Methodology · pollster report card
Which pollsters can you trust?
After every election, we compare each firm's final poll with the result. This report card is updated after every vote, here and elsewhere.
The ranking across all elections
Each firm is compared with the other firms in the same election, then averaged. To be ranked, a firm must have been measured over at least two elections: a single good or bad result can be luck.
| Firm | Elections | Average miss | Versus the others | |
|---|---|---|---|---|
| 1 | Forum Research | Quebec 2018, Quebec 2022, Quebec 2026 | 1.91 | −0.14 better |
| 2 | Léger | Quebec 2018, Quebec 2022, Quebec 2026 | 1.92 | −0.14 better |
| 3 | Research Co. | Quebec 2018, Quebec 2022, Quebec 2026 | 1.94 | −0.13 better |
| 4 | Mainstreet Research | Quebec 2018, Quebec 2022, Quebec 2026 | 2.58 | +0.25 worse |
Measured in a single election
These firms will be ranked after their second election.
| Pallas Data | Quebec 2026 | 0.92 |
|---|---|---|
| Angus Reid | Quebec 2018 | 2.60 |
| Ipsos | Quebec 2018 | 3.15 |
| Liaison Strategies | Quebec 2026 | 1.80 |
| EKOS | Quebec 2022 | 2.52 |
| Synopsis | Quebec 2026 | 2.42 |
Quebec 2026 · Oct. 5
The result and the average of the final polls
ResultAverage of the final polls
Together, the pollsters overestimated the CAQ by 2.4 points and underestimated Québec solidaire by 1.5 points.
The ranking · Quebec 2026
The rank combines five measures, explained below. The chart shows the final poll's error on each party: left of the line, the party was underestimated; right, overestimated.
PQQLPPCQQSCAQ
Léger
Sept. 30–Oct. 1 · 1,000 resp. · Online- Average miss
- 0.90
- Key race
- +0.2
- Error ÷ chance
- 0.76
- Consistency 3 wk
- −0.21
- Party order
- 5/5
- Days before vote
- 4
Accurate across the board, right on the PQ–Liberal race, and consistent over its three late-campaign waves. Its error stays below what sampling chance alone explains.
Forum Research
Oct. 3–4 · 1,227 resp. · Automated phone- Average miss
- 0.81
- Key race
- −2.8
- Error ÷ chance
- 0.91
- Consistency 3 wk
- −0.63
- Party order
- 3/5
- Days before vote
- 1
The smallest average error, and the only one to see QS above 12. But it had the PQ and the Liberals almost tied, while the PQ won by 3.8 points.
Pallas Data
Oct. 3 · 1,137 resp. · Automated phone- Average miss
- 0.92
- Key race
- +0.8
- Error ÷ chance
- 0.97
- Consistency 3 wk
- −0.02
- Party order
- 5/5
- Days before vote
- 2
Its first Quebec general election, and level with the best. Liberals and Conservatives almost exact; only QS is underestimated.
Research Co.
Oct. 2–4 · 708 resp. · Online- Average miss
- 1.44
- Key race
- +2.2
- Error ÷ chance
- 1.21
- Consistency 3 wk
- 0.00
- Party order
- 5/5
- Days before vote
- 1
The CAQ is well read, but the PQ is 3 points too high and QS 2.4 points too low. The small sample explains part of the gap.
Liaison Strategies
Oct. 2–3 · 1,000 resp. · Automated phone- Average miss
- 1.80
- Key race
- +1.2
- Error ÷ chance
- 1.95
- Consistency 3 wk
- +0.24
- Party order
- 5/5
- Days before vote
- 2
PQ exact. The CAQ at 18% misses the late-campaign slide, and QS is 2.4 points too low.
Mainstreet Research
Oct. 2–3 · 2,011 resp. · Automated phone- Average miss
- 2.84
- Key race
- +0.3
- Error ÷ chance
- 4.12
- Consistency 3 wk
- +1.28
- Party order
- 3/5
- Days before vote
- 2
Good PQ–Liberal race, but the CAQ at 20.3% is 7 points too high and ahead of the Conservatives. With the largest sample of the group, an error that size is not chance.
Synopsis
Sept. 24–27 · 1,000 resp. · Online- Average miss
- 2.42
- Key race
- +4.2
- Error ÷ chance
- 2.16
- Consistency 3 wk
- +0.15
- Party order
- 5/5
- Days before vote
- 8
The only final poll taken more than a week before the vote, before the last Conservative surge and the last CAQ drop. Once that date is accounted for, Synopsis is about average.
Party by party
| Firm | Field dates | PQ | QLP | PCQ | QS | CAQ | Average miss |
|---|---|---|---|---|---|---|---|
| Léger | Sept. 30–Oct. 1 | 29.0 | 25.0 | 20.0 | 11.0 | 14.0 | 0.90 |
| Forum Research | Oct. 3–4 | 26.0 | 25.0 | 21.0 | 13.0 | 13.0 | 0.81 |
| Pallas Data | Oct. 3 | 29.0 | 24.4 | 20.4 | 10.5 | 14.6 | 0.92 |
| Research Co. | Oct. 2–4 | 31.0 | 25.0 | 20.0 | 10.0 | 13.0 | 1.44 |
| Liaison Strategies | Oct. 2–3 | 28.0 | 23.0 | 20.0 | 10.0 | 18.0 | 1.80 |
| Mainstreet Research | Oct. 2–3 | 27.0 | 22.9 | 19.3 | 8.9 | 20.3 | 2.84 |
| Synopsis | Sept. 24–27 | 30.0 | 22.0 | 17.0 | 13.0 | 17.0 | 2.42 |
| Result | Oct. 5 | 28.0 | 24.2 | 20.7 | 12.4 | 13.3 |
Voting intentions in each firm's final poll, in %. Highlighted: more than 2 points from the result.
Not ranked
Nanos (Sept. 14), Pollara (Sept. 17) and Segma Recherche (Sept. 17) did not publish a poll in the final ten days. They cannot be judged on the result without penalising them for the date. Once that distance is taken into account, none stands out from what was expected at that date.
Riding polls
Segma Recherche published local polls in the final week: Jonquière (average miss 2.9 points, about 400 respondents), Chicoutimi (average miss 3.6 points, about 400 respondents) and Dubuc (average miss 4.8 points, about 400 respondents). It had the right winner in each.
Preliminary results; updated with the official results.
How we rank
The average miss alone rewards a lucky firm and ignores the date. So we rank each firm on five measures, then average the ranks.
- Average miss. The average gap between the final poll and the result, across the main parties.
- Key race. The error on the gap between the top two parties, the race that decided the government.
- Error on smaller parties. Missing a party at 12% by 2 points is a sixth of its vote; missing a party at 28% by 2 points, a fourteenth. This measure counts the error relative to the party's size.
- Error versus chance. We divide the error by what sampling chance alone would produce. Below 1, the firm could not do better with its number of respondents. Above 2, the gap comes from somewhere else.
- Consistency over three weeks. Each poll from the final 21 days is compared with the normal error at its date: the further from the vote, the larger the expected error. A negative number means better than expected.
Polls as published by the firms, voting intentions after allocating undecided voters.