e-Democracy Symposium – Recounting/Auditing By Machine – The Devil Is In The Details

Yesterday I attended the spring symposium at the UConn Law School by the Connecticut Public Interest Law Journal: e-Democracy: Democratic Values In A Digital Age. There were three panels:

  • Campaigning On The Web
  • The Mechanics of Voting
  • Political Speech and the Internet

I found all three very interesting. Christine Stuart has an excellent article on the first panel at CTNewsJunkie I completely agree with her highlighting of Tim Tagaris, statement: “It’s Americans with computers, [not 20 somethings]”. The other point that rang true to me was that political blogging leads to personal interaction not isolation. I learned of the panel based on personal contact that was the result of political blogging, that was the result of personal contact that was the result of blogging…

The third panel discussed laws and regulations concerning political contributions, blogging, and the relationship of bloggers to campaigns. The summary was that political blogging is not regulated or restrained because, unlike television, it is open to anyone and thus does not need such regulation. I asked if the panel agreed that all their conclusions were based on “Net Neutrality”? Matt Stoller gave an impassioned response articulating current erosion of net neutrality.

The Second Panel: The Mechanics of Voting

The second panel covered the risks of electronic voting. Some of the issues we cover regularly here. The panelists were:

  • Lawrence Norden, Brennan Center for Justice, lead author of several significant reports on electronic voting. Mr. Norden also has worked to cause New York to make better choices in electronic voting.
  • J. Alex Halderman, Ph.D. candidate at Princeton University, and participant in the Diebold portion of the CA Top-To-Bottom Review.
  • Senator Gayle Slossberg, Co-Chair of the Connecticut General Assembly Government Administration and Elections Committee, and one of the authors of our audit bill, PA 07-194.

I appreciate all the panelists and their work for democracy. Also in attendance for the panel: Deputy Secretary of the State, Lesley Mara, several registrars, and more than several town clerks. I also appreciate their taking the time to be involved and informed. I was on the edge of my seat for the whole panel, wishing for more elaboration on some of the points, wishing it was more of a debate that I could join in. I fear that the lack of time, lack of common understanding, and background information between the panelists, and between the panelists and the audience may have left some significant misimpressions. Since I don’t have a transcript, you will have to go with my summary of what was said and of course my own impression of what might have been misinterpreted:

  • I completely agree with Mr. Norden’s statement that there is a lot we can do with what we have in time to improve the security of elections in time for November 2008.
  • I completely agree that Connecticut is much better off with optical scan than we would be with DRE (touch screen) voting equipment. I also agree that we have procedures that are better than those in many other states.
  • However our procedures are woefully inadequate to protect us. A chain-of-custody is only as good as the weakest link. Our procedures are inadequate, by their nature procedures are unenforceable, and have not been enforced. They have been regularly violated in the custody of ballots, audits, and memory card security.
  • Our 10% selection of districts for our post-election audits provides an easy claim to the strictest audit in the Country. That is misleading since the 10% is followed by a secret drawing of three or 20% of races to be audited, the law is full of loopholes, is articulated in inadequate, unenforceable, frequently violated procedures, and has proven in practice to provide no confidence that an error or fraud would be recognized. As Mr. Norden pointed out 10% can be overkill in some cases, unfortunately our audit law for the most part falls far short of the Brennan Center’s standards. Even that 10% is misleading since it is followed by a secret random drawing of three or 20% of races to be actually audited. And further while 10% for a state wide race would often be overkill, it is woefully inadequate to provide deterrence in state legislative races and most municipal races.

Recounting By Machine – The Devil Is In the Details

Once again, as has been the case since the November election we heard calls from officials for recounts by machine rather than hand counting. We also heard, once again, from Senator Slossberg that the Legislature may consider changing from manual recounts to machine recounts.

Here is where I saw the possibility of a large gap in understanding and context that may lead to misinterpretation of some of the remarks of the speakers.

In answer to a question of mine to Mr. Norden, where I asked how recounts by similar machines with similar memory card could discover problems, he answered that is was possible in cases where the machine used in the election was adjusted incorrectly or scanning incorrectly. I fear this may have left an impression with some that recounting by machine was fine. However, his answer, while technically accurate, has no relevance to catching errors or fraud having to do with software in the computer or in the memory card. Those would not be detected by running the same votes through properly setup, identically functioning computers and memory cards. In addition, if a variance were found based on an improperly functioning scanner, the question would then be which machine was more accurate – the original or the second one?

