Saturday, November 18, 2017

Queensland: A Hard Election To Model

Primary vote aggregate Labor 34.7 LNP 32.8 Green 8.8 One Nation 17.3 Other 6.4
If Newspoll/Galaxy preference assumptions are correct Labor should just win majority on current numbers (Projection: Labor 50 LNP 37 PHON 4 KAP 2)
If ReachTEL preferences are accurate LNP may win, though ReachTEL released 20 Nov is less clear on this

Note: Live comments on Queensland elections here on Saturday night.

It's taken me until the week before the election to get around to posting any analysis during the campaign for the 2017 Queensland election.  This is partly because of an unusually severe version of the usual problem: I've been extremely busy and there are just not enough of me to do everything I'd like to do.  It's also because this election's very challenging to model.  And, as I noted previously, the big picture isn't much help either.  The government has been chaotic, but the federal Coalition's turmoil is a massive burden for Tim Nicholls' LNP opposition.

Most of the difficulty comes from the preference problem.  A divergence of around four points in polling exists between the ReachTEL polls on the one hand and the Galaxy/Newspoll stable (and also Essential) on the other.  About three points of that, on the latest figures, is caused by preferencing differences.  ReachTEL is using purely respondent preferencing, while the others are using some combination of previous elections (Essential) or assumptions with an undisclosed and unclear basis in evidence (Galaxy/Newspoll).  The previous election had much less significant One Nation presence and also had optional preferencing, so there is really no credible baseline for "last-election" preferencing.

A further issue with Queensland has been the dearth of statewide polling.  In the last six weeks there has been only a Newspoll (17 Oct), a Galaxy (1-2 Nov) and three ReachTELs (two commissioned ones both on 13 Nov and an earlier one for which the 2PP was released but the primary votes were not.)  The Newspoll and Galaxy both had Labor ahead 52-48 and all the ReachTELs had Labor behind by the same margin.

Aggregating all polls that have released full primaries based on polling conducted entirely since September, but with a 25% downweighting for commissioned polls, and with greater weighting for November (3) than October (2) and September (1) I get a reading of:

Labor 34.7 LNP 32.8 Green 8.8 One Nation 17.3 Other 6.4

By Newspoll's preference assumptions this would result in a 51.8% 2PP for Labor, a 0.7% swing to the government.  But by ReachTEL's respondent preferences it would result in about a 51-49 result to the LNP.

A small number of seat Newspolls released today was reported favourably for Labor, because it showed them gaining Whitsunday and competitive in Gaven, while holding Ipswich West and Mansfield.  However the flip side of the reporting is that the poll also had Labor losing Bundaberg and losing Thuringowa (an otherwise vaguely safe seat) to One Nation.  Also as Whitsunday was 51-49 to Labor and Gaven 51-49 the other way, the chance one would give them in Gaven is the chance of them not gaining Whitsunday.  So, a very threadbare conclusion on the evidence in question, which based on my record in calling out sloppy Queensland poll reporting probably means the conclusion is true.

The overall average of the eleven Labor-vs-LNP seat polls released by the Galaxy/Newspoll stable has been a swing of 0.9 points against the government, with three projected losses, one projected gain and two 50-50s in seats held by the LNP.  Throw in the projected losses in Thuringowa and South Brisbane (the latter of which I don't take all that seriously) and the seat polls do not in fact show Labor in good shape.  Seat polling is an enterprise in crisis, but the least we can say is that the seat polls at best add nothing positive for Labor to what we know about the state picture.

