Resolve’s poll, however, underestimated Labor’s primary vote by at least 3 per cent and overestimated the Greens’ primary vote.

Redbridge came closest to the final primary vote result by finding Labor and the Coalition tied on 34 per cent (the current vote count has Labor on 34.8 and the Coalition on 31.1 per cent).

“We did pretty well,” Redbridge pollster Kos Samaras says.

Crucially, none of the major pollsters found Labor beating the Coalition on primary votes, when this is what occurred on election day.

“There will be some soul-searching in the polling community about that,” says Reed.

Adrian Beaumont, a psephologist for The Conversation, says: “I think the pollsters didn’t dare predict a Labor landslide.”

YouGov performed well with its final seat-by-seat poll, predicting Labor would win 84 seats (still short of the 90 or so seats it expects to win). It correctly predicted Coalition frontbenchers David Coleman and Michael Sukkar would lose their seats, though it missed the result in Wannon, where shadow minister Dan Tehan held off an independent challenger.

Newspoll, published in The Australian, came close to picking Labor’s primary vote with its 33 per cent finding, but fell in the middle of the pack in two-party-preferred terms. Labor used Newspoll’s parent company, Pyxis Polling & Insights, to run its internal polling.

The least accurate pollsters in two-party-preferred terms were Ipsos (whose polls were published by The Daily Mail) and Freshwater (published by The Australian Financial Review), whose results were outside their margin of error.

Importantly, the Coalition also hired Freshwater to conduct its internal polling in key marginal seats. Freshwater’s polling significantly overestimated the Coalition’s primary vote, giving it an overblown 37 per cent vote share.

“It was definitely wrong,” a Liberal frontbencher says of the party’s polling. “We spent millions of dollars on it and will be keen to know what went wrong.”

Both Ipsos and Freshwater rejigged their polling methods from the last election, which seems to have played a role in skewing their results.

Unusually, Freshwater used how a respondent voted in the Voice referendum as a factor when weighting its survey responses to try to ensure they represented the Australian community.

Meanwhile, Ipsos weighted its responses on how respondents voted in previous elections.

Loading

Writing in the Financial Review, Freshwater pollster Mike Turner pointed to his company’s strong record in recent elections, saying it was impossible to call every election perfectly.

While his poll overestimated how many Labor voters who voted No in the Voice to parliament referendum would defect to the Coalition, Turner says the impact would have been small, making less than a one percentage point difference.

All the polls overestimated the level of support for One Nation, which helped contribute to an artificially inflated two-party-preferred result for the Coalition. The final Essential poll published by The Guardian, for example, found One Nation had 10 per cent support, when it received 6 per cent on election day.

Reed says: “Intriguingly, those polls taken earlier – around a week out – were more accurate than those released on the Friday or Saturday, where there was limited room for them to miss a late swing.

“For some reason these later polls’ results did not follow the trend of the campaign, where Labor was making strong gains throughout.”

Turner counters that all pollsters appeared to have missed a late swing to Labor from voters who only made up their minds in the final days of the campaign.

“When looking for trends it’s important to compare apples with apples,” he says. “All the polls show a consistent trend over the course of the campaign: a significant movement from the Coalition to Labor, even in the late stages.”

Read more on Labor’s landslide election win