If there’s a race that keeps us awake at night, it’s Alaska. The state is home to one of the most important Senate races in the country, but it also has a history of quirky and often inaccurate polling.
Until late last week, the main challenge had been a lack of polling. In the past 20 days, just four nonpartisan polls have been released in Alaska — mostly by firms with middling to poor pollster ratings. Most of the competitive states we’re tracking have had at least twice as many surveys.
So, the polls in Alaska were scarce, but the ones we’d seen told a fairly consistent story. Whether conducted online, by automated script or by live-telephone interview, all of the nonpartisan polls since Labor Day had Republican Dan Sullivan ahead by 3 to 6 percentage points against the Democratic incumbent, Sen. Mark Begich.
All that changed with two new surveys. On Friday, Hellenthal & Associates released a poll that had Begich with a 10 percentage-point advantage — drastically improved from a 5-point deficit in mid-September.
The Hellenthal poll had looked like a huge outlier — until late Monday, when another local firm, Ivan Moore Research, released a poll showing Begich up by either 7 or 8 percentage points (depending on which of the poll’s two turnout models used).
Has Begich made a comeback? Perhaps. His chance of pulling out the race is up to 34 percent in the FiveThirtyEight model, up from a low of about 20 percent earlier this month.
But there are many reasons to be skeptical. Public sentiment almost never shifts so much so quickly in a general election, especially this late in the race, without major precipitating news events. A candidate might gain a couple of points over a month on the basis of a superior campaign — but usually not more than that. The Ivan Moore and Hellenthal polls also applied small sample sizes, and showed results in Alaska’s gubernatorial election and at-large U.S. House election that differed from the polling consensus.
In addition to the Ivan Moore and Hellenthal polls, two other surveys of Alaska have been released since Friday. YouGov published a poll last weekend that had Sullivan 4 percentage points ahead — down only slightly from a 6-point lead he held in its late-September poll. Meanwhile, a poll from Harstad Strategic Research, for the Democratic group Senate Majority PAC, showed a tied race, 44-44. From a nonpartisan pollster, that would be a good result for Begich — but Harstad’s polls have had a Democratic-leaning house effect of about 3 percentage points this cycle. In general, when the best a partisan pollster can do is to release a poll showing a tie, it’s a good bet that a nonpartisan pollster would show its candidate behind instead.
In some states — say, Iowa — the solution would be relatively simple. You’d defer to the local pollsters, especially if more than one was showing the same result. But Alaska probably isn’t one of those states. The average polling error in statewide races in Alaska since 1998 is 7.2 percentage points. The error in any one year has never been less than 3.3 percentage points.
Is the FiveThirtyEight model — which adjusts for house effects and weights polls based on their past accuracy — doing a good job of sorting all this data out? We can think of a couple of reasonable reactions. On the one hand, you might argue that this is a case of garbage in, garbage out. On the other hand, you might find a statistical model to be especially valuable when the evidence produces such cognitive dissonance.
The FiveThirtyEight model, in addition to estimating which candidate is ahead, also evaluates the uncertainty in each forecast based on factors including the volume and consistency of polling in each state and the number of undecided voters. The uncertainty is very high in Alaska. The 90 percent confidence interval on its forecast in Alaska ranges from a Begich win by 6 percentage points to a Sullivan win by 9 points — a 15-point span. By contrast, the 90 percent confidence intervals in Colorado and Iowa, which have more abundant and more consistent polling, span a 9-point range. The high uncertainty is accounted for in the win probabilities the model establishes for Begich and Sullivan.
What if you don’t trust the polling at all? Then you might look toward the “fundamentals” of the state. At least in the FiveThirtyEight model’s view, these favor Sullivan. Begich was elected quite narrowly in 2008, a much better year for Democrats than this one. Alaska is not friendly terrain for Democrats. Sullivan has raised about as much in individual contributions as Begich — both candidates had brought in about $5 million as of Sept. 30 — which is rare for a challenger against an incumbent. The Democratic incumbents facing the most similar circumstances to Begich — Sens. Mary Landrieu of Louisiana and Mark Pryor of Arkansas — are trailing in polls of those states.
Furthermore, although Alaska has had very inaccurate polling in the past, just about all of the misses have been in one direction. Surveys there have overestimated how well Democrats would do.
There are some counterarguments to this. Although past polls of the state have underestimated how well Republicans would do, they’ve also underestimated the performance of incumbents.
And what about Begich’s narrow win in 2008 — historically a sign that a candidate might have trouble in his next re-election bid? Yes, it came in a historically strong year for Democrats — against Republican incumbent Sen. Ted Stevens who, at that time of that election, was a convicted felon. (Stevens’s conviction was later reversed.) But it also came against an incumbent who had been in office since 1968 in a year when the then-popular Alaska Gov. Sarah Palin was on the ballot as John McCain’s vice presidential nominee.
Alaska is also small enough that a decent “ground game” can make a big difference. As Nate Cohn of The New York Times mentioned, the YouGov survey found that an absurd 43 percent of registered voters have reported being contacted by the Begich campaign. He only needs 7,500 votes to turn a 2.7 percentage-point loss into the slimmest of wins. Polls can have difficulty picking up this small of a difference.
All things considered, neither of us would wager much on either side of the betting line established by the FiveThirtyEight model, which still makes Sullivan a modest favorite. But the chances Begich will return for a second term have improved. The polling in Alaska, along with improved polling for the Democrat Michelle Nunn in Georgia, have helped to offset a generally bad run of polling for Democrats in other states and helped to keep their chances of retaining the Senate alive.