Friday, February 15, 2019

Poll Roundup: Mixed Start To Year For Struggling Government

(17/2: Ipsos update added at the bottom. Also, Not-A-Poll added in sidebar: predict the next Newspoll 2PP)

2PP Aggregate: 53.7% to Labor (last-election preferences) (-0.4 since end of polling season last year)
53.2% to Labor with One Nation adjustment
Labor would easily win election "held now"
Government is now Australia's longest continually-trailing government in polling ever

Time for another roundup of the federal polling picture in the lead-up to an election expected to be held in mid to late May.  At the end of last year the Coalition's polling had been going slowly downhill after the initial recovery from the shock caused by the messy removal of former PM Turnbull.  Early this year there were some early signs things might be improving, but a bad Essential poll this week has somewhat muddied the waters.  In this article I'll just be discussing voting intention and leadership ratings as it is long enough without covering more.

This week has been a dramatic week in parliament and we still have to see how that plays out.  However, a change to laws affecting medical treatment of people on Manus Island and Nauru is not the same thing as an incident like the Tampa, the 2002 Bali bombings and the 9/11 attacks - all of which produced substantial poll movements.  The law change may lead to a major incident (perhaps orchestrated) which could affect polling, but here is how things stand for the time being.

The polls so far this year have been:

* Three Essential polls, with headline 2PPs for Labor of 53, 52 and 55.  After considering the primaries I aggregated these at 53.1, 52.1 and 55.1.  This year, Essential has changed its source panel of survey respondents (the large list of voters from whom each survey's potential respondents are selected), but has not changed how it asks its questions or any other method details.  It will take some time to tell whether the panel change has made any difference.  It should also be noted that the second of the three polls came out a week later than normal, with an extended survey period (apparently on account of the Australia Day holiday).

* Two Newspolls, both with headline 2PPs for Labor of 53.  The second of these was notable for having a One Nation primary of only 5%.  Aside from an outlier last February, at which time the One Nation vote in Newspoll dipped from 8% to 5% then bounced back to 8% again, this is the lowest One Nation vote in Newspoll since late 2016.  It puts the party back to what it probably would have polled at the last House of Reps election had it contested every seat.  Minor softening in the One Nation vote since the replacement of Turnbull may be caused by voters who were parking their vote with the party being more comfortable with Scott Morrison as PM, but may also be caused by direct competition from other parties - most visibly Clive Palmer's United Australia Party.  Bearing in mind that Newspoll's preferencing formula assumes a strong flow of One Nation preferences to the Coalition, I aggregated these Newspolls at 53.6 and 53.9 to ALP.  Still, the government did well to break even in the first poll of February compared to the previous poll - historically, governments have gone backwards at this time on a 2PP basis 20 times out of 34, and forwards only seven.

* There has also been a Morgan of which partial details are available through the company's daily newsletter.  The 2PP is not published (but the report says Labor is "comfortably ahead").  The main primaries are Labor 36, Coalition 34.5, Greens 12.5.  Five minor parties (One Nation, UAP, KAP, Australian Conservatives and Christian Democrats) were in the readout and totalled 5.5%, with One Nation getting 3% and UAP 1%.  Others besides those listed (including independents) scored 11.5%.  Based on the polling period (a fortnight) this was probably a face-to-face poll, which creates social-desirability biases in favour of Greens and independents, and against One Nation.  Morgan polls are not very reliable and I give them a very low weighting.  I aggregated this one at 54.7% to Labor.

With all these included, after finishing last week at 52.6% to Labor based on very few polls (the closest things have got under Scott Morrison) my aggregate has now blown back out to 53.7.  Here's the smoothed tracking graph:

The government has now trailed continuously since the first polls taken just after the 2016 election.  This means it has been in arrears for 31 months, passing the previous record of 30 set by the Menzies government between 1951-1954 (during a 37-month term of government at the end of which it was re-elected).  We can say that no government that has trailed for this long now has been re-elected, but that is not saying anything useful given that no previous government has trailed for this long at all.

A more useful indicator is time to go before the election.  The government has been dipping in and out of having the status that "no government this far behind with this little time to go has won".  In the Newspoll era (since 1985) one government that was polling about this badly at about this time won, and that was the Howard government in 1998.  (Another two that were polling worse and one that was polling slightly better lost).  However, the Howard government in 1998 had the luxury of being able to win with 49.0% of the two-party vote, thanks partly to large personal vote effects from the 1996 election.  At this election, the government is coming off a loss of seats in 2016, so the personal vote effects in close seats help the Opposition, and 49.0% is unlikely to be enough.

