Australia Next federal electionby 20 May 2028in 546 days Hung parliament 96% Labor majority 1% Calendar Track record Research
Research · Australia

The division model and its tests

Abstract. Every 2025 candidate's first preferences move by half a uniform and half a proportional swing of their group, with half of any state swing shown by two or more pollsters, and the count is replayed; where Labor and the Coalition are the last two, the 2PP swing decides. We set out the formula, its effect on a surging One Nation, a worked example in Dickson, and the tests: 133 of 150 winners from the notional 2022 results with the true 2025 vote, and 784 of 887 replaying each election from the one before since 2004.

Polling Forecast · updated 15 October 2026 · 7 min read · Australia, divisions, swing model, uniform swing, proportional swing, notional results, backtest, state polls, Dickson, classic seats

Summary

The House is won division by division, and in October 2026 most divisions no longer look like a contest between Labor and the Coalition. The forecast therefore works at the level of the candidate. It starts from each division's 2025 first preferences, moves every candidate with their group's national change since 2025 (half a uniform swing and half a proportional one, with the uniform half never falling below the proportional figure), adds half of any state swing shown by two or more pollsters, and replays the count with the preference weights of From primaries to seats: preference flows and the 2PP. Where the last two are Labor and the Coalition, the winner comes from the division's 2025 2PP plus the national 2PP swing. Tested from the AEC's notional 2022 results on the 2025 boundaries, with the true 2025 national vote, the model picks 133 of 150 winners; replayed election to election from 2004 to 2022 on the actual boundaries, 784 of 887.

The model

Inputs

For each of the 150 divisions the model holds every 2025 candidate's share of the formal first-preference vote, their party and preference group, the division's 2025 2PP (Labor against the Coalition, counted by the AEC in every division), its 2025 final two, and its formal vote. The base national share of each poll group, \(n_g\), is the formal-vote-weighted mean of the division shares: Labor 34.57, the Coalition 31.82, the Greens 12.20, One Nation 6.40 and others 15.01. Independents and minor parties all move with the polls' "others".

Moving each candidate

Let \(c\) be a candidate's 2025 share, \(n\) the base national share of their group, and \(t\) the group's national share today plus any state adjustment (below), floored at zero. The uniform and proportional moves are

\[ U = c + t - n, \qquad P = c\,\frac{t}{n}, \tag{1} \]

and the projected share, before rescaling, is

\[ x = \tfrac{1}{2}\max(U, P) + \tfrac{1}{2}\,P, \tag{2} \]

floored at zero. Every division's shares are then rescaled to sum to 100. Equation (2) is half a uniform and half a proportional swing, with the uniform half never below the proportional figure, so that a falling group's small candidates are not wiped out.

Since \(U - P = (c - n)(1 - t/n)\), the maximum has a simple reading:

  • For a rising group (\(t > n\)), \(U \ge P\) exactly where the candidate was below the group's national average. There the move is half uniform, half proportional. Where the candidate was above the national average, the move is proportional.
  • For a falling group (\(t < n\)), the opposite: half and half where the candidate was strong, proportional where weak.

A group polling at 5% or more nationally with no candidate in a division is given one, with a share of \(\tfrac{1}{2}\max(0,\ t - n)\): the uniform half of its swing from nothing.

What the choice of swing does

One Nation has gone from 6.4% in 2025 to 26.5% in today's average, a ratio of 4.14. The three rules give very different answers for its candidates:

One Nation's 2025 share in the division Uniform \(U\) Proportional \(P\) Model, equation (2)
2.0 22.1 8.3 15.2
4.0 24.1 16.6 20.3
6.4 26.5 26.5 26.5
10.0 30.1 41.4 41.4

Table: One Nation's projected share before rescaling, with no state adjustment, for national shares of 6.40 (base) and 26.5 (today).

A uniform swing hands One Nation 20 points everywhere, including divisions where it barely stood; a proportional swing multiplies it fourfold where it was strong and leaves it weak where it was weak. The model sits between them where One Nation was weak and follows the proportional rule where it was strong. For Labor (34.6 to 27.7, a ratio of 0.80) a candidate on 20% falls to 16.0 (proportional), a candidate on 50% to 41.6 (half of 43.1 uniform and half of 40.0 proportional).

