Washington Post - September 20, 2000
The Poll Less Traveled
By Richard Morin and Claudia Deane Washington Post Staff Writers Wednesday, September 20, 2000
Two roads diverged in a wood, and I . . . took the one less traveled by - Robert Frost
Diverging roads make for good poetry. But diverging tracks make for lousy daily tracking polls. And that's precisely what the poll-watching public faced this week as two high-profile presidential tracking polls consistently and mysteriously reported very different results.
In the past seven days, the Voter.com/Battleground poll found Republican George W. Bush leading by 1 to 6 percentage points among likely voters.
At the same time, the Gallup Organization's daily track for USA Today and CNN just as consistently put Vice President Gore in the lead by 4 to 8 points.
Last Thursday, the two tracking polls were, well, polls apart: The Voter.com website declared Bush up by six points in the <http://www.voter.com/home/news/article/0,1175,2-12665-,00.html?GXHC_gx_session_id_FutureTenseContentServer=3fd1632e123a5685>latest Battleground poll while <http://www.usatoday.com/news/poll001.htm?GXHC_gx_session_id_FutureTenseContentServer=3fd1632e123a5685>Gallup reported Gore with an even heftier 8-point advantage.
As the days passed, the Battleground track seemed to be the odd poll out. Among the dozen major surveys reported in the past week, only the Battleground found a Bush lead of any size.
Uh, what's happening here? The all-star team of Republican Ed Goeas and Democrat Celinda Lake conduct the Battleground poll for Voter.com, the politics website. In an interview, Goeas acknowledged some start-up problems. Voter.com shouldn't have reported the fat Bush lead last Thursday, which was based on a likely voter model that "works better closer to the election." This week and for the time being, they've publicly retired that likely voter model - and, perhaps coincidentally, the horse race has tightened.
Other factors, including weighting techniques and the order questions are asked, also help explain the differences between Battleground and Gallup tracking polls.
Goeas and Lake adjust, or "weight," their survey results so that 35 percent of their tweaked sample are Democrats and 35 percent are Republicans.
Setting the parties at parity could account for the slightly more favorable outcome for Bush - in the latest Gallup poll, Democrats outnumber Republicans by 4 percentage points.
The true percentage of Democrats and Republicans is, of course, a mystery and varies as people fall in and out of love with one party or the other. In the network exit polls, self-described Democrats outnumbered Republicans by 3 percentage points in 1992 and 4 percentage points in 1996, respectively. (Goeas says they derived their estimate from their own past polling that showed Republicans with a 3-point advantage in 1992 and Democrats up 3 points four years ago.)
The Battlegrounders also don't interview on Friday or Saturday, while Gallup does. Friday and Saturday are notoriously bad nights for finding people at home and those who are home tend to be Democrats, skewing the sample, Goeas said.
In the standard Battleground questionnaire, the four-way presidential preference question is asked at the very end of their survey, after questions about the direction of the country, important issues, and President Clinton's job approval. Gallup and most other polling organizations, including the Washington Post, ask the horse race question at the beginning of their polls. That's done to avoid question "order effects" in which the answers to one question color responses to subsequent questions. Additionally, the Battleground survey is reporting about twice as many undecided voters than other surveys are showing.
Goeas remains confident that their poll is not merely sound, but superior to other surveys. He says watch this space: Four years ago, Gallup's tracking showed a 20-point Clinton advantage in September while the Battleground poll showed a 12-point lead. "Everybody came around to us," Goeas noted, claiming his poll was only off by a single point in each of the last two presidential elections.
Meanwhile at Newsweek: Gore Goes Gangbusters
If Gallup and the Battleground polls are on two different tracks, then Newsweek sometimes seems to be on a different planet.
While most polls are showing a tight race or modest Gore lead, the <http://www.msnbc.com/news/461397.asp?GXHC_gx_session_id_FutureTenseContentServer=3fd1632e123a5685>latest Newsweek poll suggested the biggest wipeout this side of Surf City: Gore was up by 14 points among likely voters, conducted last Thursday and Friday by Princeton Survey Research Associates.
In its two post-Labor Day polls, Newsweek estimated that Gore led Bush by an average of 11 points - 51 percent to 40 percent - while the average lead reported in five other major media polls put the veep's lead at a modest 2 points.
"I think if Newsweek took a poll of the Bush family, Gore would win," Fox News' Brit Hume cracked recently.
Ouch! And unfair, says Newsweek's pollster. "There's a misperception about how big the gap is between our results and others'," said PSRA Vice President Larry Hugick, who has run the Newsweek poll since 1989. Hugick notes that if you average the Gallup tracking polls conducted between Sept. 7 and Sept. 15, Gore gets 48 percent of the hypothetical vote. The Democrat's share of the vote in Newsweek's two post-Labor Day polls averaged 51 percent - a 3-point difference. Likewise, Bush averaged 43 percent in Gallup and 40 percent in Newsweek polls, again a modest difference.
Granted, the candidates' individual support is not wildly different. But it's the size of the lead that makes headlines. Since Labor Day, Gore's advantage averaged 5 points in Gallup polling - but 11 points in the Newsweek polls.
Being out there alone and far from the polling pack can keep any pollster awake at night.
"It's a concern," says Hugick. But "I've been investigating these polls exhaustively since the spring. . . . When we see a new poll, and it looks different than some others I would expect us to be similar to, I'm going to make sure there's no [problem] in our data. And then just simply, if it looks right, you gotta go with it."