From the pre-Newspoll era there is also a precedent of sorts for winning from this position: the first-term Whitlam government's Morgan poll in February 1974 was equivalent to exactly where the Morrison government is now.  But that is only a single poll, and it's possible Morgan was underestimating Labor in that year.

2004 is not a precedent.  An article by Peter van Onselen this week referred to John Howard's Coalition being behind 46-54 in Newspoll as the 2004 campaign was called (in fact this result was nearly two weeks old at the time).  But that was the year in which Newspoll blundered by using respondent preferences, contributing to a 2.7% 2PP miss on the final result (so the 46 was really a 47).  Also it was only a single poll, an outlier in a sequence that ran 48-49-49-50-46-48-50-50.  On aggregate, Labor was never that far ahead once the early-2004 honeymoon bounce for Mark Latham as leader had subsided. A professor of politics should know better than to treat a single Newspoll as a reliable indicator of where a government is standing.

What Chance Does This Government Have?

The usual pattern of federal elections from this point is that the margin narrows - eight of the last 11 elections have finished at least 0.9 points closer than the aggregated polling just over three months out, with a minor blowout in 1996 and a major blowout in 2004.  The only case where the party clearly behind in 2PP polling at this point ended up winning the 2PP was in 2001, though in 1993 the trailing party retook the lead before losing.  Polling at this stage explains about half the variation in the final result, and it doesn't become more predictive until the final weeks.  A linear regression for the Newspoll era currently suggests the government is tracking for about 48.5% with a standard deviation of 1.4%.  All else being equal, that gives it only about a 12% chance of reaching the 50.2% 2PP that I estimate it needs for a 50% chance of winning.  The chance of the government winning from here based on historical data is probably better than that, because it might get lucky with the seat distribution and win with, say, 49.5% 2PP, but to give it a 20% chance would seem generous.

The betting markets currently imply around 21-24%, which probably contains a dose of "longshot bias", but also reflects that Labor has been spending electoral capital in expectation of a likely victory.  Examples of this include Labor's franking credits changes (risking votes for greater budgetary comfort in office) and this week's action on medical evacuations (risking votes on its right flank to protect its left flank from the Greens.)  On the other hand, the famous Howard recoveries were all from things that had made that government temporarily unpopular or its opposition briefly attractive, while this government has been polling badly for its whole term.  There's a rather strong argument on that basis that there won't even be much narrowing to speak of this time around (a la 1996) and the government will simply be belted.


A brief mention of Newspoll leadership results.  Newspoll this week had Scott Morrison improving to a -2 netsat (43-45).  All nine Morrison netsats so far have been in single figures, either negative or positive.  This is the third longest such run in Newspoll history. Bob Hawke recorded 13 such netsats in 1986-7 prior to a slight lift in popularity in the leadup to the 1987 election.  The record is held by John Howard, who recorded 34 consecutive single-figure netsats in the leadup to the 2007 election, a sequence ending in defeat.  Voters had made up their mind that Howard was OK but it was time for a change.

Bill Shorten (netsat -15 (36-51)) continued his run of more benign but still slightly unflattering ratings since he saw off Malcolm Turnbull.  Perhaps surprisingly, Shorten is the first Opposition Leader in the Newspoll era to receive lasting credit for outlasting a PM in this way.  Tony Abbott found his ratings worsening at first when Julia Gillard was rolled by Kevin Rudd, while Shorten (removal of Abbott), Abbott (removal of Rudd) and Hewson (removal of Hawke) got bounces lasting only one or two polls.

Morrison continues performing well on the Better Prime Minister figure (44-35) given how far his government is behind, but given that Better PM has a skew to the incumbent of about 16 points when the 2PP is 50-50, even that indicator still suggests the government is losing.

Essential in mid-January had Morrison on net +4, Shorten on net -12 and Morrison 12 points ahead (42-30) as better PM.

Seat Polls - Can We Please Just Ban Them Until Full Release Of Details Is Compulsory?

Seat polls have proven very unreliable in recent elections (click "seat polls" tab for numerous articles on this topic) but continue to be reported with little restraint and frequently inadequate levels of detail in various parts of the media.

A further curve-ball lately has been a proliferation of "uComms ReachTEL" polls, which I shakily understand to be a different pollster using ReachTEL's technology and question formats but presumably not its weightings.  It is very difficult to get to the bottom of which of these seatpolls reported as ReachTELs are uComms ReachTELs as opposed to vanilla ReachTELs.

(Snarky note: I have so lost hope in the mass media reporting polling adequately that I will, from now on, hold everyone who commissions a poll and does not publish it in full or email it to me in full within 24 hours of release responsible for all shortcomings in the public reporting of the poll.  My email address is .)

