Saturday, February 20, 2021

WA 2021: How Lopsided Do You Want It?

The 2021 Western Australian lower house state election looks like the most lopsided state election for several years, and may be among the most lopsided ever.  A Premier with near-universal approval who won a large majority last time around takes on an opposition onto its third leader of the term amid a pandemic that has seen large polling boosts in most places that have not completely stuffed it up.  The comparison with the New Zealand landslide is impossible to avoid.  

Even with no COVID-19 and no Opposition disarray it would be a massive upset for the McGowan government to lose.  Two major patterns in predicting state results are "federal drag" (it is an advantage not to be the party in power in Canberra) and the age of the state government.  The last loss by a first term government that was in opposition federally in any state was back in 1930 and the historically expected result for WA Labor would be to gain about three seats.  That should be treated with some caution given that they have so many already, but to actually get near losing would be a massive outlier.

But there's been speculation that this might not be just a routinely heavy pasting of a hapless opposition (such as the crushings delivered by Labor Premiers Beattie, Bracks and Bacon at the end of their first terms in 2001-2) but rather something more spectacular.  Is the opposition heading for Tarago territory or worse?  The results of the opening Newspoll would suggest yes, while a uComms poll merely predicts an enormous but more manageable thrashing.  

Briefly re the upper house ...

The higher-end polling suggests that Labor could even win outright control of the Legislative Council, the chamber's malapportionment and the ridiculous persistence of the Group Ticket "Voting" demolottery system notwithstanding.  Indeed if the Newspoll is correct (which it may not be) and the swing from the lower house is replicated in the upper (which it may not be) and the swing is anywhere near uniform, Labor would win 19 seats and a majority off raw quotas, plus anything else they could pick up on group ticket preference flows.  

I will have more to say about the upper house, most of it frothing and fuming, when the ABC calculators are up and I've tested some ideas on them.  For the time being, suffice to say that nothing much has changed (just replace "Barnett" with "McGowan") since I wrote this article in 2017.  If anything, the preference harvesters are making a bigger effort this time after only jagging one seat last time around.  There simply is no safe way for any sensible person to vote above the line - a vote for any party whatsoever could potentially help elect anti-vaxxers, health quack parties, gun nuts, xenophobes, the religious right, or deceptively named and unaccountable rubbish parties like Liberals for Climate (nee Flux).  The only way to put all the even vaguely mainstream parties above all that rubbish is to vote below the line numbering every box without making a single mistake.  If doing this, prepare your vote, and if you make a mistake, ask for a replacement ballot paper.  (If it were me, I'd be sorely tempted to invoice the WA Parliament for my time when I was finished.)

The Newspoll

The Newspoll has come out with a scarcely processable 2PP of 68-32 to Labor off primaries of ALP 59 Liberal 23 Green 8 One Nation 3 Nationals 2 Others 5.  The 2PP is the largest in Newspoll history in any state, beating the 66-34 lead for the NSW Coalition soon after the 2011 election.  (And that was in an optional preferencing state, which tends to blow out the 2PPs when one party is on the nose).  Mark McGowan has a net satisfaction rating of +78 (88-10), the highest recorded in a regular full scale Newspoll (beating Wayne Goss's +57 from 1992), but slightly lower than recorded by both McGowan and Peter Gutwein (Tas) in smaller sample approval-only Newspolls last year.  McGowan leads Zak Kirkup as Better Premier 83-10, noting that this indicator does skew to incumbents (but only slightly in the context of such an enormous lead).  This is the highest lead in Newspoll history by seven points - Steve Bracks (74-8 over Denis Napthine in 2000) and federally Kevin Rudd (73-7 over Brendan Nelson in 2008) have led by 66.  Opposition Leader Zak Kirkup is on net -12 (29-41), which in the circumstances is probably not that bad really.

