Thursday, June 16, 2016

Poll Roundup and Seat Betting Watch: National Poll Drought Edition

2PP: 50.3 to Coalition (same as end of last week, +0.5 in two weeks)
Coalition would probably win election if everyone voted now, probably with small majority (seat projection 78-67-5)

The 2016 federal election is underway! Prepoll voting has already started and we're just sixteen days away from the main game.  And yet, courtesy of a long weekend and perhaps media disinterest in splashing out on polls this time around, the evidence of what is going on in nationwide voting intentions is very limited indeed.  (We do have evidence of who is paying attention though.  Check out Morgan's very believable list of the most and least engaged electorates.)

For all that trendy stuff about how we're being swamped with polls, as I write we have just one national sample that is entirely less than one week old, and that will stop being true some time tonight.  Unless the overdue Morgan finally appears (which apparently it will sometime), we may be left with the infamously trend-averse Essential as the only poll with any data less than one week old until ReachTEL and Ipsos come along on Friday night.  The non-appearance of Newspoll this week makes this the first time since 1990 that the Newspoll brand has gone this late into a campaign before switching to weekly polling.

So if federal voting intention has changed significantly in the last week, we may well not even know. It doesn't seem like it has based on seat poll results and murmers from party insiders, but it's hard to tell which of these sources of knowledge is least reliable.

Polls ... you don't know what you've got til its gone! (I dislike that song, by the way.)

National polls ...  blink and you'll miss 'em.

In the last week we had a ReachTEL which came out at 50:50 by respondent preferences, but which was in fact a strong result for the Coalition based on the primaries (I get 51.8 by last election preferences, though you can knock a few tenths off for Xenophon if you want).  We also had an Essential two-week sample which was 51:49 to Labor.  I aggregated the ReachTEL at 51.8 for Coalition and the Essential at 49.1.  The heavy weighting on the ReachTEL caused by my special rules for the last three weeks at one stage pushed my aggregate out to 50.7 but this has since come back to 50.3.  Here's the smoothed tracking graph, not that there has been anything much to smooth lately:

The half a point movement means less than it otherwise would given how few polls are in the mix at the moment. This weekend's polls will give a much better idea.  For all that we saw with the bursting of the Turnbull bubble, it remains the case that at no point under his Prime Ministership so far has the Coalition quite lost its projected majority, in comparison with the Abbott prime ministership under which Labor would have won an election at any time from April 2014 on.

Labor's best hopes are either that something big damages the government in the final weeks (some are hoping the Parakeelia business might be it, though it really seems a bit obscure and too easy to muddy the waters by claiming Labor are equally guilty) or that the polls (not just the national polls, but also the seat polls, and the internal polls) are wrong. They would look more competitive if someone tracked polls using only respondent preferences, but even that would probably not quite have them in a winning position.

I strongly agree with Peter Brent that Labor's campaign is actually pitched a lot further left than it needs to be, and that really the only reason they are not further behind is that the government's campaign has had a lot of potholes too.  It's as if Labor actually believe all those terrible Australia Institute polls purporting to show "the hitherto repressed progressive voter inside every Australian".

It's notable that the BludgerTrack aggregate gives the Coalition only 76 seats off a 2PP of 50.5%, which is based on adverse state samples in Queensland and WA.  My seat model has a lighter weighting for state polling and hence a higher Coalition seat tally off a slightly lower 2PP.

(Edit: The missing Morgan is 51:49 to Labor, presumably by respondent preferences, but primaries have not yet been published. I've aggregated it at half-value, which knocked another 0.1 off the Coalition, pending the full results)


Essential had new leader ratings with Malcolm Turnbull flipping back to a net -2 (38-40), and Bill Shorten up to -6 (34-40), which I think is his best from them since March 2015.  The Turnbull-Shorten better PM gap closed again to 11 points (40-29) but this was in the context of a possibly ALP-friendly sample.  For summaries of the rest of Essential's polling see Adrian Beaumont or Poll Bludger (or just go to their website).

ReachTEL have published their ratings for the last two weeks now and there is not much to see there except that last week had a substantial dip for Malcolm Turnbull, which this week's sample more or less reversed.


As usual we see the sad spectacle of some parties preferencing someone who is either repulsive to their supporters, or just repulsive full-stop, on House of Reps How-to-vote cards.  However whether the Greens foolishly and insultingly preference a gay Liberal behind the Christian Democrats has no actual impact on the election since the CDP will not win any Reps seats.

