How the polls look vs provisional results

UPDATE: Talbot Mills informs me that the poll that was reported by the NZ Herald as internal polling for Labour was “wrong”. I accept them at their word, so that should not be treated as an accurate reflection of their polling. This of course raises serious questions about why media prominently reported on an unpublished internal poll, that had “wrong numbers” according to the polling company that did them.

Lots of people asking how the polls did vs the provisional results. This will change with final results but for now, here it is how it looks.

This includes seven polls reported on by media in October – 1 News Verian, Newshub Reid Research, Roy Morgan, The Guardian Essential, Taxpayers’ Union Curia, Talbot Mills (Labour) and Talbot Mills (Corporate).

• National 39.0% actual vs 35.0% poll average = National +4.0%
• Labour 26.9% vs 28.1% = Labour -1.2%
• Greens 10.8% vs 12.4% = Greens -2.2%
• ACT 9.0% vs 9.0% = ACT -0.1%
• NZ First 6.5% vs 6.8% = -0.3%
• Te Pāti Māori 2.6% vs 2.9% = TPM -0.3%
• TOP 2.1% vs 2.6% = 0.6%

So (so far) National performed 4% better than the polls and, other parties slightly worse and Greens significantly worse.

But there was some variation in the polls so let’s look at individual polls for individual parties.

National – all polls had them (so far) too low, with Verian closest and Roy Morgan furthest away.

Labour – Roy Morgan had too low, others too high with Talbot Mills (Corporate) closest.

Greens – Essential and Curia very close, and others too high.

ACT – Roy Morgan too high, Essential too low.

NZ First – Essential and Roy Morgan too high.

TPM – Curia and Talbot Mills (Labour) too high, Verian too low.

TOP – Roy Morgan and Curia too high

Now what about the average absolute error for each party and poll:

So the pollster with the smallest average absolute error (on provisional, final will change) is Talbot Mills (corporate), then Curia (for the 1% who don’t know, that is my company), Verian, Reid Research followed by Essential and Talbot Mills and Roy Morgan.

However the margin of error is different for a poll result of 40% and 2% so let’s look at whether each company’s result was within the expected margin of error.

So the Curia poll and Talbot Mills (corporate) poll was within the expected margin of error for six out of seven parties. Again may change with final results. Curia had National 3.1% too low and the MOE is 3.0% so if National drop 0.1% or more on specials that will be within the MOE. Talbot Mills (Corporate) had the Greens 2.2% too high vs an MOE of 2.1% so they may end up 7/7 also.

Verian and Reid Research got five out of seven within the MOE. Verian had Greens too high and TPM too low. Reid Research had Greens too high and National too low.

Essential got four out of seven parties within the MOE. They had National too low, Labour too high and NZ First too high.

Roy Morgan and Talbot Mills (internal Labour poll reported by media on the day before the election) only had three out of seven parties within the margin of error and four outside.

Both had National too low and Greens too high. Roy Morgan also had Act and TOP too high and Talbot Mills (Labour) had Labour and TPM too high.

Of interest the two polls that had the lowest average error – Curia and TM (corporate) were both completed before the final week, while the more recent polls were less accurate. However this may change again with specials (which tend to occur more in final week) so I’ll update this once we have final results.