HUNTSVILLE, Ala. - Wes Colley, perhaps best known for running one of the BCS computers used in college football for years, has also correctly predicted the last three elections using a statistical model. This year, his model has shown Hillary Clinton with over 300 electoral votes (you need 270 to win) since Oct. 7.
Out of the last three elections, Colley only missed a single state in two of them, and he was only off by two electoral votes in the other. He's never missed the final outcome.
Instead of polling averages, which are commonly used in prediction models, Colley explains, "Our method is based on the median of the last month of polls state by state."
"If I rank the polls from biggest Clinton win to biggest Trump win and take the middle one," he elaborates, "that's what we're saying tells us what the state would do."
After watching the model for months he says, "It would be very surprising if she weren't able to pull it out at this point."
As of this writing, the model has Clinton finishing with 323 electoral votes and Trump with 215.
Still, he says Trump has a map that works for him, but he can't afford to lose any of the toss-ups.
"Trump needs to flip five straight heads, if you like, and if Hillary flips one tails, she probably wins it," he concludes.
Colley says on a scale of 1 to 10, his surprise at a Trump win would hit eight or nine, he adds, "It's just going to take a lot of things to break the right way."
Trump hit a high watermark in Colley's model back on Sept. 27, when he peaked with 221 electoral votes. Clinton had 264 at that point. 53 votes were considered toss-ups. Clinton made it back above 270 votes in the model on Sept. 29, and she hasn't dipped back below that level.
However, Colley does point out that his model might have a weakness that this election cycle would draw out.
It might not change fast enough.
Colley says the FBI's announcement from directorJames Comey has moved the needle dramatically in many models, "It does appear that has shifted the polls back toward Trump in a few states."
Because Colley's model depends on medians instead of averages, it keeps a more steady pace. This helps filter out noise, but in the case of big shift happening late, the model might not capture it all.
"It does appear this eleventh hour surprise may be having some impact," Colley notes, "That hurts us, because we're assuming everything sort of settles out by the last month. It'll be interesting. A good test of our system."
He says Trump has a way to win the electoral college, but it comes with little margin for error, "You have this collection of states, Ohio, North Carolina, Florida, Arizona, Nevada, that Trump essentially needs to win all of those or forget it. So if you hear North Carolina has gone for Hillary Clinton election night, I think you can pretty much go to sleep."
He says when picking out targets for Trump on election day, "You get sucked in by the closeness of some of these percentages. Six, well you can overcome six percent maybe, but can you overcome seventeen straight?"
We'll see soon.