In his opening remarks Mr. Halderman mentioned that is was possible to recount and audit by reading votes through a new scanner and manually counting some votes, while saving huge sums of money, and in some cases reducing the actual hand counting to as low as 1 vote in 10,000. I agree. Yet, how many in the audience understood that he was talking about a completely different type of scanner, one that is not available commercially today. Yes, we could count and audit at a much lower cost with much more confidence – in fact the same type of technology could likely be used for the initial counting and avoid the second scanning (but the not the manual sample) altogether. The important point to understand is that this is not what is on the table in Connecticut at this time, when election officials, the Secretary of the State and the Legislature discuss recounting by machine.

ComputerWorld Interview With Rep. Rush Holt

Representative Rush Holt (D-NJ) is a physicist who has been working for some four years to provide integrity and confidence in voting. While waiting for a longer term solution, he has introduced an emergency bill, H.R. 5056 to encourage states to move to paper ballots and sound audits in time for the 2008 presidential election. We strongly support the bill. Rep. Holt is interviewed by ComputerWorld: Q&A: For e-voting, Holt looks to undo HAVA’s havoc <read>

Scientists and engineers would look all look at these systems and say, “You need a paper receipt that verifies the vote.” A politician on the other hand is slow to recognize that software can’t verify itself. The politician in me saw that [the lack of ability to verify votes] was a critical flaw in our voting systems, which could undermine confidence in the entire voting process…

If there is a voter-verified paper ballot, that ballot belongs to the election system. The voter doesn’t carry it away. It’s kept under lock and key the same way that all election records are kept. It’s subject to inspection only with a bipartisan, duly formed group of witnesses…

In presidential and senate races, if there is any county of the state in doubt, then the whole state and even the whole election is in doubt. We haven’t even finished setting a national standard that would apply to all votes cast in federal elections. Right now, all we can do is encourage as many counties as possible to do the right thing, give them some incentive, and provide reimbursement for their cost of using verified paper records and audits…

The legislation that I would like to see would have chain-of-custody requirements and transparency of software so that the software would be available for independent people to check. But the best single thing we could do is to have an independent audit of randomly chosen precincts in each federal election. An audit will be the most direct, simplest way of uncovering problems even if there is a software error, be it innocent or malicious.

TalkNationRadio: Dr. Shvartsman Outlines Memory Card Failures

We have covered the memory card issues and the reports from Dr. Shvartsman’s team at the UConn VoTeR Center <here> and <here>. This week, Dori Smith of TalkNationRadio interviewed Dr. Shvartsman <audio and transcript>.

The UConn reports reported 3.5% and 8% failure rates. These are outrageous. Not only do they represent a huge and unacceptable hardware failure rate, but this is after LHS was to have programmed and tested each card. But this may be the tip of the iceberg, what kind of organization is LHS?

The Secretary of the State made public announcements that the University of Connecticut team planned to collect memory cards for a study into the failure rates of the cards. But even so, LHS staff members collected memory cards that failed during the set up and testing phase of the 2007 election. They did this even as the UCONN team was attempting to collect a random sample to test for the number of failing cards and the manner in which they failed. The actions of LHS of replacing failing memory cards during the set up of the election is not outlined in state security protocols and those actions served to remove failing cards from the UCONN study, it ruined portions of the statistical analysis.

Continue reading “TalkNationRadio: Dr. Shvartsman Outlines Memory Card Failures”

Press Conference: Bysiewicz “Tried To Rest Lingering Skepticism”

Christine Stuart has an accurate report of yesterday’s Press Conference and the post-election audit discussion: <read>

Asked why the machines tend to over count by one vote, Dr. Alex Shvartsman, director of the voter research center, said he didn’t know. “That’s a very good question,” he said, adding that the center would continue to look into the results.

But Secretary of the State Susan Bysiewicz was quick to step in an answer the same question for Dr. Shvartsman as she tried to put to rest any lingering skepticism about the results. She said that the machine overcounts may have happened in races with multiple candidates when voters unintentionally marked an additional bubbles on the sheet.

For example, a voter may have wanted to vote for three of five candidates, but accidentally made a mark in a fourth bubble, Bysiewicz said. In such a case, the machine counts the vote, but a poll worker conducting a hand recount may not count the additional vote because to them it appeared to be an inadvertent mark or smudge, she said.

Another issue was that 175 audit reports submitted to the University of Connecticut were “incomplete, unuseable, or obviously incorrect.” About 70 percent of the 958 reports submitted by 70 polling places were complete. The center decided to use about 783 of the reports to complete their audit.