State swings

Pollsters publish state breakdowns of their national samples. Where at least two pollsters have published one for a state, the model takes each pollster's latest state figures within 90 days of the newest state poll for that state, averages them, and computes the state's swing since 2025 beyond the national swing. Half of it is added to the target \(t\) for every candidate in that state:

\[ a_{s,g} = \tfrac{1}{2}\Big[\big(\bar{y}_{s,g} - v^{2025}_{s,g}\big) - \big(y_g - v^{2025}_g\big)\Big]. \tag{3} \]

The half is there because state samples are small. On 15 October 2026 the adjustments were:

State Pollsters Latest Labor Coalition One Nation Greens Others
New South Wales Newspoll, YouGov 21 Sep 2026 −0.4 −1.3 +0.7 +0.5 +0.5
Victoria Newspoll, YouGov 21 Sep 2026 −0.3 −0.1 −0.2 +0.7 −0.1
Queensland Newspoll, YouGov 21 Sep 2026 +1.5 −1.8 +1.8 −1.1 −0.4
Western Australia Newspoll, YouGov 21 Sep 2026 +1.7 +0.2 +0.1 −1.7 −0.3
South Australia Newspoll, YouGov 21 Sep 2026 −0.8 −0.6 +2.9 −2.0 +0.4
Tasmania DemosAU, EMRS, Roy Morgan 3 Mar 2026 +0.5 +1.9 −1.9 +0.8 −1.3

Table: State adjustments \(a_{s,g}\) in points, equation (3). The ACT and the Northern Territory have one pollster each and get none.

The count and the classic seats

The projected first preferences are counted as in From primaries to seats: preference flows and the 2PP: the lowest candidate is excluded and their votes pass to the others by preference weight, until two remain. If the two are Labor and the Coalition (a "classic" seat), the winner is decided by the division's 2PP instead:

\[ T_d = T_d^{2025} + \big(T - T^{2025}\big) + \Delta T_s, \tag{4} \]

where \(T\) is today's national 2PP from the primaries, \(T^{2025} = 55.22\), and \(\Delta T_s\) is the extra swing in the state: the national 2PP recomputed with the state adjustments added to the national primaries, minus \(T\). Labor wins if \(T_d \ge 50\). Otherwise the count decides. Equation (4) keeps the oldest and most tested piece of Australian election analysis, the uniform 2PP swing, where it still applies.

A worked example: Dickson

Dickson, in Queensland, was won from Peter Dutton by Labor's Ali France in 2025. Queensland's state adjustment adds 1.78 to One Nation's national 26.50, so \(t = 28.28\) against a base of \(n = 6.40\). One Nation's 2025 candidate had 4.19%:

\[ U = 4.19 + 28.28 - 6.40 = 26.07, \qquad P = 4.19 \times \frac{28.28}{6.40} = 18.51, \qquad x = \tfrac{1}{2}(26.07) + \tfrac{1}{2}(18.51) = 22.29. \]

Doing the same for every candidate gives shares summing to 94.52, and dividing by that sum gives the projected first preferences:

Candidate (2025) 2025 Projected
Labor (Ali France) 33.6 30.0
One Nation 4.2 23.6
LNP (Peter Dutton) 34.7 23.3
Independent (Ellie Smith) 12.2 9.3
Greens 7.6 8.0
Legalise Cannabis 2.8 2.1
Family First 2.2 1.7
Trumpet of Patriots 1.8 1.4
Animal Justice 0.9 0.7

Table: Dickson, first preferences, 2025 and projected, %.

The count excludes the four minor candidates, then the Greens, whose votes go mostly to Labor and the independent, then the independent, whose 14.5% splits between Labor (6.0), One Nation (4.7) and the LNP (3.8). Before the last exclusion Labor has 40.5, One Nation 30.9 and the LNP 28.6. The LNP is excluded, its votes go 77:23 to One Nation, and One Nation wins 53.0 to 47.0. The simulation gives One Nation 54% here, Labor 41% and the Coalition 5%: Dickson is a toss-up because the order of the second and third candidates turns on 0.3 points. Had the LNP stayed ahead of One Nation, the final two would have been Labor and the LNP and equation (4) would have decided it, with Labor's projected 2PP there above 50.

