“The results of pollster YouGov’s latest MRP model,” the ABC’s Casey Briggs announced at the beginning of the ABC’s Insiders on 16 February, “suggest the Coalition would be likely to win about seventy-three seats, with a lower estimate of sixty-five and an upper estimate of eighty, if a federal election was held today.” The ABC’s headline on the same story, posted online a day earlier, was less hedged: “Peter Dutton most likely to be next prime minister, according to YouGov poll.”
Seventy-three seats wouldn’t automatically make Dutton PM — the Coalition needs seventy-six to govern in its own right — but if the Coalition secured seventy-three and Labor sixty-six then the opposition leader is more likely than Albanese to garner the extra three from the crossbench. (YouGov mistakenly said the number required was seventy-five.)
The Coalition currently has fifty-four seats in the House of Representatives and Labor seventy-eight. That leaves a crossbench of nineteen, including four Greens and six teals. Also known as “community independent” or “Voices of…” candidates, the teals include NSW MPs Sophie Scamps (Mackellar), Kylea Tink (North Sydney) and Allegra Spender (Wentworth), Victorians Zoe Daniel (Goldstein) and Monique Ryan (Kooyong), and Western Australia’s Kate Chaney (Curtin). We might also include in that group Helen Haines (Indi), successor to Victoria’s original “community independent,” Cathy McGowan.
North Sydney was abolished in last year’s electoral redistribution, reducing the number of teals standing for re-election to five (plus Haines) and reducing the number of members in the next parliament from 151 to 150; the abolition of Higgins (Victoria), a seat that Labor holds, is offset by the creation of the new seat of Bullwinkel (WA), a seat Labor is favoured to win.
WHAT THE ABC MISSED
The story the ABC missed — the big story missed by all the media — is that if the Coalition were to secure seventy-three seats, it would do so (on YouGov’s figures) without having won back either of the seats it lost to the Greens in 2022 (Brisbane and Ryan) or any of the seats lost to the teals. The crossbench would be reduced from nineteen to eleven, but the five teals would be returned, as would one of the Greens (Adam Bandt, Melbourne); the three other seats held by the Greens would be won by Labor.
While YouGov did note that it was projecting the Coalition to win “none of the seats taken by ‘teal’ independents in the last election,” it saw this as a failure for Mr Dutton and his colleagues rather than a conspicuous strategic success. Not winning any of these seats “suggests the Coalition is experiencing challenges in winning support from the higher-income, highly educated voters that dominate its former heartland seats in Sydney and Melbourne” was its only comment. This may be true; but it misses the point.
Dutton’s strategy of ignoring the inner-city seats in Melbourne, Sydney and Perth that the Liberals lost to teal or Greens candidates will have been vindicated — if not wholly then in large part. The corollary: members of the press gallery, and others, who declared — endlessly — that no such strategy could work would be proved wrong.
On YouGov’s projections, the Coalition would pick up just one seat the Australian Electoral Commission classifies as “inner-metropolitan” — Chisholm (Victoria) — a seat currently held by Labor. The Coalition’s other gains, again at Labor’s expense, would come from “outer-metro” seats — Werriwa (NSW), Aston (Victoria), and Boothby (SA); “provincial” seats — Macquarie, Paterson, Robertson, and Shortland (NSW); and “rural” seats — Eden-Monaro, Gilmore, Hunter (NSW), McEwen (Victoria), and Lyons (Tasmania). This would mean that most of the Coalition’s gains would come from the “outer-metropolitan” (three) and “provincial” (four) areas to which Dutton has tried to appeal.
The Coalition’s biggest gains would come in New South Wales, where YouGov expects the Coalition to pick up eight seats from Labor (while Labor easily claws back Fowler from independent Dai Le), with three expected to come from Victoria. The other states will see the Coalition pick up just two.
If this proves broadly to be the case, the electoral pendulum will prove a good guide to how many seats Labor loses (in net terms) but not a good guide to which seats Labor loses. The pendulum works not because swings are uniform — they never are — but because the seats that should have swung (given the size of the swing overall) but didn’t are offset, roughly, by the seats that shouldn’t have swung (given the size of the swing overall) but did. Since 1972, it has been ever thus.
Although YouGov had posted the results on its website at 11.48 pm on the preceding Friday, Briggs presented its findings as an “exclusive” — something that the ABC, he said, could “reveal.” The poll, conducted 22 January–12 February, appears to have been handed to the ABC in the hope that it could be run as “breaking news” while (incidentally) promoting the pollster’s brand.
