Tuesday, July 25, 2017

Poll Roundup: Are Malcolm's Newspolls Worse Than Tony's?

2PP Aggregate: 53.2 to ALP (-0.2 since last week)
ALP would easily win election "held now"

Five weeks since the last Poll Roundup, things have not improved for the Turnbull government in opinion-poll horse-racing land. If anything, things have got worse.  We've had twin 53-47s to Labor from Newspoll and an Essential run of 52-53-54-54-53.  Closer 2PP readings from ReachTEL (52 then 51 for ALP) have arisen only because of the use of respondent preferences, and new entrant YouGov has produced a 49-51 followed by a 52-48 lead by a new respondent preferencing method off primaries that offer the government no more joy than the others.  (More on that later).  I'm not aggregating YouGov until later this week after its third poll has arrived, but my overall read of the others comes out at 53.2 to Labor this week.  Here's the smoothed aggregate:


The rot looks increasingly set in, with no large or lasting movement away from 53-47 since the start of the year.  As with the Gillard government, voters so far do not give this government credit for passing legislation or policy announcements. In polling terms, everything the government sends out comes back dead.  History doesn't say this position can't be won from, but it will probably need something large and unexpected to rebound in the government's favour.



I won't go into detail on leadership ratings and so on this time, as there is really not much to see, and I want to devote more space here to the metrics of the Turnbull Newspoll clock.  Since Turnbull used "We have lost thirty Newspolls in a row" as a justification for removing Tony Abbott, focus on how many Newspolls Turnbull has lost in a row and by how much has been intense.

Turnbull Vs Abbott Newspolls

As the self-inflicted "clock" strikes sixteen straight two-party preferred defeats,  some commentators and/or agitators have decided that April or so next year is too long to wait for Turnbull to lose another fourteen for his fate to be declared.  They've said Turnbull has already lost enough polls to be toast (Newspoll: Turnbull Has Reached The Denominator Of Doom) or they've said that while Abbott did lose 30 consecutive Newspolls, Turnbull's defeats are worse.

So, is losing a lot of consecutive Newspoll 2PPs a valid predictor of doom?  Peter van Onselen makes the point that no leader (government or opposition) has lost even 14 straight Newspoll 2PPs and gone on to win the next election.  Generally those who have lost that many in a row have been removed by their own parties without even making it to the next election.  van Onsolen writes "The only leader since Newspoll began doing fortnightly polls who has trailed for 14 or more polls and survived till election day is Howard."  What he fails to mention though is that Newspoll has only been releasing 2PPs outside of election time since late 2002.  This means that any observations about the fate of past leaders at elections are only based on a sample size of five elections.  It's really not enough to say too much.

It does turn out, however, that even going back to 1985 using 2PPs derived from the primaries (thanks again to Peter Brent for doing these on his old site some time back) the pattern holds: the most consecutive 2PP losses a leader has overcome and gone on to win the election would have been ten, by (who else) Paul Keating between April and August 1992.

Another line of attack has been that while Abbott did lose 30 consecutive Newspolls, in that time he was sometimes competitive.  Abbott's 30 losses included five 49-51 2PP scorelines but Turnbull's 16 straight losses have all been 48-52 or worse.

The problem with all of these claims is that they ignore something very significant: the switch from Newspoll being run by its own company to Newspoll being outsourced to Galaxy, this switch coming in mid-2015.  The switch involved an increase in sample size, a change from a single polling method to an amalgam of two methods, and perhaps most importantly of all, a switch to a company noted for producing polls with little poll-to-poll variation.

The following are the average 2PP movements from Newspoll to Newspoll (again using derived 2PPs where necessary) for PMs in the Newspoll era: Hawke 1.69 points, Keating 2.25, Howard 1.87, Rudd (first term) 1.75, Gillard 1.98, Abbott 1.55, Turnbull 0.74.  Spot the odd one out!

