Wooing Purple America
How the Democrats can win again, if they really want to
Nov 15, 2004, Vol. 10, No. 09 • By JOHN J. DILULIO JR.
IN 2000, the polls had Bush winning the popular vote. He went on to lose it by more than 3.5 million votes. In 2004, pollsters on election eve said the race was "too close to call." The next day, exit polls predicted a comfortable Kerry victory. Then on election night, the Bush-Kerry national popular vote split turned out to be no squeaker, but 51 percent to 48 percent. So are the pollsters all wet?
No, just damp. Both in 2000 and in 2004, most polls, including exit polls, were correct within their margins of error. But let's all finally understand just how wide those margins are. If a poll predicts a 51-48 percent Bush-Kerry split with a margin of error of plus or minus 3 points, all it is actually predicting is an outcome somewhere in the vast territory between a 9-point lead for Bush (54-45 percent) and a 3-point lead for Kerry (51-48 percent).
Polling, never an exact science, is now an increasingly questionable art. Properly conducted, a poll can capture what 250 million citizens think by interviewing as few as 1,500 people. The key is random sampling: Every voter must have an equal chance of being interviewed.
Unfortunately, many polls this election season were based on smaller-than-ideal samples. With people trying to avoid telemarketers and using call-screening devices, pollsters apparently had more trouble than usual getting people to answer their calls and hence assembling reliable samples and results. To overcome such difficulties will cost more money than some news organizations and other purchasers of polls seem prepared to spend. Instead, this year some news organizations took a cheap short-cut by averaging results from competing polls. This is like trying to build a functional car motor by assembling parts from as many fine or faulty engines of as many different makes as you can find.
The Election Day exit polls were misleading, but so were the forecasts made many months ago (all before Labor Day) by political scientists. The same was true in 2000, when most election-forecasting models had Gore beating Bush handily (winning from 53 percent to 60 percent of the popular vote). This year, several of the most respected models had Bush winning 54 percent or more of the vote, and the median forecast for the seven leading models was 53.8 percent for Bush (including one model that had Bush at 49.9). That is "only" 3 points and change off from Bush's actual election-night tally; but then a 3-point shift from the actual winner to the second-place challenger would have changed the popular-vote victor in a half-dozen of the presidential elections since 1956.
Different models embody different assumptions and crunch different data, but most include polling results (garbage in, garbage out), presidential approval ratings (subject to wartime rally effects), and various measures of individual or national economic well-being (like the change in real GDP during the first two quarters of an election year, or job growth during the first 3.5 years of a president's term). In addition, some models incorporate incumbency-advantage measures and other variables.
Still, no model accurately predicted the 2000 presidential race, and none got this year's right, either. Following the 2000 forecasts, Larry Bartels of Princeton and John Zaller of UCLA combined features of 48 different models, and were thereby able retrospectively to "forecast" the 2000 presidential election within a point or two of the actual results. Summarizing the 2004 forecasts in the October 2004 issue of Political Science and Politics, SUNY-Buffalo's James Campbell argued that "the forecasts cannot be fairly judged by whether they predicted the candidate who won the election." After all, "each forecast expects to be wrong to some degree" because "there are unanticipated . . . developments in a campaign that cause votes to shift here or there." (Memo to my bookie: Pay me no matter what team I bet on next week, especially if anything unexpected happens during the game.)
But the pollsters and professors are models of intellectual rigor next to the media pundits. Remember their endless, self-confident commentaries on the coming Dean-Bush showdown? Or roll the old footage: A "backlash" against 12 years of Republicans in the White House elected Bill Clinton in 1992; but, in 1994, "angry white males" (obviously a backlash against the "backlash") put Republicans in control of the House for the first time in four decades; then in 1996, "soccer moms" (jilted by the "angry white males"?) reelected Clinton; and since 2000, "deeply divided" Americans have conveniently sorted themselves into "red states" and "blue states" (stay tuned for word on whether "security moms" love "NASCAR dads").