I have been reading with great interest why online polling is misunderstood and unfairly maligned in response to the recent article, "A voters guide to political polling."
Both articles raise good points about the polling industry in Canada, and why it seems the majority of polls are less accurate today then they once were. I need to agree with Angus Reid that there are benefits to online polling, including the complexity of questions they can ask and the ability to use visual stimulus to solicit responses. Online polls can be fun and if we want to determine the public's opinion on their favourite colour, or why they like Britney Spears better than Madonna, they can be accurate enough to be newsworthy, I suppose. I am not a statistician so I won't bore you with the usual argument that online polling isn't truly random because of the "opt in" nature. I think we have to look at the differences between American and Canadian politics and culture to understand why online polls don't accurately reflect public opinion accurately enough when it comes to voting intentions.
Comparing Apples and Chocolate Pudding
In the last U.S. election, as was pointed out by Mr. Reid, four of the seven most accurate polls were sourced online. Comparing U.S. & Canadian politics and political culture is like comparing apples and chocolate pudding, they are both edible. Consider some numbers from the last two federal elections in the U.S. and Canada. In the 12 months leading up to the 2012 Presidential Election, Barack Obama and the Democrats raised approximately $800,000,000 from 4.2-million Americans, relying heavily on online and email solicitations to raise small and repeat donations.
During the 12 months leading up to the last Canadian Federal election in 2011, Stephen Harper and the Conservative Party raised just under $25,000,000 from under 120,000 Canadians. Again, the Conservatives rely on small individual donations and repeat donations from engaged supporters. Proportionately, 3.2 times more money is raised from 3.5 times more people in the U.S. The trend of small, online and email solicitations supporting campaigns, started in 2004 with the Howard Dean campaign and quickly grew to be a big part of all subsequent campaigns. Many campaigns in Canada have attempted to replicate the Dean/Obama email solicitation and online engagement/mobilization effort, yet the best effort ever in Canadian political history (in that regard) falls short by a factor of 3-3.5.
Now, it's quite possible that American political strategists are 3.2 to 3.5 times as smart as Canadian political strategists. To provide some context, if we look at the top crowdfunding campaigns in Canada versus the U.S. in 2014, we see that Canada outperforms compared to our population. Canadians are not averse to the internet or spending money online so how do we explain the discrepancy in political donations which is now largely online or via email. Perhaps we haven't hit a critical mass, perhaps we never will. In my opinion, Canadians don't like sharing their political opinions online with the same proportion as their U.S. counterparts, I would bet somewhere between 3.2 to 3.5 times less likely.
Work Harder, not Smarter
The notion that live phone polling costs more today than it did 20 years ago is an idea I can't possibly ridicule enough. It's true that response rate have dropped from somewhere around 20 per cent to as low as 5 per cent but the technology available today with voice over internet protocol (VOIP) and predictive dialling software more than makes up the difference in costs of those reduced response rates. If fewer people are answering polls, the answer shouldn't be to wring our hands and reminisce about the good old days of public opinion research, the answer should be make more calls. Nik Nanos correctly states that the goal should be to weigh as little as possible, or not at all, this requires bigger samples, more work and perhaps increased costs. The stampede to online polling probably has more to do with profit margins than a sincere desire to get more accurate results.
Undecided voters don't behave like they used to
There is one last factor that I believe has affected the ability of many pollsters to accurately gauge voter intentions for the last few years, the undecided voters aren't doing what they are supposed to. Traditionally, the undecided vote used to break in nearly the exact same way as the decided vote, and that is simply not the case anymore. If pollsters aren't taking this in to account somehow, predictions versus outcomes will continue to be less accurate. The last few years, we have seen a significant shift in the behaviour of undecided voters towards supporting incumbents. The closer than expected results in 2014 in New Brunswick, the results in Ontario, B.C. and Alberta saw varying degrees of incumbency swing in the closing days before election day. This may bode well for Stephen Harper this fall if that trend continues.
So how do we account for that undecided break you may ask? We pour on a few grains of salt, of course.
Which polls should you trust
With all the above said, I won't tell you which polls to believe or not believe. There are many pollsters who have a great track record of explaining why they were wrong, and few who have a great track record of not needing to explain why they were right. My online poll of which you choose to trust will be released tomorrow.