[Editor's note: Polls fresh on The Hook: As various poll results are revealed, Tyee contributing editor Tom Barrett will post them to our political news blog. Check The Hook regularly to keep up to speed in the last days of the 2009 provincial election.]

What's a voter to think when the polls are bouncing up and down like Paula Abdul on Disco Night?

B.C. election polls have put the two main parties as far apart as 17 percentage points and as close as three points.

It could be that a lot of people have been changing their minds a lot. Or it could be that there's something wrong with some of these polls.

With just days left in the campaign, and with a flurry of polls due today and tomorrow, it's worth asking which it is.

Kennedy Stewart, an assistant professor in the Simon Fraser University Graduate Public Policy Program, thinks it's more likely something in the polls.

Some pollsters, he said in an e-mail interview, "are better at selecting random samples than others.

"This is very problematic."

"The whole reason you hire a polling company and do polls is to use your sample to get an accurate snapshot of the population in which you are interested," Stewart said. "I think of it like taking a picture of someone with a camera, but when you get the developed photos back their head or feet are cut out of the portrait."

We won't know until election day who cut whose toes off. But certainly, these are challenging times to be a pollster. People are harder and harder to reach. Some people are less and less inclined to participate in polls. And if the people you aren't reaching are somehow different in their politics than the people you are reaching, you could be in trouble.

The grouchy old men factor

Let's say, for example, that the Grouchy Old Men demographic tends to vote Liberal. But because they're grouchy and they're old and they don't like to talk to strangers, they don't show up in your poll.

There's a good chance you're underestimating the Liberal vote.

Even if you make sure you've got lots of old men in your sample, you could still be off. Maybe you're interviewing a lot of members of the Lonely Old Men Who Will Talk to Anyone Who Phones demographic -- who, let's say for the sake of argument, tend to vote for the New Democrats.

Mario Canseco, of Angus Reid Strategies, said a similar thing has happened in British politics, where telephone pollsters have at times oversampled retired people and homemakers -- people who tend to be at home more, and therefore are easier to reach by phone.

The problem was, these people were also more likely to vote Conservative.

How online pollsters work

YouGov, an online pollster in the U.K., had more accurate results because its samples more accurately reflected the voting population, Canseco said.

Online polling, the methodology used by Reid in its B.C. election polls, is still a new technique. But, said Canseco, "We feel very comfortable with the stuff that we're doing."

Reid points to its track record to back up its claim that online polls are more accurate. In the 2008 federal election, for example, Angus Reid Strategies boasts that it came closer to the final result than seven other competitors.

Canseco pointed out that technological changes have created problems for telephone pollsters. Telephone polls may miss the growing numbers of people who don't have landlines, he said.

This became an issue in the run-up to last year's U.S. elections, Canseco said.

Even the pollsters who are sticking to telephone polls in this campaign use samples drawn from online panels for other polls.

The case for the phone

Evi Mustel heads up the Mustel Group, which called the last B.C. election within a few percentage points. She said about half her company's work is done online these days. In an online poll, a pollster recruits Internet users who are interested in responding to surveys. Samples for individual polls are drawn at random from the resulting panel.

The industry is still trying to determine which is the best method, she said.

"Is it telephone? Is it online? Is it a combination of the two?"

At this point, she said, "we're more comfortable using the telephone methodology for political polling....

"There's gaps in both, without a doubt. Our feeling is there's larger gaps right now in the online polls."

The people who participate in online polls have agreed to opt in, Mustel noted. (This raises the question of whether your panel reflects, in this case, B.C. voters as a whole.)

And not every home has Internet access, she said. Older people tend to be less likely to be online, but they're also the most likely to vote, she said.

As for difficulties reaching people who don't have landlines, Mustel said that research indicates that this group tends to be younger than average. But research also indicates that young people without landlines tend to have the same views as young people with landlines, she said.

So as long as your sample has enough young people, the problems are minimized, Mustel said.

'It's how you use it': Braid of Ipsos Reid

Like Mustel, Ipsos Reid, the other big pollster in this campaign, tends to do about half its polling online, said Ipsos vice-president Kyle Braid.

"It's not a case here that an online methodology or a telephone methodology is necessarily better," he said. "They both have their advantages and they both have their drawbacks."

The most obvious problem with telephone surveys is "people aren't always at home," Braid said. "The refusal rates on telephone surveys are quite high and you have to deal with the issue of people who don't have landlines.

"We've worked with those kinds of surveys for a long enough period that we're able to make adjustments and to be able to adapt to those things. If you look at the most recent federal election in Canada, the telephone pollsters did quite well, and in the most recent U.S. election Ipsos was tops in all of North America."

Online polls, Braid said, "have their own issues."

They are, he said, "generally conducted among a limited group of people. People have to opt in to take part. There are issues with going back to the same people over and over again with multiple surveys and whether that influences their opinions."

As well, as Mustel noted, "there's issues with who's represented online," Braid said.

Online polls tend to under-represent people with less education and lower incomes "and perhaps to a degree, still, older Canadians...."

"So neither methodology is perfect and like any other tool it's how you use it."

Various polls will converge: SFU's Stewart

He said Ipsos tracks its election telephone polls with online polling and the result are "really close."

"At this point, we're still more comfortable using the telephone for the vote preferences numbers," he said. "But that could change anytime over the next few years.... At some point, the phone may not be an option. But I don't think we've reached that point yet."

Kennedy Stewart, the SFU prof, said he expects the results of the last flurry of polls to converge.

"But if one or two companies are off by miles, their reputation will considerably suffer," he said.

"In some ways, watching this race between polling companies has been more exciting than the race between the parties."

Related Tyee stories:

• Voters trust Campbell on economy, mistrust him otherwise: poll
• Liberals slip to 51 seats in SFU prof's projection
• BC Liberals lead NDP by 11 points: Environics