Wednesday, June 1, 2016

Poll Roundup And Seat Betting Watch: Preferred Pet-Sitter Edition

2PP Aggregate: 50.1 to Labor (-0.1 since end of last week)
Coalition would probably win with a slight majority if election were "held now" (projection 78-68-4)
No significant move in voting intention detected in past seven weeks

In this edition:
Seventh straight week of no 2PP action
Old Newspoll's Queensland problem
Is Nick Xenophon Channeling Terry Pratchett?
Bill Shorten, Preferred Pet-Sitter
Corangamite seat poll
Punters Vs Polls - A False Dichotomy (For Now)

(Note: these roundups get pretty long, so feel free to just read the bits that interest you! I hope to have a detailed preview for the five Tasmanian seats up sometime this coming weekend)


Welcome back for another rundown of another week of polling in which, in terms of the national 2PP headline, nothing happened! There wasn't really too much to cause it to happen yet either.  There was a debate on free-to-air TV, but it was still watched by hardly anyone, and almost everyone who did watch it complained about it afterwards.

Candidate malfunctions continued at the higher than normal rate that has distinguished this campaign, with Chris Jermyn (Liberal, McEwen) wowing the crowd by trying to ambush Bill Shorten only to be ambushed himself.  The Coalition made a gloriously overcooked attack on Labor's spending commitments, parts of which were immediately dismantled, and then with no sense of shame replied that by refuting half of their original claims Labor was admitting to being half guilty.

Two polls this week received attention because they appeared to show some movement, but as it happened these were in opposite directions and were both less of a big deal than they seemed.  ReachTEL late last week jumped from 50-50 to 52-48 to Labor, but this mostly reflected in its use of respondent-allocated preferences (which can make the 2PP result more volatile).  On primaries the Coalition had lost just 0.9 of a point, with Nick Xenophon Team and not Labor the beneficiaries.  Essential this week switched from 51-49 to Labor to 51-49 to Coalition, quite a surprise given its ALP-leaning form under Malcolm Turnbull's reign so far, but the shift was probably not much more than a point before rounding.  The third poll out in the last week was Morgan, which returned a 51.5 to Labor result by last-election preferences but only 51-49 respondent-allocated, bucking a recent trend for respondent preferences to favour Labor.  Morgan again had an extremely high non-major-party vote of 30%, of which more later.

Here's the smoothed tracking graph:

(Other aggregates as they update: BludgerTrack 50.2 to Labor, Phantom Trend 50.1 but yet to include Essential, Luke Mansillo 50.55 pre-Essential, Mark the Ballot 50.7 but see this post).

For the first time since 1996 there was no regular Newspoll on the fifth weekend prior to voting, but there was the pollster's aggregated state breakdown from the last two months.  After this Peter Brent noticed something very interesting - the old Newspoll had an extremely strong habit of polling higher 2PP readings for Labor in Queensland specifically than Labor polled on the day.  This had happened in every one of 20 recordings that Peter tabulated, with an average discrepancy of four points! Labor would be dearly hoping that the Galaxy-run Newspoll has ended this curse, because if Newspoll's 51:49 to Coalition in Queensland is really 55:45, then Labor will be scratching for crumbs in the sunshine state and might even not get both of Petrie and Capricornia. (NB: After I wrote this, @Flibbety on Twitter found one Newspoll sample missing from the Brent tables - in 2010 the final Newspoll underestimated Labor in Queensland by 2.9%, so that's 20 vs 1 now.)

The other big takeaway from the Newspoll breakdowns is that things are looking bad for the government out west, with the 9% swing in WA not much smaller than the damage in other sampling.   The BludgerTrack breakdowns tell the story - the Barnett government is dragging on the federal polling, and in WA alone the Coalition is back where it was in the last days of Tony Abbott.  This puts about six seats in play and on current polling Labor could well take about four.

The Xenophon Team: Magic Or Illusion?

There's also the small matter of Others polling 34% in South Australia in this Newspoll, of which Nick Xenophon Team is polling 22%.  Xenophon himself is quoted as saying there is a "million-to-one chance" of a Lower House win for his team.  While this might imply that his noted aversion to industrial gambling comes from being incredibly bad at it, more likely he's a closet Terry Pratchett fan: as any political magician knows, one in a million chances happen nine times out of ten.

