The 2026 Victorian state election tools use the same modelling as the federal system, just recalibrated on the 2022 state election data. This article covers only what's different for Victoria – see the main Methodology page for how the underlying model works.
One Nation growth
One Nation only ran a handful of Legislative Assembly candidates (Victoria's lower house) in the 2022 state election, recording a statewide vote of just 0.28%. Accordingly, One Nation's primary vote was not individually reported in Victorian public opinion polls until January 2026, when their national support had begun climbing significantly.
Given the expectation that One Nation will run in every Legislative Assembly seat in the November 2026 state election, and also receive a much higher statewide primary vote, a couple of manual adjustments to the model are required.
Firstly, a new "Ensure major party candidates in all seats" checkbox has been added (ticked by default) to automatically add a Labor, Coalition, Greens or One Nation candidate to any seat where there was not one in the 2022 state election. While worded in a party-neutral way, this actually means adding 83 generic One Nation candidates, and 1 generic Labor candidate in Narracan (more on this shortly).
Secondly, a plain rolling poll average doesn't handle the transition period into One Nation having a reported primary vote very well. To handle this, for every date before One Nation's first tracked poll, the model estimates its support as whichever is larger: the 2022 statewide primary vote (0.28%), or the gap left over when the other four rolling averages (Labor, Coalition, Greens and Other) don't sum to 100%. Because those four series are smoothed independently of each other, that gap stays close to zero for as long as no real 2026 polling is near enough to influence them – and only starts growing once genuine One Nation polling data enters the averaging window.
The result is a smoother ramp into the first real reading (though likely still steeper than reality), rather than a flat line at zero that suddenly jumps.
Narracan supplementary election
Just days before the November 2022 state election, the Nationals candidate in the district of Narracan passed away, and the election for that seat had to be re-held a few months later. In that supplementary election, Labor did not field a candidate, which means that they start from an unrealistic base of 0%. The delayed timing also may reflect a different political context in which the vote was held.
This is a known limitation, and the result in this seat should be viewed cautiously. As noted earlier, a generic Labor candidate can be added to this seat with the "Ensure major party candidates in all seats" checkbox, but this candidate will remain at 0% in the model unless Labor receives a positive statewide primary vote swing to them.
Manual preference flow overrides
The VIC 2026 preference model draws first on 133 observed flow combinations from the actual 2022 state election preference distributions, then falls back to nearest-match and ideological-bloc rules exactly as described on the main Methodology page.
Layered on top of that is a set of 24 manual preference flow overrides from the 2025 federal and 2026 SA state elections. These overrides are enabled by default, but can be switched off in the "Edit preference flows" dropdown on Model an Election, or the "Apply federal 2025 / SA 2026 calibrations" checkbox on Track a Poll.
18 of the manual override flows are from the federal 2025 dataset, where there is simply a greater abundance of data from almost twice the number of Lower House seats; the remaining 6 are from the SA 2026 dataset, which are mostly for filling in new likely scenarios containing One Nation. The latter are the same as applied in the federal model, and are listed in full on the main Methodology page.
The 18 manually overridden flows drawn from the federal 2025 dataset are:
- Greens eliminated, [Coalition; Labor; Other, Left-leaning] remaining: 50.3% to Other, Left-leaning, 42.4% to Labor, 7.3% to Coalition
- Greens eliminated, [Coalition; Labor; One Nation] remaining: 69.4% to Labor, 15.4% to Coalition, 15.2% to One Nation
- Labor eliminated, [Coalition; Other, Left-leaning] remaining: 75.8% to Other, Left-leaning, 24.2% to Coalition
- One Nation eliminated, [Coalition; Labor] remaining: 69.9% to Coalition, 30.1% to Labor
- One Nation eliminated, [Coalition; Greens; Labor] remaining: 61.9% to Coalition, 19.9% to Greens, 18.2% to Labor
- One Nation eliminated, [Coalition; Greens; Labor; Other, Left-leaning] remaining: 56.7% to Coalition, 25.8% to Other, Left-leaning, 10.0% to Labor, 7.6% to Greens
- Other, Left-leaning eliminated, [Coalition; Greens; Labor; One Nation] remaining: 35.8% to Greens, 26.0% to Labor, 23.3% to One Nation, 14.9% to Coalition
