Here’s something to think about when looking at Nate Silver’s column on skewed polls. Going back to 1990, the elections that did not turn out as expected wound up having the worst skew in the polls. Now, that might seem like a tautological observation, since expectations are formulated largely by looking at polling data. But, the premise here is that bias can sometimes be introduced into the polls based on outside factors, including the meta-media narrative about which party is supposed to win or lose.
Going into the 1994 midterm elections, the Democrats had controlled the House of Representatives since 1955, and for all but two single congresses since 1933. It was difficult to imagine the Republicans winning the House, and even harder to foresee the absolute landslide that actually occurred. It’s not surprising that the polls in 1994 were skewed three percentage points in the Democrats’ favor. Was it a modeling problem or a lack of imagination on the pollsters’ part?
The 1998 midterms were dominated by l’affaire Lewinsky, which was more of a Beltway obsession than a national one. The polls skewed four points toward the Republicans, which was the biggest skew (for either party) in the time period we are examining here.
The 2002 midterms are the ones I remember, because the polls were much too optimistic for the Democrats. We seemed to lose every close race, including ones in which we had been narrowly favored. Perhaps the New York/Washington-based media simply failed to understand how popular Bush was at that time in the rest of the country? Perhaps the pollsters couldn’t calculate the impact of war fever? The result was the biggest skew (4%) in the Democrats’ favor in this time period.
The 2006 elections were expected to be good for the Democrats, and they were, but the polls still skewed two points Republican, slightly underestimating the ultimate size of the Democrats’ victory.
And then in 2012, the polls skewed three points Republican, which made the Obama/Romney race seem closer than it ever was.
Silver is correct that the skew goes back and forth, sometimes favoring Democrats and sometimes favoring Republicans, but the one pattern I see is that the polls tend to skew in the direction of the media narrative or, at least, hew closer to what seems imaginable than what actually takes place.
In 1994, it was assumed that the Democrats simply owned the House of Representatives. In 1998, it was assumed that the nation shared David Broder’s revulsion with the president’s adultery. In 2002, the elite media may have lost a feel for the sentiments of the heartland. In 2012, the media may have been willing a competitive race that never really existed.
So, this year, if there is to be a skew, which way is it likely to swing? The media has been giving us a steady drumbeat of predictions that the Republicans will win back the Senate. They predict low voter turnout by Democrats and they suggest that the president is very unpopular and will be a drag on the tickets. It hardly seems possible that they’ve underestimated the Republicans’ strengths .
My prediction: either the polls are accurate or they have a Republican bias. I think we can rule out a Democratic bias, despite Nate Silver’s caveats.
In the twelve elections since 1990, only four elections have seen unskewed polls. That’s thirty-three percent. Does that mean that there is a 67% chance that the polls are skewed toward the Republicans?
That kind of math will probably give Nate Silver an embolism, but I do think the media narrative has had an historical role in biasing the polling data.