Canada: Polling houses and provinces

So this week, Ipsos Reid, Environics and Ekos have all released polls. They don't have the same sample sizes and they don't cover exactly the same dates. The three houses have slightly different methodologies and tend to produce differing results. Yet despite all of this, they have pretty close numbers at the moment nationwide. For example, Ipsos Reid puts the Conservatives at 34% nationwide, Environics has them at 35% and Ekos, which typically gives the lowest Tory numbers, has them at 32.4%. All well within the margin of error. Ipsos Reid and Environics have the exact same numbers for the Liberals, Ekos has them lower by little more than 2%. All in all, pretty good numbers. So good, in fact, that we can conclude that (a) the three houses must be doing reasonably good jobs if their numbers are so similar, and (b) these numbers must be pretty accurate, if three different polling houses have produced them, more or less.

The national sample size is going, of course, to be some ten times larger than the sample size of each province. Of course as a result the margin of error is going to be greater at the provincial level. But what strikes you looking at the three houses is just how deviant their provincial numbers are from each other, in spite of their similarities at the national level. It's an interesting paradox that unreliable numbers at the provincial level can combine to produce more reliable numbers at the national level, but it's a fact of statistics that deviation gets 'rounded out' as numbers increase. Still, it's tough to make any real statements at all about what's going on provincially based on these three polls.

For example, let's look at the four-province megaregion pollsters call 'the Atlantic'. The four provinces east of Québec all obviously have different voting trends and traditions, but due to their relatively smaller population bases are still lumped together in polls. Look at each of the three pollsters, though, and you get a remarkably different picture of these provinces.

Ipsos Reid has the Liberals at 48%, domination on a level we rarely see anymore. Ipsos Reid will tell you that just slightly less than half of all Atlantic Canadians are planning to vote Liberal. Ekos, on the other hand, has the Liberals slightly behind the Conservatives - 32.5% to 33.2%, with the NDP doing historically well at 22.4% in a legitimately tripartisan race: numbers that might be plausible in Nova Scotia, but as for the other three provinces? Environics, on the other hand, not only agrees with Ekos by putting the Conservatives first but puts them far ahead, at 35% to the Liberals' 31%. Has Danny Williams's ABC campaign lost its fizzle? Or are these numbers just all over the place?

Going from one coast to the other, we have BC. Now, yes, BC is notoriously volatile. And yes, BC is legitimately three-way (or even four-way) to an extent unheard-of in any other anglophone province. But... Ekos puts the NDP ahead at 30.8% and has the Conservatives and the Liberals neck-and-neck at 27.0% and 26.5% respectively (and Green at 15.0%). Those numbers might be plausible if surprising, but contrast them with the other two: Ipsos has the NDP at a sad third with 21% behind a surging CPC at 41% and the Liberals at 25% (Green at 13%). Environics also has the NDP at third at 23%, well behind the Liberals at 30% and the Conservatives at 35% (Green at 10%). Admire that range: are the NDP polling 31% or 21%? Are the Tories polling a sad 27% or a mighty 41%?

The national numbers produced by these polling houses provide great talking points. But come election day national numbers mean nothing (remember 1993, where 16% of the national vote got the PCs two seats but 14% of the vote got the BQ 54 seats). And just how much can we trust these polling companies to give us useful provincial (in fact regional) data when it's so very hit-and-miss?