Saturday, May 21, 2022

2022 Election Day: Newspoll Says Labor Really Ought To Win This Time

2PP Aggregate: 53.2 to Labor (-0.3 since Wednesday) (not a prediction)
(Weighted for time only, no house effects or quality weightings)
Cross-poll average of polls released in last week: 52.8 to Labor 
If polls are right (they may not be), Labor should win outright, with a median 82-83 seats
Historically, Labor has underperformed slightly on average when it leads in final week polling

We've finally reached the end, and after seeing yet another nonsense article from Nine claiming that "If current polling is replicated at the federal election, a hung parliament is the most likely outcome" the end cannot come soon enough.  If there is a hung parliament, then it will be because the polls weren't replicated (at least not on primary votes for those at 51-49.)  But the end is only the start of weeks of post-counting and projection fun ahead.

The final Newspoll has arrived and it has Labor with a 53-47 lead from primaries of Coalition 35 Labor 36 Greens 12 One Nation 5 UAP 3 IND/others 9.  In the past Newspoll has sometimes rounded final poll 2PPs unusually (to halves of a point in 2013 and 2019 and even to tenths in 2010) and I don't know whether the 2PP was rounded to the nearest point or the nearest half-point.  However it matters little.  Scott Morrison is on a poor but not terrible -13 net satisfaction (41-54) and Anthony Albanese picks up six points to -5 (41-46).  For the first time ever in a final Newspoll, the two leaders are tied on the skewed Better Prime Minister indicator, 42-42.  This puts Albanese above John Howard, who trailed 45-40 when he won heavily in 1996.  



The only action from other pollsters overnight was that Morgan put out a release saying that further sampling through the week had shown no movement from their final 53-47 and then tried to cover themselves with baseless waffle about supposed Shy Tory effects which Ethan from Armarium Interreta has called out.  

(Speaking of shy voter effects there is an article touting a very silly theory that shy voters might vote for the loudest party in the room, the UAP.  The theory makes no sense because almost all polling nowadays is online and anonymous, but people making these claims never seem to bother to check how polling is conducted.  The "data analysis" method referred to is so silly that it actually predicted a majority for Labor in the 2021 Tasmanian state election, in which Labor lost outright, losing the primary vote by 20.5%.  Following this article there have been numerous reports of UAP voters being very conspicuous in prepoll, but I wasn't seeing any such reports before the article.  In any case, if the UAP is underestimated by pollsters - which they weren't in 2019 - it will be for any reason other than the one being claimed.)

We therefore have three pollsters with final 2PPs of 53-47 and two on 51-49.  (Resolve also published a last-election estimate of 52-48 but it is very clear from their FAQs that they actually regard respondent preferences as the leading estimate.  Also it seems they actually make their respondents fill out a full ballot paper, which in an online poll is just asking for dropouts and semi-donkey votes.)  The final primaries of Newspoll and Ipsos are coincidentally almost the same, though Newspoll has a slightly lower Green vote and a higher independent/others vote.  

But in almost all these cases, one or more of respondent preferences, rounding or (in Morgan's case) possibly incorrect calculations appears to have favoured the Coalition, and as a result my average of the final polls converted to last-election preferences is fully 0.6% higher (52.8) than the average of the released 2PPs (52.2).  Allowing some weighting for older polls (an idea justified by, for instance, the fake narrowing in 1996) results in an aggregated estimate of 53.2.  It may be that my weightings are a little generous to the older polls, but at this election the final polls will have a median age of between 9 and 4.5 days, so I am weighting the final polls less heavily than in the past.  Indeed Morgan's final poll's median data age is not much different to the second-last Newspoll.

In seat terms, my model gives a median 83 seats to Labor should the aggregate be replicated, and 82 should the final-poll average be correct.   Or for a straight final-poll 2PP average, 80 seats.  (The MRP model also had 80 seats off what seems to have been around 52.2 2PP).

As noted in the previous edition, there is a historic tendency for Labor to slightly underperform its final polling on average.  This has been further explored in modelling by Mark the Ballot, who on this basis projects a 2PP of 51.1% with 77 Labor seats.  Australian Election Forecasts has 52.4 and 79 Labor.  Armarium Interreta currently has 52.1% with 78 Labor seats.  Buckleys & None hasn't yet updated for Newspoll but has 51.9% with 77 Labor seats.   All these models have Labor more likely than not to win a majority, but in some cases not by much.  

