Simulation trend for median number of Democratic Senate seats
The lesson of the last two days is, the polling gods giveth, and the polling gods taketh away. On Tuesday, Democratic odds of retaining control of the Senate peaked as high as they've been since Daily Kos Elections introduced its model; they moved up to a 59 percent chance of holding the Senate, with the median number of Dem-held seats up to 51 for the first time. That golden age lasted all of one day. By the end of Wednesday, Democratic odds had fallen below 50 percent, and the median number of seats was back down to 49; that knocked them back to about the same shape they were in before the Kansas switcheroo shook things up in their favor.
If you've been following our model closely, you'll remember that there's been not a lot of volatility until now. As you can see in the graphic above, which shows the change in the median number of seats from day-to-day, there was a period of a number of weeks where it simply stayed flat at 49; during that period, Democratic odds of holding the Senate stayed stationary around 46 to 47 percent. Only last week did Democratic odds cross over the 50 percent line, and they suddenly shot up to 59 percent Tuesday, before nose-diving to 46 percent on Wednesday.
Does that volatility have an impact on every Senate race, though? Not at all; it's really only coming from four races. Problem is, those are four of the six most pivotal races. A minor fart in one of those races gets amplified into an explosion simply by virtue of their centrality to the whole system. On a day like Wednesday, where you get a bad result in four different races at the same time, the reverberations are huge.
The biggest impact was in Louisiana, where a Fox News poll put Republican Rep. Bill Cassidy ahead of Dem incumbent Mary Landrieu by an outsized 51-38 in a hypothetical runoff. This seems to have outlier-scented stink lines coming off it, since Landrieu has never trailed by more than 3 points in a nonpartisan poll before. (I can already hear your shouts of indignation about Fox News being a "nonpartisan" poll, but the key distinction for our purposes is whether a pollster performs polls for candidates or advocacy groups, as opposed to a poll for a media organization. Furthermore, going by 2012 pollster rankings, they were unremarkable, middle-of-the-pack performers in terms of average error and Republican bias.) Nevertheless, that means it goes on the pile with everything else, in just the same way that we did with the equally odd (but in the totally opposite direction) poll from last weekend that saw Pat Quinn suddenly crushing in the Illinois gubernatorial race.
Even factoring in another Louisiana poll from Wednesday—a 45-45 tie from Gravis (which works out in Landrieu's favor, since they're penalized for being an openly Republican pollster)—that 13-point deficit caused Landrieu's odds to plummet 48 percent to 17 percent. Now, a Democratic majority is by no means dependent on a Landrieu victory; the Democrats could lose her, Mark Pryor, and the three red-state open seats and still control the Senate. But the model looks at every possible permutation, and nearly closing-off one of the main routes to 50 has a big impact on the overall chances.
We'll look at the rest of the new developments over the fold:
The decline in Iowa wasn't quite as bad, though it still took Braley's odds back below 50 percent after several weeks above water. The Quinnipiac poll released on Wednesday found Braley trailing Republican Joni Ernst 50-44; although the impact was somewhat lessened the same day by a Fox News poll showing the race a tie, it was still enough to drop his odds from 60 percent to 36 percent. (This poll definitely doesn't comport with
either party's understanding of the Iowa race, based on what Republicans have been leaking to Politico. But there's no way to build whispered conventional wisdom into the model, nor should there be a way.) Also we need to account for Wednesday's Suffolk/USA Today poll of Colorado, which was the first poll in two months to give Republican Cory Gardner a lead (and a tiny one, at 43-42); that pushed Mark Udall's odds down from 77 percent to 64 percent, which also weighs on the model's overall odds.
Finally, there's Kansas, which has been the source of almost all the confusion in the Senate model in the last few weeks, and doesn't seem likely to stop. When we checked in on Monday, odds of Democrats picking up this seat were languishing at 39 percent (since there's a two-step process here: not only does independent Greg Orman have to beat Pat Roberts, which is by no means a sure thing, but Orman then has to agree to caucus with the Democrats). On Tuesday, a Public Policy Polling poll came out giving Orman a 41-34 lead over Roberts, with the dropped-out Chad Taylor pulling only 6. That boosted Democratic odds in Kansas to 55 percent; Orman's chances of winning were up to 82 percent, and then he was likelier than not to caucus with them because of their odds of hitting 50 seats elsewhere. The improved situation in Kansas was almost single-handedly responsible for the surge in the Dems' overall odds to 59 percent (though a couple nice polls in North Carolina helped).
On Wednesday, well,
Fox News showed up in Kansas too. The results weren't terrible for Orman; they had him down 40-38 with Taylor at 11. But that was enough to push Orman's odds of winning back down to 71 percent, and, once you factor in the decline in Louisiana and Iowa, Democrats were already more likely than not to lose control of the Senate, making it likelier that Orman would instead caucus with the GOP if he beat Roberts. That pushed Democratic odds of a "win" in Kansas down to 42 percent. In 13 percent of all scenarios, though, Orman is still the sole decider as to which party controls a 49 D/50 R Senate.
On top of all that, Democratic odds fell below 50 percent in two gubernatorial races. One is Colorado, thanks to another suspicious Quinnipiac poll; Democratic odds there fell from 61 to 41 percent. Quinnipiac has generally done a fine job of polling the east coast states that are in their wheelhouse, but they've always seemed to shoot right of the mark in Colorado (in 2012 as well as this year).
On the other hand, the decline in Wisconsin may be more worrisome, since it comes from Marquette Law School, who've established themselves as the gold standard for polling in Wisconsin. After a 49-47 lead for Mary Burke among likely voters in their August poll, Marquette found Scott Walker leading 49-46 in Wednesday's poll. That, very similarly, pushed Democratic odds in Wisconsin from 61 percent down to 43 percent. Together, that was enough to push Democratic odds of gaining gubernatorial seats down from 58 percent on Monday to 54 percent today; the median number of seats fell from 23 to 22, which would be a one-seat gain over the current 21.
Finally, we'd like to take this opportunity to not criticize any of our rival modelers. The distinction between a model that gives Republicans a 45 percent chance of capturing the Senate, versus a different model that gives them a 55 percent chance, isn't really a fruitful basis for allocating who's "right" or "wrong," any more so than deciding a model is more or less "right" when it fluctuates between a 45 percent chance and a 55 percent chance on different days. Either way—to repeat the same tedious caveat we make every week—you're describing the most minute variations on what is truly a coin flip.
If nothing else, all the models should probably just be thrown on a pile together and aggregated, to smooth out the different assumptions ... in exactly the same way that each model throws all the polls on the pile to smooth out their different assumptions. Luckily, to that end, Vox
has created a bookmarkable page that's true one-stop shopping, that pulls all the different models into one place and averages them out. However, I suppose the next logical step, once someone else sets up their own site to compete with Vox, is that we'll need someone to aggregate the aggregators of the aggregators. (At which point the universe will probably implode into a black hole under the weight of all the meta.)