Today's projection by state

State Seats One Nation Labor Coalition Independents Greens Other
New South Wales 46 17 16 6 6 1 0
Victoria 38 13 16 4 3 2 0
Queensland 30 20 7 2 0 1 0
Western Australia 16 9 6 0 1 0 0
South Australia 10 2 6 1 0 0 1
Tasmania 5 1 3 0 1 0 0
ACT 3 0 3 0 0 0 0
Northern Territory 2 2 0 0 0 0 0
Total 150 64 57 13 11 4 1

Table: Projected seats, 15 October 2026. "Other" is Centre Alliance's Mayo.

71 divisions change hands on the projection compared with the 2025 result. Only 20 end as Labor against the Coalition; 88 end as Labor against One Nation.

The tests

2022 to 2025 on the 2025 boundaries

The AEC published notional 2022 results for the 2025 boundaries, which allows a clean test: move the notional 2022 first preferences by the national change to 2025, count with the 2022 preference weights only, and compare with the real 2025 winners.

Input Winners right (of 150)
The actual 2025 national vote; 2PP swing in classic seats 133
The actual 2025 national vote; the count alone in every seat 131
The actual 2025 national and state votes; 2PP swing in classic seats 134
The 2025 final-poll average as published 128
The final-poll average and the 2022–25 state polls (5 states) 127
The final-poll average, corrected as the history to 2022 advised 128

Table: The division model tested 2022 → 2025.

Three things follow. With perfect national figures the model gets 133 winners right; the 17 it misses are the division model's own error. The 2PP swing in classic seats helps (133 against 131), so it is kept. And state swings help when they are right (134 with the true state results) but the 2022–25 state polls did not (127 against 128); the state swing is kept as a halved adjustment, and that choice is one of the open items in What we don't know yet: limitations of the Australian forecast.

The seats from the final polls (Labor 80 against a real 94) show the national polling miss, not the division model: the 2025 polls understated Labor by 2.7 points.

Election to election, 2004–2022

The same model replayed each election from the one before on the actual boundaries, in every division with the same AEC identifier at both elections, so that boundary changes count against it. Each column changes one choice:

Elections Our model Final polls Count alone All proportional ¼ uniform ¾ uniform All uniform Pooled weights
2004 → 2007 132/149 131 131 132 132 132 132 –
2007 → 2010 127/147 126 128 127 127 127 127 128
2010 → 2013 132/150 132 129 132 132 132 131 132
2013 → 2016 132/149 132 131 132 132 132 132 132
2016 → 2019 132/143 119 132 132 132 132 132 134
2019 → 2022 129/149 131 128 129 129 129 129 129
Total 784/887 771 779 784 784 784 783

Table: Winners right, scored on the grouping that won. "Final polls" uses the final-poll average in place of the true change (parties the polls did not name keep the true change); "count alone" drops the 2PP swing in classic seats; the uniform columns change the weight on the uniform half; "pooled weights" fits the preference weights on every earlier election's exclusions together.

The model gets 88% of winners right from one election to the next with the true national change. The 2PP swing in classic seats adds five winners over six elections (784 against 779). The share of the swing taken as uniform made at most one seat's difference in any year, because no party's vote moved far: the test cannot tell the swing rules apart, and One Nation's rise is the case where they differ most. The final polls cost 13 winners net: 2016 → 2019 alone lost 13, 2019 → 2022 gained two, and two earlier pairs lost one each. Pooled preference weights gained 3 over five pairs; the reason we do not pool is in From primaries to seats: preference flows and the 2PP.

What the division model leaves out

People. It keeps the 2025 field and moves parties. Members who have changed party since 2025 are not modelled: Barnaby Joyce (New England) now sits for One Nation, and Zali Steggall and Allegra Spender for Community Strong Australia; their 2025 votes stay with the party they stood for then. New independents, retirements and by-election winners (One Nation's David Farley in Farrer, May 2026) are not foreseen.

Boundaries. The 2025 divisions are used. Redistributions in South Australia, Tasmania and the ACT, and in Queensland, are under way; each state keeps its number of seats, but notional results for the new boundaries are not yet in the data.

Local swing. Every candidate in a state moves by the same rule. Real swings vary by division, and the simulation adds random division-level noise to stand in for that (Simulating Australian polling error: where the chances come from).

References

Australian Electoral Commission, House of Representatives results by division and candidate, 2004–2025, and notional 2022 results on the 2025 boundaries (CC BY 4.0).

Figures are as of 15 October 2026. The site's pages update with each new poll.

How to cite

Polling Forecast (2026). The division model and its tests. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/australia-divisions. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like