In terms of averages, Turnbull's losing Newspolls have had very slightly better 2PPs than Abbott's (47.2 vs 46.8).  His problem is that the static nature of the new Newspoll has greatly reduced his chances of picking up a stray lucky 50-50 or even a 49-51.  In the past, rough trots for leaders were often disrupted by outlier Newspolls that would have reset the 2PP clock if it had even existed.  For instance, here's a run of derived 2PPs for Howard in 2001:

45-46-43-45-46-47-44-48-50-46-50-46-47-51-48-49

Had the new Newspoll been in the field at the time, it's quite likely Howard would have lost all sixteen of these 2PP polls.  In reality the losing streaks were broken up by bouncy results.  Going back even further, there was the case of Menzies in 1954, who trailed badly in polling for almost the whole of his term, but was returned.  Had there been Newspolls in those days, Menzies may well have lost 50 straight.

It is under the new Newspoll's watch that Turnbull's Coalition has just recorded exactly the same 2PP five times in a row, an all-time record, but two of the three previous cases of four in a row also came since the change in Newspoll methods.  It should be clear that this statically rather-bad stream of results under Turnbull is a product of a polling method change, and not a fact about Turnbull's government.  Indeed, other polls that were in the field under both PMs have not become less dynamic in a similar way.

A final claim I want to mention here is Rita Panahi's claim that Turnbull has "made Bill Shorten look prime ministerial" (and earlier did the same thing with Kevin Rudd).  In fact Rudd was hugely popular against all leaders he opposed except Abbott.  As for Shorten:

* Turnbull has beaten Shorten by more as preferred Prime Minister in every Newspoll than Abbott did in any Newspoll after the 2014 budget (when Abbott even beat Shorten at all, which he often didn't).
* Shorten's average net personal rating of -22 is the worst by any Opposition Leader for a term since John Hewson's lame-duck term after losing the 1993 election.
* Shorten's rating on Newspoll attributes has worsened in every case since 2015, by an average of nearly ten points.

If anyone made Bill Shorten look prime ministerial, it was clearly Tony Abbott!

On the other hand, Newspolls during Turnbull's losing streak have been slightly less frequent, so his 16 are worth about 19 of Abbott's.

We Need To Talk About Respondent Preferences

A running theme in current poll-watching is the tendency of respondent-allocated preferences to favour the Coalition more than 2016 election preferences.  This is historically odd, because in the past respondent preferences have tended to favour Labor, and have usually (but not always) done so to an excessive degree compared with election results.  One recent time when the reverse applied was the Turnbull honeymoon period, but that's a long time ago now.

The obvious reason why respondent preferences could be different is the nature of current One Nation support.  One Nation preferences broke only 51% to the Coalition in the 2016 election, but they contested only a few seats, and decided in one of them that Wyatt Roy was "a little leftie" who they wanted to remove.  Historically, One Nation preferences break around 56% to the Coalition, so that's a difference worth about half a 2PP point at current PHON support levels right there.  But there could be much, much more, because One Nation has become a vote-parking lot for the disgruntled low-information right, so it could be that the proportion of One Nation voters preferencing the Coalition is actually much higher than at any time in the past.  If that's the case, voting intention is actually much closer than the headline Newspoll 2PPs are saying, and the Coalition's current level of poll-driven panic is unjustified.  However, past experience should make us treat the idea of radical preference shifts with a lot of caution.

Assessing what is going on with respondent preferences is complicated, for the following reason:

* None of the pollsters reporting respondent preferences (those being ReachTEL, Ipsos and YouGov) have reported very often since the last election.  Therefore there are not a lot of data.

* ReachTEL's June poll was found to be distributing the preferences of Nationals voters, which should in theory create a skew of nearly a point to Labor.  It is unclear if they are still doing this in their current poll.

