Tuesday, December 19, 2017

Poll Roundup: 2017 Year In Review

2PP Aggregate: 53.4 to Labor (-0.4 since last week, -0.8 in three weeks)
Labor would easily win election "held now"
Average for 2017 53.3 to Labor
By last-election preferences, Labor won all 93 polls this year

The polling year has just about come to an end so it's time for the annual roundup.  Should any further national polls appear I will edit in any necessary changes.

Since the last roundup things have improved slightly for the Turnbull federal government.  Following a brace of 55-ish results to Labor around mid-November, we've had two 53s from Newspoll, a run of 54-54-55-54-53 from Essential, 53s from ReachTEL four weeks back and Ipsos two weeks back (but the ReachTEL 53 came out at 54.7 by last-election preferences), and results from YouGov that came out to 53.1 and 54 by last-election preferences from YouGov.  (As noted further below, YouGov's 2PPs are wacky, so let's ignore them.)  The two 53s from Newspoll and today's 53 from Essential all looked like they were probably rounded down, and so my aggregate now sits at 53.4 to Labor.  Here's the (slightly) smoothed tracking graph:

The last few weeks are seen as having been good for the Coalition.  Same-sex marriage was finally passed without further complications, the Coalition won two by-elections (both more easily than was widely expected), Labor was embarrassed over Senator Sam Dastyari's Chinese connections, and two Labor MPs were referred to the High Court over citizenship issues.  The Coalition was also able (by a whisker) to escape having any more of its own MPs referred, thanks to Barnaby Joyce's re-election being declared just in time for him to appear in the Parliament and tie the vote to refer them.

My aggregate functions purely off last-election preferences and these were atypical in the case of One Nation.  If a historic average figure for One Nation is used instead, the aggregate is about 52.9.  The flow was around 65% in the Queensland election - perhaps slightly stronger once ignoring votes filtered through other candidates - but Queensland does have a history of slightly stronger flows than other states.  If that figure is used, the aggregate is about 52.4.

In management news, Galaxy has been acquired by YouGov and is to be rebranded "YouGov Galaxy".  It is unclear what impact if any this will have on YouGov's current YouGov-Fifty Acres series.  There has been some misreporting of this as YouGov acquiring Newspoll.  Newspoll is a News Corp brand which Galaxy presently administers.  This will therefore mean that YouGov owns the company delivering Newspoll, but not that YouGov controls Newspoll itself.


Newspoll showed only minimal recovery for PM Turnbull over the past month, with his net rating now at -25 (32-57).  Bill Shorten dropped three points this week to -24 (32-56).  Turnbull now has a still-anaemic seven point lead as better Prime Minister, 41-34.

Other pollsters continue to have a kinder view of these basic ratings.  Essential had him at just -3 (41-44) compared to Shorten -9 (36-45), and with a 42-28 lead as better PM.  Ipsos two weeks ago had Turnbull at -7 (42-49), Shorten -4 (38-42) and Turnbull preferred by a whopping seventeen points, 48-31.

Turnbull was seen as having received a lift from an Ipsos question that "found" that 71% of voters preferred that Prime Ministers be allowed to serve a full term rather than being replaced before an election.  This was widely uncritically reported (and not just by Fairfax), however this poll question was so badly designed as to make the result irrelevant.  The alternatives given (and don't try to make sense of that graph) were between expressing general approval for changing leaders between elections and saying one disapproved.  An obviously better design would have been to ask the voter whether they believed incumbent Prime Ministers should always be allowed to serve out their full terms or whether they should be removed mid-term by their parties in some cases.  However even then, voters may respond differently to an abstract question to when it is asked about a specific Prime Minister.  Despite general distaste for political musical-chairs games, the removal of Tony Abbott polled very well at the time.

That this (non-)result is not a filip for Turnbull should have been obvious from Ipsos' other findings, which included that Bishop (32) led Turnbull (29) as preferred Liberal leader, with Tony Abbott on 14%, Peter Dutton on 5% and Scott Morrison on 4. Obviously if 71% of voters held the view that incumbent Prime Ministers should never be removed, then at least 71% of voters, rather than 29%, would agree that Malcolm Turnbull should remain leader of the Liberal Party, at least for the time being.  Similar results on the Liberal leadership, with Bishop either ahead of Turnbull or else tied, have been seen in other recent polls.  YouGov, asking a question specifically regarding Turnbull, had 39% saying he should stay and 40% saying he should go.  (63% of Coalition supporters wanted him kept, which is not all that high really, but at least puts the lie, again, to views that Coalition supporters want somebody else.)

