Rumor has it that canvassing will be completed today in Wisconsin's Supreme Court election which will allow the declaration of an official winner.
From a data perspective, it's been interesting. By midnight on election day the incumbent held about a 700 vote lead with 99% of precincts reported. Although the AP wouldn't call the race, pundits said that it was very unlikely that the lead would change based on past voting patterns in the last 1%.
Did you ever hear the phrase "Past performance is no indication of future returns"? I guess someone should have reminded the pundits. By morning, the challenger turned things around and climbed to a 200+ vote lead with just a single precinct left to report.
Again, the pundits said that, based on past patterns in that precinct, the challenger's lead would be cut by about 100 votes leaving her with slightly more than a 100 vote lead, guaranteeing a recount. The last precinct didn't change the lead by more than a few votes.
In spite of the "Past performance is no indication of future returns" disclaimer, this was a statistically interesting result. Any time there is a major change in patterns, it should stand out to statisticians. That's not a statement about vote fraud. It's just a statement that a new pattern emerged.
But then things got even more interesting.
On the same day that people were saying "hmm, an interesting change in patterns", it was announced that ALL the votes in Brookfield (Waukesha County), had been excluded from the original count. Not miscounted but completely missed. This was over 14,000 votes that no one had noticed were missing. Unlike some other precincts, Brookfield did follow its traditional voting pattern and turned the lead from 200 votes in favor of the challenger to 7500 in favor of the incumbent.
Of course, the other side has tried to shout "fraud!" but the evidence so far points to nothing more than error. Stories I have heard vary on whether or not a number was entered in the wrong column of a spreadsheet template or it was entered into the correct column but not saved.
The Waukesha County Clerk has taken full responsibility for the human error. That's commendable but this is a big error. Really big. I think those calling for her resignation are justified in doing so. Data, especially election data, is valuable and needs to be treated as such. When an error this big has been made, saying "oops, I'm sorry" isn't quite enough. Because of the nature of the County Clerk's position, there is no demotion penalty available. Thus, resignation may be the only appropriate response.
In the Clerk's defense, however, I've got to point out what an incredibly arcane system this is. Someone counts up the votes, or takes totals off of the machines, and then types the numbers into a spreadsheet? I'm a big fan of spreadsheet modeling but it sounds kind of "horse and buggy" to build an election reporting system around this technology.
Then again, I voted on a paper ballot that day so maybe Excel ® is cutting-edge technology after all.
It would be interesting to see how many voters would have had to switch sides for the predictions before the Waukesha votes were in to be true.
ReplyDeleteIf we knew that on previous elections, only 5% of voters switched sides, with a low standard deviation, and for those results to be true would require a double digit change in voting patterns, it would have been easier to raise suspicions before we even knew about the Waukesha error. This could be done with a significance test.