Lies, Damn Lies, And Political Polls?Richard S. Dunham
Since Labor Day, Presidential polls have been consistent about one thing: Bill Clinton is ahead of Bob Dole. But when it comes to the size of the lead, surveys have fluctuated wildly, ranging from a 6% Dole deficit to a Clinton blowout by 25%. A CNN/USA Today/Gallup poll boosted Clinton's lead from 9% to 22% in just two days. "It kind of shakes your faith in polling a little bit," sighs Thomas H. Silver, publisher of The Polling Report, a newsletter on political trends.
The erratic polls have some Republicans grousing that the public is getting a distorted picture of a race that many pundits are already declaring over. Citing a recent first-place finish for a Louisiana Senate candidate who was all but counted out by the polls, the GOP claims Dole has a better shot than many media-commissioned surveys suggest.
Polling pros concede there's some truth to GOP critics' gripes. A proliferation of error-prone quickie surveys, methodological differences between polls, and rising voter volatility all can skew results. "It's standard for every loser to complain about how a poll is conducted," says Claibourne Darden, an independent pollster in Atlanta. "But sometimes, the complaint is valid."
For flagging campaigns, the accuracy debate is more than academic. Bad polling news can hurt fund-raising, grassroots enthusiasm, and turnout. "No question, your leadership and supporters get disheartened by constant negative publicity," says Republican National Committee Chairman Haley Barbour.
A problem for all pollsters is questioning a valid cross-section of voters. It's an even greater challenge for the one-night telephone polls increasingly trumpeted by the media. Among the hazards: Some people don't have telephones or are not at home; others refuse to respond. Moreover, small polls have a high sampling error, so a candidate's purported 12% lead could be as large as 19% or as small as 5%.
Then there's the built-in polling bias that favors any incumbent. Larry Hugick, media polling director at Princeton Survey Research, says some voters don't like to tell pollsters they oppose the incumbent, preferring to say they are undecided. On Election Day, these voters tend to break in favor of the challenger. "Dole is going to do well among people who tell us they're undecided," Hugick predicts.
FAULTY MATH. The GOP's big complaint concerns formulas predicting how likely a respondent is to vote. These turnout models, and other weighting factors that compensate for overrepresented or undersampled groups, are fraught with potential for error. In the Sept. 21 Louisiana open Senate primary, polls showed Republican Woody Jenkins running a poor third. But he ended up first because of heavier-than-predicted turnout by GOP activists.
Republicans say many turnout models favor Democrats. For 1996, most Republicans are using as a predictive model the 1994 congressional elections, when GOP voters were far more likely to vote than Democrats. But press pollsters prefer the 1992 election model, when Democratic turnout was higher. "We won't know until the election who is right," says Democratic pollster Anna Bennett.
But even if public polls overstate Clinton's lead, history offers Dole little solace. Since 1952, the poll leader on Labor Day has gone on to victory on Election Day. Republicans may indeed have a valid complaint about the methodology of many headline-making surveys. But it will take more than perfect polling models to rescue Dole's campaign.