Welcome to 338Canada

About 338's Methodology

[Updated September 20, 2020]


The Basics


The 338Canada model uses a mostly proportional swing with regional adjustments. For instance, if a party goes from 30% to 33% in the poll aggregate, an increase of 10%, then its score goes up by 10% in every district - if said party is at 15% in district X, then its score goes up to 16.5% (an increase of 1.5 point).

Also, the electoral history of said regions and districts is also taken into consideration - by giving probabilistic floors and ceilings for each political party.

Demographic data


But mostly proportional does not mean exclusively proportional. The 338Canada model also uses demographic data: Careful considerations is given to demographics of each district, such as median and average household income, age distribution, language most spoken at home, etc. This data is used to make statistical correlations of voting intention swings between regions and districts.

Why use demographic data? Because in some cases, it can help control some of the swing. For instance, in the 2014 Quebec election, Québec Solidaire (QS) received 51% of the vote in the Motreal riding of Gouin, but QS only received 7.5% of the vote provincially. Four years later, during the 2018 election campaign, QS was polling at around 15% - double the support it had won in 2014 - so obviously, QS wasn't going to win 2 × 51% = 102% of the vote in Gouin. In cases like this one, proportional swing needs demographic adjustments. [In the end, QS received 16% of the vote provincially and 59% of the vote in Gouin - meaning the QS vote had already reached its saturation point in the riding.]

What data is used by the model?

  • Languages most spoken at home (very useful especially, but not exclusively, in Quebec);
  • Age density curves (proportions of 18-25 years olds, 26-35...etc.);
  • Median and mean household income;
  • Population density, which helps build an "urbanity index";
  • Education levels;
  • Riding countries of birth and immigration levels;
  • Classes of workers and employment statistics;
All of the above mentioned data are made available in the Canadian census and can be found on the Statcan website.


How are polls weighted in the model?


First and foremost, polls are weighted according to their sample size and field date. The weight of a poll (w) is proportional to the square root of its sample size (n) and diminishes as time (t) passes by. Also, a polling firm rating ( f ) is attributed to each poll - this variable is an editorial decision by 338Canada (I know it can/will be a source of criticism, but I stand by it - and it is a variable, not a constant.) For more information of the 338 Ratings of Canadian Pollsters, visit this page. 

Here is the basic equation:

The gamma exponent is usually ½ (hence, the square root of the sample size), although it could be changed in cases like during the American election where Survey Monkey polls had sample sizes over 20,000 respondents (and still their polls were no better than most). The time variable t is calculated from the middle field date of a poll. The beta exponent of the time changes whether we are in pre-campaign or campaign mode. Since voting intentions can move rather fast during an actual election campaign, the beta factor increases as we get closer to election day (meaning a poll's weight diminishes faster with time during a campaign).


Star candidates


Finally, the effect of star candidates is also estimated by calculating the approximate over-performance of past candidates in similar situations (party leader, cabinet minister, high media visibility candidates, etc.). A prime example of this is former deputy Premier of Quebec Nathalie Normandeau (in the Jean Charest era). She first ran in the Bonaventure district in Gaspésie in 1998, where the PQ had won by 10 pts in 1994 (until 2018, much of Gaspésie still leaned towards the PQ). Normandeau won a narrow 160-vote victory that year, but later became a star in the Quebec Liberal Party and was re-elected in 2003 with over 60% (!) of the vote in Bonaventure. She (and the Quebec Liberals) held on to Bonaventure until she resigned in 2011. In the next general election, the PQ won back Bonaventure. It's safe to say that many Bonaventure voters didn't vote QLP as much as they voted for Normandeau herself.

This is one of many examples of a local star candidate having a significant impact on the vote. We track those down and calculate which candidates overperform their own party in any given election. This is tricky for new candidates, as their local impact is not measurable without local polling.


What are safe, likely, leaning and toss ups?


A quick note on the districts classifications of "Safe", "Likely", "Leaning" and "Toss up": these terms are determined by the odds of winning the districts according to current data.
  • Safe: > 99.5%
  • Likely: 90% - 99%
  • Leaning: 70% - 90%
  • Toss up: < 70%
Here is the record so far on how those labels have fared


On Social Media



The 338Canada Facebook page is now available here. The rules of conduct on the page are stricter than those you'd expect from a mainstream media comment section: obviously, personal insults, disrespectful language promoting hate and just plain general trolling will result in a permanent ban. Moreover, promotional videos or links, as well as overly partisan content will be removed. Finally, those who promote conspiracy theories or try to attack my credibility will be banned (e.g.: "pwned by Tories" or "Liberal shill" [yes, those are actual lame examples]). I have no time to deal with such people.

You may also follow me on Twitter here. I post in both French and English. Again, those who are disrespectful will be muted or blocked. You may disagree with my numbers and/or analysis, but there are ways to express disagreement without insults. I will gladly engage with people who respectfully challenge my findings.

Lastly, I do speak and write in English, but my mother tongue - and the only language I've used up to age 18 - is French. So please be nice to me if I say/write the occasional Frenchism/Gallicism.

I hope you enjoy the website.



Philippe J. Fournier is the creator of Qc125 and 338Canada. He teaches physics and astronomy at Cégep de Saint-Laurent in Montreal. For information or media request, please write to info@Qc125.com.


Philippe J. Fournier est le créateur de Qc125 et 338Canada. Il est professeur de physique et d'astronomie au Cégep de Saint-Laurent à Montréal. Pour toute information ou pour une demande d'entrevue médiatique, écrivez à info@Qc125.com.