Federal MRP, February–March 2026

01

Ten points of vote and forty seats in four months

One Nation's modelled primary vote rises from 17 per cent to 27 and the Coalition's falls from 24 to 21. On seats the two move much further than that, because the Coalition is now below the level at which single-member districts return anything.

The bars show first preferences at the May 2025 election and in the two MRP models, in per cent.

Both sets of central estimates total 150. The ranges come from the model's seat classifications and the report publishes no national two-party preferred figure for this wave, where the previous model published 56–44. Against the May 2025 election this model puts Labor down 5.6 points, the Coalition down 10.8, One Nation up 20.6, the Others down 4.0 and the Greens down 0.2.

The two models disagree about the shape of the parliament rather than about the direction of travel. In October and November the Coalition was losing seats to One Nation at roughly one for one and Labor was gaining from the wreckage. Here the Coalition's losses accelerate past the point where that trade continues: it drops 20 more seats while One Nation gains 40, which means the extra 20 come from Labor and the Others. Labor falls 15 seats between the two models on a primary vote four points lower.

Labor still governs in this model. Its 83 seats are seven above the 76 a majority needs, and its range of 77 to 86 sits entirely above that line, so every simulation in the band returns a Labor majority. What changes is who sits opposite: One Nation at 52 against a Coalition at nine, which is fewer seats than the thirteen the Greens and the Others held between them after the 2025 election.

The classifications say how firm any of that is. Labor leads 83 seats of which 56 are firm, 20 lean and 7 are in doubt. One Nation leads 52 of which only 23 are firm, 25 lean and 4 are in doubt — so just under half of its total rests on seats the model puts between 70 and 95 per cent probability. A reader should hold One Nation's 52 more loosely than Labor's 83 for that reason, and the published ranges, 46 to 55 against 77 to 86, say the same thing in a different notation.

02

The movement roughly doubles in every cell of the grid

The share of 2025 Coalition voters now intending to vote One Nation ran from 5 to 35 per cent in the October model. Here it runs from 13 to 55, and every one of the 45 cells has risen.

Figures in brackets are the same cells in the October–November 2025 model, from ACE-IU-2026-041. These are modelled estimates from the post-stratification frame rather than observed cell counts, and neither report publishes an interval for them. The Year 12 column is labelled "Grade 12" in the October model and "High School Certificate" here.

All forty-five cells rise and twenty-one of them at least double. The smallest rise is two-fifths, among inner-metropolitan Year 12 holders over 55, and the largest is nearly threefold, among outer-metropolitan postgraduates under 35 who go from 7 per cent to 20. The largest rise in points is 26, among rural and regional voters aged 35 to 54 who did not finish Year 12, where the figure goes from 28 per cent to 54.

The structure survives the movement almost intact. The least-educated column is still the highest figure in every row and the two degree columns the lowest, every location band still sits above the one inside it, and every age band still sits above the one below. What has changed is the level, and the level has changed everywhere at once, which is the signature of a national shift rather than of one region or cohort turning. One piece of the structure did change: in the October model the Year 12 and TAFE columns were identical in eight of the nine rows, and here TAFE runs above Year 12 in all nine. The two groups have separated.

The single figure that carries the most for a campaign is the rural and regional row for voters over 55: 55 per cent of those without Year 12, 49 of the TAFE-qualified, 45 with Year 12, and still 34 and 35 among degree holders. In that part of the country the model has One Nation taking about half of everyone who voted Coalition thirteen months earlier, with no education band offering much shelter.

03

The model stops borrowing two electorates' preferences and measures them instead

The October model used the flows observed in Maranoa for every Coalition-against-One Nation final count in the country and those observed in Hunter for every Labor-against-One Nation count, and named that as an extra source of uncertainty. This model replaces both with preferences respondents allocated themselves.

The reasoning the report gives is straightforward. One Nation's vote has risen sharply and the Coalition's has fallen, so the people now holding a One Nation ballot paper are not the same people whose preferences were observed in 2025, and flows measured then may no longer describe them. To measure the new ones it asked respondents to rank all five voting options in full. The report notes that respondent-allocated preferences have been unreliable in single polls of one or two thousand, and that a sample above 8,000 is what makes the approach worth attempting.

In Labor-against-Coalition final counts the model applies a national One Nation flow of 78.5 per cent to the Coalition and 21.5 to Labor, implemented by raising the Coalition's share of One Nation preferences by four points in every electorate. All other flows in the model still come from the 2025 election.

