Forecasting the House of Commons from imperfect polls: an overview
Abstract. The Canada section forecasts the next federal election with the rule used for every country on the site: take half the average past miss off today's polls, but only where that would have improved past forecasts made without knowing the result. Canadian final polls pass that test: since 2008 they have understated the Conservatives every time, and since 2000 overstated the NDP every time. The seats come from 343 separate first-past-the-post contests, so the forecast reads seven regions and moves every riding's 2025 result with its region. This paper maps the other ten Canada papers.
Summary
This paper is the way into the Canada section of Polling Forecast. It says what the section forecasts, why Canadian polls are corrected when German ones are not, how 343 riding results are projected from national polls, and how the uncertainty is measured. The last section is a guide to the other ten Canada papers, each of which takes one part of the forecast and treats it in full.
The short version is this. At the nine general elections from 2000 to 2025, Canadian final polls missed each party by 1.2 points on average, and the misses had a direction: the Conservatives were understated at every election from 2008 on and the NDP overstated at every election from 2000. Taking off half the average past miss, with each election predicted from the others, came closer at seven of the nine and cut the average miss from 1.21 to 0.94 points, so the forecast does it. The seats come from 343 separate first-past-the-post contests, so the forecast reads seven regions from the pollsters' own regional figures and moves every riding's 2025 result with its region, by the swing rule that got the fewest ridings wrong when past elections were replayed. 2,000 simulated elections, with errors sized from the past at the national, regional and riding levels, give the ranges and chances.
What has changed since this paper was written
Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.
- Our figures beside the polls (26 October). Our projection differs from the poll average in Canada (the correction is on), so both are shown on the hub, the leader cards and the polls page.
- When the polls missed (2 November): The Wildrose lead that vanished overnight: Alberta's 2012 polling miss and The young voters who stayed home: British Columbia's 2013 polling miss.
- The map (9 November). The map can show every federal election since 1988 on the riding boundaries of its day, and any census figure; each figure has its own page (census maps).
What the Canada section forecasts
The forecast is for the next general election to the House of Commons. The Canada Elections Act (s.56.1) fixes it for the third Monday of October in the fourth calendar year after the last polling day: 15 October 2029. The Governor General can still dissolve Parliament earlier on the Prime Minister's advice, as happened before the 2021 and 2025 elections. The House has 343 seats under the 2023 Representation Order, each elected by first past the post: the candidate with the most votes wins (s.313). A party with 172 seats has a majority; one with 12 is recognised as a party in the House, which brings research funding and speaking time.
The Liberals under Mark Carney won 169 seats at the general election of 28 April 2025, three short of a majority, against 144 for the Conservatives, 22 for the Bloc Québécois, 7 for the NDP and 1 for the Greens. Since then five members have crossed the floor to the Liberals and seven by-elections have been held; on 15 October 2026 the House has 173 Liberals, 137 Conservatives, 21 Bloc, 5 NDP and 1 Green, with six seats vacant. The forecast starts from the 2025 result, not from today's House, because a general election contests every seat.
The section answers five questions, each on its own page:
- The seats each party would win if the election were held today (the Canada home page, the ridings page and the provinces page).
- The chance of a majority, and of each party reaching official party status.
- The winner of each riding, and how sure that is.
- How different the result could be on the fixed polling day, three years away (the election-day page).
- What the House would look like if the polls were wrong the way they were at a past election (the what-if page).
Today's forecast
| Party | Poll average | Our projection | Seats | 80% range | 12 or more seats | 2025 seats |
|---|---|---|---|---|---|---|
| Liberal | 46.5 | 47.4 | 241 | 211–256 | over 99% | 169 |
| Conservative | 29.6 | 30.3 | 67 | 58–97 | over 99% | 144 |
| Bloc Québécois | 6.0 | 6.1 | 21 | 11–27 | 89% | 22 |
| NDP | 12.5 | 11.9 | 12 | 8–15 | 52% | 7 |
| Green | 2.6 | 2.0 | 2 | 0–2 | 0% | 1 |
| People's Party | 1.6 | 1.1 | 0 | 0–0 | 0% | 0 |
Table: The Canada forecast for an election held today, 15 October 2026. Poll average from the latest poll of each of four pollsters (Abacus Data, Liaison Strategies, Nanos Research and EKOS Research, fieldwork 2 to 13 October); others 1.2%.
On these polls the Liberals would win a majority of 139. The simulation gives a Liberal majority in every one of its 2,000 runs, shown as "over 99%". On the fixed polling day, with three years of possible drift, the same polls give a Liberal majority 67% of the time and a Conservative majority 18%.
One rule for every country
The British forecast starts from a finding: at the general elections from 2015 to 2024, England's final polls missed in steady directions, and a correction built from those misses would have brought the forecast closer to the result (Forecasting a British general election when the polls are wrong: an overview, Correcting the raw answers: the method behind our UK projection). The rule taken from that work is the same in every country.
