January is mostly a quiet month, so I’ll take the opportunity from time to time to recommend some interesting reading material. First up, a post from just before Christmas by Kevin Bonham that explains the importance (or, more often, unimportance) of parties’ decisions about preferences and tries to dispel some popular myths.
If you’re at all interested in Australian elections, please read the whole thing. Bonham is one of our best psephologists, and he has little patience with those who peddle nonsense and spread confusion.
His basic point is that how parties direct their preferences is less significant than is popularly believed. He estimates that changes in the flow of preferences at federal elections over time have had only about one-tenth of the effect of changes in the primary vote of the major parties. For any particular minor party, the way that its preferences flow rarely changes much, and generally doesn’t respond very well to what its how-to-vote cards say.
The particular relevance at the moment is with talk of the supposed threat posed to Labor by preferences from right-wing minor parties, especially Clive Palmer’s “United Australia Party”. Nick Dyrenfurth, head of the John Curtin centre, a Labor right think-tank, argued earlier this week [link fixed] that “the spectre of Clive … looms large over the election,” and that his party’s support is “a game changer in close contests if its 2019 preference deal is replicated.”
This is exactly the sort of claim that Bonham is trying to rebut. As he puts it:
The claim that Clive Palmer delivered government to the Coalition via United Australia Party preferences is a widespread myth on Twitter, spread mainly by accounts that seek to blame Palmer for the result rather than blame the basic fact that Labor’s primary vote was way too low.
Dyrenfurth compares Palmer’s party to the DLP, which used to hurt Labor by attracting natural Labor voters and inducing them to give preferences to the Coalition. It’s not impossible that Palmer may do that to some extent, but there’s no real evidence for it. Bonham again:
We don’t actually know what percentage of UAP voters even follow Reps how-to-vote cards but I suspect from the Senate it is quite low … The 2019 UAP flow can be explained just fine by (i) The UAP attacking Labor heavily and therefore being more attractive to voters who disliked Labor anyway (ii) The UAP’s natural support base including outer-suburban demographics that swung against Labor …
If people vote for a right-wing minor party and then preference the Coalition, it’s most probable that in the absence of that minor party, they would just have voted for the Coalition outright. Whether or not those people were Labor voters in the first place is irrelevant: what matters is not where the votes were coming from, but where – absent Palmer or the like – they were going to.
This brings us back to a point that I’ve made a number of times before: debates about preferences that are couched in terms of tactics are in fact usually about ideology. Dyrenfurth is not arguing in a vacuum; he wants Palmer to be important, because he wants to move Labor’s position closer to that of Palmer’s voters.
He and others like him want Labor to stop talking about corruption and climate change and focus on more traditional working-class issues. That may or may not be good politics, but it’s something that’s independently motivated, not a consequence of some detached analysis of preferences.
We can see that from looking at the flip side of the preference argument: if a strong vote for Palmer is hurting Labor, then a strong vote for the Greens must equally be hurting the Coalition, by luring away its natural supporters and delivering their preferences to Labor. In fact there’s more reason to believe this than in the Palmer case, since the Greens vote is both larger and more disciplined. But don’t expect to hear that argument from the folks at the Curtin centre.
5 thoughts on “Understanding preferences”
I think you have accidentally repeated the link to Kevin Bonham’s piece at the point where you intended to insert a link to Nick Dyrenfurth’s piece.
What does Nick Dyrenfurth mean by ‘traditional working-class issues’? My conception of traditional working-class issues would begin with higher wages and shorter hours: ‘Eight hours of work, eight hours of play, eight hours of rest, and eight bob a day’ (only numerically updated for modern relevance). Am I doing Nick Dyrenfurth an injustice when I suspect his conceptions are different, and less well-founded in historical evidence?
Thanks J-D, you’re quite right about that link – not sure how that slipped thru. Now fixed.
As to exactly what Dyrenfurth wants to focus on, “traditional working-class issues” is my phrase rather than his; he refers to “bread-and-butter Labor politics” and specifically instances concerns about “worsening economic outcomes, quality of life, housing affordability and healthcare.” The relationship between those things and wages/conditions, and what public policy can do about the latter, are the sort of things that a Labor think-tank maybe should be looking at.
I don’t think Palmer’s preferences mattered, but the $50 million he spent bagging Labor goes a long way to explaining the contrast between the disastrous federal election and Labor’s string of victories at the state level before that.
Yes, Bonham concedes that that’s very possible, but it’s a difficult thing to prove. I’m always pretty sceptical about the effectiveness of advertising of any sort.
One way to examine this is to use a simple model to “predict” how the state elections from 2017 – 2019 should have gone given historical trends, and compare this to how the elections actually went.
One model for this is the federal drag + age model. Here, I’m using Dr Bonham’s old model from 2014 – so it was developed before those elections (https://kevinbonham.blogspot.com/2014/07/what-kills-state-governments-age-or.html):
Dr Bonham’s regression formula (post-1989 sample):
Expected size of parliament change =.0039+.133*(fed govt different)-.019*(age of state govt)
fed govt different = binary variable, 0 if same major party as fed govt, 1 if not
age of state govt = age of state govt in years
Going through each state election:
> 2017 WA: 0.0039 + 0.133*0 – 0.019*9 = -0.1671 (expected seat swing against govt of 16.7%)
> 2017 QLD: 0.0039 + 0.133*1 – 0.019*2 = 0.0989 (expected seat swing to govt of 9.9%)
> 2018 TAS: 0.0039 + 0.133*0 – 0.019*4 = -0.0721 (expected seat swing against govt of 7.2%)
> 2018 SA: 0.0039 + 0.133*1 – 0.019*16 = -0.1671 (expected seat swing against govt of 16.7%)
> 2018 VIC: 0.0039 + 0.133*1 – 0.019*4 = 0.0609 (expected seat swing to govt of 6.1%)
> 2019 NSW: 0.0039 + 0.133*0 – 0.019*8 = -0.1481 (expected seat swing against govt of 14.8%)
Comparing expected seat gains/losses to actual, and how well Labor did relative to expectations:
> 2017 WA: expected -0.1671, actual -0.339 (Labor +0.1719)
> 2017 QLD: expected +0.0989, actual +0.043 (Labor -0.0559)
> 2018 TAS: expected -0.0721, actual -0.08 (Labor performed as expected given Tas has 25 seats)
> 2018 SA: expected -0.1671, actual -0.0851 (Labor +0.082)
> 2018 VIC: expected +0.0609, actual +0.0909 (Labor +0.03)
> 2019 NSW: expected -0.1481, actual -0.0645 (Labor +0.0836)
That averages out to approximately a +0.0534 overperformance by Labor.
> The model is primarily focused on govt performance; for Labor oppositions I’ve attributed the entire seat swing against the govt to the opposition. However in some elections the govt primarily lost seats to the crossbench instead of the opposition; an example is NSW 2019 where just 2 of the govt’s 6 seat losses went to Labor.
> The model looks at seat performance; which means that states like SA where electoral geography naturally favours one side of politics and/or where significant redistributions have happened can mess up the model. Additionally TAS uses a proportional-representation system which naturally leads to very different patterns of seat swing than single-winner systems.
I should also note that in that same piece, Dr Bonham finds that federal popularity influences state results for govts of the same party (so 2017 WA, 2018 TAS and 2019 NSW). However, given the massive error in 2019, I did not attempt to model this as it’s entirely possible that the polls were wrong for the entire 2016 – 2019 term.