This wasn't even the rosiest projection of the NXT's fortunes doing the rounds: Morgan had them on 26.5% in South Australia (compared to 20.5 and 19.5 in previous weeks). Both Essential and ReachTEL have them up nationally, although still at lower levels than Morgan.  I believe Morgan are overestimating the NXT vote as part of a systematic tendency to overstate the votes for third parties, as Morgan has had third parties combined on an average of about 29 in recent polls, while other pollsters have third party votes around 22-24 at the moment.  This overestimate of the third party vote in my view is a signal that a pollster is oversampling either politically engaged voters or fashionable views.  Morgan's SMS method is prone to the first and its face-to-face method prone to the second, just as it had a far bigger signal for the Turnbull bounce at first than anyone else.

But even toning down Morgan's figures accordingly, the Newspoll sample is still hard to ignore and what limited polling is available to me across all pollsters is consistent with an NXT vote of around 19% in SA.  That's more or less what I take to be the seat-winning level and while caution is always warranted with minor-party polling surges, if similar polling continues I will soon give them a seat in my projection.  (Re the Greens in Batman, I'm still waiting on the preferencing decisions there.)


A few new leadership stats this week, with ReachTEL just about sharing Newspoll's view now that Turnbull and Shorten are equally popular (net -7.4 vs net -7.8; the latter is Shorten's best since December 2013 and his second best ever).  Turnbull's lead on the forced-choice better PM figures narrowed to 54.9:45.1.  Essential had Turnbull back in positive territory with a net rating of +2 (41-39) to Shorten's -10 (34-44) and had a two-point narrowing of the "beauty contest" gap to 40-27 in Turnbull's favour.

But the most unusual part of Essential's leadership polling was a series of quirky questions about which leader they would trust the most in everyday-ish situations (see p 8 of the PDF).  Voters would rather have the merchant banker than the unionist negotiate their pay rise (36-27), would most trust Malcolm Turnbull to advise their children about the future (34-17) but think Bill Shorten would be most likely to lend a hand if a car was stranded (21-37) and would slightly prefer him to care for their pet (20-23).  Some of these seemingly flippant questions do shine some light on perceptions of the leaders' personalities that don't come through in the headline questions - mostly in Turnbull's favour.

This was all an amusing sequel to a rant by Mark Textor about a supposed excess of polling in Australia.  The core claim is that electoral coverage is being swamped by discussion of polls at the expense of discussion of issues, and voters can't escape from this even if they want to, because the stuff is everywhere.  (The swamping aspect is greatly exaggerated by the inclusion of polls ignored by the media and seat polls only reported in batches in the figures, but I digress).  Textor wants more polling that explores the reasons behind voter choice rather than just the raw numbers and the nonsense that is made of tiny random movements in them.  Somehow I am not sure this Essential effort was exactly what he had in mind.

Seat Poll

ReachTEL released a seat poll of Corangamite.  As with their seat poll of Lyons (see Tasmanian ReachTEL coverage) it had the ALP primary suspiciously low, but also with a respondent-allocated flow of preferences to Labor (78%) that seems on the high side, though Labor did get 71% of preferences in that seat last time.   This all came out to a 2PP of 54:46 to Coalition (no swing).  This continues a trend of mildly disappointing results for Labor in most of the seat polls from ReachTEL so far.  However, taking into account sophomore effect and the fact that the swing in WA is blowing out the national swing, the headline figure here doesn't seem greatly surprising.  The poll finds a good positive rating (net +19.6) for local member Sarah Henderson.  Her Labor opponent Libby Coker has the usual work to do on the name recognition front for a new challenger, but 16% non-recognition is not too bad, and a net rating of +1.3 with a "neutral" rating of 37% is OK at this stage.  The issues results are just about a microcosm of the nation.

Seat Betting

The usual disclaimer applies - the object of this section is to have some harmless fun with gambling odds before Nick Xenophon hacks both houses of parliament and all the bookies shut down their websites in alarm.

Firstly if you got on Labor in McEwen when they were out at $2.20 for no especially convincing reason, well done and I reckon you can spend that all at once.  At the time of writing a very small arbitrage exists on this seat if you assume that one major party or other will win it with 99+% certainty.

At present these are the seats considered on the Sportsbet and Crownbet markets to be changing hands or else reasonably close.  If a seat is tied on one market I use the other.  In cases of conflict I call it a tie, but there are currently none of those.  If a seat is close on either market (usually both parties inside $3) I class it as a close seat.

