If there's one thing we learnt from the last election, it was: don't trust the polls.
In 2019, all of them showed that Labor had the support of most Australian voters - and that was exactly wrong. The Coalition won with 51.5 per cent of the vote, pretty well the mirror image of what the polls indicated: Labor was predicted to get not far off that percentage.
It was a failure. There's not much point in polls being nearly right in a tight race.
The pollsters' only consolation might be that Burt, the Psychic Crocodile, also got it wrong. He, you may remember, was a crocodile in the Northern Territory who, with cameras primed, was offered meat attached to a picture of Bill Shorten and one of Scott Morrison. Whomever Burt chose after employing his psychic powers would no doubt be prime minister.
Burt went for Bill. Wrong, wrong, wrong - but no more wrong than the polls.
It was all the more baffling because the Australian polls had seemed to be accurate before. All but one of the 27 final polls from 2007 and 2016 called the result correctly.
As a result of the polling failure, the Association of Market and Social Research Organisations and the Statistical Society of Australia undertook an inquiry.
One of its findings was that human behaviour had been changing quietly but the pollsters hadn't adjusted their methods accordingly.
One indication could have been that people were responding to surveys more reluctantly.
"The report documents a decline in response rates for typical telephone surveys from around 20 per cent in 2016 to 11 per cent in 2019, with the polls likely to be achieving much lower response rates than this," according to Darren Pennay, who chaired the enquiry and who is a professor at the Australian National University.
Polls work by choosing a sample of people to represent the country, with the right balance of age, income, location and a raft of other factors. If the sample is big enough and an accurate representation of the broader voting population, its views should reflect national opinion.
So the trick is to get an accurate sample. This wasn't done. There was an unwitting bias. The post-mortem found that people who were more engaged in politics were over-represented (and, by implication, voters who were not that interested in politics were under-represented).
"We did find the polls most likely over-represented people who are more engaged in politics and almost certainly over-represented persons with bachelor level degrees or higher," Professor Pennay said.
"Both of these factors are associated with stronger levels of support for the Labor Party and were not reduced by sample balancing or weighting strategies."
The difficulty for the upcoming Australian election is that it is difficult for pollsters to know how many people in the population take a keen interest in politics.
Pollsters can look at educational levels - the figures are easy to get from census results - but these figures don't reflect completely "engagement in politics". For that , there is no reliable measure.
Another flaw was that pollsters were increasingly relying on machines to do the questioning, and on other unreliable ways of gauging views. Squeezed media budgets meant that polling companies cut corners, according to Professor Pennay.
Pollsters had started to use "opt-in panels" but this method had obvious flaws: "People join those panels to do polls and do surveys and they get rewarded for doing so," Professor Pennay said.
"You have to wonder whether the people who join those sorts of panels are representative of all voters, and there is reason to believe that they're not."
And companies were using "robocalling" where a computer calls a phone and the person who answers punches in answers to automatic questioning by the machine (or just hangs up). "Generally, that sort of 'robo-polling' is only done to landlines. So that's another problem, given that so many households don't have landlines."
The warning signs had been there in elections in the US and UK where the polls picked the wrong winner but, for some reason, Australian pollsters hadn't registered the difficulties heading their way: "It also seemed to be the case - perhaps lulled into complacency by a long period of relative success and a mistaken belief that compulsory voting made Australia different - that our pollsters did not heed the lessons emerging from the polling reviews into the 2015 UK and 2016 US elections."
Maybe it won't matter this time. According to two recent polls, Labor has a sizable margin over the Coalition. The latest Newspoll puts Labor ahead 53-47, while the Resolve poll gives Labor a 54-46 lead.
Professor Nicholas Biddle at the ANU Centre for Social Research and Methods said the Labor Party was in a far stronger position now than it was before the 2019 election.
A whole heap of other factors that tend to predict voting have also moved against the Coalition - like declining confidence in government (and that tends to indicate disenchantment with the incumbent).
"People's views on polices have moved more towards a policy agenda which you would expect to benefit the Labor Party. People are more likely to think global warming is an issue now," he said.
So is it all over, bar the shouting? "I wouldn't say it's all over. Over the past 12 months there was also an increase in the undecided votes."
The polling companies learnt from the last election what they did wrong. They changed their methods.
Whether they have now got it right will be clear on polling day.
Sign up for our newsletter to stay up to date.