A Token Post About Modelling The 2019 NSW Lower House Election

NSW (COALITION 52 ALP 34 GREEN 3 IND 3 SF+F 1)
Recent state polling suggests a hung parliament (approx 44 Coalition 41 ALP 3 Green 3-4 Ind 1-2 Shooters) - but there's hardly been any of it!

Update Monday 11/3: Polls over the long weekend (a uComms/ReachTEL at 51-49 to Labor and a Newspoll at 50-50) have been completely consistent with the assessments below.  

Rinse and repeat ... another state election is only weeks away and there's been virtually no public polling.  At this stage in the 2015 election cycle there had been five statewide polls, but so far this year there have been just two.  Perhaps they will come thick and fast in the next two weeks, but I have so little hope of that that I think the best I can do is write an article complaining about the lack of polling in the hope that my article becomes out of date as soon as possible.

Anyway, to briefly poke my head through a gap in what is often a very busy few weeks of the year for me (the dreaded "AGM season"), I thought I'd post some comments about where things might be at if the very limited public polling we've seen this year is anywhere near accurate - which it may not be.



State polls this year and late last year have either shown the 2PP contest as even, or shown Labor with slight leads.  The two this year were a Newspoll (50-50) and an Essential (51-49 to Labor).  For the purposes of this article I'll split the difference and call it 50.5 to Labor.  With the Essential poll the youngest, and it nearly a month old, this might well be totally wrong by now, but the feeling from the noises being made by the parties is they don't think all that much has changed.

The Daily Telegraph had a couple of YouGov-Galaxy seat polls of East Hills (50-50, 0.4% swing to Labor) and Ryde (53-47 to Liberal, 8.5% to Labor).  A great deal was made of very little with the report trying to read something into a 35-31 lead in voters saying they were more likely (rather than less likely) to vote Liberal on account of Scott Morrison's performance as PM.  Aside from this being not even reliable evidence (given margin of error issues) if seat polls behaved like random samples (and they're not that good anyway), these sorts of questions are generally useless.  Voters often respond to them in a partisan fashion, or even if answering honestly often aren't much more or less likely to change their vote based on the factor, or are mistaken in thinking they will end up giving it a moment's thought.  However, the Coalition have been making upbeat noises about East Hills off the back of migration concerns, though holding it last time seemed remarkable enough.

The standard model I use for trying to work out state elections is a conditional probability model.  It asks, for a given 2PP, what would appear to be the chance of each side winning the 2PP contest in a given seat.  Of course there are many seats where a non-classic contest either excludes the seat or means that the output is of dubious value.  Because the model only knows about objective factors (margin, sophomore effects, retirements, public seat polls at a very low weighting), it will often be inaccurate in specific seats.  It also assumes that the swing per seat will be random but normally distributed around the state swing (in contrast to the pendulum model, which assumes the swing will be the same everywhere.)  At this election, there's a strong expectation that the swing will be bigger in the bush where the Nationals are apparently in so much trouble that they have run away from their branding - however, there are some areas (like the far north) where this might not apply.  I've kept the variation in seat swings at a relatively high 4% for this reason.

For now, I'm just going to show what the top side of the Coalition part of the seat model currently looks like for an assumed 2PP of 49.5%:

The columns are, respectively, the current Coalition vs Labor 2PP, the predicted 2PP for a statewide 2PP of 49.5%, and the predicted probability of the Coalition winning the 2PP on that basis.

A number of seats where the probability is especially likely to be misleading have been flagged as having one or more caution factors:

* Lismore: because there is a view that coal-seam-gas related anger at the Coalition, seen in the 2015 election, might have gone away now, and hence the monster swings seen at the past election might be followed by below-average swings or a swing back.  (If not, this vacant seat is toast, but the model's output should be seen as overconfident).

* East Hills: for reasons discussed above and also because of the scandal-prone reign of the retiring incumbent (though that sort of thing often doesn't affect a replacement)

* Monaro: because of John Barilaro's rise to the leadership, as a result of which one would expect the seat to be porkbarrel central.

* Barwon: because the Nationals appear to be savagely on the nose in this seat, though the big threat appears to come from the Shooters, Fishers and Farmers

* Coffs Harbour: because of reports of high swings in internal polling.

This model may help make sense of one of the issues noticed in the Sportsbet betting odds.  Although Labor have been very competitive on the headline market, there have been few seats where they are favourite to win seats.  The reason for this is that while a small number of seats on small margins look likely to go, there are a swag of Coalition seats that are vulnerable.  Currently the Coalition is slated to lose Coogee, East Hills, Lismore, Tweed and Upper Hunter, all of which are also losses in my model, though more confidently so in some cases because the model is conditional on the swing being of a certain size.  The Coalition is also seen by the markets as at serious risk (Labor at $3 or under) in nine seats, while Labor is only seen as at such risk in one seat it holds (Port Stephens).

