I just finished reading Ian Ayres’ latest book, Super Crunchers: Why Thinking-by-Numbers Is the New Way to Be Smart. It was a fascinating read for me (but a fine introduction for the layman, as well). Through a series of anecdotes, Ayres provides the reader with a general overview of how data mining is changing the way everything operates, from web advertisement optimization to providing better medical care to impoverished areas of the world.

Critics contend that the type of models Ayres showcases have been around for decades (true), but that’s missing the point. Until now the access to large enough volumes of data necessary for randomization has been expensive (or in some cases) impossible. Data have become exponentially cheaper to collect and store. With the advent of modern data warehousing, data can be stored in a more general-purpose fashion. The processing power to effectively model and analyze large (and often disparate) sets of data, and test hypotheses not conceived at time of model conception is now widely available.

However, as Ayres points out, there is no need for humans to fear an end to their usefulness in decision-making. In many cases, these systems are making very arithmetic-intensive choices based on human selections and ratings.

To use an example from the first chapter: how would an accurate formula to predict the best Bordeaux vintages possibly be created without historical data on what humans think tastes best? More fundamentally, if a machine could formulate its own opinion on Bordeaux, would we care?

Yesterday we were featured on NPR's WYSO 91.3, in Yellow Springs, Ohio, with an interview by Cynthia Sperberg-Hart. I give a special thanks to her and the rest of the staff there for helping to inform voters about the importance of the secret ballot.

Listen to the interview here.

Look, I voted!

I asked the pollworkers to not record my order in the pollbook, and they told me that they weren't recording anyone's. I got excited and told them that I was partially responsible, but I'm not sure they understood ;-) For me personally, it was a very symbolic victory.

Today I voted, and my ballot was secret.

This evening, Steve Madsen contacted me again with some great news regarding Powell's decision to not record voter order:

To be fair, it wasn't my idea (though I was going to ask about it) and it was at least county-wide (Delaware). It may have even been state-wide. It's been a long day, but I believe they told us yesterday during presiding judge training that word came down from the Brunner's office Friday that the old poll books were out. Instead, we used a simple list of precinct voters and highlighted them as they came in. While we still used the numbered "authority to vote" slips, the numbers were not recorded anywhere.

I wish all my readers a fruitful and private Election Day.

Steve Madsen emailed me today with this exciting news:

I've been an election judge in Powell [Ohio] for the past couple of elections, and you'll be happy to hear that they've changed their procedures for tomorrow's election. We'll no longer be recording voters in the order they arrive, so vote matching shouldn't be possible this time around.

Although I haven't personally examined the arrangement, it sure looks like a victory for the secret ballot for Powell voters.

Although a a state-wide solution is key, Steve has demonstrated that in the meantime, local election officials and concerned citizens can still work together to make their vote secret in their own communities.

Mike Chapple and his students at Notre Dame are examining an anonymized version of our voter preference records, as well a computer program which I wrote in Ruby several months ago which automates the process of matching a pollbook record to a VVPAT record.

I have been in contact with the head of the Ohio e-voting security review (at SysTest Labs)to verify he had been made aware of the voter confidentiality threat which the ES&S iVotronic machines possess.

Miami University Alumnus Ryan Nowak has theorized that using an advanced statistical model could greatly increase accuracy at failure points (i.e. times which multiple machines undergo heavy activity). His proposal is very intriguing; we will be examining its feasibility around the middle of next week.

A Denver-based e-voting testing laboratory and experts from three universities have been hired by the state of Ohio to undertake independent evaluations of the states' e-voting hardware, software and processes. The move is aimed at finding and fixing potential problems before the 2008 presidential election.

No mention was made of our algorithm for determining how individuals voted, so I will be in touch with SysTest Labs as soon as my phone charges ;-) Read the full article here.

Hat tip to Pam Carney on this one.

James Moyer has taken part in an interview on us for Fox News. Check it out at 7p-ish EST, on The Fox Report.

There are two particular criticisms which have been raised several times, in various forms. Rather than continue to correct them at each turn, I'll just address them here.

Example one (from C|NET's coverage):

Assumption is flawed, with multiple voting machines and time taken to vote differences, matching sign in records to time stamped records doesn't provide proof of that individual's vote.

In fact, there are both beginning and ending timestamps for each vote entry, with down-to-the-second precision. Although this doesn't guarantee the accuracy of every decoded vote,

• it provides a high degree of accuracy, in the aggregate, and
• the degree of accuracy of any individual vote can be reasonably calculated (in other words, you'll know how right or wrong you are with each vote).