I've been avoiding making unconditional predictive claims at this election (it's much more fun to copy Antony Green and refuse to predict things except when I do predict them) but I loosely see things thus.  The actual result has a history of usually being closer if one side has a polling lead, and given especially that it is Labor in the lead, it would seem surprising if Labor matched the late polls or the aggregate.  However one argument that Labor might do so, or do even better (though nobody expects that) is the possibility of over-adjustment to the 2019 polling failure.  

Because of the polling failure, I feel there's a broader tail to either side in the range of things that might credibly happen than I would have assigned in most of the previous elections.  It could conceivably be that everyone is overestimating the Coalition out of caution from 2019 and Labor has a much bigger win than expected.  It could conceivably be that the polling problems are not fully solved, or even not solved at all (for instance if they actually relate to under-engaged voters), and we might yet be in for a long close night or even fortnight.  With various things I say, I get tempted to add "but it doesn't really feel like that", but it didn't really feel like that (where that is what actually occurred) in 2019 either.   But the central range of historic projections by others is Labor somewhere between the low 51s and the high 52s, probably winning a majority but maybe not quite, with a Coalition win still plausible but only an outside chance.  

Something else I mention here is some processing I did of the MRP model.  Given that it includes demographic modelling, it should be good at identifying any potential risk of certain kinds of seats producing uneven swings that make it harder for Labor to win, and hence any possible repeat of the 2019 scenario where the Coalition had swings to it in its own marginals with swings to Labor wasted in safe Coalition seats.  But from the 80 seats in the model, if I give away all the 50-50 seats, and those Labor is winning 51-49 and 52-48, Labor still forms government with 74 seats.  (That is, assuming that they would want it, which Terri Butler and Justine Elliot keep telling us they don't, but nobody believes that).  So the MRP model, if sound, is suggesting that the Coalition probably needs to win the 2PP to even govern in minority.  

It being late, I'm posting up these comments now as for starters, and I will add more in the morning.  I should probably work out who I can vote for in the Senate first however.  

--------------------------------------------------------------------------------

Saturday: Not much time for more writing today but a brief comment about Greens and indies.  The Greens' national polling is remarkably strong given the attention and votes they must have lost to teal independents, but is this accurate or is polling overestimating them as it has rather often done before?  I would be very surprised if they beat the 12% they are polling in Newspoll and it would be a heck of a result to even match it, all considered.  

Projecting Greens seats off national polling is difficult because they sometimes succeed in concentrating their vote in inner-city targets without increasing their overall vote share.  The most plausible targets for the party this election are Higgins in Melbourne (where there is strong on-paper evidence that they should be able to get into the top two, but a distinct lack of buzz about their chances of doing so despite early controversies re the Labor candidate's views on vaccines), and the Brisbane trio of Griffiths (ALP), Ryan (Lib) and Brisbane (Lib).  The big problem for the Greens here is that in all these seats they need to get over Labor after preferences from micros and into second, but if Labor's vote surges nationally too, it's harder for them to do it.  There is simply not sufficient data to model these contests, so keep an eye on them; it will be something new in terms of models for crossbench victory if the Greens manage to gain a second seat.

So much fuss about teal independents and there is not much basis for projecting their results either, because so much of the polling is commissioned seat polling.  The independent vote may well be up from 3.4% to something like 4-5%, but that doesn't say much about specific contests (and the evidence is at best indirect anyway since it mostly comes from Resolve - the others either not splitting out IND or clearly overestimating it.) The cases that best fit past experience of cases where crossbenchers gain Reps seats (seat is vacant, candidate is already high profile or opponent is in disrepute) are Goldstein and Wentworth.  In Goldstein, even allowing for the foibles of commissioned polling there is so much consensus on Zoe Daniel being ahead that I'd say she will probably win (by no means certain though, there is much we do not know about such contests).  The evidence on Wentworth is more sketchy and the incumbent probably less obviously vulnerable personally.  Curtin is another interesting one because of the candidate's Liberal family background and the general weakness of the Liberals in WA.  Kooyong has been a massive campaign but it would be a new thing if it worked in view of not ticking the historic boxes.  Perhaps we will see wins in new circumstances, but the YouGov MRP aside, there's not much to look at that's objective on this one and the internal polling has been far more variable than Goldstein.  Hughes is interesting just because the seat is such a mess (and is also as good as vacant) - then there are a string of others where we would need to be seeing something really new for teal indies to get across the line.  Winning seats for non-major candidates is hard, it will be interesting to see how many prevail.  

