出口民调为何失准？——2004年大选中的“布拉德利效应”

November 2004 A lot of people are writing now about why Kerry lost. Here I want to examine a more specific question: why were the exit polls so wrong? In Ohio, which Kerry ultimately lost 49-51, exit polls gave him a 52-48 victory. And this wasn't just random error. In every swing state they overestimated the Kerry vote. In Florida, which Bush ultimately won 52-47, exit polls predicted a dead heat.

(These are not early numbers. They're from about midnight eastern time, long after polls closed in Ohio and Florida. And yet by the next afternoon the exit poll numbers online corresponded to the returns. The only way I can imagine this happening is if those in charge of the exit polls cooked the books after seeing the actual returns. But that's another issue.)

What happened? The source of the problem may be a variant of the Bradley Effect. This term was invented after Tom Bradley, the black mayor of Los Angeles, lost an election for governor of California despite a comfortable lead in the polls. Apparently voters were afraid to say they planned to vote against him, lest their motives be (perhaps correctly) suspected.

In theory, exit polls ought to be very accurate. You're not asking people what they would do. You're asking what they just did. How can you get errors asking that? Because some people don't respond. To get a truly random sample, pollsters ask, say, every 20th person leaving the polling place who they voted for. But not everyone wants to answer. And the pollsters can't simply ignore those who won't, or their sample isn't random anymore. So what they do, apparently, is note down the age and race and sex of the person, and guess from that who they voted for.

This works so long as there is no correlation between who people vote for and whether they're willing to talk about it. But this year there may have been. It may be that a significant number of those who voted for Bush didn't want to say so.

Why not? Because people in the US are more conservative than they're willing to admit. The values of the elite in this country, at least at the moment, are NPR values. The average person, as I think both Republicans and Democrats would agree, is more socially conservative. But while some openly flaunt the fact that they don't share the opinions of the elite, others feel a little nervous about it, as if they had bad table manners.

For example, according to current NPR values, you can't say anything that might be perceived as disparaging towards homosexuals. To do so is "homophobic." And yet a large number of Americans are deeply religious, and the Bible is quite explicit on the subject of homosexuality. What are they to do? I think what many do is keep their opinions, but keep them to themselves.

They know what they believe, but they also know what they're supposed to believe. And so when a stranger (for example, a pollster) asks them their opinion about something like gay marriage, they will not always say what they really think.

When the values of the elite are liberal, polls will tend to underestimate the conservativeness of ordinary voters. This seems to me the leading theory to explain why the exit polls were so far off this year. NPR values said one ought to vote for Kerry. So all the people who voted for Kerry felt virtuous for doing so, and were eager to tell pollsters they had. No one who voted for Kerry did it as an act of quiet defiance.