Wednesday, May 1, 2019

Poll Roundup: Make Newspoll Preferences Great Again!

2PP Aggregate: 51.9% to Labor (last election preferences) (-0.5 since last week)
51.6% with One Nation adjustment
Labor is still winning, but at present polls imply only a modest majority (c. 80 seats) 
Betting markets don't think it's that close

I didn't put out a Poll Roundup last week because there were no major polls.  This week there is a lot more to discuss, starting with United Australia.

UAP Surge - Is It Real?  

After languishing at around 1% in those polls that included them until not long ago, Clive Palmer's United Australia Party has made major inroads in polling in the last month, helped by problems for One Nation, which is not running in every seat anyway.  At the current level, team yellow (unless you're in Tasmania where there are two team yellows) isn't threatening to win Reps seats, but it looks very competitive for the Queensland Senate at least.  That said, media are overplaying the value of the very useful Coalition preferences there, since the Coalition might not get that much over two quotas given the added competition from UAP, and only 30% or so of its voters will follow the card anyway.  The surge is being linked to Newspoll switching to including UAP as a distinct option in its results, which is being interpreted as the party being added to the readout (something which has been thought to overestimate minor party votes).

I don't fully agree with the readout-based argument that UAP's vote is being overstated. Firstly the main party that was historically understated in polling until it was added to the Newspoll readout and overstated thereafter was the Greens.  But the overpolling of the Greens was probably caused by social desirability bias in live polling - these days, online polls and the better robopolls don't seem to get their vote too high anymore.  Secondly, from time to time (though not that recently) I've seen screenshots of the online "readout" of Newspoll, and it often includes a brace of parties that are included in the list for the voters to choose from, but not separated out in the published poll.  Thirdly, polls have more or less accurately captured the One Nation vote at recent elections, and surely similar issues should apply there.

Where might it go from here? In 2013 PUP really only lifted off from c. 2% to its final 5% in the final weeks, so if UAP could repeat that they could be on for a very large vote, which could spell peril for Labor on preferences if voters follow its how to vote card at even modest rates.  This seems very plausible given the amount of money UAP is spending on wall-to-wall advertising. But in Queensland 2015 the party was polling well into double digits six months out and finished up with only 5%, including a modest drop-off in the last few weeks.  It also dropped back from about 7% to finish with 5% at the 2014 Tasmanian state election, where it was singly beset with candidate problems compared to the established parties.  At this election the UAP have again had no shortage of candidate issues (especially with Section 44) but they're competing in this regard with One Nation's smut parade, the Coalition's many disendorsements and a ALP chap who wants to talk about Jews lizards.

Newspoll Preferences - The Sequel

This week's Newspoll has come out with the Coalition leading 51-49 off primaries of Coalition 38 Labor 37 Greens 9 United Australia 5 One Nation 4 others 7.  However, even if we use Palmer United preferences from 2013 for UAP (instead of generic Others preferences from 2016), this would come out to 52.2 to Labor by last-election preferences.  And the well known different treatment of One Nation doesn't explain the gap either, because they are no longer getting that many votes.  It turns out that Galaxy have decided to start distributing UAP preferences 60-40 to the Coalition (the 2013 split was 53.7%), widening the gap between their preferences and last-election preferences just when the collapse of One Nation's vote was closing it.

This decision may well prove correct, but it represents a further shift away from Newspoll's long history as a last-election-preferences pollster.  In One Nation's case the (well justified) shift was based on the results of subsequent state elections and then validated in the Longman by-election.  But in the case of the UAP there is no direct data basis for it; nobody has ever seen this party's preferences flow this way.  David Briggs of Galaxy has defended the decision, citing:

* That there is a preference deal between the LNP and UAP, which there wasn't in 2013.
* That UAP voters in the current sample (albeit probably only around 100 UAP voters) picked Scott Morrison over Bill Shorten as most trusted, by a 53-13 margin.

These are good reasons to suspect a shift.  Another is that the immigration and foreign relations mix has shifted from the pro-asylum-seeker positions in PUP's 2013 campaign to attacking alleged Chinese influence in this one.   Also there are signs that the lavishly resourced UAP campaign will make a bigger effort on how-to-vote cards.  But there are some counter-arguments:

* No, there wasn't a deal in 2013, but PUP preferenced the Coalition in all 150 electorates anyway.
* UAP may be taking Coalition-friendly voters from the Others pile, so the remaining Others may on balance be slightly Labor-leaning.
* Newspoll has implemented corrections for UAP and One Nation, both to the Coalition's benefit and for easily defended reasons, but not for any possible factor that could push preferences the other way.  Greens preferences probably couldn't get much stronger, but even a few percent would make a few tenths of a point difference, while the preferences of independents might be more Labor-leaning this time given the number of high-profile indies running on climate change issues.  (On that score, I note that voters who put Kerryn Phelps ahead of both majors in the Wentworth by-election split 58.2% to Labor on a 2PP basis, despite Phelps preferencing the Liberals.)

