Tuesday, December 18, 2018

Poll Roundup: 2018 Year In Review

2PP Aggregate: 54.2 to Labor (last election preferences) (+0.2 since two weeks ago)
With One Nation adjustment: 53.6 to Labor
Labor would easily win election "held now"
Labor won all 66 public and three commissioned national polls released this year

With the release of this week's Ipsos and Essential polls, the polling year has probably come to an end.  If there are any late polls I will edit this piece and update it accordingly.

For a government that currently looks as stuffed as a Christmas turkey, the end of the year cannot come soon enough.  As the final poll of the year, Essential offered some respite having the government only six points behind (47-53) but this should be treated with some caution as there is an ongoing difference of opinion between Newspoll and Essential as to just how bad the Morrison government's situation is.  Since Scott Morrison became Prime Minister, Newspoll has had the Coalition primary on an average of just 35% and the Labor primary on 40%.  Essential, however, has had the Coalition primary only narrowly behind (on average 36.9-37.2).  On a 2PP basis Newspoll has had an average reading of just 45.25% for the Government, while Essential has had 46.6% - and this is even though Newspoll's preferencing method is more favourable to the Coalition's than Essential's.  Currently, with Newspoll and Essential coming out in different fortnights, my aggregate bobs around a bit depending on which one is out, rather than based on the Coalition making substantial gains or losses.  If this continues into the New Year I may apply corrections to both.

Since the previous poll roundup we have had:

* Newspoll at 55% 2PP for Labor twice, which I aggregated as 55.7 and 55.9 by last-election preferences after considering the primaries.

* Ipsos at 52% and 54%, which I aggregated at 52.4 and 54.4.

* Essential at 52%, 54% and 53%, which I aggregated at 52.3, 53.9 and 53.

* A commissioned ReachTEL by the Australian Youth Climate Coalition at 54%, which I didn't aggregate under current rules on polls commissioned by activist groups, though it would have made little difference if I did.

The overall result is that the Coalition continues to fall away slightly from a post-coup best in the high 46s, and is currently on a possibly generous (because this isn't a Newspoll week) 45.8%.  Here's the smoothed tracking graph, which gets rid of the Newspoll-vs-Essential ups and downs:

Things are at least as bad now as in the worst weeks under Turnbull.  On the Newspoll front, the recording of three consecutive 45-55s marks a low not seen under either Abbott or Turnbull, and last seen in the last three polls (and at several other times) under Julia Gillard.  No PM in the Newspoll era has been re-elected after polling three consecutive 2PPs of 45 or worse, though the increased stability of Newspoll means this run is probably no worse than some survived by Howard in 2001 (and also, pre-Newspoll, Malcolm Fraser in 1979).

Following the Victorian election there have been a number of federal seat ReachTELs:

* The Geelong Advertiser gave a rather weirdly reported poll of Corangamite that could be taken as suggesting a 48-52 result, but in fact that was based off calculations with some impossible preference flows (all Greens to Labor and all others and all soft ReachTEL "undecideds" to Liberal); last-election preferences come out at only around 41-59 to 42-58.

* The CFMMEU had the Liberals losing Higgins 47-53 and Kooyong 48-52.

* The Australia Institute had the Liberals clinging on in Boothby 51-49.

Seat polling is highly unreliable (with an effective margin of error on the 2PP of at least six points) and seems to be getting worse with every fresh test it has had this year.  In Wentworth, the NSW equivalent of Kooyong, every seat poll that asked about it had the Labor vs Liberal 2PP wrong by at least 8 points.

I consider there to be a strong "kick them when they're down" aspect in the commissioning of union polls in Victorian seats with recriminations flowing from a dud state election result.  Of course, these polls are likely to get bad results right now.  The point of it seems to be to scare the government into thinking that the fire of potential seat loss (including in heartland seats) is everywhere, so it doesn't know which fires to put out and will waste resources shoring up safe seats, especially against unknown independents.

Is This Government Cactus?

We now have a fairly clear pathway to an election around mid-May, though in theory the government might still change its mind and go earlier, or might collapse during the February session.

