I noticed yesterday quite a bit of blogging buzz about the new online poll out from Angus Reid's new polling outfit. Two other polls were also out this week, one from Ipsos Canada, Reid's former firm, and Decima.
Reid's new online poll would produce a Conservative majority of 161 seats (with no seat gains in Quebec) according to TC's calculations(Liberals would get 73, the NDP 25 and the Bloc 48), providing further encouragement, if any were needed, for Stephen Harper to trigger a spring election.
But there is a catch. The methodology of the Reid poll is a tad obscure. It is based on an "online panel ... recruited via an industry-leading process that incorporates a randomized, widespread invitation approach and a triple opt-in screening procedure". I think what these words mean is that they have made their best efforts to select a random sample over the internet. Nevertheless online polling is in its infancy, and while I think it will be a standard some day, it certainly isn't now.
There was an online poll conducted by Ipsos just prior to the last election. TC did a comprehensive analysis of the accuracy of all the polls taken at that time, and found that, although the Ipsos online poll had the largest sample size, it was the least accurate in predicting the national vote shares in the election (in fact, most of its errors were greater than its statistical margin of error). While overall it was fairly close, having an aggregated error of 9.6 it needs to be compared to SES, the closest to the mark, which had an error of 2.1 with a far smaller sample. The online poll did do better on its regional sub-samples, where it had larger numbers than the conventional telephone pollsters and thus a lower statistical margin of error.
My conclusion is that all online polls should be treated with extreme caution, and we really need to see confirmation in conventional polls before they should be taken as gospel.
The two other polls did reveal Conservative strength and Liberal weakness, but they both had the Conservatives at 36% compared to the online Reid poll's 40% and the Liberals averaging 29.5% compared to Reid's 26%. Interestingly, the largest error in the Ipsos online poll prior to the 2006 election came in its underestimate of the Liberal vote, reinforcing my inclination to view it skeptically.
Finally, looking at the online Reid poll and the 161 seats reinforces TC's view that it is almost (but not quite) impossible for the Conservatives (or the Liberals for that matter) to win a majority. And some of the numbers in the poll are just not believable. TC thinks that the results of the last election are usually a better predictor of what to expect next time than many polling numbers that appear in the interim and that certainly appears to be the case here in some regions. Just a slight deviation from these results would quickly reduce the Conservatives to less than a majority. However, it does look like an early election would likely produce another Harper government, simply because it doesn't appear that the Liberals have the money or the time to be properly prepared.