Provincial update

Alright Ontario, we’re on day 8, 21 more days to go. And… we might have movement! I say might because these changes are within the margin of error, so we’ll have to see what happens over the next few days. Also, if I sound excited, it’s only me as a pollster who was worried about spending 28 days saying nothing has changed. My Liberal self is holding my breath. 

Ford has reached a new peak in my polling. He and the PCs are at 51%, Crombie and the OLP are at 22%, Stiles/NDP at 18%, and the Greens are at 9%. If you compare to the baseline, the Liberals are down 3 points, and the NDP are up 2. 

To give you a bit more information about how my polling works, for the rest of the campaign, I’ll be doing 3-day rollups. I collect ~225 interviews each day and combine that data with the previous 2 days. This helps us identify trends while minimizing noise. So Crombie dropping 2 points since yesterday’s 3-day rollup indicates that last night wasn’t great for her. But it could have been a blip. We’ll see if the trend holds tonight and tomorrow. 

National polling volatility 

Now, let’s talk federal polling. For the past couple of weeks, we’ve seen a lot of national polling with a lot of volatility. Has the race tightened? Well, it depends on who you ask and how they asked voters. 

To get into the weeds, I pulled data from the polling aggregator 338 Canada. I focused on the past 2 months, which is when we’ve really started to see the polls diverge. Some pollsters include “other” as a response category, so their listed polling figures don’t add up to 100. I wanted to be sure that we’re looking at apples-to-apples comparisons, so I standardized the data (omitted anyone who didn’t plan to vote CPC, LPC, NDP, BQ, PPC, or GPC). I then grouped the polls by methodology, dividing them into IVR, online panel, and live-caller phone. 

Let’s kick off with Interactive Voice Response or IVR. This is an automated type of polling where individuals receive a phone call, hear a recorded voice, and submit their responses via voice or their keypad. Most of the below polls were fielded by EKOS, but there is one from Pallas Data. In case you’re very in the know of the Canadian polling landscape (you poor thing, you), you might notice Mainstreet Research isn’t included in this list. From my understanding, the public polls they released over this time period used a text-to-web methodology, although they do use IVR at times. 

As you can see, the polls have narrowed dramatically since the middle of December, going from a 26-point spread to just 3. Exciting, isn’t it?!

Most pollsters that release data publicly use online panels, although not the same ones. Angus Reid Institute, Ipsos, Léger, and Innovative Research each have their own panels. Innovative supplements its panel with Lucid/Cint, which Abacus and I also use. I’m not sure about Research Co.

Regardless of the panel/pollster, online polls have been pretty consistent for the past couple of months. Up until mid-January, there was a little bit of bouncing around, but the polls mostly stayed within a 5-point range for each of the 3 major parties. There does appear to be some narrowing over the past couple of weeks. Instead of a 23-point spread, as was the case in early December, we’re seeing an 18-point spread between the Conservatives and Liberals.

Only one pollster, Nanos Research, publicly releases phone polls. Their polling has been comparable to the online polls, and they, too, have observed some narrowing of the race.

I don’t share this data to conclude that IVR is wrong. I feel more confident in online polling, which is why I use it. While getting the horserace right is important, it is only one indicator of public opinion. I also like to focus on what’s happening under the surface and on persuasive messaging, which is easiest to do online.

The uncomfortable truth about polling is that it’s hard, and you really only get to test your methodology and assumptions on election day. Low stakes, right? There are so many things that can go wrong. The one that keeps me up at night is non-response error, which is when certain types of people do not answer your polls. This was the big fear of US polling misses – that pollsters weren’t speaking to untrusting MAGA supporters. 

What makes non-response so difficult to detect is that you can hit all your demographic targets and still be unknowingly off on ideological representation. In the US, some pollsters weigh on past presidential vote as a check on this, but with our multi-party system, it’s much more difficult to do. 

My theory is that non-response… or hyperresponse, I guess… is why IVR and online/phone are so divergent now. Our political events are so much more dramatic than usual, and if you’re an enthusiastic Liberal, this signals political renewal. I think IVR, with its very low response rate, is more sensitive to changes in enthusiasm, meaning they’re receiving more non-responses from people who aren’t inspired by a change in LPC leadership or who are terrified of Trump. But we shall see! 

Hope this provides some clarity on the state of the race(s)! À demain.