If we're wrong: today's polls with each past miss replayed
Abstract. Beside the projection, the Canada pages show the House if today's polls miss exactly as they did at each general election from 2000 to 2025: each election's final-poll errors are taken off today's average and the seat engine run as usual. Every lens keeps a Liberal majority today, from 228 seats (a repeat of 2011) to 266 (2006). The post-merger misses of 2004 and 2006 favour the Liberals; those since 2008 favour the Conservatives. 48 ridings change hands under at least one lens.
Summary
Our projection corrects today's polls by half the average past miss (How we read Canadian polls: the average, the correction and its test). No election repeats the average. So beside the projection, every Canada page can show the House if today's polls turn out to be wrong in exactly the way they were at a past general election: each party's final-poll miss at that election is taken off today's poll average, the shares are rescaled, and the seat engine is run as usual. There are nine such lenses, one for each election from 2000 to 2025.
On today's polls every lens keeps a Liberal majority. The misses of 2004 and 2006, when the polls overstated the new Conservative Party, push the Liberals to 263 and 266 seats; the misses since 2008, when the polls understated the Conservatives by about two points, bring the Conservatives to between 65 and 78 seats and the Liberals to between 228 and 247. A riding is "at risk" if at least one lens gives it to a different party from our projection: 48 of 343 are, most of them Liberal–Conservative contests in Ontario, Alberta and British Columbia, and Liberal–Bloc contests in Quebec.
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.
- The lenses on the map (November). The replays are also on the map as "The polls miss again", beside "You set the vote".
The idea
The simulations (Simulating Canadian elections: where the chances come from) answer the question of how wrong the polls might be in general, by drawing errors of the typical size in every direction. The lenses answer a narrower and more concrete one: what happens if the polls are wrong in one specific way that has actually happened. Each lens moves every party at once, in a pattern that really occurred, which keeps what the simulation's independent draws leave out: at most Canadian elections the Conservative and NDP misses ran in opposite directions, and a lens moves them that way.
The arithmetic
Let \(y_p\) be today's poll average, as published, and \(e_{p,t}\) the error of the final-poll average for party \(p\) at election \(t\) (poll minus result; How far Canadian polls have missed, 1988–2025). The lens for election \(t\) is
for the parties measured at that election; a party not measured then (the Greens before 2004, the PPC before 2019, "others" always) keeps its polled share before the rescale. The lens shares are run through the full seat engine: the regions, through today's regional ratios (From national polls to 343 ridings: regions, Quebec and the swing model), and every riding by uniform swing from 2025. There is no random noise: each lens is a single House.
A worked case: the 2021 lens. In 2021 the final polls had the Liberals 1.20 points too low, the Conservatives 2.62 too low, the NDP 1.35 too high, the Bloc 0.73 too low, the Greens 1.24 too high and the PPC 1.54 too high. Taking those off today's averages gives Liberal 47.74, Conservative 32.21, NDP 11.19, Bloc 6.70, Green 1.32 and others 1.19; the PPC, polled at 1.59, would go below zero and is held at the floor of 0.1. The total is 100.45, so every share is scaled down by that factor: Liberal 47.5, Conservative 32.1, NDP 11.1, Bloc 6.7, Green 1.3, PPC 0.1. Run through the seat engine, that is Liberal 232, Conservative 76, Bloc 23, NDP 11 and Green 1.
Two choices are worth stating. The lenses start from the polls as published, not from our corrected projection, because each one is a complete alternative to the correction: it asks what the result would be if the full miss of one election recurred, where the projection takes off half the average. And the lenses use only the nine elections from 2000 on. The site-wide rule keeps 1988, 1993 and 1997 out of everything that shapes a forecast, the lenses included.
