News Limited have unloaded what they had promoted, accurately I believe, as “the largest opinion poll ever conducted in Australia” – a 4000-sample monster covering four marginal seats in each mainland state. This poll has been dubiously reported, thanks to a national total calculation which has credited the Coalition with a 51.4 per cent two-party vote. This appears to be a straight average of the five states’ results provided by Galaxy, without regard to population relativities between the states or the fact that the seats targeted were 2 per cent weaker for Labor than the national total in 2007. A balanced appraisal of the results points to a swing of about 1.7 per cent, which would produce a national Labor two-party vote of 51 per cent if consistent slightly at the lower end of the recent phone poll trend. The poll shows a 2.4 per cent swing to the Liberals in the NSW seats of Eden-Monaro, Gilmore, Macarthur and Macquarie; a 1.6 per cent swing to Labor in Corangamite, Deakin, La Trobe and McEwen in Victoria; a 5.4 per cent swing to the Liberal National Party in Bowman, Dawson, Dickson and Flynn in Queensland; a 2.1 per cent swing to the Liberals in Hasluck, Stirling, Cowan and Swan in Western Australia; and no swing at all in Boothby, Grey, Kingston and Sturt in South Australia.
The rub for Labor is that the New South Wales and Queensland swing figures are right where they need to be to maximise the Coalition seat haul in uniform swing terms: over the 4.5 per cent mark needed for a tenth seat in Queensland, and just reaching the threshold that would cost them seven seats in New South Wales (it would take a further 1.5 per cent to bag an eighth). A straight loss of this many seats would single-handedly cost Labor the election. With no swing recorded in South Australia, the only counterbalancing gains would be the two Liberal marginals in Victoria, La Trobe and McEwen. The result would be a bare absolute majority for the Coalition.
However, any haul of 17 seats in New South Wales and Queensland would have to include a few they are generally expected to retain, such as Eden-Monaro and Page. Possibly some of the seats selected for the poll are a bit unflattering for Labor. There is a concentration of western Sydney in the NSW sample, an area yesterday’s poll of four seats for the Daily Telegraph showed to be tough for Labor (it appears Galaxy have conducted separate polls for Macarthur for each release). The Queensland sample also includes Bowman, which Labor has probably written off (UPDATE: Mark Bahnisch at Larvatus Prodeo says Labor is barely running a campaign, with reports appearing for weeks in the Brisbane Times that their candidate is invisible, and the local papers can’t get hold of her for an interview). Note that for all the vastness of Galaxy’s total national sample, as far as all-important Queensland is concerned the results are less sturdy than yesterday’s Newspoll, which targeted eight Queensland seats rather than four and 1600 respondents rather than 800. That poll produced a swing of 3.4 per cent against Labor compared with Galaxy’s 5.4 per cent, which in uniform swing terms would mean a difference of no fewer than four seats.
The table below shows swings recorded in state-level Newspolls and Nielsens through the first three weekends of the campaign (with one Westpoll thrown in for good measure), plus the targeted polling we have seen over the current weekend. For the former, samples for any given observation are 765 for NSW, 665 for Victoria, 585 for Queensland, 465 for WA (865 in week three, achieved by throwing in the Westpoll result) and 445 in SA, producing margins of error ranging from 4.6 per cent in South Australia’s case to 3.6 per cent for New South Wales. The composite of the most recent two Nielsen figures has smaller samples of around 250 for the smallest states. The latest Galaxy polls have samples of 800 per state and margins of error of about 3.5 per cent. The Newspoll marginals poll had samples of 600 in Victoria (4 per cent margin of error), 1200 in New South Wales (2.8 per cent margin of error) and 1600 in Queensland (2.5 per cent margin of error).
|Nielsen (2 week)||-1.7||-2.7||3||-3.4||-2.7||0.6|
UPDATE: Remiss of me not to have noted when the poll was conducted: from Sunday to Thursday, and hence not as timely as some of the more favourable recent polling for Labor.