Therefore, for the purposes of our analysis, we will use the higher error rate suggested as the maximum for central tabulation, i.e. 2 ballots lost. If we combine the two limits, for the County of San Diego, we are limited to four ballots lost.
The question of disenfranchisement is whether every voter who submits a valid ballot has their vote counted, not whether their vote was counted correctly. Disenfranchisement can occur when elections equipment skips ballots (i.e. misfeeds) and there is no correction or if ballots are lost or stolen, for example. It might be argued that random disenfranchisement will not change the outcome of the election since any truly random change will average out to no change. However, this is absolutely not the statement by the Election Assistance Commission guidelines, which explicitly state that every vote must be counted, and no ballots should be disregarded.
Method of Analysis
The question "were all ballots counted" does not consider whether the vote is counted correctly from those ballots, but simply that every ballot that is cast is attempted to be counted and is not lost in the shuffle. This must always be our first order of business in reviewing any vote counting operation.
To determine the performance of the San Diego County Ro V
in this regard, we reviewed data collected in our 5% sample regarding the number of ballots processed. The most significant figures are:
- The number of people who signed rosters at the polling place and who received paper ballots.
- Total number of signatures is Reported on Ballot Statements and signatures can be counted on the sign-in rosters to verify those figures.
- Number of paper ballots is the total number of signatures minus provisional voters and minus touchscreen machine voters.
- The number of ballots counted by precinct workers at the close of the polls
- The number of ballots scanned by scanner equipment at the Tally Center
- The number of ballots reported in the Audit Log of the central tabulator for each precinct.
- The number of ballots reported in the final canvass report by the Ro V
Issue Category I: Most Precincts Did Not Balance
This issue is handled by Issue of Concern Issue A006 - "Ballot Statements - Few balanced"
We were surprised to discover that very few precincts (15 = 17%) in our sample reported correctly reconciled the count of ballots with the number of signatures after their polling place closed. 83% of the precincts reported that their totals did not match the number of signatures on sign-in rosters, and most of those could not identify why the discrepancy occurred. See the column "Ballot Statement Notes" on All Precincts
and look for "Yes", meaning that the count of signatures matched the count of ballots. Many precincts simply gave up and did not attempt to determine why the discrepancy occurred. Others reported that they were instructed to stop attempting reconciliation after several tries because, "they can fix it at headquarters."
The following counts and observations are relevant to this issue in our attempt to make sense of this disaster.
Blank Ballot Count
The count of blank ballots received by each polling place is determined by Ro V
staff in preparation for Supply Pickup Weekend and documented on the Ballot Inventory Sheet. The Precinct Inspector counts the ballots received and confirms the count on the two-part Ballot Inventory Sheet. It is difficult to imagine that this count can be significantly in error with this two-step check. This count can be considered very reliable.
Counts for Voted, Spoiled, Unused, Provisional Ballots
Precinct workers are instructed to count voted, spoiled, unused, and provisional ballots after the polls close. Some reported in their comments that they counted ballots several times, although this is not specifically documented by the Ballot Statement form. In our analysis, we consdered these counts as "reliable" because counting ballots is not a difficult procedure.
Reliability of Sign-In Count
We counted the number of signatures on sign-in rosters by manually reviewing the rosters and making notations about the number of signatures on each page. We compared that figure with the number as reported on Ballot Statements
Why Statements did not Reconcile
- no significant variation between the count we made by inspecting sign-in rosters and that reported on Ballot Statements. This implies that this count is relatively reliable.
- Sometimes it was difficult to know if the signature existed or not, since some signed on top of the notation "Mail Ballot Received", some signatures are not much different from random markings, and some precincts added notations, such as lining-through some signatures in their counting process.
- that many precincts allowed VBM voters to sign over "Mail Ballot Received" if they had a mail ballot to surrender, and some allowed VBM voters who did not have a ballot to surrender to sign it as well.
- Many provisional voters did not sign the rosters.
Given an expectation that the reported counts of signatures and ballots would be reasonably reliable, why did so few precincts balance?
