Monday, August 20, 2018

Poll Roundup: Here We Go Again?

2PP Aggregate: 52.8 to Labor (+1.2 since last week) by last election preferences
52.2 to Labor with One Nation adjustment
Labor would comfortably win election held "right now"
That escalated quickly ...
Last night's Ipsos poll has produced a big blowout in my polling aggregate, wiping out all the Coalition's gains since June. However, Labor had already made modest gains in the wake of the Super Saturday results even before that.

Normally I put out a Poll Roundup every second Newspoll.  Last week's Newspoll had some points well worthy of comment so I was considering bringing this one forward (but got distracted by another issue).  Now with this extreme Ipsos result hitting the stands, I think it's worthwhile going one week early - especially with the noises being made in media and a feeling about that leadership problems could blow up into a challenge quickly.  Whether or not that happens, at the least we have another round of serious leadership speculation, and even if the Ipsos was twaddle it will probably turn out to be a self-fulfilling prophecy.  Events have developed quickly, with the PM shelving emissions targets in the National Energy Guarantee on Monday followed by news of possible Section 44 issues for Peter Dutton, which it seems the Dutton backers are intent on ignoring.  As I write the word is that Dutton backers think they have enough support and are deciding whether to go right away or wait until September.  In the past these things have often come to a head very quickly.

This all started with the Coalition's defeats in the Super Saturday by-elections, and especially the woeful primary vote result in Longman.   Had expectations on the right about the by-election results been more cautious, these results would by now have been a humdrum footnote in electoral history, and the Longman disaster would have been mostly written off as caused by candidate factors.  But because Liberal hopes of a victory in Longman and/or Braddon were boosted so high, especially by far-too-credulous reporting of unreliable and close electorate polling, the result has been seen as a disaster.

Throw in the chaos caused by rebellious backbenchers (and perhaps the odd Minister) over the NEG, and the PM's complete failure to placate these with his switch to a regulation mechanism, and this is all looking very familiar.  Now there is widespread speculation about a switch to Peter Dutton, who has routinely failed to much trouble the scorers in "smorgasbord" style leadership polling, and whose personal approval among the electorate hasn't been polled in ages.


Last week's Newspoll came in at 51-49 to Labor (again), however the 2PP stasis was accompanied by an unflattering loss of two points of primary vote to the poll zombie that One Nation has become.  (As the Longman vote showed, a lot of voters don't care what a rabble One Nation is or how flawed its candidates are - they vote for it to vent their disgust with the major parties).  I aggregated the poll at 52% to Labor by last-election preferences, compared to 51.8% for the previous Newspoll.  (Newspoll is using a method with a stronger flow of One Nation preferences to the Coalition than seen at the last federal election.)

What was interesting about the Newspoll was that although the 2PP didn't move, Malcolm Turnbull was clubbed on the personal ratings front.  His net satisfaction dropped from -6 (42-48) to -19 (36-55) and his Better PM lead over Bill Shorten was cut from 19 points (48-29) to 12 (44-32).  The netsat drop was Turnbull's largest in one poll since he dropped 16 points while coming down off his early honeymoon-period ratings in December 2016.  It was still only the equal 31st largest drop by a PM in Newspoll history, but that's not to say that it was nothing.

Among poll observers, for every one who makes far too much of Better PM despite its history of being a messy and skewed indicator, there's another who maintains that personal ratings are irrelevant and that only voting intention (especially 2PP) is worth anything.

Firstly, as has been known since at least the time of another Pollytics classic The Primary Dynamic, there's a very strong historic correlation between a Prime Minister's Newspoll personal rating and 2PP, which works whether one uses a PM's satisfaction rating or a PM's netsat (% satisfied - % dissatisfied).  The same thing basically doesn't work for Opposition Leaders (I think what little pattern there was must by now have been completely destroyed by Tony Abbott and Bill Shorten, who both polled large 2PP leads while highly unpopular.)

A sceptic might say "So?  Even if personal rating correlates with 2PP, that doesn't mean that relationship always applies.  If Turnbull's netsat drops 13 points but nothing happens to the 2PP, surely it's the latter bit that counts".

Possum's piece referred to the fact that personal ratings do something called Granger-causing 2PP.  This has been widely interpreted as meaning that PM satisfaction and netsat ratings are an indicator that moves ahead of the 2PP, so that if a PM's personal ratings get worse, the 2PP may follow.

What the "Granger causality" aspect actually means is this.  Suppose you have one dataset of Newspolls that has the 2PPs and the age of the polls.  You try to use this dataset to predict future Newspoll 2PPs in the series.  Now suppose you do the same thing with a dataset that includes the 2PPs and the ages of the polls, but also the PM's personal ratings in those polls.  Adding in the PM's personal ratings will help you to do a better job of predicting the next Newspoll 2PP.  (See suitably wonky explanation here.)  That, however, doesn't say anything about why this might be the case, or what sort of prediction you can come up with.

The prevailing interpretation of this has been that leader ratings causally drive the 2PP, so that if you can convince voters to disapprove of the Prime Minister you will also generally convince them to not support the government (1993 is always the exception).  However since the minimal interpretation of Granger causality is just that leader ratings contain information useful for predicting the future 2PP, I think what's going on might not be as dramatic.  One possible explanation is that because personal ratings vary over a wider range than 2PPs, they are also more likely to retain information that is lost from the 2PP figure as a result of rounding.

Newspoll again saw no real action on Bill Shorten's ratings (32-56), especially not his satisfaction rating which has stayed at 32% for the fourth poll running.  There have been three previous cases of a leader having the same satisfaction score four polls running (Keating 1996, Beazley 1999, Howard 2004).  Five in a row has never happened.

Essential and Ipsos

Essential last week came out with a 52-48 result to Labor, which I aggregated at 52.1.  I aggregated the Ipsos result at 55.1.  Ipsos is the bounciest of the current polls, but that is not to say it is especially bouncy by historic standards (it just looks so because Newspoll and, to a lesser degree these days, Essential are so stable).  There is a tendency for Ipsos to sometimes produce extreme results that lead to a Fairfax media house-poll frenzy in which the trend of other polling is ignored.  This may be connected to whatever it is in Ipsos that causes it to persistently have the Greens vote too high and to also have much friendlier ratings for Malcolm Turnbull than other polls.

The Ipsos primaries were again problematic.  Much has been made of the bizarre concept of Labor getting a 55% 2PP from a meagre 35% primary (especially when they got only 50.2% from a 38% primary in 2010) but the poll has a minor-party preference pool of 32%, including 13% for the Greens.  I don't believe the minor party vote is anywhere near that high, and nor do other pollsters.  It is very tempting to downweight Ipsos because of its primary vote weirdness but so far I have not seen much evidence that this affects the behaviour of its 2PPs.  This will be reviewed after the next election.

Essential had only fairly minor leadership changes - Turnbull's net rating down from +3 a month ago to zero, Bill Shorten up from -16 to -10, and Turnbull's Better PM lead cut from 42-25 to 41-27.  Ipsos had bigger changes with Turnbull's net rating down 19 points to -2, Bill Shorten up five to -11, and Turnbull's Preferred PM lead down 15 points to 12 (48-36).  However, these changes in the Ipsos poll could be down to it being an extreme sample.

The major question with the Ipsos poll is whether it is simply a rogue poll or nearly so.  It may seem out of the ballpark compared to the 52 of Essential and the 51 of Newspoll, but the 51 of Newspoll is a 52 when the same preferencing methods are used.  Moreover, as pointed out above, the Newspoll 2PP may have been undercooking the Coalition's problems anyway.  Would the events of the last week have done two or three points of damage?  My suspicion is not that much, but we'll need to see more polls to see.

There have also been some commissioned national ReachTELs in recent weeks - one for Greenpeace (52-48 for Labor) and one for the ACTU (51-49 to Labor).  My aggregate doesn't include polls commissioned by lobby groups and in any case I have not found full voting intentions details of the ACTU poll.

I may update this article with more polling snippets as they come to hand, or if I have time I may fill in some of the minor polling details I haven't commented on here.  In the event of a spill I may have a separate thread for that.  This depends on when the spill meeting is held as I am busy working during the day.