First things first:
• I had a piece in Crikey yesterday looking at how-to-vote cards, preference deals and the media coverage attendant to them. It considers a claim by Greg Sheridan of The Australian that any teal independent victories will be owed to ovine fealty to how-to-vote cards on the part of Labor and Greens voters; Labor’s decision to put the United Australia Party ahead of the Greens in central Queensland seats, in sharp contrast to its treatment of the party everywhere else; and One Nation’s tokenistic direction of preferences against the Liberals in five seats, without endangering Pauline Hanson’s second placement on the Liberal National Party Senate ticket in Queensland. Had I held back a day, I might have been able to note the latest full-page United Australia Party advertisements proclaiming &8220;Labor are preferencing the Liberal Party”.
• Andrew Tillett of the Financial Review reports the Liberals are boosting resources in the Labor-held seat of McEwen in outer Melbourne. However, Labor sources say they expect to retain the Victorian seats that have become the source of media chatter over the past week, the others being Corangamite and Dunkley, as “national concerns over cost of living and antipathy towards Mr Morrison” are registering as strongly here as anywhere else.
• The Australian reports the Liberals are newly concerned that the normally safe Brisbane seat of Ryan will be lost to Labor. There has also been increasing talk that the Greens could prove competitive in both Ryan and neighbouring Brisbane, together with Labor-held Griffith immediately to the south.
• Alice Workman of The Australian reports “independent polling conducted over the phone, towards the end of April” shows Liberal candidate Jenny Ware on a low enough primary vote to put her in danger of losing Hughes to independent Georgia Steele. Liberal sources are quoted saying Ware is “on the nose with locals after being picked by Scott Morrison to run, against the wishes of rank-and-file branch members”. Assuredly not in contention is the seat’s incumbent, Craig Kelly, notwithstanding United Australia Party advertising proclaiming him the “next Prime Minister of Australia”.
• The Age/Herald today reports that this week’s Resolve Strategic poll showed 32% rated the Coalition as best to manage the Solomon Islands issue compared with 29% for Labor.
Now, by popular demand, a post probing into this site’s popular BludgerTrack poll aggregate, for which a permanent link can be found on the sidebar. This presently suggests Labor holds a lead of 54.0-46.0 on two-party preferred, which I don’t think anyone seriously expects to be actual result at the election. It does, however, show a narrowing trend commencing slightly before the onset of the campaign period, though not sufficient to suggest any chance of the Coalition closing the gap. Given the record low support for the major parties, the Coalition can at least hope that parity on two-party preferred need not be required to at least hold on to minority government – and also for another pollster failure like that in 2019 (the likelihood of which is considered in a post by Mark the Ballot).
BludgerTrack is one of a number of endeavours around the place that seek to plot a signal through the noise of federal opinion polling, together with one on the Wikipedia federal election page maintained by a user called Canley; Twitter user @Gergyl’s regular posts aggregating trends both short-range and long-range; and semi-regular blog posts from the aforementioned Mark the Ballot. That’s aside from the sites Armarium Interreta and Buckleys and None which, together with Professor Simon Jackman’s betting odds model, are tracking the horse race in other ways.
BludgerTrack and its close relatives, each of which produce very similar results, use LOESS (locally estimated scatterplot smoothing) functions to trace a path through the data points that keeps the distance between the path and the points at an appropriate level. The degree of tolerance for this difference is set by a smoothing parameter, which produces something resembling a straight line if set too high, and zooms all over the place in response to each individual poll result if set too low. The industry standard for determining the Goldilocks point is called the Akaike information criterion, which I use (thanks to a library available for the R statistics package) without really understanding its mathematical intricacies.
The data points themselves consist of every opinion poll of voting intention published since July 2019 by YouGov/Newspoll, Essential Research, Roy Morgan, Resolve Strategic and, just recently, Ipsos, which are weighted according to their perceived accuracy and adjusted to smooth out the peculiarities of each series. My methods here are quite a bit less presumptuous than they were before the pollster failure of 2019, when I imagined there was value in calibrating pollsters’ historic performances. I now assume that YouGov/Newspoll, the only pollster with any track record to speak of since 2019, is essentially free of bias, and calculate other pollsters’ biases based on the extent to which they deviate from a trend measure of it. These are halved so the peculiarities of each pollster have at least some weight in the overall result, rather than it being totally centred around Newspoll. The biggest change made is to Resolve Strategic’s Labor vote, which is increased by over two points. All other adjustments amount to less than one point.
Each pollster gets a weighting based partly on how much bias adjustment they’re being loaded up with, but mostly on my subjective impression of how accurate they’re likely to be, together with consideration of how frequently they report. The latter ensures the aggregate doesn’t get overwhelmed by the more prolific polling series. The most heavily weighted pollster at the moment is actually Ipsos, which is a (presumably) high-quality pollster that has so far produced only two polls released several weeks about. However, the mainstay of the series, Newspoll, is not far behind – a Resolve Strategic poll is worth about half a Newspoll, and an Essential Research (the most prolific series over the current term) and a Roy Morgan (which has a dubious track record) are worth about a quarter each.
Where BludgerTrack goes deeper than its rivals is in providing state breakdowns (together with leadership ratings). This is done using trend measures of each state’s deviation from the national results, which are then combined with the national trends (excepting Tasmania, for which next to no state-level data is published). Unfortunately, only Ipsos offers complete state data for each poll, as the others don’t care to have their small sub-samples held up to scrutiny their margins of error can’t bear. Essential Research comes close, but it smooths results for the smaller states by publishing three-poll rolling averages. Newspoll has always dealt with the issue by publishing state-level aggregates on a quarterly basis, which are a big deal for BludgerTrack when they come along (it would be nice to see one soon). Resolve Strategic only goes as deep as Queensland, and Roy Morgan’s are not used as they only provide two-party preferred at state level, whereas BludgerTrack works off primary votes.