I will add more comments later today.

------------------------------------------------------------------

Saturday afternoon:  I give here a sample of my simple seat probability model for a 2PP of 52-48 to Labor (I do not assert this will actually be the 2PP).  There are much fancier and probably better models out there.  I thought about including the YouGov MRP in the model as I would include seat polls by established firms, but including someone else's model in a model seems like piggybacking - most likely the YouGov MRP will outperform this model easily.  The purpose of the model is to give estimates of seat totals.  I have included state adjustments for Victoria (weak swing in polling) and WA (strong swing) but have not adjusted Queensland because of the history of bad Queensland federal polling.



The model projects government losses in Bass, Chisholm, Boothby, Swan and Pearce with a slew of seats line-ball for the Coalition such that it would expect a few of them to drop (but hard to say which); of these Reid is generally expected to fall as Fiona Martin has had a terrible campaign.  

If I move up to the average of the final polls, Reid, Leichhardt, Hasluck and Robertson fall over with Dickson, Deakin, Braddon, Longman and Brisbane all line-ball.  Of these Dickson and Deakin have been relatively little-discussed, Braddon is widely considered tight but a more likely Coalition hold, and Longman and Brisbane are generally thought to be at serious risk.  Further up the tree Ryan, Banks and Bennelong are all seats on bigger margins that are vulnerable.  

In general the model expects about one Labor seat to fall in this 2PP range but has no idea which one it is (its readings for some Labor seats look overconfident).  

Overall as a rundown about Labor seats that might fall, the most attention has been on Gilmore and Corangamite, with the following also mentioned as possibilities: McEwen, Lingiari, Werriwa, Dunkley, Greenway, Hunter, Parramatta, Lyons, Eden-Monaro (largely off a seat poll that nobody takes very seriously).  If Labor wins comfortably, it could well be that nothing falls, but for a reasonably close win it would be likely one to a few would drop somewhere.  

Note the model does not project new crossbench seats (eg Goldstein is likely to fall).

---------------------------------------
A final run of seat betting (not predictive but interesting) as at 3 pm:

Expected Labor gains from Coalition (not close): Chisholm, Reid, Swan, Boothby

Expected Labor gains from Coalition (close): Robertson, Pearce, Bass, Braddon,  Brisbane

Mixed market Labor vs Coalition: Bennelong, Longman

Expected Coalition holds vs Labor (close): Lindsay, Leichhardt, Sturt

Expected Coalition holds vs Labor/Greens (close): Ryan, Higgins

Expected Independent gain from Coalition: Goldstein

Mixed market Independent vs Coalition: Wentworth

Expected Coalition hold vs Independent (close): Kooyong, Nicholls, Hughes, Mackellar, Curtin

Expected Coalition hold vs Independent/Labor (close): North Sydney

Expected ALP hold vs Independent (close): Fowler

Expected ALP hold vs Coalition (close): Gilmore, Corangamite.

By favourites that's 80-63.5-7.5.  I will do a close-seat adjusted total in my post-election coverage.


3 comments:

  1. Given they've been a decade out of office, and the general polling trend, Labor will absolutely do everything to form government even if it's a hung parliament. They need to be seen to 'win'.

    It's just my view, but I feel the Coalition might not try to hard to stay in office in a hung parliament situation. A potentially difficult next couple of years plus dealing with a large and unruly crossbench might make some on the Right feel "it's a good one to lose"

    ReplyDelete
  2. Hope you got some sleep overnight Kevin. Just reflecting on your comment on the "Shy Tory" effect. I agree that there is no evidence for this in Australia. However, many voters would be aware that their actions online can be tracked. Do you not think this might possibly affect the type of voter who is prepared to complete online polling and the answers they may be willing to give? i.e. some voters may not accept the premise that it is "anonymous".

    ReplyDelete
    Replies
    1. It's possible but my suspicion is that most people don't worry about it and that those who do in the case of online polling are scattered across the spectrum (for every paranoid right-winger who thinks the lizard conspiracy is out to get them there'll be a leftie with genuine data security concerns about sharing information.)

      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.