Opinion polls: What are they, how are they made and why do we love them so much?

That's because in 2019, the five national opinion polls spectacularly failed to predict the outcome of the federal election.

The polls each predicted a Labor victory, and YouGov's Newspoll even placed Labor at the top for 55 consecutive polls.

But Labor didn't win. In the only poll that really matters, on election night, the Coalition claimed a 51.5 majority to Labor's 48.5.

So how did the polls get it so wrong in 2019, and have we learnt our lesson this time?

ACM spoke to University of Central Queensland political scientist Dr Jacob Deem to find out.

Opinion polls are survey of the general public.

Short questionnaires are conducted over-the-phone or over-the-internet, and are supposed to be a window into the voting population's feelings at one point in time.

"The major polling companies tend to use a pool of of randomly generated respondents, so they'll generate a panel of between 1000 and 2000 participants, usually about 1500 from right across the country," Dr Deem said.

"They'll try sample so that they've got a pretty representative spread of men and women from different regions within Australia, different age brackets, and things like that to try and capture a sample that's generally representative of the population as a whole."

The surveys will ask people about their voting preferences and who they intend to vote for in the upcoming election.

"So opinion polls provide a snapshot of the electorate's mood or attitudes towards a particular party candidate or issue," Dr Deem said.

A few things worked against the polls in 2019.

Firstly, it was found that the polls oversampled Labor supporters, resulting in a firmer belief that Labor would win.

"It takes time out of people's days to participate in these surveys," Dr Deem said.

"And the people who are more likely to take that time are a bit more interested in politics [and at that time] they tended to be Labor's support base."

So the sampling was problematic and that caused an echo chamber of reporting that artificially promoted Labor as the likely winner.

"The way that these polls are picked up in the media and reported then starts to feed political views as well. And so for example, Bill Shorten himself personally didn't poll very well. He was fairly unpopular. That then gets picked up in the media and reported that Bill Shorten's unpopular," Dr Deem said.

"That leads to further reporting. And then what people are seeing in the media is that Bill Shorten's unpopular. He's not likeable. So then in the next poll, when I get asked, 'Do you like Bill Shorten', they say, 'Well, no, I don't'."

Once the dust had settled on the Coalitions victory in 2019, Dr Stephen Mills from the University of Sydney spoke to ABC News to explain how the polls managed to fall so far from the result.

He explained it was evidence that the way polls have been conducted in the past did not meet the needs of the modern political landscape.

"... in the days when everybody had landline phones, it was relatively easy to get a clear, good sample of the entire population," Dr Mills said.

"Today mobile phones and online panels, they're not really inherently representative and so the big development in polling really over the last couple of decades has been in weighting. And I think this is where a lot of polls came adrift."

In recent state elections in Western Australia, South Australia and Queensland, the newspolls proved to have returned to their predictive mojo.

In the final newspoll before the March election in South Australia predicted a Labor landslide, and that eventuated.

Then in Western Australia last year, the final newspoll did also pick the winner.

And in 2020, the pollsters predicted the Queensland state election, though not the margin of the win.

How has this improvement come about? Well, the largest poll at least has changed its way of sampling. YouGov, which runs the Newspoll survey, now runs its samples entirely online.

Campbell White, head of YouGov's public affairs and polling, told 2GB's John Stanley the polls now seek out representatives of otherwise silent demographics.

"Younger people who didn't go to university, particularly if they're on higher incomes, they're quite difficult to get in surveys," Mr White said in April, 2022.

"So you actually have to go after and make sure you've got enough of those sort of people. People in rural areas is another group, that if you don't actually actively make sure your sample's represented, it's not going to be."

Political scientists warn against it, but "the temptation always exists to use the polls to predict the election", as Dr Deem says.

It's generally considered the polls that are made closer to the election date will be the ones which are most accurate and closest to the election day result. Although, again, that was not the case in 2019.

Polls can be made at any time during a government's three year term, but are perhaps most relevant to the public during an election campaign.

"In between elections polls provide a useful feedback mechanism to help the government gauge how people are reacting to its policies," Dr Deem said.

"Polls are useful for the major parties and minor parties, to work out their election strategies, work out where they should be focusing their efforts, key seats that they might need to shore up support in or where they think they might have an opportunity.

"And for the voters as well, they there's an element of value in saying what other people think, whether people find their views consistent with other people around them as well."

Opinion polls are not gospel.

The surveys can really only shine a light on the electorate's feeling during that particular moment in time.

Polls often don't indicate the overarching performance of any one candidate or party over time.

"What they tell us is a snapshot of the electorate and their attitudes at any given moment. It can be reaction to particular things that have happened in the preceding few days," Dr Deem said.

Analysing any trends that emerge over the weeks leading to an election can be helpful in building an understanding of how a party or candidate is tracking, but again, it's not gospel.

"The difficult thing is that [opinion polls] are often used these days to predict the outcome of elections, and that's where things get tricky because it's really not what they intended for," Dr Deem said.

"And taking a one off poll as as predictions for an election can be really problematic."