Canada before the debates. Standstill?

 

Hello,

In this blog post, I will address three questions. First, I present the evolution of voting intentions since the beginning of January. Second, I examine whether there has been any changes since the start of the election campaign on March 23.  I conduct these analyses for Canada as a whole and separately for Ontario and Quebec. Finally, I look at whether the estimates differ depending on the methodologies used (for Canada as a whole only).

The graphs I present are similar to those found on media aggregators such as 338Canada, CBC Poll Tracker, or Polling Canada. However, there are methodological differences that may lead to differences in estimates. These differences are outlined at the end of this post.

What has happened since January?

The first graph shows that we went from a gap of over 20 percentage points in favor of the Conservative Party of Canada (CPC) (45% to 23%) at the beginning of the year, to an 8-point lead for the Liberal Party of Canada (LPC) over the CPC today. Major events that occurred in the meantime include Justin Trudeau’s resignation on January 6, followed by his official replacement by Mark Carney on March 9; Donald Trump’s inauguration as President of the United States on January 20 and the subsequent developments; and finally, the official launch of the electoral campaign on March 23. It is unclear whether the rise in support for the LPC began with Trudeau’s resignation, Trump’s inauguration, or even the emergence of Mark Carney as a potential candidate. One thing is certain: the increase continued at least until the start of the election campaign.

Vote intentions for the LPC now stand at around 45 percent, the CPC at 37 percent, the New Democratic Party (NDP) at 8 percent, the Bloc Québécois (BQ) at 5 percent, and the Green Party at approximately 2–3 percent. The LPC gained support not only at the expense of the CPC but also from the NDP (down 10 points) and the Bloc Québécois (down 4 points).

How does this translate to Quebec and Ontario?

How does this translate to Quebec and Ontario? The second graph shows that the LPC is also polling at nearly 45 percent in Quebec, while the other parties trail far behind. At the beginning of the year, the CPC was significantly less ahead of the LPC in Quebec. The LPC has made a rebound just as significant as at the national level, but this has come at the expense of all other parties, particularly the Bloc Québécois, which has lost nearly 15 points, compared to a 2–3 point drop for the CPC and a four-point drop for the NDP.

It is important to note that, given the sample sizes, the variation in estimates across polling firms is greater than it is for Canada as a whole.

In Ontario, the LPC also gained 20 points, but since it started at 30 percent, it is now estimated at 50 percent. The CPC would have lost nearly six points, now standing at 39 percent, the NDP lost ten points (now at 7.5 percent), and the Green Party lost two points (now at two percent).

---

What has been happening since the start of the campaign?

 
The next three graphs show the evolution of voting intentions since the electoral campaign began. The increase observed since January disappears when we consider only the polls conducted at least partially during the campaign—as if everyone is waiting, or as if voting intentions are already fixed.

The beginning of the graph shows a slight rise for the LPC, but afterward, everything stabilizes. The same goes for the other parties. Estimates based solely on campaign-period polls are not very different from those that included all polls since January. The LPC stands at 45%, the CPC at 38%, the NDP at 9%, the Bloc at 5%, and the Green Party at 2%. The gap between the two main parties is not narrowing, contrary to claims made by some pollsters who based their conclusions solely on their own polls—without accounting for the margin of error.

In Quebec, the pre-campaign movement had not yet ended at the start of the campaign, but as in the rest of Canada, voting intentions are now stable. The LPC is estimated at 43%, the CPC and the Bloc are tied at 24-25%, the NDP at 6%, and the Green Party at 2%.

Ontario is also very calm. Nothing is moving. The LPC is at 50%, the CPC at 39%, the NDP at 8%, and the Green Party at 2%.

---

But can we trust the polls? Are there systematic differences depending on the methodologies used?

The 128 polls conducted since the beginning of the campaign used the web in 44% of cases, automated telephone mode (IVR) in 23%, and telephone or a mixed mode (telephone + web, SMS-to-web) in 32%. It is worth noting that one firm, Mainstreet, which had been using SMS-to-web, recently switched to IVR mode.

The graph shows the evolution of voting intentions by mode of data collection. IVR and mixed modes tend to produce higher vote intention estimates than the web mode. Statistical analyses show that IVR mode gives the Liberal Party (LPC) 4.2 percentage points more, and mixed modes 1.9 points more than the web mode, after controlling for the timing of each poll.

 

If we consider only the 53 polls conducted since the start of the campaign, the situation is similar, but the differences are smaller. IVR mode consistently yields higher voting intentions than web mode. Statistically, IVR polls show results that are three percentage points higher, and mixed-mode polls 1.3 points higher than web polls, a difference that is not statistically significant in the latter case.

 

Conclusion

On the eve of the debate, the situation is relatively clear and stable. Liberals are ahead and sure to form the next governement if nothing changes before the end of the campaign. However, it is important to pay attention to the data collection mode used by polling firms.Liberals are not as much ahead if you look only at the web polls. How can these differences be explained? Depending on the method used, there may be variations in the types of people who are reached. Some individuals will never participate in a web survey, while others never respond to an automated phone survey. These differences may, however, diminish as the election is nearer—among other reasons, because opinions tend to change less.

With that, enjoy the debate!

 

Credit to Anthony Pelletier for data entry and graph production.

Methodology:

The methodology I use differs from that of aggregators in two key ways. In the graphs, each point represents a vote intention estimate from a pollster, placed at the midpoint of the data collection period. The lines represent estimates of vote intention based on local regression. However, for tracking polls or rolling polls, I record them only once during the collection period. Essentially, if the period spans three days, I record the estimate once every three days in order to avoid redundancy in the data, which would be statistically inappropriate. Aggregators, by contrast, tend to record the estimates every day but assign them reduced weights (e.g., one-third each day).

Furthermore, when plotting the vote intention trends, the local regression gives less weight to poll estimates that are further from the others. This procedure smooths the trend and offsets the influence of polls whose estimates deviate significantly from the rest.