Will the 2008 election be polling's "perfect storm"? Pollsters rarely say it in so many words, but when they compare notes these days, worry is the prevailing theme. Three big challenges loom that threaten to throw off survey estimates for the matchup between John McCain and Barack Obama.
George Stephanopoulos, the host of ABC News' Sunday morning show "This Week," summed up pollsters' concerns with a list of "three different undercounts" that he attributed to Democratic pollster Peter D. Hart. This year's polls, he said, "may be undercounting the number of young people who are going to vote," they "may be undercounting the African-American turnout" and they "may not be capturing those white voters who just won't vote for Barack Obama because he's black."
Stephanopoulos delved indirectly into the methodological issues I want to explore in more depth in this space between now and the election. I see three big technical worries on the minds of pollsters:
• Cell-Phone-Only Voters. Political pollsters have traditionally relied on telephone surveys to reach samples of voters with landline telephones. Four years ago, according to exit polls, just 7 percent of voters lived in "cell-phone-only" households -- those with cell phone service and without landlines. They tended to be younger and more likely to support John Kerry, but since their numbers were still relatively small, pollsters were able to eliminate any bias by weighting their samples by age.
Since 2004, the cell-phone-only population as a percentage of all adults has more than tripled and now includes nearly a third of adults between 18 and 24. With younger voters expressing overwhelming support for Barack Obama, are pollsters that depend on landline samples understating Obama's support?
• The Bradley-Wilder Effect. I have written about this issue here before, but the critical idea is not that some voters "just won't vote for Barack Obama because he's black." Rather, what pollsters fear is that in the context of a survey interview, some respondents may fail to tell the truth about their preferences due to some "social discomfort" arising from Obama's race.
In the 1980s and early 1990s, in a series of contests between black and white candidates -- including the historic gubernatorial candidacies of Tom Bradley in California in 1982 and Doug Wilder in Virginia in 1989 -- polls typically understated support for the white candidate. The black candidates would receive roughly the same percentage of the vote on Election Day that they did on the last poll, while the white candidates did surprisingly better than the polls predicted. Handicappers started to assume that most of the undecided vote in such contests would "break" for the white candidate.
Over the last 10 years, according to a paper [PDF] by Harvard post-doctoral fellow Daniel Hopkins that studied 133 statewide races between 1989 and 2006, the apparent polling bias in such races largely disappeared. But can we assume that Bradley-Wilder will remain in remission this fall?
• Likely Voter Models. The primary battle earlier this year between Obama and Hillary Rodham Clinton set all sorts of records for turnout. And that higher turnout helped reshape the demographics of the primary electorate. Exit polls showed that 18-to-29-year-olds and women made up a consistently greater portion of the primary electorate than they did four years earlier.
Will something similar happen on Nov. 4? Will, as many speculate, younger and African-American voters turn out in sufficient numbers to alter the demographic composition in ways that take the pollsters by surprise? And will the assumptions of their "likely voter models" be sensitive enough to accurately capture any such turnout wave, if it occurs?
All three of these issues have been hot topics lately, everywhere from network news shows to political blogs. Everyone has a theory or seems willing to speculate with conviction about what might happen.
Over the next few weeks, I hope to make the case that we know more than we think we do about these issues. Pollsters are collecting data right now that will help us make empirical judgments about the potential for each source of polling error on Nov. 4.
Meanwhile, we can probably dispense with the "Perfect Storm" analogy. In the movie of the same name, three different weather-related phenomena combined to produce a storm of exceptional severity. In this case, as Democratic strategist Joe Trippi pointed out in September, the potential polling foibles may work in opposite directions and "cancel each other out." A return of the Bradley-Wilder effect would work to McCain's benefit, while an underrepresentation of younger, African American or "cell-phone-only" voters will likely benefit Obama.