Monday, February 27, 2023

NSW 2023: New Polls Say Labor Majority More Difficult

NSW Aggregate: 53.0 (+5.0 since 2019, -0.9 since January) to Labor
Most likely result if election "held now" is Labor minority government (approx 44-45 seats)
All four results are still plausible, though Coalition majority appears the least likely

With the overnight release of new polling by Newspoll and Freshwater Strategy, it's time for another NSW polling roundup; I originally set the scene for NSW in my 2022 Christmas Day offering to the masses.  

What we have had this year so far of significance is five major polls, that can be divided into a January wave and a February wave.  The January wave consisted of YouGov, Resolve and Roy Morgan and the February wave is the two released overnight.  However, the Morgan was not released until 21 February.  Because the YouGov and Resolve polls were very strong for Labor, the release of the Morgan was wrongly seen as evidence of narrowing in actual voting intention.  In fact its average data age was similar to YouGov and Resolve and what it really showed was that once the three were taken together there had not been a blowout (my aggregated estimate as at Christmas was about 54-46 to Labor, and following the January polls I get 53.9).



The new polls do suggest that Labor's lead has narrowed, though on my estimates both Newspoll (52-48 to Labor) and Freshwater (53-47 to Labor) might be a little more positive for Labor than the headline 2PPs.  My 2PP estimate for Newspoll (Coalition 37 Labor 36 Greens 12 Others 15) is 52.2 and for Freshwater (Coalition 37 Labor 39 Greens 10 IND 6 SFF1 Others 8, 53-47 to Labor) is 53.3.  However there is a lot of guesswork in the 2PP preference flows (for more see below).

In January YouGov had Coalition 33 Labor 39 Greens 11 others 17 for a 2PP estimate of 56-44.  Resolve had Coalition 34 Labor 37 Greens  12 SFF 2 IND 11 (extremely likely to be an overestimate) others 5 (very likely to be an underestimate).  My 2PP estimate for this was 54.7 to Labor.  The January Morgan had this dog's breakfast: L-NP 35 ALP 32.5 Grn 9.5 ON 6.5 SFF 1.5 AJP 1.5 LC 1.0 LDP 0.5 Teal 0.5 others 11.5 (who?).  The published 2PP was 52-48 but I estimated 51-49. 

Overall giving a very low weighting to the January polls as they are now over a month old, my aggregated estimate is now 52.9 to Labor.  But there are a lot of cautions on the 2PP estimation.  Firstly Morgan has a very high One Nation estimate, but at this stage One Nation have only unveiled 13 candidates; it is not clear how widely they are running.  If One Nation do not actually run in most seats, then voters who would have voted 1 One Nation and stopped will instead vote for someone else, and this could help the Coalition considerably.  On the other hand this may also apply to polls including Resolve and presumably Morgan that have suspiciously high Independent votes - if no Independent runs then those votes may become primary votes in a way that might favour Labor and the Greens.  So these things can be argued either way (and it is also likely that some polls are underestimating generic "others").

There is also the possibility of preference shifting.  There has been some suggestion that the flow from the Greens to Labor could weaken because Greens votes might be disenthused about Chris Minns' relatively minimal and pragmatic stance on poker machine reform.  I don't think this really follows given the strong Greens preference flows to Michael "Asians With PhDs" Daley's Labor in 2019 though that said Labor did (in my view to their cost) make a much bigger deal of environmental issues during that campaign than they have so far with this one.  Another possibility is that a few Greens voters might be positively attracted to Dominic Perrottet's more ambitious stance on the issue, but it has never shown up in polling as a big one for voters.

The Australian's article about the Newspoll projected that Labor would win 42 seats if the Newspoll was correct with the Coalition dropping Willoughby to the independent Larissa Penn.  It is apparent that what the writer has done here is apply uniform swing to the current margins including the margins from by-elections.  This is not recommended because (i) by-elections frequently see a backlash against the government of the day and are not reliable predictors for a general election (ii) the margin in the case of Willoughby was Liberal vs Independent and not a 2PP margin; projection off the primary votes for a by-election Labor didn't even contest is invalid.  Oddly, it turns out that the source for this practice could be Antony Green's pendulum, which uses by-election results in some cases and not others.

