Wednesday, September 24, 2008

Which polls are most accurate?

Here's an interesting post from Worthwhile Canadian Initiative, a keen blogger who's analyzed the seemingly inexplicable gaps in party support that we see between pollsters' numbers, particulary between the Liberals and Greens. He illustrates why poll watchers will be wise to trust Nanos's daily releases most.


Stephen Gordon said...

That's not *quite* what I concluded. I think we have a good idea as to *why* Nanos' numbers are different. But I'm not at all sure that we can conclude that his numbers are more reliable.

Prairie Topiary said...

Thanks for the clarification -- you're right that my blog post title is probably a little hasty when it suggests that your analysis makes conclusions about accuracy or reliability.

Based on past observations, we do know that some poll methodologies are better at predicting election results than others, and Nanos's methodology appears to have been better than others in this regard in the past two federal elections. I believe this has to do with such things as "top of mind" voting choice (e.g., a voter may say they'll vote for the Greens when prompted with the party name, but yet vote for a larger party that's "top of mind" when in the polling booth) and actual election day party organization (i.e., where a given party has the organization to translate voting intention into actual votes). I believe that Nanos's methodology -- specifically the way it asks potential voters to indicate voting intention -- is better able to account for these factors.

Your analysis is consistent with this observation and I believe it sheds light on the discrepancies we see between different pollsters' numbers.

Stephen Gordon said...

You may well be right.

But then again, maybe not!