27 January 2008

They're down. No... No.... They're up. No, wait. They're down again...

There's a reason why some people should not write about political polls results and others should just be careful.

Let's do Peter a favour and throw him some traffic, only to illustrate how not to interpret poll results.

For whatever reason, there's no link in that post to the poll results, but for those who want them, you can find them at the National Post. It's under the second most "not news" headline right after "Danny in a snit, again".

Anyway, to cut a long story short, a poll by Ipsos Reid conducted between Tuesday and Thursday of last week has the Connies at 37% nationally with the Grits at 29%. Regionally, the survey breaks down different ways with one or another of the two major parties on top - NDPers spare me the e-mails - all of which Peter takes to suggest that there won't be an election any time soon. Yes, he tosses in some other stuff, but the poll results are the hinge on which his post swings.

Couple of problems.

First, the national margin of error, at 3.1% is typical of national surveys but it means that the range of possibilities for the result could have the Connies in majority territory and therefore feeling kinda cocky. Then again, in the Connie worst case in this poll, the parties are basically in a dead heat. So, we could be in an election or we couldn't be.

Second, that sort of stuff only gets worse at the regional level where the margin of error heads for 6%. That's pretty much in the bullshit part of the charts where basically more useful numbers could be produced by tossing darts randomly on the streets of any given small town in the said region.

If that wasn't bad enough, consider that just two weeks before, Ipsos Reid cranked out a poll for CanWest - i.e. the same client - and had the numbers nationally with Grits 35% and Connies at 33%.

Six freakin' point variation in two weeks? Wow.

if you want to wade through a post by a Saskatchewan blogger and the 70-odd comments, you'll find some things that might shed some other light on these poll results. No, Ipsos Reid is not biased. Get past all that crap, including the rather silly rejoinders from Ipsos vice president John Wright. Look at some of the comments, including one by your humble e-scribbler.

Sample size for the national stuff is running around 800 to 900 Canadians. That's the same sample size Ipsos used in the last Ontario provincial election. Basically, they can produce results that look good on paper nationally and might even be generally in the right neighbourhood.

But for all practical purposes, the poll results are useless.

Well, except maybe to keep the Ipsos Reid name in the headlines.

And to generate posts that draw erroneous conclusions based on them.

-srbp-