My own projection for the Newspoll would be 43 Labor seats.  The reason for this is that swings are influenced by both personal vote factors and seemingly random variation; all else being equal this 2PP would put a bunch of Coalition seats in the 0-2% margin range (Goulburn, Tweed, Riverstone, Holsworthy, Parramatta and Winston Hills) compared to no seats in that range for Labor; Labor would be likely to pick up, say, a couple of those.

Leaderships

In leadership polling it is notable that a scandal surrounding Perrottet having worn a Nazi uniform to his 21st birthday party has done him no harm; nobody much cares as it was long ago, he was young and has apologised profusely.  Newspoll has him with a healthy net rating of +9 (50-41).  The small Essential samples (300 per survey) are bouncier with the most recent having him on +3 (45-42) but that followed a high rating of +18 (51-33) in early February well after the uniform scandal had broken.  Minns is also rating fairly well (net +13 and +11 in Essential, net +8 (41-33) in Newspoll).  Preferred Leader scores are showing significant improvements for Perrottet with leads of 10 points in Newspoll and 12 in Freshwater, but these are only mediocre leads by historic standards because such ratings favour incumbents.  I would say that Perrottet is doing pretty well on these scores given that there is no personal rating difference to speak of on the two and given that the Coalition is still on average a fair bit behind.

Seat-based model

Currently my usual seat-based model projects Labor winning the 2PP in about 44 of the included seats on an "election held now" basis, with the Coalition winning the 2PP in 40 - that includes Kiama although the Coalition seem to be making no serious effort to defeat Gareth Ward (ex-Liberal IND).  Nine seats won at the last election by Greens and independents (three of those under the Shooters banner at the time) are excluded from the model although the Coalition has as yet unknown chances to recover at least Barwon and Murray.

Here is the seat model's current output on the Coalition side for a Coalition 2PP of 47.1:



* = incumbent retiring

This model includes the closest 2PPs for seats won by the Coalition in 2019, but of these Bega was swiped by Labor in a by-election (which disrupts the probabilities for the seat making it a likely Labor retain), and who knows what will happen in the very odd contest for Kiama.  The "Proj" column is the model's projection given that the Coalition 2PP is the current aggregate figure of 47.1.  The "Prob" column is the model's estimate of the chance the Coalition holds that seat given that 2PP.

Although Upper Hunter was very marginal at the last general election it generally isn't considered all that favourable for Labor at the moment and the model's probability for that seat is likely to be an underestimate because the model does not (yet) include any regional factors.  On the other hand other seats may be overestimates, and of special interest here is the 5% to 7% range. Further up the tree the seat of Ryde has seen a spectacular shift to the Coalition during Victor Dominello's tenure and with his retirement it would not be odd to see an oversized correction.  

I haven't shown the Labor side of the model here but it doesn't see much chance of Labor dropping anything to the Coalition if Labor is this far ahead on 2PP.  It does think there is a 10% chance that Labor fails to win the new seat of Leppington.

It should however be noted that Labor has some potential crossbench or non-classic downsides including in Lismore (which is notionally National because the Greens are notionally second after preferences post-redistribution; the Greens could also potentially win), and there is a challenge in Shellharbour from local mayor Chris Homer (IND), though Homer has only been involved in politics for a relatively short time.  There is also a potential challenge in Cabramatta or Fairfield from Frank Carbone (yet to be confirmed).

I should stress that this model is a "nowcast".  There could be considerable narrowing in Labor's lead come election day though my experience generally is that state governments that are trailing do not recover as successfully as federal governments.  Most state governments that fall well behind in polling during their term just lose.  See Australian Election Forecasts for a bearish outlook on where Labor might end up (the site as I read it estimates both a skew against Labor in polling and further narrowing, heading to an estimated final 51.2% 2PP for Labor which would be anyone's election, though probably not a majority for either side.)

Although my model has projected Labor needing upstream of 54% 2PP to form majority government, there are potential paths around that.  The main one is simply that Labor does well in critical Sydney seats, cleaning out most of the 5% to 9% range while getting little swing in its own non-marginals.  However there is no obvious demographic reason for such an uneven swing to occur.  Conceivably, swings against the government might be wasted in the same 2PP-safe Liberal seats where teals did well in the federal election.  If the Coalition can contain losses to a few and recover a couple of rural seats from the crossbench they could conceivably be not far off a majority, though this relies on a somewhat better 2PP than current polling.  

