What if the polls in the B.C. election were not completely wrong? Hold on, let me finish. It might not be of interest to anyone but the wonkiest of wonks, but there does seem to be some indication that turnout played a major role in the missed call by the polls – not in the number of people who turned out to vote, but who those voters were.
The following can be found in the fine print of Ipsos-Reid’s final poll for the B.C. election: “These data were statistically weighted to ensure the sample’s regional and age/sex composition reflects that of the actual B.C. population according to 2011 Census data.” Similar language can be found in the polls released by Angus-Reid and Forum Research in the last days of the campaign.
Thanks to Ipsos-Reid’s admirable full disclosure policy, we can determine the weights that they used for their final poll. Of interest is how they weighted the population by age: 29 per cent of the weighted sample was between the ages of 18 and 34, 36 per cent was between the ages of 35 and 54, and 35 per cent was aged 55 or older.
While these weightings ensure the sample is an accurate reflection of the population, it does not try to estimate what the voting population will look like. And the exit poll conducted by Ipsos-Reid on election day (which closely replicated the election’s results) suggested that voters are very different from the general population.
In 2009, Elections BC found that 49 per cent of voters were 55 or older, while only 16 per cent were between the ages of 18 and 34 (middle-aged voters turned out in about the same proportion as their share of the population). Turnout among the entire electorate was around 52 per cent, but in 2009 turnout among 18 to 34 year-olds was only about 30 per cent, while over two-thirds of British Columbians aged 55 or older cast a ballot. That divergence has the ability to warp a poll considerably.
But it is not merely a question of weighting. If these turnout levels were used to re-weight the final Ipsos-Reid poll, the New Democrats would still have been pegged to win (though by a smaller margin).
What the exit poll suggests, however, is that those who do turn out to the polls are different than the general population within each age group. The exit poll showed that the New Democrats did win the 18-to-34 demographic, but were behind the B.C. Liberals by five points among the 35-to-54 age group and nine points among voters 55 or older.
Interestingly, those numbers are not dissimilar from Ipsos-Reid’s final poll – at least if the voting results are shifted by one column to the right. What that suggests is that the 18-to-34s who voted had similar voting intentions to the 35-to-54s who were polled on the eve of the election, and that the 35-to-54s who voted were more like respondents 55 and older in Ipsos’s final poll. In other words, those who vote generally have a perspective more in common with the general population that is older than them.
This suggests that voting intentions of respondents 55 and older a potential better proxy for turnout than anything else. The final polls of Angus-Reid, Ipsos-Reid, and Forum Research show that the voting intentions of their older respondents almost matched the final result of the election, while the final EKOS Research poll also showed the B.C. Liberals ahead among older voters.
And if we look at the results of Ipsos-Reid’s polling throughout the campaign, it is clear that the B.C. Liberals were making inroads among voters and had, indeed, moved ahead in the final days. This was reflected in other polls, as they also showed a closing gap between older respondents as the campaign wore on. Considering that the Liberals were trailing by such a wide margin even among the oldest voters at the start of the campaign, the narrative of Christy Clark having closed a double-digit gap seems accurate.
This has important implications when it comes to polling. The polls that were in the field at the end of the campaign may have very well been reflecting the political views of the entire B.C. population, but that does not matter much when elections are decided by those who vote. If the pollsters were able to estimate who is likely to cast a ballot, they would be able to get better results.
This means that the challenge pollsters will have in subsequent elections is to develop a model that estimates voter turnout. This is something that is done more regularly in the United States, but even there huge mistakes can be made. While most polls showed Barack Obama heading to re-election in November, Mitt Romney’s internal pollsters had different estimates of who would turn out and so had Mr. Romney convinced that he was heading to victory.
Estimating who will turn out to vote can sometimes be little better than a guess. In some elections, those who turn out may be similar in profile to the general population. In other elections, they may not. For example, in the final poll of the Alberta provincial campaign by Forum Research, the only public poll to show a rapidly closing gap between Wildrose and the Progressive Conservatives, the voting intentions of respondents 55 and older gave the Tories a three-point edge over Wildrose. Among the general population, Forum had Wildrose up by two points. On the other hand, the voting intentions of older respondents in the final Forum poll of the Quebec campaign did not reflect the final result. That may have more to do with the Quebec Liberals’ traditional support among older Quebeckers, or it could have been the dynamics of the campaign itself (post-secondary tuition was a major issue, for example).
But turnout certainly appears to be one piece of the puzzle in explaining why the polls missed the call in British Columbia. Multiple exit and post-election polls (as well as B.C.’s political experience, i.e. the “10-second Socreds”) suggest that there was a small change of heart on election day itself. The expectations that the polls may have put into the mind of voters could have had an effect as well (virtually no Green or Conservative voters believed the Liberals would win a majority, for example). And the spectre of new methodologies’ difficulty in building samples that are reflective of the population, even the voting population, is part of the equation as well.
What most people will learn from the election in British Columbia is to take polls with an ever larger grain of salt than they already were. But what pollsters will learn is that the old way of doing things may not work in Canada anymore. Turnout models, like those used to some success in the United States, need to be developed. Greater disclosure, so that those who are interested can look at the numbers for themselves, needs to be encouraged. Recognition of the potential for errors, either in the polling data or in the turnout estimates, needs to be emphasized before the vote occurs. When the next major election rolls around, we will see if any lessons from the B.C. election will have truly been learned.
Éric Grenier writes about politics and polls at ThreeHundredEight.com .Report Typo/Error