Wednesday, October 29, 2008

A real-life Zelig

Zelig is a Woody Allen film about Leonard Zelig, a "human chameleon" who shows up (thanks to very clever editing) in all sorts of historical places. There are echoes of the idea in Forrest Gump (better known for the line "life is a box of chocolates").

Robert Furman, age 93, died last week. He was a real-life Zelig - as a young man, he supervised building the Pentagon, helped bring scientists to Los Alamos, tracked German scientists like Werner Heisenberg across Europe during and after World War II, and worked with baseball player turned spy Moe Berg. When the war was over, he didn't speak of his involvement but instead returned to a quiet life, eventually becoming a builder of shopping malls.

An obituary well worth reading. And a man I wish I had known.

[For a fascinating biography of Moe Berg, read "The Catcher Was a Spy: The Mysterious Life of Moe Berg." It mentions many of the same incidents listed in the obituary, with more details, although it disputes the claim in the obit that Berg spoke seven languages.]

Friday, October 10, 2008

Proud papa

I usually write about technology topics. But today, I have to shep naches: my son Daniel spent the summer on a research program at the Weizmann Institute in Rehovot Israel, and yesterday an article about his summer experience appeared in the local newspaper. Makes a father proud!

Official pollworker training

Like many technical professionals involved in the voting world, I've decided to become a pollworker. I almost wrote "volunteer", but it is paid in Virginia - $100 for a 16 hour day (and no pay for training)! I live in Fairfax County, which is the most populous in Virginia, and so probably a "best case" in terms of organization and technology.

Last night I went to pollworker training. There are four "levels" of pollworkers in Virginia - the pages, the ordinary grunts, the assistant chief, and the chief. The pages are high school students who get credit for helping (all schools in Fairfax County on election day); they can do limited tasks. The ordinary grunts, like me, can do most of the jobs that don’t require exception processing (such as dealing with machine failures, or provisional ballots). The assistant chief and chief seem to be generally interchangeable, and have the responsibility for oversight of the whole thing (setup, opening, voting, and closing), and handling all of the exceptions that occur. [Assistant chiefs are paid $150 and chiefs are paid $200/day - showing that these are very dedicated individuals.]

I was quite impressed by several things at training:

  • The folks doing the training are well organized. Last night was their 25th training session this year; they're planning on about 60 in all.
  • All pollworkers are required to go through training this year, even if they've worked before, due to the change in equipment (described more below). Having said that, about half the people in the room last night were first-time workers like me.
  • The average age of the pollworkers was well below the widely-reported national median of 74. In fact, I'd guess it was two decades younger than that. There were a few people in their 30s; most looked to be in their 40s or 50s. That seems to be a good sign.
  • The county is providing 104% ballots - that is, 104% of the number of eligible voters. That should (hopefully!) ensure that we don't run out of ballots, even accounting for spoiled ballots.
  • If a voter spoils more than about 3 ballots, they are encouraged to switch to the DRE. That makes sense, as they're having trouble following the instructions.

Presidential elections in Virginia tend to have short ballots, since we elect state and local officials in odd-numbered years. So the ballot in Fairfax will have only four items: president/VP, US Senate, US House, and one bond issue. That's good, as the expectation is for very heavy turnout.

For the past 5 years, Fairfax has been a DRE-only county, except for absentee voters. This year, we're going to a hybrid voting system this year - Premier AccuVote-OS optical scan and AVS WinVote DREs. Voters are being encouraged to use the optical scan (we're to offer an optical scan ballot, and a voter has to explicitly request to use the DREs). Precincts in Fairfax vary greatly in size from about 500 voters to over 5000. Depending on the number of voters, there will be 1-3 AccuVote readers and 3-7 DREs, plus 3-15 "privacy booths" (cardboard box dividers) for use in marking the optical scan ballots, and 10 plain old clipboards for people who don't want to wait for a booth.

The training mentioned several times that turnout is expected to be very heavy all day long, and pollworkers should vote early at one of the central locations in the county (the polls are already open) or by absentee. I hope they're right - I get very frustrated by low turnouts, especially at important elections like this one.

There are only three critiques I have of the way the election is being run in Fairfax County.

  • The AccuVote readers are being set up to reject overvotes (selecting more candidates for a race than are allowed), although the voter can, with assistance from the chief, override that and cast a ballot anyway. However, they're not set up to even give a warning if someone undervotes - for example, forgetting to vote for Senate. Voters shouldn't have to vote for all races, but they should get a warning if the machine doesn't detect a vote. This isn't a fault of the machines; it's how they've been set up.
  • There's no minimum number of votes to be cast on DREs (the trainer thought I meant maximum and I had to explain why a minimum was also a concern). As I expect most people will select the optical scan, there's a risk that if there are only one or two votes cast per DRE, the end-of-day totals will reveal those voters choices. My preference would be to use the first DRE for the first five (or so) votes, then switch to the second, then to the third, etc., until all machines have at least five votes - and after that it doesn't matter which ones get used. [The pollworkers should have the discretion to note that there's almost no one voting on the DREs, and not to even use more than one or two if the load demands it.]
  • There was no training on inspecting or watching for physical security issues. Perhaps this is part of the chief/assistant chief training, but I expected to hear something about watching out for evidence of tampering with seals before, during, and after the election.

