Anyone watching the polls during the Australian election campaign – which, let’s face it, is everyone – might be tempted to conclude that the trend is so strong in Labor’s direction that there is no way the Coalition can win.
But the pollsters themselves warn that one factor should make observers more cautious than usual in interpreting the numbers, namely that the 2025 election has the largest number of undecided and soft voters Australia has ever contended with.
Essential research has found only about half of Australians voters know who they would vote for and would not change their minds. Of the rest, 34% said they might change their minds, while 13% had not decided who to vote for.
Kos Samaras of the research firm Redbridge says the younger generations that are now dominant – generation Z and millennials outnumber baby boomers at this election – have a lower level of connection with registered parties.
One result is that the Coalition is losing votes from disengaged, younger voters to minor parties and independents, and not necessarily getting them back from preferences.
“That’s inflating the Labor two-party-preferred vote,” Samaras says, although Labor’s own primary vote remains at historically feeble levels, barely above 30% in many polls.
Samaras says the large number of undecided voters, with less than a week until pre-polling begins, means there are likely to be “unpredictable results in very specific geographies”.
The 2019 watershed
Polling’s reputation took a hit in 2019 after the Coalition’s surprise victory, and since then some of the factors that made the pollsters’ life more difficult have only got worse – primarily the fragmentation of voting behaviour and the rapidly changing ways people communicate.
According to the Museum of Australian Democracy, most modern polls use a “randomly selected, statistically average group of people” to gauge the mood of the nation, a method invented by the American George Gallup. Roy Morgan conducted Australia’s first Gallup poll in 1941, and found that almost six in 10 Australians favoured equal pay for women.
Since then, there have been changes in technology, the way the data is weighted to account for factors including age, gender, ethnicity and education, and how to account for voters that are not rusted on to a major party.
Peter Lewis, the executive director of Essential, says over time randomised phone calling was replaced by low-quality robopolls. But since 2019 the shift has been towards online surveys.
“To do research well, it’s expensive,” he says. “There are cheap, not very good, alternative options.”
In 2019, the Coalition retained power with 51.5% of the two-party-preferred vote, with Labor on 48.5%.
That was almost the mirror opposite of what polls had predicted.
Scott Morrison declared his win a “miracle”, although the truth was more prosaic.
While the polls had been predicting a Labor victory, an inquiry found that neither divine intervention nor the “shy conservative” theory was responsible for the discrepancy between the polls and the result.
It was more likely the polls were “skewed towards the more politically engaged and better educated voters”, leading to a bias towards Labor that was not corrected with weighting, the inquiry found.
It could not rule out the “herding” effect – when polling companies consciously or unconsciously converge to avoid being outliers.
In the fallout from 2019, a range of polling companies founded the Australian Polling Council (APC) to ensure transparency, professional standards, accuracy and better understanding of polling process by the media and public.
It has a code of conduct and provides a “quality mark” to members to guarantee the quality of the research.
Pollsters point out that the 2019 result was still within the margin of error, and that the 2022 result was more accurate, but that at the same time polling has become more complex.
Support for the major parties is at an all-time low, as voters flock to minor parties or independents, or hover in the undecided/swing/soft space.
Samaras says in 2025 there are about 35 “non-classical contests”, meaning they are not straightforward Labor v Coalition battles.
“You had 80-odd seats in 2010 that went to preferences,” Samaras says.
“In 2022 over 130 went to preferences.”
Samaras says trying to get gen Zs on the line is making it more and more difficult to find samples representative of the population.
“In 2022 it took me three days to get 20 people under the age of 40 to answer the phone,” he says.
“They’re on silent.”
He says there can also be issues with some cultural barriers where people are not comfortable talking politics with a stranger.
‘You have to use all your critical faculties’
YouGov, a founding member of APC, now uses multilevel regression with post-stratification (MRP) for its polling.
It surveys almost 40,000 people and combines that information with data from other sources, including the census, along with previous election results.
The results are weighted and matched to electorate profiles, then simulations and variations are run through a computer system which uses artificial intelligence to process the results and learn from them.
Guardian Australia uses Essential polling, which has also conducted qualitative research for the Labor party, and also has a poll tracker that combines all opinion polls over a certain period.
Lewis says Essential changed the way it dealt with undecided voters after 2019.
“So we’re not pretending to predict the end of the horse race,” he says.
“For the 2PP we ask people to tell us how they went at the previous election, but we also ask them to nominate preferences.”
Polls themselves should be one of a “number of inputs”, he says. “It’s not a scoreboard. An election is not a horse race. You have to use all your critical faculties.
“At the same time the credible polls work incredibly hard to give an accurate readout.”
Samaras similarly says polls should not be seen as predictions.
“Trends should be observed over time, rather than an over-fixation in the dial moving one or two percentage points in the span of a month.
“If the numbers for one party are deteriorating, that’s a trend. That tells you something’s going on.”
Rodrigo Praino, a politics professor at Flinders University, says interpreting polling is further complicated because it rests on several assumptions.
“The first assumption is that people answering tell you the truth,” he says.
“The second assumption is that people don’t change their mind between the time you ask them and when they cast their vote.
“And the third is that people don’t make mistakes.”
The experts warn against reading anything into certain types of survey: single-seat polls if they are not done with a large enough sample size; polls that predict shock outcomes, particularly in crucial seats or where high profile MPs sit; and push polling, where the way the question is framed suggests the answer.
Paul Smith, YouGov’s public affairs and public data director, says when looking at any poll people should ask: “Are they using the latest in data science to make their polling projections? And are they being transparent about their methodology and how they came to those numbers?
“If it’s not industry best practice you shouldn’t pay attention to it.”
