By day, Philippe J. Fournier teaches physics and astronomy at Cégep de Saint-Laurent in Montreal. His evenings and weekends, however, are devoted to a very different kind of expertise—election modelling.
Fournier’s reputation is built on his creation of 338Canada, a platform that has become a go-to resource for political junkies. By aggregating polling data, demographic trends, and electoral history, Fournier creates sophisticated projections that routinely outperform traditional punditry. Since launching in 2016, his model has called results correctly in over 1,600 electoral districts, achieving an astonishing 90 percent accuracy rate.
While there are other prominent polling analysts, such as Darrell Bricker, what sets Fournier apart isn’t just his track record but how he blends statistical precision with insights into dynamics such as voter psychology, capturing nuances raw polling often misses. This has made him a fixture in Canadian media. It’s also what led us to ask him to become a contributing writer.
We spoke recently over Zoom. In our conversation—which has been edited here for length and clarity—we discussed his process, his commitment to accuracy, the upcoming federal election, and the value of applying scientific rigour to the unpredictability of politics.
I want to start by talking about a correction you had us make when you first joined us. We introduced you as a “pollster,” and you swiftly emailed saying, no, “I’m a polls analyst.” For the record, can you explain the difference?
I don’t conduct polling. This is done by professional firms that have an infrastructure that I don’t have and don’t wish to have. Instead, I take their data and try to make something out of it—calculate seat projections and odds of winning. I also rate the firms. That’s the other thing. I can’t participate in the polling since I rate the polling. It’s like if Standard and Poor opens a bank. It doesn’t work. So the analogy I like to use is—they grow the wheat and I try to make bread.
How do you trust the quality of that wheat? What’s your criteria? I imagine not all polling firms are created equal.
The simple answer: I check the track record. If a new polling firm pops up, I’m going to include their polling but give it very low weight until they show that they can get close to election results, either a by-election or a general election. A firm with a great track record will have more weight in my calculation—and also more leeway if they pull out numbers that are a bit strange. If Leger suddenly shows the federal Liberals tied with the Conservatives, I would probably ask some questions, but I’ll go—Yeah, it’s Leger, right? It’s not just some obscure pollster.
What does it take to become a “Leger poll”? What does it take to become a gold standard in the polling industry?
Consistency. You have to show that your sample can be representative in many places—provincially, municipally, and federally. You have to show that when you get it right, it’s not a fluke. In the case of Leger, if you look at their track record in the past decade, it’s almost straight “A”s throughout. In the big provinces, in the smaller provinces, at the federal level, they’re never too far off. This is what it takes to be an A or A-plus pollster in my rating. You have to show that you can consistently call elections within the margin of error.
How do you go about your day as an analyst? Do you check polls first thing?
Every day, I go through the polls from almost every province. I have some help with this, because there’s a lot of data out there. I cover nine provinces. I don’t cover Prince Edward Island. It’s not that it’s not interesting, but there are very few polls there. Also, it’s a small place. I also don’t cover the territories—again, very few polls. But I check the numbers from the federal level and, for the bigger cities, the municipal level. I crunch those numbers and, federally, I update every Sunday. It’s good to have regular updates, because it drives traffic, and polls are rarely released on Sunday. During a campaign—especially a federal campaign, like the one we will probably have in the spring—I’m going to update every night, because there will be polls all the time. If you want to have a good picture of what’s going on, especially if there are significant movements in the polls, you have to put in the work to update every day.
Wait, you said spring. You think an election will happen in the spring?
I don’t think the opposition will have an opportunity to make the government fall sooner. I don’t see how the Liberals can pass another budget. We’ll see. Some politicians are Teflon. But I think the most likely scenario would be late spring, maybe around May.
Why do you even do this? You’re an astrophysicist! How did you get into this game?
