How did the polls do last week? Quite well. While we worried about the many challenges, the telephone survey again defied the odds and delivered mostly accurate results.
The final margins of the national polls varied, as random sample surveys always do, but the averages of their results came very close to the ultimate outcome (as they should, assuming no big changes in voter preference between the final survey and the election). With an unknown number of provisional and late-arriving mail ballots still being counted, Barack Obama's margin in the national popular vote stands at 6.5 percentage points. Our final trend estimate at Pollster.com gave Obama a 7.6-point advantage, a margin identical to the final RealClearPolitics average.
The state-level public surveys were just as close on average, if not more so. In closely watched and heavily polled states like Colorado, Florida, Missouri, North Carolina, Ohio and Virginia, our final trend estimates came within about a percentage point of the unofficial results. In the 28 competitive states I watched most closely during the campaign, my initial comparison shows an average difference between the final polls and the results of just 0.2 percent.
The one unambiguous implication of these results is the absence of any clear "Bradley Effect." Remember, the Effect of old was about a failure of polling, not about the role race may have played in voters' decisions. Since the polls were on target and showed no consistent pattern of under- or overstating John McCain's support, we can say with confidence that the much-discussed Effect was not in evidence last week.
The accuracy of the final polls also tells us that other worries, such as the growing percentage of voters reachable only by cell phones or the potential that a wave of new voters might fly in under the radar of pollsters' likely voter models, did little to throw off pollsters' estimates.
We can assume that the likely voter models and screens were generally on the mark, although we can say little about how and why, since pollsters were as resistant as ever about disclosing the specifics of what they did. Adding to the confusion, some intriguing and persistent differences among national polls all but disappeared in the final week. The fact that the average of all polls came so close to the final margin tells us that whatever pollsters were doing, collectively, to identify the likely electorate was reasonably close to right.
In late September, we looked at the variation across national polls in terms of age composition, with most surveys showing 18-to-29-year-olds as a share of all voters falling somewhere between 10 and 18 percent, and most tending toward the higher end of that range. The national exit poll estimates that under-30 voters made up 18 percent of the electorate, just one percentage point higher than in 2004.
The exit also showed first-time voters as 11 percent of the total -- exactly the same as four years ago. While the exit polls may have issues of their own when it comes to measuring the age of the electorate, these initial results suggest that the wave of new voters was not what some had expected.
The general accuracy of state-level surveys last week also suggests that whatever difficulty pollsters had reaching cell-phone-only households did not significantly affect their vote forecasts. The national exit poll sample included a question on phone usage and found that 20 percent of those who voted on Election Day identified themselves as "cell phone only." As the Washington Post's Jennifer Agiesta reported, "the different demographics" of cell-phone-only voters, primarily their youth, explained most of the difference. If they took the landline Election Day voters and weighted those by age to match the full electorate, Obama would hold a 5-point lead. "That's preliminary, but not definitive, confirmation," Agiesta wrote, "that weighting by age mostly accounts for the lack of cell phones in many polls."
"Preliminary" is a word we should stress in thinking about the cell phone issue, since the Election Day exit poll sample did not include those who voted early or absentee. The exit pollsters did conduct a telephone survey to reach early voters nationally, but they sampled only landline phones. Since the Obama campaign made a concerted effort to get its supporters to vote early in many states, and since they were demographically more likely to be cell-phone-only, more analysis remains to be done. Still, the overall accuracy of state-level polling suggests that any cell phone effect was minimal.
Some of the exceptions to the overall patterns may turn out to be most interesting: Three of the biggest polling errors came in western states with large Hispanic populations -- California, Nevada and New Mexico. As of this writing, our trend estimates understated Obama's margin by roughly 6 points in New Mexico and 5 points in Nevada and California. At least one pollster (PPP) has pointed to an understatement of Obama's support among Latino voters as the primary source of the problem in Nevada. It will be interesting to see if other polls showed similar patterns.
But that is a topic for another column, and with this long and historic election now behind us, we have a lot of time to sift through the results. See you next week.