Special votes, enrolment changes and outcome probabilities

Full report

NZ Election 2026: Special Votes, Enrolment Changes and Outcome Probabilities

Data and sources as at 13 November 2026. Official results are due 27 November 2026.

Summary#

The left bloc has about a 7% chance of a majority once special votes are counted. Under the pre-2026 enrolment rules the same model gives about 17%. Only two of the six closest electorates are genuine toss-ups.

• Close seats: Wairarapa (~51% chance of flipping to Labour) and Wellington Bays (~66% chance of flipping to the Greens) are the only genuine toss-ups. Hamilton East, Mt Roskill and Te Tai Tokerau are each at least 93% likely to stay with the election-night leader, and Hutt South (an 889-vote margin) is effectively settled.

• Bloc outcome: Right majority ~60%, hung Parliament ~33%, left majority ~7%. Opportunity is counted in the left bloc throughout. Under more left-friendly assumptions the left's chance rises to ~23%. Three other published models put a right majority at between 34% and 70% (Part 5).

• Enrolment law: about 60,000 votes were lost (range 40,000–80,000), including an estimated 15,000–20,000 people turned away at the booth. Based on 2023 behaviour, the lost voters would have netted the left bloc roughly 9,000–10,000 votes. That is under half a seat on average. In a result this close it still moves the left's win probability by about 10 points.

• Why fewer gains this year: about 28% of the valid specials are now overseas votes, up from 13% in 2023. Overseas votes are roughly neutral between the blocs. The domestic specials that used to lean strongly left have shrunk by about half.

Background#

On preliminary results the National–ACT–NZ First coalition holds 67 of 131 seats, one more than the 66 needed. About 399,000 special votes are still to be counted, and official results are due on 27 November 2026 [1][2].

Source: preliminary count as reported in [1].

• Overhang: National's party vote earns about 35 seats, but it won 42 electorates. Te Pāti Māori's earns about 1, but it won 5. Together that makes 11 overhang seats and a 131-seat House [1][3].

• Specials: nearly 97,000 of the 399,000 outstanding specials are overseas and dictation votes [4].

• Law change: the Electoral Amendment Act 2025 closes enrolment 13 days before election day, 25 October this year. People can no longer enrol or update their details while voting [5][6].

• Published analysis: Andrew Geddis sets out three outcomes after specials: a one-seat coalition win, a 65–65 split, or a one-seat left majority [2]. Max Rashbrooke estimates that National would need to fall from 29.2% to about 28.4% for the left's path to open [3].

Part 1: The close electorates#

Six electorates finished within 900 votes on election night. Only Wairarapa and Wellington Bays are realistic flips once specials are counted. The largest margin, Hutt South, is effectively settled, and Chris Bishop has conceded it [7].

Election-night margins#

Source: [4].

Working: specials per electorate#

Valid electorate specials per seat ≈ (399,000 − 55,000 not enrolled − ~57,500 wrong-address electorate votes) ÷ 71 electorates ≈ 4,000. The model uses a range of 3,800–4,800 [1][14][16], except in Wairarapa, where it uses 5,500–7,000 because about 6,500 specials are expected there [36].

Worked example, the largest margin: to overturn Hutt South's 889 votes on about 4,300 specials, the trailing candidate would need a net margin of 889 ÷ 4,300 ≈ +21 points among special voters.

Working: how far specials have moved one seat (Hutt South, 2017–2023)#

The specials figures add up the overseas, before-polling-day and on-polling-day special rows of each year's voting-place table. The National candidate's best result on specials was +76 votes, about +1 point, against the +21 needed this year.

Working: biggest leads overturned by specials#

In 2023 four seats flipped. The largest lead overturned was 495 votes, in Tāmaki Makaurau. In Te Tai Tokerau a 487-vote Labour lead became a 517-vote loss, a swing of 1,004 in a Māori seat. In general seats, only 30-vote (Te Atātū) and 54-vote (Nelson) leads were overturned [11]. In 2020 three National election-night leads flipped to Labour [34]. In Maungakiekie, National led by 280 with about three-quarters of votes counted, and Labour won by 635 [24][26]. None of the overturned general-seat leads reported comes close to 889.

