Poll Roundup: The Honeymoon Still Isn't Over But Could It Be Check-Out Time Pretty Soon?

Federal cross-poll average 55.2% to Labor

(Cross-poll average expected to overestimate Labor's actual standing by c.1%)

It's been a long time since I released a federal polling roundup (several attempts have been commenced then lost in the longer weeds of why Resolve gets such different results from other pollsters) but I thought this would be an appropriately random time to put one out.  A new Voice roundup is also likely in the next week or two, but to save the suspense, the No lead is continuing to build (Yes is now behind about 45-55 on cross-poll aggregate).  There are weak signs that the rate of decline may have stopped accelerating and might even be slowing but that's not enough for Yes which unless the polls are very wrong needs to start making real gains.  

Newspoll is currently on hiatus, and it is not clear when or in what form it will (at least I understand it will) return.  About six weeks ago head of polling Campbell White and analyst Simon Levy left YouGov to form a private polling firm called Pyxis Polling and Insights.  Whether they were fleeing anything undesirable or just wanted to do something new I have no information on.  There have been claims about other departures from YouGov but these, if they exist, are unrelated to Newspoll and details are sketchy.  The Newspoll then in cycle appeared following the usual three-week schedule but with a curious issue in its Voice gender breakdowns and nothing has been seen in four weekends since, beyond an article in The Australian that merely aggregated already existing Voice data. We should prepare for the possibility that Newspoll, assuming it does soon return, may behave differently in the future.  [UPDATE: It is now confirmed that Newspoll will switch to Pyxis and will cease to be associated with YouGov.]

This week we've seen new polls from Resolve and Morgan.  The Resolve poll had only a 4% primary vote lead for Labor, compared to leads between 9-14% in every other poll of theirs this year.  The primary votes were Labor 37 Coalition 33 Greens 11 One Nation 5 UAP 2 IND 10 others 2.  Such numbers would still be a thumping win for Labor if they occurred at an election (I estimate 56.3% 2PP, down 2.5 on last month) but the behaviour of Resolve relative to other polls is such that this 56.3 is like a 52.5 from anyone else, unless something has changed.

Morgan however had Labor up one on 2PP to 54.5 off primaries of Labor 35.5 Coalition 34.5 Greens 12 and IND/Other 18.  Using the batched IND/Other figure released I get 54.2 to ALP by last-election preferences, also up 1.  However I've recently seen some Morgan breakdowns (both the partial breakdowns published here and more recent) and can say that Morgan, like Resolve, is getting high Independent voting intention, presumably as a result of having Independent on the readout everywhere, which tends to drag a few percent of wishful-thinking voters into the mix.  This may help explain why Morgan's published 2PPs often, but not always, run a bit above what would be expected based on 2022 election preferences.

Essential last week also had the government up.  On their 2PP+ measure it was a large increase from 50-45 (=52.6) to 52-42 (=55.3) but on last-election estimates I got a far smaller increase of 1.1 points to 55.2.  The most recent raw primaries (not to be compared to other pollsters because of leaving undecided in) were Labor 33 Coalition 30 Greens 12 One Nation 8 UAP 2 IND/Others 8. The One Nation figure seems surprisingly high for a party with some recent dud election results.

The four-week old Newspoll was published as 55-45 but I got the last-election 2PP at 54.3.  There was some speculation that the staffing chaos might have included a change or error in 2PP calculation but it wasn't terribly convincing; the Newspoll before had had a fairly low 2PP for Labor given the published primaries and matters like rounding or minor party breakdowns could have helped explain how Labor's 2PP went up a point in a poll where their primary vote was down 2 (and the Coalition's was down 1).

As a cross-poll average of the most recent polls from these four pollsters (noting that the Newspoll is a bit old) Labor would be at 55.2% 2PP.  But I don't believe an election held "right now" would be quite so lopsided.  In particular I'm very cautious about the high Labor leads in Resolve, which continue to run ahead (even this one) of the figures of Essential, Morgan and Newspoll - polls that are hardly noted on average for underestimating Labor at elections.

Labor's post-election honeymoon has now overtaken the fourteen-months stretch for the Howard Government post-1996, but still has the Rudd Government's post-2007 stretch of just over two years ahead of it.  The Hawke Government's post-1983 stretch lasted that government's whole 20-month first term except for a brief drop about five months in, but there is strong evidence that the polls for at least the tail end of that term were wrong.  Nonetheless Labor's lead has been slowly cooling off.  One can see this on BludgerTrack which has Labor now down to 54.2, down from a peak of about 57.  BludgerTrack does not include Morgan, but including Morgan would lead to a similar result: since February Morgan's released 2PPs have declined on a regression basis in a more or less linear fashion from about 57.5 to just above 54, and my own estimates off Morgan's primaries have declined from about 56 to about 54 (see graph below).  The last sample released in July with Labor at only 52 according to Morgan was an outlier, but also to a degree a sign of the times.  

Morgan 2PPs in 2023.  Red: as released, orange: my estimates

Labor's polling honeymoon has survived all manner of false and ludicrous declarations of its demise from the usual overly eager and often biased media suspects (and a few people who really should know better). But no-one should now be surprised - in the absence of any major rally-round-the-flag type event or opposition blunder - if it soon slips gradually below the line of what should reasonably be considered still a honeymoon.  I've used an average of 54 or more based on polls that lack obvious house effects as a guide, but I would want to see it either below that level for a month, or well below it, before declaring it's the end and that we're back to competitive politics as normal.

Resolve Divergence 

The post-election divergence of Resolve from the other polls has been among the largest I've seen in a mainstream federal poll series in the last 15 years, although in the latest specific poll it has reduced and it will be interesting to see if it stays reduced or not.  There was Morgan Face to Face which skewed to Labor by a few points at most times in the past but there was an obvious explanation there (social desirability bias).  Resolve did not show any particular strong leaning compared to other polls before the federal election but since then it has tended to do so not only federally but also has tended on average to do so in three states - but not in the last poll or two prior to elections.  

I'll mention one thing that isn't the cause, at least not entirely, and that's Resolve's use of forced choice on voting intention.  It was claimed to me that Resolve differs because it forces the respondent to indicate a voting intention for somebody to continue with the survey, whereas all other pollsters allow the respondent to be undecided (though this may be after initial prodding).  The theory then is that the undecideds who are forced to make a choice make it in favour of whichever side is winning and/or the incumbent (which is doubly good for Labor when they are both those things.)

One reason this theory can't fully explain the difference is that the size of the difference is actually too large.  Not everyone polls at exactly the same time but in the context of a year with little real change in voting intention that won't matter much; the average primary vote leads for Labor have been 1.9% in Essential, 2.2% in Morgan, 3.6% in Newspoll and 9.9% in Resolve.  (I have used average primary vote leads to show there is a difference irrespective of the issue of Resolve not estimating 2PPs mid-term.) The average undecided figures for the other polls have been Essential 5.5%, Morgan 7.3%, Newspoll 6.0%.  If every single undecided voter was forced to choose and picked Labor that would only roughly account for the difference in two cases and not account for all of it in the third.  

And undecided voters don't do that - they may well tend to break to some parties more strongly than others at any given time, but not that much.  There's evidence of that from polls conducted on the ReachTEL platform, which have often been published in a forced-choice format but with the intentions of the initially undecided voters shown.  It is rare to see even half of the initial undecideds break to one party in a choice with five or six alternatives.  (Yes these are robopolls and uComms which now uses them isn't terribly accurate, but if close to 100% of all undecided voters picked a certain option under any circumstances then there would be examples of this.)

Another issue with this theory is that it implies that convergence between polls close to election day will occur because the undecided vote drops as voters make up their minds, but the undecided vote in Newspoll and Essential actually didn't do that in the 2022 final polls. 

So what could be causing it?  That's not at all clear on the available information, which is limited.  Resolve does show marked shifts in the leadup to elections based on specifying which divisions Independents are running in (this causes the independent vote to shrink to about the actual value, very accurately so far) but there's no reason that should affect the major party gap in Labor's favour.  Resolve also augments its final pre-election polls with a CATI live-phone-poll sample (not seen in its regular online series) but the CATI sample has been reported as only 20% of the sample, which is not enough to cause the divergence unless the sample is upweighted (which I have not seen any statement about) or extremely strong for the Coalition.  

(NB An argument for using forced choice on voting intention is that it replicates how a voter behaves on election day, but actually it doesn't.  A voter may deliberately vote informal (around 2% do this) or may risk a fine by not voting at all (about 10% failed to vote in 2022 but it is hard to be sure what proportion of these were deliberate as opposed to disorganised, unwell, unable to access a booth or unaware).  Probably deliberate non-voters/informal voters are underrepresented in polls, but there would still be some.  A voter who is still undecided on election day also usually has the option to spend some time considering their vote, including by doing online research or chatting to friends, but a online survey respondent isn't going to leave a survey for a few hours while they decide who they would vote for in an election "held today".)

Leaderships

While Labor's voting intention honeymoon is still going, it seems fair to say that Anthony Albanese's is over unless there is a boom in the next few weeks, especially following this week's Resolve that put his recent performance only at net +2.  (He was most recently +11 in Newspoll and +7 in Essential).  I think net +15 is a reasonable guideline for a PM's personal approval honeymoon, and on that basis Rudd's personal honeymoon ended around the same time as his party's, while Howard's ended roughly one month before.  (There is a lot of talk about leadership approval drops as supposed harbingers of voting intention drops, but this derives partly from gallery urban myths and partly from an over-simplified understanding of a thing called Granger causality.  Yes, the leadership data series contains information useful for predicting the voting intention series, but that doesn't tell you what that is or how it operates.)  Albanese's ratings are still pretty good, but they are now nothing special.  (The Morning Consult tracker which has Albanese on +21 begs to differ, but it is insufficiently transparent and showing disturbing signs of developing into a pay site, so I'll ignore it.)

The BludgerTrack ratings show that Peter Dutton, however, may have ever so gently turned a corner, with Dutton's own ratings now easing to an average -10.  Dutton is also becoming more competitive on Better/Preferred Prime Minister polling, which based on its relationship with voting intention polling skews to incumbents by around 15%.  Albanese's leads have so far exceeded that amount, but recently he was ahead by "only" 25 points in Newspoll and 21 in Resolve.  

Tale Of Two Honeymoons

As long and strong as the Albanese Government's polling honeymoon has been, it hasn't reached the heights of the Rudd honeymoon, and looks unlikely to last quite as long either.   In some ways they are similar in that both Rudd and Albanese have confronted a turbulent economy for reasons that were none of their doing.  But the current "cost of living crisis" has nothing of the emergency sense of the Global Financial Crisis which gave Rudd's honeymoon a second wind - it's just something that makes voters remarkably pessimistic about how things are going overall, even if they have been slow to take that out on the new government.  Rudd was also a rock-star leader who came to office already extremely popular, whereas Albanese's ratings in opposition were never anything special.   The opposition Rudd faced was a mess, unable to settle on a leader or a policy approach, whereas the current Coalition is still wounded by its loss of inner-city seats but at least without leadership tensions.  Meanwhile, while I'm not sure whether the Voice debate is affecting or should affect the national polls, it is at least giving the right reason to remain motivated by the thought that there is something they can win.

Lastly, there has been some silly stuff about possible Senate results in the event of a late-2024 double dissolution, the penny having finally dropped among the commentariat that an early-2024 DD is probably a bad idea.  There have been some gleeful assessments that it would result in mass seat movement from the Greens to Labor, but these are not supported by polling or modelling at this stage.  I refer people interested to Ben Raue's assessment of possible DD outcomes, which has saved me from having to write a post taking out some of the trash that's out there.  

UPDATE 20 Aug:  A federal Redbridge poll has come out with Labor 55.6-44.4 ahead off primaries of Labor 38 LNP 32 Green 10 others 21.  Awaiting more detail but for the moment that 2PP seems about right.  There is relatively little national data on which to benchmark Redbridge.