The poll’s national score — the Coalition ahead, 51.1–48.9, on a two-party-preferred basis — wasn’t very different from what other polls had reported; if anything, it was less dramatic than either the RedBridge poll or the Morgan poll, both of which had reported, on 10 February, a 51.5–48.5 split. Compared to the 2022 election result (47.9–52.1), these polls showed a swing to the Coalition of 3.6 points.
Based on that pendulum produced by Antony Green, the ABC’s chief election analyst, YouGov’s two-party split put the Coalition in line to pick up seven Labor seats (net); but on the figures produced by RedBridge and Morgan, the pendulum had the Coalition picking up eleven Labor seats (net).
YOUGOV’s ALL-SEATS POLL VERSUS REDBRIDGE–ACCENT’S “KEY SEATS” POLL
Without the YouGov poll, Insiders very likely would have opened with a roundup of polling in the Sunday papers, something it likes to do. Had it done this, it would have splashed with the front page of Sydney’s Sunday Telegraph which carried James Campbell’s account of a RedBridge–Accent “rolling tracking poll” in twenty “key seats” — seven in New South Wales, five in Victoria, four in Queensland, and four in the three other states — the first in a series in the run-up to the election. A fortnight later, Insiders would splash with the second in the News Corp-commissioned series.
The RedBridge–Accent poll — a poll that didn’t use MRP — was one that Insiders chose to ignore. What “reveal” might this poll have delivered? Across the twenty “key seats,” it had the Coalition leading Labor 52–48 — a three-point swing since the 2020 election, when the Coalition trailed 49–51. The Coalition was “doing better” in the seven New South Wales and five Victorian seats than in the four Queensland seats it polled. In New South Wales, the Liberals led 53–47, in Victoria by 58–42. But with a sample size of just 1002 (fifty respondents, on average, in each seat), the margins of error would have been very large. How well the Coalition was doing in other states was not disclosed. To see just how likely — or unlikely — these figures were as a guide to the swing in either New South Wales or Victoria, readers needed some account of Redbridge’s track record or of Accent’s track record; none was offered.
What did other polling show? Here, the Guardian’s tracking of all the published polls offers a sobering contrast to the RedBridge–Accent estimates. In New South Wales as of late 2024, the Coalition led Labor by the narrowest of margins, 50.2–49.8; in Victoria, it led by a similar margin, 50.7–49.3. Even allowing for a shift to the Coalition since then, the figures for Victoria are difficult to credit.
What makes the twenty seats selected by RedBridge–Accent the “key” seats? “These are electorates,” Accent’s Shaun Ratcliff explained, that “provide a good representative spread of different parts of Australia” and “all the intel suggests could either change hands at the election, or are likely to be competitive. At the last election these seats had a two-party preferred vote of 51 per cent.”
Not all of this was true. A closer look reveals that the twenty seats include thirteen of the twenty-seven Labor seats where the Coalition needs swings of between 0.2 and 7.8 percentage points (only three require swings of one point or less) and seven of the twenty Coalition seats where Labor requires swings of between 0.04 and 6.1 points (only two require swings of less than one point). Being spread across the country didn’t necessarily make them “representative”; the proportion of “provincial” seats was twice as great as the proportion nationally. Beyond ensuring the selection was roughly isomorphic with the proportion of seats in each state, what “intel” informed the sorting process — twenty seats from a possible forty-seven on the pendulum — is unclear.
In their own way, the RedBridge–Accent results were just as remarkable as YouGov’s. In 2022, the Coalition’s two-party vote in the seven NSW seats polled by RedBridge–Accent was 48.2 per cent; the Coalition vote in the five Victorian seats it polled, was also 48.2 per cent. On these figures, there had been a 4.8 point swing to the Coalition in the NSW seats and a swing of 9.8 points to the Coalition in the Victorian seats — though these estimates may not be exact, given that some seats have new boundaries. The finding that Labor was in bigger trouble in New South Wales than in Victoria contradicted what YouGov was suggesting; in the five Victorian seats, YouGov estimated the Coalition’s support at about 53 per cent rather than 58 per cent. That the Coalition was “clawing back votes lost to minor parties and independents” (in Campbell’s words) also ran counter to YouGov’s findings.
Were the swings recorded by RedBridge–Accent to be reflected statewide, the Coalition stood to gain five seats in New South Wales from Labor and ten in Victoria. While the shifts, as Campbell rightly observed, were “not enough to secure Mr Dutton the extra nineteen seats needed to win outright” — seven of the seats, he might have acknowledged, were (nominally) Coalition seats already — the swing “reinforc[ed] the prospect of a minority government.”
