Pollster ratings

Kiwi Polls grades each New Zealand pollster on its track record against election results. We don't run polls ourselves.

These grades are Kiwi Polls' own assessment of each pollster's record and are separate from the forecast model.

Scores

Kiwi Polls pollster grades and scores out of 100, overall and at the 2023 election
PollsterOverall2023 electionWhy
Talbot MillsA, score 86.5 out of 100A plus, score 98.0 out of 100Its one final poll at an election, in 2023, was the most accurate of the five that year (average party error 0.44 points). However, it meets only 1 of 8 public disclosure checks, and a one-election record is pulled towards the average.
VerianA minus, score 82.6 out of 100A, score 89.8 out of 100Rated over four elections (2014–2023, including as Colmar Brunton), it beat the other final polls in three of them. It fell short in 2020, when every final poll overstated the right bloc, and it meets all 8 disclosure checks.
CuriaA minus, score 82.3 out of 100A, score 86.9 out of 100Its 2023 final poll had an average party error of 0.99 points and the second-smallest bloc gap error (2.3 points). It meets 6 of 8 disclosure checks; with one election on record, its score is pulled towards the average.
Reid ResearchB minus, score 66.0 out of 100C plus, score 63.5 out of 100Over four elections, its final poll did worse than the other final polls in 2014, 2020 and 2023 and better in 2017. In 2023 it understated the right bloc's margin by 6.9 points. It meets 6 of 8 disclosure checks.
Roy MorganC, score 58.3 out of 100F, score 33.7 out of 100Its 2023 final poll had the largest error of the five (average party error 2.34 points, bloc gap error 7.1 points), and it was also well off in 2017. It beat the other final polls in 2014 and 2020 and meets 5 of 8 disclosure checks.
FreshwaterN/A, not ratedN/A, not ratedNot yet rated, as it has no election record. Its 2023 polling ended in August, too early to count as a final poll.
  • AA+ and A
  • A−A−
  • BB grades
  • CC grades
  • FD and F
  • N/Anot rated

How we rate pollsters

We rate each polling firm on how close its final poll came to the result at every general election since 2014: its last poll with fieldwork ending in the 30 days before election day and published before election day, graded against the Electoral Commission's official party vote.

Each final poll is checked in two ways:

  • Party error: the average error across every party that won at least 1% of the party vote. A party the poll didn't report counts as zero.
  • Bloc gap error: the error in the gap between National and ACT on one side and Labour, the Greens and Te Pāti Māori on the other (NZ First is left out). We divide this error by five, the number of bloc parties, so it is on a per-party scale, then add it to the party error.

Some elections are harder to poll than others, so we also compare each poll with the other final polls at the same election. Accuracy weighs a firm's own error and its result against that group equally. 2023 counts 1.5 times as much as earlier elections.

One result can be luck, so short records are pulled towards the average; the fewer elections, the bigger the pull.

The overall score is 90% accuracy and 10% transparency. Transparency is eight public disclosure checks: method, sample size, margin of error, weighting, fieldwork dates, full results, survey mode, and membership of the Research Association of New Zealand.

We don't grade lean: a firm that tends to show one side higher than other polls is not marked down for that alone. A firm with no final poll at a general election is not rated.

Grades: A+ 90 and above, A 85, A− 80, B+ 75, B 70, B− 65, C+ 60, C 55, C− 50, D 40, F below 40.

The 2023 column grades each firm's 2023 final poll on its own, with no pull towards the average.

These ratings are Kiwi Polls' own assessment and are updated after each election.

Contesting a grade

If a polling firm wants to contest its grade, email us at quasimodo@kiwipolls.co.nz with the details and we'll review it.