This is the first full-scale Newspoll of Western Australia since the poll adopted its new online-only methods and should be treated with caution for that reason, as well as because the result is so extreme, especially the primary votes.  The sample size is 1034, for an implied maximum margin of error just above 3%, but polls can be wrong by more than the margin of error (as we saw with many in 2019).

uComms

A commissioned uComms that was reported recently but conducted on Jan 30 had a 61-39 lead for Labor off primaries William Bowe reports as Labor 46.8%, Liberal 27.5%, Nationals 5.1%, Greens 8.3% and One Nation 6.9%, leaving about 5.4% for others.  There is insufficient information about how this poll has calculated its respondent preferences.  I get 59.5% (a mere 4% swing) by last-election preferences from these primaries, so one possibility is that the respondent preferences from Greens, One Nation and others voters are stronger for Labor than in the 2017 election.  However another is that Nationals voter preferences have been distributed (which is inconsistent with how the 2PP estimates were calculated for 2017.)  There is also insufficient information about how the pollster dealt with parties running in some seats and not others, especially the Nationals and One Nation - if they were offered as options across the whole state this could inflate their votes.  

There was also a report (see same link) of party polling by the Online Research Unit, which has historic staff connections with AMR, a minor online pollster that performed very well at the 2013 federal election.  This had a 62-38 lead but was not based on a quite full selection of electorates, and given that it is party polling and we don't even know which party I'm not sure it merits reconstructing.

Polling error?

Both the Premier and the Opposition Leader have been keen to discount the lopsided polling and the Premier especially has pointed to recent cases of polls being way off as a reason not to trust these polls.  But the errors in the polls that were famously incorrect (eg Brexit, Trump 2016 and 2020, UK 2015 and 2017, Australia 2019) were minor - the equivalent of a few percent 2PP - compared to Labor's current leads.  (The US national polls in 2016 were actually quite good, it was mainly the polling in specific states that was the problem.)  All these recent errors are irrelevant.  They could be repeated and then some and the election would still be a thrashing.  The Newspoll may be something else - a genuinely rogue poll that is wrong by more than those errors - but even if it is, that doesn't account for the small trickle of 60+ 2PP polls from other sources.  

On the possibility of poll error, this is a terrific read.

Basic seat modelling

As usual I have set up a rather basic conditional seat model which asks: for a given 2PP, what seat result should on average be expected?  My version is far less sophisticated than the deluxe Armarium Interrata model which throws in pretty much everything and is well worth a look, but it will reach broadly similar overall conclusions for a given 2PP.  It's been very quickly assembled given current work commitments and isn't guaranteed mistake-free.  

The main point of a conditional probability model is to avoid the limitations of uniform swing models, which (i) can go badly astray if one side is predicted to win many seats very narrowly while the other is not and (ii) ignore personal vote effects.  While I may adjust these assumptions further, for the time being I've set retirements at 2% (though in some cases this should probably be increased), sophomore effect (personal vote gained by a new MP) at 1%, sophomore surge (benefit gained from the new MP having beaten the opposing party's incumbent at the last election) also at 1% and I've set the standard deviation of seat swings after adjustments at 3.5%, though I may raise this if the high swings continue in further polling.

A number of seats are unusual.  These are:

* Dawesville: This is an unusual case where a leader has been appointed in a very marginal seat.  Looking at most relevant cases since 1950 (bar one where I haven't yet found seat results) I find that where an MP becomes Opposition Leader they will usually outperform their party's swing, though there is one very strong example (Charles Court 1974) of underperformance.  The median is 2.3%, which I have used in preference to 1%.  This plus Zak Kirkup being a first-term MP means that he would have a roughly even chance of surviving a state swing around the 4% range, but if there is a massive swing to Labor then his seat will almost certainly fall.  I am not certain Kirkup will receive a leader boost though - his case might be different if he has been chosen as leader specifically to try to save his seat. 