Model probabilities#

Part 4 sets out how these were calculated. Only Wairarapa, Hamilton East and Mt Roskill change the bloc arithmetic: each moves National's overhang by one seat. A Te Tai Tokerau flip would remove one Te Pāti Māori overhang seat, costing the left one seat. Wellington Bays is left-on-left and changes no seat totals.

Part 2: How many voters the enrolment change affected#

The best estimate is about 60,000 votes lost (range 40,000–80,000). Of those, an estimated 15,000–20,000 people were turned away at the booth (range 8,000–35,000). A further ~40,000–75,000 electorate votes are being thrown out because of out-of-date addresses, though those voters' party votes still count.

2023 compared with 2026#

The Electoral Commission does not count people turned away at the booth [15], so that figure is this report's estimate, worked below.

Working: votes lost to the law and people turned away#

1. New enrolments during voting in 2023: 271,082 enrolment forms were completed in the voting period, of which 60,928 were new enrolments (9,656 Māori roll, 51,272 general roll) and the rest updates [35]. That total excludes online enrolments. Scaling the 22% new share to all ~450,000 late enrolments and updates gives 61,000–100,000 new enrollers.

2. Unenrolled when voting opened: in 2023, about 206,000 people were still unenrolled at the close (5.3% of ~3.89 million eligible), so 270,000–305,000 were unenrolled when voting opened, and 23–33% of them enrolled late. In 2026 about 260,000 were unenrolled at the 25 October deadline (6.3% of ~4.05 million) [1].

3. Votes lost: the pre-deadline campaign brought the keenest people onto the roll early, so a lower 15–30% is applied to the 260,000. That gives 40,000–80,000 people who would have enrolled and voted under the old rules. The low end matches the ~40,000 enrolment shortfall against the 2023 rate [3]. The model uses a uniform range of 40,000–80,000, with a mean of 60,000.

4. Turned away at the booth: staff turned away only people who were not on the printed roll and said they were unenrolled. Anyone who believed they were enrolled cast a special vote, which is later disallowed [15]. Assumed: 50–70% of the lost voters still turned up, and 40–60% of those said they were unenrolled. Low 40,000 × 0.5 × 0.4 = 8,000; central 55,000 × 0.6 × 0.5 ≈ 17,000; high 80,000 × 0.7 × 0.6 ≈ 34,000.

5. Check due on 27 November: these assumptions imply roughly 25,000–30,000 disallowed votes, against the Commission's forecast of ~55,000 [14] and 16,633 in 2023 [30]. If the official figure lands near 55,000, both the lost-vote and turned-away estimates are too low.

6. Wrong-address electorate votes: ~73,000 expected in 2026 against ~42,000 in 2023 [16]. The model uses 40,000–75,000 as electorate-only losses.

Context: turnout#

Votes cast fell from 2,883,412 in 2023 to about 2,832,000, while the eligible population grew to about 4.05 million [1][18]. That puts turnout of eligible people at about 69.9% [3], down from about 74% in 2023. Much of that fall reflects general disengagement rather than the law. The law-lost estimate above uses only the enrolment-specific evidence.

Part 3: How late enrollers have voted#

In 2023, domestic special votes ran about 11–16 points more left than ordinary votes on the bloc margin. Overseas specials were roughly neutral between the blocs and mainly moved votes from Labour to the Greens and Opportunity. Late enrolments and updates made up as many as 450,000 of the ~525,000 domestic specials (up to 86%), so the domestic pattern is a close proxy for late enrollers.

Sample: 2023 party vote by vote type, six electorates combined#

The six electorates were chosen to span city, suburban, rural, Pasifika, Asian-majority and Māori seats: Auckland Central, Hutt South, Māngere, Rangitata, Upper Harbour and Te Tai Hauāuru [20]. Ordinary = electorate total minus the three special rows. Margin = (Lab + Grn + TPM) − (Nat + ACT + NZF), in points.

On-polling-day specials, which include the 110,000 election-day enrolments, were the most left-leaning group.

