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

Past-miss lenses: today's polls with each past error replayed

Abstract. Each lens takes today's poll average, removes one past House election's final-poll miss, and runs every division again. On the polls of 15 October 2026 none of the seven lenses, 2007 to 2025, gives any party a majority. A repeat of 2025's miss, which understated Labor, gives Labor 75 seats; a repeat of 2022's gives One Nation 71; the 2010 and 2019 misses double the Coalition's seats. 56 divisions change hands under at least one lens. We define the lenses, work through 2019, and compare them with the simulation.

Polling Forecast · updated 9 November 2026 · 6 min read · Australia, lenses, past polling misses, what-if, seats at risk, 2019, 2022, 2025, scenarios

Summary

The simulation says how wrong the polls might be in general. The lenses say what would happen if they were wrong in one particular way that has happened before. Each lens takes today's poll average, removes one past election's final-poll miss, rescales, and runs every division again through the same count. There are seven, one for each House election from 2007 to 2025 with final primaries. On the polls of 15 October 2026 none gives any party a majority. The 2025 miss, when the polls understated Labor by 2.7 points, comes closest: it gives Labor 75 seats and One Nation 49. The 2022 miss gives One Nation 71. The 2010 and 2019 misses, which understated the Coalition, double its seats. 56 of the 150 divisions change hands under at least one lens.

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.

  • The lenses on the map (November). The replays are also on the map as "The polls miss again", beside "You set the vote".

Why lenses

The simulation in Simulating Australian polling error: where the chances come from draws each group's national error independently and rescales. That gives every group its own uncertainty, but the errors it produces are random combinations, not ones that have happened. Real polling misses have a shape: in 2019 the polls overstated Labor and understated the Coalition at the same time; in 2025 the opposite. A lens replays one such shape in full.

Lenses answer a plain question a reader can check: who would win if the polls are wrong today the way they were wrong in 2019. They are not probabilities, they are not averaged into the simulation, and each is equally a "what if".

Definition

Let \(y_p\) be today's poll average for group \(p\) and \(e_{p,t}\) the error of the final-poll average at election \(t\) (How wrong have Australian polls been? Primaries, the 2PP and the λ test). The lens for election \(t\) is

\[ \tilde{y}^{(t)}_p = \begin{cases} \max\!\left(0.1,\ y_p - e_{p,t}\right) & \text{if } p \text{ was measured at } t, \\ y_p & \text{otherwise,} \end{cases} \qquad y^{(t)}_p = \tilde{y}^{(t)}_p \,\frac{\sum_q y_q}{\sum_q \tilde{y}^{(t)}_q}. \tag{1} \]

The lens starts from the polls as published, not from our projection; today the two are the same, because no correction is applied. A group not measured at election \(t\) (One Nation and others before 2019) keeps its polled figure before the rescale, so it moves only by the rescale. The lens shares then go through the whole division model, with today's state adjustments: projection of every candidate, the count, and the 2PP swing in classic seats (The division model and its tests).

The model uses no election before 2000, so the misses of 1987, 1990 and 1998 are not lenses. Until 28 September 2026 they were.

The seven lenses

Lens Labor Coalition One Nation Greens Others Labor 2PP
The polls today 27.7 21.4 26.5 13.2 11.3 52.2
Miss like 2007 27.2 20.7 26.7 14.1 11.3 52.7
Miss like 2010 26.1 24.9 27.1 10.3 11.5 48.4
Miss like 2013 27.6 22.8 26.7 11.6 11.3 50.8
Miss like 2016 28.0 21.6 26.7 12.4 11.3 52.0
Miss like 2019 25.3 24.3 25.5 13.5 11.5 50.0
Miss like 2022 25.3 21.7 26.6 13.2 13.3 50.9
Miss like 2025 30.4 19.1 24.6 12.2 13.7 55.0

Table: National primaries under each lens, equation (1), and Labor's 2PP on 2025 flows, %.

Lens One Nation Labor Coalition Independents Greens Other Divisions changed
The polls today 64 57 13 11 4 1 –
Miss like 2007 67 55 8 12 7 1 8
Miss like 2010 64 44 28 10 3 1 25
Miss like 2013 65 55 17 9 3 1 7
Miss like 2016 64 56 14 11 4 1 1
Miss like 2019 58 48 25 11 7 1 22
Miss like 2022 71 44 13 13 7 2 16
Miss like 2025 49 75 6 15 3 2 26

Table: Seats under each lens. "Other" is Centre Alliance, plus Katter's Australian Party where it wins. "Divisions changed" counts divisions whose winner differs from today's projection.

No lens reaches 76 for any party.

2025: Labor understated. The 2025 miss had Labor 2.7 points too low and the Coalition and One Nation too high. Replayed today it raises Labor to 30.4 and lowers One Nation to 24.6, and Labor wins 75 seats, one short of a majority. Of the 26 divisions that change, 15 go from One Nation to Labor. Its 2PP, 55.0, is almost exactly Labor's 2025 result.

2022: Labor overstated, others understated. The 2022 miss had Labor 2.4 points too high and others 2.0 too low. Replayed today it takes Labor down to 25.3, below One Nation, and One Nation wins 71 seats, seven of them from Labor.

