The correction.

Every pollster that measures One Nation recorded a lower figure in its latest release than at its mid-year peak. The size of the fall depends on the house.

Newspoll, YouGov and RedBridge/Accent show two to three points. Resolve, DemosAU and Essential show three to four. Roy Morgan, which recorded the highest peak of any series at 31.5 per cent in the third week of June, shows seven and a half.

The fall was not a straight line. One Nation reached a second peak of 30 per cent in Newspoll in late August and again in mid-September, and in YouGov in the first week of September, after its candidate won the Secret Harbour state by-election in Western Australia on 29 August with 39.1 per cent of the primary vote. The October readings are the first in which every house moved the same way at once. Newspoll fell two points, Resolve fell two, Roy Morgan fell one and a half, and DemosAU's long-fieldwork model fell four against its predecessor.

Two features of the October numbers matter more than the headline. The first is that the combined right-of-centre vote has barely moved. One Nation and the Coalition together polled 48 per cent in both Newspoll and Resolve in October, and between 47 and 50 per cent in every national poll since late August. The votes One Nation has lost have mostly gone to the Coalition, whose primary vote has recovered from 17 per cent in Newspoll in June to 20 per cent in October, and from 18 to 22 per cent in DemosAU between June and the August–September model.

The second is that the order of the three parties now depends on which poll is read. RedBridge/Accent and YouGov still have One Nation first. Newspoll and Resolve have Labor first, by one point and four points. Roy Morgan and the DemosAU model have Labor first, One Nation second and the Coalition third, with the whole field inside a band of five to six points. DemosAU's head of research George Hasanakos described the position on 11 October as "the electoral twilight zone", with three parties between 20 and 30 per cent, where "even modest swings can have a significant impact".

The DemosAU/Capital Brief MRP published on 11 October was built from 9,740 interviews taken between 4 August and 19 September, projected onto all 150 seats and run through 20,000 Monte Carlo simulations.

It is the second of these models in a row, and the comparison with the May–July edition is the most instructive single dataset of the quarter.

A four-point fall in the national primary vote produced a 36-seat fall in the central projection. Labor, on 27 per cent of the primary vote, is projected to win a majority in its own right. Ben Raue of The Tally Room, writing on 14 October, put the mechanism in one sentence: "small shifts in the vote can swing a lot of seats."

The shift is concentrated where One Nation was strongest. DemosAU divides the House into 45 inner-metropolitan, 64 outer-metropolitan and 41 regional and rural seats. In the May–July model One Nation polled 41 per cent across the 41 regional and rural seats and led in 39 of them, with a live possibility of all 41. In the August–September model it polls 34 per cent in those seats and is projected to win 17.

Seven points off the regional primary vote cost 22 projected regional seats. The reason is set out in the next act, and it is the same reason the Nationals, projected at zero to two seats in July, are now projected at four to eleven. Hasanakos said the result showed that "the obituaries that have been written for the Nationals, and indeed for the Coalition, have been premature."

One Nation came first or equal first on primary votes in 52 of the 150 seats in the August–September model. It is projected to win 24 of them. Labor takes 11 of the remainder and the Coalition 17.

The phrase DemosAU used to explain this was that One Nation "is currently more popular than the Coalition" and "also more unpopular". In a three-cornered seat, the party that finishes first on primaries wins only if it also wins the preferences of the party that finishes third. One Nation receives those preferences from neither major party at the rate it needs.

Where Labor runs third, its voters preference the Coalition over One Nation at roughly three to one. That is how the Coalition wins 15 regional and rural seats in which One Nation leads on primaries. DemosAU named two of them. In New England the Nationals lead One Nation 53–47 on the two-candidate count. In Farrer, which One Nation won at the May by-election with 39.5 per cent of the vote and 57.6 per cent after preferences, the model now has the Coalition ahead 54–46.

Where the Coalition runs third, its voters split. In regional and rural seats the model sends only 30 per cent of them to One Nation. Antony Green's analysis of the Secret Harbour by-election put the Liberal-to-One Nation flow at about 70 per cent in a seat where One Nation won, and Peter Brent's analysis of the eleven Labor-versus-One Nation contests at the South Australian election put the Coalition flow to One Nation at 66.6 per cent. The model's regional assumption is lower than either. Raue's view is that the model's flows are "probably about right" on the South Australian and Farrer evidence, with the Coalition-to-One Nation figure possibly a little low. The margin on that figure matters, because in the 15 seats it is the difference between winning and losing.

The flow in the other direction is not symmetrical. At the 2025 election three in four One Nation voters preferenced the Coalition over Labor, and the model lifts that to 82 per cent. The Coalition therefore receives One Nation preferences wherever One Nation is excluded, while One Nation receives Coalition preferences only in part. A party whose voters are reliably transferable to a rival, while the rival's voters are not reliably transferable back, is at a structural disadvantage in every seat it cannot win on primaries alone.

Labor's majority in the model rests on this asymmetry and on a margin of luck. The winning candidate polled exactly 50 per cent of the two-candidate count in nine seats, and Labor won six of them. Raue estimates that a swing of under 2 per cent would put Labor in minority. William Bowe's reading was that the 84-seat central figure "flatters Labor". Both readings are consistent with the model's own ranges, which put Labor's floor at 73 seats, three short of a majority.

