Wednesday, November 7, 2012

Why All the Polls Were Wrong

How can the vote counts be so far off in so many states?

That’s a question that some of Fox News’ pundits were asking as the election results came in last night.

Those of us who live in a “reality-based” political world have a different question to ask this morning: How did it happen that all the political polls were wrong?

For that matter, how could all the pundits be wrong? From CNN to BBC, for the last six weeks and as late as the late afternoon of Election Day, the mainstream media was looking at the election using terms such as “knife edge” and “closest ever.”

We know some of the answers. Journalism has a built-in story bias. If there is any angle to show novelty or suspense, journalists will play that up. It’s a habit a journalist cannot avoid, because it is the “good stories” that are rewarded with reader attention. Similarly, pollsters have an incentive to show that an election is close. When it is, clients pay for more polls. The frequent presidential polls in my state of Pennsylvania dried up in September when it became clear that Pennsylvania was not in play.

Aside from that, some of the poll releases were just false. Republican pollsters, we have heard, intentionally put out adjusted poll numbers to show that Romney had a chance of winning after the polling started to show definitively that he did not. Looking back, it is easy to imagine that there was similar contrivance in some of the senate-race polling. On the Democratic side, fund-raisers had an incentive to emphasize the polls that showed their side with a narrow lead. Supporters wouldn’t give with so much urgency to a cause that might win in a landslide.

If there is a built-in bias in journalism, news publishing has a much stronger bias, which is to support its advertisers. If the advertisers aren’t happy, there won’t be enough money to pay the journalists and other staffers. The corporate executives who make the biggest advertising decisions for television, newspapers, and other media lean heavily Republican. News outlets have to present a face that is friendly to the point of view of their sponsors, even when this is at odds with reality. This is something we see every day, and we have seen it especially in this fall’s political coverage. There were plenty of analysts to be found, even with a cursory web search, to explain why Obama had such an easy path to victory, but the sponsors would not have been happy if these were the pundits who appeared on the news channels.

These are the obvious answers for a “too close to call” election that turned out not to be close at all. But the more important answers are not so obvious. Pollsters in the last three weeks have a strong incentive to predict an election. The more predictive power their late polls show, the more credibility they will have next time. Yet last week, there were still polls showing Pennsylvania as a tie in the presidential election — not accurate at all in a state that the Democratic candidate carried by a five-point margin. Similarly, across the country, polls on average gave Republicans a boost of two or three points, an edge that wasn’t there in reality. Reality has a liberal bias, perhaps, but it is clear, now that many of the votes have been counted, that the mainstream polls, essentially all of them, had a Republican bias all year long. How can this be?

Pollsters imagined more Republicans than exist in reality, but how? Reputable pollsters cannot directly adjust their results based on party identification numbers, because that is one of the results they are trying to measure from day to day. Pollsters must, though, adjust for more static measures, and it is hard to get this right. One Michigan pollster early on showed Romney carrying that state. That result was produced by assuming that the average age of voters casting a ballot would be 55. Republicans are older than any other party in the United States, with an average age of nearly 60, so imagining lots of old voters was enough to give that party the edge.

There is a special problem with having a party whose voters are so old: its numbers tend to decline from pure demographics. Grand Old Party voters are more likely to die of the diseases of old age than the voters of any other voting bloc. From demographics, you would predict that Republicans would lose about 0.5 percent of their voting power every year. Republicans compound this difficulty by having less-than-healthy lifestyles, with higher rates of every lifestyle-related disease from drug abuse to diabetes, so the rate of decline could even be faster than 0.5 percent per year.

But even the pure demographic rate of decline, 2 percent every presidential election, is so shockingly fast in the context of American politics that I think political pollsters are expecting something to keep it from happening. If pollsters are discounting the Republican demographic decline from 2008, that would explain most of the difference between the national presidential polls and the actual election vote counts so far.

Republican strategists are proud to say that their political approach is not “reality-based.” Yet the decline in their numbers is a reality they must come to grips with in some fashion, or their party will just continue to fade.