Latest polls
Two polls out tonight. A One News and 3 News poll. They differ significantly.
The weighted average of the last three polls is above. What they show is that the centre-right would have 58 seats, centre-left 60 seats and Maori Party the balance of power with three seats.
Now three polls in as many days have had quite different results – Roy Morgan, Colmar Brunton and Reid Research. The Roy Morgan was over a different time period, so it is less surprising that it has a different result.
The differences between Colmar Brunton and Reid Research are a bit surprising. I regard both companies as good ones, and it is worth pointing out that their results are just within the extremes of the margin of error – ie one may just be at the high end of the normal range and one at the low end. And also recall 1 in 20 polls will fall outside the normal margin of error.
But people often ask when they differ this much, how to decide which one is accurate.
Generally my advice is to average them out. This tends to be pretty reliable, and the recent US elections found the polling averages very reliable.
You can look at how each polling company did in regard to the last election. However I’d be careful about putting too much weight on that, unless a company was way off-beam. Judging the accuracy of a company off one single poll that has a 3.2% margin of error is dangerous. It may be that the final pre-election poll was accurate, but that things changed in the final few days after the poll was taken.
However for those interested the average difference between the polling company’s results and the average result for all parties was
- Colmar Brunton 0.9%
- DigiPoll 1.1%
- Reid Research 1.3%
- Research International 1.6%
- Roy Morgan 1.6%
A difference in the average variation by 0.4% is not great, in my opinion. And as I said, I’d be very careful judging off a single poll. In the US they have polls covering scores of elections, so can get a good idea there if a company systemically has one party too high or low.
Now we can’t do that in NZ, but we can look at how each company has rated the major parties over an electoral cycle. Now again be wary of this comparison because over a three year period, the polls will be done at different times. If I get the time, I might try and do a comparison of polls done within say the same week, but for now thought a quick and dirty analysis could be interesting.
Over the 2008 to 2011 period, the average for each polling company and party was:
- Colmar Brunton – National 53.6%, Labour 31.1%, Greens 7.3% = Lead for National of 15.2%
- Reid Research – National 55.0%, Labour 29.6%, Greens 8.5% = Lead for National of 16.9%
So in the last term Colmar Brunton on average had the gap 1.7% tighter than Reid Research. Now again, that is not saying company A is correct and company B is incorrect. As I said these are polls generally taken over different time periods. But it is interesting there is some difference.