The FiveThirtyEight Senate model is launching Wednesday. We’ll be rolling it out in stages, with additional features, functionality and further methodological detail. We’ll also be unveiling our new set of pollster ratings and publicly releasing our database of all the polls used to calculate them. So there’s a lot more to come.
But if you’re looking for a headline, we have two. First, Republicans are favored to take the Senate, at least in our view; the FiveThirtyEight forecast model gives them a 64 percent chance of doing so.
The reasons for the GOP advantage are pretty straightforward. Midterm elections are usually poor for the president’s party, and the Senate contests this year are in states where, on average, President Obama won just 46 percent of the vote in 2012.1 Democrats are battling a hangover effect in these states, most of which were last contested in 2008, a high-water mark for the party. On the basis of polling and the other indicators our model evaluates, Republicans are more likely than not to win the six seats they need to take over the Senate. This isn’t news, exactly; the same conditions held way back in March.
An equally important theme is the high degree of uncertainty around that outcome. A large number of states remain competitive, and Democrats could easily retain the Senate. It’s also possible that the landscape could shift further in Republicans’ direction. Our model regards a true Republican wave as possible: It gives the party almost a 25 percent chance of finishing with 54 or more Senate seats once all the votes are counted.2
Why so much uncertainty? Consider some of the challenges that election forecasters face this year:
- The quality and quantity of polling has been poor. We have stunningly few polls in some states. In Colorado, for example, which could easily determine the balance of the Senate, no polls at all were published in August. And in many states, most of the polls we do have are from non-traditional polling firms, including those that conduct “robopolls” or Internet polls, or which have an explicitly partisan affiliation.
- In contrast to 2010, when major swing states like Pennsylvania, Ohio and Wisconsin were key to the Senate picture, this year’s races take us all over the country, including to politically idiosyncratic states like Arkansas and Alaska. These states are harder to pin down: Arkansas has begun to vote strongly Republican for president, for example, but sometimes still elects Democrats to other offices. In certain other states, polls and partisanship diverge. Ordinarily in a year like this one, you’d give Democrats no chance at all in Kansas, for example. But polls show a somewhat close race there, and there’s further uncertainty because of the presence of an independent candidate.
- Readings of the national mood are ambiguous. President Obama remains unpopular — and the president’s party has a long history of performing poorly in midterm election years. However, Democrats remain roughly tied with Republicans on the generic congressional ballot. That may reflect the Republican Party’s poor image, which remains quite a bit worse than that of the Democratic Party. But it could also mean there’s some upside for Republicans. On average in midterm years since 1990, the generic ballot has favored the opposition party by 5 percentage points by Election Day instead of being even.
- We don’t yet have a good sense for the potential turnout and enthusiasm advantage. It’s reasonably safe to assume it will benefit Republicans; their (older, whiter) demographics are usually associated with higher turnout at the midterms. But we don’t have much evidence yet about the magnitude of this effect. One way of measuring the turnout edge is to compare polls that release results among both registered and likely voters in the same state. In 2010, Republicans polled a net of 6 percentage points better in the likely-voter surveys, a historic high. Compare that with a GOP advantage of about 2 percentage points on average in midterms from 1990 to 2006. This year, however, we’ve seen very few polls to release both registered and likely-voter results in the same state. That leaves us somewhat in the dark about whether 2010’s turnout pattern was a fluke.
Another complication is the broad Senate playing field. There are, according to our model, either nine or 10 races in which each party has at least a 20 percent chance of winning.3 There are several others in which each party’s chances are at least 10 percent.
The Republicans’ edge comes from an abundance of opportunity. They are almost certain to win the Democratic-held seats in Montana and West Virginia, and very likely to do so in South Dakota. That gives them three of the six seats they need. Beyond that, they have few guarantees but a lot of good prospects:
- Republicans are slightly favored, though far from certain, to oust Democratic incumbents in Louisiana and Arkansas.
- Four more Democratic-held seats — in Alaska, Colorado, Iowa and North Carolina — rate as tossups.
- While Democrats are favored in Michigan and New Hampshire, Republicans retain some chances to win those states as well.
Democrats, by contrast, have plausible chances to win only three Republican-held seats — in Georgia, Kentucky and Kansas — and they aren’t favored in any of those races.
In some ways, Republicans’ position resembles that of President Obama against Mitt Romney during the 2012 presidential campaign. Even though Obama’s edge was relatively narrow for most of the campaign in swing states like Ohio, Florida, Nevada, Virginia and Colorado, he needed to win only a couple of them to have a winning electoral map, whereas Romney, with fewer electoral votes in his base, would have had to almost sweep them.
In the end, Obama won almost all the swing states instead, save North Carolina. Something similar could happen in the Senate this year — most or all of the competitive races could break in the same direction. This is a reasonably common pattern: In 1994, for example, Republicans benefited from a late surge. In 1998, Democrats did (under President Clinton, which demonstrates that midterm surges don’t always work to the benefit of the opposition party). In 2006, Democrats gained ground in Senate races almost up to the last minute.
This is the advantage of having something like the FiveThirtyEight model — we can run thousands of simulations of these outcomes each day, accounting for the historical extent to which the results in different states are correlated with one another. How often, for example, do Republicans win six Democratic-held seats, but then get tripped up because they lose Georgia or Kentucky? How often does a state like Michigan or New Hampshire put the GOP over the top even though other states seem like easier “gets”? How often do Republicans sweep almost everything — or get shut out beyond Montana, West Virginia and South Dakota? These scenarios are hard to calculate by hand, but they’re the model’s bread and butter.