I did a master’s degree in astrophysics at Laval University around 2004 and got myself a part-time job teaching physics at Cégep de Saint-Laurent. I was twenty-four, and I decided that I should get pedagogical skills to last in this job. So I signed up for the education certificate at Concordia and took a political science class. I thought it was very interesting. Completely different from physics and astrophysics. Different setting as well. Usually, I sat down with my calculator, my pens and pencil case, ready to take notes. But here it was just people talking. I came to understand that I really, really love numbers in politics—more than I like politics itself. I finished my education degree and went back to teaching. Then, during the 2016 US presidential election, I followed the campaign on a day-to-day basis. And I noticed a huge discrepancy between what the talking heads and the columnists were saying and what the numbers were saying. The numbers were suggesting a close race, but many people just couldn’t believe a so-called simpleton like Trump could win. That’s when I realized that the media, including Canadian media, needed more scientists to talk about politics.
Since the time you started doing this, has polling changed, or has the way of collecting information changed? And have you needed to adapt to those changes?
The adaptation is more on the side of pollsters. Polling is an ever-changing industry. There was a golden age in the ’70s, ’80s, and ’90s, when everybody had a landline and you could just call them. Cellphones have now made it harder to reach people, especially young people. But pollsters have to find a way—otherwise they go out of business, right? Still, there’s no question there’s been a Darwinian winnowing. If you look at the firms we had maybe fifteen years ago, half don’t exist anymore. Many of them were really good, but they could not adapt to the new reality.
Polling now happens online.
It’s more and more of an online thing, yes. Polls are still done over the phone, and the firms that conduct them are quite good. There’s Janet Brown, in Alberta, who still does phone polling, and her numbers are always really close. Nanos Research also does phone polling at the federal level. So it still exists. You have to have people in call centres sitting at their desks with their headphones and calling numbers.
I guess I should have started with this question, but—as simple as you can make it, what is polling?
Imagine you have a big reservoir, and imagine that reservoir is filled with coloured balls—green, blue, yellow, red. And let’s say, at random, you start picking out balls. One is red, one is blue, then a blue one again. At some point, since you are picking at random, the sample that you are building will look like the entirety of the reservoir. The odds that you will keep picking, say, red balls at random are nil. It doesn’t happen. This is the nature of statistics. And so, if you’re a pollster and you’re trying to take a representative sample of the population, then you have to make the sample as random as possible. If randomness is involved, your sample will be close to the total within a margin of error that can be calculated.
What’s the trick for achieving that randomness?
You select respondents in a way that ensures every voter within the target population has an equal chance of being included in the sample. This is typically done through methods like random digit dialling for phone surveys or using algorithms in online surveys which pick out respondents from a large pool of potential voters. But true randomness is impossible to achieve with perfection, so pollsters also apply statistical weighting to adjust for underrepresented groups, based on demographic factors like age, gender, and region. This helps make the sample more reflective of the broader population.
And without randomness, findings become skewed.
Yes. That’s why polling relies on an infrastructure that needs maintaining. People move, die, are born, become eighteen. If your sample stops being representative of the overall population, your numbers fail. And if your numbers fail during an election, that’s bad publicity, and suddenly you get fewer contracts. Pollsters make money by showcasing how well they do in an election and turning to, say, an airline company and saying—Look, you can trust our numbers. We called it correctly.
Leger uses its success in political polling as a calling card for other industries?
Oh, 100 percent. They don’t do political polling for profit. There’s very little profit in it, unless you get to poll for a political party, and even then, you can’t rely only on that. Political polling is like advertising. Leger invests in it to prove their salt as market researchers. There was a new poll from Leger recently, and in the first pages, they put the result of the 2021 federal election next to their final numbers and reminded readers that they were the closest firm in the country. And it’s true. They were the closest.
Has your way of analyzing polls changed over the years? Are you a little bit bolder now that you feel comfortable reading the signs?
I don’t think my method has changed much. I make adjustments on a regular basis. Every election, I try to learn with every missed call or mistake. We had the BC election recently, and my projections were pretty good, considering there was a new electoral map, a new party, a new contender. But there were some surprises. And I always ask myself: Was there a way in the numbers to make that call accurately? If yes, then I try to correct the course.
