Wednesday, November 23, 2022

Victoria Lower House 2022: Final Days Rolling Poll Roundup

2PP Polling Aggregate (not necessarily accurate) 55.0 to ALP (last-election preferences).
Aggregate of pollster-released 2PPs (ditto) 54.1 to ALP.
Labor appears overwhelmingly likely to win.
Labor majority more likely than not if past polling/results history holds up

This post will track polling for the Victorian election released in the final week of the campaign.  A section dealing with each new poll that I see will be added to the top of the post, however polls will not be added during the day on Thursday because of a field trip.  As I start there is only one poll two polls to discuss, following the common second-last-week drought in state elections, but I am sure more will be added in coming days minutes and that one poll is interesting enough to be worth putting out an article based on it alone.  

Until the release of this week's Resolve poll there had been nothing significant since my previous roundup.  There had been an odd Lonergan poll for the Victorian National Parks Association (incidentally the sponsors of the most accurate poll in 2018, a ReachTEL) but that poll with a Green vote of 19% cannot be taken seriously as a voting intention poll.  Firstly the poll did not ask voting intention questions first up but asked them "Immediately after main body questions", the main body questions covering a series of environmental issues and hence being very likely to skew the voting intention questions in favour of the Greens.  Secondly it is not clear whether the voting intention results reported are those weighted by "Age x Gender, Location" or were reported simply as sample size information.  

I have added a #pollshapedobjects section at the bottom to cover any even more useless offerings.

Betting (26 Nov)

A final look at betting odds to see how they go!  The following projections are combined from two bookies.  Betting odds are not reliable but it is interesting to track their performance. I call a seat "close" for betting purposes if there are two or more candidates below $3 on any market.

PROJECTED ALP LOSSES (Not close)
Bass (Coalition), Nepean (Coalition), Hawthorn (Coalition favoured vs IND), Northcote (Green), Richmond (Green)

PROJECTED ALP LOSSES (Close)
Melton (IND), Pakenham (Coalition)

PROJECTED COALITION LOSSES (Close)
Benambra (IND), Kew (IND), Caulfield (IND or ALP, Coalition favoured in one market but above $2), Ripon (ALP - tied in one market)

PROJECTED CLOSE ALP HOLDS (vs Coalition unless stated otherwise)
Ashwood, Bayswater (tied in one market), Box Hill, Morwell, Ringwood, Pascoe Vale (Green), Albert Park (Green)

PROJECTED CLOSE COALITION HOLDS
Brighton (IND), Mornington (IND), Glen Waverley (Labor)

PROJECTED CLOSE INDEPENDENT HOLDS
Shepparton (Coalition)

If all favourites win Labor wins 51 seats, Coalition 26.5, Greens 5, Independents 5.5

With adjustments for close seats, Labor 49.7 Coalition 27.9 Greens 5.6 IND 4.7 (Caulfield split .3/.3/.3)

Labor majority and minority are at 1.40 and 3.60 with Coalition at 24 and 12 on one market.  Another has 1.55/4.50/21/9.  The Coalition offerings here would be longshot-biased (election contenders at those odds win far less often than the odds imply).   Headline odds are 1.10 or 1.11 for Labor vs 7.00.  

Internal Polling Claims

There have been various loose reports about parties saying things about their internal polling, with both sides expecting Labor to lose several seats and some suggesting that they might lose 10-15.  Internal polling claims are often inaccurate.  The most detail has been concerning Liberal internal tracking which is said to have the party with a 37-33 primary vote lead in a tracker of about 25 marginal seats. However it is not clear whose seats these are. There were also reports of a 50-50 2PP from this tracker but such a 2PP is unlikely to be correct as in 2018 Labor gained an average of 7% per seat on preferences from the primary vote, and the final polls with their various 2PP methods have an average gain of 6.3%.  Earlier in the campaign there were implausible Liberal internal polling claims where the party was supposedly ahead by 16.6% in Sandringham and 15.3% in Pakenham on something that looked like a 2PP with bits missing.

Newspoll

Quite a surprise with the final Newspoll bucking the narrowing narrative and coming out at 54.5 to Labor, off primaries of Labor 38 Coalition 35 Greens 12 Others 15.  By last-election preferences that's 55.6 to Labor, but the Newspoll figure is based on respondent preferences which appear to be weaker for Labor (whether because of changes in party/independent composition or for other reasons).  This has bumped my headline aggregates up to quite high numbers.    Daniel Andrews' netsat has taken some damage but is still a very solid -2 while Matthew Guy's is -25.  Andrews' Better Premier lead remains a mere 51-35.

