Election Day is finally upon us, and with that comes our final look at the Election Outlook model—or at least the final look in regulation time, as it's almost certain that Louisiana will go to a runoff and likelier than not that Georgia will as well; if so, we'll continue to predict the odds in those races in the coming weeks. Unfortunately, we aren't wrapping up with particularly good news; the Democrats' odds of retaining control of the Senate have fallen considerably, to the point where the polls simply need to be wrong in order for them to pull it out.
That isn't to say that it isn't possible that the polls could be wrong, simply that there's no way for the model to account for that in a fair and consistent way. (Believe me, if the model is "wrong," I'd be as happy about that as anyone.) It could happen because pollsters simply aren't adequately reaching segments of the voting population who are nevertheless likely to vote Democratic, or it could happen because persons who hadn't intended to vote earlier got contacted and activated through Democratic GOTV efforts.
Two states where one or both of those possibilities are especially plausible are Colorado and Alaska. Colorado, as I've emphasized repeatedly in the last month, is a state where pollsters have tended to underestimate Democratic chances in the last several cycles; most 2010 polls put Republican Ken Buck ahead of Democrat Michael Bennet (who went on to win), and many 2012 polls had Mitt Romney beating Barack Obama here. Colorado has a large Latino population who may be hard for pollsters to pin down (especially pollsters who only use English-language operators and/or who only call landlines), and Colorado's switch to full vote-by-mail removes one more barrier that might have kept an "unlikely" voter in the non-voting column.
Similarly, Alaska is a notoriously hard state to poll, and Mark Begich has been devoting a lot of effort to under-the-radar GOTV efforts, especially among Alaska's large and far-flung Native population. There are some positive signs in early voting numbers in both Colorado and Alaska as well.
Nevertheless, for the Democrats to retain control of the Senate, they'd have to hope that both of those against-the-polls outcomes happen. Then on top of that they'd have to hope that either the polls are also wrong in Iowa (the polls have been closer in Iowa, but there's also a fair amount of certainty to them, given how many polls there have been and how consistent they've been, and given how Iowa is a whiter and older state than Colorado and Alaska, and thus easier to model), or that Greg Orman not only wins in Kansas but decides to caucus with the Democrats, all without losing the still-not-secure races in North Carolina and New Hampshire. Pulling one rabbit out of a hat is somewhat plausible, but pulling three rabbits out of hats in succession would be a feat for the ages, though not the basis for a sound strategy.
We'll look at specific changes to the model over the last week, over the fold:
The overall odds of retaining control of the
Senate fell from 31 percent to 10 percent over the last half a week, with the median seat count falling from 48 to 47 during that period. That may seem like a very steep drop for such a short period, especially since the odds in any one individual race didn't change that much. Part of the problem is that almost every one of the major races got at least slightly worse for the Democrats. That goes beyond the core Colorado-Iowa-Alaska trio, but also North Carolina, where Kay Hagan's odds remain above water, but not much better than a coin flip, falling from 65 percent last Thursday to 56 percent now. Almost every poll of that race falls within the range of a 1- or 2-point lead for Hagan; even though the center mass of all those polls points pretty conclusively to a Hagan win, the error bands on the trendlines overlap enough to allow a fair number of Thom Tillis wins in all our simulations.
Also, the alternate routes are starting to close off, with odds in Arkansas, Georgia, and Kentucky all falling into the single digits (most notably from 15 percent to 2 percent in Arkansas, where the trendlines now project to a 54-46 victory for Tom Cotton). Individually, it wouldn't matter much if one of those races got cut off, but when all of them start to get cut off at the same time, it means the Alaska/Colorado/Iowa play gets closer and closer to being the only path to 50 left.
The rest of the problem, though, is that we've slowly been removing random noise from the model. Rather than relying on the "fundamentals" and slowly removing the fundamentals' thumb from the scale, our way of adding uncertainty to the model was simply by including a certain amount of random noise, which decreases as we get closer to Election Day. Well, we've gotten to Election Day, and all the noise has been removed, leaving us with nothing but the polls shaping the model's outcome at this point. And the polls don't offer a lot of paths to 50, other than the Alaska/Colorado/Iowa double-bankshot approach.
Turning to the gubernatorial races, Democratic odds have also dwindled a bit here as well, though to more of a status quo, away from the potential big gains we were looking at a month ago. Odds of gaining seats fell from 57 percent to 48 percent since last Thursday, and the median number of gubernatorial seats fell from 22 to 21, which would reflect no change from current conditions.
If you look at individual races and consider whether or not Democratic candidates are above the 50 percent mark or not in each race, though, we're still on track for a gain of one; Democrats would be on track to pick up Pennsylvania, Alaska, Kansas, and Maine, while losing their hold on Colorado, Massachusetts, and Arkansas. Florida, where Charlie Crist has had better-than-even-odds for most of the last month, just narrowly slid under water, from a 60 percent chance last week to 47 now.
Florida and Maine are the two absolutely closest races, much more so than any Senate race. They're both on track to be almost tied; our trendlines project Rick Scott leading in Florida 46.8-46.5, and Mike Michaud leading in Maine 44.1-43.9. These races are both complicated by third-party candidacies, which keep the numbers from adding up to 100.
In Maine, Michaud may get a last-minute boost from independent candidate Eliot Cutler's decision to stand down; Cutler didn't exactly drop out, and at any rate he was way too late to get his name removed from the ballot (in fact, many people had already cast early votes before Cutler's semi-withdrawal). But Cutler did encourage his supporters to consider looking elsewhere, and Cutler's most prominent backer, Angus King, promptly shifted his support to Michaud. That may help get Michaud over the top. What happens in Florida, though, is anyone's guess; Libertarian candidate Adrian Wyllie isn't dropping out, and Scott and Crist both remain unpopular. You might start steeling yourself to seeing the words "Florida" and "recount" in close conjunction again.
12:49 PM PT: One last update: we ran the most-likely outcomes again, as we did several weeks ago. Today, the single likeliest outcome is the worst-case scenario, where the Democrats win only the safe seats and lose everything else (including NC and NH), for a net of 45 seats; this happens in 12 percent of simulations.
However, there are many more paths to get to 46, 47, 48, or 49, which is why the model points toward 47 as the median. For example, the second most common pathway, which happens in 10 percent of simulations, involves the Democrats losing AK, AR, CO, IA, and LA, not picking up KS, and holding NC and NH, for a net of 47.
The most common scenario where the Democrats hold an unexpected seat is the fifth-most common one, which happens in 3 percent of simulations, where the Dems hold IA (but not AK or CO) in addition to NC and NH. Finally, you have to drill all the way down to the 29th-most common scenario before you find one where the Dems win (happening in one-half of one percent of simulations); interestingly, it doesn't involve Orman being decisive and tipping it toward the Dems. Instead, it involves the Dems holding AK, CO, and IA, and then Orman winning and caucusing with the Dems anyway because they're already in the majority, for a net of 51.