The nine lenses today
| Lens | Liberal | Conservative | NDP | Bloc | Green | PPC | Seats: Liberal | Conservative | Bloc | NDP | Green |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Polls as they stand | 46.5 | 29.6 | 12.5 | 6.0 | 2.6 | 1.6 | 238 | 68 | 22 | 13 | 2 |
| Our projection | 47.4 | 30.3 | 11.9 | 6.1 | 2.0 | 1.1 | 241 | 67 | 21 | 12 | 2 |
| Miss like 2000 | 46.3 | 29.6 | 11.9 | 6.8 | 2.6 | 1.6 | 234 | 70 | 25 | 12 | 2 |
| Miss like 2004 | 50.7 | 27.6 | 10.6 | 6.4 | 1.9 | 1.6 | 263 | 50 | 20 | 10 | 0 |
| Miss like 2006 | 50.2 | 28.4 | 11.6 | 5.1 | 1.8 | 1.6 | 266 | 55 | 11 | 11 | 0 |
| Miss like 2008 | 46.8 | 32.4 | 11.0 | 6.3 | 0.7 | 1.6 | 231 | 78 | 22 | 11 | 1 |
| Miss like 2011 | 45.8 | 31.6 | 12.6 | 5.8 | 1.4 | 1.6 | 228 | 78 | 21 | 14 | 2 |
| Miss like 2015 | 48.4 | 30.6 | 11.1 | 5.9 | 1.3 | 1.6 | 247 | 65 | 19 | 11 | 1 |
| Miss like 2019 | 47.9 | 32.2 | 9.9 | 6.3 | 1.7 | 0.7 | 235 | 74 | 22 | 11 | 1 |
| Miss like 2021 | 47.5 | 32.1 | 11.1 | 6.7 | 1.3 | 0.1 | 232 | 76 | 23 | 11 | 1 |
| Miss like 2025 | 47.6 | 32.0 | 10.3 | 6.2 | 1.9 | 0.8 | 233 | 75 | 22 | 11 | 2 |
Table: Vote shares and seats under each lens, 15 October 2026. "Others" are 1.2% throughout. 172 seats is a majority.
What the lenses show
Every lens keeps a Liberal majority. The smallest Liberal total, 228 seats under a repeat of 2011, is still 56 above the line. With a seventeen-point lead in the poll average, no single past miss comes close to closing it: the largest gap error since 2000 was 5.8 points (How far Canadian polls have missed, 1988–2025).
2004 and 2006 point the other way. At the first two elections after the merger the polls overstated the Conservatives by about two points and understated the Liberals by two to three and a half. Replayed today, those misses would give the Liberals 263 and 266 seats, more than our projection, and cut the Conservatives to 50 and 55. They are also why the Conservative correction is under half a point: averaged over 2008 to 2025 alone the Conservative miss is −2.08, and half of that would be about a point.
The recent misses favour the Conservatives. Each lens from 2008 on except 2015 adds about two points to the Conservatives against the polls as published, and they win between 74 and 78 seats under the 2008, 2011, 2019, 2021 and 2025 misses, against 67 on our projection. The 2015 lens is the exception: that year the Liberals were understated by more than the Conservatives, and the lens gives the Liberals 247.
The NDP sits near the status line under every lens. Official party status needs twelve seats. The NDP has 10 to 14 under the lenses and 12 on our projection. Only the 2011 lens, when the NDP was polled almost exactly, gives it more than 12, and the 2000 lens gives exactly 12; under the other seven it falls below. Since the NDP has been overstated at every election since 2000, a repeat of almost any recent miss would leave it short.
The Bloc moves with how Quebec's share is shifted. From 11 seats under the 2006 lens (when the Bloc was overstated by a point) to 25 under 2000 (understated by 0.8). A point on the Bloc nationally is more than four points in Quebec, and Quebec's three- and four-way ridings turn on that.
Seats at risk
A riding is at risk if at least one of the nine lenses gives it to a different party from our projection. On today's polls 48 of the 343 ridings are.
| Lenses that flip the riding | Ridings |
|---|---|
| 1 | 18 |
| 2 | 15 |
| 3 | 4 |
| 4 | 1 |
| 5 | 5 |
| 6 | 4 |
| 7 | 1 |
Table: Ridings at risk by the number of lenses under which they change hands, 15 October 2026. The other 295 ridings keep the same winner under every lens.