Sample: left–right margin by electorate#

Working: national special-vote ratios#

For each party, the share of specials is backed out from the preliminary and final counts. The ratio compares that with the party's preliminary share:

Formula: sₚ = (final votesₚ − preliminary votesₚ) ÷ S, and rₚ = sₚ ÷ (party p's preliminary share)

Here S is the valid special votes. A ratio above 1 means the party did better on specials than on election night.

2020 preliminary shares: Labour 49.1% and National 26.8% are the final results less the changes reported in [25]; Greens 7.6%, ACT 8.0%, NZ First 2.7% and Māori Party 1.0% are election-night figures [26]. The 490,000 valid specials (of 504,621 cast [27]) is an assumption.

Working: separating overseas from domestic (2023)#

The overseas ratio is taken from the sample above, applied to the 78,030 overseas votes [30]. The national domestic ratio is the remainder:

Formula: domestic ratioₚ = (S × rₚ − O × overseas ratioₚ) ÷ (S − O), with O = 78,030

The national back-out matches the sample's domestic ratios closely except for TPM, where the sample has only one Māori seat. Nationally, 2023 specials ran +14.3 points more left than ordinary votes, and domestic specials +15.9. Opportunity's national 2023 specials ratio was 1.34 [27][30], consistent with the sample figures.

Working: what the lost voters would have done#

Applying the domestic ratios to the 2026 preliminary shares gives an estimated split for the lost voters: Labour 32.0%, National 26.1%, Greens 16.6%, ACT 8.0%, Opportunity 7.7%, NZ First 7.5%, TPM 1.2%.

• Their left–right margin is +8.2, against −6.7 for the preliminary count: a gap of 14.9 points.

• Net cost to the left per 60,000 lost votes: 60,000 × 14.9% ≈ 8,900 votes, or about 10,000 counting Opportunity.

• That is about 0.35% of the final vote, under half a seat on average.

Part 4: Seat model and results#

Under current rules the model gives a right majority 60% of the time, a hung Parliament 33%, and a left majority 7.0%. With the old enrolment rules restored, the left's chance rises to 17%.

The law shifts the average seat count by only about 0.15 seats to the left and 0.25 from the right. Because the result sits on the 65/66 line, that small shift moves the probabilities a lot.

Method#

1. Split the 2026 specials into two pools. Overseas and dictation O ≈ 92,000–97,000 valid votes [4]. Domestic Dv = 302,000 − D, where D is disallowed votes, drawn from 40,000–70,000 [14].

2. Project each pool's party shares from the preliminary share π, adjusted by the Part 3 ratios and normalised to 100%:

Formula: domestic shareₚ ∝ πₚ × (domestic ratioₚ)^(α·κ) × e^εₚ, and overseas shareₚ ∝ πₚ × (overseas ratioₚ)^κ × e^εₚ

• α (dilution), drawn from 0.5–0.9. In 2023, up to 86% of domestic specials came from late enrollers. In 2026 the share is roughly two-thirds, made up of people who enrolled in the weeks before 25 October and weren't on the printed rolls, and they are less left-leaning than election-day enrollers.

• κ = 1 + 0.35Z, Z ~ N(0, 1): how strongly specials lean this year, shared across every part of the model. This covers the spread between 2017, 2020 and 2023.

• ε ~ N(0, 0.05): party-level noise.

3. Final votes = preliminary + O × s^os + Dv × s^dom.

4. Seats: Sainte-Laguë allocation of 120 seats among parties with at least 5% or an electorate seat. National and TPM each get the larger of their entitlement and their electorate wins. House size = sum.

5. Electorates: the change in a candidate's margin = n × lean, with n ~ 3,800–4,800 valid specials per seat, or 5,500–7,000 in Wairarapa (Part 1).

   • Labour v National: lean = (Dv ÷ (Dv + O)) × α × L_dom + η. Here L_hist = 4% + 4% × Z (the same Z), L_dom = L_hist ÷ 0.87 with overseas votes treated as neutral, and η ~ N(0, 5%).

   • Te Tai Tokerau (TPM v Green): lean ~ N(+1%, 7%), n ~ 3,500–4,800. Wellington Bays (Green v Labour): lean ~ N(+4.5%, 4.5%), n ~ 4,500–6,000.