For each party \(p\) and past election \(t\), let \(e_{p,t}\) be the average of the final polls minus the result, in points; a positive value means the polls overstated the party. The correction takes off half of the average past miss:
where \(y_p\) is today's poll average, followed by a rescale so the shares keep the same total. A party is corrected only if it has been measured at two or more elections. The rule is kept only if it passes a test: each past election is predicted with the correction learned from the other elections alone (leave-one-out cross-validation; Stone, 1974), and the correction stays only if its average miss per party beats the uncorrected polls by more than 0.01 points.
Only elections from 2000 on take part. The model's cutoff year is a site-wide setting, 2000 by default: earlier elections (for Canada, 1988, 1993 and 1997) are kept and shown, marked "not used by the model", but shape nothing.
Why Canadian polls are corrected
The evidence is the record of final polls at the nine elections, 82 polls in all (How far Canadian polls have missed, 1988–2025):
| Election | Liberal | Conservative | NDP | Bloc | Green | PPC |
|---|---|---|---|---|---|---|
| 2000 | +0.3 | 0.0 | +0.7 | −0.8 | – | – |
| 2004 | −3.5 | +2.3 | +2.1 | −0.3 | +0.7 | – |
| 2006 | −2.2 | +2.0 | +1.3 | +1.0 | +0.8 | – |
| 2008 | +0.3 | −2.4 | +1.7 | −0.2 | +1.8 | – |
| 2011 | +1.0 | −1.8 | 0.0 | +0.2 | +1.2 | – |
| 2015 | −1.4 | −0.8 | +1.6 | +0.2 | +1.3 | – |
| 2019 | −1.0 | −2.4 | +2.7 | −0.3 | +0.9 | +0.9 |
| 2021 | −1.2 | −2.6 | +1.3 | −0.7 | +1.2 | +1.5 |
| 2025 | −1.2 | −2.4 | +2.2 | −0.2 | +0.7 | +0.8 |
| Mean | −1.0 | −0.9 | +1.5 | −0.1 | +1.1 | +1.1 |
Table: Average final poll minus result, points. Before 2004 the Conservative line is the Progressive Conservatives and the Alliance together.
Unlike in Germany, where the sign of the miss flips from one election to the next for every party (Forecasting the Bundestag: an overview), the Canadian misses repeat. The NDP was overstated at every election, the Greens at all eight where measured, the People's Party at all three; the Conservatives were understated at all six from 2008, and at the last three every single final poll had them too low. The test shows that a correction learned from such misses helps:
| Election | Polls' average miss | Corrected |
|---|---|---|
| 2000 | 0.44 | 0.53 |
| 2004 | 1.78 | 1.58 |
| 2006 | 1.45 | 1.27 |
| 2008 | 1.30 | 1.09 |
| 2011 | 0.85 | 0.94 |
| 2015 | 1.04 | 0.60 |
| 2019 | 1.35 | 0.83 |
| 2021 | 1.45 | 0.96 |
| 2025 | 1.25 | 0.66 |
| Mean | 1.21 | 0.94 |
Table: Leave-one-out test, average miss per party, points.
Closer at seven of nine elections, and closest at the four most recent. Today the correction adds about half a point each to the Liberals and the Conservatives and takes about three quarters of a point from the NDP and half a point each from the Greens and the PPC (How we read Canadian polls: the average, the correction and its test). It is the same rule reaching a different verdict from Germany's, not a different rule.
What decides the seats
Where the votes are. A party wins nothing for second places. At the 2019 and 2021 elections the Conservatives won the most votes and the Liberals the most seats, by 36 and 41, because Conservative votes piled up in Alberta and Saskatchewan. National shares alone cannot forecast the House.
The regions. Canadian elections swing by region: the NDP's sweep of Quebec in 2011, the Liberals' clean sweep of Atlantic Canada in 2015, the NDP's losses in British Columbia and Ontario in 2025. When past elections were replayed with the true change, moving ridings by their region's change instead of the national one cut the ridings given to the wrong party from 373 to 268 (Testing the seat engine: the plurality check, 2025 and the swing replays).
Quebec and the Bloc. The Bloc Québécois stands only in Quebec's 78 ridings. Its national share of about 6% is a Quebec share of about 27%, so a point nationally is about four and a half points in Quebec. Quebec is the one four-party region, and fifteen of the 48 ridings that change hands under at least one past polling miss are there.
The 172 and 12 lines. 172 seats make a majority. Twelve make a recognised party, which the NDP lost in 2025 with seven and is projected to win back today with exactly twelve.
The forecast, step by step
The pipeline has eight steps. Each is stated briefly here and developed in its own paper.
Step 1: the average. Each pollster's latest national poll within 21 days of the newest counts once, rescaled to 100:
Canadian releases give decided and leaning voters; the undecided are already left out. Nanos and Liaison publish weekly rolling samples, and one vote per pollster keeps them from outweighing monthly pollsters (How we read Canadian polls: the average, the correction and its test).
Step 2: the correction. Equation (1), switched on because it passes the test.
