Monday, December 8, 2008

Minnesota Brainstorming

As the Minnesota Senate recount grinds agonizingly on, one thing seems clear: there is no way to decide who truly won a race that's this close. I've been trying to think up ideas, but I'm afraid my best brainstorming hasn't solved the problem.

Running all the ballots through the machines and accepting the count is not the solution. Machines make mistakes. When you see a ballot that's marked with only a two-stroke "x" in the oval for the Senate choice, you know that's something the machine might not pick up. But when you see that the voter put only "x"s in the ovals for every election on the ballot, you know that the voter intended to vote for the candidates marked with the "x"s. As you look at images of the disputed ballots, you see many similar situations where the voter's intent is clear, but where it's understandable that the machine might not have counted it. Human judgment is necessary.

But humans make mistakes too. Judgment calls can be clouded by partisanship, intentional or unconscious. Even if everyone involved is counting with perfect honesty, there are still screw-ups. Humans can put ballots in the wrong pile. Counting a pile of 1000 ballots, humans might come up with 997 or 1004. Some ballots disappear.

And by the way, while it might seem that the solution to human partisanship is to trust the machines and say that it's just too bad if voters are too dumb to fill out their ballots correctly, remember that the apparently impartial solution of trusting the machines is actually partisan; the partisanship is just implicit rather than explicit. Democrats (at least according to conventional wisdom) have a higher error rate in filling out their ballots than Republicans, so putting the whole thing in the hands of the machines has the effect of giving a small edge to Republican candidates.

If the vote totals were separated by 1%, none of this would matter. 1% seems like a whole lot in the context of this recount. Even half a percent -- the statutory limit for automatic recount -- would be a luxurious margin. In this case, as in Florida's 2000 presidential race, the vote totals are separated by less than 1/100 of 1%. Even a margin of 0.02% would give me a lot more confidence that we would know the true winner at the end. The 0.01% difference seems impossible to resolve correctly.

Holding a runoff where no candidate reaches 50% (as Georgia did), would have solved the problem of this race, as would instant runoff voting, in which voters also vote for their second and third choices, and in which those votes immediately count if no one breaks 50%. Here, Dean Barkley's 400,000 votes would have been immediately distributed between Coleman and Franken and would, in all likelihood, have clearly broken the tie.

But neither of these methods really solves the problem. They would have worked this time, but might equally well have created trouble another time. For example, if Coleman had 44%, Franken had 42%, and Barkley had 14%, the election would have been clearly decided under the existing first-past-the-post rule, but distributing Barkley's votes under instant runoff might have created precisely the 0.01% difference that we see now. A candidate might have a clear plurality margin in an initial election but the runoff might result in a tie. So runoff and instant runoff (although perhaps desirable for other reasons) are just as likely to create an almost perfectly tied vote as it is to break one.

Similarly, one is tempted to say that the rule should be, "if the vote totals are separated by less than 0.01%, redo the election." But does that really help? Again, it just pushes the marginal cases out to a different line. There could be just as must trouble deciding whether the margin is above or below 0.01% as there is deciding which candidate is ahead under the current standard.

A better rule might be to vest someone (presumably the Secretary of State) with discretion to decide whether the election is so close that it should be redone, with the statutory guidance that redos should occur when the margin is about 0.01%. That would allow for a decision to be made without agonizing about determining the exact margin. But of course the fatal flaw with this method is that it would be impossible to find someone who could be trusted to exercise the discretion impartially.

Perhaps the best thing I can come up with is this: if the machine count on election day produces a margin of less than 0.5%, hold a recount, and if the margin is less than 0.01%, do the election over without a recount. That way the costs of a recount are avoided when it seems likely in advance that the election cannot be decided accurately. But again, even that method doesn't really solve the problem. It takes care of the present case, but still leaves trouble in the case where the initial machine count produces a margin of 0.02%, but the human recount is more closely tied.

Sorry, there's no solution. No matter what you do, there are going to be cases in which the count is too close to whatever line you've drawn.

I guess the ultimate solution is too increase the accuracy of voting on election day to the point where we can really trust the machine count and not have recounts even if it's really close. But that was the point of touch-screen voting, which has turned out to be a disaster.

The real solution is to win by a bigger margin.


Anonymous said...

There is indeed a way to truly decide who won a race this close. But not until we start using verified ballots.

It is a simple, at least straightforward, matter to eliminate the challenging of ballots: voters should be required to verify their ballots before leaving the polls.

If the tallying machine accepts the ballot as complete and countable, the voter should be required to examine a summary printout listing the names of the candidates and issues to whom or what the ballot will apply, as well as any under-votes. The voter could then approve it, or reject it, and either revote or correct the ballot.

Once verified, that ballot would become machine-only re-countable, and humanly unchallengeable, except in the case of physically damaged ballots no longer machine readable.

Human judgement would not be necessary - at least by perhaps orders of magnitude less than in the Minnesota case.

Ballots not so verified would be thrown out - with the voter still present.

This sort of verification was routine in the not too distant past with computer programming by punched cards. Cards were first punched by a typist from handwritten pages on one machine, and the holes were read on a second non-punching verifier.

Only a novice programmer - or a fool - would try to compile a program without verified cards. There were always mistakes.

Why do we tolerate unverified ballots?

Requiring verified ballots is the only way to truly decide close elections - by preempting questionable ballots, and eliminating ballot challenges.

But what is the likelihood that we will someday have ballot verification? About the same as having an accurate count in Minnesota - zero!

aaronson said...

Hey I know of a really cool website called the voting site . The sites mission is to evangelize the instant run off voting. Users can create there own instant run off elections and vote in other peoples instant run off elections. The site does a good job of showing how instant run off voting works.