Saturday, May 18, 2019

2019 Federal: Late Night Live Commentary

Coalition has won the election - small majority likely but perhaps a minority

Apparently won Coalition 74 Labor 64 Green 1 CA 1 KAP 1 IND 3

Seats Assumed Won By Coalition From Labor: Longman, Herbert, Bass, Braddon, Lindsay
Seats Assumed Won By Labor From Coalition: Gilmore, Corangamite, Dunkley
Seat Assumed Won By IND From Coalition: Warringah

Seats currently in doubt (projection = AEC projection)

Eden-Monaro (ALP) - Labor leading on projection
Boothby (Lib) - Liberal leading narrowly on projection
Chisholm (Lib) - Liberal leading narrowly on projection
Cowan (ALP) - Labor leading on projection 
Lilley (ALP) - Labor leading narrowly on projection
Macquarie (ALP) - way too close to call on current projection
Wentworth (IND) - Sharma (Lib) currently well ahead and expected to win

If all current leads hold Coalition will win 77 seats, Labor 67, KAP 1, CA 1, Greens 1, IND 3, with one unclear.


I've finished my live coverage for the Mercury (thanks very much to them for having me!) so I'm continuing live on here for a few hours and will unroll postcount threads tomorrow - not as many of them as I expected.

What we've seen is a massive polling failure and a result that (whatever exactly it is) was not predicted by anybody much to my knowledge, except for the sort of people who predicted Donald Trump.  This is looking a lot like a mirror image of the expected result - the Coalition 2PP is tracking for about 51.6% 2PP and about the same number of seats as Labor were expected to get.

But this isn't the same thing as Trump (where the national polls were more or less right and the serious errors were localised), this is a national total polling failure more similar to Brexit or to recent UK national elections.  This after national polls in Australia have been so reliable for so long. Betting markets failed as well - initially expecting Labor to win by more than Labor's leads at the time showed, and only converging to what the polls were saying in the end.   Internal polls failed, both quantitatively and qualitatively.  Psephologists failed too, beyond that we warned that the polls were ridiculously herded and therefore couldn't be trusted.

Comments will follow below the dotted line, scrolling to the top:


1:40 Senate: A quick look at where the Senate races are going, though there will be much more detail on those that remain unclear in coming days:

NSW looks like 3 Coalition 2 Labor 1 Green.  We will have to wait to see what Jim Molan has by way of BTLs but the size of the Coalition vote as that unrolls but he would need a huge lead to be competitive.

Victoria is currently 2 Coalition 2 Labor 1 Green and the last is in doubt.  Presently the Coalition leads Labor and a gaggle of micro-parties including One Nation and Derryn Hinch, but nobody has very much there.  On current numbers the Coalition should lift on postals and should also benefit from the Greens soaking up left preferences, so they should win, but it isn't yet clear.  If they drop back it's possible Derryn Hinch could still be in the mix, but it seems very difficult for him.

South Australia looks like 3 Coalition 2 Labor 1 Green.

Western Australia looks like 3 Coalition 2 Labor 1 Green.

Tasmania looks - very much - like 2 Coalition 2 Labor 1 Green and Jacqui Lambie.

Queensland ... this one is messy!  Currently we have LNP 2.555 quotas, Labor 1.649, Greens 0.818, One Nation 0.692.  So 2 LNP 1 Labor and very probably 1 Green, with LNP, Labor and One Nation left to fight for the last two seats.  The LNP is behind to start with, but may pick up on postals and UAP preferences.  So Queensland could end up with 3 Coalition and One Nation, but perhaps Labor will lift.

ACT 1 Labor 1 Coalition  (though the Liberals are somewhat short of quota, they are far enough ahead of the Greens to easily not be caught)

NT 1 Labor 1 Coalition

So in total: 17 Coalition, 13 Labor, probably 6 Green, Lambie with three in doubt, of which either the LNP or One Nation must win one (if not more).  I'll look at what that does for the Senate balance soon.

Adding in those already there the Coalition will have at least 33 seats. The non-Green crossbench will contain 2 Centre Alliance, 1 One Nation, Cory Bernardi and Jacqui Lambie.  The worst case for the Coalition is failing in Victoria and Queensland in which case they'd need all six out of either six or seven non-Green crossbench votes to pass bills.  The best case is winning in both in which case they would need four out of five or six non-Green crossbenchers.

1:20 Wentworth: I did have some comments up here about the Rose Bay PPVC being likely to knock off Kerryn Phelps' lead, but it turned out that it was much smaller than in the by-election, so currently Dave Sharma's lead is much narrower than I expected.  He's still doing better for the same booths than in the by-election, but by-elections lack absents and out-of-electorate prepolls, which will help Phelps.  So this may actually remain very close.

12:20 Hunter: There is a rogue situation in Hunter with One Nation challenging the Nationals for second place.  Labor has won the 2PP.  The Nationals have preferenced One Nation so if One Nation can actually come second they could win.  However I suspect that with postals and UAP preferences the Nationals will hold, but the seat will need careful scrutiny.  How on earth did One Nation get 20% in Hunter?  (1:50: The Nationals have improved so I have taken Hunter off deathwatch for now.)