- Voters did not sign the rosters - Although it is a requirement that voters sign the roster, some did not and still received a ballot. This was particularly true of provisional ballot voters.
- VBM voters signed the roster over the "M" or "Mail Ballot Received" designation. There was no place for VBM voters to sign if they had their ballot to surrender.
- VBM voters who just turned in their envelopes did not sign anything, so these were untrackable.
- VBM voters sometimes signed both the provisional roster and the white/blue/pink rosters if they did not have a ballot to surrender.
- Voters sometimes returned their ballot to the wrong precinct, in locations where multiple precincts were conducted in parallel.
- Ro V staff said it may be the case that voters left with their ballot and did not turn it in. However, this sounds like a very unusual case as it makes no sense for voters to come to the polling place and not vote.
Issue Category II: Many Ballots were not scanned
Ballot Statement for Seq# 1563 documenting 565 ballots counted.
Top of the Scanner tape for Seq# 1563 documenting only 550 ballots scanned -- 15 ballots lost.
This issue is handled by Issue of Concern Issue A044 - "Many Ballots Not Scanned"
Polling place workers count the number of ballots in their attempt to complete the Ballot Statement form and reconcile the numbers. Counting ballots seems like a relatively simple procedure particularly if repeated to check the count, as many polling places reported. It is likely that they counted multiple times mainly because they could not find the source of the reconciliation discrepancy.
However, when the number of ballots was compared with the number of "Cards Cast" from the scanner tapes, we found that 429 ballots were lost in our 5% sample. Extrapolating this over the entirety of the election, a total of over 9,000 ballots were lost, more than five per precinct on the average.
Only 4 precincts had no apparent loss of ballots when they were scanned in our 85 precinct sample (4.7%).
To cite a clear example, consider Precinct Sequence 1563
which is one of the few Ballot Statements that reconciled completely. The precinct had no touchscreen ballots. The ballot statement discloses 565 ballots counted and that matches the number of non-provisional signatures (565). However, the scanner tape from the Tally Center says there were only 550 cards cast.
Therefore, 15 ballots were lost in this precinct alone.
What happened to the ballots:
We asked the Ro V
this question and they refused to explain it, saying that it was not a record that could be provided in a California Public Records Request. We refuse to accept stonewalling, and would like to know what happened to these ballots, as this is the first step to producing a reliable canvass.
The options are:
- Ballots were never in the "voted" box - This would imply that workers in over 95% of the precincts cannot count the ballots and that they consistently miscount by missing ballots. Since we did not find any obvious deviation from the number of signatures on rosters and the counts provided by the precinct workers, the concept that they cannot count ballots is viewed as VERY UNLIKELY.
- Ballots were removed because they were "Write Ins" - At first glance, this seems to be likely, despite the fact that the scanner tapes include values for write-ins. If this is the case, then the removal of these ballots makes it very difficult to track lost ballots in the scanning process.
The Post Election Manual Tally procedure detected errors and lost ballots in 29% of the precincts. The Ro V confirmed these errors when they reran the precincts through the scanners, magically finding the ballots. Instead of seeing this as confirmation of a serious problem, the Ro V said there was no problem at all because the rescan corrected it, and they refused to answer our question regarding where the ballots went. (See "Manual Tally Procedure").
- Ballots were misfed through the scanner, either accidentally or intentionally - This is a distinct possibility. The scanners were designed for one-at-a-time manual feeding. If an operator feeds them too quickly, they may skip a ballot, or if the operator sees a vote they do not like, they could skip the ballot and place it directly in the output tray. The Ro V confirmed that both of these were possibilities.
- Ballots are removed from the ballot box prior to scanning - 6% of the tamper-evident seals are either tampered with or incorrect and the tracking of these seals is minimal, with no followup. However, the number of precincts with tampering was not high enough to account for the large number of precincts with lost ballots (95%).
, the Ro V
mentioned the fact that write-ins and damaged ballots were removed from the rest of the ballots and this may be the explanation for this discrepancy. However, there are no documents (that we are aware of) that track this or report on the number of write-ins and their disposition.
