WASHINGTON - For political pollsters struggling to gauge American voter sentiment during the 2008 presidential primaries, it has been a season of confident prediction and humble contrition.

John Zogby knows both feelings all too well.

Just before California Democrats voted in their presidential primary on Super Tuesday, Zogby's poll for the Reuters and C-SPAN gave Barack Obama a 13-percentage point advantage over Hillary Clinton - suggesting an easy victory for the Illinois senator in the nation's most delegate-rich state. Clinton won by 10 points.

Hillary Clinton speaks at a YMCA in Philadelphia Friday. Polls for the state's Democratic primary show Clinton leading, but with Barack Obama gaining on her. But can the polls be trusted?

Getty Images

Email to a friendPrinter friendly

Font:

• *
• *
• *
• *

"I screwed up California," Zogby says matter of factly.

He's not alone.

The Democratic presidential race has proven to be a minefield for some of America's most prominent pollsters, who've have been confounded by a volatile electorate, a huge turnout by new voters and a series of last-minute campaign surprises. 

From New Hampshire to South Carolina to Ohio, polls have gotten it wrong time and again.

Now, with 10 days before the potentially-decisive Pennsylvania primary, a batch of surveys shows Obama steadily narrowing the gap on Clinton. But can we believe them?

"At this point, voters have had so much thrown at them you would think they are past the point of high volatility," says Charles Franklin, a University of Wisconsin political scientist and co-developer of Pollster.com. "But you never know."

After weeks of considerable hand-wringing, pollsters have come up with some theories about why their snapshots of the Democratic electorate have been so fuzzy.

The first explanation is straightforward. Primary campaigns are historically tougher to poll than general election campaigns because the behaviour of undecided voters is far less predictable.

In a general election, undecided voters tend to break toward Democrats or Republicans based on their partisan 'leanings.'

But in this year's Democratic race, undecided voters have been genuinely torn because Obama and Clinton are generally well-liked by party activists.

The result of that indecision?

In New Hampshire - where every major pollster incorrectly predicted an Obama victory - 18 per cent of voters made up their mind on voting day.

In Ohio, 12 per cent of voters made last-minute decisions. Clinton ended up winning the Buckeye State by 10 points, despite several polls showing a neck-and-neck contest.

"What we've seen all year so far is that - in terms of accuracy and in terms of predicting the outcome - there has been a pretty consistent under-estimation of support for the first-place finisher," says Franklin. "Late deciders are breaking disproportionately for the ultimate winner."

Complicating things further for pollsters have been problems getting an accurate sample of the electorate.

Obama's popularity among African Americans and youth fuelled an enormous surge of new voters, who typically fly under the radar of many pollsters.

"I think this year's greater influx of new voters has challenged the pollsters more, and made the 'likely voter models' more difficult," Franklin says.

The problem has been most evident in states Obama has won by large margins. In South Carolina, Obama's margin of victory was 15 points greater than most polls predicted on the eve of voting.

Days before Wisconsin's primary in February, Obama won by 17 points despite one survey that showed Clinton with a six-point lead.

But inconsistencies have also arisen in states won by Clinton.

In California, Zogby admits he under sampled Latino voters - a solidly Clinton constituency - and over sampled African-Americans.

"There are so many moving variables here," Zogby says.

"When we say that polls have a 95 per cent confidence rate, there is always that five per cent. And California was our 'five-per-cent' blunder."

Extensive early voting in many states also made it difficult for pollsters to accurately take an election-day snapshot since ballots had been cast over an extended period.

And then there are the truly unpredictable factors - like Clinton's teary-eyed episode in a New Hampshire diner and Obama's 'NAFTA-gate' controversy in Ohio. Both events occurred just prior to voting and are thought to have triggered last-minute swings.

Franklin says it's easier to spot polling inaccuracy in 2008 simply because there are so many more public opinion surveys conducting surveys.

Just since April 1, there have been 10 separate polls in Pennsylvania, with results ranging from a Clinton-Obama tie to a 14-point Clinton lead.

What has been consistent to most of them, however, is a trend showing Obama closing the gap.

"Obama's upward trajectory is bigger than Clinton's upward trajectory. I believe that trend," says Franklin. "The real problem is knowing whether that trend is going to flatten or accelerate."

Despite the high-profile miscues, both Zogby and Franklin say most of this year's polling has actually been spot on. Of 21 states where Zogby regularly polled, his firm nailed the result in 16.

"By and large, I think we have done fine," he says. "There's always been someone who has gotten it right. But no one has gotten everything right all of the time ... Each of us has taken a turn falling on our face."

© Canwest News Service 2008