Many Polling Problems in the Midyear Elections

By NATE COHN / New York Times News Service

With four months to go until this year’s midterm elections, perhaps the only thing clear about the fight for the Senate is that it will pose challenges to public polling.

There’s always the possibility that the polls could miss the outcome in a close contest. Polls have missed the result in three close Senate races in the last two cycles. But this year is particularly challenging. The rapid growth of partisan polls has contaminated the polling averages in states where surveying public opinion is already difficult. Many of these partisan polls employ dubious weighting and sampling practices. The combination will make it even harder for polls to nail the result.

So far this year, 65 percent of polls in Senate battlegrounds have been sponsored or conducted by partisan organizations, and an additional 10 percent were conducted by Rasmussen, an ostensibly nonpartisan firm that leans conservative and has a poor record.

The problem is partly economic. The cost of high-quality polling has surged, mostly because of declining response rates, just as the traditional sponsors of nonpartisan polls - local newspapers and universities - are facing revenue problems. Partisan organizations have largely filled the vacuum, and their incentive is to advance their message, not to sponsor costly, rigorous survey research.

The partisan polls pose a serious challenge to polling aggregators and election modelers. Traditionally, even a simple polling average tends to match the election results. But if the partisan polls are biased, and Democratic and Republican polls do not balance each other out, then the averages might be biased.

The possibility that partisan polls could skew the averages is perhaps best illustrated in Arkansas, where every Republican poll shows the Republican in the lead, and every Democratic poll shows the Democrat in the lead.

The five nonpartisan polls haven’t offered much clarity, with a net 16-point margin separating the most extreme findings. Most polling averages show the Republican, Tom Cotton, ahead because there are more recent Republican surveys.

The number of nonpartisan polls will increase as the election draws nearer, but the problems confronting pollsters won’t go away, given the unusual slate of states and contests this year.

Alaska, for instance, may be the hardest state in the union to poll. It is notoriously challenging, perhaps in part because of lower landline telephone coverage in far-flung Native communities.

In 2008, the polls were less accurate in Alaska than anywhere else in the country. They showed Mark Begich, then the Democratic challenger, leading Sen. Ted Stevens, a Republican, by a 10-point margin. In the end, Begich prevailed by 1 point.

Pollsters also have a mediocre record in Colorado, where public polls have underestimated Democrats in seven of eight statewide contests over the last decade. It’s been worse in the five contests since 2008, including the state’s 2010 Senate contest, in which the Democrat, Michael Bennet, trailed in the polls but prevailed by nearly 2 points.

With Mark Udall, a Democratic incumbent, ahead by 1.2 percentage points in the latest Huffington Post Pollster average - about the same poll margin as President Barack Obama had before he won the state by 5.4 points in 2012 - it’s fair to wonder whether the race is really this close.

Analysts have hypothesized that the Colorado problem is about Hispanic voters. There are certainly reasons to think the polls miss Democratic-leaning Hispanic voters: Many pollsters do not conduct bilingual interviews, and Spanish-speaking Hispanic voters are more likely than English-speaking Hispanics to lean Democratic. Hispanic voters are also less likely to have a landline.

On the other hand, Hispanic voters are only 14 percent of eligible voters in Colorado, and it is hard to generate a 4- or 5-point error out of such a small share of the electorate.

In North Carolina and Louisiana, there’s a different challenge: the candidates themselves.

The uncertainty in North Carolina revolves around a Libertarian candidate, Sean Haugh, a pizza deliverer who has never held public office, has not aired a single advertisement and has not reported a single dollar raised (despite soliciting contributions by bitcoin, which is apparently legal). He drinks beer in an unusual YouTube video announcing his candidacy.

Nonetheless, Public Policy Polling and Civitas have included Haugh in their surveys since he won the Libertarian Party nomination in May. The polls found that between 9 and 11 percent of voters support Haugh.

It is hard to imagine that this reflects authentic support, given that he is a largely unknown figure. Instead, his apparent strength probably reflects the well-known tendency of respondents to be more likely to select a third-party candidate who is named in a survey than to vote for that candidate.

Nonetheless, the decision to include Haugh in polls may become something of a self-fulfilling prophecy. His standing in recent polls produced a flurry of articles speculating on whether Haugh could spoil the chances of Thom Tillis, the Republican candidate. The North Carolina Association of Broadcasters, which will host a debate, said that the group would allow candidates to participate if they had more than 10 percent in the three polls closest to the day of the debate.

If Haugh continues to attract so much more support in the polls than he could be expected to win on Election Day, the effect could be to inflate the standing of Kay Hagan, the Democratic incumbent, in the polls. Haugh, then, is more likely to spoil the polls than Tillis’ chances.

(BEGIN OPTIONAL TRIM.)

A different problem plagues polls in Louisiana. The Nov. 5 election is a so-called jungle primary, when multiple Republican candidates will appear on the ballot. To win outright, a candidate must receive at least 50 percent. Otherwise, the state will hold a December runoff between the two candidates receiving the most votes.

The sheer number of Republican candidates increases the number of undecided voters. Some Republicans don’t know which Republican they’ll support. Some persuadable voters open to supporting Mary Landrieu, the Democratic senator, might wait to make up their mind until they’ve surveyed a vast field.

The five polls of the jungle primary so far this year have found that an average of 20 percent of voters are undecided. That number will probably remain higher than in other states all the way through the election, increasing the uncertainty about the outcome.

(END OPTIONAL TRIM.)

In spite of all of these challenges, the polls might still be fairly accurate this November. A large number of imperfect polls can, together, produce meaningful information if the errors tend to cancel each other out.

But even if the polls are as accurate as they have been lately, it would not be unusual for the polling averages or models to miss one or two Senate races, as they did in both 2010 and 2012.

This year’s fight for control of the Senate could easily prove to be closer than the 2012 presidential election or the 2010 and 2012 national Senate contests, which were all decided by the margin of several states.

If the fight for Senate control remains as close as it is today, even the typical modest errors of the last few cycles could easily lead the aggregators and models to miss on the overall control of the Senate.