Methodology
How we measure opinion
A survey is only as good as its method. Here is ours, explained without jargon — because a reader who understands how a number was produced is better placed to judge what it is worth.
Recruitment: “river” sampling
We recruit our respondents directly on the web, as they browse, rather than from a panel of paid respondents. This approach, known as “river” sampling, reaches people who do not answer surveys every week — which reduces the risk of relying on “professional respondents” whose habits end up distorting the measurements.
In return, a sample recruited this way does not fall from the sky perfectly representative. No one should claim it does. That is precisely why the next two steps exist.
Adjustment: weighting
Once the responses are collected, we adjust the sample so that it reflects Quebec’s population as described by the best public sources available:
- census data — age, sex, region, education, language;
- attitudinal and behavioural anchors — for example, past voting behaviour, whose actual result is known — which correct imbalances that demographics alone cannot see.
Each person is assigned a weight, and we monitor the effect of this weighting on precision: an adjustment that has to work too hard is the sign of a deficient sample, and we treat it as such.
Quality control
A web survey also attracts bots, rushed answers and distracted participants. Our instrument records minimal technical paradata — device type, browser language, response time — used solely to detect these cases: questionnaires completed at a speed no human could manage, answers that contradict one another, automated attempts. Responses that fail these checks are set aside before any analysis, and the number of exclusions is disclosed in the methodology report.
Why we never speak of a “margin of error”
The classic “margin of error” (the familiar “± 3%, 19 times out of 20”) rests on a theorem that requires a probability sample — one in which every citizen had a known chance of being selected. A voluntary web sample — ours, like that of most surveys published today — does not meet that condition. Attaching a probabilistic margin of error to such a sample conveys a precision the method cannot justify.
Instead, we report modelled credibility intervals: a range that expresses the uncertainty of our estimates given the sample size, the weighting applied and the model’s assumptions — and we say plainly that they are assumptions. This is deliberate transparency, aligned with the disclosure standards of the survey research profession.
The methodology report
Every survey we produce comes with its complete methodology report: field dates, recruitment mode, sample size before and after quality control, weighting variables and their sources, the effect of weighting, and the exact wording of the questions. The report is delivered to the client with the results; when a survey is made public, so is the report, in full. If any information needed to judge a number is missing, that is a defect in the report — write to us.
A method validated continuously
We put our method to the test by checking our measurements against verifiable external references — official results and recognized benchmark data — before anything is released. This validation work is permanent. We claim no track record of predictions: it will come when it exists, and it will be published as is.
Last updated: July 2026.