A simple cross-poll aggregate with no weighting at all would be about 54.3 to Labor.  Australian state polling does not skew to either side on average and only about one state election in four in the past 12 years had an error on the cross-pollster average that was big enough to get from there to more-likely-than-not hung parliament territory (which, just about following Australian Election Forecasts, I now suspect starts around 52.5) But half those errors would fall in either direction so some case could be made that a hung parliament is still rather unlikely.  

Redbridge (Not A Poll)

The Herald-Sun published an analysis by Redbridge projecting a rather gloomy outlook for Labor (41-48 seats) and claimed it to be "bombshell new polling" but there was nothing in the article to make it clear that any new polling was included or if so where or when it was taken or with what numbers.  So that's very nice and all but I can't aggregate it.  

Morgan

Minutes after I hit publish on this article, Morgan released an SMS poll taken over the previous two days with a 55-45 2PP obtained by unstated means from primaries of Labor 38 Coalition 32.5 Greens 12.5 Teal INDs 4.5 DHJP 1 UAP 0.5 (not running in lower house so shouldn't even be included) leaving 11 for other Independents and other parties.  Once again the 2PP method (whatever it is) seems to be selling Labor short because Labor would win thumpingly on these numbers (my last-election estimate is 57-43).

Morgan SMS polls had a poor record in previous state elections and lack transparency, so I am giving them a very low weighting in my aggregate and only including one of them at a time.  As a result of this the impact of this one on my aggregate is near-zero.  It is also hard to avoid suspecting Morgan has a house effect even in its published 2PPs, as they have consistently exceeded neighbouring polls by whoever else and by no small margin either.  On average since late August Morgan 2PPs have exceeded the average of other polls by 2.2%.  Perhaps Morgan is right and everyone else is wrong but you would not have enriched yourself betting on such a theory at previous elections.

Morgan has published a set of predictions where they predict 48-29-5-6 (Labor-Coalition-Greens-IND) but this is based on a uniform swing model that does not take any account of personal votes, which overwhelmingly favour Labor in close seats.  Assuming Morgan's four seat losses to the Greens and independents are correct (which seems a reasonable reading of their poll), I would expect 50 Labor seats off a 2PP of 55%.   

There have been comments on social media suggesting Morgan is miscounting seats here on account of the treatment of notional seats.  I think they are actually doing something like what I do, which is only taking notional seats into account where a seat is newly created.  On this basis Labor does start with 56 seats, including Bass and Bayswater.  

Resolve

Resolve was released via The Age with primary votes of Labor 36 Coalition 36 Greens 10 IND 6 Others 12 and a respondent-allocated 2PP of 53-47 to Labor.  Determined to learn nothing from similar misadventures at the federal election (or perhaps to continue making money out of the "it's close" theme whether it was actually valid or not), media reported this as the major parties being "neck and neck" when the 2PP shows that Labor would clearly win comfortably on these numbers, with a better than even chance of a majority.

A version of the poll numbers to one decimal (Labor 36.2 Coalition 35.8 Green 10.1 other 17.9 respondent allocated 2PP 52.7) was inserted on the Victorian election Wikipedia page by a user called Polliewaffle whose contributions mainly concern polling (especially Resolve).  These decimal numbers were then overwritten by a user called Futuramaking2999 and later reinstated by Polliewaffle although there was no link to a source that includes those numbers.  It is unclear to me if they have been published anywhere besides Wikipedia but in any case I have it on reliable authority that they're official.

By last-election preferences I estimate the rounded-primary version at 53.5 to Labor and the one-decimal version at 53.8.  However, a significant issue here (and the most interesting feature of this poll) is that Independents have gone from 12 to 6 while Others have gone from 6 to 12.  For Others to be twice the rate of Independents is very different to the last Victorian election and most other polls, but it is very similar to the federal election.  The drop in Independents mirrors that that occurred at the federal election once Resolve started offering the actual per-division candidate options rather than the generic "independent" option everywhere.  However at the federal election the major immediate beneficiaries of that in Resolve were the Greens, with One Nation and UAP picking up only a little.

If the Others breakdown was similar to the federal election (ie mostly One Nation and UAP) then that would reduce the preference flow to Labor off IND/Others from about 53% to something like 49% (I get it as that for the 2022 federal election), which is a 0.7% hit on Labor's last-election 2PP.  However the problem here is the candidate mix - One Nation are contesting only five lower house seats and UAP zero.  The Others are the Animal Justice Party and Family First Victoria in every seat, Freedom Party Victoria in almost two-thirds, Labour DLP in just over a third, Victorian Socialists in a quarter, and a dribble of low contesting rates from the Liberal Democrats, Shooters, DHJP, Reason etc.  It's difficult to believe this lot (an average of four Others parties per seat) are good for 12% between them, so I think this Others reading is unlikely to be correct.  There is a need for public clarity (or otherwise) to confirm that the same ballot paper simulation method was used as in the federal polling.  Any information as to what Others are getting large chunks of this 12% would be interesting.  

