What's in a Poll?

One of the big surprises in this month's B.C. provincial election was just how far off the opinion polls were from the actual polls on Election Day. Combined with the surprising results of last year's provincial election in Alberta, some have gone so far as wondering if polls are even capable of accurately measuring where the electorate is - in the west anyway - in deciding who they want to run their provincial governments. But the pollsters themselves are saying, "don't worry, we've still got game."

Difficult to believe? Maybe, but the pollsters themselves have an explanation. I subscribe to an e-mail newsletter from Abacus Data, a data collection firm that crunches numbers across various sectors, including the political. In a blog post, Abacus CEO David Coletto says that with "constant refinement and improvement we are up to the challenge of accurately predicting the next election."

So what happened? Coletto notes that his firm didn't do any B.C. polling following the April 29 leaders debate, but being confident that other pollsters were seeing the electorate as presented through the latest results, Coletto, like everyone, thought it was going to be an NDP landslide. But lessons have been learned since that fake-out-turned-Liberal victory. First of all, Coletto concluded that in the future, all responses should be counted, including undecideds. Since the number of undecideds leading into the B.C. election was as high as 20 per cent, that's a pretty significant figure to put in the "recycling bin," as Coletto calls it.

Corletto also suggests that he and other people in his business should more closely parse those undecideds: What are their values? What are they looking for in a government? What do they think of the leaders and the parties? He also says that pollsters should do a better job of distinguishing between polling groups, such as who's more likely to vote as based on demographics, and also consider peoples opinions on the themes and issues of an election and how they might line-up with party affiliation. Corletto notes that although his company's own polls were showing that tremendous NDP lead, a closer look at the numbers showed possible warning signs to suggest that that lead wasn't as solid as it looked.

It's an interesting dissection that's worth reading in detail. I think it does well to remember that there's more to polling than just asking people who they like in a horse race. Indeed, it seems that the nature of political polling is just as complex and unpredictable as politics themselves.