Saturday, June 3, 2017

Modelling The Queensland Election Off The Senate Election And Polling

For most aspects of the Queensland state election, we can use the usual modelling tools: the pendulum, two-party preferred polling, seat win probabilities based on expected margins and so on.  Whoever wins the two-party preferred vote by a substantial margin should win the election, while if the 2PP vote is very close there might be a hung parliament, but won't necessarily be.

The difficult modelling task is to try to work out what seat share the high One Nation vote in recent state polling converts to.  If the One Nation vote share keeps falling, this will soon be a non-issue.  But for the time being, let's assume it doesn't.  It's also worth having a look at whether the Greens have any chance of winning seats in a state where they have yet to win a state seat in their own right.

In the previous instalment, I reckoned that Pauline Hanson's One Nation's current level of polled state-election support in Queensland (17%) probably wouldn't be good for all that many seats.  I suggested that if the current primary vote levels were applied to the seats as they stood in 2001 (ie as swings from the 1998 election, One Nation's last big success), 17% would only have been good for about three or four seats.

That was a rather crude way to model things, given the redistributions and passage of time since then.  So I suggested booth-by-booth modelling off the 2016 Senate results (as the most recent indicator of how One Nation support breaks down by seat in a contest with statewide results) could be a useful way to go, but that I didn't have the computer skills to do it.  Fortunately for us all, Alex Jago does, and he's prepared an excellent resource that estimates the 2016 Senate votes by post-redistribution state boundaries, considering only how the Senate voters ranked four parties: Labor, the LNP, One Nation and the Greens.  See the link for discussion of the data issues involved in doing this.

Although the primary Senate booth vote for parties other than these four was a whopping 22.2%, most of those voters preferenced at least one of the "big four", with only about 4% of all formal votes reaching none of these four parties.  These won't be exactly the same voters as the 6-7% whose primary preference is currently "others" - indeed there might not be that much overlap between them - but neither of these are particularly large vote shares, so I suspect we can treat them as the same.  The breakdown of Senate preferences from those who put "other" first appears to have been fairly even by party - I get about 26.4% to One Nation, 22.1% LNP, 17% Green and 15.4% Labor, with the rest to exhaust.  The current Others voters will probably preference a bit like that too - a slight conservative lean, but nothing massive.

The data also shows us something useful about how Greens voters see a choice between Labor, One Nation and the LNP.  Basically they don't like One Nation any more than they like the Coalition, with 54% preferencing Labor ahead of the other two, 5% the LNP, 4% One Nation and 37% none of the three.

It's also important to bear in mind that the Labor vote shares in the Senate in Queensland in 2016 were pretty low, which is why the LNP wins the two-party Senate vote in so many seats.  This is consistent with a House of Reps result in which Labor only scored 45.9% two-party preferred in the state, compared with 51.1% at the 2015 state election, and similar levels in recent state polling.

How would One Nation win seats?

There are two major groups of possible One Nation seat wins - seats where the final two parties are One Nation and the LNP with Labor eliminated, and seats where the final two parties are One Nation and Labor, with the LNP eliminated.

We don't know how preferencing strategies will play out exactly for the election just yet, but under compulsory preferencing it's likely that Labor will generally put One Nation last and then try to attack the LNP if it preferences One Nation ahead of Labor.  Whether the LNP will decide to preference One Nation anyway in search of One Nation preferences, and risk the constant distraction of being blamed for preferencing every fruitcake who gets through PHON's hopeless candidate screening, is yet to be seen.

Beating Labor on LNP preferences is easier because these tend to be more helpful.  In the 1998 election, Liberal preferences (8 seats) split 52.2% to One Nation, 22.6% to Labor, 25.2% exhaust. Nationals preferences (14 seats) split 60.8%-19.1%-20.1%. (Source: Antony Green) As an average, if this sort of flow based on a deliberate decision to preference One Nation is repeated, One Nation might see preference flows off the LNP of around 69 to 74% vs Labor, depending on how one reallocates the 1998 exhaust.  Maybe this could even be a bit higher, based on the (spurious) "One Nation aren't so racist anymore" defence used by Coalition figures who support preferencing One Nation.

On the other hand, if the LNP doesn't preference One Nation, it will probably be a lot lower.  At the Senate election, 58% of Queensland LNP voters exhausted their vote without it reaching either Labor or One Nation.  Those whose votes reached one of the two split virtually equally (though those whose votes reached Labor tended to number more boxes, for what it's worth.)  I doubt that we will see an equal split at the state election, but even 60% can't be guaranteed without knowing the preference strategies.