* YouGov are doing something unusual, and getting 2PPs that so far have leant to the Coalition by 2.4 and 4.3 2PP points compared to last-election preferences from the same primaries.  They are asking voters to fill in a simulated ballot paper with full preferences.  It is unknown whether they are using a single or a rotated ballot order, but perhaps a single order with a high degree of "satisficing" by respondents might explain the difference from other polls here.  That said, a further complication with YouGov is that their "others" primary vote category has at least once included an item "Christian parties", meaning that their "others" might be a little bit more right-leaning than they seem.

A useful way to look at respondent skew in ReachTEL over time might be their recent commissioned seat polls that have published 2PPs.  The set linked to (of Cook, Curtin, Dickson, Flinders, Kooyong and Sturt and finding the Coalition easily holding all) shows an average respondent-over-last-election gap of 1.6 points, but this is greatly blown out by a 7.7 point gap in Cook, where One Nation forms over half of the third-party vote and Scott Morrison is supposedly getting 64% of all preferences.

Hanson Is The Sensible Centre of The L-NP

A very interesting piece of polling by JWS Research has asked voters to place themselves and various parties and politicians on axes of "politically left" vs "politically right" and "socially progressive" vs "socially conservative".  There is a certain skew in that respondents tended to rate all the parties as right-wing and conservative on average, but ignoring that the most interesting cluster involves Abbott, Turnbull, the Coalition and One Nation.  Voters rate Turnbull as well to the left of his party and Abbott to the right of it on social issues especially.  But they rate One Nation closer to the Coalition average than either the PM or the ex-PM.  This mirrors a view that Abbott and Turnbull are close to opposite extremes within their own party.

Phillip Coorey has suggested that this places Turnbull closer to the "sensible centre" of the political spectrum and flags this as an argument for keeping him as leader.  Conservatives might well object that once you remove everyone who will never vote Liberal in a fit, Turnbull is well on the left among those remaining, and that someone midway between Turnbull and Abbott (say, Scott Morrison) might be able to reach out to the centre and right at the same time, in a way that neither Turnbull nor Abbott can.

Speaking of Abbott, polling continues to show that his antics are widely rejected by voters, with almost half the sample of a recent Newspoll insisting that Abbott should zip it, and voters preferring Malcolm Turnbull's values to Abbott's by an overwhelming margin (just as they preferred Turnbull to Abbott 68-32 in a recent ReachTEL - tellingly, only One Nation voters choose Abbott).  Moreover polling of preferred Coalition leaders continually finds that Coalition voters support Turnbull more than supporters of other parties do, putting the lie to lazy claims that no-one who supports Turnbull would vote for them.  (This is a typical pattern - party supporters will tend to support the current leader at most times).  However polls that show voters considering Abbott to be unfit to return to the top job miss the point.  Abbott isn't in this personal validation crusade to be popular and isn't likely to stop based on what the voters think or even what it means for the next election.

There are many other polling matters I could probably cover but these ones will do for this month.  In a few days I will be introducing YouGov into my aggregate, though I'll be calculating my own 2PP off the primaries, and using a starting weight of about 0.5 or 0.6 in view of the poll's odder aspects so far.

YouGov update (26 July): I have been able to include YouGov faster than expected.  Their third poll (50-50) shows a respondent-preferences difference of 1.7 points, reducing the average difference so far to 2.8 points.  However the average primaries for the three polls so far also suggest the primaries may skew slightly to the Coalition.  Including YouGov made only a 0.1 point difference to the aggregate.

5 comments:

  1. Thank you for this very good article.

    ReplyDelete
  2. Thanks, Kevin. Insightful and worthwhile.

    ReplyDelete
  3. Can the increase in sample size alone explain the reduction in standard deviation of the polls? If not, what percentage does it account for?

    ReplyDelete
    Replies
    1. No and about a third. The increase in sample size should have made Newspoll about 20% less bouncy but it has become about 60% less bouncy (depending on how that is measured - one option is simply poll-to-poll change but another is difference from cross-polling aggregate; both methods give roughly the same answer.)

      Delete