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

ALP Close Holds: Richmond (three-cornered), Lingiari
Coalition Close Holds: Banks, Reid, Page, Gilmore, Lindsay, Robertson, Macquarie, New England (vs IND), Bonner, Brisbane, Forde,  Dickson, Leichhardt, Lyons, Braddon, Deakin, Corangamite, Dunkley, La Trobe, Cowan, Stirling, Swan.

Here's the colour-coded tracker.  A pale shade indicates the seat is tied in one market, and grey indicates an overall tie (there are none at present).  I've added a line which gives the expected Coalition seat total as adjusted for differences in close seats.  For these purposes a close seat is counted as 0.7 to the leading party and 0.3 to the trailing party (0.2 for the Coalition in Richmond).  Note also that Mark the Ballot is doing daily tracking of numbers of seat favourites.

(NB as noted in comments I initially missed Swan from the Coalition close hold list, so the Close Seat Adjusted figure for this week should be 76.9.  I'll fix that in next week's table.)

Factoring in close seats allows for what I think is a more accurate view of what the seat betting markets "think" than just going by numbers of favourites.  On my aggregate, the current 50-50ish national polling started in the week of the April 20 scrape above, and over the next two weeks seat betting markets adjusted to evidence of the national polls.  Since then they've pointed to much the same bare Coalition majority as projections like mine and BludgerTrack.

Sportsbet now has an exact totals market that sums to a slightly silly 254% but with 79 and 80 Coalition seats as the individual most likely totals and a weighted average (for what it's worth) of 78.1.  The line market has the Coalition winning by 12.5 seats (c. 79).  The William Hill banded seat market is still pointing to just under 77, and their exact totals market also points to 77.  Although the Sportsbet 2PP market continues to expect a Coalition 2PP of 51.6, all the evidence from markets specifically about seats suggests punters and/or bookies (depending on who is more to blame for these odds) expect on average that the Coalition will win with a very small majority, which would probably mean a 2PP of about what we have now (50-50).

A lot is said about the accuracy of punters vs polls at predicting election outcomes.  It's a silly comparison at the best of times because betting is forecasting while polling is taking a snapshot of current voting intentions.  But at the moment with the punters leaning towards a Coalition victory while the polls are 50-50ish, the difference is actually an illusion.  Since 50-50 probably means a Coalition win, at the moment the aggregated polls and the markets are saying much the same thing.

Seat Model Example Output

Finally here's some current output from my seat model, a version of which predicted 140 seats correctly last election.  This output lists the Coalition seats the model rates most likely to fall on the assumption that the current 2PP is correct.  (Please don't bet on Labor in Barton at 100-1 on this basis since that assumption might be false!)

The model gives the swing required, and then an adjusted swing-required figure that takes into account sophomore effects, retirements, redistributions and a 50% loading for state polling (as of last week; I haven't updated those weightings yet).  The model's estimate of the Coalition's chance of retaining that seat if the 2PP is 49.9% is shown on the right. This estimate takes into account seat polls where they are known.  Some of the individual estimates are bound to be rubbish because of factors outside the scope of the model, but if the final national polls are accurate then any errors should cancel out.

Mostly, this list gives similar forecasts to the seat betting markets, but there are three notable exceptions: Cowan, Brisbane and Eden-Monaro.  For Brisbane it's possible the market is aware of the track record of Queensland polling (but the seat is a vacancy, which is why the model's keen on it falling).  For Eden-Monaro the fact that the former member is recontesting may be enough to explain the difference (my model doesn't currently account for this.)  But the case of Cowan is an intriguing one - can anyone explain why betting markets expect this seat to be held?

Cowan Update: As of the afternoon of 1 June, Cowan flips bigtime on Sportsbet going to ALP $1.35 but Crownbet has not yet moved.  I doubt that will remain the case very long.  Should Labor be favourite next week Cowan will be added to next week's colour tracker.


  1. I feel like Cowan has slipped off the radar a bit in recent elections due to Labot not really throwing resources at it since the retirement of the very popular Graham Edwards: Liz Prime underwhelmed in 2007, Chas Hopkins was a last-minute recruit after a candidate mishap in 2010 and turned out to be a goose, and their 2013 candidate was nobody I've ever heard of (and I'm about as wonkish as they come). Anne Aly is a great candidate and Luke Simpkins is also a goose so I can see why it'd come back now with a swing on in the west, but after a decade of it being well down the list of watched seats, not surprised some punters might forget Labor has a pulse up that way.