Issues of Concern related to Ballot control and scanning
Issue A006 - Ballot Statements - Few balanced
Only a few of the Ballot Statements had figures that balanced, and only one precinct had all the ballots correctly scanned. See Unscanned Ballots
for a full analysis of this problem.
According to federal election guidelines, disenfranchisement of even one voter, or more than one lost ballot in 808,407 ballots cast due to equipment failure is not allowed. Tracking this problem starts with the sign-in rosters and Ballot Statements
completed at the polling place.
Polling workers are volunteers and are not specifically suspect of malfeasance, but improved training and simplified reconciliation procedures should be established to minimize the discrepancies in the count of ballots.
- Issue A006: Proposal to Ro V
Suggestions to Improve Ballot Statement Reliability
- Avoid multiple-precinct polling locations or separate them more distinctly so it will be very unlikely that a voter would turn in ballots to the wrong precinct ballot box.
- Simple ballot scanners that detect the precinct and will cause an alert if a voter attempts to deposit a ballot into the wrong ballot box could be used.
- Scanners could also provide a machine count of the ballots.
- Provide a worksheet to reconcile smaller aspects. Current reconciliation attempts to reconcile the total number of signatures with the total number of voted ballots.
- Separately reconcile provisional ballots and provisional sign-in sheets.
- Separate provisional ballots into two groups:
- VBM voters who did not surrender their VBM ballots who are listed with that precinct
- Other voters who are voting in the "wrong" precinct.
- Subtract provisional and touchscreen voters from total signatures to determine total paper ballot signatures and reconcile that with the number of actual ballots.
- Maintain sign-in rosters for ALL VBM ballots returned to each polling place.
- Currently, separate sign-in rosters are not maintained for VBM voters who do not wish to vote in person.
- Sign-in rosters will allow tracking the VBM ballots and provide a way to determine where in the system they are lost, if indeed they are.
- Provide sign-in locations for VBM voters who are listed as members of the precinct and designate if mail ballot was surrendered or not. Those who do not surrender their mail ballot should also sign the provisional roster and obtain a provisional ballot. With this designation, duplicate signatures can be easily located.
- Avoid the current practice of "signing over" the "MAIL BALLOT RECEIVED" designation in the signature box.
- Avoid the practice of putting a line through the signature box, which some precincts did apparently to help count them.
- Include a checkbox that can be marked by Precinct workers to designate that the scribble represents the voter's signature. Some signatures are so bad, they are hard to discriminate from random markings.
- The Ro V should produce a report summarizing the count of signatures, the actual number of ballots returned, and an overall percentage of error. This report should be mandated by the So S. The Registrar should encourage precincts to do better (The registrar is required to compel the accounting per Election Code 14405: "(a) The members of the precinct board shall account for the ballots delivered to them by returning a sufficient number of unused ballots to make up, when added to the number of official ballots cast and the number of spoiled and canceled ballots returned, the number of ballots given to them. The officers receiving returned ballots shall compel this accounting.") Precincts that produce accurate reports that account for all signatures should be acknowledged with certificates of merit or other prizes. Similarly, those precincts that do not do well will need additional training or staffing changes so the failure does not repeat in the next election.
Issue A044 - Many Ballots Unscanned
When the number of ballots as counted by poll workers was compared with the number of "Cards Cast" from the scanner tapes, we found that 429 ballots were lost in our 5% sample. Extrapolating this over the entirety of the election, a total of over 9,000 ballots were lost, more than five per precinct on the average.
Only 4 precincts had no apparent loss of ballots when they were scanned in our 85 precinct sample (4.7%). See Unscanned Ballots
for a full discussion of this issue.
In addition to the related Issues of Concern: Issue A007 - "Ballot Statements - Removed Write-ins Not Documented"
, Issue A014 - "Tally Center - Write-Ins Removed and Not Counted"
, Issue A015 - "Tally Center - Damaged Ballots"
, and Issue A019 - "No Comparison of Ballot Count"
, it should be observed that the general strategy of the Ro V
to count ballots is problematic.