- Other, Left-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning] remaining: 27.0% to Greens, 24.0% to Other, Left-leaning, 19.2% to Labor, 17.8% to One Nation, 12.0% to Coalition
- Other, Left-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Right-leaning] remaining: 33.4% to Greens, 23.3% to Labor, 17.0% to One Nation, 15.6% to Other, Right-leaning, 10.7% to Coalition
- Other, Left-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Right-leaning; Other, Right-leaning] remaining: 35.3% to Greens, 26.9% to Other, Right-leaning, 19.7% to Labor, 9.8% to One Nation, 8.3% to Coalition
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation] remaining: 48.4% to One Nation, 21.1% to Coalition, 15.7% to Greens, 14.8% to Labor
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning] remaining: 43.1% to One Nation, 20.2% to Coalition, 16.6% to Other, Left-leaning, 10.9% to Labor, 9.1% to Greens
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Right-leaning] remaining: 37.1% to One Nation, 22.1% to Other, Right-leaning, 17.8% to Coalition, 12.2% to Labor, 10.8% to Greens
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning; Other, Right-leaning] remaining: 35.8% to One Nation, 17.4% to Coalition, 16.4% to Other, Right-leaning, 13.2% to Other, Left-leaning, 9.6% to Labor, 7.6% to Greens
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Right-leaning; Other, Right-leaning] remaining: 37.1% to Other, Right-leaning, 25.7% to One Nation, 16.2% to Coalition, 10.7% to Greens, 10.2% to Labor
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning; Other, Left-leaning] remaining: 38.6% to One Nation, 26.1% to Other, Left-leaning, 20.5% to Coalition, 9.4% to Labor, 5.4% to Greens
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning; Other, Right-leaning; Other, Right-leaning] remaining: 31.2% to Other, Right-leaning, 25.3% to One Nation, 15.2% to Coalition, 11.2% to Other, Left-leaning, 9.5% to Labor, 7.6% to Greens
- Other, Right-leaning eliminated, [Coalition; Greens; Labor; One Nation; Other, Left-leaning; Other, Left-leaning; Other, Right-leaning] remaining: 38.5% to One Nation, 22.0% to Other, Left-leaning, 15.6% to Coalition, 13.2% to Other, Right-leaning, 6.4% to Labor, 4.3% to Greens
None of the 24 overrides displace a qualifying observation from the 2022 Victorian dataset with n ≥ 5 instances.
Ecological modelling for age/sex demographics
The federal model overrides the ecological regression estimates for age and sex specifically with rates drawn from the Australian Election Study (AES), a large post-election survey – because ecological regression is known to be less reliable for age and sex than for other Census variables, due to compositional confounding (age and sex correlate with almost everything else at the booth level).
There is no Victorian state-election equivalent of the AES. Therefore, the Victorian model uses booth-level ecological regression for age and sex too, the same method as every other demographic variable. Treat age- and sex-based swing estimates with significantly more caution than the federal tool's equivalent figures for this reason.
The 2022 backtest
Running the preference distribution model back over the 88 Legislative Assembly districts from the 2022 state election demonstrates the internal consistency of the model – i.e. when trained on 2022 primary vote data, that reliably converts those values into realistic preferences that correctly predict almost every seat.
| Metric | Result |
|---|---|
| Correct final 2CP pair identified | 86 / 88 seats (97.7%) |
| Correct winner within correctly paired seats | 85 / 86 (98.8%) |
| Correct winner overall | 87 / 88 seats (98.9%) |
| 2CP within ±1pp of actual result | 51 / 86 correctly-paired seats (59%) |
| 2CP within ±2pp of actual result | 76 / 86 correctly-paired seats (88%) |
| 2CP within ±5pp of actual result | 86 / 86 correctly-paired seats (100%) |
| Mean absolute error (2CP) | 1.05pp |
| Systematic bias (2CP) | −0.41pp (model slightly underestimates winner's margin) |
In the 2022 backtest, High-confidence simulations got the correct pair and winner in all 55 cases, and Medium-confidence simulations got the pair and winner right in all 23. Low-confidence simulations got the pair right in 8 of 10 and the winner in 9 of 10 – exactly the behaviour the reliability rating is designed to produce, with every miss falling in the Low tier.
The two wrong pairs (Hawthorn, Kew) are both inner-metropolitan Liberal-held seats with a crowded field of Other, Left-leaning independents: the model gets the eventual Coalition winner right in both, but eliminates Labor a round before the independent vote actually consolidates – the same failure pattern as Cowper and Forrest in the federal backtest.
The one wrong winner within a correctly-paired seat is Bass, which was highly marginal – Labor won by just 202 votes in reality, but the model predicts a Coalition win.