Assuming a May election, after last week's Newspoll, my polling aggregate was such that no government had ever won from that far behind with that little time to go.  This was interpreted as me saying that I believed the government was "finished", but I wouldn't argue that just because recovery had never happened from a position meant that it could never happen.  Rather, I think a grim outlook for this government's chances of re-election arises from a combination of the bad polling, persistent chaos and the lack of any concrete reason to believe recovery is possible.  It's not a first-term government, it's not facing an unstable opposition or one with risky policies, polling hasn't been volatile except during the "spill" shock, and so on.  Recovery to an honourable loss (48.5-51.5 or so) is still quite possible given the history of federal voting intention leads often narrowing, but I wouldn't take that for granted either.

As of this week, the Ipsos and Essential results have brought it back to about the same position as the Whitlam government at the same point in old Morgans in late 1973, but that's the only precedent of any sort for recovery on this time frame at the moment.  A downturn in Whitlam government polling in late 1973 and early 1974 coincides pretty well with an oil price embargo shock at the time.

The AFR reported that a government source as saying their tracking polls are showing it is not as bad in marginal seat tracking polls as the national 45-55 results seen in Newspoll.  There are two things to note here - firstly after almost every election, the party that has had a disappointing result suddenly finds that its tracking polls were rubbish.  Secondly, as Andrew Catsaras has put it "the swing has to be somewhere".  If there is a national swing of 5% but the swing in the marginals is only, say, 3%, then there will be bands of "safe" seats with 7% swings, and given that the standard deviation of seat swings is around 3%, this means that somewhere seats on 10% will fall.  Thirdly the narrative of the swing in the marginals being different to the swing everywhere else is one of the eternal tropes of pre-election speculation, but it only really worked like that in 1998,


Scott Morrison's personal ratings continue to be strong given the carnage, but we have to bear in mind that he is still new and still being given the benefit of the doubt.  In the previous Newspoll he achieved something not achieved by any PM since John Howard when his netsat moved from negative to positive in the same term as PM (-8 to +1, it has since gone back to -3).  Howard moved from negative to positive fifteen times, but none of Gillard, Abbott or Turnbull ever managed it at all, while Rudd only did so briefly after a few years out of the job.  Bill Shorten's Newspoll personal ratings remain fairly poor (lately at net -13 then -15) but he continues to do about ten points better than when Malcolm Turnbull was Prime Minister, suggesting either a dividend for seeing off another PM or that the Coalition is too consumed with internal tensions to attack Shorten effectively.  Morrison recorded a surprisingly high Better PM lead in the previous Newspoll (46-34, the largest ever lead for a PM whose party trailed 45-55 or worse) but this has since come down to 44-36.  This is still high given the 2PP, but when you have a single-digit lead as Better PM, you are, in general, losing.

Essential found that perceptions of Morrison had worsened slightly since September on all bar two of the questions asked, but still found him comparing favourably to Shorten on all attributes except "Out of touch with ordinary people" and "Narrow minded".  Essential last fortnight also found better personal ratings for Morrison (up four points to net +8) than Shorten (down two to net -8) and had Morrison ahead 40-29 as better PM (again, this indicator skews to incumbents).  And Ipsos had very similar figures with Morrison on net +8, Shorten net -9 and Morrison ahead 46-37 as better PM.

In terms of election issues Ipsos found a 43-44 support-oppose response to Labor's policy on negative gearing, which doesn't really tell us anything given Labor took similar policies to the last election and escaped any backlash for doing so.  My biggest point of surprise here is that 87% of voters supposedly know what negative gearing even is.  A 43-48 response on capital gains tax changes also doesn't tell us anything about whether the issue is a game-changer.  In general, the measurement of whether election issues will actually change people's votes in Australian polling is very primitive and unscientific.


Now, on to the 2018 summary (click here for 2017).  For all the moaning in certain media circles about too many polls, the year saw just 66 national "public polls" released, down 27 on 2017, mainly because Essential switched to fortnightly.  In all we had 26 Essentials, 21 Newspolls, nine Ipsos, five ReachTELs and five Morgan-SMSs.  There were also at least three national activist-group commissioned ReachTELs.