Example two (from ColumbusING):

Wouldn’t work. The way the voting process works essentially randomizes voters. Knowing the order in which people voted does not tell you what machine they used. Without that link (order w/ machine), the information is useless.

At any given time, each voting machine will have one Voter Verified Paper Audit Trail. For the machines discussed here, that is a roll of paper which contains the beginning and ending timestamps for each voting session.

James Moyer, my colleague in the voter identification project, has requested that his contact information be listed here for inquiries from members of the media, as well as concerned citizens.

Email: james@moyer.com.

Phone: (330) 242-3657

On behalf of Jim Cropcho:

I've been getting requests to discuss this issue, so I'd like to make my contact information more accessible.

Email: jim+publicballot@findholmes.com.

Phone: (419) 277-4081

Today is a very big day for The Public Ballot. We have are now on the front page of News.com (CNET's news service).

An in-depth article by Declan McCullagh, with quotes from Stanford Professor David Dill and Berkeley Professor David Wagner. Insight is also provided by Lorrie Cranor, director of the Usable Privacy and Security Laboratory at Carnegie Mellon University and featuring Kurt Olsen's elegant chart.

A special thanks to Ed Felten for his thorough analysis.

Read our article here. Expect a follow-up post soon!

Here is a short list of links to sites on which The Public Ballot has been discussed.

Jon of ColumbusING has a great, simple summary of the process. Thanks!

Independents Unbound

Care2

Black Box Voting

Democratic Underground

BuzzFlash

Digg

Jennifer Brunner, the first Democratic secretary of state in 16 years, stressed that "with paper ballots, there is no chance that votes can be matched to the names of voters. With electronic voting, there's a minuscule chance that this matching of names and votes could take place. But that chance is so remote that Brunner's requirement is a solution in search of a problem."

Grr.

Read the full Colmbus Dispatch article here.

In a Columbus Dispatch article this morning, Ohio Secretary of State Jennifer Brunner (and company) advocated the use paper ballots (versus electronic) for provisional voting. More interestingly, she had this to say about the electoral system:

"The biggest concern I have here, which I am trying to protect, is the secrecy of the voter's ballot. I've got to make sure the system is one people have absolute trust in."

Well, that sounds good, I guess. In discussing provisional ballots, Matthew Damschroder, director of the Board of Elections said,

"...records linking voters' names and identifying codes are kept separately -- in buildings miles apart -- from voting machine paper tapes printed with those identifying codes and voters' ballot choices" (paraphrased by Dispatch author Barbara Carmen, and quoted by me).

Ha! Sounds like our next project.

Thanks to James Moyer for the heads-up.

A Down-to-Earth Explanation about Hacking the Secret Ballot

My previous post on the matter was just a cut-and-paste job with our press release, which was overly formal, according to Yury; others (myself included) agreed. To recap:

It is possible to see out how someone voted. This includes a particular person’s selection of candidates, as well as their decisions on ballot initiatives, referendums, etc. To be clear: this is tied to their name and address!

• Who could that ‘someone’ be?
• They reside in one of ten specific Ohio counties (or in any other state which employs a similar system).
• They voted at the polls (i.e. not by absentee ballot).
• They voted in the past decade or so.
• How accurate is your technique?
• Generally speaking, the technique is highly accurate – my best guess is 95%+, but the depth of technical analysis required for more precise figures has yet to be performed.
• What do you mean ‘anyone’ can do it?
• A person only needs to go to a county courthouse or place where an election’s ballots are stored, and perform the necessary aggregation and analysis. In our experience, it was pretty easy. We just called a county courthouse, asked to examine the ballots, and had a meeting scheduled within the week. Still, the precinct we decoded was small, which made it easy to perform the data aggregation by hand. It’s not difficult to do this on a mass scale; it would simply require some scanning/OCR equipment.
• So have you guys done this?
• Yes. In order to show people that this can actually be accomplished, and isn’t just theory, James Moyer and I have figured out how people voted on a particular ballot issue in a small precinct in Delaware County, Ohio. Immediately before voter confidentiality was breached, each voter was assigned a unique, anonymous ID.
• Wait, did you just say you can tell how people voted?
• For the second time, yes. Please read my previous post. This is a big deal.

If I can actually get people to read about this, then I’ll post an in-depth technical follow-up (“mini-whitepaper”?), which will include raw and processed poll data, as well as a step-by-step discussion of the methodology involved. Additionally, a list of affected Ohio counties will be provided.