The polls v the poll: performance credibility [UPDATED: +iPredict]

Uber-blogger and National's polling guru, David Farrar, has made a table of how close the pre-election polling was compared to the figures on the night. The table shows the final polls in the last week. The red indicates the forecast was off by 1.5% or more.

The most obvious observation is how far off the polling organisations were with National's vote - they kept them above 50% when they were never going to get that outright majority - and how they didn't pick up the extent of NZ First's late surge. This was largely due to the polls not taking into consideration the undecideds - a major fault that tends to exaggerate National's actual level of support. The other stand-out is how askew the Horizon polling is - due, I understand - to the methodology of using an online audience that participate as if it were a game.

I made my predictions on 31st October, just before the lists were officially announced - a month before the election - so they cannot be expected to be as accurate as the other polling done in the last week, but they still measure up quite well given the time lapse and mean a lot more than Farrar's bragging over predictions made only a day before the election. Here is what I forecast (with the difference between actual in brackets):

[UPDATE: I've got an email from iPredict on how their dodgy insider trading gambling/market did so I've added in their data (also from one month before the election) after my own and prefaced it with "i"]

Because I'm not perfect and still want to claim I got it right I've allowed a 1% margin for the big parties and a half a percent for the minor ones.

(+/- 1%)
46% National (-1.99%) i-0.89%
32% Labour (+4.87%) i+1.47%
8% Greens (-2.62%) i+0.48%

(+/- 0.5%)
5.0% NZ First (-1.81%) i-2.11%
3.0% Mana (+2.00%) i+1.00%
2.5% Act (-1.43%) i+1.93%
1.0% Maori (-0.35%) i-0.15%
1.0% United Future (+0.39%) i+0.29%

1.5% All others (approx. -1.5%) i n/a

The only ones within my own margins of tolerance to claim I got them right were Maori and United Future - much the same as that reflected in Farrar's table.

National:
Only the Roy Morgan poll was closer than me - not too bad. The specials tend not to favour National either so when the official results are released in a week that should come down a bit closer to where I am, but not much, maybe a quarter of a percent. I said I couldn't see the Nats pulling higher than 47%, so I was wrong on that, but by less than a percent.

Labour:
Well I was way off. I concluded that Labour's vote would hold up and that a collapse, which I defined as a sub-30%, would be averted. Indeed I thought that all the way through and was confident about a good Labour showing when Goff performed well in the head-to-head debates. However I acknowledged that a meltdown was always on the cards and it would end up favouring mainly the Greens and then NZ First as well as Mana. The polls consistently showed Labour weaker than I thought they were - I should have listened to the polls in this case. I think a lot of Labour voters did turn Green and even more stayed at home, perhaps believing the dire predictions and thinking it was a wasted trip to the polling station. Perhaps a severe erosion is a better description of happened than a catastrophic collapse - it wasn't as if they were ever polling that high to start with.

Greens:
With Labour eroding away the Greens rose significantly and for once their soft, youth, stoner vote held up and deserting middle class Labourites pumped them well beyond what I thought they were capable of. Doubter. Mr Bradbury actually made a call about a week or two out on this blog that they would get 10.5% - so he nailed that one. They tend to do well on specials so the magnitude of my error will grow once they are counted. The polling organisation captured their vote strength quite well (although Roy Morgan and TV3 were out further than I was).

NZ First:
Apart from Roy Morgan they were all well out - I had picked their rise well in advance. If it weren't for that infamous cuppa between Banks and Key he wouldn't have lifted above the 5 - 5.5% I was picking.

Mana:
It doesn't help being a partisan trying to pick your party's own success. 3% was always on the high side - the logistics of turning out a vote that high with a new organisation with no dough was always pushing it up hill. The pollsters however under-rated Mana significantly.

Act:
They went from troubled to tragic to implosion during the four weeks of the campaign and if I could have revisited the prediction they would have been marked well down. Out of the road-kill was a bleed to National and maybe a quarter percent to Conservative. They mostly scuttled to National, but that just re-jigs things on the right without moving the total bloc vote: on my prediction Nat+Act=48.5%, on the night they got 49%.

Conservative:
Surprised everyone. I thought they could get to 1%, but 2.76% is amazing for a party that no-one knows about, except they are conservative. At least for most people; but the Christian vote knows who to follow and that good showing is mainly due to them.

Maori Party & United Future:
We all got this about right, but I thought Maori Party would dip just under 1%.

Also see Morgan Godfery's data on polling of the Maori seats to see how wild some of those figures are - albeit most were taken in September. I note that Mana only bet the Maori Party in the party vote in a handful of general seats and only one Maori seat, Hone's.

The low turnout was the decider really. The elderly keep turning out reliably while the young, the flaky youth flitter and didn't make it on the day, this leaves the oldies with a disproportionate inflation of their impact and so Winston and the Conservatives and National all did well at the expense of the others. In future things look bad for the left because of these trends. There will be more elderly as a proportion of the population and the turnout continues to drop, so more conservative elements will dominate if things don't change.