Riding polls versus projections

We are getting an unprecedented number of riding polls. Forum is releasing some every few days, Environics is polling a couple of ridings multiple times and Mainstreet is also providing its fair share of data, including four new polls in the Ottawa region this morning. You also have some from insight West and Sigma. Even better, some ridings have been polled by more than one firm. For instance, Ottawa West Nepean has been polled by Environics, Forum and Mainstreet, with relatively similar results.

Riding polls are, by nature, less accurate than national polls. Mostly because of the usual small sample size (less than 400 for most Forum polls). Although Mainstreet and Environics are both polling as many as 800-900 respondents at times. Still, by experience, riding polls can be very wrong (hey, remember Brandon-Souris?). Not sure why. One reason could be that it's hard to make sure you are polling people really in one riding. After all, phone numbers are all mixed up.

Nevertheless, with such a high number of polls, in particular in Ontario, we can get a relatively good idea of the dynamics of this election by looking at these polls. We can also see if the projections are off. After all, some people trust riding polls more than my "made up" numbers!

Let's look at the very recent polls. I'm talking of the one published in the last 2 weeks. You also have my projections using the national and provincial polls at the times (click on it for a version that almost doesn't look terrible).



The results? Well first of all, notice how close I am with my projections in average. I know some of you could think that this is expected, but it remains very satisfying. In average, my projections are around 4 points in absolute deviation. This is a level of accuracy that I'd be proud of on Oct. 19. In terms of seats projected, I have literally the same count, although two mistakes cancel out. And since I'm sometimes above and sometimes below, it averages out.

Of course, these riding polls could be way off (and some of them likely are given the sample sizes), but you'd expect the average to at least work out.

What is true also is that the Green Party seems quite high on Vancouver Island (other, older riding polls have shown the same). I'll have to make adjustments. It might not give the Greens any additional seat, but it'll increase their chances in some ridings.

Also, what we see is that my projections do not underestimate the Liberals, neither in votes nor seats. If anything, my projections seem to slightly underestimate the Conservative vote in Ontario. So again, I understand other polls aggregators/projections show the Liberals much higher (some, like the Signal of the Toronto Star, even have the Liberals leading!!), but the numbers here should convince you that they are wrong. I'm not saying the Liberals aren't increasing or that they won't be higher in 2 weeks, I'm simply saying that as of right now, the Conservatives would win the most seats. And as i was showing yesterday in the Huffington Post, the only chance for Justin Trudeau to become prime minister on Oct. 19 is to lead in Ontario by a good 8-10 points. These riding polls show us that this is more likely not the case currently. Which isn't surprising as the poll average also doesn't show that (although some recent polls have given the Liberals comfortable leads, like Abacus yesterday).

Now let's look at the Liberals gains since the start of this election. I averaged the poll numbers by months, here are the results:

Average August
29.3
30.1
33.8
6.8
Average September
33.7
31.7
26.4
5.7
Average October
32.1
37.8
23.2
6.2

Of course, one issue is that different ridings were polled. For instance, we haven't had an October riding poll in Alberta yet. So I also averaged only in ridings that were polled in September and October (by the same firm or not). Here are the results.

sept
33.5
37.2
22.8
6.4
oct
33.0
40.6
19.6
6.9

What we see is consistent with what the national polls are indicating. The Conservatives are mostly stable since September while the Liberals are rising and the NDP is falling. However, remember that the projections do already account for the Liberals' good fortune in recent weeks. So again, these riding polls show no reason to believe I'm underestimating the Liberals. I understand it might be frustrating for some to see the Liberals increasing and somehow still being stuck at less than a 1% chance of finish with more seats than the Conservatives. But I really have nothing against this party, I don,t have an agenda. It's just that my model and simulations show the Tories as favourite. This post should at least convince you that I don't have a systematic bias.