The Overhang @overhang-aonz.bsky.social · 13/10/2026There's a bunch of fun historical stuff in this once, including the case for why 2005 remains the most important election post-MMP. 030
The Overhang @overhang-aonz.bsky.social · 13/10/2026In your inboxes now: Part 2 of the series on modelling the election, this time looking at the electorate party vote. theoverhangaonz.substack.com/p/the-overha... 252
The Overhang @overhang-aonz.bsky.social · 12/10/2026I am a serious political analyst whose ideas demand to be taken seriously: 4173
The Overhang @overhang-aonz.bsky.social · 06/10/2026New poll benchmark: Roy Morgan (Sep 2026). Good poll for National, the Greens, and ACT. Bad poll for Labour. 000
The Overhang @overhang-aonz.bsky.social · 05/10/2026The 'benchmark' here is basically asking "what would this firm report if the consensus trend was accurate?" So it takes the prior consensus and adds the house effects for the firm concerned. 031
The Overhang @overhang-aonz.bsky.social · 05/10/2026Good poll for the Greens, bad poll for National, about par for the rest. Reinforces the trend seen in the 1News-Verian poll last week. 183
The Overhang @overhang-aonz.bsky.social · 05/10/2026This is more of an issue when looking at the candidate vote (as it makes splitting behaviour less reliable) but may have a round-about impact on the party vote. 000
The Overhang @overhang-aonz.bsky.social · 05/10/2026There have been some suggestions that voters should game their party/candidate vote split with a view to maximising “their side’s” share of seats via an overhang. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026This disenfranchisement will mean fewer special votes (that tend to favour the left). But how many is very difficult to say (and may end up bitterly contested if the results are close). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Changes to enrolment rules mean voters will not be able to enrol or update their enrolment during the early-voting period (12 days before election day) and cast a special vote as they would in past years. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026They’re still eligible to vote (and some still will) but many won’t, especially with barriers to getting an overseas postal ballot having increased. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Since 2023 around 350,000 people have left the country. Those emigrants are overwhelmingly younger than the voting population as a whole. This will have a disproportionate effect that isn’t necessarily captured in polls that weight to mirror 2023 Census demographics. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026There's at least three factors that give me more pause than usual this election: 1) Emigration 2) Enrolment rules 3) Tactical voting 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026While they can become sport-for-nerds shit, they’re still a better empirical basis for assessing electoral (and political) realities than conventional wisdom, the vibe, or the "mood on the ground”. But they can't tell us everything. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026To sum up: This model and others like it are best-efforts, good-faith attempts to quantify how likely certain elections results are given what we know from the past, about the present, and projected into the future. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026We also produce a set of 20,000 simulated election results (using a Monte Carlo simulation). But that's more use for the electorate and Parliament/coalition modelling (more on that later). 100
The Overhang @overhang-aonz.bsky.social · 05/10/20264) Projects the polls forward to election day (adding volatility error). 5) Projects an election forecast (adding consensus bias and error). As of the latest poll, that gives you a range of outcomes that looks something like this: 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026To bring all this together, our model: 1) Adjusts each poll for its firm’s house effects. 2) Weights the polls by recency and sample size. 3) Calculates a 'nowcast' consensus estimate (with sample error and drift error). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026We do factor in the direction of these misses where they’re statistically significant (the Greens and National) but the main point is the size of the remaining uncertainty, even with zero hours left on the clock. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Other parties don't show the same reliable patterns. Labour do anywhere from +4pp better to -2.5pp worse. Similarly NZ First have been -2pp under to +2.5pp over. (Combined "Other" micro parties reliably do better, but largely because many polls exclude them). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Compared to the polling average on election day the significant trends are: - National doing on average -1.5pp worse (although that's less consistent than it used to be) - The Greens doing on average -1pp worse (fairly consistently and sometimes as much as -2pp) 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Despite averaging and adjusting for house effects, the polls: can still be wrong, frequently are, and in consistent ways. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026This volatility term is correlated between parties. If National are rising: Labour (and/or ACT) are probably falling. The same goes for Labour and the Greens/NZ First. Interestingly though, at least over the 2017-2023 cycles the same doesn’t hold for National and NZ First. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026To factor this in, our modelling includes a ‘volatility’ term that decays towards zero as we approach the election: slowly at first (as the chances of An Event remain realistic) then rapidly as we move into early voting (and the changes of a significant break diminish). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Even in the final stretch (measured here at 41 days out - the timing of the latest 2026 poll as of writing) late swing is still possible, with the majors moving an average of ±1.4pp and the minors an average of ±1pp. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Over an election campaign (the last 100 days), the major parties move by an average of ±3pp and by as much as ±9pp in extreme cases. The minors move on average ±2pp, and up to ±6pp in extreme years. This 'volatility error' narrows as we approach the election. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026More important than slight (unintentional) biases in results for various parties, what's much more material (at this point, at least) is the mundane fact that voters can still change their minds. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Accounting for this is pretty simple: remove the house effect from a poll before you added to the smoothed average. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Other than Roy’s peculiarities other consistent trends are: - Reid Research and Talbot Mills tending to overstate Labour’s support relative to the rest. - Verian tending to overstate National. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Picking on Roy Morgan (who show the most pronounced departure from the consensus) you can clearly see the trend in better performance for the minors (especially the Greens) and worse results of the majors (especially Labour). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026For a simple illustration of this, you can compare each firms polling to the trend in all other polls: 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026This is not the same as an intentional thumb on the scale. Rather, it’s an observed historical trend that results from differences in polling methodology (collection methods, question framing, choice of demographic weights, etc). 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Over time, we can identify persistent leans in a firms polling that cannot be explained by pure chance (sample error). These leans are called "house effects". 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Taking an average has two advantages: 1) It increases your effective sample size (to around 5,800 if all six firms report at once) reducing sample error and down-weighting any rogues 2) It lets you cross-compare polls to identify persistent biases... 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026To calculate this, we weight the polls for recency (decaying from full weight on the day, down to zero weight after 60 days) and sample size (which only matters for Roy Morgan who poll at n≈870 while the others are at n≈1,000) then apply a smoothing filter to fill in the gaps. 110
The Overhang @overhang-aonz.bsky.social · 05/10/2026In any given year, some firms may be more or less accurate. But (and this is key) you can't tell in advance who that will be. What is reliably more reliable is the average of all firms, beating the average final poll 9/10 times so far under MMP. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026On average across all firms, years, and parties, the final batch of polls before an election are out by by 1.3pp per party versus the election result. Timing and some residual biases mean that's a little less accurate than pure sample error. But it still beats "well I reckon..." 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026My main point is this: the polls are not lying to hurt your feelings. Opinion polling in New Zealand elections has historically been reliable, and in the absence of other information can be trusted as a guide to voters’ behaviour. 100
The Overhang @overhang-aonz.bsky.social · 05/10/2026Out this morning via the newsletter: an introductory write up on polls, whether to trust them, and how we use them in our modelling. theoverhangaonz.substack.com/p/the-overha... 142
The Overhang @overhang-aonz.bsky.social · 02/10/2026Which given that was a "oh shit there's a poll out quick throw some assumptions at it" approach, I'm pretty happy with how the adjust model and the poll compare. 000
The Overhang @overhang-aonz.bsky.social · 02/10/2026In terms of the split assumptions in the candidate model: - 2023 splits as a baseline - Added a generic NZF candidate - Gave White the generic candidate premiums - Gave Wong Raf Manji's 2023 splits from Ilam (And then renormalised to 100). 100
The Overhang @overhang-aonz.bsky.social · 02/10/2026Similar to National's results in Hutt South, the difference is so large I'm sceptical it'll hold in practice. This would be well outside the historic norm for all three parties. 110
The Overhang @overhang-aonz.bsky.social · 02/10/2026Big difference here is the party vote for 🔵National (32 vs 23 expected, 9pp up) and the 🟢Greens (19 vs 25, 6pp down). Given the state of the nationwide polls and the party's historical relative performance in the seat that's a pretty significant difference. 100
The Overhang @overhang-aonz.bsky.social · 02/10/2026Model vs poll: Mt Albert Spinoff-Curia poll showing a much closer candidate race here, but in all honesty our candidate model here is four guesses in a trenchcoat. 110
The Overhang @overhang-aonz.bsky.social · 02/10/2026Worth noting the poll (for efficiency reasons I assume) only targeted areas with a Māori population over 25%. In 2023, voters on the Māori roll (across all Māori seats) in those areas were more likely to favour Labour and TPM than the Greens or National. 000
The Overhang @overhang-aonz.bsky.social · 02/10/2026On the party vote combined "Other" are far ahead of where we'd expect based on historic patters, but given that's a handful of voters split over 10 parties, I'm not putting too much weight on it. 100
The Overhang @overhang-aonz.bsky.social · 02/10/2026Model vs poll - Te Tai Tonga Candidate vote reasonably aligned except for 🟢Te Morenga who's 5pp down on our expectations. Given the candidate model here is 🤷"I guess Ferris will do about the same as Harawira averaged"🤷 that's not so bad. 120
The Overhang @overhang-aonz.bsky.social · 30/09/2026NZ First being well back is probably closer to the truth, and maybe suggests their lift in the party vote is not replicating in the Māori seats. May need to revisit the assumptions there. 011
The Overhang @overhang-aonz.bsky.social · 30/09/2026The "other" figure on the party vote is the strangest, but given the difficulties in polling electorates (especially Māori ones) that's not entirely surprising. 110
