Comments by Jerry Berkman to
Post Election Audit Study Working Group

Introduction

I do not have reason to believe any of our current elections officials are dishonest. On the other hand, Monterey County Registrar Tony Anchundo, said in 2005:

"There is obviously going to have to be some trust and faith in the elections official, or in this case, it's me."

The Auditor

The Time Problem

1% Manual Tally

• Safe precinct selection - In the interest of transparency, many Counties are switching to using 10-sided dice or selecting ping pong balls as a way to publicly select random precinct to tally. In Alameda County, since precincts are not numbered sequentially starting at 1, the random number chosen is used to select the precinct via a table, e.g.:

  Random Number	Corresponding Precinct Number
  001	2001
  002	2002
  003	2003
  006	2006
  etc.

  This method assumes all precincts are listed in the table. However, I doubt if any observers check this. With a preliminary statement of votes cast (SOVC), it is not even possible to check this. Even if an SOVC is available, checking the table against the SOVC would be difficult for a large County. This vulnerability strikes me as an easy way for an elections official to thwart the manual tally, just by leaving a precinct out of the table.
• Complete canvass beforehand - All parts of the canvass other than the 1% manual tally must be completed before any part of the manual tally is begun. All absentee ballots and provisionals must be fully processed and the results of the canvass must be published before beginning any part of the manual tally, including random selection of precincts.
• Publish results of canvass beforehand - The results must be published on the Internet as a CSV (comma separated values) file or in another format which is easy to feed into to common spread sheet software. PDFs do not meet this criterion. The results must specify the totals separately in all the categories that are tallied separately. This must be available before selection of precincts for the manual tally. The number of overvotes, undervotes, voided DRE ballots and blank ballots should be published by precinct on the web. The canvass results in Alameda County are over 5000 pages, so printed copies would not be a good alternative for a large county.
• Publish results of 1% tally - The results of the manual tally must be published on the Internet including a description of discrepancies and the reasons for the discrepancies before certification.
• Follow selection immediately by tally - The selection of the precincts should be held at the site containing all the ballots, and the audit should begin immediately after the selection. (In Alameda County, the selection was done on Friday afternoon so that staff would have the weekend to get the ballots from the warehouse. That is both not transparent and depends on more trust of the elections official and their staff than is necessary and desirable.)
• 1% Manual Tally Procedures - The Elections Code does not define "manual tally" and does not specify the procedures to use for it. EC 15277 does specify a procedure for a "manual vote count", but nowhere in the Elections Code is that related to a "manual tally". However, from the descriptions by Registrars Deborah Seiler, John Tuteur, and Cathy Darling at the public hearing, it is clear the procedures in EC 15277 are not followed in Solano, San Diego, Shasta Counties, and Napa Counties. And it was not followed when I observed in Alameda County.

  In Alameda County, in May 2007, the Board of Supervisors passed a resolution requiring the Registrar to "conduct a 100% recount of the Direct Recording Electronic (DRE) votes cast in the November 2006 election." (http://www.acgov.org/board/bos_calendar/documents/min06_08_06spmtgworksessionROV_BudPresentation.pdf) The intent, we thought, was to check the machine totals matched the votes as verified by the voters. But instead of having the staff read the ballots, the Registrar had them use bar code scanners and software for the comparison.

  Since "manual tally" is undefined, I think it is important to specify the manual tally for VPATs involves humans reading the voter verified text. It may also be useful to standardize allowed procedures.

• Tally all ballots by precinct - EC 15360 (b) allows absentee ballots cast on DREs (early voting ballots) to be included in the manual tally by DRE, instead of by precinct. This leads to composite figures and no one will have any idea if the figures for such a DRE are reasonable or not. E.g., if the preliminary results for my Berkeley precinct show 50% Republican votes, that would be pretty suspicious. But if a DRE from early voting shows 50% or 40% or 60%, we would have no idea if that is reasonable because the area covered is unknown. Also there is no published SOVC (statement of votes cast) which gives totals by DRE to compare against. All ballots must be included in the manual tally by precinct.
• Transparency - The process is not transparent. In Alameda County, the observation area was more than six feet from the nearest row of tables used in the manual tally, and about thirty feet from the last row. I could not see or hear anything meaningful, even as an official EC 15004 observer. At the end of the tally, we asked about tallying the VPATs. We were told they had been done. So either they were done while no one was observing, or out of sight at the far side of the room. In some other Counties, e.g. San Francisco and Yolo Counties, observers can walk around and actually see what is going on, but not in Alameda County.
• Accuracy of optical scanners - We are often told the manual tally of optically scanned ballots matches exactly the machine count. After reading papers by Prof. Douglas Jones, University of Iowa, I don't really think this is likely with paper ballots marked by voters. (see http://www.cs.uiowa.edu/~jones/voting/)
• Comparability - Currently, I don't think the totals from the manual tally are comparable to any other numbers, e.g. to the totals on the SOVC or to the totals on the poll tapes posted outside the precincts on election night after the closing of the polls. If I am correct, this makes the tally much less useful in verifying that the results of an election are correct.
• Provisionals - In some Counties, provisionals are cast by machine. In others, by paper ballots. In few are they included in the manual tally. The Election Code still does not explicitly say whether provisionals are included in the manual tally. A Secretary of State interpretation or regulation is desirable.

