Forecasting Australia's House from imperfect polls: an overview
Abstract. Australian final polls have missed each main party's primary vote by about 1.4 points at the seven House elections since 2007, overstating Labor at five, but a correction learned from those misses would have made our forecast worse, so the projection takes the polls as published. What decides the seats is the preference count: with One Nation level with Labor, most divisions are three-way contests. This paper sets out what the Australia section forecasts, how each step works, where the uncertainty lies, and how the other Australia papers fit together.
Summary
This paper is the way into the Australia section of Polling Forecast. It says what the section forecasts, why the rule that corrects British polls leaves Australian polls alone, why the seats are worked out by replaying the preference count in each of the 150 divisions rather than by a single two-party swing, and how the uncertainty is measured. The last section is a guide to the other Australia papers, each of which takes one part of the forecast and treats it in full.
The short version is this. At the seven House elections from 2007 to 2025 for which final primary votes survive, the average of the final polls missed each party's primary vote by about 1.4 points, and overstated Labor at five of them. But the misses changed direction from one election to the next often enough that a correction learned from them would have made our forecast worse: predicting each election with the correction learned from the other six was further from the result on average (1.35 points per party against 1.33 for the plain polls). So the forecast takes the polls as published. What it spends its effort on is the step from votes to seats, which in Australia has become unusually hard. In October 2026 the polls put One Nation level with Labor on first preferences and well ahead of the Coalition, and in a three-way contest the 2PP says little about who wins. The forecast therefore moves every 2025 candidate's vote with their party's national change and replays the count in every division.
What has changed since this paper was written
Updated 9 November 2026. The rest of the paper is as first published, except where a sentence below says it was corrected.
- Our figures beside the polls (26 October). The correction is off in Australia, so our projected shares are the poll average, and the pages say so rather than repeating the figures.
- When the polls missed (2 November): The miracle that was a sampling problem: Australia's 2019 polling failure.
- The map (November). The map can be coloured by any census figure; each figure has its own page (census maps).
What the Australia section forecasts
The forecast is for the next election to the House of Representatives: 150 single-member divisions, each won by full preferential voting. A party or group needs 76 seats for a majority. The current House first sat on 22 July 2025 and expires three years later. To hold the House election together with the usual half-Senate election, polling day must fall between 7 August 2027 and 20 May 2028, and the site uses the latest of those dates, 20 May 2028, for its polling-day forecast. The Prime Minister chooses the actual date.
The section answers five questions, each on its own page:
- How many seats would each party win if the election were held today? (The home page, the divisions and the states.)
- What is the chance that any party, or any combination of parties, reaches 76? (The every-outcome page.)
- Who is likely to make the final two in each division, and who wins it?
- How different could the result be on polling day, which may be 19 months away? (The election-day page.)
- How would the half-Senate election go? (The Senate page.)
As of 15 October 2026, built from the latest national poll of each of six pollsters, the projection is:
| Party | Primary vote (poll average) | Seats | 80% range | Seats in 2025 |
|---|---|---|---|---|
| One Nation | 26.5 | 64 | 49–70 | 0 |
| Labor | 27.7 | 57 | 47–69 | 94 |
| Coalition | 21.4 | 13 | 8–24 | 43 |
| Independents | 11 | 8–14 | 10 | |
| Greens | 13.2 | 4 | 2–7 | 1 |
| Centre Alliance | 1 | 1–1 | 1 | |
| Katter's Australian Party | 0 | 0–1 | 1 | |
| Others | 11.3 |
Table: The projection on the polls to 11 October 2026. "Others" in the primary vote includes independents and minor parties. The 80% range is the middle 80% of 2,000 simulated elections.
No party reaches 76. One Nation would be the largest party, 12 seats short; no party has a majority in 95% of the simulated elections. Labor's two-party-preferred (2PP) against the Coalition, worked out from the primaries with the 2025 preference flows, is 52.2%, against 55.2% at the 2025 election.
One rule for every country
The British forecast starts from a finding: at every general election from 2015 to 2024, England's final polls missed in fairly steady directions, and a correction built from those misses would have brought the forecast closer to the result (see Forecasting a British general election when the polls are wrong: an overview and Correcting the raw answers: the method behind our UK projection). The rule taken from that work is the same in every country.
For each party \(p\) and past election \(t\), let \(e_{p,t}\) be the average of the final polls minus the result, in points, so that a positive value means the polls overstated the party. The correction takes off half of the average past miss:
where \(y_p\) is today's poll average, followed by a rescale so the shares keep the same total. A party is corrected only if it has been measured at two or more elections. The rule is kept only if it passes a test: each past election is predicted with the correction learned from the other elections alone (leave-one-out cross-validation), and the correction stays only if its average miss per party beats the uncorrected polls by more than 0.01 points.
