PoliCast predicts: Labor Minority Government
Dr Sam Vaughan
Intro
At long last, we have a date for the federal election: May 3rd. For most of the run-up to the election, poll-watchers told us that the race would be a tight one, with no clear winner in the two-party preferred (2PP) contest between Labor and the Coalition. What is clear is that the Coalition has gained significant ground since 2022, so the contest is likely to be a lot closer this time.
However, commentary is largely confined to the overall popularity contest of “who would you vote for out of the two parties”, usually at a national level. In reality, which party forms the next government is the result of 150 individual contests, most of which involve preference flows to determine the winner. To get a majority, increasing your vote isn’t enough – you have to flip seats and get to the magic number of 761.
Here we present a new model, PoliCast, that aims to convert polling numbers into realistic seat outcomes. This allows us to investigate the likely makeup of the next parliament if an election were to be held today. Based on the polling up to 4th of April 2025, we predict the most likely outcome if the election were held today would be a hung parliament and a Labor minority government. There are, of course, uncertainties in our model, but it is very hard to see either party reaching an absolute majority; the ALP’s sagging support does not translate into enough Coalition gains.
Below we outline how our model works, dive into the predictions, and suggest some seats to focus on.
How the model works
Our model PoliCast is almost entirely driven by statistics. By this, we mean we have tried to reduce the assumptions to a minimum and rely on historical data to determine the likely correlation between opinion polls and the results in any given electorate. It works by translating swings towards or against parties at a national and state level down to party votes at individual voting booths in each electorate. This allows us to turn an opinion poll into a series of likely outcomes for every single seat.
Our model concentrates on the three largest parties in Australian politics: the Labor Party, the Liberal Party2 and the Greens. We use historical election results to see how the vote for these parties changes at every polling booth in the country at each election. This data allows us to measure a degree of similarity between a booth and every other booth, essentially asking questions like “if we see a large swing towards the ALP at booth number 1, how likely are we to see a large swing against the Liberal Party at booth 2?”. We place this information into a “covariance matrix”, ending up with over 400 million data-points characterising the connections between polling centres across the country.
Importantly, we can also measure the degree of correlation between a given polling booth and the overall election results nationally and in each state. This is crucial, because opinion polls give us a measurement of the expected national and state-level results if an election were held on that day. Our model uses these polls as an input to generate a realistic series of simulated results for every electoral booth based on these ‘birds-eye-view’ polling numbers. In technical terms, we use the joint probability distribution of polling booths, state polls and national polls to find the conditional distribution of the polling booths, given a set of state and national poll results.
We then combine electoral booths together to work out the number of likely first-preference votes in every seat. Once we have these, we use historical preference flows from the 2022 election to see which candidate would be elected.
The correlation, say, between the Greens vote at a booth in Fitzroy and the Nationals vote in Port Hedland is probably not very strong. This uncertainty is captured in our model, so when the national polls swing +1% towards the Coalition, this might result in a 2% ± 1.5% (1 sigma) result for the Liberal candidate in Goldstein. This means that, based on the poll, we believe that 66% of the time the swing will be within 1.5 percentage points of +2% – sometimes higher, sometimes lower. We sample every booth in the country 4000 times, randomly selecting a primary vote for each party according to the expected swings and the errors. This gives us 4000 distinct parliaments to look at. It’s from this set of hypothetical parliaments we can draw our predictions, and also understand the uncertainties in the model.
For the moment we’ll be assuming that every previous party or independent stands again in a given seat. Once we have a list of people running, we’ll be able to update the model more fully and include realistic preference flows based on how-to-vote cards from the major parties.
Inputs to the model
As inputs to the model, we use publicly available polls from a variety of sources. In particular, the excellent compilation data at the Poll Bludger has been invaluable.
We combine the polls together to create a single estimate of the true support for the ALP, Coalition and Greens at any point in time, using methodology inspired by the work of Prof. Simon Jackson in his 2005 research paper. You can find a full write up of how we aggregate different polls together in a previous PollIQ article.
Results
Perhaps the most interesting result from our recent simulations is that for now the era of majority government seems to be over. Out of 4000 simulations, just over 5% resulted in majority government – 5% for the ALP and only 0.03% of the time for Coalition. This means that there is a 95% chance of a hung parliament, based on the current polling numbers. Whether Mr Albanese or Mr Dutton have the confidence of the House will almost certainly come down to the cross bench.
The ALP end up being the largest party in 93% of our simulations, while the Coalition are the largest party 4.5% of the time. In 2.5% of simulations, the two parties are tied.
We can also see the likely size of the Labor and Coalition party rooms.
- 73% of the time, the ALP is within 5 seats of a majority.
- 2% of the time, the Coalition is within 5 seats of a majority.
- 16% of the time, the ALP is within 2 seats of a majority.