Two patterns in that table do the work. Coalition preferences reach One Nation at 60 per cent in the inner suburbs and 79 in the country, so the further from a capital city a contest is, the more a Coalition elimination helps One Nation. Labor preferences run the other way and flatten: 72 per cent to the Coalition in the inner suburbs falling to 64 in the country, which is the same regional weakness the October model described in words when it put Labor's rural flow below 60 per cent.

The Others column is where the model's assumptions are doing the most and resting on the least. In rural and regional Labor-against-One Nation contests the Others flow is 53 to 47 and in the equivalent Coalition contests it is 54 to 46, both close enough to even that a small error changes which candidate the model puts in front. Those are precisely the seats section 01's One Nation total is built from, and precisely the contests for which Australia has the least observed history.

The seat classifications were rewritten in the same report. The October model defined its three categories by two-candidate margins alone: in doubt at a winning share of 50 to 51 per cent, lean at 52 to 54, firm above. This model defines them by a calculated probability of victory — in doubt below 70 per cent, lean between 70 and 95, firm above 95 — and gives the old margins as a rough equivalent. The categories carry the same names in both reports and no longer mean quite the same thing.

04

A central forecast of nine seats inside a range of nine to seventeen

Every other party's central estimate in this model sits inside its range. The Coalition's sits on the floor of its own, which says the published nine is the least favourable outcome the simulations produced rather than the middle of them.

Firm, lean and in doubt sum to the total leading for every party, and the totals sum to 150. In doubt is a probability of victory below 70 per cent, lean 70 to 95 and firm above 95, against an electorate vote share interval of about ±5 points at 95 per cent.

The Coalition's nine seats are five firm, one lean and three in doubt, and its range runs from nine to seventeen. A central forecast equal to the bottom of a range is an unusual thing to publish and it has a plain reading: the model finds almost no outcome in which the Coalition does worse than nine and a good many in which it does somewhat better. A party reading this should take nine as a floor rather than as an expectation, and anyone citing the headline should say so.

The same table makes One Nation's total the softest in the model. Twenty-five of its 52 seats are classified lean, meaning a probability of victory between 70 and 95 per cent, and four more are in doubt. Labor's 83 is the opposite: two-thirds of it is firm. If every lean seat in the model fell the wrong way for the party leading it, One Nation would lose half its total and Labor would lose a quarter of its.

The retention table underneath these figures is stable in a way the seat totals are not. Labor holds 76 per cent of its 2025 voters, the Greens 82, One Nation 97 and the Others 61. The Coalition holds 56 per cent, and 37 per cent of its 2025 voters say One Nation. The report also notes that its estimate for the New South Wales seat of Farrer should not be read as a prediction of the by-election then pending there, because by-election dynamics differ from a general election's.

05

Where the vote sits, state by state and cohort by cohort

One Nation leads the Coalition in every state and territory except the Australian Capital Territory, and leads Labor in Queensland. The model publishes no interval for any of these figures and this briefing computes none.

Every row sums to 100 in the report. The report also publishes the three intermediate income bands, which run from 28 per cent Labor and 27 One Nation at $45,000 to $75,000 to 31 and 24 at $125,000 to $200,000.

The Coalition's figures describe a party with one constituency left. It takes 34 per cent of those earning more than $200,000, which is the highest figure it records anywhere in the model and thirteen points above its national 21. Everywhere else it sits between 15 and 24 per cent, and in regional and rural Australia it is on 21 against One Nation's 37. It reaches 21 per cent in New South Wales, Victoria, Queensland and the Northern Territory, and its lowest figure is Tasmania at 15.

One Nation's lowest figures identify the same boundary from the other side: 11 per cent in the Australian Capital Territory, 17 among postgraduates and in the inner suburbs, 18 among undergraduates and 19 among under-35s and the highest income band. Its highest are 38 among those who did not finish Year 12, 37 in regional and rural Australia and 33 among the over-55s. Education separates its vote more than income does, and location more than either.

Queensland is the only state where the model puts One Nation ahead of Labor, at 31 to 25. In every other state and territory Labor leads on primaries, and in the Australian Capital Territory it leads One Nation by 31 points. A margin that runs from six points behind in one jurisdiction to thirty-one ahead in another is the clearest statement in the model of how unevenly this vote is spread, and an unevenly spread vote is the condition that turns 27 per cent into 52 seats.