The polling record between July and October allows the fall to be dated and, in part, attributed.

The first and largest movement was among women. Between the June and August DemosAU polls Hanson's rating among women went from slightly positive to 26 positive and 52 negative, and One Nation's primary vote among women fell ten points against three among men. DemosAU's release listed the events of the intervening period: the 17 June National Press Club address, the Europe trip funded by Gina Rinehart, the CPAC London speech, and the 22 July radio interview in which Hanson said domestic violence was "a two-way street". The September poll recovered three of the ten points among women.

The second movement was on the question of readiness. Resolve asks respondents which party they expect to win the next election. In October 20 per cent named One Nation, down eight points in a month, level with the Coalition and twelve points behind Labor. Hanson's own net rating in Resolve fell six points to +4. Jim Reed of Resolve said that "One Nation thrived on controversy and attention and we simply haven't seen them as much lately."

The third movement is the one described in Act I, the Coalition's partial recovery. Its primary vote is up three to four points across most series since June, and the August DemosAU poll found that 67 per cent of 2025 Coalition voters now intended to vote Coalition, up from 57 per cent in June. The Coalition's policy record over the quarter is one explanation commentators have offered. On 24 August Angus Taylor and Nationals leader Matt Canavan announced the Coalition would abandon net zero. On 6 October they announced a migration policy with net overseas migration held to 100,000 for two years and the humanitarian intake halved, which Hanson called "a photocopy" of her party's position. The polling cannot say whether voters returned to the Coalition because of these announcements, but the timing is consistent with it.

None of this reached the One Nation base. The August DemosAU poll found 96 per cent of 2025 One Nation voters intended to vote One Nation again. The party's vote in regional and rural Australia, at 34 per cent, remains the highest of any party in any region. Newspoll's July–September quarterly breakdown, pooled from four polls and 4,967 respondents, has One Nation on 36 per cent in Queensland, up four on the April–June quarter, with Hanson preferred to Albanese 52–48 as Prime Minister in that state. Roy Morgan's October fortnight has One Nation at 29.5 per cent in Western Australia and 28 in Queensland, against 19 in Victoria and 14.5 in South Australia. The party has won two by-elections this year on primary votes near 40 per cent. The correction has come off the top of the vote, in the capitals and among women, and it has not reached the floor.

One Nation has surged twice before, and both surges decayed. The relevant question is whether the 2026 decay is following the same path.

Three things distinguish 2026 from the earlier episodes. The base is higher by a factor of three to four. The 1998 and 2017 federal peaks were in the range of 11 to 12 per cent, and the 2026 peak was 30. A fall of four points from 30 leaves a party that is still the second or third largest in the country. A fall of four points from 11 left a party at 7.

The party has converted polled support into seats in this cycle. It won four lower-house seats and 22 per cent of the vote in South Australia in March, the federal seat of Farrer in May and the state seat of Secret Harbour in August. In 1998 and 2017 the decay set in before any comparable test, and in 2026 it set in after three.

The decay to date is slower as a share of the vote. Western Australia in 2017, where the party lost eight points in five weeks, remains the steepest decline on record. The next two months will show whether the 2026 slope steepens.

The precedent that does hold is the one Part 3 named. Polls measure stated intention and elections measure behaviour, and for a party that depends on winning three-cornered contests the gap between the two is wider than for either major party.

The assessment in Part 3 had two halves: One Nation had crossed the lower threshold of being a contender for real political power, and it was not on track to form government at the next federal election. Both halves stand, and the data since has moved the second half further from government.

On the polling to mid-October 2026, One Nation is the second or third party nationally on primary votes, within a few points of both the others. On DemosAU's model it wins 22 to 35 House seats, the Coalition 25 to 44, and Labor 73 to 85, which is a majority at the centre of the range and a minority at the bottom. In regional and rural Australia One Nation polls 34 per cent, more than any other party, and wins 17 of 41 seats, fewer than half. In 15 of the seats it leads, Labor voters' preferences elect a National or a Liberal instead.

The party's floor has held since July. Its ceiling has fallen. The model's May–July edition described a parliament in which One Nation was the alternative government. The August–September edition describes one in which the Coalition is again the larger non-Labor grouping on seats, while One Nation remains the larger on votes. That configuration of more votes and fewer seats is stable only as long as the two parties' voters preference each other asymmetrically. If the Coalition's voters begin to preference One Nation at the rate One Nation's voters preference the Coalition, the 15 seats change hands and the model changes with them. Nothing in the data to date shows that happening.

The Victorian election on 28 November is the first statewide contest since the correction began. The three most recent state polls, from RedBridge, Resolve and Freshwater, have the Coalition on 28 to 30 per cent, Labor on 24 to 25 and One Nation on 22 to 24, with Warren Pickering preferred as Premier by 8 to 14 per cent of respondents. Roy Morgan's seat model has One Nation winning 11 of the 88 lower-house seats. The abolition of group voting tickets, passed in August, removes the mechanism by which micro-parties previously won Legislative Council seats on small primaries, and leaves One Nation's upper-house prospects to its primary vote, which on these numbers is sufficient for several seats. The federal series will record whether a state result on 28 November moves the national number, as Secret Harbour did in September. Every poll ahead of it is on the Victorian election polls 2026 tracker.