Anyway this Fairfax-Ipsos poll on leader-booting gets the rarely awarded Triple Boot Fish, collecting Wirrah Awards for Fishy Polling and Fishy Poll Reporting by (mainly) the sponsoring outlet, with a special bonus Wirrah for Worst Graph.  (Oh yes, there are graphs worse than mine.)

(image source)


On to the review section.  This year saw 93 federal voting intention polls released (down fifteen on last year): 49 Essentials, 19 Newspolls, 13 YouGovs, eight ReachTELs and just four from Fairfax-Ipsos.

In discussing the 2PP results, I am splitting the results into YouGov and the rest.  Of the eighty non-YouGov polls, the Coalition lost the 2PP in all of them.  It managed a single 49 from ReachTEL via an unlikely flow of respondent preferences.  Its worst was a 44 from Ipsos and everything else fell in the range 45-48.  The rather odd YouGov 2PPs - based on respondent preferencing with 75% of One Nation preferences supposedly flowing to the Coalition - gave the Coalition five wins, five ties and three losses, with a high of 52 and a low of 47.  However, all of these were losses by last-election preferences, and indeed by last-election preferences I have the Coalition losing all 93 polls this year with a best result of 48.5 and a worst of 44.6.  Overall I have the Coalition's average aggregated last-election 2PP for the year at 46.7.

No government has been re-elected after such a dud calendar year on the 2PP front since Labor in 1991, and no Prime Minister has won following one since Menzies in 1962.  However, only six previous governments have had a calendar year this bad at all, and of those three won the next election.  So it is much too early to assume from this bad year on the 2PP front alone that this government is doomed.

The Newspoll netsat story is a grim one, with both Malcolm Turnbull and Bill Shorten being in negative territory for the entire year.  Turnbull averaged -23 with a best of -13 and a worst of -30, Shorten averaged -21.5 with a best of -15 and a worst of -28.  Shorten has now been sub-zero for 34 months in a row, and has just snuck past Tony Abbott to trail only John Howard (as Opposition Leader, 41 months) and Paul Keating (as Prime Minister, 49 months) in that particular hall of shame.  Turnbull has now had a negative netsat for 21 months in a row, which places him fourth among Prime Ministers.  Turnbull was the "better Prime Minister" for the whole year again but his average lead of 10.1 points (best 15, worst 2) was not really a lead at all, since the indicator skews to incumbents by around 16 points on average.


Again I note some current betting figures, for historic interest rather than because betting is necessarily predictive.  Currently Labor is favourite and by more so than last year (1.48 vs 2.40, 1.62 vs 2.20 and 1.50 vs 2.60 are some prices available), implying a perceived 37-42% chance of the Coalition being re-elected.  One site still has 1.50 vs 2.30 on Turnbull facing a leadership contest before the next election, but has Turnbull at 1.80 to still be leader at the election (Bishop $3, Abbott and Dutton 5.50, Morrison 6 etc).  The markets still expect the next election to be in 2018 (and if so probably November or October) but less strongly than they did a few months ago.  While the odds continue to have Anthony Albanese as slightly more likely than not to be the next Labor leader, I still can't find anyone fielding on whether Bill Shorten will take Labor to the next election.

The overall picture

This was a year in which very little happened in the national polling picture.  The Coalition started the year with a February polling crash of the sort quite often seen before.  It spent most of the first half of the year in the 47s and most of the second half in the 46s, with the federal Budget completely failing to deliver an unwisely anticipated bounce.  The Government recorded the worst polling of its term in November amid the citizenship crisis, but has since recovered slightly.

The road ahead

The government has now been behind in polling for a year and a half, and well behind for most of that time.  Someone might say that this means they are cactus at the next election, and someone else might say it is irrelevant as we are too far from the next election for polls to be predictive.  We don't know if we're really, say, 10-11 months as opposed to 14-16 months out, but in any case, the truth is somewhere in between.  Governments that poll badly early in their term have a historic record of losing about half the time.  When I try to narrow that down to governments that polled really badly in the first half of their term, there are two problems: firstly there are very few data points, and secondly some of the worst-polling governments won.  Things may get clearer if the government still trails by this amount a few months out from the next election, but we may not know that we are a few months out at the time.