Duplicating/Enhancing/Replacing/Converting Ballots

I don't know for how many ballots these procedures are used. I don't think this is documented, and it should be with the statistics posted on the Internet.

I believe most Counties use two workers to check each other when duplicating damaged ballots. I'm not sure if that is the case for all Counties or whether it applies to all the types of duplication/enhancement. I don't see where witnessing is required in the Elections Code (except for EC 15211 in the case of duplicating paper ballots onto punchcard ballots!) And certainly, the 1% manual tally does not check this was done correctly.

Here are the categories I know of where the original unaltered ballot is not checked in the manual tally:

• Remakes - Paper ballots which are torn, bent, or otherwise defective are remade/duplicated according to the provisions of EC 15210 (including using a serial number to correlate the remade ballot and the original ballot). John Tuteur, County Clerk of Napa County, testified in the public hearing that they had to remake almost one fourth of the paper ballots in the last election.
• Enhancements - Sometimes a paper ballot is not scanned properly because the voter did not mark the ballot dark enough, or used a red pen when red is not scanned properly, or circled the bubble instead of filling it in. Via a liberal interpretation of EC 15210, many Counties (all?), these ballots will be fixed by staff. However, EC 15210 only requires the ballot to be "corrected" so that it is read properly. It does not require a remake unless "necessary". Some Counties will enhance these ballots by filling in the bubble, instead of duplicating the ballot. Some also use white-out to cover up stray marks interfering with the scanner reading the ballot.
• Conversions - In a County which has only DREs in the polling place, a voter who does not trust DREs may insist on a paper ballot. In some Counties, the staff will then later enter these ballots via a DRE! (This was done in Riverside, see: http://www.votetrustusa.org/index.php?option=com_content&task=view&id=1423&Itemid=60)
• Emergencies - EC 19005 provides that in an "electrical failure or other emergency", ballots may be marked by pencil or ink, and the elections official may later "duplicate" the ballots.
• Special absentee voters - According to EC 3103.5, special absentee voters (e.g. overseas military voters) may return their ballot via FAX and it is then duplicated by the elections official's staff. This looks like a limited time experiment, as it expires in Jan. 1, 2009. I do not know if anyone used this section.

Items Not Audited:

• Totals - The 1% manual tally only checks the results for a few precincts. The totals for the County are not audited.
• Central tabulator and central optical scanner logs - Appropriate logs (e.g. audit logs, system logs, and Windows event logs) should be audited on central tabulators and central optical scanners. When this was done in Arizona, it showed apparently illegal activity - printing summaries reports before 8 p.m. on election day. (http://www.bbvforums.org/forums/messages/1954/46569.html)
• DRE and PBOS tabulator logs - Appropriate logs should be audited on a random selection of DRE and PBOS devices used in precincts in the election. BlackBoxVoting sued for logs in Florida, and found 100,000 errors in the logs on a small number of machines and found 1,475 calibrations performed while the polls were open. (http://www.bbvforums.org/forums/messages/1954/19421.html) This type of information must be available.
• Software audits - EC 19223 directs the Secretary of State to "conduct random audits of the software installed" on DREs to make sure it the approved software. This should be part of the audit. Although fraud-proof verification is difficult, any tests would be an improvement and would cause the vendors and Counties to take notice.
• Precinct materials - Are the rosters audited? Are the required signatures on the materials checked against expected signatures?
• Return centers - I'm not sure this is in the scope of the working group, but I'll add it in anyway. The Elections Code allows Counties to set up multiple return centers and counting centers. Election materials are driven by poll workers, at least two per car, to the return or counting centers (EC 15202). BlackBoxVoting has an interesting video showing potential for fraud at a receiving station in King County, Washington. (http://www.bbvforums.org/forums/messages/2197/14299.html) The following should all be documented and audited:
  • chain of custody of election materials coming into and leaving the return center,
  • who is working at the return centers and who else is present,
  • and who drove the materials from the return center to the counting center. (this can be only one person per car!)
• Chain of custody - The chain of custody of ballots is not audited. Actually, I doubt a complete chain of custody is kept, especially on Election night. I suspect that the precinct materials are checked in, but then I doubt each person processing the materials has to sign for them.
• Signature verifications and rejections - These needs to be auditable and audited.

More Ideas

One possibility would be to split the federal election canvass and especially, the manual tally of the federal election, from the same for non-federal races. This would create other problems for elections officials.

Tiered Audits - Tiered audits would help, for very close races and for landslides. For very close races, the manual tally could be expanded. For landslides or uncontested races, perhaps not a full 1% manual tally is needed. This would require changing the Elections Code which requires at least 1% of the precincts to be recounted.

Same day registration - Provisionals create major problems and are distrusted by most people. Same day registration would eliminate most provisionals. Same day registration is used successful in several states, e.g. Minnesota and New Hampshire, with few if any problems. We should consider this for the future (it obviously requires changes in the Election Code).