In December of 2000, Caltech and MIT performed a voting accuracy study, which calculated the “known error” rate for various kinds of voting processes. They didn’t assume any byzantine failures or malicious actors, no externalities such as roadblocks or disasters or suspiciously designed ballots. The study was simply an evaluation of the inherent faults of the vote casting and counting methods. The results showed that each method was accurate to between 1 and 2 percent, with punch cards the best, hand counted paper the worst, and e-voting at about the middle of the pack. Keep in mind that these are “known” errors, cases in which it was discovered to be certain that the vote cast was not correct. These are by definition a strict lower bound, so the real error rates are certainly higher… their exact value cannot be known.

A recent University of Washington study using these minimum criteria to examine the 2004 elections showed that the results of three Senate campaigns, one Gubernatorial election, and the presidential electors of three states were statistically useless. In other words, if you ran time backwards and held these elections again in the same exact circumstances, they would be just as likely to come out the other way.

The important point here is that falling inside the margin of error is not as simple as saying that the results are “possibly wrong.” It’s equivalent to saying that we don’t actually know what the result was. If you have a telescope that can only see objects at a certain distance at 1-meter resolution, should you be relied on to read the newspaper over my shoulder? Better analogy: would you let me (20/600 uncorrected vision) drive your car without my contact lenses?

Now take a look at the (still tabulating) results from some key referenda in tonight’s California special election:

73 N Minor’s Pregnancy 2,563,070 48.9 2,674,283

74 N Teacher Tenure 2,464,243 46.6 2,812,781

75 N Public Union Dues 2,574,991 48.8 2,695,427

These have spreads of 2.2 points, or +/- 1.1 from even, 6.8, or +/- 3.4, and 2.4, or +/- 1.2, respectively. If a scientist published these types of results from an experiment that used methods of the above error rate, they would be lucky to just have their data called inconclusive. What does that say about our voting process? Should we really be ammending our state constitution by these criteria?

Update: The final spreads ended up being a bit wider, but let’s not be fooled into thinking they are authoritative. “My side” won, but does that really mean anything? I maintain that the types of margins we see in these elections are roughly equivalent to a guess as to what the populace thinks at any given moment. The results are especially worthless when you consider the massive volatility in pre-election poll data, indicating the number of voters who voted not out of conviction, but something closer to confusion. This process does not work, and must be changed.

Obligatory-bright-side-comment: Congratulations to new Virginia governor Tim Kaine, whose victory is well beyond any statistical questioning.

Related Posts:

• franchisers over enfranchisement (January, 2006)
• Most amusing news story of the month (October, 2006)
• Polling High School Seniors (January, 2006)
• Election Update: Do-Over on First Ballot (February, 2006)
• oh hell yeah: new red-blue map (May, 2006)

This entry was posted on Wednesday, November 9th, 2005 at 1:22 am and is filed under mathematics, politics. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.