The big preferencing stories are that:

* The Liberals are preferencing Labor ahead of the Greens everywhere
* The Greens are preferencing Labor ahead of the Liberals in 139 of 150 seats (Hotham is apparently one of the exceptions, I don't have a full list yet)
* In South Australia, Labor, Liberal and Nick Xenophon Team are issuing open or alternate tickets with regard to each other.  So voters will make their own decisions - I suspect it's still highly likely voters for both majors will lean heavily towards NXT over the other major.  Not getting Labor card preferences will blunt the Sharkie attack in Mayo somewhat, but possibly not very much.

In SA the Greens are preferencing Labor ahead of NXT, but relatively few Greens voters follow cards anyway.  Perhaps even fewer would do so if they knew that such a vote could in some seats help the Liberals to beat NXT and thereby elect a Liberal Government.  The reason for this is that to win Liberal seats NXT has to make the last two, and they won't do this if Greens preferences push Labor ahead of them.

As concerns the Liberals' decision to preference Labor, by leaving the decision as late in the campaign as possible the Liberals have caused Labor to devote effort to shoring up Labor-vs-Green seats.  However for the cost of letting Labor off that hook for the last three weeks, the Liberals get a decision that dovetails neatly with their (objectively unconvincing) claim that the election is a choice between a Liberal majority and a mess.

The matter of Senate preferencing is very interesting and I hope I can find time for a separate article about that, or else link to someone else's.

Finally, for free entertainment, check out this stuff from Warringah (a seat where Labor has never exceeded 41% two-party preferred).

Seat Polls

This week's ReachTEL seat poll featured the seat of Grey (Lib 13.5%) and had a stunning 32.7% NXT vote after allocating the undecided, based on which NXT would win 54:46.  This even though the poll's ratings for the local member Rowan Ramsey were strong (a +21.5 net rating).  This was a highly amusing result for me because my modelling attempt for NXT had Grey as, all else being equal, the last seat in SA where NXT might stake its claim.  The size of the electorate would probably make it hard for a new party to cover, and might also make it a very hard seat to poll, so it's possible the result is just plain wrong.  But if it is anywhere near correct then there is no seat in SA is completely safe from the third-party uprising.

A reason to wonder about the Grey poll sample is the implied 2PP Labor-vs-Liberal result.  It can be calculated from the 2PP figures supplied (and it is fantastic that these figures were included even though Labor aren't in play in Grey) that Labor would obtain a 10.3% 2PP swing.  This seems hard to believe, unless NXT is acting as some kind of "gateway drug" for ex-Liberal voters to preference Labor (a thing I doubt would ever happen on that scale).

The Grey poll had "creating jobs" at the top of the usual ReachTEL issue list, the first poll I have seen anywhere with that result.

The Fairfax presses had a bunch of ReachTELs with the Coalition leading in Deakin 52:48, Lindsay 54:46, Corangamite 51:49, Bonner 56:44, tied in Cowan and trailing in Dobell 49:51. The low average swing (about 1%) was taken as evidence of a gross disparity between the national polls and the situation in individual seats, but with the national swing at barely over 3%, most of the difference could actually be explained by four of these six seats being "double sophomore" seats.  That Labor is struggling to convert 50-50ish 2PPs into an even share of seats is not news; it is rather business as normal given the result of the last election.  (Alas, primary results for these polls seem not to have surfaced yet.)

That said these results still look a touch on the disappointing side for Labor, but not so you'd read much into it when seat polls were so systematically skewed in 2013.  This election there is much less evidence of discrepancies between seat polls and the local polls.

Sketchy details of a Mayo poll in the same batch had Jamie Briggs on only 37.6 to Rebekha Sharkie (NXT) 24.4% with Labor on 19.5 and the Greens 10.4.  On such figures Sharkie would need less than 70% of preferences and would most likely win, the second recent Mayo ReachTEL with this result.

Commissioned Polls

The following should be treated with scepticism since the forces commissioning them stand to benefit from the result that is released.  Had the polls been bad, they would just have been mothballed.

