Poll aggregation suggests the momentum against Labor has slowed, but with the rocks now too near for the ship to be turned around. Their last hope: the polls are wrong.
Yesterday’s Essential Research release concludes what I’d normally regard in the off season as a weekly cycle of poll results, making this an appopriate moment for a situation report. News reports yesterday were full of talk of a shift back to Labor based on a three-point lift in their primary vote from Newspoll, which neatly encapsulates what’s wrong with mainstream journalism’s reportage of opinion polling. For all The Australian’s success in marketing it as the gold standard, Newspoll is not – and indeed, does not claim to be – any more immune to statistical noise than any other poll. This particular shift was undoubtedly more noise than signal, coming as it did off a below-par result in the previous week’s poll.
For a more meaningful read of the situation you should refer to a poll aggregate, of which the one most readily to hand is of course BludgerTrack. However, you can find much the same story being told by Julian King at Pottinger (who has the Coalition at 53.6%), Simon Jackman (53.5%), Mark the Ballot (52.7%), Kevin Bonham (52.3%) and the AFR’s Poll of Polls (51.8%) (with apologies to any I’ve missed, like Andrew Catsaras’s which only appears on Insiders). Much of the variation comes down to the weight given to Essential Research, which is excluded altogether by Mark the Ballot and downweighted by BludgerTrack for having gone off trend.
Mark the Ballot’s charts offer a better visual representation of what emerges from the numbers coming out of BludgerTrack, which is that the rate of Labor’s decline slowed this week. The current output of the BludgerTrack model has Labor down 0.4% since this time a week ago (note that this isn’t exactly the same thing as the “last week” comparison on the sidebar, which reflects the model’s result at that time rather than its current retrospective evaluation) compared with 0.7% in the week to August 17, 0.7% in the week to August 10 and 0.5% in the week to August 3. As Mark the Ballot and BludgerTrack concur, the rot set in around mid-July, perhaps a fortnight before the election was called.
Of course, it’s a little too late in the game for a mere slowing of the momentum against them to do Labor much good. It would take a black swan event to return Labor to parity in the 12 days still available to them, the potential nature of which is by definition impossible to foresee. Labor’s other hope of course is that the polls and the aggregates derived from them are fundamentally wrong. A favourite argument of those predisposed to this view is that the online polling of Essential Research is telling a different story from polls using other methods, having had Labor at a post-Gillard high of 50-50 for the past two weeks. Nate Silver, it is noted, gave online polling the highest collective score in his post-match report after the presidential election, finding an average error of 2.1% compared with 3.5% for live interview phone polls and 5.0% for “robopolls”.
However, I would observe that the pollsters at the top end of Silver’s list are a mix of live phone interview and internet polls, with the live phone interview average dragged down by a number of poor performers at the bottom end (the most spectacular example being Gallup). Given the strong performances of Newspoll and Galaxy at recent state elections, at least on two-party preferred, it would require a leap of faith to conclude that either belongs in the latter camp. If anything, the risk appears to be on the downside for Labor. Essential Research aside, the big anomaly of the polling picture is that national polls have been kinder to Labor than electorate-level ones, a phenomenon by no means unique to the robo-polls (and it should be noted that the disparity seems to be lower in the case of Galaxy’s automated polling, suggesting that “house” as much as “method” bias has been at work here).
Now to some observations on the state-level projections. Labor received a large bounce in New South Wales and Queensland after Kevin Rudd’s return, bringing them into parity with the 2010 election result in the former case and well in front of it in the latter. However, in Victoria the move to Labor was more subdued. Three weeks after the Rudd comeback, BludgerTrack had the Coalition’s national lead at 50.5-49.5 and pointed to swings in Labor’s favour of 0.6% in New South Wales and 2.4% in Queensland, while going 3.7% in the other direction in Victoria (remembering that the 2010 election gave Labor its best result in Victoria since the Second World War). So far as BludgerTrack is concerned, the decline in Labor’s fortunes since has been driven by New South Wales, where the swing has now caught up with Victoria’s. That being so, one could perhaps hypothesise that it’s the Daily Telegraph wot’s winning it.
As for Queensland, that 2.4% swing to Labor is still there as far as BludgerTrack is concerned, inconsistent as that may be with reports of internal polling, the weekend Newspoll showing Labor going heavily backwards across eight Liberal National Party marginals, and a series of grim results for Labor coming out of Griffith and Forde. I keep waiting for new data from Queensland to square the circle by showing the other recent results to have been anomalous, and it keeps not happening. Of the last nine data points available to me from Queensland, which go back as far as August 7, only one fails to show a swing to Labor. The one exception is the sturdiest result of the bunch – an 800-sample Galaxy poll conducted late in the first week of the campaign. The rest are sub-samples from national polls, some (but not all) with very small samples.
The samples are smaller still for Western Australia and South Australia, although there are enough of them that it would be hoped that aggregating them would get you fairly close to a real world figure. On this basis, Labor’s once promising position in Western Australia relative to the 2010 result seems to have deteriorated significantly. In South Australia, Labor is rating a little better than media chatter (not to mention the weekend’s Hindmarsh poll) suggests they should be, though not to the extent of indicating an improvement on the 2010 status quo in terms of seats.
One final thought. With Labor holding 72 seats out of 150, it seems very likely that they will need a swing in their favour if they are going to win the election. Out of 17 national and 39 electorate-level polls conducted during the campaign (not counting a small number of electorate polls involving Greens or independent sitting members), not a single one has shown such a thing.