4
GET EXCELLENT POLLING, FOCUS ON THE PRIMARY VOTE
At this point things are getting ugly in the cave. You see it now. The wrong people were in the wrong portfolios. You got suckered into a ‘big target’ instead of a ‘smart policy’ strategy. Your campaign communicated a thicket of policies instead of a couple of crucial messages which your policies strategically fitted into and reinforced, so voters mostly heard static from you. Meanwhile, the Liberals’ message about Labor’s alleged franking credit ‘death tax’ policy pervaded the electorate like marsh gas though, as did sharper messages about Labor’s uncosted climate change policy. Why weren’t you told? Where’s the pollster? The one who said you were going to win?
Political polling in Australia has a strong and credible record. The 2019 election result was a shock because public opinion polls suggested Labor would win, and for three years had pointed to a Labor victory, but this did not happen.
The most prominent, Newspoll, showed Shorten Labor level with, or ahead of, the Turnbull Government in every survey after election day 2016, and ahead of the Morrison Government every survey from the time Scott Morrison succeeded Turnbull as prime minister through to election day 2019. A Labor success looked so certain that a few days before the election, on Wednesday 15 May, Sportsbet paid winnings to punters who had backed Labor to win. The next day, Sportsbet declared Labor had the election ‘run and won’.1
On Saturday 18 May 2019 Australian voters made not just Sportsbet, but Labor and the pollsters too, look like monkeys. Not that many Liberals, Prime Minister Scott Morrison excluded, thought differently. Liberal treasurer Josh Frydenberg privately tipped a coalition loss and feared losing his own previously safe Melbourne seat of Kooyong right up to election day.
Despite the polls, many Labor people were uneasy about the likely election outcome but deferred to the conviction of Bill Shorten and his staff that Labor would win. Some unconvinced senior Labor figures did try to intervene behind the scenes, on a range of fronts, but public opinion polling provided an implacable shield for Shorten and his team to resist internal suggestions for campaign changes. So what are these juju devices possessing such incredible power?
Professionally designed and ethically conducted opinion polls survey a statistically representative sample of voters to produce a rough snapshot of how people say they would vote if an election were held at the time of the survey. Though they are the most scientific guide to how parties and leaders are performing between elections, they are also human constructs, and thus fallible, fraught with flaws, and famously figure as weapons in leadership battles.
They reflect the opinion of the people surveyed; some lie, and some are more adventurous with a theoretical vote disclosed to a pollster than with the one that counts come election day. Opinion polls are not predictive, though it is human nature to slide into treating them as though they are. Rather, polls are like a progress score in a football match but, as we all know, a team can lead for an entire game only to lose in the dying seconds, as Labor appeared to in 2019.
Polls have a disclaimer on the results that may be higher or lower within the range specified in what pollsters call a ‘margin of error’. A poll that has a 2.5 per cent margin of error, for example, and has Labor on a two-party preferred vote of 50 per cent, means the actual result could range anywhere from Labor on 47.5 per cent to Labor on 52.5 per cent.2 That is very imprecise: 47.5 per cent would be a thumping loss, 52.5 per cent a landslide victory. The ‘margin of error’ is both statistically legitimate and a handy fig leaf for pollsters to cover their embarrassment should their polling prove terribly wrong. Because the doubt created by the margin of error undermines the poll’s authority, and therefore its magic, when disclosed it is usually in the tiniest of small print.3
Further, different polls conducted by different companies use differing methods that are disclosed to varying extents, and sometimes barely at all. Polling practices have been subject to tremendous disruption given changes like the decline of fixed telephones in households, and the rise of mobile phones, which make representative samples that much harder to get. Even before the digital era though, pollsters vigorously argued about which techniques delivered the most reliable results, their own always asserted as superior to those of rivals.