Rob Oakeshott made a very last-minute and much ridiculed announcement that he was going to have a go at roughing up the Nats in Cowper (13.2%).  Cowper has been National-held for almost a century except for a single Labor term in 1961-3 after the incumbent, ex-stopgap-PM Sir Earle Page, declined to retire despite being 81 years old and terminally ill.  There was much scepticism (mainly founded in views of Oakeshott's waffly speeches) towards a GetUp! ReachTEL that had (after undecided) the Nats Luke Hartsuyker on 42.6%, Oakeshott 25.6, Labor 14, Green 8.4, Christian Democrats 3.7.   Hartsuyker would only need 23% of preferences on those figures, but there would be real doubt that he would get them.

GetUp! also had a ReachTEL of Dawson (LNP, 7.6%, held by wannabe turtle-saviour George Christensen who is for some strange reason much disliked by the left) with a respondent-allocated 50-50 2PP. Poll Bludger calculates the 2PP as a much less remarkable 53:47 to Christensen by last-election preferences (which are quite rubbery given the 10% Palmer United vote in 2013, but still probably better than the

The Greens released a Lonergan poll - effectively one of their internals, since the Greens often use Lonergan for polling - which purported to show them trailing but competitive in their major target Liberal seat of Higgins (Lib, 9.9% vs ALP).  The timing of this release looked like an attempt to shore up morale in the wake of the adverse Liberal preferencing announcement. (Green strategy frequently considers supporter morale more important than actually getting a good result).  The 47-53 result would if real suggest a 7% "swing" from the Liberal-Labor 2PP, which seems on the high side, but it's plausible these kinds of inner-city targeted Greens campaigns might be more effective at peeling off Liberal primaries than ALP campaigning in a relatively safe 2PP seat.  Part of Higgins overlaps the Victorian state seat of Prahran, which the Greens knocked off with a "swing" of 5.4%.

Lonergan's limited public polling was terrible in 2013 (though not that bad in the NSW state election) and the Greens have a history of releasing polls that usually have their vote higher than they actually get.  I think it's possible the Greens will get about this close in this seat but actually knocking it off is much much harder than the Prahran win.  Meanwhile the loopiness of the dumb Australian twitter left was such that many of them declared Kelly O'Dwyer to be toast on the basis of a cherrypicked-for-release seat poll that still showed her winning.

One would expect the Greens are sitting on any number of seat polls that are much less flattering that they haven't released.  When it comes to seat polls released by parties, the gleaming object that you see in public view is just the tip of a large and very dirty iceberg.

Seat Betting

Once again, the point of this section is to test the theory that markets provide accurate information, and not to make claims that they do.

The most important thing in this week's betting is the headline rate - despite the scarcity of new national polling evidence, Labor has drifted to implied chances of about 18%.  This is actually in line with my historical poll-based model, which based on the slight aggregate shift in the last week is now giving Labor chances of only either 21% (simple version) or 16% (party-based version).  However my model will increase Labor's chances sharply should the new polls this weekend put it back above 50-50.

At present these are the seats considered on the various markets (now using Sportsbet, Crownbet and Luxbet) to be changing hands or else reasonably close.  If a seat is tied on one market I use the others.  In cases of conflict I call it a tie.  If a seat is close on any market (usually both parties $3 or closer) I class it as a close seat.

Loss (Coalition to Labor): Barton*, Paterson*, Petrie
Close Loss (Coalition to Labor): Solomon, Dobell*,  Page,  Eden-Monaro Macarthur,  Capricornia, Burt, Hasluck, Cowan
Loss (PUP to Coalition): Fairfax

ALP Close Holds: (none)
Coalition Close Holds: Banks,  Gilmore, Robertson, Macquarie, Brisbane, Forde, Hindmarsh, Mayo (vs NXT), Lyons, Braddon, Deakin, Dunkley, La Trobe, Swan

In the past week only Hindmarsh has flipped (and the ALP-Greens contest in Batman because the Liberals are preferencing Labor).  However the important change is that a large number of seats formerly on death-watch are drifting and there is no longer the massive imbalance in the number of narrowly expected wins.  As a result on my close-seat-adjusted measure the Coalition is in its best position on the markets since the polls closed up:

There is still a strange discordance between the Sportsbet 2PP market (which has the Coalition headed for 51.6%) and the remaining markets.  That 2PP would be likely to produce about 84 Coalition winners (disregarding any losses to the crossbench), but no other type of market backs this. A similar Luxbet market also has a strong 2PP, projecting about 51.2%. The William Hill banded market points to 78 Coalition seats, as does their exact seat market.  The Sportsbet exact winning margin market points to 79-80 Coalition seats.