Currently, the Ro V
uses 150 hand-fed Diebold scanners and 150 volunteers feeding those scanners, likely because the Ro V
already had purchased these scanners. It would be superior to use fewer high-speed automatic feeding scanners which are common in private industry today. This is the configuration suggested by the Elections Assistance Commission, and would likely vastly improve the throughput and reliability of the scanning process. This is the general direction suggested by the Open Canvass Proposal
, where the ballots are simply scanned and turned into digital images before any vote recognition is attempted.
It is critical that the Ro V
address this issue, as losing 9,000 ballots is a serious concern in the election.
Please explain how and why there are so many ballots that are not scanned.
We recommend that you consider discontinue using the Diebold scanners that were originally designed for use in precincts and exhibit an excessive error rate. Instead, consider using high-throughput digital scanners with automatic page feeding capability. These commonly run at 100 to 200 pages per minute, instead of about 6 pages per minute with the existing scanners.
Since these scanners exhibit a failure rate far in excess of the guidelines of the Elections Assistance Commission, perhaps the manufacturer, Premier Election Solutions (formerly Diebold Systems
) should provide corrections to their equipment or provide other equipment to reduce the failure rate.
- Issue A044: Legislative
The number of unscanned ballots discovered in San Diego County Ro V is extremely distressing and if this is currently allowed by law, something is seriously wrong with our elections code. The Election Assistance Commission has provided a number of requirements for election equipment manufacturers. These requirements should be resolved to state law so that counties have recourse to demand that the equipment manufacturers provide solutions that meet those requirements or describe why it is technically impossible to meet them.
Issue A007 - Ballot Statements - Removed Write-ins Not Documented
did not include a count of the number of write-ins and damaged ballots removed. Precinct workers are requested to look through the ballots and remove all write-ins and damaged ballots and place these into a separate group, which is to be banded. The incoming inspection activity has two check-offs that relate to this: 1) write-ins removed and 2) write-ins banded. However, no one counts these so we know how many have been removed. This should be recorded on the BallotStatements
. In addition, these ballots (nearly 9,000 ballots) are withheld and not processed for an extended period of time. During this time, the "precincts reporting" figure should not claim that the precinct has been fully reported.
- Issue A007: Proposal to Ro V
Document the number of write-ins and damaged ballots and their disposition for all precincts in future elections. The Ballot Statement may be a good place to record the number. Until these ballots are included in the count, the "precinct reporting" figure should not be 100% (See Issue A013: Tally Center Reception Count Misleading.)
- Issue A007: Legislative
Law should mandate that elections officials document the number of write-ins and damaged ballots and their disposition for all precincts in future elections. In addition, the "Precincts Reporting" figure released to the media, should not include any precincts that have not been fully counted, including the write-in ballots. (See Issue A013: Tally Center Reception Count Misleading.)
Issue A014 - Tally Center - Write-Ins Removed and Not Counted
At the Tally Center, write-in ballots are apparently removed permanently prior to scanning rather than scanned first and processed later or quickly processing and uniting them with the rest of the precinct. It makes reconciling the number of ballots problematic because there is no notation as to how many "write-ins" are removed nor what is done with them. Approximately 9,000 ballots appear to be removed at this point. The scanners do tally the number of write-ins and there are many write-ins that were noted on the scanner tapes. We consider this an open issue as we do not have adequate information to understand exactly what happened, and the number of ballots -- 9,000 -- is certainly significant. We are not sure how the write-in ballots, received on election day, are processed. Only 146 ballots were removed and documented as "damaged" and the balance of nearly 9,000 ballots removed at this point are not documented.
Ballots should not be taken from the boxes prior to scanning. Currently, many counts do not match (9,000+ ballots missing), but are magically corrected later. See [[#IssueA026}[Issue A026 - No Reconciliation Procedure]] for details on how this should be reconciled.
