Elections

Australian elections are preferential and multi-party. More and more of them are decided on the crossbench. We model them that way.

How the model works Applied to Victoria Forecasts Analyses Forecast programme

Victoria88-26 is live. First-preference estimates for every party in all 88 Victorian electorates are published now. Win probabilities follow once nominations close on 9 November and the how-to-vote cards are registered. Polling day is 28 November. The full post-election diagnostics publish in December 2026.

How the model works

The 98th Percentile election model is one probabilistic description of an election, fitted in three stages. Each stage conditions on the one before.

Pooling the polls. Every published voting-intention poll is pooled into a daily estimate of each party’s vote. The model corrects for each pollster’s persistent lean. It also allows for the whole polling industry being wrong in the same direction at once, as it was at the 2019 federal election. That common error is estimated from past elections and kept in every published range.

Distributing the vote across electorates. Each party’s statewide vote is spread across the electorates, starting from how each area voted at a recent election and what the Census says about it. The electorate figures always add back to the statewide figure.

Preferences and seats. A full preferential count is simulated in every electorate. Every seat is treated as a general multi-candidate contest. The classic two-party contest falls out as a special case rather than being assumed. Seat results are published only once the real candidate field and the how-to-vote cards are known.

Every published figure is a summary over many thousands of simulated elections. Each carries a plausible range: the model puts a nine-in-ten chance on the result falling inside it. The full method →

How our election models differ

Three commitments separate this work from a poll of polls.

First, we never use a two-party-preferred figure. Pollsters compute it under their own preference assumptions. A two-party summary also describes a two-party contest, and several Australian jurisdictions are no longer having one. We model the primary votes and simulate the preferences ourselves.

Second, we never assume uniform swing. Where a large share of seats will not finish as Labor against the Coalition, uniform swing has nothing to say about them.

Third, pollster leans are estimated jointly with an industry-wide error, anchored to past election results. The model can tell pollsters disagreeing from pollsters being collectively wrong.

Applied to Victoria

Victoria elects its Legislative Assembly on 28 November 2026. It has 88 single-member electorates under full preferential voting, fought on the same boundaries as 2022.

The polls. Six pollsters publish Victorian state voting intention: RedBridge and Accent Research, Resolve Strategic, Freshwater Strategy, DemosAU, Roy Morgan, and Newspoll conducted by Pyxis. Each poll is reviewed before it enters the model. The industry-wide error is anchored to the 2010, 2014, 2018 and 2022 elections.

One Nation. Victorian pollsters did not prompt for One Nation until early 2026. That changed the measuring instrument for every pollster at once. The model treats 1 February 2026 as a hard break, and no earlier poll informs the One Nation series. Its geography comes from the party’s 2025 federal vote, adjusted for each electorate’s education profile. How far to adjust is calibrated on the 2026 South Australian election, where the same surge has already run its course.

Preferences. Win probabilities wait for the how-to-vote cards. In June 2026 a RedBridge and Accent MRP projected 12 One Nation seats. Applying different preference flows to the same primary votes produced 27. Identical first preferences, a difference of 15 seats. The preference model decides the chamber, so it must rest on the real cards.

Every choice specific to Victoria, with its assumptions graded, is set out in Victoria88-26: the Victorian application →

Forecast programme

Election	Date	Status	Notes
Victoria Legislative Assembly, 88 seats	28 Nov 2026	Live	First preferences published now. Win probabilities after 9 November. Diagnostics December 2026.
New South Wales Legislative Assembly, 93 seats	Mar 2027	Planned	Next application. Optional preferential voting means exhausted ballots must be modelled.
Federal House of Representatives, 150 seats	Due by 2028	Planned	Full preferential, so supported by the current method.

We do not forecast upper houses. Victoria’s Legislative Council has moved away from group voting tickets, and the replacement system has no Victorian precedent. A separate approach to upper houses is under investigation.

Use of AI

I used large language models, Anthropic’s Claude Opus 5.5 and Claude Fable 5, to test this work for narrative flow, logical consistency, bias and errors. They also assisted with drafting and with the site’s code. Responsibility for the content is mine alone.

John Ward (The 98th Percentile)