The audit found that 66.4 percent show a discrepancy of 0 to 1 vote between the machine counts and hand counts; 89.4 percent show a discrepancy of 5 votes or fewer; and 31 records, or 4 percent, show a discrepancy of 10 or more votes.

Bysiewicz said overcounts happened in races where candidates were cross-endorsed by two parties.

Lingering “Skepticism”:

  • The machines do not “tend to overcount by one vote”. They tend to count very accurately in most cases. That “one” is an average that can be misleading. I was sitting on the edge of my seat waiting for a reporter to ask what the top counting errors were. The Coalition report and Dr. Shvartsman’s had about the same numbers, his overcounts were: 72, 54, 28, 26, 26, 22, 16, 14, 14 In our report we did not include the 72 because it was an obvious counting error.
  • Checking after the press conference, the analysis by the Secretary of the State’s office is not yet complete – they have not checked all 31 races to see how many were candidates listed on more than one line – the overcount of 72 votes was due to misclassifing votes of a candidate listed on two lines and was offset by an undercount of 74 votes on another line. (that is why we did not list it as a discrepancy in the Coalition report)
  • Finally, even if all of the 31 highest overcounts were candidates listed on more than one line, it is not in itself an indication that they were all due to human counting errors – some could be due to election programming errors – without further research the cause would all be speculation.

Post-Election Memory Card Tests, 8% “Junk Data”

Last week Dr. Shvartsman of the UConn VoTeR Center released a report on Pre-Election testing of memory cards, that report showed 3.5% of the memory cards had “Junk Data” (the cards could not be read) and that many election officials had difficulty following pre-election testing procedures.

This week Dr. Shvartsman released another Post-Election study of 100 cards collected for evaluation after the election. These cards were not randomly selected. The results were similar to pre-election testing with more “Junk Data” cards at 8%. <read>

Once again we appreciate the work of the Secretary of the State’s Office and Dr. Shvartsman’s team in implementing this program. We have much sympathy for election officials who had difficulty complying with the program this time as procedures were added close to the election. We have no sympathy for LHS Associates. Two things seem clear:

  • We should expect that compliance with procedures by election officials will improve dramatically over each of the elections in 2008, making this program a unique and valuable addition to election security
  • We should be very very disappointed with the performance of LHS in quality control — with years of experience programming elections in other New England states — they should have met commitments to Connecticut to provide tested cards to Connecticut municipalities. There is no reason to give them a pass for the November 2007 election.

The report was released at an 11:00am press conference at the Secretary of the State’s office. Also available at the conference was a yet to be posted analysis of the post-election audits by the VoTeR Center. The data presented was generally consistent with the data compiled and reported by the Coalition. We will cover that report when it is released in its final form.

Bysiewicz Prepared To Eliminate Manual Recounts

Update: The word “prepared” does not not necessarily imply “will”.   Apparently no decision has been made.  However, we now have a member of the GAE committee who has suggested machine recounts, the Secretary has mentioned it before, and the GAE leadership mentioned the possibility at the press conference announcing public hearings.  This remains a bad idea which we hope will be quickly rejected.

*****

The Secretary of the State of Connecticut is prepared to discontinue manual recounts in close races. In an e-mail to advocates, good government groups, and the leadership of the Registrar of Voters Association, Deputy Secretary of the State Lesley Mara wrote:

From: Lesley Mara [Deputy Secretary of the State]

Subject: Memory Card audit report

I have received questions from some of you regarding recounts. As you all know, our regulations regarding optical scan machines call for a review of every ballot in the event of a recount. Where all parties agree that a ballot is clearly and properly marked in a way intended to be read by the scanner, that ballot is put in one pile. All “questionable ballots” are placed in another pile and, of course, are hand counted. Those properly marked are run through a new scanner. This procedure was used most recently in the Courtney/Simmons recount.

The Secretary exercised her authority to specify another procedure this past November during the first statewide use of optical scan technology, given that we had not finalized all of the pieces we now have in place with [the University of Connecticut] UCONN (e.g., post-election audits of memory cards). Since November, we have spoken to legislators and election officials and UCONN and we are prepared to return to the original procedure. Of course, we welcome your thoughts on this issue as well when we meet tomorrow.