I could put out a UAP-and-One Nation adjusted figure in my aggregate but I've decided not to, because while Galaxy's judgement may be correct, arbitrarily assigning preferences isn't science.  After considering the primaries (and revising my formulae to deal with what Newspoll is doing with UAP) I aggregated this Newspoll at 52.0 to Labor by last-election preferences.  Last-election preferences seem probably wrong at this election (so I am using the One Nation adjusted figure for seat modelling still), but they have a formidable track record (including at other elections where they looked to be probably wrong). The last time the Newspoll brand deviated from them (2004) it came a cropper.

Other recent national polls

Other recent national polls have been a Galaxy taken in the middle of last week with a 52-48 2PP to Labor (which I aggregated at 52.7 by last-election preferences), two Morgan face-to-faces both 51-49 (aggregated at 51.3 and 51.2) and this week's Essential also at 51-49 (aggregated at 51.4).  The Essential poll's primaries would normally have come out at 51.9 to Labor, but Essential is still using pure last-election preferences, so probably the primaries were a shade better for the Coalition than the rounded primaries appear, and they just scraped over the 48.5 bar on 2PP to get rounded to 49.  I aggregated this poll at 51.4, but with a breakdown for UAP I probably would have taken a tenth or two off that.

So despite this week's slew of 51s, even if one assumes a strong shift from previous-election preferences, Labor's current position is still probably better than 51. Here's the smoothed tracking graph by last-election preferences:

By last-election preferences I have it at 51.9 to Labor; if I included Newspoll-style adjustments for both ON and UAP it would probably come out to 51.2.

Is This Like Clinton vs Trump?

We continue to see narrowing in the 2PP as the election approaches; do not adjust your set, elections frequently end up closer than leadup polling suggests.  That said, sometimes when federal elections have been close (eg 1990), it has been a surprise with polls continuing to predict a not-so-close result right up to election day.

Using my One Nation adjusted figure (48.4% for Coalition) my historic regression still gives the Coalition only a 17% chance of getting to my estimated win target for them (50.3) though really that might mean, say, a 25% chance after factoring in the chance that they might get lucky and win off less than that.  The projection of a final 2PP somewhere in the 48s (now up to 48.9) has been pretty stable over time.  Even at this stage though, the relationship between past polling and final results is pretty messy.  At three of the past 11 elections, governments have made gains from here that if repeated this time would give the Coalition a better than even shot of victory.  However one of those was making up ground from a way-behind position, which is easier than making ground across the 50-50 line.

The last three governments in a worse position than Morrison's is in now lost heavily.  However two of the governments in just a very slightly better position (Keating 1993 and Howard 1998) won.  Howard had a large advantage in personal votes from the previous election, which Morrison doesn't have.  Keating had the advantage of having Fightback! to campaign against.  Labor is taking risks at this election, but it doesn't seem to be on that level.

The recovery in government polling has created an unusual situation.  Changes of federal government are normally decisive, so a narrow loss (as implied by current polls) would be unusual.  Even looking at old Morgans, the last six governments to lose were way behind at this point, except for Labor in 2013.  In 2013 Labor were about as far behind as the Coalition is now, but were going backwards rapidly as the bounce for bringing back Kevin Rudd disappeared and the Labor campaign unravelled.  I am actually surprised that a government that has been so chaotic is polling competitively at all, especially given discontent over slow wage growth, but perhaps they have just stumbled on the sort of leader they should have had all along.  Both Abbott and Turnbull failed the test of being able to bridge the party's right-wing and moderate factions - Abbott in the eyes of the public, and Turnbull in the eyes of his own party.

The current situation has echoes of Trump vs Clinton, where late in the game Nate Silver warned that Trump was only a normal polling error from victory, and people who thought the election was in the bag for Clinton ridiculed him for giving Trump too much chance.  As it happened Trump didn't even need the normal error - a smaller error nationwide plus errors in a few key states did it for him.  Scott Morrison isn't much further off winning than that (of course just because one in four chances happen sometimes, doesn't mean they happen all the time).  However, a normal error in the projection the other way would see a decisive Labor win, on a similar scale to 2007 at least.