By region the at-risk ridings are in Ontario (17), Quebec (15), Alberta (6), British Columbia (6), Atlantic Canada (3) and Manitoba and Saskatchewan (1). By contest, 28 are Liberal–Conservative (17 projected Conservative with the Liberals second, 11 the other way round), 14 are Liberal–Bloc (10 projected Bloc, 4 Liberal), and the rest are Liberal–NDP or Liberal–Green.
The ridings page lists each one with the years of the misses that flip it. The at-risk ridings fall into a few groups, each tied to one kind of miss.
Liberal gains that the recent misses take back. Several ridings the Conservatives won in 2025 are projected Liberal by less than two and a half points, and every lens from 2008 on that adds two points or more to the Conservatives returns them: Oxford in Ontario and Edmonton Southeast, Edmonton Gateway and Edmonton West in Alberta change hands under the 2008, 2011, 2019, 2021 and 2025 lenses, and three more ridings also under 2000.
Conservative holds that 2004 and 2006 would take. Ridings projected Conservative by between half a point and six points, such as Edmonton Northwest, Fundy Royal, Calgary Skyview and Essex, go Liberal only under the post-merger misses, when the Conservatives were overstated.
Liberal–Bloc ridings in Quebec. The 2006 lens, with the Bloc overstated by a point, gives the Liberals all ten ridings projected Bloc with the Liberals second, among them Mirabel and Saint-Jean. The 2000 lens, with the Bloc understated by 0.8, gives the Bloc four ridings projected Liberal, among them Repentigny, Terrebonne and Shefford.
The NDP and the Greens. Vancouver Kingsway, projected NDP by 1.3 points, goes Liberal under seven lenses, every one in which the NDP was overstated by more than a point. The Greens' projected seat in British Columbia goes Liberal under six. In the other direction, the 2011 lens, the one election at which the NDP was polled almost exactly, gives it two Ontario ridings, Windsor West among them. A riding flipped by many lenses is close on today's projection and sensitive to the Liberal–Conservative or Liberal–Bloc balance; one flipped by a single lens usually turns on a party that only that election's miss moved. Repentigny, projected Liberal by less than a tenth of a point over the Bloc, changes hands under five of the nine.
Lenses and the correction
The correction and the lenses use the same record in two different ways. The correction takes the average of the nine misses, halves it, and applies it to the central projection, because the test shows that doing so would have improved past forecasts (How we read Canadian polls: the average, the correction and its test). Each lens takes one miss, whole, and shows its consequence. The correction is a best guess; the lenses are a set of concrete alternatives around it. A reader who believes the next election will look like 2021 can read the 2021 lens; one who thinks the post-merger pattern of 2004 and 2006 could return can read those.
The lenses are not weighted and not averaged. They are not probabilities. The probabilities come from the simulations, which draw errors of the same typical size in every direction and add regional and riding noise; the lenses carry no noise at all.
How to use them
The home page shows the lenses as dots beside the projection and offers a "pick a year" control. The what-if page lets a reader set any polling error by hand, party by party, or start from a past miss and adjust it, and shows the House and the ridings that change hands. Both run the same seat engine as the forecast (From national polls to 343 ridings: regions, Quebec and the swing model).
Three cautions apply. A lens replays a national miss, spread over the regions in proportion to today's regional pattern; a real miss may be concentrated in one region. A lens replays the miss of the final polls, a few days before polling day; three years out, the polls can move much further (Polling day: how far Canadian polls move before an election). And a lens is only as good as the final polls behind it: all of them come from Wikipedia's tables (The Canada data: sources, grades and checks).
Figures in this paper are as of 15 October 2026; the pages update with each poll.
How to cite
Polling Forecast (2026). If we're wrong: today's polls with each past miss replayed. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/canada-lenses. Live figures change as new polls arrive; the date above is when the text was last revised.