One Nation

Currently the only seat where I can find any seat betting that has One Nation favourite is Buderim, the seat of their state leader Steve Dickson.  However, Buderim is not strong ground for One Nation historically and it isn't clear that having a state leader in a seat is enough to overcome that.  In general, I have had about 19-20% as the tipping point at which One Nation start to win a lot of seats.  The polling at present has them a little shy of that.  Using Alex Jago's Senate-based calculator (the Senate preference results skew to Labor by around 1.5 points on average compared to the Newspoll/Galaxy preference assumptions) the tip from the current level of One Nation support is for five One Nation wins, but one of them is Traeger which is being left with Katters Australian Party.  The rest are Callide, Hinchinbrook, Lockyer and Mirani.  Callide is extra-winnable because of the loss of Jeff Seeney's personal vote (and One Nation have shown up as polling competitively there) while the others listed haven't had any released polling I'm aware of.  Mirani is one that's cropped up in these Senate simulations but for whatever reason isn't on the frontline of betting.  With so many vagaries in converting Senate to state results it's deeply unlikely they would win exactly those seats but the simulation should give a fair idea of about how many seats they are good for, even if not those exact ones.

So at the moment One Nation will probably win a few seats, but probably won't win a lot.  They are not far short of the winning-lots level but would probably have to surge a few points in the last week to get there.

There's an impression in the seat betting (which is not predictively reliable, but in this election, what is) that One Nation are more of a nuisance to the LNP than Labor in seat terms.  Of the 21 seats One Nation are at $5 or less on one exchange, 12 are seats the LNP would be expected to win, including 4 of 6 where they're at $3 to win.  However, because the LNP probably will form government with One Nation if they have to, and Labor are very unlikely to, losses to One Nation are a lot more painful for Labor.

Labor's Task To Retain

I count Labor as starting with 49 notional seats, including three notionally Labor seats occupied by the LNP, and also Pumicestone and Cairns which are technically on the crossbench.  I count the LNP as starting with 42, including Nicklin which is being vacated by an independent, and Dickson's seat of Buderim (won by the LNP last time).  Two seats are held by KAP.  Ignoring any possible losses to the Greens, Labor's target for forming majority government is 47 seats, a loss of two notional seats.

A fair amount has been made of the fact that the One Nation strategy of preferencing against sitting members harms Labor more than the LNP, since in many LNP seats One Nation could come second anyway making its allocation irrelevant.  However, there are quite a few close seats where One Nation aren't running or won't poll much.  I don't find that it makes a huge difference in terms of the number of expected 2PP wins for each side, and the fact that a lot of One Nation's LNP-held targets are well up the LNP side of the pendulum means that the LNP must have been expending effort in seats it would normally not need to bother defending.

Against that aspect, there is an advantage to Labor, and that is personal vote effect.  Labor MPs who won seats from LNP members at the last election will have obtained personal votes, whereas the defeated LNP members' personal votes will have been factored into the LNP's 2015 vote in their seats.

I expect personal votes to count for less this election than normal.  Firstly, most of the new Labor MPs defeated LNP MPs who had only been in for one term themselves, and hence would not have built much of a personal vote up themselves.  Secondly, redistributions mess with personal votes, and the Queensland one which has changed 89 seats into 93 would have had large impacts on them.  Nonetheless, sophomore effect is a theory and a fact, and Labor would have won outright last time but for it.

After factoring all these things in, if there is no swing I have Labor winning the 2PP contest in, on average, 49 of the 91 non-KAP seats (in other words, no change).  For the 0.7 point swing projected using Newspoll's preference assumptions, I get 51 Labor 2PP wins (so, say, something like ALP 50 LNP 37 PHON 4 KAP 2, if no Greens get up).  For the 3.1 point swing to LNP that ReachTEL keep getting, however, I get 51 LNP 2PP wins.  And obviously if it's somewhere in the middle things get very close.  (I personally slightly prefer Labor's chances at this stage.)

Bearing in mind that the majors are each likely to drop a seat to a few seats to One Nation, it's still quite possible either side can win this outright.  However, if the 2PP lands somewhere around 50:50, a hung parliament becomes a lot more likely.

The Greens

The Greens have not had the success so far with inner-city Queensland seats that they have had in NSW and Victoria.  However the Adani coal mine proposal gives them one hell of a trump card to deploy in such areas.  The Greens will be energised in the final week by their party winning the Northcote by-election in Victoria - a win that was no great surprise, but the margin (approaching 56-44) was a big one.

