Voters, polling, and Canadian democracy: where have all the flowers gone?

Ekos study provides insights from the 2011 election

The 2011 election may be remembered for many remarkable events, not the least of which was its outcome which surprised pollsters, pundits, and parties to say nothing of voters themselves. Many were left scratching their heads, including the polling gurus who (in theory) should have had the best information and insights of anyone. 

Into this apparent miasma steps Ekos poll-meister Frank Graves who has published a report entitled "Accurate polling, Flawed Forecast" (http://www.ekos.com/admin/articles/FG-2011-06-17.pdf). In it, Graves investigates why polling results from the 2011 federal election poorly predicted the actual electoral outcome, and what lessons this provides, not only for the methodology and utility of polling, but for the future of Canadian democracy.

Graves' springboard is the generally held view that in 2011, federal election polls performed poorly, particularly in not predicting a Conservative majority government. In the final days of the campaign, many polling companies, including Ekos, predicted a Conservative minority government, or even the possibility of an NDP-lead coalition. What went wrong? Graves examines three possible scenarios that could account for the perceived failure of the polls:

1. The polling results were flawed by systematic sampling or measurement error;

2. There was a significant last-minute shift in voting that threw off previous forecasts; and

3. The discrepancy between polling and electoral results was due to a differential voter turnout, and that the composition of actual voters differed from that of all eligible voters.

To tests these, Ekos conducted additional polling (May 20-28, 2011 with almost 2,800 participants). In their report, Graves examines each of these hypotheses in turn.

A. Systematic sampling or measurement error

Maybe there is a flaw in polling methodology itself? To test this, Ekos conducted sample-bias polling (Fig. 2.2 of the report) illustrating how those polled in 2011 actually voted in 2008. This showed both stability and little variance from actual percentages of votes cast (although the proportion of NDP voters polled appears slightly low), an indication that their polling techniques were not misrepresenting the proportions of political party supporters in the general population. Graves also examined Ekos measurement techniques, such as asking vote intention questions in a neutral and consistent manner, and ensuring that their samples include "the entire (non-institutionalised, charter language speaking) population of voters". Graves says "We therefore conclude the testing of survey bias hypothesis with a high level of confidence that the gap was not a product of faulty data or sample bias." In other words, the methodological and statistical chops were in good order.

B. Late shifts in voting

The claim has been advanced in some circles that a significant number of voters switched from Liberal to the Conservative at the last moment in the election in order to head-off an NDP-lead coalition government. Could this be true? Figure 3.1 of the Ekos report indicates that 56% of those who voted had already decided who to support before the election campaign began; 15% (26% of youth; 20% of those in Quebec) decided after the leader's debate; 18% (28% of youth) in the last week of the campaign; and only 9% (16% of youth) on the last day (the so-called "late shifters").

Although 21% of Canadians changed their mind as to who to support during the campaign (Fig. 3.2), only 21% of these (i.e., 21% x 21% = 4.4% of total voters) shifted their support on election day, an amount that Graves calls "picayune." Fig 3.2 of the Ekos report shows which parties voters shifted away from, but not which parties they shifted to. It is nonetheless clear that Liberals suffered the greatest losses since 36% (of that 21% who changed their minds) shifted away from the Liberals. In the case of the Conservatives it was 20%, the NDP 20%, the Green Party 11% and the Bloc Québécois 12%.

Only a miniscule 6% (Fig. 3.3) of the 21% who shifted, in other words 1.26% of all voters, did so because they were "concerned about an NDP-led coalition" (and in Ontario polling there wasn't a single "late-shifter" who did so for this reason). In contrast, 43% did so because "It was time for a majority government" and 32% because "We need to stay on a sound economic trajectory."

Ekos also polled to find out if polling itself had had an impact on the election (Fig. 3.4). They found that 76% of voters said the polling results hadn't influenced them at all, whereas 13% said polling had influenced them to a "moderate extent", and 8% to a "great extent." Ekos concluded that polls "had no clear bearing on the final outcome of the election." Although 21% (13+8) is a non-insignificant proportion of voters, Ekos informed me that when they examined the demographics of those who say the polls influenced their decision, there was no clear evidence that any one party benefited from their impact.

Graves concluded that "The counterfactual hypothesis of what this election would have looked like if the polls hadn’t been available to voters is that things would probably have been the same."

