If you'd asked me three or four years ago who the most endangered Democratic incumbents in 2014 would be, I'd have guessed Mark Begich and Kay Hagan. Begich, after all, won in 2008 only by the skin of his teeth against Ted Stevens, who'd been convicted on corruption charges only weeks before; Begich, it seemed reasonable to conclude, wouldn't be lucky enough to face another convicted felon six years later.
And Hagan won by a more convincing margin, but only against the backdrop of a huge Democratic turnout surge in what had, to that point, been a red state at the presidential level, and with a huge assist from an Elizabeth Dole campaign that imploded during the stretch run, most notably with one of the most legendary political ad backfires of all time. Certainly it wouldn't have occurred to me that Mary Landrieu or Mark Pryor — who didn't even bother having an opponent in 2008 — would be in any trouble ... or that Max Baucus or Tom Harkin would retire, to say nothing of the fact that the Democrats would have trouble holding those seats.
Well ... fast forward to now. Mark Begich seemed a few weeks ago to be on track to eke out an unlikely victory, running a pitch-perfect campaign and leading most polls. However, just in the last week, Begich seemed to run into a polling brick wall, trailing in three straight polls (from Public Policy Polling, Rasmussen Reports, and a Republican internal). That wall got stacked even higher since our last Polling Outlook writeup last Thursday, with an apparently non-partisan poll from local pollster Hellenthal Research that put his Republican opponent, Dan Sullivan, up 46-42. That pushed Begich's odds in Alaska down to 26 percent, from 36 percent last Thursday.
Pundits who subscribe to the "Game Change" theory of politics would probably point to the flap over an overbearing attack ad that Begich had to pull (an ad that probably got more attention inside the Beltway than in Alaska). Instead, there's probably a much simpler reason for Sullivan's rise: The GOP primary finally rattled to a conclusion in late August, and Joe Miller's supporters, who prior to the primary never would have admitted to be willing to vote for RINO squish Sullivan, are now totally fine with voting for said RINO squish. If you need further proof, just look at the trends in the toplines over the months; Begich has always been polling in the low 40s, with the same base of support, but Sullivan suddenly shot up from the high 30s to the mid 40s in the post-primary period.
On the other hand, then there's Kay Hagan. No one has been raving about her ads the way they have with Begich, and through May and June she was having trouble pulling away from her opponent, Thom Tillis. But she may have simply lucked into having a bad opponent, again: Tillis suffers from his association with the state's unpopular Republican-controlled state legislature, where he's House speaker. His numbers fell when the legislature went into a prolonged special session over the summer, and never really recovered; never the most charismatic fellow, he then proceeded to run a very mansplaining ad against Hagan and followed that up with an even more mansplainy debate performance.
And now two more polls, since our last Polling Outlook writeup, show the race getting away from Tillis. Hagan led by 3 in a CNN poll, and led by 4 in a poll from Gravis Marketing (effectively a lead of 7, since they're considered a Republican partisan pollster). That pushed her odds from 73 percent up to 81 percent, the highest they've been yet, and in the same zone as not-entirely-locked-down-but-not-too-worrisome races like Michigan and New Hampshire. (In fact, Hagan has actually trailed in fewer polls in the month of September — one YouGov poll — than Jeanne Shaheen has in New Hampshire — two polls from Republican internal pollsters, Magellan and Vox Populi.)
We'll talk about the gubernatorial races, as well as 538's new pollster ratings, over the fold:
The Alaska result has a bigger effect on the Senate Democrats' overall odds than the North Carolina results, though. That's because the Democrats need to hold
both those races to hold the Senate, absent some other bank-shot approach that seems unlikely by now (i.e. salvaging Louisiana, or flipping Kentucky). If one of those races falters even while the other gets better, that still makes the overall odds worse; in this case, Democratic odds of retaining control of the Senate kept falling from 37 percent last Thursday to 33 percent now. 48 seats is both the median and modal number of Democratic-controlled seats after the election.