The model learns from no election before 2000, the site-wide cutoff set on the admin page. For Australian primaries that leaves seven elections, 2007 to 2025: in 2001 and 2004 only election-eve 2PP figures survive, and the final polls of 1987, 1990 and 1998 are kept for the record but not used.
Why Australian polls are not corrected
The evidence is the final-poll error for each party at each of the seven elections:
| Party | 2007 | 2010 | 2013 | 2016 | 2019 | 2022 | 2025 | Mean |
|---|---|---|---|---|---|---|---|---|
| Labor | +0.6 | +2.2 | +0.3 | −0.2 | +2.4 | +2.4 | −2.7 | +0.7 |
| Coalition | +0.9 | −2.9 | −1.2 | 0.0 | −2.8 | −0.3 | +2.3 | −0.6 |
| Greens | −0.8 | +3.1 | +1.7 | +0.8 | −0.4 | 0.0 | +0.9 | +0.8 |
| One Nation | +1.0 | −0.1 | +1.9 | +0.9 | ||||
| Others | −0.2 | −2.0 | −2.4 | −1.5 |
Table: Average of the final polls minus the result, primary vote, points. One Nation and others are measured from 2019, when pollsters named One Nation.
The table seems to suggest that polls overstate Labor and the Greens and understate the Coalition. The test shows how far that pattern can be trusted. Correcting was closer at only two of the seven elections (2010 and 2013) and further out at five, including 2025, when Labor was understated by 2.7 points after three elections of overstatement. On average the corrected figures missed by 1.35 points per party and the polls as published by 1.33. The correction fails, so our projection is the poll average. How wrong have Australian polls been? Primaries, the 2PP and the λ test sets out the record and the test in full.
Until 28 September 2026 the test also used 1987, 1990 and 1998, and over those ten elections the correction passed, but only because of 1987 and 1990, one Newspoll each, both overstating Labor and understating the Coalition. The owner then set the rule that elections before 2000 count for nothing that shapes a forecast, and the verdict reversed.
What decides the seats
If the average is left alone, the forecast must earn its place in the step from votes to seats.
For fifty years most Australian divisions ended as Labor against the Coalition, so a uniform swing in the 2PP was a fair guide to seats. That is no longer true. With One Nation on about 26% of first preferences, the forecast today has only 20 of the 150 divisions ending as Labor against the Coalition. The projected final two are Labor and One Nation in 88 divisions, the Coalition and One Nation in 19, the Coalition and an independent in 8, Labor and the Greens in 5, and other pairs in the remaining 10. Where three candidates are close, the one excluded third decides the seat, and a national 2PP says nothing about who that is.
Dickson in Queensland shows how fine the margin is. On today's figures Labor leads on first preferences with 30.0%, One Nation has 23.6% and the Coalition 23.3%. The Coalition, 0.3 points behind One Nation on first preferences, is excluded third; its votes flow to One Nation by about three to one, and One Nation wins 53.0 to 47.0. Had One Nation been excluded instead, the seat would have been a Labor against Coalition count. The division model and its tests works through this example.
The forecast, step by step
The pipeline has seven steps. Each is stated briefly here and developed in its own paper.
Step 1: the average. Take the latest national poll from each pollster whose fieldwork ended within 21 days of the newest poll. Each poll's primary votes are grouped as Labor, the Coalition (Liberal, LNP, Nationals and Country Liberal), the Greens, One Nation and others (independents included), the undecided are taken out and the shares rescaled to 100. The average for group \(p\) is
where \(J_p\) is the set of those polls that report the group. Each pollster counts once. On 15 October 2026 the six pollsters were Roy Morgan (11 October), Resolve Strategic (10 October), Newspoll (8 October), YouGov (6 October), RedBridge (2 October) and Essential (28 September).
Step 2: our reading. Equation (1) with the correction switched off, because it fails its test.
Step 3: the 2PP. Labor's national 2PP is its own primary plus each other group's primary times the share of that group's votes that reached Labor in the 2025 2PP count. From primaries to seats: preference flows and the 2PP.
Step 4: the divisions. Every 2025 candidate's first-preference share moves by half a uniform and half a proportional swing of their group, with half of any state swing shown by two or more pollsters' state breakdowns. The division model and its tests.
Step 5: the count. In each division the lowest candidate is excluded and their votes pass to the others by preference weights fitted on all 826 exclusions in the AEC's 2025 distributions of preferences, until two remain. If the two are Labor and the Coalition, the winner comes from the division's 2025 2PP plus the national 2PP swing. Run on the actual 2025 first preferences, the count picks 149 of the 150 winners. From primaries to seats: preference flows and the 2PP, The division model and its tests.