The county election officials seem to have come to the conclusion that at least one part of our arguments against DREs was right: they scale better. Even if they don't agree on the reliability & security issues, they now understand that more voters means more pencils, instead of more $5000 machines!

Virginia still has some serious election problems, including possibly the worst audit law in the country (it's generally speaking illegal to look at the paper ballots, except after all the election results have been certified, and then only if there's a margin of victory more than 10%), and a recount law that's almost as bad, but at least having the optical scan ballots is making it possible to do the audits and recounts if we can change the law in the future.

I'm looking forward to election day - sorry I won't be able to keep up on the minute-by-minute developments since I'll be working at the polls, but excited to be part of the process!

Watch for my post-election-day report on how things go...

Wednesday, October 08, 2008

DC Council releases their findings

As I wrote about yesterday, the DC Council is looking into what went wrong in the primary election last month. Today, the results of their investigation were released; the report can be found here. If I worked for the vendor, Sequoia Voting Solutions, I'd be unhappy - the council accepted the recommendations of the experts (including me), and generally rejected Sequoia's excuses.

The report concludes with three main recommendations.

1. The District should set up an independent voting-technology experts to perform a forensic study of what went wrong.

2. The Board of Elections should improve its procedures to do a better job detecting problems like those in the preliminary results, and address them before releasing the results.

3. The Board of Elections should set up effective post-election audits. (Kudos to Larry Norden from NYU for his excellent information in this area).

4. The Board of Elections should provide better training to pollworkers. Pollworker training is critical everywhere, and its importance is usually underestimated.

5. The Board of Elections should set up procedures to ensure a rapid but accurate release of preliminary results.

6. The Board of Elections should have policies in place for public communication in case something goes wrong.

I'm pleased with the results of the DC Council investigation. Now comes the hard work - putting these recommendations into practice!

Tuesday, October 07, 2008

The DC voting mess continues

Way back before the financial market collapsed (say, last month), the District of Columbia held a primary election. For reasons still unexplained, there were some very strange preliminary results - huge numbers of overvotes in one part of the city. The vendor, Sequoia Election Systems, has alternately blamed the problem on static electricity, errors by the Board of Elections, and reporting the results too quickly - but at all times have denied that it was due to a problem in their hardware or software. The DC Council established an ad hoc committee to examine what happened and try to get to the bottom of the issue.

There's been lots of coverage of the problem in The Washington Post: here (Sep 11), here (Sep 12), here (Sep 12 again), here (Sep 22), here (Sep 25), here (Sep 26), here (Sep 30), here (Oct 2 editorial), and here (Oct 4).

The last of the above articles is coverage of a public hearing held by the special committee, where they heard from interested citizens, the vendor, and several voting system experts. I was honored to be one of the experts invited to testify, and my testimony can be found about 54 minutes into the video.

I made four basic points:

1. We shouldn’t blindly take the vendor’s word for what happened, because the vendor's explanations don't make sense. Sequoia hasn't been able to reproduce the problems that correspond to any of their explanations, be they static electricity or mistakes by the election officials.

2. We shouldn’t blindly take the vendor’s word for what happened – the explanations may be wrong. We’ve seen at least one case where a vendor example where a vendor’s initial claims were proven wrong. After the spring 2008 primary in Ohio, the vendor, Premier Election Systems, blamed election officials for incorrect results, where one precinct’s votes were lost. Later, the vendor changed their explanation, and blamed the anti-virus vendor for the error. Eventually, the vendor admitted that it was a bug in their software. Right now Sequoia is saying it’s not a software error – but that might change as the DC Council and Sequoia learn more.

3. We shouldn’t blindly take the vendor’s word for what happened – prior studies have indicated that the problem seen in DC was a possible problem. In particular, the California Top To Bottom Review, sponsored by the California Secretary of State, noted that Sequoia software “seems not to check whether vote counts stored on Memory Packs received from the MPR are consistent with the number of voters, so an erroneous Memory Pack will corrupt the final tally instead of being detected”. The California report then goes on to note that the Sequoia software does not perform any sanity checks to ensure that each Memory Pack only contains data from a single precinct – so a corrupted Memory Pack that is somehow corrupted can not only affect that precinct, but can cause the erroneous results to propagate into other precincts.

4. We shouldn’t blindly take the vendor’s word for what happened – we’ve seen similar failures before with Sequoia voting systems. In Alameda County California, the February 2008 primary election had a very similar failure, where a corrupted memory card registered an absurd number of votes in one precinct, and reprocessing the memory card caused it to give correct results.

I then recommended that the Council obtain an independent expert forensic study to figure out what went wrong, along the lines of the investigations of the 2006 Florida CD-13 election or the 2008 New Jersey primary.

I don't know what's going to happen next, but I hope the Council moves forward quickly, so there can be some preliminary answers before the election next month!