For Labor preferences, we see a similar picture in reverse.  In 1998 Labor preferences split 46.2-17.9 to Coalition forces against One Nation, with 35.9% exhausting.  That's a 72% flow if exhausted votes are ignored, or 64% if they're split evenly.  Again the "they're not so evil anymore" argument provides some reason to suspect that flow might be more even, but on the other hand the lack of an option to exhaust a vote might result in many Labor supporters following the card in the LNP's favour.  At the Senate election, most Labor votes preferenced neither PHON nor the LNP, with those preferencing one or the other breaking slightly to the LNP (about 52-48).  There was also one House of Reps preference flow measured, in the seat of Maranoa, where Labor preferences split only 50.3% to the LNP against One Nation although the LNP were preferenced on Labor's card.  Maranoa is probably not typical, being a sprawling seat where Labor isn't competitive.

All up then, Labor preferences might favour the LNP strongly over One Nation or they might not, but they are very unlikely to help One Nation.  LNP preferences might favour One Nation strongly over Labor or they might not, but they are unlikely to help Labor.

Possible One Nation Target Seats

I'm assuming One Nation won't win seats where the LNP beats it on primaries, whether it beats Labor on primaries or not.  (The possible exceptions would be seats where a fifth party, like KAP, does well on primaries but is eliminated behind One Nation.)  As a starting point for a model of the Queensland state election off the 2016 Senate election combined with current polling, then, I'm using the votes that reach each party first (columns BT to BX in Alex's spreadsheet) and applying a swing to get them to the most recent Galaxy poll (which was discussed further in the previous article.)

Ignoring state-level candidate effects I've treated One Nation as competitive if the Senate result as adjusted for current polling has them within five points of the LNP primary, and also has the modelled Labor primary below 50%.  I've assumed One Nation get 24% of Green votes if the final contest is One Nation vs Labor (this is modelled off the Senate results, and is probably too generous) and an arbitrary 55% of Others votes in the same case (this on the other hand may be a bit stingy.  Or not.)  On this basis I've estimated the share of LNP preferences One Nation would need to beat Labor.
Ordered by the projected One Nation primary margin against the LNP, these are all the seats where the polling-adjusted model has One Nation within five points of the LNP's primary based on current polling, and the projected share of LNP preferences they would need to win if they did make the top two (which in many cases on the primaries given, they would not.)

Here are the takeaways from the chart above:

1. The seats One Nation most threatens to win at present are marginals.  No safe LNP seats appear on the above list, and where safe ALP seats appear, One Nation may get into second but the preference flow required appears too steep.

2. The model only has One Nation ahead of the LNP in three seats where the preference target is winnable.  However none of them are laydowns.  In Lockyer, if Labor underperforms and comes third, One Nation might lose on Labor preferences.  (I think this is unlikely with the LNP incumbent quitting though.)  Traeger looks highly winnable for One Nation except that it is occupied by Katters Australian Party, who surely won't be going quietly, and Mirani looks a chance but the preference flow required is no sure thing.

3. Another seven seats are flagged as winnable if One Nation can do better on primaries than the model projects, which would also help reduce the preference targets.

4. Because the key target seats are Labor-LNP marginals, the model isn't conducive to deliberate tanking in specific seats by either major party.

Overall though, two seats with One Nation as favourites and another eight as possible but not that probable (plus the chance of the odd seat here and there that will completely buck the model) isn't that crash hot off 17%, and isn't any better than my model based off 1998 comparisons.

And if One Nation's vote share falls further - which only a fool would bet against - then it is going to be very hard for the party to get more than the odd seat.

As with my previous models, this one also suggests that there is a massive difference between One Nation on 17% and One Nation on 23%.  Using the polling at the time One Nation had that vote, the model places them ahead of the LNP in 23 seats.  Discounting four where the preference flow would be much too hard, and all six where Labor comes third, still leaves 13 probable wins.

The inevitable disclaimers apply.  Sitting member effects, good and bad candidates and different issues mixes will create different votes in particular seats to what a Senate and polling based model suggests.  The seat outlooks are examples only and could be wildly wrong in particular seats.  But they do underline a general point: the current level of One Nation polling is probably only good for a handful of seats.  If One Nation were exceptionally unlucky, they might even poll 17% and win nothing.

What Chance The Greens?

There has been quite a bit of coal-powered noise about the Greens' chances of picking up South Brisbane, and a little too about McConnel (nee Brisbane Central).  In either case the winning plan would be to intensify the inner-city primary vote, beat Labor into third on primaries and then beat the LNP on preferences.  Generally in such contests flows from Labor to Green are as good as Green to Labor, or even slightly better because more Labor voters follow how-to-vote cards.

If we look at the four-party-preferred Senate votes in these two seats, the Greens were very close to Labor in both of them (0.6% behind in McConnel and 1.8% behind in South Brisbane).  However, the Greens' recent state polling is not really holding up with their federal result (considering the amount of Green vote at federal elections that is hidden in the primaries for left-wing micros). Meanwhile Labor's primary was in the context of a heavy statewide two-party loss to the Coalition.  If the state election turns into a landslide for the LNP and the Green vote picks up then we might see something more competitive here, but for the time being, the model has the Labor-to-Green gaps blowing out to 10 points in McConnel and 11 in South Brisbane.

