Angry in the Great White North

A new poll is reporting the pre-election race is a dead heat:

A Conservative lead in popular support seems to have evaporated this week, suggests a poll conducted after Paul Martin's national TV address and while the Liberal-NDP budget pact was being worked out.

The survey by GPC Research had the two parties in a virtual dead heat: 27 per cent of respondents said they would vote Liberal and 25 per cent said they would opt for the Conservatives if an election were held. Those numbers would represent a break with every other poll done since a spring election became a distinct possibility in the wake of the unfolding sponsorship inquiry.

With a margin of error of 2.8 percentage points, the GPC poll puts the parties in a statistical tie.

I won't say the standard nonsense about the only poll that matters is election day, or that we shouldn't be looking at polls, etc, etc. Polls do matter and we watch them religiously. Having said that, not all polls are equal:

[Gail Haarsma, vice-president of research for GPC] said the numbers represent her firm's first foray into polling about elections, although GPC did polling for its private clients during the election last year.

"This time we thought with the environment the way it was, we would go out and find out what the landscape was prior to a potential election call," she said.

Uh-oh, newbies.

Recall that during the last US election, Zogby poll after Zogby poll showed John F. Kerry with a commanding lead. What did they know that the other pollsters didn't? As it turned, nothing. Their polling technique was flawed. I would put GPC into that category until I see more polls from established firms.

Remember too that polls done for news organizations are different from polls done for parties (which is why you always hear about "internal polls" as if they're different -- turns out they are). When a sample is selected for a poll, it is done with care. Some many houses in so many neighbourhoods, with so many people in these economic brackets or social classes, etc, etc. When the call goes out, if the person doesn't answer, the pollster is supposed to keep trying. For internal polls, that is what they do. But for media polls, they general rule is to keep calling down the list. As sample homes are skipped, the data set drops in size, decreasing your confidence, and the data is skewed, as critical representative households are skipped.

Who knows? Maybe GPS does extrapolations, too. That can really mess things up.

So let's see what more polls say. If they are all trending the same way, then we've got something to talk about. But to say the lead is evapourating on the basis of one media poll from an inexperienced firm that seems to be going against the trend polls from the major firms have already established is to jump to conclusions.

But the headline that suggests the election might be a real race does sell papers.