Sunday, April 8, 2018

Newspoll Number 30: Rolling Comments

"The one thing that is clear about our current situation is the trajectory. We have lost 30 Newspolls in a row. It is clear that the people have made up their mind about Mr Abbott's leadership"


Normally I am now doing federal polling roundups every second Newspoll (here's the latest) but the event that is very likely to happen demands its own thread.  In polling history, this is something very novel - a Prime Minister who seems about to meet the same standard of polling failure that he used as a justification for removing his opponent.  Judge for yourself from the link above how central a justification it was, but I reckon it was more than an aside.

Once we have the Newspoll result I will update this article and there are likely to be comments on various claims that are made as a result.  One remarkably silly false claim circulating on social media is that Abbott's 30 Newspolls plus Turnbull's 30 Newspolls will equal 60 consecutive Newspoll losses.  Turnbull did not start losing every poll immediately; his losing streak commenced 21 polls in.


How likely is it?

Based on the history of Newspoll, it's overwhelmingly likely Turnbull will chalk up another 2PP loss on Sunday night.  In Newspoll history, three-point 2PP jumps for the incumbent government of the time have happened 91 times out of 702 opportunities (I'm using Peter Brent's 2PPs derived from primaries for the pre-2PP era Newspolls, and excluding cases straddling an election at which government changed hands.)  That 13% rate might seem high, but the frequency has reduced sharply since the Galaxy-run Newspoll with a mostly larger sample size commenced in mid-2015.  In that time there have only been two such jumps out of 54 opportunities (less than 4%), and one of those was for a change of Prime Minister.

But three-point jumps become more likely to occur if the previous Newspoll was an outlier that was unusually low for the government, and that is not the case here.  Compared to a rolling average of the previous four Newspolls (with the most recent weighted double), jumps large enough to get us to 50-50 have happened 65 times in Newspoll history (9%), but excluding the Abbott to Turnbull transition there hasn't been one on the new Newspoll's watch.  The standard deviation of differences between the 2PP and the rolling average using purely Turnbull data has been only 1.09 points, which suggests only about a 1% chance of jagging a 50-50 or better.  I wouldn't take that estimate to the bank though, since it's always possible that voter knowledge of the event itself could influence the outcome, but it would be generous to give the Coalition more than a 5% chance of avoiding another Newspoll 2PP defeat.

Is it a record?

In Newspoll terms, Tony Abbott's 30 2PP defeats in a row, which Turnbull now threatens to equal, is not quite the record.  Labor under Julia Gillard lost 33 in a row in 2011-2012.  (The Coalition also lost 53 in a row in opposition from 2007-10, but that was with three different leaders.)  

In time terms, another loss this week will give Turnbull the longest continuous time period of Newspoll 2PP deficits by a PM since the pollster started in late 1985; again, I am including converted 2PPs for those polls for which no 2PP was published.  He will have been behind continuously for 18.5 months.  Looking at converted 2PPs for old Morgans for the years prior to the advent of Newspoll, there is actually only one case of a PM who would have been behind for longer - Menzies in the 1951-1954 term would have lost the 2PP in every Morgan taken over a period of 30 months.  However that 30 months included only ten polls.  (This didn't stop Menzies winning the election.)  

Who wore it better - Tony or Malcolm?

Tony Abbott's defenders (in particular the @TeamTAbbott twitter account) have been keen to talk up ways in which Turnbull's losing Newspolls have been worse than Abbott's.  Here are several facts in this regard:

* The average Coalition 2PP for Abbott's 30 losses was 46.8; under Turnbull the average has been 46.93.  Turnbull's average 2PP for his 30 losses will be higher than Abbott's for any Coalition 2PP above 43, a tie for 43, and below only if the 2PP is 42 or worse.

* The 2PP for Abbott's losses was more variable than it has been for Turnbull.  Abbott's losses had a standard deviation of 1.49 points but for Turnbull it has been only 0.82.  This can be spun both ways: that Abbott had more competitive 2PPs than Turnbull (five 49s to nil) and that Abbott had more very bad 2PPs than Turnbull (a 43, five 45s and nine 46s compared to just two 55s and five 54s for Turnbull).  To some degree this difference would have been caused by the old Newspoll being more volatile than the Galaxy version, but in Abbott's case there was a much stronger tendency towards runs of better or worse results than for Turnbull.  So Abbott's good spells were more competitive than Turnbull's but his worse runs were also more potentially disastrous.

* Abbott's average net satisfaction rating for his 30 Newspoll losses, of -24.9, was worse than Turnbull's so far (-23 so far), and will remain so unless Turnbull's netsat is -80 or worse (which will not happen).  Again Abbott's netsats were more volatile (standard deviation 8.3 vs 4.5) and this was mainly caused by Abbott having runs of very bad and less bad ratings.  The average poll-to-poll netsat change for Abbott was 4.7 points compared to 4.0 points for Turnbull.

