Why opinion polls got the federal election so wrong and how to fix them

By the final week of the campaign all major pollsters had Labor slightly ahead on a two-party preferred basis with between 51.5 and 53 per cent of the vote.

While most polls overestimated the primary vote for the Coalition, commentators and analysts were talking about a slim majority for the incumbent government.

Some Coalition insiders were still convinced they could pick up seats, and Labor internal polling suggested it might seize up to 78 seats.

But only one polling company, YouGov, predicted Labor would thump home to victory with a significantly increased majority in the lower house - and even it underestimated the size of the win.

YouGov's polling was largely dismissed in political circles at the time.

But nearly two weeks after election day, Labor looked set to govern with at least 90 seats to the Coalition's 42.

Whether maths is your strong suit or not, it doesn't seem to add up.

How can the pollsters get the numbers roughly correct but not foresee the wipeout of a major political party?

"[YouGov] was the only pollster to say it would be the worst Coalition result since the Liberal Party was founded," YouGov public data director Paul Smith said.

He said their approach allowed them to predict results in specific seats, not just provide a generic national view.

"[Our modelling] understands the detailed differences in different seats - because the election is not about who's going to get 52 per cent of the votes.

"It's about what happens in each seat," he said.

ACM, the publisher of this masthead, relied on YouGov data for its election coverage in partnership with AAP.

This polling was based on the multi-level regression with post-stratification (MRP) technique, which combined survey data from 35,185 voters with demographic information, such as education and income levels, cultural background and employment.

Most Australian opinion polls rely on responses from between 1000 to 2000 voters.

The MRP system also modelled preference flows and the statistical likelihood of a particular outcome.

Opinion polling has been around since the 1800s but only since the second half of last century has it become big business.

Political parties fork out huge sums of money to pay for polling during election campaigns, which then determine where they direct resources.

Some Liberal MPs are reportedly calling for their party's pollster, Freshwater Strategy, to be ditched over its poor advice.

Freshwater had the Coalition on a 37 per cent primary vote, while the actual count was closer to 32 per cent.

But Mr Smith said both major parties were suggesting the YouGov polling was wrong in the final days of the election campaign.

"And we've been proven right," he said.

"If you want to show, seriously show, to politicians what the public think in every electorate, you need an MRP poll.

"We have an increasingly more complex electorate that will vote in multiple ways and we need a more sophisticated method of modeling public opinion to capture that accurately."

The primary vote for major parties has been dropping steadily for well over a decade and this has accelerated in recent years.

In 2007 a little over 85 per cent of the Australian electorate voted for either Labor or the Coalition first.

That had declined to just 68 per cent by 2022.

In 2025, it looks likely to dip below 67 per cent despite Labor's huge seat haul.

The Tally Room founder Ben Raue said election polls were still useful for predicting the outcome of "classic" electorates.

The difficulty comes in converting the national two-party preferred, or 2PP, prediction into seats.

"I do think the 2PP gives you a reasonably good idea of how seats that are classic will perform," he said.

"But the number of seats that are classic is getting smaller and smaller."

In a significant number of electorates an independent was now among the top two candidates.

"There's probably gonna be 35, 40 seats where ... it's not as simple as just saying it's Labor versus Coalition," Mr Raue said.

"It is a constant challenge, but I think it's also it's a bit of an art, and because we have a voting system where we are choosing candidates, choosing MPs, one by one - and so small changes in votes can shift a lot of seats."

He said people needed to "temper their expectations" of opinion polls, but they were nonetheless useful and acted as a referendum on the major parties.

"At a national level, I think they can do very well," he said.

"I think it was really informative during the campaign that Labor was losing support and got to a point where they were in some danger from Peter Dutton, and then the campaign went well for them and they bounced back up.

"And we know all of that because of opinion polls.

"And, even if they didn't quite get the magnitude right, they definitely had the direction of travel right."

William Bowe, the man behind The Poll Bludger blog which political operatives watch keenly during election campaigns, said opinion polls still had merit.

"I'm still defensive of the polling industry because they were more right than anybody else," he said.

"The polls [this election] weren't actually that far off," Mr Bowe said, particularly when a margin of error was taken into account.

"By international standards this was a good result."

But he said polling had changed since the 1980s and '90s when scam calls were rare and people were more likely to answer their phone to a stranger and take the time to respond to questions.

"You didn't have the relationship with your phone that you do today," Mr Bowe said.

Now, opinion polls used a mix of phone and online surveys, often relying on pre-registered panels of people who were offered prizes and other inducements for participating.

Mr Bowe did not predict polling companies would make big changes to how they operated following the 2025 election.

"I don't they're going to reinvent the wheel," he said.

But the situation was different in the US, where recent opinion polls have been disastrously wrong, partly because it was so hard to get core Trump supporters to participate in surveys.

He said while YouGov did well in its predictions overall this election, it struggled more with seats where independents had significant primary votes because its modelling relied on demographic data.

YouGov wrongly predicted Wannon in south-west Victoria and Cowper on the NSW mid north coast would be picked up by teal independents.

It also thought the Greens would retain three seats, but the party only held onto one.

"That [method] doesn't work very well with independents," Mr Bowe said.

The success of an independent candidate depended on the individual running, including their local profile, which could not be modelled with information about the electorate population alone.

Communication about how opinion polling worked and what purpose it served was key, Professor Scott Sisson, director of the University of NSW Data Science Hub, said.

But polling also needed to acknowledge the growing shift away from the major parties.

"We're going away from the two parties," he said.

"It's reasonably straightforward to do a three party [preferred] version of this.

"That would give you more information because I guess people are thinking a little bit more beyond the two parties."

If existing polling was not taking into account all factors, it might only tell part of the story and could miss things, Professor Sisson said.

"There's more nuances going on, to be honest, if we want a bit more sophistication in the conversation in the political landscape in this country, it has been more than, you know, one guy effectively talking to another guy."