I have continued to have an eye on Maiwar because of its Prahran-like results history, though it's debatable whether LNP voters in the suburbs in Maiwar are as tractable as Liberals are in Prahran.  A ReachTEL showing the Greens falling well short should not be treated as definitive.  There is a history of ReachTEL poll errors in inner-city electorates where the Greens are strong, of which Northcote tonight (a nine or ten point 2PP miss) is yet another case.  A Greens internal has them competitive but Greens internals were never that reliable either.

The fight for South Brisbane and McConnel (where the Greens are taking on Jackie Trad and Grace Grace) stepped up with a surprise Galaxy showing a 51:49 Green lead in South Brisbane, but the Greens are coming from a very long way back there against a candidate with a large personal vote, so I'd need to see a lot more than a 1-point seatpoll to give that one away.

So all up the Greens still have a few chances but none of them are easy.

Which Pollster Will Blink?

We've had a long run of the Newspoll/Galaxy stable on the one hand and ReachTEL on the other getting completely different 2PPs because of different preference assumptions.  Pollsters are especially careful with their final poll because it is the one they're judged most by, though I've often argued this is silly.  It will be interesting therefore to see if either pollster comes back to around 50:50 this week, or if they both stay where they are.

Update: 20 Nov New ReachTEL

After churning out endless 48-52s ReachTEL now in the final days comes out with a 51-49 for Labor, off primaries of Labour 34 LNP 30 One Nation 17 Green 10 KAP 3, leaving 6 for Others, so I assume the undecided have been addressed.   By Newspoll methods this would be about 53.3-46.7 for Labor, so this also suggests some easing in the differences between the two pollsters' preferencing assumptions, but not much.


  1. "There is a history of ReachTEL poll errors in inner-city electorates where the Greens are strong." There is also a history in Brisbane of the Greens saying "we're going to win this one this time!" and still coming third. And I say this as someone who has voted Green for the last few elections, and has joined the party and is campaigning actively this time. I still predict no Green wins. Brisbane will have its Prahran and Northcote moments in a few years, but not yet.

    1. There is a long history of the Greens getting ahead of themselves with their expectations in many different parts of Australia and it is often irritating to people outside the party. In 1992 the Tasmanian Greens had a slogan "Go, Go, Go, Green Government". Their vote went backwards considerably.

    2. I wonder if this is beginning to have a bit of a reverse effect. I too have spent years being irritated by Greens over-excitement and they certainly still do it (I remember Di Natale crowing about the possibility of two Senate seats in Queensland on election night 2016, which was never a possibility on the results), but I think it's ingrained in a lot of people to automatically doubt their claims even though they are sometimes right these days. I certainly underestimated them in Northcote and I think most people did in Prahran and Ballina. That said I can't see them winning any of their targets this year - it seems to me that making some solid non-Noosa 2PPs would be a fair result.

  2. Kevin - another aspect to consider is preferences on HTV cards. Of the four large parties, One Nation & The Greens do the worst. One Nation will be placed last by both ALP and Greens across every electorate. Similarly, The Greens will be last on LNP and ONP cards. The LNP does slightly better than the above mentioned two as it will be only second-to-last on ALP and Greens HTV's. Labor has the advantage here over the LNP - potentially solid Greens preference flows across all seats. LNP does not enjoy that kind of support from ONP, who are recommending preferences to Labor in many seats, and will not be contesting 1/3 of the seats (although I'm not sure which side it advantages by not contesting that 1/3). Even though ONP could presumably poll up to 18%, and the Greens below 10%, it is doubtful that the benefit of ONP preferences to LNP would be strong enough to counter the Greens-ALP arrangement. I'm also speculating here that the preferences of the remaining minor parties and independents will split around 50/50 between the two major parties, so no huge advantage to either side there.
    The other factor to consider is how effective the various parties are going to be in manning booths across the country, to get their HTV's into hands of voters.
    Your thoughts on all this?