C. Actual voter composition differs from eligible voter composition

So, if polling accurately represented the political viewpoints of eligible voters, then the difference between outcome and polling must therefore be a result of incorrect forecasting. How could this be? If the entire polling sample was unbiased, then the projected support (through simple arithmetic) of the 38.6% of Canadians who didn't vote must be approximately, Conservative 24.8%, NDP 32.1%, Liberal 24.3%, Green 9.3%, Bloc Québécois 7.0%, and Other 2.5% (Fig. 4.1).
 

Graves then adduces a variety of evidence that indicates that "this hypothetical distribution of vote intention amongst the non-voters is highly plausible." The most important of these relates to the percentage age composition of the people who didn'tvote. The proportion of people who didn't vote according to age classes (Fig. 4.2) was: age 65+ = 4%; age 45-64 = 7%; age 25-44 = 15%; age < 25 = 24%. Thus, the younger voters were, the more likely it is that they didn't vote. Almost a quarter of people under 25 didn't show up at the polls.

A surprisingly accurate measure of this appears to be the distinction between those who have both landlines and cell phones, and those who posses cell phones only(Fig 4.3). The proportionate support amongst the former is quite close to the actual 2011 electoral results; the proportion of the latter is much closer to that of the non-voting segment of Canadians (and the report makes the apparently plausible claim that the majority of "cell phone only" users are under 40). For example, Conservative supporters comprised 24.8% of the non-voting constituency (Fig. 4.1) while cell phone only Conservative supporters comprised 24.1% (Fig. 4.3), a nearly identical proportion. Similarly for the NDP the proportions are 32.1% versus 33.1%; for the Liberal Party it is 24.3% versus 21.1%; and for the Green Party 9.3% versus 7.6%.

All this leads Graves to conclude that the difference between polling results and electoral outcomes resulted from statistically sound sampling methodology (including an accurate sampling of "cell phone only" users) but that those predominantly younger voters, turned out at the polls in far smaller proportions than did older voters.

At the same time the Ekos report also released the hitherto unreleased results of their "commitment index" (Fig. 5.1) that factored into polling results a variety of parameters related to how likely it was that eligible voters would actually turn out and become actual vote. Employing this commitment index lead to results that much more closely approximate the actual 2011 election outcome. Although calculated before the May 2 election they were not released at the time since Graves lost his nerve "due to inconsistency between the overall index and 'absolutely certain to vote' [measurement] (which turned out to be quite unreliable)."

On the basis of this analysis Graves poses a series of provocative rhetorical questions:

"If there is a widening gulf between the voting and non-voting populations, where do we hear the voice of the non-voter? Can we just say that this large group – nearly 40 per cent of all voters and over half of under 50 Canada can be safely ignored? What if the factors producing non-voting are not simply "laziness and apathy"? What if the lack of voting is linked to alienation and conscious political strategies designed to suppress the interest of those voters? What if there is a mutually reinforcing tendency to further weaken "next Canada's" interest in federal government by virtue of a federal agenda which systematically undervalues and deemphasises their interests and values and emphasises the interests and values of its constituency? Does this become a particularly troubling problem at a point where our highly unusual demographics have produced a voter whose median age is around 60? Are we fashioning the future Canada in the image of those who are disengaging or those poised to inherit the positions of authority and influence in short order?"

Project Democracy recognizes the critical importance of working to engage young people in the political process. This may be an uphill battle, not only because of the erratic and unrepresentative outcomes of the first-past-the-post electoral system, but perhaps also because of what Graves suggests may be conscious political strategies designed to suppress the interest of younger voters. In the last election campaign, we certainly witnessed an emphasis on the part of the Harper Conservatives on law and order issues, militarization, purported economic stability, the abolition of the long-gun registry, and other issues that have very little resonance for younger voters. In contrast issues such as the environment, childcare, affordable education, electoral reform, and other issues that are of concern to younger citizens were conspicuously absent from the Conservative playbook.

This emphasis to recast the political agenda in such a way as to alienates and disengage the younger electoral constituency is cause for considerable concern and indicates that the gulf between the governing and the governed may be rapidly widening. While the population of seniors and baby-boomers ages, the members of Gen X and Gen Y find their interests marginalized by the Harper government, with the consequence that they increasingly eschew the ballot box. Such a trend would be an ill omen for the state of Canadian democracy. If political and electoral systems disengage from the citizenry they are supposed to serve, the quality of governance is certain to suffer – with unpredictable consequences for the state of the nation.

Christopher Majka

Project Democracy