As the Australian federal election approaches, political polls are coming thick and fast. This page will be regularly updated so you can track who is polling up, how the independents are faring and how the parties stand with different demographics.
This first chart is based on a poll averaging model developed by political scientists at the University of Sydney. It factors in sample sizes, previous results and “house effects” (bias towards a party) of each pollster.
While the first chart shows the estimated state of play when our model was last run, the next shows a timeline of the two-party preferred (2pp) vote since 2022:
Looking at two-party support alone can obscure one of the biggest stories of the last election: almost a third of votes were for independents and other parties.
The 68.5% primary votes share for Labor and the Coalition is an all time low and the continuation of a steady decline since the two parties claimed 98% of votes in 1951.
The chart below shows the primary votes for Labor, the Coalition, Greens and others/independents. It is based on the same model as our main tracker, starting with the vote share at the last election. Use the drop-down menu to see what has changed over different periods since the election.
To get a sample that reflects the nation at large, pollsters collect a lot of demographic information, including age, sex, location and education.
Polling companies occasionally release two-party preferred measures for these sub-demographics.
The following charts use simple rolling averages to try to find the underlying trend in two-party support. There has been no adjustment for sample size, house effects, weighting or release date.
The first shows support by the age group of the respondent.
The next chart groups respondents by education – those with no tertiary education, those with a Tafe or technical education, and those with university education.
The chart below groups respondents by sex – male or female. As the numbers are rolling averages, they will not always add up to 100.
The final demographic category is state. Data is not available for all states, largely because of their size. Tasmania, for instance, makes up about 2% of the population. A representative sample of 1,000 Australians would have far too few Tasmanians to provide a robust estimate.
The final table shows the two-party preferred share for all of the polls that feed into our models.
Notes and methods
The main poll tracker is based on work by Dr Luke Mansillo and Prof Simon Jackman. You can find their paper here.
The model in different disciplines is called a hidden Markov model or a state space model and employs a Kalman filter algorithm that uses a series of measurements over time, including statistical noise and other inaccuracies, to produce estimates.
These types of models are often used in fields such as robotics, economics and medicine to create estimates from noisy measurements.
Each newly published poll is treated as a new measurement, with the model factoring in new data in the context of what has come before.
The model begins with (is anchored to) the vote share for each party at the last federal election.
Only polls with a defined sampling procedure, reported sample size and fielding dates have been included in our dataset.
Sample sizes are adjusted to account for non response.
The model calculates house effects for each pollster dynamically, by finding systematic differences to what would be expected, given the current average.
The two-party preferred vote is adjusted to remove unknowns or non responses, leaving only Labor and Coalition shares.
The model is run 1,000 times for each update.