Betting

Betting is not reliably predictive but I note some odds for interest.  One site is taking odds on the total number of seats won by either side and remarkably has 51-55 Labor seats as the shortest-priced outcome.  (Seat totals markets tend to be plagued by high bookie rake and are probably best avoided even by anyone who knows what they are doing).  

Another has markets on some individual seats.  This has Labor gaining East Hills, Penrith, Parramatta and Riverstone but behind very narrowly in Camden and Goulburn and also failing in Holsworthy, Winston Hills, Tweed, Upper Hunter and Oatley, which suggests only 42 ALP favourites.   One of these sites is wrong here!  (This one also has the Liberals tied with the independent mayor Michael Regan in Wakehurst.)

Betting markets had Labor clearly odds-on for a majority last week but have run away to 2.00 and 2.30 with the release of new polling.  In the order ALP majority-ALP minority-Coalition minority-Coalition majority the implied probabilities from the two outlets are 40-36-19-5 and 43-36-17-4.

Crossbench mysteries

This NSW election is fascinating because the 2PP range within which neither party might fail to win a majority is relatively wide, but it's hard to say what the crossbench will actually do.  Two of the independents, Greg Piper and Alex Greenwich, might normally be flagged as more to the left than the right.  However both have flagged poker machine reform as critical to their support should it be relevant.  Labor might shift its position to try to get the crossbench onside, but would they be trusted? (Greenwich has also flagged banning gay conversion therapy as critical to his vote, but both parties have already agreed to that.)  This means that outcomes that see Labor winning say 40 to 43 seats are difficult to model as it isn't obvious who forms government off those numbers.  While it would be assumed the Greens would side with Labor lest their own voters destroy them, I have yet to see an unequivocal statement that the party would throw the current government (of the sort seen in the federal election).  

But surely this cannot be close?

The narrowing Newspoll overnight was met with a lot of scepticism on social media from some Labor supporters.  Quite a few of the more rusted-on persuasion tried to claim that Newspoll had predicted close results in the federal and Victorian elections that then didn't occur, but this is actually false - YouGov and Newspoll strongly projected Labor majorities in both cases while non-Murdoch outlets were buying up shares in Hung Parliament Club.  The other reason for scepticism is that the government's standing in public debate has appeared shabby: scandals, preselection infighting and lately even farce as the media plays Where's Wally games with the Premier's brother.  The elephant in the room here is federal drag (a theory and a fact) which says that the government's age-related scars are more likely to be forgiven by voters when the party is not in power federally.  Again as noted in the Christmas edition federal drag theory suggests a 12-year old government that is federally in opposition will take a bit of damage, but only enough to give it an about even chance.  Federal drag models are not all that predictive, but they do provide a template to say that reasonably close polling isn't obviously silly.

It could also be that the issues mix at the moment helps the Liberals to use federal drag against an otherwise popular new federal government.  As with the rest of politics for the past several months this election is about cost of living more than anything else; it dominates Freshwater's issues poll by a large margin.  Federal drag seems to be fuelled by the perception that same-party state governments won't stand up to the feds and I think personal economic issues could make this a more acute concern than normal.

I may update this article for any further polls this week or next.  It is pleasing to see a higher volume of polling for this NSW election so far than 2019 and I hope this continues.  

Added: Pittwater Seat Poll

The AFR have reported on a Freshwater Strategy seat poll of Pittwater but the report is inadequate, lacking details such as sample size, polling date, polling method and exhaust rate.  It also confusingly refers to "Two-party preferred, with no preferences" when the "with no preferences" makes no sense and the former expression should be two-candidate preferred.  The poll has the Liberals leading independent Jacqui Scruby 52-48 off primaries of Liberal 41 IND 30 Labor 16 Green 4 others 9.  As Scruby would need to gain at .380 preferences per vote to win, which would be fairly hard if the 9% others are right-leaning, the 2CP margin seems credible.  The report says that 25% of Labor preferences are going to the Liberals but does not make it clear if this is out of all Labor votes or just those with preferences; I would expect the latter as the former would appear too high.  

