The Polls May Have Been Off, but the Australian Federal Election Exposed a Growing Geographical and "Chattering" Divide
The victory for the Coalition led by incumbent Prime Minister Scott Morrison surprised many pundits and voters, and was dubbed a ‘miracle’ by the PM himself. But how did the narrow re-election of a sitting government in good economic times shock conventional wisdom?
Failure to understand the limits and vulnerabilities of opinion polling was a factor. But also gauging the national political mood from a metropolitan base can distort judgement because of the growing divide in outlook between the populations of major metros and voters in suburban areas, small towns and rural communities. These same factors figured in the political surprises of the UK’s referendum vote to leave the EU in 2016 and Donald Trump’s victory in the 2016 US presidential election.
There are a number of reasons why the result should not be viewed as surprising. Perhaps the most obvious is that the polls always suggested the election outcome would be a close call. Labor and the Coalition were rarely more than four points apart over the course of the campaign. Everyone knows that polls have error built into them, and the recent high profile polling ‘misses’ in the UK and US offered good reminders of this. Yet polls are often reported without the necessary health warnings or disclaimers – and in the absence of serious reflection on the methodological foundations upon which they are built.
The result also should not have surprised us given the long-established insights from political science that incumbent governments are more likely to be re-elected when voters judge the economy to be performing, when they and their leaders are viewed as more competent to manage the economy and when they emphasise the issues that voters care most about. With the rise of populism across a number of established democracies, and widespread disillusionment with the political class, these old truths about elections may have been forgotten.
In a close election the polls suggested a close result, and were only about 2% off in the two-party preferred vote – around the average for poll errors worldwide. What led the polls to mislead here? While the average poll error was not that unusual in its magnitude, the polling industry consistently showed Labor in the lead – with all the final polls suggesting it would form the new government. The polls systematically under-estimated the Coalition and over-estimated Labor. There is a parallel here with the experience in the UK’s 2015 general election where the polls systematically over-estimated support for the British Labour Party, which the official inquiry into the election polling found was due mainly to unrepresentative samples.
On top of this, the ‘spread’ of the poll estimates was extremely narrow throughout the campaign – with Labor’s vote share never deviating outside the 51% to 52% range. Simply due to sampling error alone, one would have expected modest variations from this which could well have shown the Coalition in the lead (and perhaps led to more caution over the predictability of the result among commentators). Such a unvarying pattern is suggestive of some form of ‘herding’, or over-modelling, although it is difficult to be sure of the cause without looking inside the pollsters’ methodological black box.
A parallel can also be drawn with the US 2016 presidential election where national polls performed fairly well by historical standards but failures with state polling in the Midwest led to over-confident predictions of a Clinton win. This was reinforced by the use of polls in high profile forecasting models – which put the probability of a Clinton victory as high as 90%. A recent study has shown that such probabilistic forecasts lead people to over-estimate the likelihood of front-runners’ winning. In state-polling for the Australian federal election, there were substantially larger poll errors in Queensland – which significantly impacted on the national result.
Beyond misreading of the polls, a final possible reason for the widespread surprise may in part be that – like in many other liberal democracies – many voters, media, commentators, bloggers, opinion framers and academics are increasingly living in enclaves of the like-minded, both online and the places where they reside. One of the striking patterns of the election result was the large swings against Labor in towns like Grafton, NSW and Mackay, Queensland. In contrast, Labor’s vote held up better in major metropolitan centres – such as in Sydney, Canberra, Adelaide, Perth and McNamara in south Melbourne. This mirrors a trend observed across many other countries – including the US, UK and France – where cities are increasingly supporting left-parties whereas smaller towns and peripheral areas are drifting to the right.
There was also a significant disconnect between the dominant mobilising narrative of the chattering classes which focused on “climate emergency” and the views of the ‘quiet Australians” as Morrison termed them. The ‘quiet Australians” rejected Labor’s climate policy as a post materialist pre-occupation of the affluent Wentworth-Warringah elites. Morrison’s great achievement in this election campaign was to convince them that a Shorten government would threaten economic stability and they voted for the devil they knew.