Friday, November 5, 2021

Poll Roundup: Albanese Is Still Rating Below Morrison - Is That A Problem?

Since my last federal polling roundup, there has been quite a lot of new polling but the Morrison government's position doesn't seem to have improved.  We have seen:

* Newspoll (early October) 53-47 to Labor (Coalition 37 Labor 37 Green 11 One Nation 2 others 13)

* Newspoll (late October) 54-46 to Labor (Coalition 35 ALP 38 Green 11 One Nation 3 others 13)

* Essential (late October) 49-44 to Labor using Essential's 2PP+ method (Coalition 37 Labor 36 Greens 10 One Nation 3 others 8 undecided 6).  I get 52.2 to Labor by standard last-election preferences with undecided removed.

* Essential (early October) 46-45 to Labor using Essential's 2PP+ method (Coalition 36 Labor 34 Greens 9 One Nation 4 others 8 undecided 9.  I get 51.2 to Labor by standard last-election preferences with undecided removed.

* Morgan (mid-late Sept) 54-46 to Labor (respondent preferences) (Coalition 36 Labor 36 Greens 12.5 PHON 3.5 others 12).  This would be c. 53.1 to Labor by last-election preferences

* Morgan (early Oct) 53-47 to Labor (respondent preferences) (Coalition 37.5 Labor 36 Greens 11.5 PHON 3 others 12).  This would be c. 52.1 to Labor by last-election preferences

* Morgan (mid-late Oct) 54-46 to Labor (respondent preferences) (Coalition 36.5 Labor 35 Green 13.5 PHON 3.5 others 11.5.) This would be c. 52.7 to Labor by last-election preferences 

* Resolve (late October) Coalition 37 ALP 34 Green 11 PHON 3 IND 9 Others 5.  Resolve doesn't estimate a 2PP, and overestimates independents as discussed last time.  I estimate 52.1 to Labor by last-election preferences.

Labor is well ahead - as a rough estimate without taking into account house effects or the reliability of different polls I put this lot at an aggregated 52.7 (which is what Labor got in 2007, and would be good for about 81 seats to 64 for the Coalition.)  But that doesn't mean a lot because governments tend to recover and this is only the sort of lead that at this stage says the election is still a contest.  If the polls are as wrong as last time Labor might not be ahead at all, but global house effects don't tend to repeat from election to election in that manner.  (Incidentally I am working on a possible aggregate that will take account the detailed history of various polls, but I'm not sure when it might emerge.)

Essential's three-monthly data dump was interesting in that while Labor had a substantial lead in the current poll (partly as a result of using respondent preferences), for the most part the series has had relatively slight leads for Labor, much less than from Newspoll and Morgan and closer than either to the implied results from Resolve.  Had Essential been releasing data at the time taken, the commentary about Resolve's divergent results would have been a bit different.  

Further Adventures Of Team Yellow

An aggregation posted by Mark the Ballot (full article) now shows Others (including One Nation) continuing to rise and up four points in three months, and now taking points not just from the Coalition but also apparently from Labor (faster than Labor can take them from the Coalition too).  The surge is still mostly attributable to the United Australia Party, who have been reported as high as 5% in Essential after earlier reports had them around 3%.   Very much higher figures have been claimed in some Redbridge seat polls in Western Sydney, but those polls rely on readout lists that only include the options Labor, Liberal, Greens, UAP, "A Local Independent" and "Another Minor Party" - ie UAP is the only right-wing minor party option offered upfront.  Sampling issues with robopolling may also be at play here.

As well as UAP, some of the increase in Others may be down to so-called "climate independents", but evidence on that is less clear, since the only poll that carves out support for indies also overestimates them.  Even with the recent rise, the total aggregated Others result is only about a point higher than the 2019 election result, and fourth-party bubbles in mid-term polling have a long history of popping by voting day, so don't adjust your sets just yet, but still ... 

What Might Comparative Net Ratings Tell Us?

As regular readers would have seen me say a lot of times before, Better Prime Minister scores that compare two leaders are innately skewed to the incumbent and polling commentary based on them tends to be severely distorted.  The Australian's Twitter account highlighted that "Scott Morrison has a strong lead over Anthony Albanese in the latest #Newspoll" but in fact Morrison's lead of 14 points (48-34) is slightly below the all-time historic lead for incumbent Prime Ministers (14.8 points).  It was a large lead by the standard of PMs whose parties are trailing 46-54, but that's not much to write home about.

Recently I've been interested in the properties of comparative personal popularity, because personal rating questions ask how each leader is doing the job they are currently doing, so this should work somewhat differently to an indicator that compares two people's suitability for a job that only one of them has done.  Especially I thought I'd look at this claim by Chris Wallace:

"The more popular leader doesn’t always win the election. However, the last four times government changed hands in Australia – in 1983, 1996, 2007 and 2013 – the more popular leader prevailed. So, to win the next election Albanese, less popular than Morrison, will have to buck history. If he does, he’ll be a hero. If he doesn’t, Labor people will rue not having put a more appealing leader in place sooner."

The most obvious issue here is that just because Albanese is slightly less popular than Morrison now (the gap down to five points, -9 vs -4, in the last Newspoll) doesn't mean he will still be less popular at the election.  A lot can change by then.  Another thing is that four elections isn't that much history to buck.  Government doesn't change often in this country and has recently changed quite decisively when it does, and that tends to go with a popular Opposition Leader or an unpopular incumbent. However, there's no reason government can't change by a whisker (as it did in 1913).  Also, the less popular leader did win in a landslide in 1975 but that comes with the usual asterisk.

These are the federal final poll results for comparative net ratings for the Newspoll era, showing Government results based on the Prime Minister's net satisfaction lead vs the Opposition leader.

The more popular leader won nine times out of twelve.  The exceptions are all incumbents - Keating beating Hewson in 1993 because Keating was unpopular but the election wasn't about him, and Howard beating Beazley in 1998 (but losing the 2PP) and Latham in 2004 despite them having high ratings.  

This is the relationship between netsat difference in final Newspolls and election 2PP:

It's actually a slightly stronger relationship than the ones for PM netsat and Better PM, but weaker than the relationship with polled 2PP.  Looking at the full range of Newspolls:

Comparative netsat is also very slightly skewed in federal polling.  Although the average government 2PP in all federal Newspolls has now slipped to 49.6, the mean PM netsat lead has been around three points and the median around one point.  All else being equal, if the 2PP is 50-50, the PM's netsat is about seven points higher than the Opposition Leader's, so a PM being more popular while trailing on 2PP isn't uncommon at all.

I also had a look at state level results.  At state level when the Opposition Leader has a higher netsat than the Premier, they win 65% of the time.  When they have a lower netsat, they win 16% of the time (or 17% if you count Queensland 1995 as half a win rather than excluding it from the sample.) Every single case of a less popular Opposition Leader winning the election outright has been against a mid-term replacement Premier after the original Premier left under a cloud:

* John Fahey replaced Nick Greiner, and lost to the less popular Bob Carr
* Joan Kirner replaced John Cain, and lost to Jeff Kennett
* Denis Napthine replaced Ted Baillieu, and lost to Daniel Andrews (1 point lower netsat)
* Rob Kerin replaced John Olsen, and lost to Mike Rann
* Carmen Lawrence replaced Peter Dowding, and lost to Richard Court

Including state and federal data combined, the crossover point at which an opposition leader becomes more likely than not to win the election seems to be about a 14% lead, which is high given the historic relationship between 2PP and comparative netsats.  This is partly because governments are more likely to win with a 2PP below 50% than oppositions are, but I suspect there's more to it than that.  

Something else at federal level is that historically once the leaders are well established, the netsat gap hasn't been very different on election day compared to 5-7 months out.  There have been big changes when one leader is new or the leaders change, such as in 2016 when Malcolm Turnbull crashed from a 60-point lead to a 9-point lead in this timeframe, but excluding such cases the average shift has been small (12% in the last seven months and 7% in the last five).  Furthermore there's been a tendency for Opposition Leaders to go slightly backwards in such comparisons.  So while Anthony Albanese could well be more popular than Scott Morrison by election day, the historic chance of that is less than 50%.

Anyway for those who like the cricket stats side of things I've added the longest streaks of comparative netsat wins to the Federal Newspoll Records Page.  Scott Morrison's current streak of 26 wins against Albanese on this measure is the fifth longest in terms of number of polls (one more would make it equal fourth) but the third longest in terms of time.  In a poor sign for the measure overall Kim Beazley had the longest streak of wins as Opposition Leader, but that did straddle a win 

There is a lot of evidence that Opposition Leader ratings are less predictive than Prime Minister ratings (indeed they seem to only become predictive around election time) and Armarium Interreta have found that an "approval margin" of the PM's netsat minus half the Opposition Leader's netsat is a better predictor than the straight difference between the two.  

Climate200 Overheats A Poll

There was an outbreak of dubious seat polling claims last week when extraordinary claims about seat polling in Flinders, Goldstein and Kooyong entered the mass media through a Nikki Savva column in the Age entitled "PM’s net zero plan driven by slogans and seats, not conversion or conviction".  This was followed up by an advertisement in the same paper seeking candidates, which stated that "if an election were held today, a candidate of the calibre of Zali Steggall or Helen Haines (Indi) would win in Kooyong." I quote the relevant section of the column, which I decline to link to:

 [..] Without a name attached, an independent candidate could expect to garner from 7.5 per cent to 12.5 per cent of the vote.

Voters were then asked how they would vote if there was an independent candidate of the calibre of Warringah’s Zali Steggall or Indi’s Helen Haines. In Flinders, where doubt remains that Hunt will recontest, the expected vote jumped to 24.3 per cent, in Goldstein 28.5 per cent and in Kooyong 26.3 per cent.

Liberal voting intention dropped markedly once respondents were asked about voting for an independent like Steggall/Haines. In Flinders it fell by 8.7 per cent to 28 per cent, in Goldstein by 9.6 per cent to 28.5 and in Kooyong, by 8.5 per cent to 25.9 per cent.

So an “independent like Steggall/Haines” is behind, but competitive in Flinders, level pegging in Goldstein and ahead in Kooyong.

Firstly, even assuming we could take these numbers seriously, an "independent like Steggall/Haines" would in fact win easily on any of these numbers.  Zali Steggall garnered 79% of preferences over Tony Abbott and Helen Haines received nearly 75% of non-Nationals preferences.  If a vaguely centrist climate-concerned independent in an inner-city seat makes the final two with a primary vote in the mid to high 20s, a Liberal incumbent will generally need to poll at least in the low 40s to survive.  

But more importantly, these numbers are wildly far-fetched.  They would require primary vote swings against the Liberals of 19% in Flinders, 24% in Goldstein and 32% in Kooyong.  Even the 2019 model Tony Abbott, who in personal vote terms seems to have been a worse candidate than the average case where a candidate is disendorsed but remains on the ballot, still polled 39%, and that was a 22% swing against him over two elections compared to the one that he won.  Furthermore, both Greg Hunt in Flinders and Josh Frydenberg in Kooyong had significant independents running against them in 2019. 

It took a few days for the facts to gradually come out, but they did, thanks to Redbridge (the pollster) being a member of the Australian Polling Council and complying promptly with disclosure requirements.  The voting intention poll referred to was the raw results of a question (2.14) which praised Steggall and Haines in the preamble.  It followed another question (2.10) which also praised Steggall before asking whether voters were aware of her (a more useful test of awareness would be to ask the question without praising her), and we don't even know what questions 2.11, 2.12 or 2.13 were but perhaps they were more of the same.  In any case, this kind of question is not a test of current voting intention.  At best, it looks for maximum receptiveness - what share of voters are even open to voting a certain way under optimal conditions for the candidate (a dream campaign and a dream run from the media).  

This kind of thing probably has some sort of message-testing value if a candidate is trying to gauge if there is even any hope that they might win - dream runs for indies do sometimes occur.  However, it would also be a convenient method to use if one wanted to inflate the potential vote of a candidate for whatever reason (like driving up supporter morale, attracting media coverage or attracting candidates) and for this reason I think these sorts of message-testy polls should be handled with extreme care by all involved and that anyone making any use of one should release it with full details of all questions in the entire poll at the same time (not days later) that it enters the public domain.  At this time, there is no evidence that the Liberals will be losing any of these seats.

Secondly, the headline poll results for Kooyong are of note because in these the numbers after redistributing undecided are Liberal 38 Labor 31 Green 15 Ind 8 UAP 7 (vote inflated by limited readout).  On these numbers, Kooyong would probably just fall to Labor (c. 51-49) but these numbers don't look right.  Labor had such primary vote leads over the Greens in 2001-2007 but since then the gap has closed:  8.9% in 2010, 5.8% in 2013, 0.9% in 2016 and the Greens beat Labor by 4.4% in 2019, albeit with star candidate Julian Burnside.  Labor beating the Greens by 16% in Kooyong is extremely unlikely.  

I am also rather cautious about the Age/SMH's article report of Labor internal polling in Higgins.  As usual with internal polls there are no details supplied so we don't even know if this was anything bigger than a 200-voter tracker, or even for sure that the poll ever actually happened.  The claimed results are Liberal 41 Labor 26 Green 19 UAP 5 IND 4 (are the other 5 others or undecided?)  However the Greens beat Labor into second in Higgins in 2016 and may have done so again in 2019 but for an incumbent vacancy, negative publicity for their candidate and the success of Labor's campaign in the inner cities.  Labor's current Higgins candidate (I'd keep half an eye on that space) has been serially controversial and I'll be surprised if Labor beat the Greens by 7% in Higgins.  I'm not even convinced that they will beat them there at all.  

Not To Be Outdone, The Herald Sun ...

Sadly Nine's uncritical coverage of Climate200 polling that simply did not find what it or they said that it did, was trumped by an even worse piece of polling coverage by the Herald Sun, which I also decline to link to.  In an article entitled "Voters oppose ‘Big Australia’ push according to Australian Population Research Institute survey" the paper reported various supposed negative findings about immigration, transgender issues and the proposed Indigenous Voice to Parliament without any discussion of:

* The polling method and polling firm used

* The dates of sampling (other than the subheading misleadingly claiming the polling to be "recent" when it actually came from July 23-30)

* The wording of the vast majority of questions reported on (only one was quoted)

* The views of any independent polling expert or anyone else on the poll's veracity

While these omissions were pretty much the same as those in Savva's article re Climate200, what made this an especially bad case was that the question wordings should have been obviously suspect to anybody reading the poll, and involved a degree of punching down towards marginalised groups.  For instance, how could anyone reading the report not notice that this is - intentionally or otherwise - a skewed question:

This question could skew responses in at least the following ways:

1. It refers to "separate representation" which can be read as implying that Indigenous Australians will not be voting in or represented through normal elections (a form of apartheid).

2. It employs quotes around "voice" that can be read as scare quotes, as if the "voice" is either more or less than it seems.

3. It presents a pro-voice answer with no arguments for it but presents an anti-voice answer with two arguments, one of which involves unsubstantiated fearmongering.

4. It fails to mention that the Voice would be advisory.

5. It refers to the purpose of the Voice as being to advance Indigenous interests, a term that might be taken as denoting direct action on matters of material benefit.  In fact the Voice is intended to function more as a way to ensure voices are heard, with any action taken on those voices being a matter for the parliament.

Plenty of the questions are not obviously bad in isolation, but far too many of them are the sort of thing one expects when a lobby group writes its own poll questions.  The poll began by giving respondents a preamble that drew attention to a sharp increase in recent immigration, which can easily be read as an argument that recent immigration rates are unnatural or dangerous, and could contaminate many of the subsequent questions.

The poll also has voting intention findings (Coalition 40 Labor 32 Greens 10 One Nation 7 others 11), but the voting intention question was buried at Q25 by which stage the nature of the prior questions could well have affected the outcomes.  Especially they may have boosted the One Nation vote, which would also be boosted by a depleted set of options in the readout and by having One Nation on the readout in seats they do not run in.  In any case, despite the poll's bizarre clustering of parties as "Coalition plus One Nation plus Others", this would only be a narrow Coalition win (52.2%).  

There is also some weird stuff in Tables 2 and 3 where the report treats others as "predominantly conservative parties", "right-leaning parties" and so on, when in fact others is a mix of right-wing and right-ish parties, minor left-wing parties, and independents - and as a result only slightly favours the Coalition on preferences.  Table 2 is supposed to show the right doing better on the votes of non-graduates but the Coalition primary among non-graduates is lower and on a 2PP basis there's very little in it.

Fortunately APRI did publish its own question wordings in full, but more information on weighting methods would have been useful, and for these purposes it's a shame that the Online Research Unit (whose panel was used in the poll) isn't a Polling Council member.  

I've been very busy with work and other things lately but hope to have time for more articles soon!

No comments:

Post a Comment