Scorecard
We grade our own work.
A poll that is never checked against the result is an opinion with a sample size. After every election we poll, this page shows how far off we were, in which direction, and what grade that earns under a rubric we published before the votes were counted.
Graded cycles
Results against reality
Coverage rate (shares within stated margin): — · 0 polls graded
| Election | Final poll | Our margin | Certified result | Error on margin | Named all candidates | Grade |
|---|---|---|---|---|---|---|
| Nothing to grade yet. Our first graded cycle will appear here after the next election we poll. Every final poll we release in the 21 days before that election is already committed to this table. Get the first scorecard by email | ||||||
The commitment
How the self-audit works
Five rules, fixed before the first graded election.
- 01
Every final public poll is graded. No exceptions, no selection.
Any horse-race poll we release in the final 21 days before an election is graded against the certified result. We cannot decline to grade a poll we did not like, and we cannot grade only the ones we did.
- 02
The error is signed, on the margin and on each candidate.
We publish how far off the gap between the top two candidates was, in which direction, and the signed error for every named candidate, including minor-party and independent candidates. Undecideds are not allocated to make the numbers look better.
- 03
Grades are set by a rubric published before the election.
The rubric below is locked before our first graded cycle. If we change it later, the change applies only to future cycles, and the old rubric stays on this page.
- 04
Published within 30 days of certification.
Each graded cycle gets its own entry with a short written post-mortem: what we got wrong, what we think caused it, and what, if anything, we are changing. Misses get the same prominence as hits, and the same email to subscribers.
- 05
The margin of error is itself on trial.
Across all graded polls, roughly 95 percent of our final shares should fall within the stated margin of the result. We report that coverage rate here as the cycles accumulate. If it runs well below 95 percent, our margins are too small and we will say so.
Grading rubric
What the letters mean
Error is measured on the gap between the top two candidates, in percentage points, against the certified result. The second column is the primary test; the third is a modifier.
Minor-party and independent candidates are graded separately on their own shares: within the stated margin is a pass, outside it is reported in the post-mortem. We also report the coverage rate, the share of all our final estimates that fell inside the stated margin, which should hover near 95 percent if our margins are honest.
| Grade | Absolute error on the margin | Condition |
|---|---|---|
| A | ≤ 2.0 points | Leader correct; every named candidate within the stated margin |
| B | ≤ 4.0 points | Leader correct or race within margin; at most one named candidate outside the margin |
| C | ≤ 6.0 points | Error within roughly the margin on the gap |
| D | ≤ 8.0 points | Outside the margin on the gap; leader may be wrong |
| F | > 8.0 points | Or: wrong leader by more than the margin on the gap |
Release alerts
Be there when we grade ourselves
Subscribers get the first scorecard the day it publishes, with the post-mortem, whether the grade is an A or an F.
No spam. One email per release, and you can leave any time.