Preferences are the whole game

The statewide vote and the seat count disagree, and preferences are where they part. The Coalition leads Labor in every Victorian poll taken since Ben Carroll replaced Jacinta Allan as premier on 28 July. Nine two-party preferred readings — Labor and the Coalition head to head, with every other party's votes distributed as preferences — from five houses span 51–49 to 57–43, and the post-change average is 53.9–46.1. That average implies a uniform swing — the same statewide change applied to every seat — of 8.9 points. The Coalition needs 7.6 points, a two-party figure of about 52.6, to reach 45 seats.¹

On that arithmetic the election should already be decided. Victoria does not count votes in a single statewide pool, though. Both published seat models, fielded before the change of premier, put the Coalition on 39 of 88 — six short of a majority — even though it leads the primary vote, the first-preference vote. The uniform-swing reading and the seat models are not in error. The conservative vote is splitting.

Coalition two-party preferred, post-change average: 53.9–46.1 — Eleven polls fielded after 28 July 2026; nine published a two-party figure, from five houses.¹

The lever nobody has measured

Ace Strategies ran the sensitivity the models do not. Holding every primary vote exactly where it is and moving only the preference flow from One Nation — where a party's votes go when its candidate is excluded — the Coalition lands anywhere between a bare majority and a loss. On the 2025 federal Victorian flows DemosAU adopts, One Nation 75–25 to the Coalition, the Coalition reaches 53.0 and clears the line with room to spare. At 70–30 it reaches 51.8, still short. At 65–35 it reaches 50.7. At 60–40 it reaches 49.6.¹

Nothing else changes in that table. No primary vote moves. Between the 53.0 that the 75–25 default produces and the 49.6 that a 60–40 flow produces is 3.4 points of two-party preferred, and the Coalition's whole majority requirement is 1.3 points above its current average. Whether it forms government turns on where One Nation voters' later preferences land.

One Nation's preference flow to the Coalition, and what it does to the result
ON flow to Coalition | Coalition two-party preferred | Implied uniform swing | vs the 7.6 line
60–40 | 49.6–50.4 | 4.6 | Short
65–35 | 50.7–49.3 | 5.7 | Short
70–30 | 51.8–48.2 | 6.8 | Short
75–25 (default) | 53.0–47.0 | 8.0 | Clears
80–20 | 54.1–45.9 | 9.1 | Clears
85–15 | 55.2–44.8 | 10.2 | Clears¹

Three of those six rows leave the Coalition short of 45 seats, and all of them are inside the range that published evidence supports. The default in the table is not a guess: it is the 2025 federal Victorian flow, the one DemosAU adopts, with One Nation 75–25 to the Coalition.¹ The ABC's own election coverage describes One Nation preferences at roughly 65–77% to the Coalition and historically around 60%.²

Antony Green's July reading of South Australia was lower still: around 70% of One Nation preferences went to the Liberal Party and only around 65% of Liberal preferences went to One Nation. His conclusion, in his own words, was that preference flows between One Nation and the Liberal Party are weaker than between Labor and the Greens.³

The one published seat model that declares its calibrations sits below the 75% default, not on it. The Swing Is On runs 5,000 randomised simulations per seat with 24 manual preference-flow overrides, and its One Nation→Coalition calibration is 69.9%, with Greens→Labor at 79.9% and Labor→Coalition at 71.5% in three-cornered contests. Taking Ace Strategies' sensitivity table at face value, a flow of 69.9% sits between the 65–35 and 70–30 rows, both of which leave the Coalition short of 45. On the model's own numbers, its headline call is a Coalition minority government on 37.3 seats of 88.⁴

Of the 56 seats Labor won in 2022, the party is the most-common winning party in 35 of The Swing Is On's simulations. In the 21 where Labor trails, the Coalition leads the count in 15 and One Nation in five. Its five most exposed Labor seats are Bass at a 0.9% chance of retaining, Melton at 1.3%, Bayswater at 6.8%, Pakenham at 7.4% and Sydenham at 13.5% — and One Nation's five best chances, Melton at 83.5%, Bendigo East at 63.7%, Cranbourne at 56.1%, Sydenham at 55.5% and Mill Park at 53.9%, are all Labor seats.⁴

The asymmetry that does the work

The reason a split right primary hurts the Coalition is that the two sides of politics do not coalesce at the same rate. Greens preferences to Labor run at roughly 82–93% across the recent Victorian polling, depending on the stage of the count and which party is being excluded.² One Nation preferences to the Coalition run at roughly 65–77%, historically around 60%.² The left holds together about ten points harder than the right does.

Resolve's August poll measured both sides of this directly. In contests between Labor and the Coalition, One Nation preferences favoured the Coalition 77–23. In contests between the Coalition and One Nation, Labor preferences favoured the Coalition 73–27. A Coalition that can reach the final two gains on preferences; a Coalition sitting in the crossfire does not.⁵

Where Labor falls to third in a seat, the final contest is Liberal–One Nation, not Labor–Liberal, and the flows that matter run the other way. The federal by-election in Farrer in May 2026 is the worked example: Nationals preferences went 69.1% to One Nation and Liberal preferences 59.0%, and One Nation reached 50.2% on Nationals preferences alone.⁶ On those numbers, the weaker side of the asymmetry is the one that carries the seat.

One Nation is contesting 87 of the 88 lower-house seats, with only Northcote unnamed.⁷ The party is running where it cannot win, in seats where its primary vote may push Labor out of the final two and hand the seat to the Coalition on preferences, and in seats where it will finish second itself, where Labor's preferences to the Coalition decide the outcome. Either way the seat turns on a flow nobody has measured.

The only counted One Nation vote in Victoria

There is one Victorian contest where One Nation has actually been counted. At the Nepean by-election on 2 May 2026, with no Labor candidate on the ballot, One Nation's Darren Hercus took 24.5% of the primary in a safe Liberal seat. The Liberal member held it 63–37 after preferences, against an independent.¹

A single by-election is not a flow estimate. Nepean had no Labor candidate, no Green finishing third and a Liberal incumbent with the advantage of an open field. It is the only place in Victoria where a One Nation primary of that size has been put through a real distribution, and the answer — the seat did not change hands — is the more useful half of the result.

The seat-level modelling that would settle the question does not exist yet. The Swing Is On runs 5,000 randomised simulations per seat and applies its own preference distribution engine, which is the closest thing published. Its seat-level leader win probabilities range from 66.8% in South-Eastern Metropolitan up to 86.8% in Eastern Victoria, and the seats it flags as most uncertain are provincial and outer-metropolitan — which is exactly where the three-cornered contests are. The model's headline call for Victoria is a Coalition minority government on 37.3 seats of 88.⁴

A minority government on 37 seats and a majority on 45 are both consistent with the same poll average, separated by how preferences break in perhaps a dozen seats. No Victorian count has measured a One Nation preference flow at seat level, so the sensitivity table above is Ace Strategies' projection rather than a reading of any vote, and 2026 will be the first result that can test it.