How a polling analyst projects the results in each riding

Robert Martin is an analyst with Mainstreet Research. He uses the company's polling data to create seat projections, as you can see on this polling dashboard created by iPolitics and QP Briefing. 

Less than two months before the provincial election in Ontario, iPolitics called him to ask how he does it.

(In the interest of full disclosure, Mainstreet is a shareholder in the ownership of iPolitics and QP Briefing.)

As simply as you can, please tell us how you arrive at those riding-level projections.

Basically, we look at top-line polling and regional cross tabs — how each party is doing in each region — and at demographics: age, income, and education. So, based on the demographics of the riding and how each party is doing in a region, we're able to make projections down to the seat level.

This is a little different than some other traditional election models. We rely a lot more on the demographics to project how seats move from election to election, whereas other models are a little more static and reliant on vote share. But this model uses demographics to augment that and to complement the vote-share changes, in order to figure out in which ridings a party will outperform or underperform.

All of this depends on people answering the poll questions as if the election were today, right? 

Yeah.

Will the election be held today?

Not that I'm aware of, no. And that's actually an extremely important point to remember. We're talking here on April 11. We're still almost two months away (from the provincial election on June 2), right? So a lot can change between now and then. And our model is based on what the data says today. This is what we think would happen with the election being held today.

In some other countries, particularly the U.S., some people try to predict results using approval ratings and things like that. Here in Ontario, those trends don't really exist.

In 2018, a massive surge in provincial NDP support almost tied the party with the Conservatives. Eventually, it lost. You see changes that are completely unpredictable, so we're going with what the data says on any given day and projecting from that. Tomorrow it might say something very different.

Let's talk about things that would change the seat distribution, such as the Liberal-NDP split.

Yeah. Right now, the poll says they're functionally tied. They're within one per cent of each other, sitting around 25 points each to the Conservatives' 39. That's a huge problem for both parties.

If you look back to 2018, the NDP won 40 seats. That was because they were able to coalesce the anti-(Doug) Ford vote. So, when you have that, you're able to squeeze out those extra voters in those seats and take on the Tories. But now, they're helplessly splitting the vote, and it's not uncommon to see a riding where the Conservatives are at about 35 per cent and the Liberals and NDP are both tied in the high 20s.

I personally don't believe that will keep happening, as we get closer to the election. I think we'll see one party functionally eat the other party; I just don't know which one it will be. I could be wrong, but that's what tends to happen in Ontario.

So what does your model take into account? Does it look at the last election?

The last election is massively important. Many people will try to throw in data from other elections, like federal elections, but I'm a purist. I go to the last provincial election for the overwhelming majority of data.

The only difference is when you have to deal with personal votes. Say you have a former cabinet minister, like Charles Sousa. And in St. Catharines —

Jim Bradley.

Right. He's from my neck of the woods. He massively over-performed expectations, but he's not running again. There are a lot of examples: Liberals who should have been blown out, but weren't because of their personal name brand in 2018. So, if they're not running again, you have to adjust your expectations of the Liberals, because they won't do as well.

It's also extremely important to use the last election in Ontario specifically, because Ford Nation is a very distinct brand from the federal Conservatives. Ford isn't viewed the same way Andrew Scheer or Erin O'Toole was; he's way more popular. In Etobicoke North, the capital of Ford Nation and his seat itself, the federal Liberals won over 50 per cent of the vote. A lot of people are clearly pro-Liberal federally but pro-Doug Ford provincially. And if you start mixing in data from other elections, you start losing that.

So, essentially, you take the 2018 results, factor in how things might change according to current polling, then throw in concerns particular to individual ridings?

Yeah, that's exactly how it works, except maybe less emphasis on the effects of individual candidates. We try to have very minimal impact from candidate effects. It's basically this: You look at the past vote; you look at the demographics; you look at the demographic polling; and you figure out who should win, and that's that.

Does it matter that, in 2018, we had a very strange situation wherein the incumbent premier essentially declared defeat before election day?

Not really. The results are the provincial top line, so, for every one of those voters Kathleen Wynne functionally gave away, they're not counted as Liberals; they're counted as who they voted for. So I'm not concerned about effects like that.

What's the difference between a safe seat, a likely seat, and a toss-up?

We give safe seats a probability of over 95 per cent. It's very unlikely any such seats would move, which is why they're safe, such as those in rural Ontario.

A likely seat has a 75 to 95 per cent chance of holding. Again, not as safe as a safe seat, but, unless something really weird happens, such as a candidate effect, those seats will probably go how they're probably going.

The narrow seats are between 55 and 75 per cent. Those are the ones where, if the polling is off by a couple of percentage points one way or the other, they could flip. If the Liberals over-perform their polls by three points, for instance, you might see some narrow Progressive Conservative holds. Suddenly, those could go to the Liberals. Or vice-versa.

The seats you really want to focus on are the toss-ups, which no party has more than a 55 per cent chance of winning.

To be clear, we're still talking as though the election were being held today, not if the seats will be safe two months from now.

Yes, exactly. I don't know what the polls will say on June 2, right? If anybody does, please tell me; that'd be great.

But, in my opinion, it's a fool's errand to try to predict what that will be, because anything could happen between now and June 2. COVID cases are increasing and could go even higher. They could also decrease. There could be any number of gaffes. And, two federal elections ago, we had (Prime Minister Justin Trudeau) in blackface, which was completely unexpected.

Just a couple of months ago, Russia invaded Ukraine, right? Anything could happen in two months, and to try to predict an unpredictable event, then figure out how that would affect the election, is insane. So we're sticking with what the numbers say, and this is what the numbers say right now.