On Friday I published my weighted poll of polls, so it's only fair for me to face up with how it actually went. This is based on the election night vote, and my previous calculations are based on the final vote (including special votes). It's unlikely that special votes will change the result significantly, but history suggests that National and NZ First will end up down slightly on the election night results, and Labour and the Greens up slightly. I'll come to that in a minute.
Firstly, I'll give you what I wasn't allowed to give you on Friday - the final results for our own UMR poll. Of course people are going to say that we've played with the results to fit the election night numbers, but these exact numbers were shared with our client on Friday.
- National: 47.4%
- Labour: 26.3%
- Green: 11.9%
- NZ First: 6.8%
- Conservative: 3.9%
- Maori: 1.7%
- Internet Mana: 1.0%
- ACT: 0.8%
- United Future: 0.1%
Ok, back to the weighted poll of polls. That was based on the finding that, in previous elections, National and the Greens had consistently done worse on election day than in the previous polls, while New Zealand First had done better. That led to a conclusion that National would get 45%, Labour 26%, Greens 11% and NZF 9%. I thought that the Conservatives would get between 4% and 4.9%, that the Maori Party might well win a list seat, and that Internet Mana would be entitled to one or two extra seats if they held Te Tai Tokerau.
Before I draw conclusions on that, I'll first consider what the final count is likely to be after special votes. The current totals are based on 2,112,522 votes counted so far, with 254,630 special votes still to be counted. At the last three elections, 92%-93% of those special votes have been declared valid for the party vote, so it's likely that the total number of votes in the election will increase by around 235,000 - i.e. around 10% of the vote remains.
I'm not sure we should place too much weight on the suggestion that the high number of advance votes would affect special votes. For a start, the number of special votes cast in 2014 is similar to the number cast in 2011 (263,469). I think a lot of the advance votes were cast by people who were pretty busy on Saturday, but would have found the time to vote on the day if they had to. Most special votes are cast by those who can't be at one of their electorate's polling places on the day (perhaps because they're overseas) or because they've only just registered to vote. I'm not certain, but I imagine that votes cast in advance outside your own electorate still count as special votes (I could be wrong on that).
I think the specials will be at least slightly beneficial to Labour and especially the Greens as they have been in the past, because factors that make people cast special votes are the same as they were. I'm not talking about a dramatic change, just a slight tweak to the results.
People who are overseas are more likely to support the parties they knew about before they went overseas, which advantages long standing parties (National, Labour, Greens, NZ First) over new ones (Conservatives & Internet Mana). New Zealanders living overseas also logically will be more likely to vote if they are planning to return some time in the next few years, which should mean that those who are on OEs may be more likely to cast special votes than those who have (say) retired to the Gold Coast. That should tend to benefit parties that do better amongst young people (Labour & the Greens do relatively well amongst under 30s, National doesn't do as well as they do in some other age groups, NZ First and the Conservatives don't do well at all).
If you live in New Zealand, then the smaller your electorate the more you're likely to be outside it on election day, so there tend to be more special votes in geographically small urban seats than in big rural ones. Labour and the Greens tend to do better in urban areas than they do in rural areas. In 2011, for example:
- The top electorates for special votes were Wellington Central, Auckland Central, Epsom, Rongotai and Mt Albert
- The lowest general electorates were Whanganui, Rangitata, East Coast, Invercargill and Wairarapa.
- On election night 2014, for example, the party vote in Wellington Central was National 38%, Labour 24%, Greens 28%. In Whanganui it was National 48%, Labour 25%, Greens 7%.
If we assume that special votes break something like National 46%, Labour 27%, Greens 15%, NZ First 6% (which I think is reasonable on past track records), then after special votes I'd pick:
- National just below 48%.
- Labour still on 25%, but perhaps rounding down to it rather than rounding up
- Greens rounding up to 11%
- NZ First either rounding down to 8% or rounding up to 9% (i.e. I think they'll be something like 8.3% to 8.6%)
Although that's pretty speculative, it's enough to comment on the applicability of the weighted poll of polls and the logic behind it.
- The traditional observation that the Greens will fall on election day and NZ First will rise has happened again.
- The conclusions on the Conservatives, Maori Party and Internet Mana were right - I didn't have information on Te Tai Tokerau, but if they had won that seat they would have been entitled to one list MP.
- Labour was a bit too high.
The elephant in the room, however is National. I've been arguing for months that National tend to do worse on Election Day than they do in the preceding polls, and that was clearly supported by previous data. It hasn't happened, however, in 2014. My conclusion on National proved to be wrong, and we do need to think about why.
- It seems likely that other pollsters have tweaked their methodologies reflecting the traditional overestimation of National's vote, and perhaps over-reached themselves. The art of ensuring a representative sample of actual voters is far more complex than many people suppose, and pollsters are always looking at what they can do to improve.
- My earlier conclusions didn't necessarily mean that the polls were wrong on National - as I said at the time, it was also valid to conclude that some voters switched from supporting National a few days out to supporting other parties on the day. Perhaps this time those people didn't switch. The visceral reaction against Kim Dotcom in the last few days may well have shored up National's vote in that way, by persuading those voters who might have switched at the last minute in previous elections not to do so.
Like I say, the previous model was right on the Greens and New Zealand First, but wrong on National. That doesn't mean that we should necessarily dismiss the model in future elections: it's still the first time (going back to 1999) that National hasn't fallen on Election Day. As others have pointed out, one of the challenges we have in New Zealand when trying to emulate Nate Silver's sterling work in the US is that we don't have anywhere the same number of polls. Putting this data into a new model should mean that we're able to come up with a better model last time.
One last point - this election shows that rumours of the death of traditional polling have been exaggerated (to paraphrase Mark Twain). Here's the truth for our poll plus the five public polls:
- All of them were within the margin of error for both National and Labour
- Four were within the margin of error for the Greens, and those that were out were too high
- All were within the margin of error of what I think will be the final vote for NZ First.
- All were within the margin of error for the Conservatives.
- The parties that some have suggested would be under-represented in landline-only polls (the Greens and Internet-Mana) actually did worse on Election Day than they did in the landline-only polls.