A sketch of the coming rollercoaster
How support for parties could (emphasis on could) evolve between final opinion polls and the special vote.
Hello again
As set out in yesterday’s final polling update, big-picture the election doesn’t look especially in doubt. However:
The analysis based on past polling error could be wrong: 2023 could be a 3σ event, maybe there has been a structural break in the reliability of polls.
There is still a plausible-but-narrow path to victory for Labour even within this model.
Even if the model’s central estimate is right, the path way from final polling to final results will be bumpy.
With that in mind, let’s take a look at how results could evolve tomorrow based on how they’ve evolved in the past. We’ll take it step-by-step from polling to the final results.
Polling error
Polls can be wrong, they often are. As discussed in the last post this uncertainty is what gives Labour any path to power.
There are some patterns in polling error. In short:
Labour and NZ First lift a bit (~1.4% and ~0.3% on average)
National and the Greens drop a bit (~-1.3% and ~-1.1%)
If those patterns hold, it’s possible (although I can’t stress this enough – not guaranteed) that Labour will do a bit better than expected, especially early in the evening.
Advance votes
Trends in advance voting
Advance voting (voting before election day) has been growing in popularity since the rules around it were liberalised in 2011. The numbers so far have been down versus 2020, but modestly up versus 2017.
A note here that this excludes special votes taken ahead of election day, which will be included with special votes in the count.
This could be due to a combination of two factors:
Advance votes as a proportion of total votes are returning to trend, and on-day vote will be proportionally higher; or
Advance vote as a proportion of total votes will be in line with 2020, with a significant decrease in turnout overall.
Before we know the numbers on election day, it’s impossible to say for sure.
As a share of all votes, advance votes reached a high of 58% in 2020. But even under the somewhat more normal conditions of 2017, they had reached 38%. We’ll find out tomorrow whether this trend persists, but I’ll take a completely baseless punt and say it’ll be closer to 48%.
Partisan lean of advance voting
Over time, the partisan lean of the advance vote has reversed: ACT, National, and other conservative parties used to do better, Labour used to do worse.
As noted above, before 2011 you had to have a reason for advance voting. Possible reasons associate with age (like surgery) or economic status (overseas travel). These factors tend to correlate with conservative voting, so help explain the pre-2011 trend.
Since liberalisation, the early vote initially shifted to mirroring the on-day vote. But in 2017 and even more so in 2020, the early vote has favoured the left. Again, a combination of two things could be going on here:
The various demographics of voters who vote early could differ from voters who vote on-day in a meaningful way that correlates with partisan (eg: young vs old).
A late swing over the course of the campaign could shift voter preferences – as in someone who voted early would have changed their minds had they voted on election day.
It could be possible to tease these factors apart with further research, but I’m not aware of any.
What it could mean
At any rate, this effect could make for a dramatic few hours between 7pm and midnight Saturday. The advance vote is counted over the course of election day, and released fairly promptly after polls close. On-day votes come in over the course of the evening.
It’s difficult to say with certainty what this means in partisan terms, but if 2017 and 2020 replicate, then an initial bout of excitement for the left and disappointment for the right could slowly ebb as the night goes on.
There’s a theory that because small polling places get counted quicker, and these tend to be in more rural areas, that National do better in the early returns for on day voting, but I can’t find any real evidence for that from polling-place level data (slightly shittier chart done in a hurry for 2020 shown below).
Special votes
Trends in special votes
As shown above, special votes were a relatively static ~11-13% from 1996 to 2014. In 2017 and 2020, they have increased to around 17%. In 2020, this was in part due to on-day enrolment for special voting being legalised. You might have thought this could lead to an increase even over and above 2017 levels, but it didn’t seem to.
Partisan lean of special votes
The partisanship of special votes shows a much clearer pattern: since 1999, progressive parties generally (the Greens, Labour, Te Pāti Māori, Mana, and TOP) do significantly better overall (by ~10% combined).
In 2017 and 2020, this has favoured Labour more than it has the Greens.
In terms of the difference the specials can make, for a while the consistent pattern was the Greens gaining a seat usually at the expense of National. With the rise in special voting in 2017 and 2020, there has been a two-seat switch.
(This was done based on re-rerunning a Sainte-Laguë calculation for the ordinary and post-special vote, and may be subject to errors. I couldn’t find reliable records of the change given time constraints).
What it could mean
If results on the night are close with the left narrowly behind, this consistent trend could lead to a nervous few weeks while special votes are counted and validated.
How things could evolve
So, bringing all that noise together, a very uncertain and not guaranteed view of what could happen over the next two weeks.
When the advance vote drops
The advance vote released in a big bag after 7pm could look good for Labour vs their polling, a combination of their tendency to over-perform at least a little and the lean of the advance vote. Call it conservatively +2% up.
National conversely could look a bit worse given the reverse tendence. Again very ball park -2% down.
Same for ACT but given they’re a much smaller party, call it -1%.
The Greens and NZ First could be about even, as their two effects (polling and early vote) offset.
Te Pāti Māori maybe even, given limited swings.
As the night proceeds
Labour will drop a bit if the on-day vote is more about demographics than swing in the final two weeks. Conversely, it’s more about a swing at the end, then maybe they’ll go up. Same effect for the Greens.
Conversely, if the shift is about demographics National and ACT will go up. If it’s about swing they might go down. If it’s about both it could be flat.
NZ First will probably start improving regardless.
If that thing about small polling places is true (somewhat doubtful), then it’ll go better for National early on, then edge back a little bit as big urban booths drop.
In any event, the magnitude of this swing could be larger than in 2020, as on-day votes should be a larger % of the total vote.
The longest two weeks of the year
Individual candidates in close races (more on that later this evening) will have an especially nervous wait, but the shape of Parliament overall should be relatively settled. Odds on National (maybe ACT or NZF) lose 1-2 seats combined, maybe 3 in a screaming fit but don’t count on it. Who among the left picks them up will depend.
To sum up
This is all largely a structured reckon to temper expectations when everything goes live.
Just because Labour might look like they have the horses at 7pm does not mean it’ll end that way. Similarly even if National have them by Sunday morning, if they’re only 2 seats up (from either not needing NZ First, or in an extreme case of not having the numbers even with them) then fuck me, what a fortnight.
Anyway, remember to breath and drink plenty of water.