It was a slow half-week for gubernatorial races; most of the competitive races didn't see any new polls at all. The only big mover, in fact, is one that really hadn't been competitive at all up until this point, the Massachusetts gubernatorial race. That's entirely thanks to the new Boston Globe poll, taken by SocialSphere, which found Democratic nominee Martha Coakley actually trailing her Republican challenger Charlie Baker, 40-38. That may have surprised a lot of people, and inspired a lot of "will Coakley choke again?" headlines — but it really wasn't that new, as this wasn't even the first Globe poll to have Baker in the lead (it also happened once in August) and a number of their other polls saw Coakley leading only by low single digits.
Overall, though, that only pushed Coakley's odds down from 92 percent to 84 percent. This is one of the most overpolled races in the country, and other pollsters have found it a less competitive race; MassINC's three polls in September, for instance, gave Coakley leads of 9, 9, and 10 points. Still, this is a race that needs to be taken seriously; the trendlines still show a much closer race between Coakley and Baker than it looked like during the spring, and Massachusetts did spend much of the 90s and 00s electing moderate Republicans to its state house.
Overall Democratic odds declined only slightly, from a 54 percent chance of gaining gubernatorial seats to a 52 percent chance. The median number of seats is 22 (which would be a gain of 1 over the current number), although 21 is now the modal result.
Finally, I'd like to briefly mention the new Pollster Ratings released by 538. This is a valuable asset, and you should bookmark it; you can tell at a glance whether a particular pollster is methodologically sound and how big that pollster's typical bias is (and in what direction it usually goes). It can help you be better informed when looking at our Polling Database, or reading about a particular pollster in a Polling Outlook writeup or Steve Singiser's Polling Wrap.
That said, though, we don't include any sort of pollster ratings in our model (other than a blanket 3-point correction for all partisan or internal polls, based on research how much more internal polls deviated from the actual result than nonpartisan polls did, and based on the principle that internal polls only get leaked when they serve to advance a campaign's preferred narrative). Nor do we plan to start doing so, at least not this year. That's partly because, as in many other areas, past performance is not a guarantee of future results. Assuming that it past is prologue is just another potential source of error.
A case in point is Rasmussen Reports, a pollster that had a particularly noticeable Republican bias to its results in 2010 and 2012. It's under new management this year, though, and its results tend to hew more closely to the critical mass of what other pollsters are seeing (though it's unclear whether that's because of better methodology, more herding, or just dumb luck). Would the accuracy of our model be improved if we penalized Rasmussen for their sins of the past? There's no easy way to know, at least not before the election.
Another possible solution is to correct pollsters for their apparent bias, but focusing on how far a pollsters' polls deviate from the critical mass of other pollsters this cycle. That might be more fair to firms like the new-and-possibly-improved Rasmussen, but it's still a potential source of error. A firm that's consistently out on a limb could be simply wrong ... or it could be getting everything right while everyone else is herding around an incorrect set of assumptions, and, again, there's no way to know until the election is over. A case in point might be SurveyUSA's polling in 2010 and especially in the 2009 gubernatorial races, where they were putting out ridiculously bad numbers for Democrats early in the cycle, but then everybody else caught up with them, in what was a ridiculously bad year in reality. The same thing could be happening this year with, say, Quinnipiac, which seems to have suddenly shifted to a very GOP-friendly likely voter screen, but who could just as easily be proven right rather than wrong about that a month from now.
Finally, there's the matter that poll averaging buffs out most of the underlying differences. As Nate Silver pointed out in his description of how the ratings work, most pollsters aren't bad but rather are simply average. If you round up a large enough collection of enough pollsters' average offerings, you have a pretty clear picture of the state of the race. As I've said before, just about everyone's predictive models of the 2012 presidential election got every state right, despite the constant presence of pollsters like Gallup and Rasmussen shooting way off the mark; there was a large enough sample of average pollsters doing average work that the outliers got outweighed by everything else.
The only trouble that models had in 2012 was in the few Senate races (Montana and North Dakota) where there simply wasn't enough information, and on top of that, what little information there was tended to be wrong. The fact, for instance, that Mason-Dixon had been a fairly accurate and methodologically-sound pollster in the past, didn't give any notice that they were missing the boat in Montana in 2012. Could that be happening in underpolled races this year in hard-to-poll states, like, say, Alaska? Sure, it could ... but when you've got races like North Carolina and Arkansas that are getting polled several times a week by the full spectrum of pollsters, it's much less likely that a Montana-type situation is brewing there.