   • Basis: in 2023, Labour–National marginals moved by about −1 (Hutt South), +2 (Te Atātū), +1.4 (Nelson) and about −1 (Mt Albert) and about −3 (New Lynn) points on specials. In 2020 three National leads flipped to Labour, and in 2017 Hutt South specials favoured Labour by about +10 [8][9][11][24][26][31][34]. The 5% seat-level spread is deliberately wider than 2023's range. The 2023 shifts averaged close to zero, so the 4% central lean rests on the stronger Labour shifts of 2017 and 2020.

6. Old-rules counterfactual: the same random draws, plus V_lost ~ 40,000–80,000 party votes at the full domestic pattern (Part 3). Each seat also gets (V_lost + A) ÷ 71 extra electorate votes at lean L_dom + η, where A ~ 40,000–75,000 wrong-address votes (Part 2).

7. Runs: 60,000 simulations with paired draws, so the two scenarios differ only by the law's effect.

Validation#

• Preliminary inputs reproduce the published seats: National 42 (35 earned), Labour 36, Greens 16, ACT 13, NZ First 12, Opportunity 7, TPM 5, for a total of 131 [1].

• 2023 pattern at full strength (α = κ = 1, no electorate flips): Labour 35, Greens 17, NZ First 11, Opportunity 8. That gives right 66 and left 65 of 131, with National on 28.8%. This matches Geddis's typical pattern of NZ First losing one seat and the Greens gaining one [2]. David Farrar's projection keeps the right on 67 [3]. NZ First's 12th seat is marginal, which explains the one-seat difference.

• 2020-strength lean (κ = 1.6): right 66 and left 66 of 132, with National on 28.4%. That matches Rashbrooke's threshold for the left's path [3].

Most likely seat outcomes (current rules)#

Sensitivity#

In both scenarios the law costs the left roughly 10–14 points of majority probability.

Part 5: Other published models#

Three other public simulations of the special votes put a right majority at between 34% and 70%, against 60% here. They start from the same preliminary count and differ mainly in how far they discount the specials' usual left lean for the enrolment change. None estimates the votes lost to the law or an old-rules comparison.

The outcome categories differ slightly between models. The middle band covers a tie or any result where neither bloc has a majority without Opportunity.

• The lean assumption drives the spread. Hide's right-majority figure is 62% at the full 2023 lean and 78% at half [36]. The electionhq.nz model's own "lean as in 2023" scenario gives the right a majority 65% of the time, against 34% in its central case [37].

• Wairarapa is the hinge seat. In Hide's runs the right keeps its majority 94% of the time if National holds Wairarapa and 47% if Labour wins it [36]. He estimates about 6,500 specials still to come there, well above the national average, and this report's model now uses 5,500–7,000 for that seat.

• NZ First's 12th seat is marginal. Hide gives NZ First a 40% chance of keeping it [36], consistent with the validation in Part 4.

Limitations#

These are model estimates, not official forecasts. The Electoral Commission publishes no probabilities, and the official count on 27 November will settle everything.

• Small sample: the overseas ratios and Opportunity's ratios come from six electorates. TPM's domestic behaviour rests on a single Māori seat in the sample; the national back-out is used instead.

• No direct data on late enrollers: ballots are secret. Domestic specials stand in for late enrollers because up to 86% of 2023 domestic specials came from that group.

• Lost votes are partly unmeasured: people turned away at the booth were never counted [15]. The 40,000–80,000 lost-vote range and the turned-away estimate rest on assumed shares, not counts. The official disallowed-vote total on 27 November will test them.

• Assumed inputs: some inputs are assumed rather than published: the disallowed totals, how many of the 399,000 specials will be valid, the 2020 valid-special count, and the size of the dilution factor α.

• Electorate swings are broad-brush: they come from a handful of past marginals and ignore local candidate effects and vote-splitting campaigns. The assumed 4% lean to Labour is above what 2023's marginals showed; a lower lean would reduce the left's chances further.