* Darling Range: This is what I call a "disrupted seat" because it was won by the opposition from the government in a by-election.  I investigated disrupted seats prior to Queensland 2015 (tsk tsk, that article thought the LNP was still winning) and found that they behave quite strangely.  On average the swings in them fall in between what would be expected if using the previous election as a base and what would be expected if using the by-election as a base.  However when there is a large swing to an Opposition at a general election, disrupted seats tend to record inflated swings.  What happens to a disrupted seat when there's a big swing to a government at the following election I can't say much about, because that's not a situation that normally comes up!  Anyway I prefer to model disrupted seats off their previous-election margins and owners but with a penalty applied for them being disrupted. In the case of Darling Range four different methods I used to set this penalty produced values between 2.5% and 5.9%, with an average of 4.1%, so I am treating the seat as notionally Labor on a margin of 1.7%.  If there is a meaningful swing to the government, Darling Range should be recovered.

* Kalgoorlie: This seat is Liberal on a margin of 6.2%, but it was previously National.  The Nationals lost the seat to the Liberals because they came third by about 3% after preferences.  If there is a bigger swing proportionally against the Liberals, as implied by the uComms poll, then in theory the Liberals could fall into third instead.  The Nationals might be a stronger opponent for Labor in that case by a few points 2PP because Liberal to National preferences generally flow more strongly than National to Liberal (Pilbara 2017 was an exception.)  However, the Liberals' Kyran O'Donnell should get a personal vote bonus, and the beaten Nationals candidate in 2017 was unusually high profile (Tony Crook).  Moreover I am sceptical of the pattern in the uComms poll for the reasons noted in the section about it above.

* Geraldton: Here the 12.5-year Liberal incumbent Ian Blayney has switched to the National Party.  The Liberals beat the Nationals by around 10% to make the final two in 2017 but Blayney's decision at least takes his personal vote from the Liberal column and puts it in the National column.  That shouldn't make ten points of difference, but it might get close enough for the same issue as in Kalgoorlie above to come into play.  At this stage I've adjusted neither, but it's possible I'm underrating the conservative side by treating these as if they were Liberal vs Labor seats.

* Retirement seats:  Among the retirement seats at this election are some long-serving local members and also some heavy hitters.  Labor loses 20-year incumbents Fran Logan (Cockburn), Mick Murray (Collie-Preston) and the one with the most remarkable personal vote history, Peter Watson (Albany).  Labor also loses Treasurer Ben Wyatt (Victoria Park) and the Liberals lose former Treasurer (and leader, though not at an election) Mike Nahan (Riverton).  It will be interesting to see if there are some outsized personal vote effects in these seats.

Current output

Here are some example conditional outputs for this model at the moment.  I show only the non-Labor seats, though if the swing is small Labor might lose a marginal seat by chance somewhere.  (All Labor's marginals except Albany and Darling Range are buttressed against this by sophomore surge effects.)  The probability is the model's chance of the holding side winning the 2PP - it takes no account of possible disruptions by independents and so on.  

Firstly, the most optimistic scenario for the conservatives that there is any public polling support for yet, which is the uComms poll primaries but using my last-election preference estimate (59.5 2PP to Labor):



The "prob" column gives the modelled chance of holding the seat for this 2PP.  In this scenario the Liberals probably lose Hillarys and Riverton.  Dawesville is a tossup and if it doesn't fall there are several more that might.  The Nationals probably lose Geraldton, leaving a most likely seat result of Labor 45 Liberal 9 National 5.  

Now to what the Newspoll 2PP looks like:


This would be absolute carnage - the Liberals would retain at most four seats, but could even be wiped out entirely, while the Nationals would probably hold three or four.  The most likely seat distribution would be 53-2-4.  (AI have told me that their model projects both the Liberals and Nationals to just under 3 in this case. I suspect if one takes primary votes into account - and the fact that the Nationals are down to just 2% in the Newspoll - things get worse for their seat distribution.)  

Now here's a weighted average (60% Newspoll 40% uComms):


In this version Labor wins about 50 seats while the Liberals and Nationals win 4-5 each.  