* One of the starker differences between the pair comes in the area of Better Prime Minister ratings. During Abbott's 30 consecutive losses, he lost the Better PM "beauty contest" to Bill Shorten 16 times, tied five and won nine, for an average deficit of three points.  Turnbull however has led Shorten in all 29, with an average margin of +9.8 points.  Part of the explanation is that Shorten has been more unpopular during Turnbull's losing streak than Abbott's (average Shorten netsat -20.1 compared to -11.3) but there is still a large difference between Turnbull's Better PM scores and Abbott's even after adjusting for this.

* Turnbull's 30 losing Newspolls would cover 18.5 months while Abbott's covered only 16 months.  While this is being attributed to Newspolls being less frequently released (and even connected to the ludicrous urban myth that claims Newspoll switched from "fortnightly" to "monthly"), the main explanation is simply that Turnbull's streak included both the 2016-7 and 2017-8 summer breaks while Abbott's included only one summer break (2014-5).

* The Coalition's primary vote has been lower on average during Turnbull's losing streak, at 36.5% (high of 39, low of 34) compared to 38.3% during Abbott's (high of 41, low of 35).  The main difference has been strong primary support for One Nation during most of Turnbull's streak.

* Prior to losing 30 Newspoll 2PPs in a row, Abbott as Prime Minister had won four and lost five.  Turnbull as Prime Minister prior to his current losing streak had won nine, tied seven and lost four.  

An Unlucky Benchmark

Turnbull's choice of 30 Newspoll losses as a benchmark for rolling Abbott was an unfortunate one, because while in one respect it could have been easier for him to avoid the same fate (his 30 Newspolls have been taken over a longer period), in most regards it has been harder for Turnbull to avoid repeating it during a prolonged spell of poor polling.  

The first reason for this is that the Galaxy-run Newspoll (which commenced just a few months before Abbott was removed) has a larger sample size and is less volatile on that account alone, meaning that it has been more difficult for Turnbull to jag the odd rogue 50-50 by chance.  Former Nielsen pollster John Stirton suggests further that Galaxy may employ some kind of smoothing that reduces volatility further, and although there has been no confirmation of this it is difficult to explain how steady the new Newspoll's results have been in any other way.  (This uncanny steadiness was also a property of Galaxy's pre-Newspoll federal polling.)

A further disadvantage for Turnbull has been the aberrant preference performance of One Nation at the 2016 election, and possibly since.  Because some conservative pro-Abbott voters have started parking their vote with One Nation, but One Nation preferences in the few seats the party contested in 2016 split rather evenly, the 2PPs polled by Turnbull may have often been lower (by, say, half a point to a point) than how people would really vote at that time.  This probably doesn't apply to the whole sequence as there appears to have been a change in Newspoll's One Nation preference calculation methods from late 2017 onwards.

However, had Turnbull been observing the changing behaviour of Newspoll closely at the time he made his claim he could have known that in future, long runs of consecutive Newspoll losses would become more common.  

The Undercard

The build-up to the much-anticipated 30th Newspoll has seen a few other pollsters and/or poll sponsors jockeying for attention.  I had a number of concerned comments about the return of Morgan Face-to-face last week, and there's been another one of those out this week.  Well, sort-of, because this week's Morgan is actually only half new and also reuses half of last week's sample, a la Essential, due to "circumstance".  Morgan also says the results have been stable over the last few weeks, but that can only be the case in a 2PP sense, since the primaries have changed quite a bit.  We can now be clear that the poll is using respondent preferences (since the current poll at 51-49 to Labor would be 53-47 by the last-election method).  I am still holding off on aggregating this series until I see whether it is here to stay beyond the 30 Newspoll losses.

A Fairfax-Ipsos poll also came out, with a last-election 2PP of 52-48 to Labor (based on batched preferences) but 50-50 by respondent preferences.  Excluding the YouGov-Fifty Acres series this was the first 2PP tie from any pollster since September 2016, but it's rather hard to describe the respondent-preference figure as a headline 2PP as such.  The poll also featured flattering support for Turnbull keeping his job, with 74% of Coalition supporters and 62% of all voters saying he should be kept.  However, Ipsos has a history of favourable results for Turnbull personally compared to other pollsters, and there was more of this with the personal ratings (a positive net approval, 47-43 compared to 38-53 for Shorten, and a 52-31 preferred Prime Minister rating.)  

While some Fairfax sources in their usual house-pollster-spinning mode have been saying that this poll shows the Coalition is seriously in the game and will be noticed in Canberra as evidence of such, I don't think that is likely.  I suspect that to the extent Fairfax-Ipsos is noticed at all, beltway observers think it's that rather weird poll that comes out rarely, has the Greens too high, and bounces a lot.