Step 3: regions. For each of seven regions, two readings: national swing from the 2025 regional result, and the pollsters' own regional figures, each measured against the same release's national figure and added to our national share. The two are blended with weight \(n/(n+2)\) on the polls for \(n\) pollsters, and the regions are scaled to add back up to the national shares (From national polls to 343 ridings: regions, Quebec and the swing model).
Step 4: ridings. Each riding's 2025 result moves by its region's change in points (uniform swing), and the candidate with most votes wins. A party with no candidate in a riding in 2025 stays at zero there, which keeps the Bloc in Quebec.
Step 5: simulation. 2,000 simulated elections, each with a national error per party the size of its past final-poll miss, a regional error fitted on past departures from national swing, and riding noise fitted so that past replays would have missed as many ridings as they did (Simulating Canadian elections: where the chances come from).
Step 6: polling day. Each party is pulled part of the way back towards its 2025 result (today's polls count for 43% of the gap at three years) and the ranges widen with the drift expected over the time left (Polling day: how far Canadian polls move before an election).
Step 7: lenses. Today's polls are replayed with each of the nine past misses (If we're wrong: today's polls with each past miss replayed).
Step 8: pollster ratings. Every pollster is rated on its final polls and its openness, for readers; the ratings do not weight the average (Rating Canadian pollsters: accuracy, the lead and openness).
The uncertainty
The simulation gives three kinds of answer. The first is the seat range: the band within which a party's seats fall in the middle 80% of runs. The second is the chance of each outcome (a majority or minority for either main party) and of each party reaching twelve seats. The third is each riding's chances: on today's projection 31 ridings have a projected winner with less than a 60% chance, and summed over the country the simulation expects 42 projected winners to lose.
The three layers of error give the simulation its structure. A national error is shared by every riding, a regional error by every riding in its region, and only the riding noise is local. Ridings in the same region therefore tend to fall together, as they do at real elections.
Beside the simulation, the lenses replay each past election's whole pattern of errors on today's polls. Every lens keeps a Liberal majority today, from 228 seats (a repeat of 2011) to 266 (2006), and 48 ridings change hands under at least one.
On polling day the uncertainty is much larger. The typical miss per party of polls taken three years before polling day is about 6.6 points, against about 3.9 a month out.
What we do not claim
We do not claim that Canadian polls will miss the next election as they missed the last six. We claim only that the misses have been steady enough that taking off half the average would have helped at seven of nine elections, each predicted without its own result.
We do not claim riding-level precision. Uniform swing from 2025 ignores candidates, incumbency and floor-crossings, and moves every riding in a region by the same amount. Even given the true regional change, it got about one riding in nine wrong at the last replay.
We do not claim that the regional readings are precise. In most regions they rest on two or three pollsters, in Atlantic Canada on one.
We do not claim that the "today" forecast predicts polling day. It answers a hypothetical question; the election-day forecast answers the real one, with much wider ranges, and the date itself may change.
Finally, nine elections are few, and every final poll in the record comes from Wikipedia's tables. What we don't know yet: limitations of the Canada forecast collects these points in one place.
A guide to the papers
The Canada library has eleven papers. This one is the introduction.
How far Canadian polls have missed, 1988–2025 sets out the evidence: the 92 final polls at twelve elections since 1988, how they were chosen, the error party by party and on the gap between the leaders, how much of it is shared by the industry, and each pollster's record.
How we read Canadian polls: the average, the correction and its test is the core methods paper: the average, the correction, the leave-one-out test and how sensitive it is to λ.
From national polls to 343 ridings: regions, Quebec and the swing model covers the regional step and the riding swing in detail: the two readings of each region, where the regional figures come from, the Bloc and Quebec, a worked riding, seats by province, and the Alberta and Manitoba provincial forecasts.
Testing the seat engine: the plurality check, 2025 and the swing replays reports the tests of the seat engine: the plurality check at twelve elections, the reproduction of 2025 and the swing replays that chose uniform swing and set the noise.
Simulating Canadian elections: where the chances come from explains where the chances come from: the three layers of error, how each is sized, and what the simulation leaves out.
Polling day: how far Canadian polls move before an election measures how far Canadian polls move before polling day and describes the pull-back and the widening ranges.
If we're wrong: today's polls with each past miss replayed describes today's polls with each past miss replayed and the ridings at risk.
Rating Canadian pollsters: accuracy, the lead and openness describes the pollster ratings and how releases are read automatically.
The Canada data: sources, grades and checks documents every source, its grade, its licence and the checks it passes.
What we don't know yet: limitations of the Canada forecast collects the known weaknesses and what would address them.
References
Canada Elections Act, S.C. 2000, c. 9, ss. 56.1, 313.
Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B, 36(2), 111–147.
Figures in this paper are as of 15 October 2026; the pages update with each poll.
How to cite
Polling Forecast (2026). Forecasting the House of Commons from imperfect polls: an overview. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/canada-overview. Live figures change as new polls arrive; the date above is when the text was last revised.