Polls have a latent power that becomes blatant in the hands of political combatants. They can be devastating when used to destabilise leaders. Think of Malcolm Turnbull’s campaign to dislodge Tony Abbott as prime minister using Newspoll, and in turn the efforts of multifarious Liberals to dislodge Malcolm Turnbull as prime minister with the same argument when his own numbers faltered.
The sustained campaign by Kevin Rudd’s supporters to undermine Julia Gillard’s prime ministership is a more nuanced story of polling’s use in destroying a leader. Whenever the Gillard Government’s polling improved, Rudd camp leaks, backgrounding and public interventions hostile to Gillard depressed Labor’s poll results again, eroding Gillard’s underlying position. This created the circular argument that Rudd’s return to the prime ministership was necessary to ‘save the furniture’ – that is, reduce the electoral damage going to the 2013 election which Gillard would have created. But her diminished standing was largely wrought by the Rudd camp’s destabilisation of her in the first place. Gillard could not occupy the high moral ground on this, however, given her semi-regular own goals as prime minister as well as her destabilisation of Kim Beazley as Labor leader firstly in favour of Mark Latham in 2004, and again, this time for Rudd, in 2006. Rudd’s overwhelming 2013 election loss to Tony Abbott, reducing Labor to its second lowest primary vote in the post-war period, may have been karma but this was no comfort to Labor voters who have endured the Abbott, Turnbull and Morrison governments since.4
Internal party polling, whose secrecy gives it extra cachet to the journalists who very occasionally get access to it, is also sometimes used to destabilise leaders. Internal Labor polling was leaked by Labor sources in the destabilisation of Bill Hayden for the benefit of Bob Hawke in the early 1980s. Internal Liberal polling leaked by Liberal sources triggered the final collapse of Alexander Downer’s leadership in favour of John Howard in 1995.5
All those examples rely on the use and abuse of negative opinion polls, but mediocre or ‘good enough’ polls can also play a significant role. After the 2016 election, for example, Labor hoped the Turnbull Government’s polling would not decline to the point where Turnbull, nearly beaten by Shorten in 2016 and considered beatable in 2019, was replaced as Liberal leader – hope dashed when Morrison succeeded Turnbull in 2018. It was not the absolute size of the Liberals’ deficit in the polls that was Turnbull’s problem. After all, in the four fortnightly Newspolls up to the week Turnbull fell, the government was only slightly behind on 49 per cent two-party preferred to Labor’s 51 per cent.6 Rather it was the protracted run of mediocre polling that made Turnbull vulnerable, with his government failing to win a Newspoll since election day 2016 – especially potent since it was the polling pattern Turnbull himself used to overturn Tony Abbott as leader in 2015. In turn the Liberals hoped Bill Shorten, whom they perceived as beatable, would lead Labor to the 2019 election – a hope that was fulfilled, helped by those same Newspolls showing a small, but persistent Labor lead.
Like everything else, political opinion polling has a history. It is a by-product of democracy.7 Absolute monarchies and dictatorships have little need to know what people are thinking, but those seeking to win democratic elections do. Try to conduct and publish an independent political opinion poll in Xi Jinping’s China and see what happens. Conversely, try and imagine politics in contemporary democracies without polls – almost inconceivable.
World War II saw the beginning of nationwide political polling in Australia. US President Franklin Roosevelt observed the emerging utility of political opinion polls while in office and Australia’s founding Minister to Washington, Richard Casey, who knew Roosevelt, shared his view after deep initial scepticism.