Sportsbet now expects all of Jacqui Lambie ($1.25), Pauline Hanson ($1.45) and Derryn Hinch ($1.85) to be elected to the Senate.  Luxbet has Lambie at $1.12, with Lazarus at $2.50 and Bob Day at $2.70.  Ladbrokes however thinks Lazarus is more likely to win than Hanson.  In Hinch's case this has all been bolstered by a dream ballot draw and a flow of how-to-vote card preferences from the major parties. There's a need to look closely at where the major parties are likely to land in Victoria to see how likely (or not) this is to actually help him, but I haven't had time to try to model that yet, and probably won't have time for detailed modelling for Senate races other than Tasmania.

For the coming weekend's national polls I'm considering doing a rolling poll roundup from tomorrow night, and then a separate article with seat betting and my current seat model outputs later next week.  Everything is subject to time and the demands of the four contracts I am working on around all this!

Morgan SA and Greens Update (16 June):  Morgan has released aggregated figures for some seats where NXT is doing well, and aggregated NXT figures for all SA seats, based on polling from April until now.  The sample sizes for individual seats are probably tiny.  NXT averages 21.5%, 3.4% below its 2013 Senate result, and a level at which my NXT model was expecting them to get about three seats.  Based on the Morgan breakdowns, NXT would win Mayo easily, might win Barker depending on exclusion order, would probably miss Sturt as they would need both exclusion order and preference flow to go their way, and would lose Grey (contra to the headline) and Kingston because they wouldn't get out of third.  That said the Labor primary in Grey looks too high, so we may say Grey is in play.  Maybe! The pollster attributes the Grey result to the plight of the Whyalla Arrium steelworks, which is the kind of thing my last-election NXT model is blind to.

Generally Morgan has had higher third-party votes than all other pollsters and this may well mean the 21.5% is an overestimate.  Then again, since NXT apparently picked up steam in polling through May and the survey window goes back to April, maybe not by much.  Even if I take, say, a couple of points off the NXT vote in these samples to account for this, they still win Mayo, but are removed from contention in the others.

Morgan has also released details of another SA seat, Makin, in its State of the Nation document.  (LARGE PDF DOWNLOAD) On the results as given (ALP 30 Lib 25 NXT 22 Grn 12.5 Other 10.5) NXT would need to overtake the Liberals to win on Others preferences and also those of such Greens as did not follow the party card.  That's not implausible, but the same comments about Morgan possibly having NXT a touch high apply here too.  And, again, the individual seat sample sizes are probably more or less useless.

The same document contains a breakdown of NXT support by state, with the SA figure this time given as 20.5% and all others between 2% and 3.5% (in some of these states, like Tasmania, NXT aren't even running.) This is followed by some 2PP breakdowns which I'd assume are the overall 2PPs for those states, though they might be the preferences of the NXT supporters.

There is also some detail on Morgan's polling of the Green vote.  Top Greens seats are given as Batman 42% Melbourne 41.5% Wills 33 Grayndler 30 Richmond 29 Gellibrand 27.5 Sydney 27.  On such numbers the Greens might well win Batman even without Liberal preferences, but Morgan rampantly overestimates the Green vote so that should be ignored until it is reproduced by an accurate pollster.  Indeed the implication of this sample on that basis is actually that the Greens would not win a single seat, but for Melbourne that conclusion seems unlikely to be true.


  1. The Greens have also hit up supporters for extra money for signage in Higgins. I'd say that poll is realistic in so much as its the best case scenario so far, that is a narrow loss.

  2. Labor seems to be ahead in all betting markets in Eden-Monaro as far as I can see - might want to check that.

    Also, Luxbet and Ladbrokes now have seat betting too. Luxbet has been up for quite a while and the odds there seem to move a lot more frequently than the other markets.

  3. Thanks, I've changed that. Might have been a clerical error caused by Crownbet's useful but sometimes confusing habit of listing the incumbent first. I'll check out the others.

  4. Any comment about Michael Danby's effort today? Handing out "rogue" htv cards at pre-polls that preference the Libs ahead of the Greens (against ALP HQ instructions)?

  5. Well it's not illegal and they can hardly disendorse him over it this time, nor would they want to pick a major fight now that might just lose the seat. So I think they have to ignore it and hope he doesn't run third (which seems unlikely). Whether he will be a protected species after the election if he's returned remains to be seen.

    Labor candidate for Durack has given head office the finger (and some pretty short shrift) too: Preferencing Nats ahead of Libs.


The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.