With the next Australian federal election just around the corner, there are wide variety of different opinion polls out there trying to gauge which way the political winds are blowing. Knowing which to trust is tricky, however. Here at PolIQ we’ve built a statistical model to measure the support for the major parties by combining different polls together, making it easy for you to keep track of who’s fortunes are looking up and who isn’t faring so well. We’ll keep this page regularly updated as we get closer to election day.
Two Party Preferred: too close to call
Each marker point here represents a poll by a different polling organisation. The solid lines represent our best guess at the true level of support each party would receive if an election were called on a given date. The details of how we arrive at these values are explained at the bottom of this page.
We can see a sharp decline in two party preferred support for the ALP. As of November 2024, the two parties are neck and neck, a 2% swing against Labor since the 2022 election.
First Preference Voting Intention: a shift away from Labor
The first preference data shows the ALP’s post-election honeymoon period has well and truly worn off, with Anthony Albanese’s party suffering a 7% first-preference swing against them since their high point in January 2023. Peter Dutton’s Liberal/National coalition on the other hand have seen a steady increase in support, currently sitting at their highest first-preference voting intention since the 2022 election. The Greens are also faring well, with a small but significant gain in support since 2022 to their current 13% national vote share.
If the election were held tomorrow:
These models allow us to investigate the predicted vote share for each party if the election were held tomorrow.
The two party preferred vote is a a toss up. If an election were to be held tomorrow, we find there’s a 56% chance that the ALP would win the two-party preferred vote and a 44% chance that the LNP would win.
We predict that the Labor Party are likely to receive between 29-31% of the national first-preference vote share, with the LNP receiving 37-38% and the Greens 12.5-13.5%.
Methodology
This model is based on Prof. Simon Jackman’s influential 2005 research paper on the topic of combining opinion polls over time, with a few tweaks and changes.
The model is predicated on the idea that each political party has a parameter which represents the true fraction of the population who would vote for them on any given day. This parameter is generally hidden from us, however: it can only be measured at elections. We treat opinion polls as a noisy and possibly biased measurement of this ‘latent’ parameter. We assign each individual polling organisation a ‘house-effect’ term, which is the offset that organisation would report from the true support value.
We model the latent voting intention for each party as a Gaussian Random Walk. Importantly, we model political parties simultaneously, including in the model information about the historical correlations between the vote share of different parties. Shifts in public opinion don’t occur in a vacuum, and a decrease in support for one party is likely to be reflected in increased support for another. Including these correlations result in more precise measurements of the quantities we’re interested in.
We aggregate polls from a variety of sources, including the excellent resource at PollBludger.net. We fit the model using the probabilistic programming language Stan, performing full posterior inference using a modified Hamiltonian Monte Carlo sampling method.