Vote Intention Poll Performance: New Zealand General Election 2020
How did the pre-election polls for the 2020 Parliament perform?
On 17th October 2020, New Zealand electors cast votes for the new Parliament. The COVID-19 pandemic delayed this election.
How well did the pre-election polls perform? This article discusses their accuracy.
Three research companies conducted vote intention polls during the election campaign. The last poll finished on 15th October 2020 — two days before the election.
According to the Electoral Commission, Labour took 49% of party votes. In preliminary results, the National Party had 27% of party votes. ACT New Zealand finished third, on 8% of party votes.
To measure accuracy, I look at:
- LAB Lead: Labour minus National party vote intention shares, deducting the actual difference.
- Two-party mean absolute error: the average absolute difference between estimates and actual shares. This is for vote intention share estimates of Labour and National parties.
- Five-party mean absolute error: the calculation extends to NZ First, ACT, and Greens.
There are three companies to consider:
- Reid Research: this is a mixed-mode sample. Each quota sample is of 1,000 eligible New Zealand electors. The company contacts 700 via phones, and 300 more through an internet panel. The weights are for age, gender, and geography. The weighting targets are from from Statistics New Zealand population figures.
- Colmar Bruton: this is a telephony poll. The target population is of eligible New Zealand electors with access to a phone. The weights are for age, gender, region, ethnic identification, and phone access. The company asks an unprompted party vote intention question. The question is: “Firstly thinking about the Party Vote which is for a political party. Which political party would you vote for?”
- Roy Morgan Research: this is also a telephony poll. The sample is a “NZ-wide cross section”. It is unclear what the weights were. They asked: “If a New Zealand Election were held today, which party would receive your party vote?”
The table only shows the last poll for each company:
Of the final polls:
- All three companies underestimated the Labour share, and overestimated the National share.
- Reid Research and Colmar Bruton underestimated Labour’s lead by over seven points.
- All three companies had a five-party mean absolute error below three points. This score reflects closer estimates for NZ First, ACT and the Green Party.
Reid Research and Roy Morgan Research publish estimates to one decimal place. These samples of are around 1,000 people. For a simple random sample, the margin of sampling error for a true proportion of 50% is about three points. There are non-sampling errors too. This is false precision.
Following the pandemic, there was an increase in Labour’s vote intention share.
For the company average, the error in the Labour lead was like the 2015 UK General Election. Given all polls suggested a large Labour lead — underestimating its size — people may not see it as a ‘miss’.
The research companies should investigate the reasons for this survey error. One possible cause is the use of weighting targets of adults for samples of voters.