  1. Macquarie is now showing as ALP ahead (by0.2) as of 19 minutes ago.

  2. Antony Green, in the course of the ABC coverage, put the undoubted failure of polling down to the difficulty of obtaining a reliable sample with a mobile population who are no longer tethered to land-lines. That seems an inadequate response given that it's not one poll that's now suspect; the similarity in results means that *every* poll taken in the course of the Turnbull/Morrison government may have been wrong by the same margin, which he estimated at 3% nationally (but was 8-12% in Queensland). In the process, though, he avoided answering the question, "Why on earth would anyone EVER trust another opinion poll?"

    I think that it's over to you and other such analysts, who need to provide an answer to that question. I am reminded of when no-one predicted the Labor win in Queensland a few years ago - and raise an eyebrow at the thought that the state most grievously misread this time was Queensland, again.

    For a starter, is it enough to simply double error margins? Or sample sizes? Or rejig the correlation between sample sizes and error rates to a looser value? Those would all seem to be relevant questions if Antony's surmise as to the cause of the error is correct. But what else might be the cause, and how do we correct the problem?

    1. I get about 3% nationally too as the amount of error. Maybe a bit more.

      It's important to keep in mind that until now our national polls had an outstanding record. But it will take them a while to recover trust through successful results. Among the process of winning back that trust will need to be admitting that they have been too opaque about how they do polling (Galaxy especially) and that they need to open up their processes to public examination. Galaxy/Newspoll in particular need to explain why their polls are under-dispersed compared to random variation. The others need to explain whether they were herding to Galaxy/Newspoll and if so why.

      And yes it is indeed about a range of different methods since Galaxy (landline/mobile), Newspoll operated by Galaxy (landline/online), Ipsos (landline/mobile), Essential (online) and Morgan (face to face) have all failed.

      The Queensland 2015 state fail was different. Pollsters sampled the primary votes in Queensland 2015 very accurately. The mistake was in the preference flow modelling. But there is a very long pattern that Labor polls well in Queensland in federal election leadups and then falls over on election day.

  3. The "polling error" also occurred in the exit poll (Galaxy/Channel9) which had Labor leading 52-48.

    This is the same result and error as all of the other polls, but can't be blamed on phone lines etc....

    So either unlucky or lazy sampling, or voters aren't giving their true opinions.

  4. Hi Kevin,

    I really enjoy your updates, thanks for the hard work!

    Do you think embarrassment to admit over the phone that they will vote for a party like One Nation may partly explain the disconnect between polling and results? They primary swing to the Coalition seems smaller than the bleed to the fringe minor parties.

    1. No because the polling failure occurred across all modes including online and robopolling where there is no reason to be embarrassed.

    2. In the UK people keep hypothesising the "shy Tory" voter who is embarrassed to admit that he/she votes Tory. I doubt that embarrassment holds back ON voters - they're all too willing to shout at canvassers for other parties outside polling places. But perhaps they (and some Liberal voters) don't respond to pollsters, not because they're embarrassed but because they're essentially antisocial and think "how I vote is nobody's effing business". More like Surly Tories or Snooty Tories than shy ones, I expect.

  5. I notice Bass has ben presumed won the the Liberals. There's only 320 votes in it. Is there some doubt over this seat?

    1. Very minor doubt. Hospital votes should bring it up to about 500, then there's the postcount which in 2016 favoured Liberals greatly and in 2013 did nothing. It is very rare for postcounts to favour Labor by enough to save it here.

  6. OK, so the pre-election polls were all wrong and, since they were all near-identical, at least part of that has to be explained by herding - by being more frightened that an outlier would be wrong than excited by the prospect that it might be the only one that was right. Verrry risk-aversive behaviour! But the only (I think) exit poll that was reported also joined the herd, saying Labor was ahead by 52-48! Surely there'd be no inhibitions about just giving a straight report of the outcome there? Any suggestion as to what could be happening there Kevin?

    1. I suspect a combination of problems with getting a representative sample (by any method), oversampling of politically engaged voters, herding and possibly unpublished data-smoothing practices.

  7. Will be interesting to see how the spin on the Newspoll website changes after this. Comments like this:

    "Because it’s Newspoll, you can be sure it’s right.
    Our faultless record in election polling is proof."

    1. That comment is on the website of the old Newspoll which ceased running the poll in 2015. Since then Newspoll has been administered by Galaxy.

  8. Hi Kevin,

    For the ACT you say the Liberals are "far enough ahead of the Greens to easily not be caught". However, (on the ABC site) they are only currently 0.31 quotas ahead while the Labor surplus + Pesec combine for 0.33 quota. You would expect these to flow strongly to the Greens. While the Liberals are in front, what is it that makes you so confident in their lead?

    An interesting scenario would be Labor winning a 3rd seat in Victoria plus a 2nd in Queensland, and also the Greens winning off preferences in the ACT - that would give the two parties 38 seats combined.

    While its definitely not the most likely path, there is a chance for Labor + the Greens to somehow end up with a blocking majority in the Senate from all this.

    1. Based on the current lead the issue is that the Labor preferences won't nearly all flow to the Greens (last time it was 74-20 with the rest exhausted) and nor I'd expect would Pesec's. At the best, whatever they have might benefit the Greens at say 0.6 votes/vote between them. The Liberals will also benefit from UAP, FACN and postals, though the Greens should benefit from SA.

      The count is very incomplete so the only chance is if the primaries change a lot in what is left. Haven't looked at where the votes come from and what is still outstanding.

    2. There are only 7 groups on the ballot paper and one of them is Anning's candidate. I'm guessing the exhaust rate might be lower.

    3. For sure. However some of the exhaust is caused by voters not voting to 6, so there will still be some.