We have just seen the first leadership debate of the campaign.  Despite dreadful presentation values, I thought the debate was OK by the low standards of these things, and at least there was a reasonable amount of interaction and spontaneity.  The debate was judged by a hopelessly small panel (48) of "undecided voters", little better than a focus group, especially as 11 of them were still undecided.  As a result Bill Shorten's victory margin of 25-12 wasn't even statistically significant.  These debates have had no real predictive value at federal elections and generally change nothing unless somebody blunders.  I was going to write some more detailed comments re the debate but I think this piece is going to be long enough already, so maybe discuss it in comments if interested.

Newspoll this week saw Morrison's netsat down insignificantly to -1 while Shorten's rose slightly to -12.  Although Shorten has had -13 three times this term, -12 is his best since he polled the same thing in May 2016.  He last polled higher in May 2015, when his ratings had just started to slide and just before they were severely damaged by ABC's release of The Killing Season.

For once, Better Prime Minister was interesting.  Despite the closing of the 2PP gap, Morrison's lead over Shorten closed to eight points, 43-35.  That's about the sort of lead a PM should have given the current 2PP, whereas normally Morrison (and Turnbull before him) has led Shorten by more.  So that's some evidence that Shorten's iffy approval ratings are no longer washing through to Better PM.  That doesn't mean much predictively, but if there's something indefinably bad about Shorten that is putting voters off beyond what we know about him through the other ratings, Better PM isn't capturing it.

Essential also found Shorten closing the gap, to 40-31, as close as it has had it since Morrison became PM.  Morrison scored well on Essential's quirky leader-relateability questions.

Seat Polls

Seat polling is unreliable, and one seat poll lately was especially so.

Since the last roundup there was a brace of Newspolls where the United Australia Party, in all four cases (Herbert, Deakin, Pearce, Lindsay) was beating the PUP 2013 vote.  However, three of these seat polls did have implausibly low Others votes of only 2%, suggesting that UAP (and probably every other party too) was being overpolled in these seats.  The national polling is more likely to be accurate.  The polls (51% 2PP for Coalition in Deakin, 50% in Pearce and Herbert, 49% in Lindsay) were consistent with a pattern of swings against the Coalition in its own marginals, but close contests in a couple of troublesome Labor marginals.  Ludicrously, Simon Benson in the Australian claimed Michael Sukkar was "almost assured" of re-election despite this flimsy lead in a seat poll taken four weeks out.  He must have not checked what happened with seat polls in 2016.

There has also been a Galaxy with the Coalition ahead 53-47 in Sturt, and various reports of internal polling (those with numbers including Labor's internal polling having them ahead 55-45 in Dunkley and 54-46 in Lyons).  And there's been an Australian Forest Products Association ReachTEL with Labor behind 49-51 in Braddon, immediately following which Labor moved to withdraw support for further forest reservations or a Tarkine National Park in Tasmania.

And there has been a strange saga of a fake "ReachTEL" involving independent challenger Louise Stewart in Curtin (see William Bowe here for a summary).  I haven't seen the full figures of the fake poll but there were red flags in the screenshot versions sighted that I didn't pick up on immediately, including that female voters were supposedly several points more likely to vote Liberal without the (female) Liberal candidate even being named, and that the gender breakdowns didn't average to the total assuming correct scaling.  Big credit to Nathan Hondros (WA Today) for making suspicious noises about this poll more or less from the start and a very good job has been done by many WA journos pursuing a murky and messy story that has at the least severely embarrassed the candidate.

Seat Betting

Betting isn't a reliably predictive indicator, and I was thinking of not paying it much attention this cycle after its poor performance in 2013 and indifferent performance in 2016, but what it's doing this election is interesting.  Basically, markets do not place much weight on the narrowing in the polls this week and are still expecting Labor to win comfortably.  This applies to seat total markets (which currently have Labor on for a tally in the mid-80s) and also to totals of individual seats.  If we can read an argument into this it's (i) that this government has polled so badly through its term that any case of it polling well is treated with extra caution (ii) at recent elections polls have had it closer than it was, just about fence-sitting in SA and NSW and underestimating the margin in WA and severely so in Victoria.

