So, that was fun. One minute we’re promised a half dozen toss up races to determine control of the United States Senate, and the next Democrats are ducking under their desks as Massachusetts and Maryland elected Republican governors. Let the “Very Important Pundits” take turns on cable news assigning blame for the losses. I’m more interested in why the polls didn’t tell us the wave was coming.
Most of the electoral forecasts that aggregated polls predicted a likely Republican takeover of the Senate, but each race was supposed to be close. Alaska and Iowa were supposed to be 1-point races. Only 2-3 percent separated the candidates in Colorado. New Hampshire and North Carolina were close, but Democrats were favored in those races. And that was just the public polls.
Privately, Democratic pollsters were not nearly as bearish. They spoke of the problems polling Hispanics in Colorado, the native population in Alaska, and the new African American influx into the Atlanta suburbs. There was complicated talk about the messed up statistical modeling in the public polls. We’ll be fine in North Carolina, they assured all who paid to listen to their advice.
The big picture gave no hint of the coming electoral wave. Barack Obama was no more unpopular than he was in 2013. The country told pollsters they didn’t particularly favor either political party. Voters told pollsters they hated congress more than two-day hangovers.
Right up until the polls closed, all the king’s horses and all the king’s men were confident in their analysis that the election could go other way. Paul Begala, a smart man of good character, confidently told a national television audience that voters didn’t hate Democrats so much that they wanted to reward Republicans.
Lifeguards and surfers agreed; there was no wave coming. Then a tsunami came and wiped out the entire city.
The only reason the 2014 midterms weren’t worse is that Democrats had not won back all their losses from 2010. In other words, they didn’t have much left to lose, but Republicans nearly cleaned them out.
The Senate races in Alaska, Colorado, and Iowa were not close. Republicans won those big. North Carolina wasn’t particularly close either, but the Republican won that, too. You know what was close? Mark Warner barely won re-election in Virginia, a race that was on no one’s radar.
Democrats spent the entire election complaining that polls were skewed against them, when in fact they were skewed in their favor. According to Nate Silver, who hasn’t been wrong yet, the average Senate poll overestimated the Democrat’s chance by 4 percent.
This isn’t the first time this has happened, and before you start yelling about one party or the other, it happens to both. Polls were biased toward Democrats by 3-4 points in 1994 and 2002 and towards Republicans by 2-5 points in 1998, 2006, and 2012. In fact, it was the 2012 bias that probably led to the polling community to overcompensate by adjusting their turnout predictions for 2014, leading to a whole bunch of bad polls.
Here’s how that works. Imagine a state where 10 people live. If you assume six of them are white men over the age of 60, then you can assume that a Republican will win. If you assume six of them are African-American women of any age, then the polls will favor a Democrat. In extremely rudimentary terms, this is where mistakes begin.
These days, polls are aggregated and examined by guys like Silver, so there is a tendency to not want one’s poll to appear wrong by deviating from the consensus. These are called “outliers” and are immediately suspicious. So pollsters agree with each other’s assumptions, which is called “herding.” And that’s how they can all be wrong at the same time in the same way.
We’re never going to get this completely right because humans are at the root of it, though it shouldn’t take a brain surgeon to get a poll close enough to be useful in an election. If I’m going to see my team wiped off the map, I’d at least like to know it’s coming.
Jason Stanford is a regular contributor to the Austin American-Statesman, a Democratic consultant and a Truman National Security Project partner. You can email him at stanford@oppresearch.com and follow him on Twitter @JasStanford