Reading a poll: what a percentage really allows you to conclude
Sample, timing, margin of error and seats: a method for reading Canadian political polls.
A poll can inform a debate without predicting its outcome. The headline figure summarizes responses from particular people, collected on particular dates using a specific method. Our approach is to recover those conditions before interpreting a lead or a change.
Start with the methodology
Statistics Canada recommends examining the target population, sample selection and possible errors. On a new policy, responses from all adults, registered voters or people certain to vote do not answer exactly the same question. An open survey of website visitors calls for particular caution. [1]
Identify the method behind the margin
Federal standards for telephone surveys distinguish probability and non-probability samples. A conventional sampling margin of error cannot be applied indiscriminately to every panel. These standards govern the government research covered by them; they serve here as a methodological reference, not a legal rule imposed on every political poll. [2]
Compare like with like
Our reading checklist fits on one line: who commissioned the poll, who conducted it, what dates does it cover and what exact question was asked? Responses collected before an event do not measure reactions to it. To follow a trend, start with surveys whose methods and questions are similar.
Separate percentage points from percentage growth
Fictional example: moving from 30% to 33% is a gain of three percentage points. The relative increase is 10%, because 3 ÷ 30 = 0.10. Both calculations can be correct while creating different impressions. Neither calculation alone establishes that the observed change exceeds the poll’s uncertainty.
National votes are not seats
Elections Canada describes a system in which the candidate receiving the most votes in a riding wins its seat. The fictional table below illustrates the consequence: A wins two of three ridings with 40% of the total vote. This is neither a real poll nor an election projection. [3]
| Riding | Votes for A | Votes for B | Seat won |
|---|---|---|---|
| 1 | 60 | 40 | A |
| 2 | 60 | 40 | A |
| 3 | 0 | 100 | B |
| Total | 120 (40%) | 180 (60%) | A: 2 · B: 1 |
What else to ask
Before sharing a headline announcing a shift, look for the method, the size of the relevant subgroup and comparable earlier results. If information is missing, say so rather than inventing a margin of error. A poll is useful when it makes clear what the data show, along with their limits.
Sources and method
The documents below support the explanations. Neutral News created the calculations and fictional scenarios for teaching purposes. They are not forecasts.
- [1] Statistics Canada · questions about surveysAccessed September 13, 2026
- [2] Public Services and Procurement Canada · survey standardsAccessed September 13, 2026
- [3] Elections Canada · political systemAccessed September 13, 2026
Report an error or provide additional information ↗ · All explainers