Sunday

12:30 Peter van Onselen has tweeted "Newspoll WOW!" He is notorious for frequently tweeting variants of this comment even when an upcoming Newspoll (which he has seen) shows little to be carried away about.  In this case assuming he has seen it, it's unclear whether the WOW is for an unusual result (such as, say, Shorten hitting the lead as better PM) or simply for the 30 losses. [EDIT: Around 5 pm PvO tweeted that he did not actually know the Newspoll numbers, so the latter.]

6:00 Stupidity of the day: Many people on Twitter are quite seriously claiming that Bill Shorten will have won or the Coalition will have lost 60 consecutive Newspolls.  This is of course untrue, because of the non-losing phase after Turnbull first took over as PM.  When rumbled on it they instead switch to saying that the Coalition will still have lost 60 Newspolls, even if not consecutive.  In fact including the non-consecutive ones they will have lost 69, won 13 and tied 7 since coming to office in September 2013.

9:38 Social media reports that Patricia Karvelas has confirmed the Newspoll is loss number 30.

10:08 Account James_J on @PollBludger says Newspoll 52-48 with primaries L-NP 38 ALP 37 Green 10 PHON 7.  An account with this name was a reliable source on old PB site, I understand it's the same person.

11:00 Turnbull leads Shorten 38-36 as better Prime Minister.  This equals Turnbull's smallest lead so far and also equals Shorten's best score against Turnbull (but his best score against Abbott was 48).

11:22 Both Turnbull and Shorten have a -25 netsat with 32% satisfied and 57% not satisfied (yet another case lately of both leaders having exactly the same scores.)  Shorten's dissatisfaction score is the highest it's been since March 2017 and was last higher in January 2016.

A Neglected Record

Bill Shorten has racked up his 50th consecutive Newspoll Better PM loss.  This is in fact an all-time record; the previous high was Simon Crean's 47 in a row against John Howard.  As Better PM skews to incumbents, this record is of dubious value, but I mention it anyway.

6 comments:

  1. I think your summary on the Ipsos poll is valid, as you say Fairfax pay for it so they have to get their monies work.

    Also the duration of 18 months can be explained by the 2 summer breaks.

    ReplyDelete
  2. kevin will this loss tonight bring a total of 60 newspoll loses since the 2013 election and is there a graph or any info about this available buddy

    ReplyDelete
    Replies
    1. It will be Newspoll loss number 69 in total for the Coalition since the 2013 election, as well as 13 wins and 7 ties.

      Delete
  3. I'm surprised that people still fall for PVO's manufactured "OMFG Newspoll!!" tweets.

    ReplyDelete
  4. One of the interesting effects you have identified is the reduced volatility in Newspolls.

    I was surprised to read (Gratton via The Conversation)

    "The Fairfax-Ipsos poll had the Coalition trailing 48-52% on the two-party vote, when preferences were distributed, as is usual, on the basis of the last election. But distributing preferences according to how people said they would allocate them brought the result to 50-50%." I note this is still not LNP leading.

    Given the polls tend to be summarised (should that be trivialised) to a 2PP figure... Is the basis of current poll for primary vote, but last election for distribution of preferences really robust? Given a preference distribution is sought in the poll.

    Does it actually make a difference. Are both Abott and Turnbull Nil from 30 regardless of the methodology?

    ReplyDelete
    Replies
    1. Historically last-election preferences have usually been very accurate (see http://kevinbonham.blogspot.com.au/2015/09/wonk-central-track-record-of-last.html) but there are exceptions from time to time. At the moment it is likely that One Nation preferences would flow more strongly to the Coalition than they did in 2016 and that this could mean that "last-election" preferences have been underestimating the Coalition by somewhere between half a 2PP point and a whole 2PP point - perhaps even more than a point at the peak of the One Nation vote. However, Newspoll appears to have done something to adjust for this issue and over the last seven polls it seems they are no longer using the 2016 election One Nation preference flow but instead using something else, perhaps derived from the Queensland and WA state elections.

      Using respondent preferences creates a few issues. The first one is that it makes the 2PP result more prone to random variation - each poll will only sample a few hundred voters who do not support the major parties, so the split up of those preferences is volatile. Some respondent-allocated polls will give 2PPs that are two or maybe even four points different from the last-election result, but that doesn't mean the difference would really be anything like that big.

      The other problem is that respondent preferences are not very reliable anyway; they tend to exaggerate the impact of shifts in how people give preferences. One possible reason for this is that some voters follow How To Vote cards.

      Generally Abbott's government did worse on respondent preferences than last-election preferences. As for Turnbull, if Newspoll had used respondent preferences it's quite possible that he would have had a win or tie somewhere in the 30.

      Delete