E-voting Security Fixes Will Get Us Nowhere Without Stats

[Wilms]

E-voting Security Fixes Will Get Us Nowhere Without Stats

At the American Association for the Advancement of Science meeting, a statistician made a forceful argument that her field can help us do a better job of ensuring fair and representative elections, but only if we decide to let it.

By John Timmer | Last updated February 23, 2009

The recent American Association for the Advancement of Science meeting included a session entitled "Science for Public Confidence in Election Fairness and Accuracy" and, as might be expected, computer science made a significant appearance. Ed Felten of Princeton, whose work in the area we've covered extensively, spoke and emphasized the limits of what computer science can do, and how the ultimate goal should be to ensure that electronic voting systems are verifiable and auditable. Of course, that raises the question of what you do with the auditing information, which is where Arlene Ash, a biostatistician at Boston University's School of Medicine, came in. It turns out that we already have excellent statistical tools for detecting problematic patterns of voting—the legal system just chooses to ignore them.

Felten and his fellow Princetonian Andrew Appel have been at the forefront of efforts to explore the security of e-voting systems, and Felten's talk was largely a recap of past news that we've covered extensively. When it comes to the equipment itself, it's mathematically impossible to verify that the code they run will behave properly under all circumstances, which means that the best we can do is provide a verifiable and auditable record of the vote, allowing problems to be identified retrospectively. Even that's difficult to reconcile with our expectations for anonymity; in describing the challenge of creating an algorithm that simultaneously encrypts and anonymizes the votes, Felten said, "we've reduced this to a previously unsolved problem—we're really good at that in computer science."

Until that problem is solved, many states are opting for optical scan voting or printing voter verifiable receipts, which can allow a post-election audit to identify significant problems. But running these audits raises a whole new series of issues, some of which are less a technical challenge than a matter of how carefully we want to listen to what an rigorous analysis of a vote tells us.

~snip~

This, from Ash's perspective, represents the crux of the problem. We have sophisticated statistical tools that we rely on for everything from medical research to verifying the flow of money through Las Vegas casinos but we simply haven't chosen to mandate that they be used to verify election results. Even in cases like the elections in Sarasota, where they were deployed, the results were deemed legally irrelevant unless they provide an indication that election results were distorted by malice or intent. Sloppiness or incompetence, apparently, is acceptable, despite our country's promise to respect the intent of the voters.

There are cases where post-election audits are rigorous—the American Statistical Association's collection of election-related materials calls New Jersey's law "the most 'statistical' audit procedure enacted by any state." But, of course, this is the same New Jersey that has failed to act on a court order that would ensure that its voting machines actually print something that would allow an audit.

The clear take-home message of this session was that optimizing the computer science aspects of building a better voting machine is a really hard challenge, and it won't ever eliminate the sorts of UI design issues that regularly cause election results to reflect something besides the will of the voters. We do have tools available to us that can help us recognize when there is a disconnect between voters' intent and results, but we haven't taken them seriously; we don't ensure their use, and we don't legally demand that their results be considered. Until we see that change, it appears that our commitment to accurate elections is superficial.

~snip/more~

http://arstechnica.com/science/news/2009/02/e-voting-fixes-will-get-us-nowhere-without-statistics.ars