Step 6: the simulation. In each of 2,000 simulated elections every group's national share is moved by a random error the size of its past final-poll miss, each state and each candidate a little more, and steps 3 to 5 are run again. Simulating Australian polling error: where the chances come from.
Step 7: polling day. The election-day forecast pulls each party part of the way back towards its 2025 result, because at the House elections from 2010 to 2025 that came closer than the polls as they stood, and widens the ranges for the time left. Polling day: how far Australian polls move.
The uncertainty
The simulation gives three kinds of answer. The first is the 80% seat range in the table above. Ranges are wide for small errors in votes, because with three parties between 21% and 28% the order of exclusion in dozens of divisions turns on a point or two. The second is the chance that a combination reaches 76:
| Combination | Seats in the projection | Chance of 76 or more |
|---|---|---|
| Coalition + One Nation | 77 | 48% |
| Labor + Greens + independents and Centre Alliance | 73 | 47% |
| Labor + Greens | 61 | 5% |
| One Nation alone | 64 | 3% |
| Labor alone | 57 | 2% |
Table: Chances of a majority, from the 2,000 runs. No Coalition and One Nation agreement exists; the combination is shown because One Nation now polls above the Coalition.
The third is the division picture. On today's figures 57 divisions are safe (95% or more for the projected winner), 45 likely, 29 lean and 19 are toss-ups below 60%. Beside the simulation the site replays today's polls with each past election's polling miss (Past-miss lenses: today's polls with each past error replayed); 56 divisions change hands under at least one of the seven replays.
On polling day the picture is very different. The election-day forecast assumes, as the record advises, that each party moves most of the way back towards its 2025 result. That gives Labor 91 seats and an 84% chance of a majority. For One Nation, polling four times its 2025 vote, the pull-back is an assumption and not evidence, and Polling day: how far Australian polls move says so.
What we do not claim
We do not claim that Australian polls are unbiased. They overstated Labor at five of seven elections and understated it badly in 2025. We claim only that the direction of the next miss has not been predictable from the last seven.
We do not claim division-level precision. The forecast keeps the 2025 field of candidates and moves parties, not people. It cannot see new independents, retirements or members who have changed party.
We do not claim that a single number describes One Nation's rise. How its vote spreads across divisions and how Coalition voters rank it against Labor are the two largest uncertainties in the forecast, and neither can be tested on past elections, because nothing like its rise is in the record.
We do not claim that the "today" forecast predicts polling day. It answers a hypothetical question. What we don't know yet: limitations of the Australian forecast collects these points in one place.
A guide to the papers
The Australia library has eleven papers. This one, Forecasting Australia's House from imperfect polls: an overview, is the introduction. The others are described below in roughly the order in which the forecast uses them.
How wrong have Australian polls been? Primaries, the 2PP and the λ test sets out the evidence on which the rest stands: the final polls at the House elections from 1987 to 2025, how they were sourced, the error party by party on primaries and on the 2PP, the gap between the two largest parties, the record of each pollster and the leave-one-out test that rejects a correction.
From primaries to seats: preference flows and the 2PP covers the step from primaries to a 2PP and the preference weights between groups that drive the count, with their stability from 2004 to 2025.
The division model and its tests describes the division model, the half-uniform, half-proportional swing, and its tests on 2025 and on every election pair from 2004.
Simulating Australian polling error: where the chances come from explains where the chances come from: the 2,000 runs, the errors drawn at each level, the seat ranges, the majorities and what the design leaves out.
Past-miss lenses: today's polls with each past error replayed describes the past-miss replays and what each would do to the House.
Polling day: how far Australian polls move asks how far Australian polls move before polling day and explains the pull-back towards the last result.
The half-Senate: a simplified quota count describes the half-Senate forecast and its simplified count.
Rating Australian pollsters, and how to read a poll release describes the pollster ratings, each pollster's lean, and how to read an Australian poll release.
The Australian data, its sources and the law on polls documents where every number comes from, the law on polls, and the source grades.
What we don't know yet: limitations of the Australian forecast collects every known weakness of the Australian forecast and says what evidence would address it.
References
Australian Electoral Commission, House of Representatives results and distributions of preferences, 2004–2025 (CC BY 4.0).
Commonwealth of Australia Constitution Act, sections 13 and 28.
Jennings, W. and Wlezien, C. (2018). Election polling errors across time and space. Nature Human Behaviour, 2(4), 276–283.
Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. Journal of the Royal Statistical Society: Series B, 36(2), 111–147.
Figures are as of 15 October 2026, on polls to 11 October. The site's pages update with each new poll.
How to cite
Polling Forecast (2026). Forecasting Australia's House from imperfect polls: an overview. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/australia-overview. Live figures change as new polls arrive; the date above is when the text was last revised.