What makes the model that you’ve built so accurate? Is there a secret sauce?
There’s no secret sauce. My methodology is on the website, and everybody can read it. It’s pretty transparent. You’ll never see me brag online about how well I did in an election, because if the data is good, the results should be okay. I also always remember that many of my students, and now former students, are watching, and if I start to be the hot-take guy who knows that something is absolutely true—that so and so will lose or their rival will win—I will not be true to the process. I try to see polling through the eyes of a scientist. Always be humble before the data, and remember that we’re dealing with human beings. As an astrophysicist, I’m used to dealing with complex stuff. But humans, being irrational, can be far more complicated than electrons or galaxies. When you have a poll, always remember: this is an approximate thing; this is not a perfect thing.
But what are you looking for in the numbers? What do you need to see to start making projections or predictions?
My first thought is: Where does this data come from? Then I look for comparison between the new data and past trends. Is there continuity in the numbers, or are new tendencies emerging? Obviously, looking at horse-race numbers or government satisfaction is different than crunching data on more complex issues, such as tariffs or NATO. Then if I see unexpected trends, I ask myself whether this is merely noise or part of a sustained shift? Good polling requires regular polling. We won’t know whether opinions are shifting or improving or eroding if we look at a snapshot. We have to interpret every poll within context—and this is sometimes what is missing from mainstream media analysis of political polls.
Do you tend to get into fights about your polls with conservatives or liberals? Do you get pushback from political parties?
Honestly, I don’t give a shit what people think. I have good connections with political parties in Quebec at the federal level—some in Ontario as well. We exchange information sometimes, and it’s very polite and respectful. But if some randos online are unhappy with my numbers? So be it. Usually, it’s because their party is trailing and they don’t like it.
The frustration, I imagine, has to do with the sense that you have sway in how people feel about an election.
Look, I’m just providing information. Do polls influence voters? That’s a good question, and the answer for me is: information influences voters, but lack of information also influences voters. Say I show that the NDP vote might hold the key to the Liberals beating the Conservatives—if enough people switch. If it’s true, it’s information, and people can decide to do whatever they want. There are many NDP voters that will never vote for the Liberals, even if they know they split the vote. That’s not my problem. People are free to do whatever they want.
What’s the most useful way for someone to understand what polls represent?
In an election, parties will do everything they can to convince you that they have the momentum. But if a poll is done right, if it’s done by professionals, it’s the only objective information you have about the state of a campaign. Everything else is spin. And so when I hear that polls have influence, it drives me nuts. Pollsters are simply giving you the score of the game. Could you imagine going to a Habs–Bruins match, and it’s three–two for the Bruins after two periods, and Habs fans say, Shhh, don’t tell me the score; you’ll influence the outcome!
What’s the story of Canadian politics right now? Is it as obvious as it seems? Liberals sinking, Conservatives rising, NDP struggling?
It depends on how much resolution you want your picture to have, but sometimes Occam’s razor enters the fray, and the simplest explanation makes the most sense. In Canada, unlike the United States, we don’t have term limits, but we have natural term limits. Usually, it’s a decade in power, and then people get sick of you and vote you out. Right now, the Liberals have been in power almost a decade—a decade that included crises we hadn’t seen before. I mean, what was the worst thing Pierre Trudeau had to deal with? The October Crisis? I don’t want to diminish Quebec separatism, but you really want to compare that to war in the Middle East, war in Ukraine, and a global pandemic? Sure, maybe Liberals made self-inflicted mistakes, but the rest of it is about being in power a decade, a rough ten years. Sometimes it’s as simple as that.
What’s the version with a bit more resolution? What do we see when we zoom in?
That Conservatives are much, much better campaigners than Liberals—and have been for a while. That Trudeau has had to deal with an increasing number of conservative premiers who disagree on many issues, and it makes everything harder. But when Pierre Poilievre becomes prime minister, his base will love him, others will not as much, and after a few years, the tables will turn. It’s a natural cycle in Canadian politics.
