Tuesday, October 20, 2015

Poll Roundup: Investments Attack On PM Fails

2PP Aggregate: 52.5 to Coalition (+1.4 in a week*, +6.1 since Abbott)
(Highest value since Turnbull became PM*)
Coalition would comfortably win election held now

(* Changes based on revised figures for previous weeks)

This week we've seen three new regular polls plus a partial result from another, and I've made some new decisions on how to deal with the recent behaviour of Morgan polling, so it's time to update my assessment of how the Turnbull bounce is going again.  This was a week in which new Prime Minister Malcolm Turnbull was attacked over the failure of his large financial investments collection to entirely avoid connection to the Cayman Islands.  It was pointed out (among the fine details of the Cayman situation) that this was also true of investments held by Labor politicians, leaving it unclear what exactly Labor was saying other than that the PM is rich.

Whether it was seen as grubby class warfare as has been widely alleged, whether it was actually paying the PM's economic skills a backhanded compliment, whether no-one can actually follow it all or cares, whether maybe even Labor had some kind of point buried somewhere but the commentariat weren't interested - for whatever reasons the attack has done no immediate polling damage, with no net change in two polls and what looks like an extremely good result in a third.



The main new polls this week are a 53-47 from Ipsos (their first national poll in two months), a 55-45 from Morgan and a 51-49 from Essential.  The Ipsos and Morgan respondent-preference figures were each one point higher.  There's also a Morgan phone poll for which they've released a 2PP of 54-46.  I think that 2PP is by respondent preferences, and from brief attempts to reverse-engineer the unpublished primaries, I suspect the last-election result was about 53-47.

Last week I mentioned that the strange behaviour of the Morgan poll is making aggregation especially difficult at the moment.  While Tony Abbott was PM, this poll (conducted by a combination of face-to-face and SMS mobile phone polling) tended to skew to Labor compared to other polls, but since Malcolm Turnbull became Prime Minister it has been a raging bull for the blue team.  On average, the three Turnbull-era Morgans have had the Coalition about three and a half points higher than the average of all other pollsters.  Some of this could come down to the weeks in which Morgan has been polling compared to the other pollsters, but every Morgan reading has been higher so far than every reading from every other pollster.

After much thought I've decided to apply a two-point house effect in Morgan for post-Turnbull polls for the time being, as well as the downweighting mentioned last week.  It's entirely possible that Morgan will start behaving like all the other polls now and cause me to revise that immediately, but at least then we won't have to worry about its odd behaviour anymore.  I've back-applied the change to every week since the change at the top as well, and the backdated readings since Turnbull became PM now look like this:

(Abbott baseline 46.4), 50.8, 51.1, 51.7, 51.1

After the changes, and considering primary votes, I aggregate both Ipsos and Morgan at 53 exactly, and Essential at 50.8, and the Morgan phone result at 53 (but downweighted to 1/4 normal value because of small sample size and having to estimate the last-election 2PP).  The current reading is 52.1 to Coalition.  If I do not apply the new house effect to Morgan, I get 52.6 to Coalition.  If I ignore Morgan entirely, I get 51.7.  There is a lot of bouncing around in the week to week results but the smoothed tracking graph gets rid of this.  So far as it's concerned, the Turnbull bounce has yet to peak, though with Newspoll's muted response to the surge so far, the rise this week might get flattened off by next week's readings.


Figures from other aggregates include: Mark the Ballot 52.7, Phantom Trend 53.3 (edit: 53.8 post ReachTEL), Bludger Track 51.9.

For the ultra-wonks Mark had a long and interesting piece (for those not scared off by that warning) which partly covered the surprising results of a 2013-election anchored model.  That model takes the assumption that the last election was a perfect observation of the house effects of pollsters at the time (an assumption not made in my model or Poll Bludger's, nor in Mark's headline model).  I'd expect such a model to produce a 2PP about 0.8 points higher than a sum-to-zero model, but Mark finds about twice that difference: the "anchored" model gets a Coalition lead approaching Morgan's.  Assuming it's not just a calculation bug, I wonder if the near-total turnover in polling houses and methods since the 2013 elections, together with strange behaviour from some of the pollsters, may have produced a risk for complex statistical models of house effects to get sucked into feedback loops that slightly distort their results.