The story in 2PP terms is simple: Labor won the lot.  The best the Coalition managed was 49%, which happened ten times from four different pollsters, all of them while Malcolm Turnbull was Prime Minister, and six of them in July.  The worst under Turnbull was the 45 from Ipsos immediately prior to his demise.  Since Scott Morrison became PM, the best for the Coalition has been a pair of 48s from Ipsos and Essential in mid-November, and the worst were two 44s from Newspoll immediately after Turnbull was removed.  On average my aggregate was 47.3 under Turnbull (range 46.1 to 48.7), 45.6 under Morrison (range 43.9 to 46.7) and 46.7 for the year as a whole, exactly the same as in 2017.

Prior to his removal, Malcolm Turnbull's average Newspoll netsat lifted to -15 with a best of -6 and a worst of -25. Scott Morrison has since averaged net 0.4 with a best of +7 and a worst of -8.  Bill Shorten averaged -20 with a best of -13 and worst of -25, but as already noted he has done better (with an average of -15.4) since the departure of Turnbull.

On the Newspoll better Prime Minister front, Shorten trailed Turnbull in every poll on this skewed indicator by an average 10.9 points.  Shorten led against Morrison in Morrison's first poll but has not done so since, with an average deficit against Morrison of 7.3.


Again I note some current betting figures, for historic interest rather than because betting is necessarily predictive.  Labor is now a heavy favourite to win the next election, with some example odds including 1.14 vs 5.00, 1.10 vs 6.00, 1.14 vs 4.50 - quite a bit of variation there but implying an 80-85% perceived chance of Labor winning.  In practice, probably implying more because election odds are prone to "longshot bias".

Seat betting currently has the government losing Boothby (SA), Forde, Flynn, Petrie, Dickson (!) (Qld), Gilmore, Banks, Page and Reid (NSW), Corangamite, La Trobe and notionally Labor Dunkley (Vic), Hasluck and Pearce (WA), with ties in Dawson (Qld) and Swan (WA) and no market currently in Chisholm (Vic).  With Wentworth as good as tied, that suggests an expected result somewhere around 85 Labor, 60 Coalition and six crossbench, though I haven't taken close seats into account.  The presence of "Non Jane Caro Independent" at 2.80 in Warringah is also worth keeping an eye on.

In summary

The government has now been behind in polling for nearly two and a half years.  Assuming it is still behind when polling resumes, it will become the longest continually trailing government ever.  It has only been briefly competitive, most notably in July this year.  After the failure to win seats in the Super Saturday by-elections (and especially the disappointing swing in Longman), Malcolm Turnbull was removed a few weeks later without waiting for much more polling evidence.  However, the spill process was a mess that was bewildering to voters, and ended with the Prime Minister being removed and replaced by a compromise candidate who wasn't even the primary challenger.  Polling has not at any stage suggested that this circus was acceptable to voters.

Some of the initial poll shock from the ousting of Turnbull faded over the next five weeks, but that plus any honeymoon effect for the new Prime Minister could only do so much before the loss of the Wentworth by-election kicked in.  The government has since then barely matched it with the worst points of the Turnbull era, and we will have to wait until February to see if there are any signs of it being able to improve ahead of the election.  A more common pattern in previous years has been for government polling to get worse when parliament resumes rather than better.


  1. I have no idea why pollsters ask the stupid PPM question. I have and never will have a say on the PM unless he/she happens to come from my electorate. In other words in my humble opinion the question and answers supplied are total rubbish.

  2. In my probably ill-informed opinion PPM is a bit useless unless one can somehow quantify incumbency bias. Sure it can be used to justify ones belief their team will win, or the other team lose, but theres a reason we all ask if someone is already acting in the position when deciding whether to apply for a job. Maybe a better line of questioning will be who will you be voting for, why, pick one of the local candidate, the leader, the policies, it's part of my identity man.

    1. The incumbency bias is about 17 points (I usually give it as 16 but have just checked the regression and it has gone up to 17 now). That is, when the 2PP is 50-50, the PM is on average about 17 points ahead on 2PP. When the opposition is ahead 53-47, it's usually about level, and for the current 2PP Morrison would normally be behind. (That he isn't is because of 1. he is still new and getting the benefit of the doubt more than his party is 2. Shorten is fairly unpopular 3. Possibly changed properties of Newspoll under Galaxy administration.)