The main reason the Coalition has a lot more seats at risk is the unusual nature of the pendulum, which has that brace of twelve seats on 6% to 10% margins.  If the state swing is approaching 5%, random variation suggests that some of them will go, but we cannot know which ones.  As a result, adding up seats on a betting exchange to see who is favourite in how many seats is not an accurate guide to "what the market thinks".

For a 2PP of 49.5%, my model has the Coalition losing about seven of their 52 2PP seats to Labor (or perhaps in the case of Lismore, the Greens) leaving them with about 45.  It has Labor possibly losing one somewhere, but probably not, pushing Labor up to 41 (assuming Labor, not the Greens, would win Lismore).

From here some comments could be made about the seven seats held by other parties or independents:

* There is no evidence of risk to the independents in Sydney or Lake Macquarie and both are on big margins so I assume they will retain.

* The 2PP shift to Labor is mainly a result of the Coalition losing votes to Ind/Others, and to a lesser extent to the ALP.  Although the Greens' primary vote is down a bit on last election, it's hard to see anything dislodging them in Newtown; Balmain (4.7% Green vs ALP) is less safe but doesn't seem like

* There are two seats that have been disrupted by the Coalition losing them to a third party and an independent.  The previous safe nature of these seats on a 2PP basis is completely useless in modelling the margin because of this.  The arguments for Orange returning to sender are stronger than Wagga Wagga because of the very narrow by-election margin in Orange, but we don't have enough evidence to say that seats won by minor conservative parties behave the same way as seats won by opposing major parties at by-elections.  With the Nats being so much on the nose, there's a general view that the Shooters might hold Orange, but I don't think we can be too confident on that one.

* The biggest chance of a government gain from the crossbench appears to be Ballina, on the basis of the size of the swing there last time and the view that it was caused by temporary factors, and the problems the Greens are having in NSW.  But Ballina is not only underwater against the Greens, but against Labor as well (Labor won the 2PP there 53-47).  The Greens also have a history of doing better in seat defences than the state or federal swing implies, and are yet to lose a single-member seat anywhere that they won at a previous general election.

While the government might on balance recover, say, one seat from the crossbench if it's lucky, it is also in enough trouble in others to suggest that on average there will be losses elsewhere.  The main problem is Barwon (Shooters) but there is also risk in Murray (Shooters), Tamworth and Dubbo (rural independents) and North Shore (independent Mayor campaigning against a weak incumbent - though I'm not sure Felicity Wilson's credibility issues are that interesting to voters generally).

So off a 2PP of about 49.5-50.5 a possible outcome would be a parliament with about 44 Coalition, 41 Labor, 3 Green, 3-4 indies and 1-2 Shooters.  Given that the Greens will not support the Coalition, that would be basically a tie, and someone would probably end up taking government with a crossbench agreement that gave them confidence and supply by 1-2 seats.  So there's good reason to believe that this is very close.

What does an easier Labor win look like?  It probably looks like getting a bigger swing, or else getting lucky in the Coalition's 6-10% range, and then most likely forming a minority government that depends on the Greens or one or two indies (but doesn't need both).  Even if there is a blowout to, say, 53-47 in Labor's favour, my model still doesn't think that's enough for majority government unless they can rope in Ballina and Balmain.

What does an easier Coalition win look like? Probably it looks like the huge decline in Coalition primary vote since 2015 (see Bludger Track) going mainly to Shooters, One Nation and indies in rural seats that are unloseable to Labor.  It would also require the swing in the Liberal seats in the crucial 6-10% range being kept down because they are not where the primary vote action is.  If the Coalition can save one or two of the six very close seats, and get run close by indies and Shooters in bush seats without losing too many of them, then it might come out still in or only just short of majority.  If it can hold losses to, say, seven - and lose to Shooters in preference to independents - the Coalition could be in a strong position, since it is hard to see how a Labor/Greens/Shooters alliance would be stable.

Of course, things might have shifted - let's see what we get by way of state polls this week!

(As for the Upper House, I hope to have a look at that this coming week sometime too.  The most interesting aspect of the ballot draw was the Liberal Democrats drawing to the right of the Liberal-National ticket, meaning that David Leyonhjelm will have to fight for a seat.  Beyond that, the impacts of the ballot draw are probably being overstated.)