    1. Labor voters did quite strongly preference One Nation in Maranoa at the federal election (it was about 50-50) despite the Labor HTV preferencing the LNP. Of course that is not a typical federal electorate and not one where Labor would have been putting in a big HTV effort, but I am not sure how strong the flows from Labor to LNP in preference to One Nation will be. Maybe 70-30ish?

  3. Its very interesting that in the only 2 seats where One Nation is leading in the electorate level polls, the LNP is preferencing Labor (Logan & Thuringowa). Since its unlikely to be coincidental, the question is what is the ALP giving the LNP in return. The ALP was already committed to preferencing ON last so it can hardly be this. The only explanation I can think of is that the ALP must have committed to ensure its HTV cards are widely distributed in the electorates where the LNP is under threat...Callide, Nanango, Burnett, Hinchinbrook, Lockyer and Gympie. If this suspicion is correct it will be extremely difficult for ON to win these seats as they'll need to finish well ahead (10%?)of the LNP to have any chance of winning. Something similar will apply in Thuringowa and Logan, nearly impossible.

    With this in mind I therefore predict that ON will win just one seat, Maryborough. After this their best chances would possibly be Keppel and Mirani on the central coast where they are not especially strong but might still have a chance of outpolling the LNP and riding in on their preferences. I had thought Ipswich West and Bundaberg were very strong chances for them but the published electorate polling shows them short. Perhaps this is why the ALP didn't request LNP preferences in these two?

    1. I'm currently having a look at the most recent polling and what it says for PHON (especially given the preferencing difficulties) but I am increasingly feeling that even my 4 seats from them for last week is too much and it won't be a massive shock if they wipe out completely or only win one seat.

  4. Here are the ONP preference lodgements:

    I believe these will deliver the ALP the seats of Whitsunday, Burdekin, Glasshouse, Toowoomba North and probably Redlands. Should also seal the deal (if it wasnt already) in Mansfield and Mt Ommaney). Of course the common refrain is that ON won't be able to hand out their own HTV cards in these seats. Don't worry! In most of the above they will gain at least 15%. It the ALP can change the flow from 60-40 against to 60-40 in favour thats a 3% swing. I'm sure the ON candidates will have had a sudden flurry of new volunteers, helpfully, if surrepticiously organised by the ALP. Quite likely the LNP is doing the exact same in Pumicestone, Mundingburra, Barron Rive, Townsville, Keppel, Mirani and Bundaberg where this situation is reversed....though they might not want the volunteers to be too enthusiastic in the case of the latter 3, lest they find their preferences distributed instead.
    Looking at the seats, the ALP definately got the much better of it...even in seats like Burdekin and Mansfield where ON could have claimed that since they were notational ALP seats they would distrubute against the ALP - they haven't!

  5. My aggregate has a rather soft weighting for the impact of PHON's preferencing decisions. If I get time tonight I'll try a more aggressive version and see what it does.

  6. Curiously the three seats my modelling (adapted from Alex Jago's) gives PHON the best chance in have had no seat polls - Lockyer, Hinchinbrook and Mirani. In all of these the LNP is in theory third.

  7. What would be the result if voting was Proportional

    1. It would depend on what sort of proportional system was used (whole state as one electorate, if so with threshholds or without, or multiple electorates each with a number of members.) However in general such a system would be better for the LNP since it plus One Nation plus KAP have nearly 50% of the primary vote combined and hence would be well placed to form government.

      The other important point here is that campaign tactics would be different. Probably the LNP would not bother preferencing One Nation in a proportional system unless it was a Group Ticket Voting nonsense one that forced them to. Also there are many candidates who ran as local independents for single seats - one to three of whom will win seats - and in a proportional statewide system those indies would have probably had to form some kind of alliance or party.