Notes on the Atrocities Like a 100-watt radio station, broadcasting to the dozens...
Friday, January 30, 2004
POLLSTERS CHANGE METHODS TO IMPROVE ACCURACY
by Herm Tupper
Atlanta. (API) -- A month ago, Paul Antolini told Gallup researcher Debbie Fields that he expected John Kerry to win Iowa, despite impressions then that Howard Dean and Dick Gephardt were fighting for the win. Those candidates, he said, would be soundly beaten. Ten days ago, when that very thing happened, Fields wasn't surprised, "Atlantans speak for America."
In recent years, polling companies have become alarmed by how far their findings often differ from election results. They blame changes in telephone use patterns and a cagey electorate who increasingly mislead researchers. With more and more Americans switching to unlisted cell phones, polling companies have found themselves speaking with an unrepresentative sample of the elderly and unstable. "Thanks to phone company telemarketers, no one answers the phone anymore," explained Fields. "Just lonely old people and nuts. It's really amazing that the polls are as accurate as they are."
To handle these evolving changes, research firms have been privately experimenting with alternative methodologies. Gallup, for its part, has discovered that Atlantans like Mr. Antolini are the norm, not the exception. Polls in that city have accurately predicted all six of the last elections in the country. Although it will do limited polling in the rest of the country, for future national elections, Gallup will only report findings from Georgia's largest city.
Gallup is not alone. Zogby International is pilot-testing a new system that will increase the margin of error by 50%. "Statistical models no longer accurately reflect the true margin," said Virendra Rao, head of research for the company. "Gallup's method is risky. It will work as long as Atlanta mirrors the country. Eventually one will change, though, and they'll learn about that only after the fact." The new Zogby system, Rao explained, reflects the actual volatility in the electorate.
One downside, however, is that in close elections, the results will be nearly useless. Current tracking polls in South Carolina, for example, show Senator John Edwards leading General Wesley Clark by 5 percentage points--well within Zogby's 7-point margin of error. So Zogby really doesn't know who will win? "No one does," said Rao. "And our poll reflects that."
But the most experimental is a new methodology the American Research Group is testing. Called PsychTrack, this complex method employs a system based on the findings of behavioral scientists, sociologists, and cultural anthropologists. Explains researcher Cheryl Mackeson, who is heading the pilot system, "studies have shown that the way people walk, use gestures, and dress--and even the expressions on their faces--are interpretable."
According to ARG, Dean voters tend to lumber and grimace and wear flannel, while Kerry voters stare, walk slowly, and dress more conservatively--Brooks Brothers and Claiborne. "Clarkies sashay," said Mackeson. "Which you wouldn't predict." ARG said its exit studies were more accurate than polling in the Northeast, Midwest, and West Coast, although the South had baffled researchers. "It's like the voice recognition software that can't understand Southerners. They mystify us."
But will any of this result in more accurate results? Pollsters are cautiously optimistic. "As long as I have Paul Antolini's phone number," said Gallup's Debbie Fields, "I'll have a pretty good idea what to expect."