2010 and 2019: the Coalition understated. Both misses had the Coalition about three points too low and Labor too high. Replayed today they lift the Coalition to 24 or 25%, close to One Nation, and the Coalition's seats double, to 28 and 25. In the 2010 lens the rising Coalition takes eight divisions from One Nation, while One Nation, with Labor and the Greens lower, takes eight from Labor.

2007, 2013 and 2016: small misses. These change 1 to 8 divisions. They do not measure One Nation and others, so they move only Labor, the Coalition and the Greens.

The lenses show a pattern that the simulation's ranges only summarise. Lowering Labor helps One Nation more than the Coalition (the 2022 lens: Labor −13, One Nation +7, Coalition 0). Raising the Coalition costs Labor at least as much as One Nation (2010: Labor −13, One Nation 0; 2019: Labor −9, One Nation −6). And the largest single change in today's House would come from a repeat of 2025.

A worked example: the 2019 lens

At the 2019 election the final-poll average had Labor 2.40 points too high, the Coalition 2.84 too low, the Greens 0.38 too low, One Nation 1.04 too high and others 0.22 too low. Taking those errors off today's average by equation (1):

Group Today 2019 error Lens
Labor 27.67 +2.40 25.27
Coalition 21.42 −2.84 24.26
One Nation 26.50 +1.04 25.46
Greens 13.17 −0.38 13.55
Others 11.25 −0.22 11.47

Table: The 2019 lens, %. The 2019 errors sum to almost zero, so the rescale changes nothing visible.

The three largest groups end within 1.2 points of each other. Run through every division, the lens gives One Nation 58 seats, Labor 48 and the Coalition 25, and changes the winner in 22 divisions. The largest movements are eight divisions from One Nation to the Coalition, four from Labor to One Nation; three each from Labor to the Coalition and from Labor to the Greens; and two from One Nation to Labor. A miss that lowers Labor and raises the Coalition does not simply move seats between those two: with three parties near 25%, it reshuffles who is excluded third in dozens of divisions.

Lenses beside the simulation

The seven lenses span a range of outcomes close to the simulation's 80% ranges (Simulating Australian polling error: where the chances come from):

Party Lowest lens Highest lens Simulation 80% range
One Nation 49 (2025) 71 (2022) 49–70
Labor 44 (2010, 2022) 75 (2025) 47–69
Coalition 6 (2025) 28 (2010) 8–24

Table: Seats under the lenses and in the simulation, 15 October 2026.

The lenses reach a little beyond the simulation's 80% range at both ends for Labor and the Coalition. That is what one would expect: each lens is a whole past miss, with the two largest parties' errors in opposite directions, while the simulation draws each group's error separately and the rescale only partly links them. The two views agree that no single past miss gives any party a majority, and that the outcome most sensitive to polling error is Labor's.

On the 2PP the lenses run from 48.4 (2010) to 55.0 (2025) for Labor, around today's 52.2. Under both the 2010 and the 2019 lens Labor's 2PP against the Coalition falls to 50 or below, but in neither does the Coalition come close to forming a government, because most divisions are not Labor against the Coalition: under the lenses only 14 to 26 divisions end as classic contests, where the 2PP decides.

The divisions at risk

Across the seven lenses, 94 divisions keep the same winner every time and 56 change under at least one:

Lenses under which the winner changes Divisions
0 94
1 27
2 13
3 13
4 2
5 1

Table: Divisions by the number of past misses that change their winner.

Deakin changes under five lenses; Aston and Robertson under four; Bean, Brisbane, Fisher, Goldstein, Griffith, Hughes, Hume, Lyons, Macarthur, Macnamara, Makin and Wannon among those changing under three. The map has a risk mode that colours every division by this count, and each division's page lists the lenses that would change it.

They are largely the divisions the simulation calls close: 46 of the 56 have a projected winner's chance under 75%, and all 19 divisions under 60% are among them. That is expected, because a division a past miss can tip is one where the projected winner's margin is small. The lenses add what the simulation cannot: which way each division tips under each real pattern of error.

The what-if tool

The what-if page starts from any lens and lets the reader move each party's error by hand. It sends the shares to the same model the site uses, returns the seats and the divisions that change, and reports the 2PP the shares imply and how many divisions end as Labor against the Coalition. Under today's polls that is 20 of 150; under the 2025 lens 16; under the 2010 lens 26.

What the lenses cannot show

Misses without precedent. Every lens is a miss of a size and shape already seen. No past miss involved a One Nation at 26%; the record of polling it at this size is empty. A miss on One Nation larger than 1.9 points, the 2025 figure, is not a lens.

Older elections. The 1987, 1990 and 1998 misses are not used, by the owner's rule. They were larger on the Coalition (−3.1, −4.0, +3.7) than any since.

Preferences. A lens moves primaries only. A miss in how preferences flow, which From primaries to seats: preference flows and the 2PP identifies as the weakest part of the model, is not a lens.

Probabilities. Seven lenses are seven scenarios. They say nothing about how likely each is; the simulation does that, with its own limits.

References

Australian Electoral Commission, House of Representatives results and distributions of preferences, 2004–2025 (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). Past-miss lenses: today's polls with each past error replayed. Retrieved 21 November 2026 from https://www.pollingforecast.com/research/australia-lenses. Live figures change as new polls arrive; the date above is when the text was last revised.

You may also like