- 0.5% of the time, the Coalition is within 2 seats of a majority.
And this figure gives the Parliament which occurs most often in our simulations (although note that this exact make-up only occurs 7% of the time):
One advantage the ALP have is the fact that the Coalition don’t have a natural other party to form a government with. Anthony Albanese, on the other hand, could try and form a centre-left coalition with the Greens.
In fact, this might be the ALP’s best way to stay in the Lodge: the ALP and the Greens combined can form a government in 35% of our simulations. Even then, however, there is still likely to be a key role to play for the Independents: 65% of the time neither the Coalition or the ALP plus the Greens have over 75 seats.
Caveats
The biggest weakness of our model – and all statistical models and polls, really – is that they cannot account for major changes in the circumstances of an individual seat3. This can include changes in candidates, the scale of a campaign, and even the money spent by different political groups. If an incumbent is arrested during the campaign or Trumpet of Patriots recruits Chris Hemsworth to make a play for the seat, our covariance matrix is not going to pick up these new dynamics. We call out a few seats below where this is likely to be an issue.
Similarly, the independents and other likely crossbenchers are the hardest to model reliably. They have less history behind them, so both their primary vote and preference flows are more uncertain. Seat level polls are also much less common- and less reliable- than state and national polls. These races depend heavily on the campaign and local issues. We can reliably predict the chances of a suburban Liberal holding her seat when there is a national swing toward the Coalition, but how easily can a Greens MP hold Brisbane? Win or lose, this one will be unprecedented.
Seats most likely to change hands
The chart below shows the 20 seats with the highest probability to change hands if an election were held today. The colour of the bar shows the incumbent party, and hovering over each bar gives the party most likely to benefit if the seat does flip. Note that Bullwinkel is a brand-new seat, so by definition will have a new incumbent 100% of the time.
Key Seats to watch for the ALP
It is almost certain that the ALP will lose the seats of Gilmore in NSW (where they have a majority of only 373 votes), Lingiari in the NT (866 votes) and Lyons in Tasmania (1334 votes) to the Coalition. This is already enough to push them below 75 seats and into a minority government.
The most likely outcome from our simulations also has them losing Bennelong (99% of the time), Paterson (90% of the time) and McEwen (77% of the time) to the Coalition. If things are going really badly for the ALP they could also lose Richmond to the Nationals and Macnamara to the Greens.
The most likely seat for the ALP to pick up is to retake Brisbane (from the Greens). Policast predicts that this happens with a 38% chance. They also have a 26% chance to take Moore in WA from the Liberal Party.
Key Seats to watch for the Coalition
Our model predicts that the Coalition has a 73% probability to pick up all six seats of Gilmore, Lingiari, Lyons, Bennelong, Paterson and McEwen. Next most likely on the list would be the Nationals picking up Richmond in NSW (which happens 73% of the time) and Parramatta in Sydney (18% of the time).
The seats they are most likely to lose are Bradfield and Moore, with a 26% and 25% probability respectively. Moore saw an 11% swing towards the ALP in 2022, and the overwhelming state support for the ALP in Western Australia might be making some federal Liberal MPs nervous. In Bradfield, The Liberal Party were run very close in 2022 by independent Nicolette Boele, who has confirmed she’s running again.
Key Seats to watch for the Greens
We predict that the Greens have a 36% chance to add to their current tally of 4 seats. The most likely pickups would be in Macnamara, Richmond, Cooper or Wills. This has to be balanced against the chances of them losing Brisbane or Ryan: Policast gives the ALP a 38% chance of retaking Brisbane, and a 7% chance to take back Griffith.
A good example of local conditions trumping statistics might be Wills. Our model predicts the Greens have a <1% chance of picking up Wills, though this isn’t our prediction. They are running a high-profile candidate (former mayor and state Greens leader Samantha Ratnam) and by all accounts are running a much more aggressive campaign than in 2022. The putative ~4.5% swing needed is thus more likely than the model predicts- a lot will depend on preference flows from Liberal voters.
Races to keep an eye on
Macnamara is a fascinating case. It’s currently held by the ALP with a 12.2% two-party preferred majority against the Liberal Party. Yet Policast gives it a 44% chance of turning Green!
In 2022, the seat was a three-way contest between the ALP, LP and the Greens- only 0.8% of the vote total separated the three parties in the penultimate round of voting. If the ALP gets knocked into third place, their preference flows are very likely to see the Greens candidate triumph in the two-candidate preferred contest.
- Or 75, and put an opportunist from another faction into the Speaker’s chair. See: Slipper, Peter. ↩︎
- Including the Country Liberal Party in the NT and the Liberal-National Party in Queensland ↩︎
- Or more globally – if Australia is invaded by Madagascar or there’s a global thermonuclear exchange in the next five weeks, our confidence in the model may decrease accordingly. ↩︎