Casey wrote in 1940 to the Australian newspaper proprietor and, briefly, wartime Menzies Government director-general of information, Keith Murdoch, about the virtues and apparent reliability of the Gallup Poll and the Fortune Poll ‘based on the fact that they have been proven right, within small limits, by subsequent public votes at many elections’.8
This fell on receptive ears. Nascent US polling developments were evident to Murdoch from his own American travels.9 He sent an employee, Roy Morgan, from Melbourne to New Jersey to learn about polling from George Gallup at Princeton. On return Morgan became managing director of Australian Public Opinion Polls (The Gallup Method), reporting to the general manager of the Herald and Weekly Times newspaper group which Murdoch oversaw.10
Australia’s first Gallup Poll was published in 1941. From the outset Labor was suspicious of political polling, not least because of Keith Murdoch’s involvement.11 Two-thirds of a century later many Labor supporters rued the role of Newspoll, published by Murdoch’s son Rupert’s flagship local paper, The Australian, in Labor’s 2019 election loss.12
The Newspoll published in The Australian pointed to a Labor win throughout the election campaign, its margin actually widening on election eve to show Labor ahead on 51.5 per cent of the two-party preferred vote compared to the Liberal and National coalition parties’ 48.5 per cent.13 The actual result was effectively that with the party positions reversed: Labor got 48.47 per cent of the two-party preferred vote and the Liberal and National coalition parties got 51.53 per cent.14 Newspoll was not the only one embarrassed by the gulf between its numbers and the actual election result but, as the dominant survey in the Australian media, its error was the most obvious and significant.
With decades of bitter historical experience of The Australian’s slanted reporting of Newspoll – results favourable for the Liberals triumphantly splashed on page one while positive results for Labor were downplayed or ignored in favour of reporting lesser, cherry-picked aspects favourable to the Liberals – many Labor supporters wondered whether this was a new and especially nefarious bit of Newspoll-related bastardry. After all, if Newspoll had pointed to a possible loss at any time since the previous election, Shorten Labor may have rethought its approach and campaign, or at least been more open to the arguments of those who believed it urgently needed to do so. Given the Morrison Government won by just two seats, the material impact of this underlines polling quality as another factor that may have made the difference between Labor winning and losing in 2019.
Labor’s election review found that ‘the publicly available Newspoll figures had a persistent technical error that overstated Labor’s primary vote, understated the Coalition’s primary vote and consistently suggested Labor was in an election-winning position’.15 This ‘created a mindset dominated by high expectations of a Labor victory, and this affected the Party’s ability to process research findings that ran counter to this’ – that is, to accept the need for changes to the campaign.16 In a preferential voting system like Australia’s, how pollsters estimate preference flows – that is, to which major party votes for minor parties and independents end up going – is another major source of uncertainty, further complicating the accuracy of the polls.
Labor’s election review, when it was eventually conducted and released, acknowledged an ‘industry wide failure’ and noted Newspoll was not alone in wrongly showing Labor persistently in the lead.17 Immediately after the election, however, it took a Nobel laureate astrophysicist, ANU vice-chancellor Brian Schmidt, to point out the essential truth about opinion polls: they are human constructs and the people who design and manage them are prone to herd behaviour.18 Schmidt pointed out that every single one of the 16 polls published by various pollsters between the election being called and held showed Labor ahead on somewhere between 51 per cent and 52 per cent of the two-party preferred vote inclusive. When Schmidt said that ‘mathematics does not lie’ it was because ‘the polls have been manipulated, probably unintentionally, to give the same answers as each other’ because of confirmation bias. ‘I say unintentionally because humans are biased towards liking to get the same answer as everyone else,’ Schmidt said, pointing to the practice in large physics and astronomy experiments of ‘hiding the answers of an analysis from researchers until they are completely done to avoid this effect’. In other words, pollsters perform their own version of the ‘groupthink’ so damaging to party leaders and therefore to their party’s electoral prospects; and the academic world, but not the political world, has devised a method of dealing with it to make research results robust.
