Show the working.
How PolityIQ distinguishes polls, experimental averages, models, election results and rehearsal data.
The federal polling average.
The Canadian headline average uses the latest eligible national voting-intention wave per firm, with fieldwork ending within the trailing 30 days. A poll’s weight is the square root of its total sample, capped at 2,000, multiplied by a recency factor with a 14-day half-life.
Weight = √min(n, 2,000) × 2−days since fieldwork end / 14
At least three eligible firms are needed. Polls need a known sample size and explicit start and end dates, and a sufficiently complete national decided or committed voter result. Regional polls, missing dates and missing sample sizes do not enter this average. Each firm contributes one latest eligible wave; overlapping tracker releases are labelled.
The headline presentation combines the PPC and other parties for comparison across firms. Individual stored poll records preserve separately supplied party categories. Published percentages may not sum to 100 because of rounding.
This is an experimental description of voting intentions. It is not a calibrated election forecast, seat projection or probability of winning. The separate two-year archive uses its own eligible historical records; its trend is not the same series as the headline average.
The federal seat projection.
The separate Canadian federal projection applies nationally adjusted regional polling swings to Elections Canada’s official 2025 riding results. It covers 343 ridings with a 172-seat majority threshold. Regional samples, source dates, national fallback and the full formula are disclosed on the projection page.
This is an uncalibrated, deterministic model. Riding percentages are modelled support, not riding polls. There are no win probabilities, candidate effects or calibrated uncertainty intervals. Share cards retain the date, projection label and PolityIQ branding.
Approval is not voting intention.
Government approval, government satisfaction, prime minister job approval, preferred prime minister and leader favourability are separate measures. We retain the named subject, respondent base, question summary and published categories. We do not combine these questions into a single approval score.
Total survey sample and question or regional sample are distinct. Unknown values remain unknown. A national sample is never presented as the sample for one province. Question summaries paraphrase the original; the source report contains the wording.
Collect, validate, retain.
Scheduled collection researches original publishers and submits structured records. Validation checks dates, permitted sources, geography, percentages and duplicate waves. Accepted records publish from permanent storage without a website deployment. Conflicting revisions are held for review and earlier accepted records are retained.
Listed sources are a discovery registry, not proof of a working feed. Failed or incomplete checks are visible on the polling desk. Historical backfill is ongoing, so neither the registry nor the archive guarantees that every poll has been found.
Counts, calls and models.
Election desks display the status provided by their source. Leading means ahead in the available count; elected or called labels depend on the identified source. Official certification belongs to the relevant election authority. Counts can be incomplete, delayed or corrected.
US outlooks, vote simulators and Brazil’s vote-transfer tool have their own assumptions on the relevant page. Their numbers are not live results or a promise about the eventual outcome. Polling areas on Toronto’s polling map must not be read as measured ward-level results.
Rehearsal modes use synthetic data and are labelled. They do not write simulated votes into the live results store.
When a source fails.
Desks retain previously verified snapshots where available and show delay or availability messages. A retrieval time is different from a source’s own update time. Election-result pages generally refresh while open; refresh and coverage details are specific to each desk.