Volatile Polls Are Giving Blurry Electoral Pictures

If it’s true, as pollsters say, that polls are simply snapshots of opinions, then the snapshots coming out of this year’s neck-and-neck presidential election resemble a tumultuous family reunion.

In Florida recently, two different polls taken about the same time had two different leaders. Two polls in Illinois taken about the same time showed Democratic nominee Al Gore leading over Republican nominee George W. Bush, one by 7 points and the other by 2 points. And a national daily tracking poll by Gallup--the people who made polling a household word--has had more swings than the Nasdaq.

So what’s happening?

According to pollsters and other experts, it’s a mix of genuinely uncertain voters and the inexact science of polling itself. In this statisticians’ world of medians and margins, sampling errors and confidence levels, polling numbers come with so many footnotes that they seldom appear to mean what everyone wants to know--which candidate is winning.

The problem seems especially pronounced this election, with polls at both the national and state levels adding more confusion than insight for those who would plumb the mood of voters.

There are two keys to the fluctuations.

The first stems from who is getting polled. Most polls try to track the opinions of likely voters and adjust the results--using census data--to try to mirror the breadth of the population being surveyed. Each poll follows its own criteria for making those decisions, which can lead to differences among their results.

For example, each poll uses its own set of questions to winnow out likely voters from random telephone calls to homes and businesses. Yet even getting people on the phone has become more difficult in recent years as more people hang up on telemarketers and pollsters alike, according to pollster Del Ali of Research 2000 in Rockville, Md.

“You get a lot more rejection,” Ali said. “Each year is harder simply because people are more stressed and you’ve got to really convince people that you’re not trying to sell them something.”

Sampling Error, Confidence Level Affect Fluctuations

The second key to the fluctuations is the margin of sampling error and confidence level, the methods pollsters use to vouch for themselves. The confidence level, usually pegged at 95%, reflects the level of faith the pollsters have that their results would be repeated within the margin of error had the entire targeted population been polled.

The margin of error refers to how widely those results might actually vary. A margin of error of 4--usually written as plus or minus 4--means that if the poll was repeated across the general population, the results would likely be within four points above or below the current findings.

In other words, 45% for Bush means the actual support ranges from 41% to 49%. Likewise, a 43% for Gore means the actual support falls somewhere between 39% and 47%.

That range is wide enough to drive a third-party candidate through, and would seemingly make the polls unreliable. Yet pollsters say the margins usually aren’t that pronounced in reality, and are good indicators of how public opinion stands at that moment.

As a result, polls remain key barometers for those who are drawn to politics as a spectator sport, tracking the daily fluctuations in support and analyzing the latest spats over details.

“That’s the fun of it--the daily bounce,” said Gerry Chervinsky, a pollster for KRC/Communications Research in Newton, Mass. “What happened today that might move the support along toward the finish line? But people want to know what each candidate stands for, and to get a sense of them personally.”

Compounding the confusion, however, is the perception that current national tracking polls are counting more support for Bush than he is receiving in the state-by-state polls, which are the numbers that really matter in the fight for electoral votes.

For example, recent national polls give Bush a lead mostly within 2% to 7%. But in Pennsylvania, Oregon and Michigan, polls show Gore with a slight lead. The difference lies in the scope of the polls, with state polls more affected by local issues and local biases--such as the fate of salmon in Oregon or Texans’ support for their governor, Bush.

All in all, Bush leads the polls--although many of the leads are within the margin of error--in 28 states and Gore leads in 17 states and the District of Columbia. In five states, the polls are too close to pick a leader. But many of the Gore states--including California--are more densely populated and carry large blocs of electoral votes.

Thus while Bush has a seemingly large lead state-by-state, the race is much tighter for the 270 electoral college votes needed to win the White House.

And many of those leads could evaporate overnight, reflecting the shifts in support that have helped fuel swings in the national tracking polls.

Tracking Polls Draw Attention

It’s the tracking polls that have drawn the most attention this election year. Likely voters are surveyed daily and the results of each day’s poll are averaged with the results from the prior two days. Pollsters believe that allows them to measure mood shifts as they are occurring. In a seesawing election, then, the polls themselves would seesaw.

Most other polls, including The Los Angeles Times poll, are taken less often and tend not to reflect such wide swings.

One pollster likened the phenomenon to watching the daily fluctuation in Wall Street versus watching the weekly averages.

Gallup began compiling daily tracking polls just after Labor Day, when voters traditionally start paying closer attention to the campaigns.

And so far, the Gallup polls have come up with pretty wild swings. In one recent shift, the poll--conducted for CNN and USA Today--found Bush leading 51% to 40%. Three days later, the poll gave Gore a 1-point lead.

On Thursday, Bush had a 7-point lead, followed Friday by a 13-point advantage. But on Saturday, his lead was back to 7 points.

Some critics argue that those swings indicate flaws in the poll. They have speculated that such factors as the days and hours the questions were asked and the ratio of Democrats to Republicans among those polled might have skewed the results.

But Gallup officials stand by their work, arguing that the swings reflect genuine uncertainty among the electorate. Frank Newport, editor in chief of the Gallup Poll, said party identifications also shift because questioners ask voters how they identify themselves, which opens it up for people not enrolled in a party to still claim kinship with one.

‘People Changing Their Minds’

Although Gallup’s swings have been wider than some other tracking polls, nearly all have shown some volatility, pollsters said.

“People are changing their minds,” Newport said. “We’re measuring something people are supposed to be changing their minds on, trying to pinpoint a flexible attitude out there and from day to day it can move around.”

Newport speculated that if Gallup had been doing daily tracking polls during the tight 1960 and 1980 presidential campaigns, they also would have shown wide daily swings.

He dismissed critics of the current polls as missing the point.

“It’s an interesting look at what’s really going on” in the minds of voters, Gallup said. “Rather than saying, ‘That can’t be right,’ we should be saying, ‘That’s an interesting insight.’ ”

Arnold Barnett, a professor and statistician at the Massachusetts Institute of Technology’s Sloan School of Management, said that although wide fluctuations in one poll could indicate a problem with how the poll was conducted, nearly all the polls this year are showing some fluctuations.

Taken together, he said, that indicates people really are shifting their allegiances.

“Any one poll is not decisive,” Barnett said, adding that each poll is evidence of voter attitude. “If you get lots of bits of evidence, the cumulative effect can be very potent.”