19 November 2008

Recounts in HD-8 and HD-58 are not likely to succeed

Updated. Ballots in two Montana legislative districts — HD-8 in Kalispell, and HD-58 in a little town, Laurel, a bit southwest of Billings — will be recounted, with the HD-8 recount beginning on Tuesday, 25 November. Neither recount is likely to change the outcome of the election.

But much is at stake, so both political parties are playing long shots.

In HD-8, Democrat Cheryl Steenson leads incumbent Republican Craig Witte by 20 votes out of 4,044 counted. In HD-58, incumbent Republican Krayton Kerns leads Don “Doc” Woernerby 22 votes out of 4,620 counted. Both men are veterinarians. If Steenson loses her recount, but Kerns wins his, the Republican Party wins control of Montana’s house of representatives, which is currently divided 50-50. If Steenson and Woerner both win, the Democratic Party controls the house.

Because of Montana’s peculiar law governing tied elections (the governor picks the winner), the requirements for winning the recount are higher for Witte than for Woerner. In HD-8, assuming no new votes are found (a risky assumption that I’ll discuss below), Witte wins if 11 votes are subtracted from Steenson’s total and added to Witte’s. In HD-58, Woerner ties — which means he wins, thanks to the tie breaking system — if 11 votes are subtracted from Kerns’ total and added to Woerner’s.

Montana’s recount law requires hand counts, which will catch tabulating machine errors. Correcting machine errors will change vote totals, but the totals won’t change much because optical scanning machines (both Flathead and Yellowstone Counties use ES&S Model 650 tabulators) seldom make errors. The federal standard for optical counters is no more than one error in 250,000, or an error rate of 0.0004 percent. That’s a laboratory specification, of course, and higher error rates can be expected in the field.

Calculating the error rate can be tricky. An incorrectly programmed, or maliciously programmed, machine will read a vote for Jones but record it for Smith. An electrical malfunction might cause a valid vote not to be recorded. A build-up of dust and paper fibers might block a sensor, causing a valid vote not to be recorded. An oval filled correctly with the wrong kind of ink or pencil, or not filled enough with the right kind of marker, might not be recorded as a vote. A ballot with a valid vote recorded by the tabulator might be disqualified if the ballot contains an illegal identifying mark or some other defect that was not recognized by the machine.

Moreover, humans counting the votes may find valid votes on ballots for which the tabulating machine recorded no votes — under-votes in elections jargon; in HD-8, there were at least 184 under-votes; in HD-58, approximately 157 — but that actually contain valid votes. Examples include an oval with a check mark or X in it, or underlining or circling a name. Voters accustomed to old style paper ballots sometimes use old style methods to mark their optical scan ballots.

Such errors are more frequent than one might expect. Minnesota’s audit of its 2006 election revealed an error rate of 0.056 percent. A study of New Hampshire elections from 1946 to 2004 found an error rate of one-half percent for optically scanned ballots that were recounted by hand. The spread is an order of magnitude.

Applying the error rates to HD-8 and HD-58 we get these average number of expected changed votes:

At the lower end of the range, it doesn’t matter how the changes are distributed. There are not enough votes to change the outcome of the election. At upper end, however, there may be enough changed votes to flip the outcome if the changes are biased toward one candidate. If the changes are not biased, if they are evenly distributed, the outcome is not changed.

So, let’s examine the probabilities using the New Hampshire error rate. In HD-8, Witte trails Steenson by 20 votes. If 21 votes change, 16 must change in Witte’s favor, and only 5 in Steenson’s favor, for Witte to win. What is the probability that at least 16 of the 21 changes will favor Witte? After a short visit to Assumptionland, where we conclude that the district is evenly split and that the odds are even that a voter will choose Witte or Steenson, we can make a calculation. The chance that Witte will get 16 of the 21 votes is 1.3 percent. Not good odds — and we have yet to factor in the odds that at least 21 votes will change. That likelihood is only 54.4 percent. Combining the probabilities cuts the odds down to 0.7 percent (1 in 143).

In HD-58, Woerner wins if 18 of the 24 votes flip. His combined odds of that happening are 0.6 percent (1 in 167).

These numbers change if the assumptions are varied, so the example above is illustrative, not definitive. And there’s more than one way to assess the probabilities. But the numbers won’t change all that much, something the statisticians in both parties know.

So why defy such poor odds? First, because so much is at stake, and the potential reward justifies the expense. Second, because both parties probably hope to challenge enough ballots to change the outcome(Update: see Minnesota Public Radio’s primer on ballot challenges and voter intent). The Republicans are likely to send to Kalispell a squad of legal toughs with instructions to challenge as many ballots for Steenson as possible. And the Democrats are likely to deploy to Billings their legal toughs with instructions to challenge ballots for Kerns. Steenson and Kerns will have their own lawyers present, so things could get lively in the recounting rooms.

I think Woerner and Witte both know the jig is up. In fact, Witte admitted as much to the Daily InterLake’s Jim Mann:

Woerner said much the same thing to the Billings Gazette: “I realize the odds are not in my favor, but I want complete closure,” he said. “It’s a good civics lesson.” Yellowstone County Elections Administrator Duane Winslow agrees: “It’s not going to flip. [22 votes is] a huge margin.”

If the two political parties were not grasping at straws, hoping to avoid sharing power in the house, I suspect that neither Witte nor Woerner would complicate his life with a recount for which there are lottery ticket odds of winning. These are cases in which not exercising the right to a recount is probably the wiser choice.