Saturday, April 23, 2011

More on online vs phone polls

There was an interesting analysis of online vs phone polling on the Carleton School of Journalism election commentary web site by Christopher Waddell who suggests online polls might overstate support for the NDP:
In the 2006 election, Decima Research conducted a series of experiments comparing polling results from an online panel it had assembled with those obtained from traditional telephone polling. The goal was to see how accurately online polls could match telephone results and to try to figure out how online polls should be weighted compared to the traditional demographic weighting done for phone polls to ensure the pool of respondents matched the demographics of the country.
On election day, Decima asked the members of its online panel to report how they voted in an attempt to create a national exit poll. About 10,000 people responded and by 6:30 pm in Ottawa – three hours before the polls had closed everywhere west of New Brunswick – it was obvious Stephen Harper and the Conservatives were going to beat Paul Martin and the Liberals. There there was a clear gap in the share of the vote each attracted and while Liberal vote share was below 2004 results, the Conservatives were significantly higher.

By the end of the night the Liberal and Conservative vote shares in the online poll were pretty close to their actual results. The one discrepancy was the NDP which did several percentage points better in the online results than it did in actual voting results across the country. This difference was most pronounced in Atlantic Canada where the NDP was 7-8 percentage points online better than it was in the ballot box.
Intrigued, TC compared online and phone polls conducted during this election cycle and found that the online polls gave the NDP two extra points:

CPC Liberal NDP Bloc Green Other
Online 37.7 26.2 20.1 8.9 6.3 0.7
Phone 38.9 27.9 18.1 8.4 5.5 1.2

This is not to suggest that there is no NDP surge. There is indeed movement to the NDP in all polls. It is evident in the numbers in the previous post, which compared four phone polls with the 2008 results. As my earlier post on the subject discussed, there are numerous problems with online polls.  It seems significant, given the experiment described above, that HarrisDecima currently polls by telephone.