Wednesday, May 30, 2018

Poll Roundup: Unwanted Records On Both Sides

2PP Aggregate: 52.3 to Labor (last-election preferences) 
51.8 with One Nation adjustment 
Coalition has improved 1.2 points in last eight weeks and now in best position since late 2016
However Labor would still almost certainly win election "held now"

It's been a little while since the last federal poll roundup (highlighting the major issue of Newspoll preferencing changes); I meant to do one in Budget week but was simply too busy with other things.  In the last five weeks the Turnbull government has kept the modest improvements in national polling that it made during April, but there has not been clear evidence of anything more.  Despite a lot of media excitement about the possibility of a quick election off the back of some possible success in the Super Saturday by-elections on July 28, we are so far not seeing anything in aggregated polling to get so excited about.

The recent national polls have been:

* Two Newspolls with headline figures of 51-49 then 52-48 to Labor.  As the previous article notes there has now been official confirmation that Newspoll is using a preference distribution for One Nation that is derived from recent state elections.  My aggregate's headline figure uses last-election preferences, and on that basis I aggregated these at 52.1 and 52.9 to Labor respectively.  The 51-49, as with the previous 51-49, was off primaries that would have normally come out to 52.4 by last-election preferences, but the fact that the published 2PP was 51 again suggested something a bit closer.  (An alternative view is that Newspoll might have changed methods twice, but we can't conclude that reliably off just two polls if so.  Also, this week's difference between the two methods was "only" 0.9).

* A ReachTEL at the start of May (52-48 to Labor by respondent preferences; I got 52.7 by last-election preferences off the primaries)

* An Ipsos in Budget week, 54-46 to Labor by batched last-election preferences (aggregated at the same).  Those who wanted to saw this as a tonic for the 51-49 Newspoll of the same week, but it's very clear which one was more outlying on a primary vote basis.

* A three-week run from Essential of 53-52-51 for Labor, all of which I aggregated at pretty close to face value.

Last week's Essential had the all-important (according to some) 4 in front of the Coalition primary, the first poll by anyone to do so since October 2016, with the exception of one of the briefly resurfacing Morgan Face-to-Faces back in March.

The Newspoll marked the 33rd consecutive 2PP loss for the Coalition under Malcolm Turnbull, tying the all-time record set by Labor under Julia Gillard.  Turnbull's 33 losses took 20 months compared to 16.5 for Gillard, but Turnbull has also been disadvantaged in the comparison by the lack of bouncing in the current Newspoll.  Indeed Gillard's losing streak was only busted by an implausible sequence of 45-50-46-50-49 in September-October 2012.  On aggregated polling, Labor under Gillard trailed from March 2011 until Gillard's removal in June 2013.

The Australian paid little attention to this as it was consumed by a record set by Bill Shorten (see below).  Here's the smoothed tracking graph:

Recently we have seen some volatile behaviour by One Nation (what a surprise!) with the party threatening the Coalition's preference supply over the way that Queensland LNP supporters are blaming One Nation for Labor's outright win in the 2017 state election.  One Nation has every reason to be aggrieved by such comments since the poor follow-rate of One Nation how-to-votes meant that the number of seats won by Labor on One Nation preferences was most probably one at most.  On the other hand, at least three seats were probably won by Labor because of the LNP's preference decisions (against the Greens in South Brisbane, against One Nation in Thuringowa and to One Nation ahead of Margaret Strelow in Rockhampton.)  I make the point about One Nation's threat to preference against sitting members because it gives another reason for caution about assuming that the 60+% flows from One Nation to the Coalition in state elections will necessarily repeat in the federal sphere.  In 1998, One Nation flows at state and federal levels were quite different.


The Australian this week celebrated Bill Shorten's 66th consecutive Newspoll with a negative net satisfaction rating, an all-time record for an Opposition Leader (beating Tony Abbott's 65 straight) but not really a patch on Paul Keating's 109 in a row as PM.  As with Turnbull's 33 2PP losses in a row, this is another record that needs to be put in the context of Newspoll being less volatile since its switch to Galaxy in 2015.  It's also a rather strange record, because to trail for that many Newspolls in a row an Opposition Leader needs to last, and the really bad ones usually don't.  Also, in time terms it's not quite a record (John Howard in his first term went just over two months more below zero).

For all the media battering he has received lately (most notably and deservedly over Labor's candidate vetting failing to live up to his triumphalism about it) Shorten's personal approval ratings in Newspoll haven't moved in ages - his satisfaction rating has been between 32 and 34 since last August, and his dissatisfaction has been between 52 and 56 over that time.  However, there has been movement on Malcolm Turnbull's personal ratings, which have improved to term bests of -11 (39-50) last fortnight and -10 (39-49) now.  Turnbull's 49% dissatisfaction is the first time either leader has had less than half of the sample aggrieved with them since the election, ending the longest run (in time terms) with both leaders at 50% dissatisfaction or more in Newspoll history.

There was more dramatic movement in the messy and overrated Better Prime Minister indicator last fortnight with Turnbull suddenly jumping from a 3-point lead (38-35) to a 14-point buffer (46-32), which this week stretched to an equal term-high 17 points (47-30).

It turns out this is the equal tenth largest poll-to-poll increase in a PM's Better PM lead over the same Opposition Leader in Newspoll history.  The following are all the increases that were as large or larger:

(Click for clearer version.  Note: in 1994 there was a three-poll period during which either Better PM was not surveyed or else the results have been lost.  Keating's position relative to Downer improved by 36 points over four polls during this time, so it's possible that there would have been further entries for Keating vs Downer had the data been available.)

[Update 25/12/2019: I have found two of the missing results and Keating-Downer 5-7 Aug 1994 goes to the head of the class with an 18-point swing from 30-47 to 37-36, caused by Downer's disastrous Central Australian trip.]

It can be seen that what has happened here tends to happen for a reason - none of these gains are difficult to explain.  However the current one could be down to Labor's embarrassment over Section 44 disqualifications, or to favourable response to the Budget, or even to the Coalition polling more competitively than before.

Ipsos' leadership figures are often Turnbull-friendly even when their voting intention results are anything but, and it's really hard to see why the use of live phone polling should still be causing this after so long.  (Indeed, if there was once a social-desirability bonus in praising Turnbull, one would think these days it's more the opposite.)  I suspect it's more of a sampling quirk/error, perhaps allied to whatever causes them to have the Green vote so high.  In this case Turnbull had a +12 net approval (51-39), Shorten was exactly the other way round, and Turnbull led as preferred PM by 52-32.  Turnbull's net rating was up eight points since the last Ipsos six weeks ago and the rest were little changed.  Essential, however, does not find such a large difference between the leaders and had Turnbull on a net -2 and Shorten on net -4 early this month.

Shorten again scored poorly in a Newspoll of preferred Labor leader, polling 23% to 26% for Anthony Albanese and 23% for Tanya Plibersek.  However, he still held the lead among ALP voters.  It would be interesting to see some two-leader comparisons (Shorten vs Albanese and Shorten vs Plibersek) to see to what extent those preferring one rival over Shorten also prefer the other.


There is a very large volume of Budget polling and I don't intend discussing it all.  The general media message was that the Newspoll Budget polling was good for the government, but it was actually a rather mixed bag.  The Budget joined the Budgets from 2004 to 2007 only in getting a (albeit smaller) net thumbs up from Newspoll respondents in terms of expected impact on their own finances. However the Budget didn't score all that highly when it came to perceptions of its economic impact, and in terms of the question about whether the Opposition could do any better, it got the equal weakest rating of any Coalition budget (equal with the 2014 stinker.)  It comes out in quite an unusual place on the graph of personal vs economic impacts:

Overall voters seem to have seen this as a vote-getting Budget that was generous but not all that economically responsible. Also of interest is the discrepancy between Newspoll's economic management ratings of Turnbull vs Shorten (48-31 to Turnbull) and Scott Morrison vs Chris Bowen (38-31 to Morrison).  The difference between the two ratings is largely down to Coalition supporters.

More Poor Poll Reporting In The Oz

It's disappointing to report that the national broadsheet, home of the most predictively reliable voting intention poll, has besmirched itself by commissioning another dubiously worded issues poll and allowing repeat offender Simon Benson to go silly with the results.  The Australian's recent poll on company tax cuts itself has been panned for sins of omission in the preamble wording (eg see Adrian Beaumont's writeup).  However it is the use made of it by the newspaper that is the bigger problem here.  The poll question stated the tax cuts as a fait accompli, and then asked how they should be implemented.  This makes the "Not at all" option underestimate opposition, since some respondents might not have supported the tax cuts in the first place, but might believe the government should nonetheless go ahead with what it has said it will do, especially when asked when it should be done.  For the same reason, adding the "As soon as possible" and "In stages over the next 10 years" options together would not meaningfully show the rate of support for the plan even had the plan been accurately described.

Yet this didn't stop Benson from delighting the culture warriors with a claim that tax cuts were more popular than same-sex marriage (ignoring, alongside the preamble's defects, that an opt-in postal plebiscite is not a poll and that generic polls on same-sex marriage actually tended to show higher support than the actual outcome).  And this was the second time in two polls that Benson was talking nonsense.  Last fortnight, Newspoll found 46-43 opposition to a referendum on allowing dual citizens to become MPs  but Benson reported this as "Most voters are also opposed", and also referred to "A majority of voters, 46 per cent".  Not only is 46% obviously not a "majority" or "most voters", but the three-point margin is within the poll's margin of error anyway.

The tax cuts poll was widely claimed to be "push-polling" on social media but (for the umpteenth time) it wasn't.  A push-poll is a message in the guise of a poll, distinguished by the use of a huge "sample" size (to get the message out as far as possible) and by the sponsor's disinterest in making any use of any "results".    Polls with normal sample sizes but dubious wording (which might then generate a result that could be used to persuade politicians or voters that something had popular support, or simply as a news hook to grab publicity for the sponsor's views) are something else.  I call them "skew-polls".  The terms "advocacy polls" and "hired gun polls" are also used overseas, but advocacy polling can include commissioned polls that are fair and well designed, though in my experience those are a very small minority.

There is a long track record of right-wing media commissioning dubiously worded polls to make political points, but the Murdoch press are far from the worst offenders, just perhaps the most influential.  (It was interesting to see One Nation senators scurrying after the poll result and trying to conduct their own opt-ins to test it.)  I see a lot more of these things from environmental groups.  

No comments:

Post a Comment