But there's more.  The federal model suggests that Labor should run about seven points above its statewide 2PP in South Brisbane and a point or two behind it in McConnel, but Jackie Trad (as an incumbent) actually beat Labor's statewide 2PP in 2015 by 12.7 points and Grace Grace beat it by 2.1.  Some of this is probably down to personal votes for Trad at least and probably Grace as well, but another factor here was the inner-city appeal of Malcolm Turnbull at the 2016 election in the Brisbane federal seats.  The Turnbull factor - oh yes he once was an electoral asset somewhere - would probably have taken a lot of federal primary votes from Labor.

The evidence suggests that on anything like the current polling, the Greens won't win these seats.  Labor's recent Adani announcement, which may have been partly aimed at shoring up inner-city electorates against a Green surge, shouldn't do them any favours either.  Inner-city intensification of the Green vote is something that we have seen a lot, but the fact that Trad beat the Greens on primaries by 21 points in 2015 and Grace by 23 should make us extremely cautious about Green prospects of actually winning either seat this time.

On paper, the Greens' least unrealistic target seat seems to me to be Maiwar (LNP vs ALP 3.0%).  Maiwar is made up of parts of Indooroopilly and Mount Coot-tha.  The LNP incumbent for Indooroopilly, former transport minister Scott Emerson, is contesting Maiwar, but the Labor incumbent for Mount Coot-tha, environment minister Steven Miles, is out of there and running for Murrumba instead.

The primaries in 2015 in Maiwar (Antony again) were LNP 47.9%, Labor 29.1% and Green 20.4%. This is almost identical to the results in the Victorian state seat of Prahran in 2010; four years later Prahran fell to the Greens.   Moreover, Alex Jago's data shows the Greens actually outpolling Labor in Maiwar in the Senate in 2016 (by two points on a four-party basis).  And as concerns the Mount Coot-tha part, whatever personal vote Saxon Rice may have picked up in a term is gone.

Labor seem to be wary of the danger this time, picking a candidate whose skills seem to cover off neatly against those of the Greens contender.  It's also difficult to see the seat going anywhere if there is not a statewide two-party swing as there was in Prahran.  And if there is a swing to Labor statewide, I think they'll be quite keen to pick off Maiwar for themselves.  But still, a three-way contest worth keeping an eye on.

Once again, huge thanks to Alex Jago for extracting, processing and posting the Senate projections.


PS: As I was writing this, another ReachTEL rattled out with the LNP on 35.3% (after ignoring the "undecideds" pending any detail on their breakdown), Labor on 31.9, One Nation 17, Greens 9.4 and others 6.3. 2PP 51-49 to LNP.  As I noted in the previous issue, there's been some scepticism about recent ReachTEL state primary vote readings, but let's assume this is real - does a weaker vote for Labor make a difference to the model?

The answer so far as One Nation is concerned is "nope".  If anything, a weaker Labor vote is just bad news for them - it means a stronger LNP vote (all else equal) and makes it more likely that close fights for second will resolve in the LNP's favour.

PS 1 July: Alex has advised a minor error in the spreadsheet previously linked to, such that the ALP vote should have been 0.44 points higher overall.  This makes very little difference to the rough conclusions I am drawing about the likely seat scope of a PHON vote around 17%, which are a lot rougher than that on a seat by seat basis anyway.


  1. On the point of Green targets, I wouldn't rule out Noosa, where they have come second the previous two elections. Whilst I doubt they can win, if the Libs get a little too close to One Nation then small l voters could well move away from the libs with One Nation eating away at the Libs and Labor primary votes cementing them in second whilst weakening the Libs primary vote. That being said the Greens could well end up second in Maiwar, which isn't unrealistic. I Doubt their chances in South Brisbane where Trad has quite a reasonable personal vote, in Mcconnel i see them finishing third behind both the LNP and ALP.

    1. Noosa is an interesting case because the Greens have performed strongly there at state elections but their performance there in the Senate election was not so strong, coming in at about 15% 4PP. Perhaps the area was not as heavily targeted in federal campaigning. Probably at LNP 6.9% it is not marginal enough to make it an easy target for either Labor or Green but it is also worth keeping an eye on.

  2. I think Labor are winning at the moment. Those ads about Campbell Newman are sticking but I think if Labor want a decisive win they need one step more. The LNP are on the back foot though. The LNP advertisement is very bland with them just sitting at the table and really I didn't remember what any of them said because it was a nothing ad. The LNP definitely need to pick up their game because Labor are one more strong advertisement away from a big win.