Wednesday, September 17, 2008

Globe and Mail Poll of Polls

I have been watching the Globe and Mail poll of polls on their website since the start of the election and been wondering a bit about their methodology.

First of all, they include a poll from Innovative Research that no one I know of can seem to find a copy of. The poll is the worst for the Conservatives and a good for the Liberals. Without knowing dates for it or the size, it is hard to say how relevant it is.

Second, their math seems a bit odd. On a simple averaging of the polls the Liberal support is a bit high and the Conservatives are a bit low. They then say sample size is taken in account. When I do this with the same polls they have listed, the Liberal support is too high on their site.

Third, when the polls come out each day, they seem to wait to post them and keep a Liberal favorable one close to the top. I would think you should list them in order of how big they are on each given day. It would also be nice if they showed the size of each poll. As an example, the smallest poll released yesterday was the Nanos one, it was first on the list. Ekos was bigger than Nanos and Harris/Decima combined and it was listed third.

Fourth, their regional breakout numbers are weird. Why add the north to BC? Why lump all three prairie provinces together? Those two decisions make the numbers for those areas much less useful. It would also be nice to know which polls they are using for the regions. Nanos does not break out BC from the praries - an odd thing in itself - so Nanos numbers could not be used for the regional numbers. Only Ekos has sample sizes large enough for the regions that make reporting on them better than throwing a dart. Because of the huge difference between Ekos and other numbers in wieghting the regional numbers on the poll on polls does not seem to jive with the national numbers.

It would be nice if they applied a little more statistical analysis to their poll of polls.

No comments: