Tuesday, September 22, 2015

Poll Roundup: Turnbull Shift Puts Coalition Back In Front

2PP Aggregate: 51.3 to Coalition (+4.9 in a week)
Coalition would easily win election "held now"

It's one of those weeks where we start things off with the graph (and especially notice the far right side of it):

So far we've had five voting intention polls since former Communications Minister Malcolm Turnbull neatly disposed of Tony Abbott and became Australia's 29th Prime Minister.  ReachTEL came out with a 50:50 2PP immediately after the change.  Galaxy last week and the Galaxy-run Newspoll this week came out with 51:49 to Coalition.  Morgan in its usual fashion when something happens, threw an outlier, 53.5:46.5 by last-election preferences (though given the primaries, that was generous to Labor), and Essential put out a one-week sample at 50-50.

These polls followed on from some tidying-up of the entrails of the Abbott era, in which Morgan released a 56.5:43.5 to Labor poll from the Abbott government's final week, and Essential released their final two-week polling average at 52:48 to Labor.  Essential attracted some derision for again failing to move off a modest result while Abbott's Prime Ministership was falling to pieces, prompting speculation they wouldn't move this week either (which was not that far wide of the mark).  Considering the polls' headline rates, Morgan had actually moved twelve points (from 43-57 to 55:45 respondent-allocated) to Essential's two! The Morgan 55:45 is clearly nonsense: had the Coalition really led by that much, they would have won Canning by more.

Because Morgan chose fittingly to release two one-week samples, it's important to note that last week's sample was not only small but exclusively face-to-face (a method that tends to skew to Labor).  I've downweighted last week's accordingly.  (This week's is also said to be exclusively face-to-face in the report, but the long-term summary says multi-mode.) On this basis my final aggregate for the Abbott era was 53.6% to ALP.  All the Abbott-era data are now gone from the current aggregate.  (Some other aggregators have also scrubbed all the old stuff, while Phantom Trend's model is so dynamic that it didn't need to.  At present PT has 51.4 to Coalition, Mark the Ballot 51.3, BludgerTrack 51.5, Andrew Catsaras 51.2.)

There were actually six polls that asked about voting intention, but oddly only five that published it.  It's rather disappointing that when Morgan rushed out a better-Prime-Minister SMS poll (of which more later) almost as soon as Turnbull was appointed, they didn't even bother to release voting intention results. That's although they clearly had the data to calculate them, and although they had done so when releasing a similar poll for the switch from Gillard to Rudd.  If Newspoll did this sort of thing it would be a significant scandal in poll-watching circles.  The lack of a similar outcry in this case shows the lower attention paid to Morgan results.

Anyway I had a go at reverse-engineering the primaries from the party breakdowns in that poll and found that something around 45 for the Coalition, 30 for Labor, 15 for the Greens and 10 for Others (for a 2PP of around 53:47) appeared to fly reasonably well.  I have no idea what sort of scaling (if any) Morgan normally would have applied from there, but it's actually rather similar to their primary polling result this week.  Morgan's frequently radical results may look like just one long cry for more media attention, but I have a more innocent, if not all that flattering, explanation.  The SMS polling component may just tend to capture the more motivated and interested members of Morgan's panels of voters at any one time rather than the less engaged ones. In this way, SMS polling may have a lighter dose of the problems of opt-in online-newspaper "polling".

After taking into account the primaries and house effects (which will be reviewed after sufficient Turnbull-era polling), I've aggregated the new polls as follows: Morgan 54.7 to Coalition, Newspoll 51, Galaxy 50.8, ReachTEL 49.9, Essential 49.8.  This all comes out at 51.3 to Coalition.    The suggested result would be an easy win for the Government given that would probably win with 49, and in real terms their lead is as large as Labor's was two weeks ago.  The question is how long it lasts.

Compared with previous shifts:

The five-point swing so far is most similar to the six-point shift for the reinstatement of Kevin Rudd in 2013, which arrived almost fully-fledged, peaked within a few weeks, and then rapidly deflated as the Rudd return unravelled during a chaotic election campaign.

Other bounces for the mid-term removal of a Prime Minister:

* Although one pollster (Nielsen) had an eight point bounce for Rudd to Gillard 2010 based on two extreme readings a month apart, overall aggregated polling moved by only just over one point, and that was gone within a month or so.

* Newspoll data for the change from Hawke to Keating suggests that it took a few months to make a reliable difference (and I remember there being a lot of recriminations in the first few months, with Hawke's removal deeply controversial despite his stumbling performances and persistent bad polling).  When it arrived a few months late, the permanent shift from the dire polling at the end of Hawke's reign was worth about five points.

* Morgan primaries for the Coalition were generally a few points higher in Billy McMahon's early polls than in John Gorton's last poll in 1971, but the change was at the expense of the DLP and would have been worth less than a point on a 2PP basis.

Preferred Prime Minister

In the last week we've been swamped with preferred/better Prime Minister polling in spite of its record as an unreliable predictor and biased towards the incumbent.  In considering these figures we should bear in mind that for a 51:49-ish lead, we'd expect pollsters other than ReachTEL to show the PM twenty points or so up, under normal circumstances.  We are seeing even larger leads than that, either despite or because of the new PM being untested in that role.

I have a table of Preferred/Better PM margins by pollster (in the form Turnbull-Shorten) below.  All but ReachTEL allowed a don't know option, but probably voters aren't that likely to message Morgan back to say they have no opinion.  The two readings in italics are my inferred estimates, noting that Galaxy usually has extremely high undecided rates for Greens/Others voters.

There's a pattern in the Labor column that the more prone to motivated-response a polling method is (Morgan, Essential) the more the Labor voter prefers Turnbull, and the more forced it is (ReachTEL) the more they pick Shorten.  This is also consistent with motivated response to SMS polling being the explanation for Morgan's very high Coalition and Turnbull results this week.

The preference for Turnbull among Greens and Others voters is consistent with the shift seen in Morgan's respondent preferencing.  That, however, might also be down to disgruntled Andrew Bolt types parking their vote with Others, as suggested by Peter Brent on Twitter.  (Speaking of Brent, his essay Time For Liberals To Move On From John Howard Infatuation this week is especially excellent.)

Approval Data

There is not a lot of new leader approval data.  Newspoll ventured forth with the first net satisfaction ratings for Turnbull, giving him a +18 netsat (42-24) but not surprisingly with the highest undecided rate (34) for a PM since the first few weeks of John Howard.  Newspoll did not poll approvals in the first fortnight for Rudd (the first time), Gillard or Abbott as new PMs.  Newspoll found Bill Shorten's netsat little changed on -25 (29-54) but that was better for Shorten than ReachTEL, which had him on -29.8 (down 9.2 points in two and a half weeks), his worst reading from them so far.

Essential had attribute stats for Turnbull and Shorten showing that impressions of Turnbull are up across the board, those of Shorten are down across the board, and Turnbull flogs Shorten on almost every attribute except a tie for "trustworthy".  (Turnbull being judged two points more aggressive is probably a positive in these unusual circumstances.)

Approval Of Change

Both ReachTEL (51.1:31.7) and Essential (58:24) found voters approve of the removal of Tony Abbott in favour of Malcolm Turnbull.  Both found approval of the change stronger among Labor and Greens supporters than Coalition supporters, but that Coalition supporters tended to approve of it.

 Essential found Others supporters approved (48:34) while ReachTEL found they didn't (24.7:56).  Although there has been much talk about pro-Abbott Coalition supporters defecting to support other parties in disgust, ReachTEL implies that no more than three points of Coalition support initially moved to the Other column, and Essential suggests it's now even less.  Possibly ReachTEL's methods are more likely to pick up this kind of voter but it's also possible that even such a modest shift was very temporary in nature.  The Canning by-election also showed no sign of defections at all, with the Christian right parties failing to improve their vote, though the choice of a Christian candidate by the Liberal Party (and attacks by the left alleging he is a Creationist on very circumstantial evidence) may have discouraged such defections.

Essential also found 34% of voters saying Turnbull's elevation made them more likely to vote Liberal compared to 14% saying less likely, but that Others voters were more likely to say less likely.  Whether the 30% of Others voters saying this are splitters from the Liberals, or existing supporters of micro Christian parties, is unclear but I suspect mainly the latter.

Newspoll Records

The cricket-stats department can advise that these Newspoll results were notable in the following ways:

* The disposal of Tony Abbott ended a streak of 30 consecutive Newspolls in which the Coalition has lost the 2PP vote.  (Claims it was only 29 on social media were based on errors in graphs on The Australian's website.) This statistic, prominently used by Turnbull, exactly matched the streak of 30 consecutive losses covering Turnbull's own time as Opposition Leader.  The largest number of consecutive Newspoll 2PP losses by the same leader is 33 by Julia Gillard.

* The five-point shift to an incumbent government was the first since the six-point shift when Kevin Rudd was brought back in June 2013.

* Turnbull's +18 netsat was the highest since Julia Gillard's +19 the first time her netsat was measured in July 2010.

* Turnbull's 34 point Better Prime Minister lead was the highest since Kevin Rudd led then-new Opposition leader Tony Abbott by 37 points in December 2009.

* Bill Shorten's 21 point Better Prime Minister score was the lowest since a score of 14 in November 2009 by ... Malcolm Turnbull!

However we're still waiting for something I regard as a real indicator of Prime Ministerial popularity.  Not since Rudd in March 2010 have we seen over 50% satisfied with the PM in a Newspoll.  It was too early for Turnbull to have a real chance to crack that one this fortnight, but next fortnight will be a more real test of whether Turnbull is actually really popular, or just preferred by default.

Other Polling

Galaxy kicked ex-Treasurer Joe Hockey while he was down with a finding that 48% to 23% think Tony Abbott should have sacked him, and another that 41% to 16% preferred Scott Morrison to take over.  Labor voters especially supported both these concepts.

Essential found a strange boost in support for a republic, coinciding with the arrival of a republican Liberal PM.  They also found more Liberal supporters supported a republic than before, but this is undoubtedly due to pro-Turnbull voters returning to the fold.


Last week I reported that voters in my Not-A-Poll on Abbott's career prospects had mostly thought he would last longer than he did.  They also thought the Coalition would stay behind in the polls for longer than it has, and I suspect the same view of the Liberal Party's tolerance for failure was behind both of these things.  Congratulations to the 7% who successfully picked Sep-Dec 2015 as the time the Coalition would retake the lead (especially if you made your tip before last week); my own vote cast at the start of the year was for Jan-March 2016.


The Canning by-election has been thoroughly dealt with on the Canning live thread but a few general points deserve some summary comments here.  Firstly, the by-election margin is about what was to be expected for the general state of federal polling and indeed very slightly less than would be expected if we also consider the federal government's plight in WA polling and the personal vote of Don Randall.  It's a neither here-nor-there result, and attempts to portray the swing as larger than normal are mostly based on bogus baselines that fail to distinguish between government and opposition vacancies.

There is now no realistic doubt that the government would also have retained the seat under Tony Abbott.  Polling when interpreted properly suggests the impact of the change was actually only about two points, and even this is complicated by only one pollster having polled after the switch to Turnbull, and by evidence of house effects in polling.  The election results so far don't show any evidence of the change having much impact either way.  This doesn't mean the polls were wrong in finding a shift; it could just be that such a modest change has been lost in the noise of changing patterns of who votes by which method, and how much effort parties put into postal campaigns.  On the night I thought we might see a signal from the shift in pre-poll voting, but I was expecting the impact of the shift could be larger than either my own model or the polls in fact predicted!  (It clearly wasn't.)

The warning sign in the by-election results has been noted by William Bowe (may be paywalled).  The Liberals copped a 10% hiding in the city of Armadale, which might be seen as a proxy for any number of other outer-suburban "low-rent redoubt[s] of a white working class that has been progressively squeezed out of the inner suburbs".  A Liberal Party headed by a rich Sydney silvertail might have trouble appealing to the notorious "battler" vote, but the party would do well to learn the lessons of Labor's Western Sydney obsessions and realise these seats aren't the be-all-and-end-all of elections in this country.

Seat Betting Opens

Sportsbet recently opened odds on all House of Representatives seats.  At this stage the odds will not be much nourished by market wisdom (or otherwise) and will mainly reflect the views of the bookmakers' modellers, who are frequently pretty bright cookies.

The market currently predicts the Coalition to recover Fairfax from Clive Palmer and to lose Solomon (NT), Petrie, Capricornia, Bonner (Qld), Hindmarsh (SA) and Hasluck (WA), with most doubt about the last three.

Close seats (defined as more than one party at $3 or less) on the Coalition side are Eden-Monaro, Dobell, Reid, Barton and New England (vs IND) (all NSW), Flynn,  Brisbane, Forde, Leichhardt, Herbert (Qld), Deakin, Corangamite and LaTrobe (Vic), Swan and Cowan (WA), and Lyons (Tas), On the Labor side Parramatta is considered at significant risk.

This piece is probably more than long enough for now.  Over coming weeks when there is less polling action I may have some comments about what (in my view) the Turnbull Government needs to do to best distance itself from the Abbott mess without causing a conservative rebellion, and how it is going at doing that.


  1. Kevin what do you think about Parramatta being at risk?

    Labor won the seat in 2013 despite the Coalition doing well in NSW. It hasn't voted Liberal since 2001.

    Its interesting to note too that ultra marginal Reid and Barton are in the Coalition's column at the moment.The similarity between Parramatta, Reid and Barton (aside from all being in Sydney) is that they are very ethnically diverse. I wonder if this is a further trend of ethnic Sydney voters being far more inclined to vote Liberal compared to a generation ago.

    1. I think the argument in the bookies' modelling is that New South Wales wasn't showing as much swing under Abbott as the rest of the mainland states. Replacing an unpopular NSW-based leader with a popular NSW-based leader shouldn't change that. So if the Coalition is expected to win the 2PP by a modest amount nationally at the next election, then that means an expected swing of about zero in NSW - which means an ALP seat on 0.6% could plausibly fall. It's also possible it could be adversely redistributed. So I can see where the modellers are coming from.

      I note along similar lines that the current Phantom Trend model actually has the Coalition ahead in Parramatta and Labor ahead nowhere in NSW - this off a national 51.4% 2PP.


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.