Traditional Polling Methods Do Not Work for Nontraditional Candidates

01/30/2008 01:12 pm ET Updated May 25, 2011

The New Hampshire polls indicated that John McCain would handily beat Mitt Romney, with the rest of the candidates coming in as also-rans. That's exactly what happened. Barack Obama was said to be ahead of Hillary Clinton by double digits. All the polls said so, including the internal polls of the Obama and Clinton campaigns. The polls were wrong, and Clinton won the New Hampshire primary.

Why were the polls so right on the Republican side and so wrong on the Democratic side? One possibility is that the polling methods used in New Hampshire were fine. Something happened in the last few days that overcame their predictions. That is where all of the pundits and media went. They said that Clinton overcame a double-digit deficit in a day or two. Perhaps her emotional display shortly before the primary made her a more sympathetic figure. Perhaps she did better in the last debate than Obama. Perhaps Clinton got her people to the polls whereas Obama did not. Maybe Obama's lack of experience finally registered with the notoriously independent voters of New Hampshire.

The same kind of thing happened in Nevada. The polls predicted a Romney victory and it happened. Clinton and Obama were supposed to be neck in neck. Instead, Clinton beat Obama. Now we hear that Obama's supporters didn't make it to the polls. We hear that the rough and tumble of the Clinton team damaged Obama. Once again, the polls were accurate measures but late developments changed the predicted outcome.

We move on to South Carolina and Florida. McCain was predicted to win Florida with Romney a close second. That is what happened. And what of Clinton and Obama? As in Nevada, the two were supposed to be neck in neck in South Carolina. Obama crushed Clinton. Now we hear that voters were offended by the Clintons' negative tactics (the same that seemed to have worked in Nevada), particularly as they may have regarded "race." In all of these cases, the polls are presumed to be correct with last minute events accounting for their inaccuracies. Somehow, these last minute events always affect the Clinton and Obama predictions but never the McCain Romney predictions.

In science, there is always a second possibility when a measure fails to predict a behavior: The measure was off. If that is the case, then nothing special happened in the last few days of any of these races. The measure was not accurately measuring voting behavior to begin with. The pundits have mentioned one possibility of this sort. People may have been dissembling to the pollsters. Past upsets of the sort that took place in New Hampshire (for example, the Bradley electoral defeat in California a while back) have a disturbing factor in common: In each case, the polls had African American candidates comfortably ahead, but the African American candidate unexpectedly lost. Maybe people told the pollsters they would vote for Obama but in the privacy of the voting booth they did not do what they told the pollsters they would. To put it bluntly, unadmitted racism raised its ugly head. In South Carolina it happened in reverse but now it was racial pride. After all, as Bill Clinton said, Jesse Jackson won South Carolina twice in the 1980s. About half the voters are African American. Perhaps they just voted for one of their own, Barak Obama, and did not admit they would do this to the pollsters.

But there is a second way a measure can fail to predict behavior that no one has commented on. There may be something wrong with the measure itself. A great deal of psychological research has shown that what people say about gender and race does not always match how they behave toward women and African Americans. This is not because people are lying--they genuinely believe they are not sexist or racist. And on the surface most of them (and us) are not. Psychologists understand this attitude-behavior discrepancy in terms of explicit and implicit processes.

Traditional polls measure explicit processes. They measure how people say they feel about race and gender--how people think they will vote. But they do not measure implicit, underlying attitudes. Psychologists repeatedly find that white people who report they have no racial prejudice will still act less comfortably in the presence of an African American than in the presence of another white person. Ditto for other racial attitudes and behaviors, and for gender-related ones as well. Such discrepancies can be strengthened by surreptitiously bringing up racial or gender stereotypes (a procedure called "priming" by researchers).

These implicit attitudes and priming effects might not show up in polls, but they are readily detected by other kinds of measures (such as differences in response time to race-related words or reactions to gender-related stimuli people are unaware of having experienced). There are many such examples in the psychological research literature (some are reviewed in Drew Westen's book The Political Mind). Most important as Super Tuesday and the November election draw near, these findings hold for most of us, not just those who are overtly sexist or racist.

What this all means is that traditional polling might be wonderfully predictive in a traditional race (i.e., two Caucasian men), but woefully inadequate in 2008 when we have the historically unprecedented case of an African American and a woman competing. Extraordinary times require new, innovative methods, and traditional polling alone just won't be enough. Traditional polls will not work for non-traditional candidates.

If I were you, I would take whatever the polls say about Clinton vs. Obama on Super Tuesday with a grain of salt. I would readily accept their findings for the Romney McCain race however.

Joel Weinberger, PhD

Robert F. Bornstein, PhD

Derner Institute of Advanced Psychological Studies

Adelphi University, Garden City, NY