News Corp sees a minority government as the worst possible outcome of the coming election — worse, even, than having a majority Labor government returned. Happily, the poll appeared to be on News’s side. A minority government, Campbell added, was something “almost half the voters in key electorates agreed would be bad for the economy and people like them.” So, as well as keeping “readers in the know,” the results offered a way of mobilising them even if one had to resort to asking a double-headed leading question. And News had grounds for concern. In an Essential poll conducted three weeks later, opinion was divided over “the most likely federal election outcome”: 58 per cent tipped a majority Labor or Coalition government; 43 per cent, a minority Labor or Coalition government.
What made the YouGov poll newsworthy was not that it showed Labor losing its majority — something predictable from the moment it came to office, and shown by almost every poll for the best part of a year. Nor, as the RedBridge–Accent poll shows, was YouGov alone in showing the Coalition poised to form the next government, even if in minority.
What made the YouGov poll newsworthy was that it used multilevel regression with post stratification, or MRP, a technique for correcting model estimates for known differences between a sample and the population from which the sample is drawn, even when the sample is “highly unrepresentative.”
I explain MRP in more detail and look at YouGov’s track record using this technique overseas and in Australia, elsewhere in Inside Story this week.
THE PREDICTION PROBLEM: SEAT POLLS OR THE PENDULUM?
In “a close election like this one,” RedBridge director Kos Samaras is paraphrased as saying, “nationwide two-party preferred polls [are] not the most reliable tool for assessing the political landscape.” On his assessment, “tracking key seats offers a far more detailed and accurate picture of how parties are performing where it truly matters.”
The last “close election,” 2022, showed that this isn’t necessarily true. After the 2022 election, YouGov said its RPM poll had registered a national two-party vote of 52.2–47.8, something it hadn’t announced before. If this was what it showed, then anyone who consulted an electoral pendulum might have expected Labor to win seventy-six seats — an underestimate of one, not an overestimate (based on YouGov’s seat-by-seat polling) of three. In short, the national two-party vote would have been a better guide to the result than the seat-by-seat figures.
Seat-by-seat polls taken a long way out from an election are of limited value — and not only because things can change. (In 2022, late changes by voters seemed to have favoured come-from-behind independents at the Coalition’s expense.) Pollsters are forced to make too many assumptions, most obviously about the nature of the various contests before the parties or candidates have been announced. Respondents may find themselves choosing a party that won’t be on the ballot come May. Or they may hesitate to support a party — or an “independent candidate” — without knowing or being reminded who the candidate is, a process known in the trade as “prompting” or “aiding.”
Pushing their polls’ news value, especially when their brand is not well established, and determined to promote their polls as the best of the lot, pollsters are also open to other temptations. They will feel the urge to select measures of their records that flatter them (seat-shares rather than vote-shares, especially first preference shares rather than the two-party vote, and the proportion of all seats called correctly, as Labor or Coalition, even though most of these seats were never in doubt); claim to know which seats hold the “key” (a RedBridge–Accent conceit); call seats that are too close to call (YouGov called every seat, including those split 50-50); and assume their sampling of particular groups (YouGov’s coverage of Vietnamese and Hindi speakers in Fowler) is adequate.
Market research organisations are publicity seekers; the higher their media profile, and the stronger their claims to being prescient, the greater their chances of picking up lucrative contracts from corporates, governments and NGOs — including political parties, publicly funded.
To those wanting to conduct a poll, the absence of a media contract is not necessarily an impediment, especially when — thanks to the internet — the cost of entry to the industry has never been lower and the dissemination of the results never easier. While some are happy to conduct polls and publicise them through their webpage — YouGov’s initial move — market researchers increasingly conduct polls in the hope that their results will be accepted with gratitude by some media outlet, published prominently, and above all reported uncritically.
Such hopes are rarely disappointed. Polls, especially to do with elections, are newsworthy. To legacy media hard-pressed for resources, polls handed out gratis as exclusives are a boon. And few journalists, many of whom spend much of their time rewriting press releases, garnished with a couple of quotes from relevant players, have the skills — or, increasingly, the time — to do much more than spot the key finding (who’s ahead and who’s behind), note some of the other results (sometimes with an eye for the politics of their own outlet) and move on to stories that might require more effort.
Pollsters can then look forward to having their handiwork recycled by other media, picked up by politicians, and noted — again, largely uncritically — by bloggers and others trying to keep track of what poll has just been published and whether it says the same as some other poll published sometime before.
And the ABC? When it comes to taking in other people’s polls, the ABC and its very able data analyst Casey Briggs need to decide whether they are in the breaking news business — the “reveal” — or the news analysis business, reflecting in an informed way on what’s been published rather than allowing themselves, however unwittingly, to be publicity agents for the latest poll.
Taking a bit of time over this opens the possibility of spotting problems with the polls that an audience should be alerted to. It also allows for the possibility of informing the audience of a bigger story — one that the pollsters themselves may have overlooked. •