1. Warringah

Wirrah Award For Fishy Poll Reporting

A GetUp! ReachTEL of Warringah was initially reported (by both the SMH and the Guardian) without the name of the pollster or any details of primary votes, however these details were later added in the SMH results (but not the Guardian's).  The SMH result did, to its credit, mention concerns about the reliability of seat polls, while the Guardian completely failed to.

The poll found Abbott trailing independent Zali Steggall 46-54 2PP off primaries of 37.7 Liberal, 22.3 Steggall, 15 Labor, 9.6 Green, 4.9 for Moylan-Coombs (another independent), 5.7 other and 4.9% "undecided".  The independent candidates were rightly named but the party candidates weren't, though in the case of Abbott it's hard to say if this would hurt his vote or help it. Yet again the curse of the ReachTEL undecideds struck with these unprocessed headline figures resulting in a spurious claim of an "almost 14%" swing against Abbott, when it was probably more like 12% (although we don't know because inadequate details were published).

On the primaries as published, Steggall would be getting about 83% of preferences. Kerryn Phelps got 79.4%, and that was with the benefit of some voters who would have preferenced her putting her 1 for strategic reasons, so this isn't unrealistic.  (In 2016 the less competitive ALP and Greens still managed 72% of preferences against him.)  If there is solace for Abbott here it is that the seat polls in Wentworth understated the Liberal vote by 5% on average, which if repeated would lift him to a level where he'd probably just hang on.

The polling in Wentworth recently was a trainwreck, both in terms of accuracy and the levels of public detail available.  Warringah is an important contest that many Australians would have almost as much invested in as the result of the entire election!  It's very disappointing to see public information about voting intention in this seat start on such a sloppy footing and I hope that all involved (including the pollster with its treatment of "undecided") will lift their game. Date: (I can't find that information either but sometime before Feb 10). Sample: 622

2. Pearce

... and another one ...

Christian Porter's seat of Pearce has frequently popped up on the deathwatch and is currently at $1.58 to change hands.  A GetUp! uComms ReachTEL recorded a relatively benign 52-48 in Porter's favour, a swing from the last election of only 1.6%.  However the reporting outlet, in this case the West Australian, has not published the primaries online.  I haven't been able to find them anywhere else, so can't comment about whether the respondent preferences were believable or not.

Worse, the outlet gave no indication that seat polls are inaccurate, and in fact treated this flimsy evidence as the end of a theory that Porter might seek to switch seat, and even as proof that he was safe. Safe, when the average error of classic 2PP seat polls in 2016 (3.1%) easily exceeded his margin in this one?  When the aggregated swing in WA polling of federal voting intention is 6.6%?   Porter could well win (indeed his tight margin from 2016 might be an irregularity) but to call him safe on this is a bit like saying, based on a five-minute checkup by a first-year med student, that a patient certainly doesn't have cancer. Date: Jan 16 (and took nearly a month to emerge) Sample: 674.

3. Herbert

We turn to a poll reported with an adequate level of detail (for once) and it's a Newspoll of Herbert, the closest seat at the 2016 election. Continuing the trend of finding not much going on on the 2PP front that didn't serve Galaxy too well in seat polls in 2016, the Newspoll found a 51-49 result to Labor (a 1% swing) off primaries of 32 apiece for the majors, 9 each for One Nation and Katters Australian Party, 8 for United Australia, 7 for Greens and 3 for others.  The pollster would be expecting the improvement in Labor and decline in LNP primaries to be offset by much more conservative One Nation preferences than last time.  The kicker was an approval rating poll for former MP Clive Palmer, finding him with a net rating of -41 (24-65) and suggesting that even if his big-spending campaign can continue to take votes from One Nation and the LNP, he won't be a threat in the seat. Date: Jan 24. Sample: 509.

4. Higgins

A uComms ReachTEL commissioned - get this - "by a ginger group within the Liberal Party who want former Member for Higgins Peter Costello to return to his old seat" found the Liberals trailing 48-52 in the race for Kelly O'Dwyer's seat.  Poll Bludger gives primaries after fixing undecided at Liberal 40.3, Labor 27.1, Green 19.3, leaving a rather hopeful 13.3% probably looking in vain for a good independent.  On these numbers, the Liberals would be likely to lose to Labor (though 48-52 seems a little bit harsh), but ReachTEL has often badly underestimated the Greens in inner-city seats.  Date: Jan 24 Sample: 860

5. Flinders (1)

Another uComms ReachTEL, this time conducted for the CFMMEU, had Labor leading in Greg Hunt's seat with a 51-49 2PP, but this was before Julia Banks entered the race and may therefore be redundant.  The Liberal primary was 36.8% prior to distribution of the undecided, so probably about 40% in reality.   An issue with these union polls is that we have no idea whether the unions are releasing all the polls they have conducted or just the ones they want to Date: Jan 24. Sample: 627.

