Wednesday, October 16, 2019

Election 2019: Where are we headed?

Polls and Seats: Divergent messages

The polls in this campaign appear on the surface to be telling a similar story: a close election, outcome uncertain. However, different surveys, examined closely, reveal dramatic contrasts.

Here are two examples covering the same October 8 to 10 period (the day after the English debate to the day of the second French debate), one survey from Nanos Research, the other from Angus Reid. I provide a seat projection below based on the provincial/regional numbers from the two polls. What matters, however, is the dramatic difference between the two: one a Liberal plurality of seats won, the other Conservative.


There is also higher uncertainty because the polls are more variable at the level of regional details. So which survey should one trust? Even past performance is not an absolutely reliable guide since a given polling firm can have a bad result. That said, the best performing firm in elections since 2006 has overwhelmingly been Nanos. With the exception of 2008 when they had a bad result, Nanos has either been the most accurate pollster, or just a whisker away from being the most accurate. The table below tells us what happened in 2015. It compares the national numbers in the polls for each party compared to the election result. Overall the polls performed well. The firms are ranked by total absolute error. The red numbers indicate numbers that fell below actual vote shares, the black are above. The polling was fairly accurate, mostly within the margin of error.


I have been tracking polls compared to election results for some time. Generally polls conducted by phone either with a live interviewer or electronically by a method called Interactive Voice Response (IVR) have been the most accurate as in the table above. However, on occasion, an online survey outperformed the others.

I still regard online polls as experimental partly because there is no standard commonly accepted methodology. For example, some of the internet panels (meant to stand in for the Canadian or provincial population as a whole) are randomly recruited on the phone while others are self-selected or put together using other methods.

I would like to see full transparency in this area, which does not happen now: the size of the panel (including the size of provincial and regional sections), how it is selected, the number of surveys distributed and response rate. Online polls are attractive to polling firms because they are cost-efficient so they are probably here to stay. Online polls do generally report a margin of error. However, it should be remembered that the margin of error is simply a function of sample size. Online poll samples are drawn from the limited population of the internet panel, whereas phone or IVR polls sample all of the population with access to a phone (land line or cell). This difference is something that appears to be widely misunderstood. When you see margin of error in an online poll the population that it applies to is the population of the panel not the population of Canada. Sometimes a poll such as the Reid poll above will describe the margin of error using fudge language such as: "For comparison purposes only, a probability sample of this size would carry a margin of error of +/- 1.8 percentage points, 19 times out of 20." In fact the margin of error does apply to a sample of the size noted but the population it applies to is that of the internet panel.

Election trends: growth in support for the NDP and the Bloc

It is clear by now that the NDP has dramatically improved its prospects in the last couple of weeks, in part because of Jagmeet Singh's performance in the debates and his deft response to the Trudeau blackface controversy and other issues.  The Bloc has been growing too, in part because the new leader, Yves-François Blanchet, is an effective communicator, but also because of the appeal in Quebec of the province's xenophobic Bill 21 (it imposes restrictions on the headgear of religious minorities) especially to older, more rural residents (see the demographic numbers on page 5 of this Leger survey). However, Chantal Hebert also makes a convincing case that other factors are at play.

I don't think the NDP is doing better because of Singh's debate performances per se. Instead, it has rejigged the campaign because his performance has drawn voters to take a new look at the positions he has emphasized. Voters don't look at elections as beauty contests (I get the impression some reporters do), but rather as a way to solve problems they face, so the NDP's promises in areas like dental care or housing and other issues are ultimately what is delivering the support to the NDP not the qualities of his debate performance.

In a way this should not be a surprise. All governments disappoint and it has been clear for some time that some of the electorate has been casting about for an alternative. The NDP does appear in current circumstances to be drawing at least some support from both the Liberals and the Conservatives reflecting the new dynamic.

Historically, the NDP has done well in elections following Liberal majorities. This happened in the elections following Liberal wins in 1968, 1974, 1993 and 2000. It is not a new phenomenon. At the moment the NDP would not match its 2015 seat total of 44 but there are several days to go. Intuitively, I have wondered if this election might produce a scenario like that in 1972 when the Liberals of Trudeau the elder finished a single seat ahead of the Stanfield PCs. The next two years brought a Liberal government dependent on support from the NDP.  However, increasing Bloc strength suggests it may be a House of Commons where neither major party can govern with the support of just one opposition party. This suggests a period of parliamentary instability to follow.

It will be remembered by some that there was considerable speculation about how well the Green Party would do following their by-election win in May. The situation is actually the same now as then. The Greens have improved in the polls since the last election but they only have real strength on Vancouver Island. Their leader Elizabeth May has stumbled during the campaign in part because of her dishonesty. For example, she claimed a new Quebec recruit from the ranks of the NDP was a sovereignist not a separatist, although the two words have the same practical meaning (the new recruit used 'separate' in French himself). She has gotten away with a great deal of this kind of use of political weasel words for a long time, but the real problem for the Greens is that they just don't have much organizational muscle off the island.

The Conservatives could have improved their prospects if they weren't so determined to be a party well to the right. Andrew Scheer doubled down on this approach in the English debate - it seems he is somehow trying to thread the needle - but this behaviour overall is self-marginalizing in the context of the reality of Canadians' political views. It also inspires some fear among progressives leading many of them to look to cast a tactical ballot, even though that is a practical alternative in only a small number of circumstances.

High turnout at advance polls mean some of the result is already baked in, i.e. developments in the last week cannot change votes already cast (two million ballots were cast in the first two days of advance voting). That suggests the current numbers already guarantee no party will control a majority in the House of Commons come October 21.