We should be cautious about extreme polling and whether it will entirely last until the election.  It may generate an underdog effect, but this was not the case with other famous thrashings such as NSW 2011 and Queensland 2012.  Also, state governments excluding Tasmania have historically tended to slightly under-perform compared to their polling.   There's also always the potential for COVID stuffups during the campaign.  But for me the Newspoll especially raises the question of whether scaremongering about "total control" will actually work, or whether, at this moment, it might imply a mandate for strong leadership, and voters might like the sound of it.  

Note: I will be online on election night and expect to have some comments here.  However post-election comments will be limited because of fieldwork commitments starting March 15.

19 comments:

  1. Hi Kevin

    Has early voting opened?

    Also do you know the record number of seat gains for a government that was in ckear majority prior to the election? I think the record for a government seat gain is Queensland 2001 but they were in minority.

    ReplyDelete
    Replies
    1. Early voting generally opens on Wednesday; I am unsure if it opens earlier at any centres.

      The Coalition under Joh in Queensland increased its majority by 22 seats (27% of the parliament) in 1974.

      Delete
  2. Thanks Kevin. Of course it would've been to Joh to stitch up that kind of record.

    What would be your odds presently on the Libs being wiped out in the WA lower house? Is Vasse the last hold out? Or Carine?

    ReplyDelete
  3. 1%? Maybe 2%? The Newspoll is probably either over the top now or at least over what will be recorded on election day. Even if this isn't the case the chance of a total wipeout is still fairly remote. But there's also the slim possibility that it ends up even worse. Major parties have sometimes completely wiped out in Canadian province elections, albeit under first past the post.

    Probably Vasse is the last holdout. A case can be made that my model is actually overestimating the Liberals' position in Cottesloe. The incumbent in 2017 was then-Premier Colin Barnett and he may have had a significant personal vote that his replacement won't have. Then again given how unpopular Barnett was by the end he might have lost that personal vote by the election.

    ReplyDelete
  4. Fascinating Kevin. 2 questions - pls remind me how many people fit in a Tarago? And who the heck is/are Armarium Interrata?

    ReplyDelete
    Replies
    1. Hi Jack,

      You can fit 100 people in an original Issigonis Mini - Maxwell Smart in the front and 99 in the back!!!

      dedwards

      Delete
    2. This comment has been removed by the author.

      Delete
    3. Morning! We're not Bonham, or Raue, or PollBludger - perhaps somewhat unfortunately. It would be fun to be them, or to co-opt their twitter base. You can call us Armarium for short. We are indeed a "mysterious consortium" of uni students, doing this not-so-mysteriously and with a flavor of frantic, on spare scraps of time. We're into politics, economics and data modelling, and we publish articles and occasionally forecasts - that's all here under https://armariuminterreta.site/.

      (reposted as the previous comment came up as being posted by Unknown)

      Delete
  5. Seven. And Armarium Interrata are a mysterious consortium - there are apparently four of them, the names of at least two of whom are known to me; at least one comments here sometimes. I'll let them introduce themselves if they feel so inclined. Nobody already well known in these circles. Very interesting work.

    ReplyDelete
    Replies
    1. Thanks Kevin. I think a Tarago will be quite big enough - even without doing Max Smart tricks!

      Delete
    2. Hi Kevin,

      (for those who don't know me, I'm Ethan of Armarium Interreta, I think Rebekah just introduced us above)

      Specifically for Geraldton, our model estimates that if the contest ends up as Labor vs National, Blayney is expected to do about 1% better than if the contest was Labor vs Liberal (thanks to the assumed stronger Liberal > National pref flow).

      Currently the model thinks that there's about a 30% chance of the Geraldton election being a Labor v National election (thanks to the much larger drop in the National primary - even though the National drop is smaller in absolute terms, the swing is concentrated in much fewer seats, which makes it much bigger in those seats than the Liberal primary drop), so if you want to adjust for that, one way is to simply add 0.3% onto the Lib/Nat 2pp estimate. Given how marginal Geraldton is, and how big the estimated swing to Labor has been so far in polling (5 - 13%), this is unlikely to make much difference in the Liberals/Nationals' collective chances of winning/holding Geraldton.

