Incredible Shrinking Lead

OK, I can't resist: another blog entry about the election.

Norm Coleman's lead over Al Franken is down to 174 votes! That's out of about 2.8 million votes cast -- a difference of about 6/1000 of 1%. It doesn't come much closer than that.

Coleman previously urged Franken to waive the recount and concede, but that was ridiculous. Ted Stevens has graciously conceded in Alaska, but he's down by about 3,700 votes out of 300,000. There's a big difference between being down by 1.2% and being down by 0.006%. The common statutory standard is that a difference of less than half a percent justifies a recount at state expense. 0.006% is a whole lot less than half a percent.

Moreover, if one assumes that Democratic voters are more likely than Republicans to mismark their ballots, in a way that makes it difficult for machines to count, then a recount could make a big difference to Franken. Even a slight edge could overcome a 0.006% difference.

So I think a recount is totally justified. The one thing that seems peculiar to me is one feature of the actual recount process. The part that makes sense is steps 6-10 in the process: each ballot is inspected by a team of humans to determine who the voter really intended to vote for. That's fine -- machines can make mistakes, and a human should look at each ballot and determine which pile it really belongs in.

But then step 11 says that humans will also count up the ballots in each pile. Now that seems like a bad idea. Humans make mistakes too. Humans are better at exercising judgment, but machines are better at doing routine, boring tasks that don't involve judgment, like counting how many ballots there are in a huge pile. Imagine counting a pile of about 1000 ballots. Would you come up with exactly the right number?

I would have the humans sort the ballots into piles by deciding which candidate each ballot votes for, but then I would have machines count up the ballots in each pile. Mistakes are possible either way, but I would trust the humans to make fewer mistakes sorting the ballots into piles and the machines to make fewer mistakes in counting the ballots in each pile.

Of course best of all would be to do an empirical study of which method involves fewer mistakes, rather than consulting one's intuition. But there's no time for that now. People are counting big piles of ballots. Let's hope they come up with the right answer.