Editor's note: This article is reprinted from The Denver Post, Sunday, Nov. 24.

Elections that defy historical precedent are hard on the business of pundits and pollsters. The 2002 mid-term elections were expected, as they historically have, to go against the incumbent president’s party. At best it was assumed that the stalemate created during the 2000 presidential election would continue. But once votes were tallied, it became clear that, in fact, President Bush helped carry the day.

Despite a late-changing political environment and several close senate and gubernatorial races, local and national polls performed well and managed to make mostly accurate calls within the margin of error. However, it was the spectacular misses that received most of the attention.

From Colorado’s perspective, one of the most memorable scenes of the 2002 election was Tim Russert on Meet the Press Sunday before the election announcing the latest MSNBC Zogby poll showing Democrat Tom Strickland ahead of Republican Wayne Allard by 9 points, 53 to 44 percent. The panel of premier beltway pundits picked Tom Strickland to win. They reasoned that an incumbent with late election support in the low 40-percent level could not win because late-deciding voters usually go to the challenger. The conventional wisdom proved wrong for the challenger.

Despite Zogby’s national reputation as a first-rate pollster, in 2002, he finished the election cycle by admitting several inaccurate calls, which he largely blamed on sampling errors.

There were signs in the final weekend from national polls that showed Republican support on the move, galvanized by President Bush’s nationalizing the election and the strength of the Iraq war issue. Over the weekend, the New York Times, the country’s leading liberal newspaper, showed a 47% to 40% Republican advantage. An ABC News poll reported on Sunday, Nov. 4 that Bush’s pre-election 67 percent approval rating was the highest of any president in more than 50 years on the eve of a mid-term election.

Colorado Election Polling
Polling is a blend of science, such as statistics and psychology, and an art developed over decades of trial and error. The nation’s highest profile polling mistakes—picking Dewey in 1948 and a close Reagan/Carter race in 1980—spurred improvements. Pollsters today must increasingly deal with “non-voters” and “non-responders.”

In Colorado, turnout in the Allard vs. Strickland race was only 50 percent; out of 2.8 million registered voters, 1.4 million voted. As turnout declines, interviewing the voting portion of the electorate is a prime challenge. The Denver Post/9News/KOA News Radio poll, conducted by Ciruli Associates, addresses the problem by sampling from a list of registered voters only those who have voted in recent statewide elections (plus newly registered voters). Respondents are further screened by determining their voting history and commitment to vote in the upcoming election. This technique, perfected after more than a decade, produces accurate samples of likely voters.

The methodology lead to accurate forecasts of Marilyn Musgrave’s landslide; the close Feeley/Beauprez race and its slight Republican advantage; and the defeat of bilingual-limiting Amendment 31. Our final Friday night before election day poll showed Gov. Owens with 62 percent; he won with 63 percent. The final poll also showed all the major ballot issues, except for campaign finance limitations, struggling or losing substantially.

Notably, Allard outperformed final published weekend polls. Strickland won in Democratic strongholds such as Denver and Boulder, but was swamped by Allard’s exceptional vote advantages in Douglas (24,000), El Paso (53,000), Weld (13,000) and Larimer counties (10,000).

After 11 nights of tracking in our survey, Allard averaged a 3-point lead, including a high of eight points. The race closed among decided voters over Halloween (even Allard called Halloween a “crazy night” when his own tracking poll showed him losing to Strickland). Although Allard won our final Friday night survey by three points, Strickland was up one point in the four-night average. The final poll still showed 13 percent of voters undecided. Hence, the winning Strickland nights in the last week were a blip, and the overall trend showed Allard positioned to win by two to three points. The strong national movement generated by President Bush and the superior Republican get-out-the vote effort pushed Allard even higher.

Lower Rates of Responses
While not having an accurate sample of likely voters is a key source of missed calls in this election, non-response also is a problem.

In every poll some percentage of people are not available or choose not to participate. Rates of non-response for academic, commercial and political polls have increased during the last decade. Call waiting, cell phones and the proliferation of surveys all have contributed.

Media and political polls near election day are especially subject to non-response because they are conducted over only a few nights or, in some cases, hours. One of the best methods to address non-response is to conduct call backs, in which respondents are called three, five or even up to 10 times to reduce non-response.

To compensate for higher non-response rates in rapidly conducted polls, such as tracking polls which allow only 24 hours to complete interviews, our firm uses a sample of previous voters clustered by geography. Hence, if a new person is substituted for a non-responder, the person is a likely voter who lives in the same community. Also, we ensure the sample uses proportions of Democrats, Republicans and unaffiliates, gender, ethnicity and geography in alignment with historic registration and turnout data. Periodically, small adjustments are needed, however none were needed in our final tracking survey.

There is no evidence, however, that some groups such as Republicans are less likely to participate in political polls. In fact, a number of studies have shown that persons less likely to participate in political surveys also are less likely to vote (i.e., the young, the mobile and the less affluent).

Although non-response is a challenge, proper sample selection, interview methods and adjustments to demographic variables can ensure a good track record for election polling. Publishing response rates can be useful, but are further illuminated by sample selection and weighing techniques.

How polls are used by the media also is subject to considerable controversy. Media conduct their own polls to be independent of candidate polls and enhance understanding of election dynamics, voter opinion and campaign strategy. The Denver Post chose to keep polling data in perspective by integrating tracking poll results into larger analyses, rather than feature results like sports scores.

Polls Key to Republican Strategy
For all the criticisms of media-reported polls, use of polls by politicians has never been more extensive or important. In fact, the entire Republican mid-term strategy was built on state level polling. Early in 2001, White House and Republican Senate leaders pored over polls from 10 targeted senate seats. Polling data helped guide the White House to recruit Norm Coleman in Minnesota, Elizabeth Dole in North Carolina and Jim Talent in Missouri.

National polling helped shape the Republican message to emphasize homeland security and the war in Iraq, and craft defensive positions on social security and corporate corruption. Most importantly, state level tracking polls guided the President’s final travel blitz and the movement of campaign money. For example, polls showed that Elizabeth Dole needed money in North Carolina, Georgia needed a presidential visit (it got two) and New Jersey was a lost cause.

The White House and Republican Senate and House campaign committees are not fretting that polls have lost their usefulness.

Conclusion
The 2002 election reminds us that polls are not perfect and are subject to random error and bad technique. The 2002 election was especially challenging given rapid changes in voter opinion and conventional wisdom being proven wrong. Despite a few spectacular misses, polls were vital to politicians planning strategy and helped media and voters follow races with an insider’s view.

[top] [archives] [home] [send this page to a friend]