      Delete
  6. Hi Kevin,

    Thanks for an excellent article.

    Your two paragraphs on the upper house are really good.

    I have always voted below the line, and continue to do so, beacause the revised Above the Line Senate system still allows parties to nominate and elect Senators.

    I start numbering with the highest number, and give that to the candidate I dislike the most. That way they never get any part of my vote.

    dedwards

    ReplyDelete
  7. Forget the Tarago Kevin!

    A Holden Barina will suffice for lower house WA Lib road trips, post election.

    Thanks for the great modelling that is Armarium Interreta.
    Very slick.

    ReplyDelete
    Replies
    1. I suspect a motorbike and sidecar may be enough for the Liberals, Dustboot. The Nats can then have the Barina. (See discussion on pollbludger of the shift in the old folks' vote and the list of seats where it could matter.)

      Delete
  8. What seats will be non traditional runoffs? The Liberals seem to be preferencing Greens over Labor in every seat though the usual suspects for strong Green seats will likely be ones where ALP win on primaries.

    ReplyDelete
    Replies
    1. Coming second is better than coming third, at least for political parties. In any seats where the Greens come second they are in a with a decent chance of consolidating their vote next time, through greater prominence.

      Delete
    2. The theory is that once they come second, they can position the seat in the following campaign as a contest between the two parties, and say "The Greens can win here". I think it's more a self fulfilling prophecy as the Greens dedicate significantly more resources to a seat on the basis of a theory. Even when they do try it doesn't always pan out, with Greens slipping back to being also rans in Lismore (NSW), Higgins (Federal), Noosa (Qld), Heysen (SA) and Fremantle (WA), though I think this election they will rejoin the 2PP runoff there.

      This status doesn't always mean the Greens try. In 2011 there was a huge number of safe LNP seats where Greens beat Labor into 2nd in NSW where they obviously weren't going to be hot favourites the election after. This could happen this election in ultra safe ALP seats where Greens in the low 10s might be Liberals on single digits? But it's not like the Greens are putting much into Warnbro and Armadale, nor are there many Liberal votes left to swing.

      Delete
  9. It should be noted that Dawesville is a marginal seat only because of the 2017 election, previously a fairly safe Liberal seat that provides some of the core Liberal vote in the federal seat of Canning. Zak Kirkup was a new candidate.

    ReplyDelete
    Replies
    1. Jeremy,

      While that's true, I don't think "marginal" vs "fairly safe" is a good way to think about which seats are likely to fall. A seat which was "marginal" at an election where the Liberals win 60-40 is not going to be a tossup if the next election sees any kind of swing to Labor.

      Instead, I will point out that Dawesville has been about 6% more Liberal-leaning than the WA statewide 2pp (in other words, the difference between the 2pp in Dawesville and the statewide 2pp is usually about 6% in the Liberal's favour), with the sole exception of 2008. Hence, if Labor is winning statewide by more than 56-44, we should expect Labor to be favoured in Dawesville (barring exceptional circumstances e.g. Labor candidate disendorsed).

      Even if you take into account that Kirkup just won his seat and the new-leader bonus Bonham mentions above, that pushes up the margin Labor needs to win statewide to about 58-42. Labor is currently ahead by over 60+% 2pp in all public polling; hence we should assume Labor is favoured in Dawesville.

      (not to mention that there may be reasons to be bearish on Kirkup holding his seat. For example, districts with a high population of seniors swung hard to Labor in last year's QLD state election, and Poll Bludger reported on polling which suggests that Labor is tipped to gain more with senior voters than with younger voters; Dawesville has the highest proportion of over-60s in the state and if they revolt Kirkup may well be the first to fall.)

      Delete

The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.