- Issue A014: Question to Ro V
Why are write-in ballots removed prior to scanning rather than scanning first and processing later or quickly processing them and uniting them with the rest of the precinct. Is there any logging of the count of ballots removed in this manner for each precinct and the disposition of each one?
- Issue A014: Legislative
Mandate that elections officials maintain a correct count of all ballots from each precinct and maintain these ballots as a group, without removing write-ins prior to scanning. Any ballots removed for inspection by elections officials should be logged. The precinct should be considered fully counted until these ballots are included in the count. The PEMT should included these ballots in the tally (They are not included now). The scanners in use by the Ro V can detect and report if a ballot has a write-in for a race, so there is no need to separate these in advance.
Issue A015 - Tally Center - Damaged Ballots
On the Tally Center Log sheet, there is a checkbox for "Damaged Removed." However, we don't know why any ballot would become damaged and not be treated as "spoiled." A spoiled ballot, by definition, is one that is filled out incorrectly or damaged and a new ballot is issued to the voter. Why would there be any ballots that are "damaged" and a new ballot not issued to the voter? However, 146 ballots were separated as "damaged" instead of "spoiled" in the November, 2008, election. (But another 9,000 ballots were removed as well.)
- Issue A015: Question to Ro V
146 ballots were separated as "damaged" instead of "spoiled" in the November 2008 election. When are ballots considered damaged and not spoiled? Why were another nearly 9,000 ballots removed that were not accounted for as "damaged" in the Tally Center Log Sheet?
- Issue A015: Legislative
Law should mandate that ballots for a given precinct should be counted and included in the canvass without separating out write-ins and damaged ballots, and if they are separated, this should be clearly documented. All ballots for the precinct should be included in the PEMT.
Issue A019 - No Comparison of Ballot Count
There is no comparison of the number of ballots scanned, as reported on the report tape, with the number of ballots counted at the precinct nor by the count as read by the central tabulator. Comparing the count of ballots as counted by polling place workers with the number scanned can eliminate the problems documented by the PEMT - One Percent Manual Tally, i.e. missing ballots. Only 4 of 85 precincts in our 5% sample scanned the number of ballots (Ballots Cast) that were counted by precinct workers. Comparing the count may have eliminated this problem (although this problem is more likely due to the removal of write-ins.)
Workers do not routinely compare the count of ballots, in terms of the actual count and the count reported on scanner report with the count of ballots input to the GEMS central tabulator. Our project reviewed the count of ballots printed on scanner tapes with the audit report in our 5% sample and found no discrepancy. However, there have been recently reports of problems in Ohio using the same equipment. Ohio Voting Machines Contained Programming Error That Dropped Votes (Washington Post)
This article said: A voting system used in 34 states contains a critical programming error that can cause votes to be dropped while being electronically transferred from memory cards to a central tallying point, the manufacturer acknowledges. The problem was identified after complaints from Ohio elections officials following the March primary there, but the logic error that is the root of the problem has been part of the software for 10 years, said Chris Riggall, a spokesman for Premier Election Solutions, formerly known as Diebold.
- Issue A019: Question to Ro V
In future elections, we request that the Ro V maintain an accurate ballot count (i.e do not remove write-ins and damaged ballots without noting the count of ballots removed) and compare that with the count of ballots actually scanned. This should eliminate the 29% failure rate as detected by the PEMT.
Related issues that are described in full in other Categories of Issues of Concern
- Issue A019: Legislative
Election law should mandate that election processing procedures include a step that compares the count of ballots as counted manually and the count of ballots scanned for that precinct by electronic scanning equipment as soon as the scan is complete. This should eliminate the 29% failure rate as detected by the PEMT.
- Issue A008: Ballot Statements - Office Use Section Not Explained -- Ballot Statements have an "Office Use Only" section does not include an explanation to allow the public to understand how it is being used.
- Issue A022: VBM Ballots not precounted during scanning
We were told that the scanner operator pre-counts the ballots as they run them, but this did not occur.
- Issue A004: No Ballot Inventory Report -- No report is generated regarding whether any errors occurred during SPUW, including the count of ballots.
Next section: Audit Log Discrepancies