We’re aware that there are a lot of other Senate forecasting models. I think some are pretty good. I have methodological disagreements with others — but I’d trust them before I’d believe the typical television pundit. Still, a few of them share a flaw: They assume the result in each state is independent from the next one instead of being partially correlated.4 Models that fail to account for the correlation between states will have too narrow a spread in their outcomes and neglect the possibility that I mentioned before — a late surge by one party, as has occurred several times in recent elections.
The models also vary in the probability they assign to a Republican takeover. There are several others which, like ours, put Republicans’ chances somewhere in the range of 60 percent to 65 percent — and others which show Democrats with a slight edge. Soon we’ll be releasing a more detailed description of the FiveThirtyEight model and how it compares to the others. But most of the differences boil down to one or two factors.
First is whether the model treats polls of registered and likely voters in the same way. Our model doesn’t. This might seem like a trivial detail, but it isn’t — not when so many Senate races are so close. Historically, as I mentioned, likely-voter polls are more favorable to Republicans in midterm years. So our model shifts registered-voter polls toward Republicans to make them more comparable to surveys of likely voters. For the time being, this adjustment is based on the historical average of likely voter polls being 2 points or 3 points more favorable to Republicans than registered-voter ones. The model will automatically update its estimate as we collect more data on what the turnout gap looks like this year.
That’s enough to potentially change the outcome in several states. In Iowa, Colorado and Alaska, for example, we have Democrats very narrowly favored on the basis of the raw polling average — but their edge is so narrow that Republicans pull ahead slightly if we adjust the registered-voter polls in these states to a likely-voter basis.
The other consideration is which factors, if any, the model incorporates above and beyond polling. These fall into two broad categories: those that are associated with a particular state, and those that are applied to all races nationwide.
The state-by-state factors are not bad for Democrats — some are even pretty good. One variable that our model and several others incorporate is fundraising. On the basis of individual contributions from the public (not counting PAC contributions or outside spending), the Democratic nominee has raised more money than the Republican in almost all of the most competitive Senate races.5
Several models also incorporate various measures of candidate quality. Ours, for instance, evaluates it by considering the highest office the candidate has been elected to in the past. The FiveThirtyEight model also takes into account the candidates’ liberal-conservative ideology as measured by various systems and how it compares to voters in their states. Democrats have nominated some pretty good candidates by these measures, but the GOP has, too. In contrast to 2010 and 2012, when problematic candidates helped undermine Republicans’ chances, this year the party has nominated fairly moderate candidates in Iowa, Georgia, North Carolina and Alaska in lieu of more conservative alternatives.
Some models also incorporate national factors, like presidential approval ratings, or the historical tendency of president’s party to perform poorly at the midterms. Those that do so tend to show poorer results for Democrats.
It’s very possible to go overboard with these factors — similar models in presidential years have a mediocre track record. Our model does not look at presidential approval ratings, for example. It does, however, include a modest adjustment to account for the tendency of the president’s party to perform poorly in midterm years. (This is a new feature in the model this year.) Specifically, it assumes the generic ballot has some tendency to revert to the historical mean of favoring the opposition party by 5 percentage points by Election Day.
Is it too late in the political cycle to expect such a shift? Yes and no. Remember those late-breaking elections like 2006 — historically they’ve been somewhat more likely to favor the opposition party. But in many other elections, the state-by-state indicators have usually incorporated most evidence about the national mood by this point in the race. The effect of additional mean-reversion, as our model calculates it, is fairly minimal — it raises Republicans’ chance of a Senate takeover to 64 percent from 60 percent otherwise.
In some ways, this helps to explain the Democrats’ predicament. As measured by the generic ballot, the political climate is fairly neutral right now — better than it normally is for the president’s party at the midterms. But trying to defend the many red-state Senate seats they won in 2008 is a tall order under any circumstance. Republicans could win the Senate — with one seat to spare — by winning Alaska, Arkansas, Louisiana, Montana, North Carolina, South Dakota and West Virginia, all of which voted for Romney in 2012. Even in a neutral political climate, Democrats might lose several of those states. If the climate shifts further toward the historical average of favoring the opposition party, their losses could be much greater, including losses in purple states such as New Hampshire and Iowa and mildly blue ones such as Michigan.
Still, Democrats have plenty of campaigning left to do. Our forecast rates about 10 Senate races as still highly competitive, as compared with five to seven in most of the other models I’ve looked at. Some of this is because of the issues I described earlier. The margin of error is larger in states where there’s been little recent polling, or where those polls are of low quality. It’s larger in states where the polls disagree more with one another. It’s larger in states where the polls diverge from the “fundamentals” of the state — as in Kansas, for instance. All of these problems are pertinent this year. There hasn’t been much polling and the polls we do have are a middling lot. There are states like Georgia where recent polls show everything from a 9-point lead for the Republican, David Purdue, to a 7-point lead for the Democrat, Michelle Nunn. And a lot of the most important Senate races are in states with idiosyncratic demographics. This is the sort of year in which there are likely to be several missed calls — it would be a minor miracle if any of the models, certainly including ours, manage to go 35-for-36 or 36-for-36.
Nonetheless, Republicans have the more robust map and more tolerance for error. That’s why the odds are in their favor.