Added: Resolve

Polls seem to come in clusters and we also have a Resolve poll with a rather small sample size of 802 and a much stronger result for Labor than the others.  The primary votes are Labor 38 Coalition 32 Greens 11 IND 13 other 7.  It's extremely difficult to estimate a 2PP with a massive IND vote much of which won't go to actual independents; I get roughly 56.1 to Labor.  The report might be taken as implying a 55-45 2PP but it isn't spelled out.  However, there has been a pattern of house effect against the Coalition in Resolve's federal and state between-election polls since the federal election, and I estimate this is worth just over 2%.  With Resolve included but adjusted for apparent house effect my revised 2PP aggregate is 53.1, on the cusp between 44 and 45 Labor seats.   Without the house effect it's 53.5. The poll is similar to others in finding both Perrottet and Minns with good ratings.

Added: Morgan

Another one for the cluster with a new Morgan SMS taken Feb 24-28 with primaries of Labor 33.5 Coalition 32.5 Greens 11 One Nation 8.5 AJP 3 Teal Independents 3.5 others 8.  Clearly Morgan has offered a shorter list of options here than usual, driving up the ON, AJP and Teal votes, but also the One Nation statewide vote is very high.  As noted above there is a question of how many seats One Nation will actually run in (only 13/93 are so far confirmed).  If One Nation run in all seats then my 2PP estimate for this poll is 53.7 to Labor, but if they run in relatively few and many of their voters vote for the majors instead then it comes down towards Morgan's published 2PP which is 52.5 to Labor.  

The poll also shows Perrottet on a net +6 approval (forced choice, 53-47) but trailing Minns 46-54 as preferred Premier.  It is common for forced-choice preferred leader questions to lack the skew to incumbents of those that allow a don't know option. What isn't clear is how this works: do they send further texts asking for follow-ups and only count the respondents who answer?  How many respondents answer the leadership questions?  All details that Morgan should be publishing.

8 comments:

  1. Regarding your post linking to Lismore being a notionally Greens seat - this is not correct. On the redistributed 3PP the result would be:
    NAT: 43.5%
    GRN: 28.4%
    ALP: 28.1%

    However, while the 2PP between Labor and the Nationals would be ALP 2.0%, the 2CP between the Greens and Nationals would be NAT 2.4% vs GRN. Thus, in this situation, after Labor is excluded, the Nationals would have gone on to win the 2CP vs the Greens. Source:
    https://www.parliament.nsw.gov.au/researchpapers/Documents/NSW2019_FinalRedistribution.pdf

    ReplyDelete
    Replies
    1. Thankyou, corrected! (And in previous article too.)

      Delete
  2. Have you looked at the impact of opv if Labor polls higher or equal to the liberals on primary votes. And the greater ability of greens and Labor to direct their preferences away from coalition candidates

    ReplyDelete
    Replies
    1. It's actually harder to generate very strong preference flows between the left parties/candidates under OPV because some proportion of voters always just vote 1 and stop. This is especially difficult in seats where Labor is excluded because Labor voters are often not used to giving preferences. If Labor is equal with the Liberals on primaries, Labor wins the 2PP but often not by as much as it would under compulsory preferencing. An exception was Queensland 2015 because right wing minor party voters did not support Campbell Newman.

      Delete
  3. Is federal drag something that's seen consistently in polling, or is it something that emerges closer to election day? (e.g. is there a late swing relative to polls in line with what federal drag would predict, or is it baked in)

    For some reason this election is reminding me a lot of SA 2014. Really messy seat by seat election for government seeking a 4th term against a federally dragged opposition.

    ReplyDelete
    Replies
    1. SA 2014 was also a relevant comparison because the government had a 2PP to seats conversion advantage (rather larger than in NSW). What I've found with federal drag is that federally dragged state governments are much more likely to tank in polling over the course of the final year of their term, but I don't think there's an effect by the time it gets to final polls. Have not looked at, eg, one month out.

      Delete
  4. Can you explain why the Roy Morgan polls seem to show a much better primary vote for the coalition than Labor in recent times in nsw.? All other opinion polls post sept 2022 do not show the same

    ReplyDelete
    Replies
    1. The last few Morgans showed the Coalition up by 2, level and up by 2.5, mainly because they have a lower Labor primary vote than other polls. However with the most recent Newspoll having the Coalition up by 1, it's not like they are massive outliers. Morgan are not transparent enough about their poll to offer any theories about why they have a low Labor primary, although it's possible that the number of parties they have on the readout everywhere has something to do with it.

      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.