Here are the current seat betting expectations, as of late last night (see here for methods):

Expected ALP classic-seat gains (not close): Gilmore, Robertson,  Chisholm, Reid, Dunkley

Expected ALP classic-seat gains (close): Petrie, Dickson, Forde, Hasluck, Boothby, Dawson, Bonner, La Trobe, Pearce, Swan, Leichhardt, Deakin, Stirling, Flynn, Corangamite

Expected Coalition classic-seat holds (close): Casey, Sturt,  Canning, Bowman, Aston,  Hinkler, Hughes, Flinders, Ryan, Menzies, Grey, Brisbane, Capricornia, Banks, Page

Expected Coalition hold (three cornered) (close): Higgins

Expected Labor loss to Coalition (close): Bass, Herbert
Expected Labor hold vs Coalition (close): Lindsay, Braddon, Solomon

Expected Coalition loss to IND (close): Cowper, Farrer
Expected Coalition gain from IND (not close): Wentworth
Split market Coalition vs IND: Indi
Expected Coalition hold against IND (close): Warringah
Expected Coalition hold against Shooters (close): Calare
Expected Labor hold vs Green (close): Macnamara

Here's the colour-coded seat favourites tracker:

Key to colours (more may be added):

Red - Labor favourite in all markets
Orange - Labor favourite in some markets, tied in others
Dark blue - Coalition favourite in all markets
Light blue (none yet) - Coalition favourite in some markets, tied in others
Grey - all markets tied or different favourites in different markets
Purple - IND favourite in all markets
Pink - IND favourite in some markets, tied in others

The other weird thing going on here is that markets seem to think they can pick Coalition seats on 6% that will fall and also a Labor seat on 5.3% that will go.  If markets were able to buck the pendulum to this extent it would be quite remarkable. 

Update: I have now found a further bookmaker offering seat odds, BetEasy.  The most notable difference is Labor is favourite in Bass. (also a note that Braddon flipped on Sportsbet today.)


  1. Liberal Party Ad on high rotation up North last few days saying "Shorten wants to spend 25 million on an AFL Team in Hobart" and finishes with "Vote against spending money on Hobart AFL" There is nothing of course in the Labor policy about an AFL Team being placed in Hobart Good old parochialism card is being played to lure Bass voters.

  2. There is also the "issue" of HTV cards. Might only be my observations at various polling booths but not a lot seem to be taken from any candidates. Could this be due to the Tasmanian elections banning them so they aren't really a feature in our political sphere and therefore making up our own minds?

    1. Yes. Tasmanian voters have lower follow rates than the rest of the country.

  3. Hi Kevin,

    thanks for the analysis.

    Even with this "tightening" isn't the Coalition primary vote of 38% still extremely low? (42% primary in 2016 and lost 14 seats).

    Is this the UAP factor at play? To think the UAP will pick up 5% of the vote nationally is ambitious isn't it?

    1. Regarding the UAP we know that they've polled 5% before (in 2013). Compared to then, positives are the Coalition's greater state of disarray and the much bigger campaign spend. Negatives are their chaos in the 2013-6 term, Queensland Nickel and competition from One Nation. So there are arguments for them doing better and arguments for them doing worse; I'm not sure which of these is stronger. It is notable that some other polls (Essential and Morgan, though Morgan is worth very little) don't have them as high.

      Regarding primaries it is all about the 2PP. People used to say Labor needed 40% to win; they won in Queensland with just over 35%. In 2016 they had no strong preference flow sources that were polling anything useful. If this time UAP and PHON poll well and their preferences flow to the Coalition, that can make up for losing a point or two on primaries from last time. But probably not four.

    2. Hi Kevin,

      understood. Thanks for the clarification. 2013 was a very different ball game. Palmer United offered an alternative then. It quickly unravelled after that.

      Maybe the UAP will pick up 5% at the detriment of the PHON vote. South of the NSW border I don't expect that to be the case, but who knows.

  4. As I read the article, especially the section on the betting markets, a thought occurred that you might find relevant - that there have been so many opinion polls with labor winning in a canter that the markets (and perhaps the public) may be assuming that the errors which aggregate to constitute the margin of error would bias strongly toward labor, and hence that labor will end up winning more close contests than it loses on the night. Even if that's not actually the expectation by the bookmakers, it may be the public perception, which the bookies have to price into their odds in order to get enough money betting on the coalition to cover the bets on labor and vice-versa.

    Journalists have a rule - don't be part of the story if you can help it. But pollsters and bookies don't (and can't) abide by that rule since their activities help shape the narrative that they are (theoretically) measuring and quantifying. Smart bookies factor that into their calculations even if they do so by rule of thumb.

    This possibility suggests that there may be other similar factors that, if properly accounted for, might make betting odds track poll results far more closely than the raw data and past experience would suggest!