How the pollsters did

I blogged on Friday the final polls by the five public telephone pollsters and the final poll by Horizon. A fuller analysis will be done once we get final results, but for now I’ll do a quick analysis of how each pollster did for each party.

Note that this is not comparing apples and apples entirely. Those pollsters whose final poll was earlier in the election period naturally do not pick up what happens in the final few days. And there are other factors at work such as sample sizes. So this is not about saying who is “best” and “worst” but just a quick look at were they broadly in the right ballpark for the various parties.

This shows the actual result, and the (absolute) difference between the final poll for that pollster and the final result. Where the difference was greater than 1.5%, I have highlighted them in red.

This is just one of several ways to analyse it. One can also total up the differences for each pollster. Also you can count how many had a result within the margin of error for that poll. I’ll comment on each poll result.

Roy Morgan

They were the pollster that got NZ First closest. They had National and Greens too high and Labour too low. They did not record results for the Conservative Party at all, but otherwise were pretty good.

Fairfax Research International

National significantly too high, but Labour pretty accurate. Undershot NZ First and did not report on Conservatives. Other Minors within range.

3 News Reid Research

Like everyone had National too high (but within margin of error) and like most had the Greens too high. All other minor parties within 1.5% except NZ First whom they had at half what they got.

One News Colmar Brunton

Overall seemed to get things closest. National 2% too high and NZ First 2.6% too low, all others less than a 1% variance.

NZ Herald Digipoll

Also did well. National too high and NZ First too low, but did have them over 5%. Slightly more variance with the minor parties but none greater than 1.5%.

Horizon Poll

Of the nine significant parties, Horizon only got two of them within 1.5% – the Maori Party and United Future parties. They were the least accurate with National (14.2% out), NZ First (4.1% out), ACT (1.7% out) and Mana (1.8% out). They also had Conservatives at close to double what they actually got.

Very amusingly, Horizon are boasting how they consider their poll to have been highly accurate. It staggers me how anyone can put out a poll which had National only 5% ahead of Labour and then could claim it was “close to forecast” when the actual result was a gap of 21%.