Ipsos A Rich Vein For Turnbull

While the Ipsos poll did not hit the primary vote heights seen in Morgan, it was nonetheless as good a poll for the new PM as anyone would have realistically expected.  Quite aside from the 53:47 headline 2PP (this from a poll which on average leant slightly to the Coalition under Abbott), the leadership figures are glowing:

* Malcolm Turnbull recorded a net approval of +51 (68-17).  This is the highest net approval for a PM in the combined Nielsen/Ipsos series since Kevin Rudd polled +52 (74-22) in March 2009.  It is also the lowest disapproval for any new PM in the last 20 years.

* Bill Shorten recorded a net approval of -24 (32-56).  This is a new worst, in all regards, for Shorten in this series.  Furthermore the Turnbull to Shorten net approval difference (75 points) is the highest for at least 20 years, and exceeds the highest difference ever recorded by Newspoll (66 points, though the comparison is tarnished by Newspoll's higher undecided rates).  It is not an all-time record, as @sorceror43 reminds me that in October 1984 Bob Hawke had +53 to Andrew Peacock's -49, a difference of 102 points.  Two months later Peacock was popular and the election was close.

* Turnbull leads Shorten as preferred Prime Minister, 67-21.  This is the largest lead since Kevin Rudd recorded exactly the same result against, er, Malcolm Turnbull, in November 2009.

* Attribute figures show Turnbull leading Shorten by an average of 27 points, with knockouts like an 80-39 lead on "has a firm grasp of economic policy", 75-30 on "strong leader" and 83-51 on "competent".  Only the social policy question (60-55) implies room for improvement.

This is classic honeymoon-bounce polling and it remains to be seen how well it persists.  I also think that the difference between this Ipsos and the slightly anaemic responses of the other pollsters (bar Morgan) might be driven by method differences, and hence that some of the above response to Turnbull might be soft.  Live phone polling is more likely to pick up voters who say they like Turnbull because it is fashionable to do so rather than out of real conviction.

The Ipsos has attracted criticism based on its low Labor primary votes, especially in Victoria.  These should be taken in the context of a week in which Labor did something conspicuously silly, but even so, Ipsos is prone to getting high Green votes.  If it has the Labor primary too low and the Greens too high - which might easily be the case by a few points either way -  this mostly washes out on a 2PP basis since 83% of Greens preferences go to Labor.  That said, Green support is up since the 2013 election, and those who crossed from the Liberals to the Greens have almost all gone back, so it's quite plausible Labor has lost some primary votes to the Greens.  The 2013 Labor primary was bad (33%) but at times under Gillard it was well below 30.

As for the state breakdowns, Ipsos and before it Nielsen have often been targets for criticism because they actually publish state samples with small sample sizes and big error margins.  If Newspoll did the same every time it released a poll then it might often receive similar criticisms.  But it doesn't, so we don't know whether last week's 50-50 might have included a state result for Labor that was hard to credit somewhere.

There has been some attention to gender changes in the Ipsos poll.  The current poll shows just a two-point 2PP gender gap, compared with seven points two months ago.  However these changes are based on relatively small samples (around 700 of each gender each time) and the much larger Morgan samples (such as they are) have shown no such shift in the gender gap in voting intention.  The three Morgans since the change at the top have males 6.3 points more likely to vote Coalition.

North Sydney By-Election

A by-election is coming in North Sydney with the departure from parliament of former Treasurer Joe Hockey.  This is expected to be held before Christmas.  North Sydney (Lib, 15.9%) has been held by conservative parties of one stripe or another since Federation except for 1990-1996 when it was held by independent local former mayor, Ted Mack.  Despite Mack's vast local popularity, even he almost lost the seat at his first defence.

Labor is reported to be considering a possible low-key run for the seat but my advice to them on this one is to duck.  Given Labor beat the Greens by less than 5% for second place in 2013, there's a real risk of embarrassment by either the Greens or an indie should Labor run for the seat and not throw a lot of money at it.

Other polling

Ipsos found continuing support (54-33) for the China-Australia Free Trade Agreement.

This week's Essential Report (PDF link) includes various findings about trust in agencies and also a slight increase in apparent support for metadata retention.  It also finds 56:17 approval for Malcolm Turnbull's handling of "the threat of terrorism in Australia", compared to 46:33 for Tony Abbott in March.

Morgan's phone poll showed high support levels for the right to strike, especially and predictably on the left, and also showed slight net support for increased immigration.  It showed that Coalition and Others supporters are more likely to think immigrants are changing Australia's supposedly homogeneous "culture and way of life" for the worse than for the better.  It found that voters across the board tended to support population increase.  Although the tiny sample of Greens voters were more likely to reserve judgement, it's surprising that very few supported zero net population growth.  The poll also found predictable party skews on Muslim, asylum seeker and family reunion immigration.   At least, these are predictable once you notice that for modern Greens, international humanitarianism is actually more important than the environment.

It wouldn't surprise to get another ReachTEL soon; in any case I will probably post an update after the next Newspoll.

ReachTEL/Morgan Phone Woe For Shorten (Oct 23)

A new ReachTEL is out, as is a new Morgan Phone.  The ReachTEL has a 2PP of 53-47 off primaries of 46.7 Coalition, 33 Labor, 11.2 Green, 0.9 PUP. 8.2 Other.  I aggregated this at 53.3 using the decimal primaries.  This is really a major result, because until now it's been possible for someone to say they don't trust Morgan and consider Ipsos bouncy, and that the others are bunched around 51, so maybe the Coalition doesn't have much of a lead.  That's gone now.

There is also a second Morgan Phone, with their usual intermittent leadership polling, and a presumably respondent-allocated 2PP of 54.5-45.5.  For aggregation purposes I'm assuming the gap between preferencing methods here was again a point (reverse-engineering attempts won't ever give a definitive answer), and treating the two Morgan Phones this week as a single larger sample-size poll with a 2PP of 53.3. So in fact we really only have Essential and Newspoll holding out against figures of 53 and up from both Morgan methods, ReachTEL and Ipsos, and on that basis my aggregate goes up to 52.5.  This will also be its value following the weekly reset tonight.

ReachTEL had Turnbull up seven points as preferred Prime Minister (now 68.9-31.1, which given ReachTEL doesn't normally favour incumbents is a huge lead.) ReachTEL also had 40.2% to 26% believing Labor should change leader.  This includes 40.6% to 30.3% among Labor voters, bearing in mind that these are the core of Labor voters left over after those who were just hiding from Abbott have gone back to Turnbull.

That's probably a more damning finding for Shorten than the findings of Morgan Phone.  Despite a track record of reliability as a voting intention pollster, Morgan Phone has lately continually produced leadership results that are more extreme than those of other pollsters.  It's rather challenging to credit that only 9% of all voters think Labor should keep its current leader til the next federal election, including only 10% of all Labor voters, with both placing him behind the placebo option Wayne Swan as well as the familiar suspects Plibersek and Albanese (the former leading 27-23 including 34-21 among Labor voters).  It's doubtless bad for Bill, but can it be that bad?

The Morgan phone poll finds that Turnbull's ascension has overwhelming support with support for bringing back Tony Abbott as Liberal leader reduced to parity with "can't say" on 8%.  Even if Turnbull were to fall under a bus (as a result of his love for public transport?) only 10% want Tony to return to the ring, with more preferring Morrison and a lot more Julie Bishop.  Morgan phone also has Turnbull preferred PM over Shorten 76:14 (a 16-point increase in his lead), and with a +50 net rating (66:16), compared to Shorten on his worst in this series so far, -37 (25:62).  ReachTEL meanwhile has Turnbull at +31.6 and Shorten recovering slightly to -23.4.  ReachTEL doesn't replicate Morgan's finding of a net negative (-4) for Shorten among ALP voters, having his standing with ALP voters at +29.7.

Morgan draws comparisons with the famous "Mr 18%: Why on earth does this man bother?" John Howard Bulletin cover from late 1988.  It's not really a fair comparison though: Howard was up against a well-established PM who was well and truly de-glossed after five and a half years in office, while Shorten is up against a new PM in honeymoon phase who replaced an ex-PM almost everyone is glad to see the back of.

And speaking of that ex-PM, ReachTEL found voters weakly preferring (45.6:40.8) that he follow Joe Hockey out the door.  Coalition voters prefer him to stay.





No comments:

Post a Comment