Scattered thoughts on the Senate, the western Sydney electorate of Fowler, Bob Katter, the informal vote rate, and the fine art of poll aggregation.
Time for a new thread, so here’s some very scattered thoughts that it occurs to me to share at this late hour:
I had a piece on the Senate result in Crikey yesterday, and have been keeping a low profile on Poll Bludger in part because I’ve been busy fielding inquiries from media outlets eager to hear an election wonk’s take on the whole affair. If you’d like to comment on the progress of late counting in the Senate I’d encourage you to do so on the dedicated thread, or at least re-paste your comments there after leaving them on this one.
I’d also like to encourage those with particular insights to offer on late counting in close lower house seats to share the love in the relevant comments threads, which can serve as useful clearing houses for information for those of us trying to keep up. Note that these posts can be accessed through links near the top of the sidebar.
So what the hell happened in Fowler? There was, as we know, a much milder swing against Labor in western Sydney than media hype and certain local opinion polls had primed us for. However, that scarcely explains the thumping 8.8% swing enjoyed by Labor journeyman Chris Hayes. What presumably does explain it is Liberal candidate Andrew Nguyen, chosen by the party with a view to snaring the Vietnamese vote in Cabramatta, who suffered swings approaching 20% in that very area. As to what Vietnamese voters might have known about Nguyen that the Liberal Party did not, I cannot even speculate. However, it won’t be the only question the party has to ask itself about its candidate selection processes in New South Wales, for the second election in a row.
It wasn’t a very good election for Bob Katter, who failed in his bid to bring new allies to Canberra and had his seemingly impregnable hold on Kennedy cut to the bone. One reason of course was that he was squeezed out by Clive Palmer (with due apologies for the unattractiveness of that image). However, another was very likely a preference deal he cut with Labor which in the event did neither party any good. I would also observe that this is not Katter’s first failed attempt at empire-building. At the 2004 Queensland state election, Katter organised an alliance of independents with a view to activating discontent over sugar industry policy, and the only one to poll a substantial share of the vote had done nearly as well without Katter’s help at the previous election. Even the much-touted successes of Katter’s Australian Party at last year’s Queensland election involved it a) absorbing probably transient protest votes which formed part of the huge swing against Labor, and b) electing two members who could just easily have won their seats as independents. Katter’s constituency would evidently prefer that he stick to being an independent local member, and limit his broader ambitious to bequeathing the family firm to his son.
As well as witnessing an explosion in the micro-party vote, the election has at the very least seen the rate of informal voting maintain the peak scaled at the 2010 election. Limiting it to ordinary election day votes to ensure we’re comparing apples with apples (pre-poll and postal voters being generally more motivated and hence less prone to informal voting), the informal vote rate has progressed from 4.18% to 5.82% to 5.92%. Presumably the Australian Electoral Commission will be conducting a ballot paper study to let us know how much this is down to proliferating candidate numbers leading to inadvertent mistakes, and how much to disaffection leading to deliberate spoilage of ballot papers.
If I do say so myself, my BludgerTrack poll aggregate performed rather well. The Coalition’s two-party preferred vote is at 53.15% on current counting, which is likely to edge up towards the projected 53.5% as the remaining votes come in. Better yet, there’s a good chance the state seat projections will prove to have been exactly correct, allowing for the fact that the model did not accommodate non-major party outcomes such as the possible wins for Clive Palmer in Fairfax and Cathy McGowan in Indi. No doubt this is partly down to luck. There was some imprecision on the primary vote, with the Coalition about a point too low and the Greens about a point too high (though the model in fact scaled down the latter from the pollsters’ published results), with the circle being squared by a preference allocation method that proved over-favourable to the Coalition, based as it was on the 2010 election result (although I’m pretty sure it still performed better than a method based on respondent allocation would have done).
Nonetheless, the model was certainly successful enough to confirm the wisdom of its basic premise that the best way to read the campaign horse race is to a) only pay attention to large-scale polling, i.e. national and state-level results, b) adjust pollsters for bias according to their past performance where sufficient observations are available from recent history, c) instead use the pollster’s deviation from the aggregated poll trend where sufficient observations are not available, and d) weight the results of each pollster according to how historically accurate/consistent with the trend they have been. As to the performance of the polls themselves, I’ll have a lot more to say about that when all the votes are in. In the meantime, here’s a broad brush overview from Matthew Knott at Crikey.