When it comes to political polling, the model each pollster devises includes many discretionary choices: how to design the sample (who, where, age and so on, as well as how many of each type of voter), how to conduct the poll (face-to-face, telephone, online), how to deal with ‘undecided’ voters, how to notionally allocate preferences since under our preferential voting system the two-party preferred vote is crucially important, and so on. ‘I don’t know why the polls so badly missed the election’s actual result,’ Schmidt said. ‘But whatever led … five polling companies to illegitimately converge on the same answer, must be a significant contributor. All five need to have a thorough and independent investigation into their methodologies, and all should agree to better reflect uncertainties in their future narratives.’19
This has not happened, of course, and in any meaningful way won’t because the value of political opinion polls to media outlets is their newsworthiness. Media organisations want an attention-grabbing story, with figures just different enough from competitors’ polls to make news, but not so different that it would look like a freak or ridiculous result making paying for and publishing the poll commercially worthless. They want polls that play down or hide the ‘uncertainties’ Schmidt says should be highlighted.
After the election one polling firm, Lonergan Research, admitted suppressing a poll that accurately captured Labor’s dire standing in Queensland, to avoid the embarrassment of looking so different from other polling firms’ findings.20 Its chief executive, Chris Lonergan, confirmed Schmidt’s ‘herding’ thesis, saying pollsters weighted their data ‘until you get a result which is in line with the result that you would like to publish, which is not out of step [with competitors]’. British pollster Survation admitted to supressing a poll after the 2015 UK election for the same reason.21
It is up to us to remember one of the big lessons from the 2019 election polling debacle: focus on that part of the poll least affected by discretionary polling design decisions and pollster groupthink, namely the primary vote – and only then, secondarily, consider the two-party preferred vote. Newspoll overestimated Labor’s primary vote as well as its two-party preferred vote: it had the primary vote at 37 per cent in its final poll compared to Labor’s actual catastrophic 33.3 per cent election result. However, at least the primary vote data was not further complicated by Newspoll’s faulty preference flow assumptions then used by it to calculate Labor’s two-party preferred position. That primary vote data contained important information. Labor’s primary vote at elections since Federation is laid out in Table 2 (see pages 46–7).22
These figures show that Labor on less than a 40 per cent primary vote has won elections just twice in history. It happened once under Bob Hawke in 1990 in the context of recession and record home loan interest rates, with Labor only securing 39.4 per cent of the primary vote. It happened a second time under Julia Gillard in 2010 when Labor’s 38 per cent primary vote, harnessed to Gillard’s exceptional skill at winning over and keeping in tow a clutch of independent MPs, enabled her to form a minority government. In 120 years of Australian political history, it only happened twice. That Newspoll showed Shorten Labor’s primary vote under 40 per cent for nearly six months before the 2019 election was a clue – a very big clue – that a Labor win was far from guaranteed.
Recall the tendency of party leaders to think they know better than leaders before them about how to win? Profound self-belief is of course an essential trait in successful politicians – but those who appoint ‘believer’ staffers, and spend all their time with ‘believer’ colleagues, and exclude those bringing a dash of realism or useful historical perspective to the table, can count on ‘surprise’ election results more often than others. Groupthink leads to selective data vision of the kind which happened in Shorten Labor’s 2019 campaign: they saw the ‘winning’ two-party preferred Newspoll figures in bold and were blind to the worrying, and in historical context deeply alarming, primary vote figures. Even if Newspoll persistently ‘overstated Labor’s primary vote (and) understated the Coalition’s primary vote’, those overstated Newspoll primary vote figures for Labor were plenty bad enough to demand attention and action.23 Australia has a preferential voting system, but you don’t get the benefit of preferences if your primary vote isn’t high enough for your candidates to survive long enough in the count to benefit from them.
