If you want to know precisely how it works, we have a full explanation, but the short version is that we calculate smoothed trendlines for each race, based on all polls this cycle, and then run Monte Carlo simulations to find how likely various outcomes are. That, unfortunately, means the model is only as accurate as the polls themselves, but the hope is that, by November, there will be enough polls accumulated that their sheer volume will overcome any individual polls' biases or errors.
The big mover, again, is Alaska, where Mark Begich's odds of victory fell from 82 percent to 58 percent. Frankly, the 58 percent mark feels more right, intuitively; although most analysts would agree that Begich has a slight upper hand here, it's still a red state at the presidential level, one which he barely won in 2008, and he's facing a credible opponent.
What happened here is, in one word, Rasmussen. There's been very little polling of Alaska, compared to the other major Senate races, and the polls have been dominated lately by a good run of PPP polls for Begich and a too-good-to-be-true YouGov poll that put him up by 11. Rasmussen Reports, however, came out with a poll on Monday that put Dan Sullivan up, 47-45, the first public poll to give Sullivan a lead. Add that to the very small pile of polls in Alaska, and that all averages out to a much closer race than before.
No doubt some of you are objecting that that's just more evidence that we should be banning, or at least downweighting, Rasmussen, if they can just barge into a race and skew the whole thing. Well, that's not the way poll aggregating works, especially since Rasmussen can giveth just as much as they taketh away. Case in point is Rasmussen's Wednesday poll in Arkansas, which moved the needle there quite a bit, boosting Democrat Mark Pryor's odds from 37 percent to 49 percent. Rasmussen's poll, giving Pryor a 44-43 edge, is actually the first public poll to give Pryor a lead since May. (The battle in Arkansas has mostly been waged through leaked internal polls, and a number of recent Democratic internals also gave Pryor a lead. Bear in mind, though, that we do downweight polls released by campaigns or their supporters, on both sides of the aisle.)
Iowa is the other Senate race making a move. It's not a big move, from 51 percent to 45 percent odds for Bruce Braley, but it is an important psychological barrier, in that he now has less than a 50 percent shot. In fact, he now has slightly worse odds than either of the endangered red-state incumbents, Pryor and Mary Landrieu—though perhaps that's fitting, since, historically, open seats have proven much easier to flip than beating an incumbent. Interestingly, this small drop happens despite Braley not actually trailing in either of the two polls that came out this week (a PPP poll where he led 41-40, and a Suffolk poll that was tied 40-40), but they're both so close that throwing them on the pile diminished his small previous edge.
In terms of overall odds, this combination pushed the Dems' chances of controlling the 50 seats needed to hold the Senate from 47 percent to 45 percent, as Begich's move down seems to outweigh Pryor's move up. The median number of seats resulting from all those Monte Carlos is still 49, while the modal number of seats is still 50. In other words, Senate control is still a complete and utter coin flip; in the grand scheme of things, a slight flicker between 45 and 47 percent odds does very little to alter the fundamental coin flip-ness of the Senate situation.
The governor's race situation just got noticeably better, though, and we'll discuss that over the fold:
Another big mover in the gubernatorial races is Michigan, where Democratic ex-Rep. Mark Schauer is facing Republican incumbent Rick Snyder. On paper, this has always looked like a top-tier pickup possibility for the Democrats, since it's a light-blue state, and Snyder dinged up his moderate veneer by signing right-to-work legislation.
The polling hasn't cooperated though, with Schauer often coming within low-single-digits of Snyder but not actually leading at any point since a Dem internal in February. On Wednesday, however, EPIC-MRA (one of the better among the many dubious Michigan-only pollsters) released a poll showing Schauer leading, 45-43. That juiced Schauer's odds quite a bit, from 25 percent to 41 percent. This is one of those cases where we'd want another poll to confirm before getting too optimistic, but, if correct, Schauer's peaking at just the right time.
Finally, there's Wisconsin, where the Mary Burke surge appears to be real, too. This race was essentially deadlocked in our trendlines, with Burke having 52 percent odds against Republican incumbent Scott Walker, but on Wednesday, Marquette Law put out a poll with Burke leading Walker 49-47 among likely voters. The poll also gave Walker a small lead among registered voters (which is unusual, especially in a midterm, but it shows that Democrats are actually winning the enthusiasm gap in Wisconsin), but our model privileges LVs where they're available. That served to break the deadlock, pushing Burke's odds up to 57 percent.
Cumulatively, those four improved states pushed Democratic odds of gaining governorships up from 41 percent to 52 percent overall. And for the first time since we started the model, the median number of Dem-held governorship ticked up to 22 instead of 21. They currently hold 21, so that would be a net gain of one.
If you want to delve into greater detail about the previous polls this cycle, please check out Steve Singiser's polling database, which provides all the data our model draws on. We'll provide another written update next Monday, but the polling database, and the permanent Senate and gubernatorial pages of the model, are updated every day, so please check back frequently to see the battle unfolding.