6. Flinders (2)

As I write a new uComms ReachTEL of Flinders is being reported on Twitter with Labor supposedly beating Hunt 52-48 and Banks supposedly beating him 56-44.  After redistributing the undecided, primaries are 40.7% Greg Hunt (Lib), 31.1% Labor (no ALP candidate endorsed yet), 17.0% Banks, 5.8% Green, 5.4% others.  The stated 2CPs would require Labor to get 66% of preferences (ambitious since they got 71% last time, but more than half of those were from the Greens) while Banks would get a totally implausible 93% (though on those numbers she wouldn't be second.)  Tactical voting could be a big talking point in this electorate.  The poll also asked voters who they voted for last election, on which basis it had the Liberals 4% too low and Labor 7% too high, so there is room to doubt how representative this sample is.  Nonetheless it does hint that Banks could be dangerous - if she can get into second.  I may have more comments about this poll later.

Date: 13 Feb. Sample: 634.


That clears the deck for this year so far and I'll be back in another month or so with more polls that have accumulated in that time.  As the election approaches this will normally switch to a weekly article, and I will also start following seat betting (which is unreliable but interesting to keep an eye on) more closely.


Sunday - Ipsos Update: The 51-49 to Labor Ipsos released tonight deserves some immediate comment.  The main things to know about Ipsos are (i) that the pollster is volatile compared to the other significant pollsters, (ii) that the pollster's primary votes have a skew towards the Greens and, to a lesser degree, generic Others.  Ipsos' 2PPs don't skew to one side or the other, but they are bouncy.

Also, the Coalition was a bit lucky to get 49% off this set of primaries (Coalition 38 Labor 33 Green 13) - this would normally come out to 48.4 by last-election preferences, and I aggregated it at 48.6.  As a result my aggregate has so far dropped seven tenths of a point to 53.0 by last-election preferences, or 52.4 with the One Nation adjustment.  However, it's possible that the events of the last week have made a big difference and my aggregate is being conservative.  Or it's possible that Ipsos is being volatile again.  When there is a single poll that differs from the rest and there is an obvious event that may have caused it, an aggregate cannot say whether the poll is right or not - and in the case of Ipsos it's much more difficult to infer anything than it would be if we saw a three-point jump in Newspoll.

Looking at the record of Ipsos so far this term, there have been seven cases prior to this one of it being a point or more away from my aggregate even after including it.  In two of these Ipsos ran ahead of changes in other polls generally, while in four of these the movement implied by Ipsos never showed up in other polling, suggesting that Ipsos was simply being bouncy.  The cases where Ipsos was predictive include two of the three cases it stuck out on the Coalition's side (narrowings in April and September 2018) but no case where it stuck out on Labor's side.  Make of that what you will.  The seventh case was inconclusive: an apparent massive outlier in Labor's favour just before Prime Minister Turnbull was removed. (I suspect it was partly right and partly overreacting in that case.)

The Ipsos result was also somewhat contradicted by a Queensland-only federal Galaxy which had very strong numbers for Labor (52-48, a 6.1% improvement on the 2016 election, though Labor does have a history of underperforming in Queensland compared to lead-up polling.)


  1. Hi Kevin,

    Thanks for the IPSOS update.


  2. Warringah Getup Poll

    The poll was done about 2 weeks ago. Our house is in there somewhere.
    I calculate as follows:
    Prefs to
    Candidate Reported %ages Lib Steggall
    Lib 37.7 39.6% 39.6%
    Steggall 22.3 23.4% 23.4%
    ALP 15 15.8% 4% 12%
    Greens 9.6 10.1% 3% 8%
    Moylan Coombs 4.9 5.1% 1% 4%
    Other 5.7 6.0% 4% 2%
    Total 95.2 100% 51.3% 48.7%
    Undecided 4.8

    I have assumed high preference flows to Steggall from all but “Other”

    Abbott Wins

    The Getup figure for Abbott's primaries is quite believable, and he has been pulled deep into Tiger Territory, especially given the report that a poll commissioned by Moderate, very prominent, Liberals in the seat, put his Primaries at 25% (!). I think that report might have been a plant, but the source has a good reputation.

    The Steggall vote is very much lower than I would have expected, probably because Moylan- Coombs is getting some and because the poll was done only about 3 days after the Steggall launch. No Independent with fewer than 21% of the primaries has ever won a seat against the two majors. Steggall just gets past that. She needs a primary well into the high 20s to be competitive. Phelps won with 29%.

    It is worth noting here has been considerable argy-bargy from one of the Steggall support groups (Vote Tony Out) towards Moylan-Coombs. They are supposed to have buried the hatchet, but I am not sure where.