• Bloc definitions: Opportunity is counted with the left, but it has not committed to either side. A "hung" result (no bloc with a majority) would leave Opportunity or NZ First deciding. Recounts, and any judicial challenge, could still change close seats.

Sources#

All sources were opened in full on 12/11/2026 or 13/11/2026 except 13, 15, 17, 19 and 33, as noted against each. The report does not rely on 13, 17, 19 or 33; the relevant figures come from the other sources cited. Source 15 is cited only for the point that people turned away are not counted. The model scripts (hist.py, national.py, sim2.py) were supplied alongside this report.

1. 2026 New Zealand general election, Wikipedia: preliminary results, enrolment figures

2. The hangover: key takeaways and reactions from Election 2026, The Spinoff, 8 Nov 2026

3. Election analysis: who campaigned well, can the left still win, Max Rashbrooke

4. Fewer than 500 votes in it: the electorates still too close to call, 1News, 8 Nov 2026

5. Voters urged to enrol early as same-day enrolment scrapped, PMN News

6. Electoral law changes, Ministry of Justice

7. Chris Bishop slams vote-splitting campaigns as he farewells Parliament, 1News, 10 Nov 2026

8. Hutt South 2023: candidate votes by voting place, Electoral Commission

9. Hutt South 2017: candidate votes by voting place, Electoral Commission

10. Hutt South: behind the scenes of an electorate down to the wire, NZ Herald, 2020

11. National loses two seats, meaning NZ First is crucial, The Spinoff, 3 Nov 2023

12. The wait begins as Opportunity knocks, interest.co.nz, Nov 2026

13. Electoral Commission defends enrolment efforts, NZ Herald, Oct 2026 (could not be opened: the site blocks automated access; figures taken from 1, 5, 14 and 16)

14. Enrolment closes 25 October, Newswire, Sep 2026

15. Tens of thousands could be turned away from voting – but no one is counting them, Newsroom, 3 Aug 2026 (could not be opened: the site blocks automated access; read from a search extract only)

16. How the new enrolment rules could affect the election result, 1News, 16 Oct 2026

17. Hundreds of thousands of Kiwis still not enrolled, NZ Herald, Oct 2026 (could not be opened: the site blocks automated access; figures taken from 30)

18. Special votes: National and ACT lose majority, RNZ, 3 Nov 2023

19. Minor parties celebrate record-breaking victories, Newshub, Nov 2023 (now redirects to Stuff with no article text; figure confirmed by 29 and 30)

20. Electoral Commission 2023 party votes by voting place: Auckland Central, Hutt South, Māngere, Rangitata, Upper Harbour, Te Tai Hauāuru

21. Preliminary results for the 2017 General Election, Electoral Commission

22. National drops 2 seats, Greens/Labour pick up 2, RNZ, 7 Oct 2017

23. Final election results: negotiations now on a knife-edge, 1News, 7 Oct 2017

24. 2020 General Election official results, Times (Electoral Commission release)

25. Track NZ Election 2023 live results, NZ Herald: 2020 preliminary-to-final shifts

26. Election 2020: Labour claims victory, RNZ, and Labour has the numbers, Kaniva Tonga (RNZ): 2020 election-night figures

27. Preliminary results for the 2023 General Election, SunLive (Electoral Commission release)

28. Live updates, 15 Oct 2023, The Spinoff: 2023 ordinary-vote shares

29. 2023 General Election official results, Scoop (Electoral Commission release)

30. Official results for the 2023 General Election, Electoral Commission

31. Live updates, 3 Nov 2023, The Spinoff: New Lynn and flipped seats

32. Hutt South 2020 official result, Electoral Commission

33. Electoral Commission admits voting-booth errors, The Post, 6 Nov 2026 (no article text retrievable; not relied on)

34. The special votes are in and it's bad news for the cannabis referendum, and National, The Spinoff, 6 Nov 2020: 2020 seats overturned on the final count
35. OIA response 2023-99: enrolments during the 2023 voting period, Electoral Commission

36. Specials, Rodney Hide, Brash & Mitchell, Nov 2026

37. NZ election 2026 special votes simulation, electionhq.nz

38. nz-2026-specials-model, mikethemerry, GitHub