The same thing said last year applies: changing the Prime Minister has been a trick that has sometimes rescued governments that are polling poorly, but to do it two terms in a row would look more than a little bit silly and would only make sense if the party concluded it was facing certain defeat under Turnbull and wanted to minimise the damage.


  1. excellent work Kevin.

    Two things if i may. Changes in support you allude are within the margin of error so are statistically meaningless are they not?

    Secondly in New England it was a re-election not a by-election as the member did not resign, retire nor die but was declared ineligible by the court of Disputed Returns yes?

  2. 1) That depends on which changes you mean and what you mean by "statistically meaningless". At most times the margin of error of an aggregate is somewhere just under a point if the aggregate is treated as a random sample of equally weighted respondents. However since the respondents are not truly random and are weighted by the pollsters and the polls are then weighted by me, the real MOE is somewhat higher in theory and is also not precisely knowable. But this is further complicated by the "underdispersed" behaviour of Newspoll which, somehow, acts like a poll with a larger sample than it has.

    It might therefore be argued that, say, the 0.8 point improvement in the last three weeks is within the "real" MOE and therefore not necessarily real, but that ignores the fact that the start value of 54.2 was not a one-week reading but rather it stayed at that level for three weeks. One could, in theory, test via simulation what the chance is that an aggregate will hold at a level for three weeks then drop 0.8 in two weeks if there is actually nothing going on.

    However I'm not inclined to do this, because any aggregate reading implies a probability distribution and the implied probability distribution for a reading of 53.4 at this time is different to one of 54.2, even if there is not necessarily statistically significant evidence that the underlying value has changed. I might not say this if the experimental evidence showed that voting intention was generally static, but it doesn't.

    2) The High Court still calls it a by-election and the AEC still calls it a by-election. I'm not going to argue with them.

  3. Sorry I was under the impression from Antony Green it was a re-election for the above reason.

    Thus how can you compare what happened in Bennelong with any other be-election as the sitting member recontested the seat.

    1. Probably best to see my comments on Bennelong in other articles re that. Based on normal examples where a sitting member doesn't recontest, a sitting member recontesting in an urban seat is probably only worth a point or so. Of course, this varies between different sitting members and there is some evidence that Alexander may have had a somewhat larger personal vote than that.

      It is worth mentioning that there have been at least nine by-elections with recontesting sitting members. The two this year and the Jackie Kelly one in 1996 are well known, but between 1903 and 1920 there were six others, and in four of those cases the incumbent lost.

  4. Kevin,

    how can you compare Section 44 re-elections ( oky Alexander resigned before the CofDR could rule so in essence it is a re-election as the member was never a member) with normal by-elections?

    Sorry but things between 1903- 1920 do not rate IMHO

    1. We have many cases from normal elections where incumbents have not recontested their seats and from those we can get a handle on the typical impact of loss of personal vote compared to when a member recontests. Of course, things might be different in by-elections where voters might be more annoyed about an incumbent quitting, but there is actually no evidence of that, since the average swing against for a member resigning mid-term (and not recontesting) is much the same as the average swing when a sitting member dies.

      The difficulty with Section 44 by-elections is that there is an argument that the incumbent might gain a sympathy vote from being disqualified on a technicality, in addition to the personal vote benefit of recontesting their seat. However, it isn't clear to me if this is actually true or not. The Kelly case happened with the government in its honeymoon period, the opposition to Joyce was weak and Alexander had about the normal swing against him.

      Anyway I was not seeking to make comparisons in this article, just to point out that retaining these seats by a greater margin than generally expected has been seen as a good outcome for the Coalition. As for the historic cases, I mention them partly because I have seen false claims that there have only ever been three by-elections with recontesting incumbents, and partly because there were many early cases of disqualified incumbents losing by-elections, at both state and federal level - even when the disqualifications weren't their fault. It might be that standards have changed since but those suggesting there is a sympathy vote for being disqualified need to show that.