Continue reading “Bysiewicz Prepared To Eliminate Manual Recounts”

96.6% Of Memory Cards OK, Half Of Registrars Follow Last Minute Procedures

At the request of the Secretary of the State, Dr. Alex Shvartsman and his team at the UConn VoTeR Center tested memory cards which were shipped to UConn by registrars across the state. <full report>

From the summary:

The total of 522 cards were received and tested by the VoTeR Center, out of which 378 cards were received before the election. Out of the total number of cards, 18 cards, or 3.5% were found to contain “junk” data, that is, they were unreadable, which is easily detected by the tabulators as such, and could not have been used in the election. The rest of the cards, or 96.6%, were found to have been properly programmed for election. These cards contained valid ballot data and the executable code on these cards was the expected code, with no extraneous data or code on the cards.

About half of the cards were found to have been tested and set for election-the intended state of the cards following the prescribed testing procedure. Most of the remaining cards were tested by the towns but not set for election; while this is not a problem, this suggested that the relevant towns/districts either misunderstood the instructions or did not follow the instructions. One card was found in the state set for election but with non-zero counters, indicating that the district tested he card in election mode and did not reset the card. This is a potentially problematic, but detectable situation, since proper procedures require that the “zero counter” report is produced at the start of the election

Full Disclosure: I played a role in suggesting this program to the Secretary of the State’s office at a meeting in October. Professor Michael Fisher of Yale and TrueVoteCT greatly improved on my suggestion. Continue reading “96.6% Of Memory Cards OK, Half Of Registrars Follow Last Minute Procedures”

Legislature To Hold Public Hearings On Optical Scan Elections

The leadership of the Government Administration and Elections Committee(GAE) announced a series of five public hearings at a press conference yesterday. One hearing will be held in each of the five congressional districts in the state. Towns where the hearings will be held were announced with no specific dates or locations. Their goal is to have the hearings during the 2nd and 3rd week of February. The towns will be West Hartford, Norwich, East Haven, Norwalk, and Danbury.

I attended the press conference but did not take verbatim quotes so I will summarize.

This is very good news for several reasons.

  • The GAE clearly recognizes that the public as well as registrars have concerns.
  • The GAE indicated that they have heard from registrars but also want to hear more from the public.
  • The areas of concern recognized so far include: registrars’ concerns for time and cost of recounts and audits; voters concerns with election day privacy, with the optical scan machine integrity, and with the integrity of the audits. (our coalition report was mentioned in the press conference)
  • The GAE is open to hearing the public and, if warranted, changing the law in the short session.
  • It is a true bi-partisan effort supported by the committee co-chairs and the ranking members. Democrats Caruso, Slossberg, and Urban. Republicans Freedman and Heartherington.

I was also pleased with the questions from the press, asking about Connecticut vs New Hampshire, and the outsourcing of elections to LHS. The GAE members said that we may well learn from the experience in New Hampshire.

On the other side of the ledger, the GAE is open to considering an optical scan based recount in close races. They say it is only one suggestion and they are not at this point recommending it. We have seen that the machines usually count accurately, yet they also frequently count inaccurately both in New Hampshire and in Connecticut. We certainly hope this idea is quickly abandoned in favor of physics, science, and reason.

Update: Stamford Advocate Story <read>

“[Secretary of the State Bysiewicz] Susan’s been the biggest cheerleader for this system and at times I don’t think she’s been able to step back and look at it objectively (and) hear some of the problems that are occurring,” Caruso said.

TalkNationRadio On NH, CT, Outsourcing, and Audits

Dori Smith interviews Brad Friedman, Dr. Alexander Shvartsman, and yours truly. She talks to Brad about the recount in NH, outsourcing to LHS, the chain of custody in NH, and implications for Connecticut. She talks to Dr Shvartsman about NH procedures, vendors on election day, and his soon to be released report on memory cards in Connecticut. She talks to yours truly about the Coalition report on the results of our observation of the recent audits in Connecticut. The election report starts about 10 minutes into the broadcast. <summary text and link to listen>

Correction: What Do Elections Cost?

I recently wrote elsewhere that:

  • A good audit costs $.25 to at most $.50 per ballot cast – (less than the cost of one mailing by a candidate or one of those newsletters from our state representatives and senators)
  • The costs of running an election is over $3.00 to $20.00 per ballot cast
  • Candidates spend $millions of our money to get our votes
  • Special interests spend $millions more, some say, to get their votes
  • The costs of the wrong candidate being elected involves innocent lives and millions in Connecticut, $trillions in Washington, and thousands of lives in war
  • The value of integrity and confidence in our elections? Priceless

Apparently the $3.00 per ballot cast may be way low for Connecticut. Continue reading “Correction: What Do Elections Cost?”