But even if we do take this one completely at face value it is not as conducive to a hung parliament as many responses to it have suggested.  A modest Green primary would, all else equal, make Green seat wins from Labor more difficult (beyond perhaps Northcote and Richmond on Liberal preferences), and ditto for Independents.  If the 12% Others tended to the right then they would not win seats and Labor would be actually more likely to get a majority off 52.7 (on my model they would win the 2PP in about 49 major party seats).  If the 12% Others were more to the left then the respondent 2PP would be incorrect and Labor would win more seat 2PPs, albeit with more risk of losing some to the Greens. 

If the respondent preference 2PP is correct (something that history suggests always treating with caution) then this Resolve poll is the first poll that is seriously close to converting to a hung parliament, although if the primaries and 2PP were as stated, a Labor majority of a few seats would still be somewhat more likely than not.  

Labor's lead in aggregated polling has been coming down pretty fast - in the Australian Election Forecasts aggregate from 59% to 54.2% in four weeks.  Some of this is just because of more reliable polls coming into the field rather than aggregates being Morgan-heavy, but still if Resolve is fully accurate and the trend continues then we could well end up with a, say, 52-48ish result at which point a hung parliament is a tossup chance.  (AEF currently assesses the chance of a majority as 70.6%, a figure which has been over 95% in the past.)

I should note again that the pendulum is very much against the Coalition which makes it very hard for them to win the majority of seats on 2PP even if they make it from here to 50-50, let alone avoid losing any seats to unfriendly crossbenchers in the process.  So the chance of the Coalition winning from here still looks to be very small, and one that would depend on a significant polling failure.  The wide spread of polling values seen in the campaign suggests that that is less likely than in tightly herded polling sets like that seen in the 2019 federal polling fail.

Daniel Andrews' lead as better Premier contracts from 49-29 to 48-34, not an especially large change in the context of a large 2PP shift from the last Resolve poll.

Resolve polls appear pretty volatile and this one has some curious properties.   Resolve has only been tested at one election though its final poll did well there.  Nonetheless it has been taken as a big signal in the absence of other fresh polls (even by me; it carries a weighting near half in my aggregate pending more data.)

#pollshapedobjects

This section is devoted to rubbish polls.

1.  Herald-Sun (I refuse to link to it) has printed a report about a supposed "independent exit poll" of Mulgrave for the campaign of independent Ian Cook (of slug-gate fame, re which I can add that the slug looked like a Limax maximus).  This is supposed to find him ahead of Daniel Andrews 57.2-42.8 off a sample of 159 (consistent with 91-68).  The first point there is the small sample size; even if it were a perfectly taken random sample the margin of error would be 7.7%.  But also it is from prepolls, which in 2018 disadvantaged Andrews by 3.4% compared with other votes - and even that's ignoring what time of day the prepolls were sampled, how they were sampled and so on.  There is no evidence the person taking the exit poll had any idea what they were doing.  

2. Neil Mitchell internal polling claims.  Mitchell claimed to have seen leaked Labor internal polling  showing the party was losing Oakleigh and Hawthorn to the Liberals, Albert Park to the Greens, Point Cook presumably to Joe Garra and in danger in Werribee, and also that there were swings against it in the regions including Mildura.  Oakleigh is on a massive margin and there is no reason Labor would be polling in Mildura so this appears to be nonsense (Kos Samaras with experience of Labor internals practices has emphatically dismissed it).  Mitchell is opposed to Labor and has played this game and lost before.

3 comments:

  1. Resolve's final federal poll in May also had a significant change in methodology - which I remember scoffing at, but probably helped it to end up getting so close to the final result. ("The survey added several hundred telephone responses to the customary online responses to give it a larger base than the 1408 eligible voters polled in the last Resolve Political Monitor, which was conducted from April 26 to 30.") No suggestion that's been repeated here.

    ReplyDelete
  2. According to a discussion on the Wikipedia talk page for the 2022 election, Polliewaffle is claiming to be Jim Reed, director of Resolve Strategic. No idea if that's actually true.

    ReplyDelete
  3. On the morning of election day, the Nine papers are claiming "a late surge in support for the Coalition in the final days" of the campaign - on the basis of the Resolve poll that they published on Tuesday, and nothing more recent. There may have been more farcical media coverage of a Federal or State election in Australia than what we have seen in relation to this one, but I really can't remember when.

    ReplyDelete