  2. I have been confused by the odds in Cowan as well, it's really hard to imagine liberals holding it while the swing is so large in WA. Talking to my volunteer friend in labor they are feeling pretty confident in taking it.

    Any idea why Stirling is so close labor haven't announced a candidate yet, and I don't think they plan on putting much effort in to winning it at this point.

    Another seat with weird odds is Canning with the swing on so hard in WA I would expect labor to at least be competitive.

    Should swan be listed as a close coalition hold on your list?

  3. This comment has been removed by a blog administrator.

  4. Thanks for the correction re Swan, this has been fixed. I've deleted what I assumed was a duplicate comment. (Comments here are manually approved by me so it sometimes takes me a while to clear them.)

  5. Stirling - it is on 9% so with much of the WA polling suggesting the swing may be that much or greater, I guess the concern is that if the swing blows out Labor just might win it no matter who their candidate is or what effort they are making. Candidate factors don't have a huge impact (unless the candidate is especially stellar or awful).

    Canning - this is an interesting one because the margin from the last election was disrupted by a by-election. But by-elections usually see a swing against governments anyway. Hastie has had limited time to build up a personal vote but the unusual prominence of the by-election probably means he has very high name recognition. So I think the seat should be treated as on a margin slightly less than its last election margin, but not much less.

  6. Very quiet campaign so far in South Australia. No one seems to be talking about the election at all. Xenophon will do well, especially with the over 50's and there are plenty of that demographic here in South Australia. The Abbott Liberals said some very unkind things about South Australia and like Tasmanians we do not forget such things and it is very noticeable how quiet the Liberal MP's are here. Not even any junk mail as yet. Labor is not that popular either, not with a Labor state government, that has been in forever and not seeming to be doing anything much at all. Many people will not hesitate to vote for Xenophon and for South Australia to have some non-old party representation in Canberra.

  7. Sportsbet is favouring Labor in Page

    1. Thankyou! Another one to add to next week's list. So many at-risk Coalition seats ...

  8. I'm not sure how your analysis incorporates retiring local members, but it seems way out on Dunkley. The very popular Bruce Billson is retiring after 20 years, the Liberals pre-selected an outsider rather than a local, and Labor's odds on Sportsbet have continuously tumbled from 7.50 with Billson to 2.30 today. Momentum seems all one way here.

    1. At the moment the model treats all sitting member retirements equally until there is some form of evidence like a seat poll that comes along and tells it otherwise. And yes it is blind to subjective evidence about the quality of the opposition candidate (eg if the candidate is a blow-in). That said the difference between the model's current estimate for the Coalition's chances in Dunkley (67%) and the betting odds estimate (59% ignoring the minor parties) isn't actually all that large - it's equivalent to about half a 2PP point.

      There is good evidence that long-term incumbents (12+ years) have higher personal votes on average, but only about half a point higher than the average (and that is primary vote and not necessarily 2PP). Peter Brent's measure of personal votes in 2013 ( is rubbery but doesn't suggest Billson was actually super-popular.

      I will be surprised if Dunkley doesn't get a seat-poll during the campaign, which may help blow any cobwebs out of the model.

  9. G'day Kevin,
    I've noticed a few polls paid for by Getup and some unions in the last few weeks and my instinct is to be very sceptical of any poll put out by a partisan organisation (either left or right wing) but maybe my instinct is wrong. Are you aware of any analysis of the accuracy of similar polls previous elections?

  10. My reaction is the same: treat all polls released by parties, lobby groups, businesses or anyone else with something to gain from a particular result with extra caution. I'm not aware of any systematic analysis of such polls in previous elections but this election there may well be enough such results to conduct one. From memory union polls for the Victorian election didn't do too badly while the Greens have sponsored a number of polls showing them winning things that they then sometimes win and sometimes don't. In the US context 538 have reported that internal polling is less accurate than public polling and have attributed that to pollsters getting tempted to give the clients what they think the clients want to hear.

    My main concern about partisan polling is that even if it is all conducted properly (which I believe a lot of it is), somehow one rarely sees an activist group release a poll that does not suit its agenda. So the risk is that the results we're seeing released have been cherry-picked from a larger pool of such polls.