Will polling prove accurate in 2014?

None of the high-tech, number-crunching forecast models seeking to determine which party will win a Senate majority will be useful if the raw material — individual polls — are wrong. But that could be the case this year, as Carl Bialik reported in an important piece today.

Many people distrust polls. How can a thousand respondents represent the views of millions? Aren’t many pollsters biased? Widely divergent polls — just yesterday, one survey showed a 10-point spread in the Senate contest in Kansas while another showed a dead heat — help fuel skepticism.

Actually, polls have been very reliable. That’s why Nate Silver, HuffPost Pollster and others have been able to predict the outcomes of recent elections. Polling averages generally provide an accurate reflection of candidate preferences just before elections. We shouldn’t ask polls for more than they can deliver. Polling about 2016 presidential preferences now, when voters haven’t thought about the race, is a waste of time. The same goes for many policy questions about which voters have no real opinions, in which case their answers are heavily influenced by question wording. But ask a representative sample of voters in October about an election in November, and the results generally provide a good prediction.

Or used to.

Fewer nonpartisan polls are being conducted. Response rates are down. New techniques, such as online surveys, and technological challenges, such as the switch from land lines to mobile phones, have introduced potential trouble. In addition, pollsters believe that it’s harder than usual to estimate the composition of the electorate this year, according to Bialik. That’s because polling firms have less experience in many of the 2014 Senate battlegrounds (for example, Alaska) than in presidential swing states such as Pennsylvania or Ohio. Midterm elections, which have lower turnout, are always harder to predict than presidential contests in any case.

Because we don’t know how much uncertainty these elements introduce, we can’t really gauge their effects. Estimates may prove less precise, which means we would need a larger estimated lead to be confident that a particular candidate is really poised for victory. What’s more, it’s possible that polling error might systematically favor one party or the other (say, if pollsters underestimate turnout of one partisan group of voters, or if a demographic group loyal to a party changes its behavior, causing it to be underrepresented in polls.

To put this in perspective, current polling averages are within four percentage points for Senate races in Alaska, Arkansas, Colorado, Georgia, Iowa, Kansas, Kentucky, New Hampshire and North Carolina. Some of those contests (such as Colorado, which is a dead heat) could go either way — even if the polling is correct. But if something is systematically wrong, then 4-point contests such as New Hampshire (where the Democrat leads) or Kentucky (where the Republican leads) could turn out very differently, even if the polls don’t change before Election Day. I wouldn’t bet on Republican Scott Brown in New Hampshire or Democrat Alison Lundergan Grimes in Kentucky based on the polling to date, but I also wouldn’t be surprised if the trailing candidate ended up winning in any of these closely contested races.

Bottom line? The polls are still likely to be correct. And the uncertainty makes relying on polling averages even more important. But I don’t expect to know which party will have a Senate majority — however large — until the votes are counted (or even later).

[Cross-posted at Bloomberg View]

Our ideas can save democracy... But we need your help! Donate Now!

Jonathan Bernstein

Jonathan Bernstein is a political scientist who writes about American politics, especially the presidency, Congress, parties, and elections.