Who wins
how many
Median seats in each party's simulations, with the range covering the middle 80%. The dashed line is the 47 seats needed for a majority in the 93-seat Assembly.
All 93,
one tile each
Grouped by region. The colour is the favourite; a stronger colour means a surer favourite. Select a tile to find the seat in the table.
Seats on the edge
All 93 as a table →Every seat,
every party
Projected primary vote share in each Legislative Assembly electorate. “Other” is everyone else on the ballot: Shooters, independents and minor parties. A party with a blank cell is not expected to stand. Sort by name or by any party's vote.
Two forecasts,
same dice
Every forecast here comes in two versions. Polls only uses the polls, past election results, the candidates standing, preference flows and each electorate's history. Polls + context uses all of that plus a small amount of information about the economy and the federal political climate.
Both versions run the same 50,000 simulated elections with the same random numbers. Any difference between them comes from context, not chance.
Why include the economy at all?
Voters judge governments partly on how life is going. Studies of Australian state elections, covering nearly two hundred contests, find that economic conditions, both nationally and in the state relative to the rest of the country, shift incumbent governments' chances of re-election. Research on federal elections finds similar, modest effects.
The federal government matters too. Since the 1960s, state results have tended to run against whichever party holds office in Canberra. A state government that shares its party with an unpopular federal government often pays for it.
Why so cautious?
The same research is clear that these effects are small and noisy. Once you account for what voters already think, objective economic data adds little, and some popular beliefs (for instance that interest-rate cuts reward governments) don't hold up in the evidence. So the context adjustment is:
- Small and capped. It can nudge a forecast but never overturn it.
- Uncertain by design. Each simulation draws its own strength for the effect, so context widens the range of outcomes as well as shifting it.
- A prior, not a poll. It adjusts the starting expectation, never the polls themselves.
- Fading. As more good polling arrives, polls take over and context fades towards zero. Close to election day it usually moves the result by only a few tenths of a point.
The specific indicators and how much each counts are not published.
What context changed
The Council
21 of the 42 Legislative Council seats are up in 2027; the other 21 members continue. The top row shows the continuing members and the bottom row the median forecast for 2027. This is a group-level quota approximation, not a full replica of the official count.
The polls
The weighted statewide poll average each week, with a band covering the middle 80% of plausible values. Dots are individual polls. Polls missing a sample size, dates or question wording are listed but left out of the average.
All polls in this forecast
How the polls become a forecast
The forecast is built from polls alone. It blends the latest polls with the average of the past year of polling, leaning further towards the latest polls the more good, recent polling there is. Past election results do not pull the statewide forecast; they decide where each party's vote falls seat by seat.
Which polls count
Published statewide voting-intention polls for the Legislative Assembly. A poll needs a sample size, fieldwork dates and a published question so it can be checked; polls missing any of these are listed in the poll table with the reason, but left out of the average. Polls more than a year old are ignored.
How much each poll counts
- Recency. A poll's weight fades smoothly with age, and fades faster in the final six months of the campaign, when opinion is settling.
- Disclosure. Pollsters that meet the Australian Polling Council's full disclosure standard count most; pollsters that publish complete methods without that standard count a little less; polls with only partial methods count much less and are flagged.
- Who paid. A poll commissioned by a party or an interest group is discounted unless it is released in full.
- Sample size. Bigger samples have smaller sampling error, and that is the only way sample size enters. Every poll also carries a floor of non-sampling error, so one very large poll cannot swamp the rest.
- One pollster is one pollster. That error floor is shared across all of a pollster's polls, so ten polls from one firm do not count as ten independent views.
- House effects. Once a pollster has several usable polls, its consistent lean relative to the other pollsters is estimated, cautiously, and removed before averaging.
Recent polls and the past year
- The baseline is the past year of polling, averaged with a long memory, so it reflects where opinion has settled rather than the latest swing. Past election results are used only if there is no polling at all.
- The share given to the latest polls grows with the amount of good, recent polling and with how precise that polling is. The rest goes to the past-year baseline, so a single unusual poll cannot swing the forecast.
- A gap in polling does not erase the polls. If weeks pass with no new poll, the last polls keep their relative weight; the time since then is treated as extra uncertainty.
- Time to election day matters. The further away the election, the more opinion can still move, so the latest polls are a less certain guide and the past-year baseline keeps more weight.
- Every simulation adds error: sampling error, pollster error, movement before election day and a statewide polling miss shared by every seat.
- Big shifts carry big uncertainty. A party far from its last election result, as One Nation is now, gets wider error in both directions. In past One Nation surges the final polls have been about four points out, sometimes too high and sometimes too low, so the model widens the range rather than pulling the forecast back towards the old result.
From statewide vote to seats
Each seat starts from its 2023 result and moves with the statewide forecast. Where a party had no 2023 candidate, the seat starts from that party's 2023 Legislative Council vote in the same district, which every voter cast, so One Nation's local strength comes from real votes rather than a regional average. Until nominations close, each party's chance of standing in a seat is simulated; One Nation is expected to contest almost every seat.
Where the forecast differs from the latest polls, it is because the past year of polling differs from the latest polls, not because of past elections. As new polls arrive, the forecast follows them.
Take the data
Both files include both models: polls only, and polls + context.
Statewide top line
Chance of a majority in the Legislative Assembly, chance of having the most seats, projected seats (median and 80% range), statewide primary vote, and Council totals.
Seat by seat
Projected primary vote share for ALP, Liberal, National, Greens, One Nation and Other in all 93 electorates, plus each seat's favourite and its chance of winning.
How it works
- Start from the map. Every seat begins from its 2023 primary vote on the current boundaries, which says where each party is strong; the statewide level comes from the polls.
- Read the polls. Polls are averaged with more weight on recent, larger and better-documented polls, adjusted for each pollster's habitual lean, and steadied by the average of the past year of polling. How the blend works →
- Simulate 50,000 elections. Each simulation draws a statewide result, then regional, neighbouring-seat and local swings, so seats move together the way real seats do.
- Count preferences. NSW uses optional preferential voting. Each simulation eliminates candidates one by one, passes on their preferences and lets some ballots exhaust, so the final two can be anyone, including One Nation, Greens or independents.
- Count the outcomes. A probability is the share of simulations in which something happens. Odds are fair odds with no bookmaker margin.
Polls are noisy and every model is a simplification. A 25% chance happens one time in four.