| RESULTING IN | |||
|---|---|---|---|
| OPPOSITION | GOVERNMENT | ||
| 1901 | 19.4 | ||
| 1903 | 31.0 | ||
| 1906 | 36.6 | ||
| 1910 | 50.0 | Fisher Government | |
| 1913 | 48.5 | ||
| 1914 | 50.9 | Fisher Government | |
| 1917 | 43.9 | ||
| 1919 | 42.5 | ||
| 1922 | 42.3 | ||
| 1925 | 45.0 | ||
| 1928 | 44.7 | ||
| 1929 | 48.8 | Scullin Government | |
| 1931 | 27.1* | ||
| 1934 | 26.8° | ||
| 1937 | 43.1 | ||
| 1940 | 40.1 | ||
| 1943 | 49.9 | Curtin Government | |
| 1946 | 49.7 | Chifley Government | |
| 1949 | 46.0 | ||
| 1951 | 47.7 | ||
| 1954 | 50.1 | ||
| 1955 | 44.7 | ||
| 1958 | 42.9 | ||
| 1961 | 48.0 | ||
| 1963 | 45.5 | ||
| 1966 | 40.0 | ||
| 1969 | 47.0 | ||
| 1972 | 49.6 | Whitlam Government | |
| 1974 | 49.3 | Whitlam Government | |
| 1975 | 42.8 | ||
| 1977 | 39.6 | ||
| 1980 | 45.1 | ||
| 1983 | 49.5 | Hawke Government | |
| 1984 | 47.5 | Hawke Government | |
| 1987 | 45.8 | Hawke Government | |
| 1990 | 39.4 | Hawke Government | |
| 1993 | 44.9 | Keating Government | |
| 1996 | 38.8 | ||
| 1998 | 40.1 | ||
| 2001 | 37.8 | ||
| 2004 | 37.6 | ||
| 2007 | 43.4 | Rudd Government | |
| 2010 | 38.0ˇ | Gillard Government | |
| 2013 | 33.4 | ||
| 2016 | 34.7 | ||
| 2019 | 33.3 | ||
For much of Newspoll’s existence its surveys were conducted in-house by a News Corp owned company. These days News Corp contracts out the conduct of the Newspoll survey to YouGov Galaxy which, to the surprise of many Labor people, was also engaged to do polling for the Shorten Labor opposition including during the election campaign. This private ‘campaign track’ polling had a similar flaw to Newspoll, but on a smaller scale: it pointed to the public Newspoll published in The Australian being wrong about Labor’s level of support.24 ‘[The] fact the campaign track reported a Labor two-party preferred vote that was less optimistic than published polling every night of the campaign, provided warnings about key problems,’ Labor’s election review found. ‘Seat polling during the campaign also provided early warning Labor’s campaign was struggling, particularly in regional Queensland.’ Yet even as the ‘campaign track’ polling done for Labor by the very same pollster that produced Newspoll drew into question Newspoll’s accuracy – raising doubts about whether the Labor win Newspoll forecast would actually occur – the Labor campaign didn’t review, refocus and reorient the campaign. Averting their eyes to the electoral devastation unfolding in Queensland during the campaign, the Shorten camp banked a win that had not, would not and did not happen. Love of imminent power proved blind.
The penultimate word goes to YouGov Galaxy managing director, David Briggs. ‘While all research has a margin of error,’ Briggs wrote in a pinned tweet four days after the election in a memorable bit of fig leaf application, ‘the research conducted by this company across tracking polls and in marginal seat polls provided the ALP with an accurate appraisal of the state of its vote across a number of seats that would eventually be lost by Labor. What Labor strategists and decision makers within the campaign chose to do with these findings are matters for them.’25 Translation: We might have got it wrong in Newspoll but Labor was warned about the situation through our ‘campaign track’ and marginal seat polling and did nothing about it.
The fact remains, YouGov Galaxy did not alert the public to or correct its flawed Newspoll methodology – not until well after the election. It continued to publicly project through Newspoll the likelihood of a Labor victory right up to and including election day, shaping powerful expectations which undercut the ability of Labor realists to persuade Shorten to reshape the campaign.
So if you want to win an election, get excellent polling. Read all of it rather than cherry-picking. Pay absolute attention to the primary vote because it is the least likely to be subject to pollster error – and even more because if your primary vote isn’t high enough, your candidates don’t survive long enough in the count to benefit from preferences, no matter how many you get. Finally, as the next chapter shows, pay close attention to how and where the vote is distributed, because there is no such thing as an ‘Australian voter’.