Wednesday, May 25, 2016

Poll Roundup and Seat Betting Watch: Not Even A Dead Cat Bounce

2PP Aggregate: 50.0-50.0 Dead Heat (-0.1 for Coalition since end of last week)
Coalition would probably win with small majority if election "held now" (seat projection 78-68-4)
No significant move in voting intention detected in the past six weeks

This was a week in which it seemed like things might be starting to happen! First Peter Dutton said some rather provocative things about (supposedly) illiterate refugees.  This was widely regarded as a Lynton Crosby "dead cat" maneuver. The idea of that is that if you are hopelessly losing the argument, just say something ridiculous and disgusting and then everyone will talk about that instead of the issue you were losing.  A tiny minority of calmer voices pointed out that there was no need for Dutton to "throw a dead cat" since there had been nothing happening in the campaign worth distracting attention from, and suggested instead that Peter Dutton was just being Peter Dutton.  But still, perhaps all this was something voters would react to?

Then there was the NBN raid on ALP offices, a seemingly juicy matter that raised questions about the possible politicisation of policing referrals, and made people wonder whether the Prime Minister knew it was coming, and if he didn't know who knew, and who they told or didn't tell.  This, the commentariat told us, could only end badly for the Government, since at worst it looked like an unAustralian denial of fair play to a party in the middle of an election campaign, and at best it still focused attention on the government's failure to even deliver a second-rate NBN for anything near what they said they would.  Perhaps this then would finally set the polls alight?

Nope.  The verdict from the two polls that have been in the field since the above episodes has been still no change to speak of (Labor's primary even dropped a point on both).  A common response to this has been that the public aren't paying attention (how dare the voters ignore matters that have been deemed important.)  That's probably true, but it is not the only possible explanation.  The public could be paying some attention to the campaign but the various events in it could be cancelling each other out.  The public could be paying attention to the campaign but failing to find anything they actually care about (because this stuff doesn't really affect them, and they are sick of the media telling them what they should be thinking).  Or, it's always possible that voting intention has changed but neither of these polls have noticed.  Both Essential and the new Newspoll are often steady to a fault.

What I keep hearing anecdotally is that voters just find the campaign boring, especially when they have to comprehend another five and a half weeks of it.  It's not that the campaign lacks for policy differences or content (compared to, say, the farcically dumbed-down 2010 effort from both sides); it's mainly that the nature of debate is so dry that there is little for anyone but policy wonks to really get stuck into.  And even for policy wonks, there's not all that much depth to see.

Sampling for last week's Ipsos (51-49 to Coalition, 50-50 by respondent preferences) and ReachTEL (50-50 by respondent preferences, and I get exactly 50.0 by last-election preferences) was done before the raid news broke, and since then we've had 51-49 to Labor from both Newspoll and Essential.   All this continues the pattern in which Ipsos leans a bit to the Coalition, Newspoll and Essential a bit to Labor and ReachTEL not predictably to either side.  I aggregated the Ipsos at 50.6 to Coalition, the ReachTEL at 50.0, the Newspoll at 49.1 and the Essential (with an adjustment) at 50.1.  There's a very strong case for deducting a point or so from Ipsos and adding a point or so to Newspoll, but I'm avoiding it (at the cost of the odd wobble of a tenth or two tenths of a point) since they are both polling often enough that they should cancel each other out anyway.  Newspoll is more heavily weighted, but the "global house effect" feature in my aggregate corrects for that.

Here's the smoothed tracking graph.  It's especially smooth this week. The beast that has climbed such mountains and plumbed such depths in a pretty tumultuous term has now spent six weeks almost asleep on the 50-50 line.

Readings from other aggregators are (scores for Coalition): Bludger Track 50.0, Andrew Catsaras 50.0, Luke Mansillo 49.95, Mark the Ballot 49.9, Phantom Trend 49.6


If this week's voting intention figures were boring, the leadership stats were anything but.  This especially applies to Newspoll which had Malcolm Turnbull and Bill Shorten equal on net satisfaction, each on -12 (Turnbull 38-50, Shorten 37-49).  For Shorten this was a gain of seven points in a fortnight - a fortnight in which he impressed many former left-wing sceptics mainly by just being competitive and focused, and also a fortnight in which he was able to show some human touch after a car accident happened right next to his convoy.  It was Shorten's best netsat in just over a year, and only a point off Turnbull's worst as Prime Minister.  The "better Prime Minister" gap also kept closing, and is now a modest fifteen points (46-31), a more or less meaningless lead in the context of this measure's skew to incumbents.

The better PM figures, for what they're worth, set an all-time record, in that the Turnbull-Shorten gap has now closed by 34 points in six months.  There has never (in 30 years of Newspoll) been a case of the gap between a PM and an Opposition Leader closing this much this fast in the direction of the latter, or indeed this much that way in anything under ten months.

What does this new record mean?  Well. basically, for all the mythical nonsense bestowed upon better/preferred PM readings, they are actually a messy multi-purpose indicator.  All else being equal the PM should lead the Opposition Leader by about 16 points, and this then goes up or down depending on how popular the two leaders are and what the two-party-preferred score is.  (It lags behind the PM's netsat and moves with the other two).  All this record narrowing of the better PM score is telling us is that we've had an unusually strong combination of the PM becoming less popular, the Opposition Leader becoming less disliked and the 2PP moving towards the Opposition, all at the same time.

Likewise, it's probably time for another reminder that the relationship between Opposition Leader ratings and voting intention is a very weak and messy one indeed; it's normally the PM's ratings that matter the most.  The fact that many Labor voters who disliked Shorten mid-term are warming to him makes little difference if they were never going to vote for the Coalition anyway.

In the same six months, the difference between the netsats of the two leaders has fallen from 69 points to zero.  This gap isn't quite a record (Keating went from having a netsat 27 points better than Hewson's to being 50 points behind in four months in 1993) but it's been a big relative shift all the same.  But Shorten's big netsat recovery is nothing much compared to that of Andrew Peacock in 1984 (Peacock gained 61 points in two months.)

Other pollsters are showing similar changes, though not yet buying Newspoll's claim that the leaders are now seen equally.  ReachTEL had Shorten at a net rating of  -10.5 (up nine points in two weeks), his best rating since January 2014! Turnbull was more or less unchanged on -6.5 and the forced-choice "better PM" narrowed another two points to 55.6% to Turnbull.  (I think this lead might mean a bit more than the one he has in Newspoll.)  Ipsos has little change for Turnbull (up two points to +10 (48-38)) but Shorten is up five points to -6 (40-46).  Ipsos has Turnbull's lead down as preferred PM down to a modest 47-30, another poll in which the lead has closed by 34 points in six months.

We also have plenty of new attribute polling.  The Australian (Federal election 2016: Turnbull ‘arrogant’ but voters like him) has polling showing Turnbull leading on six of the nine attributes listed, with Shorten leading on "cares for people", "in touch with voters" and (in reverse) "arrogant".  Oddly, both compare poorly on many criteria with ratings for Julia Gillard and Tony Abbott, including those taken in April 2012 when both Gillard and Abbott were very unpopular.  My conclusion from this is that the Newspoll methods changes have resulted in a lower tendency for voters to agree with these statements about leaders.  This could be both a result of the general ease of being negative when pushing buttons on a robopoll, but also the fact that the online segment has to present a neutral option upfront.

Essential has done a fresh leader attributes poll only three weeks after the last, finding Turnbull down on various scores but Shorten little changed.  Turnbull tends to score higher on the comparison whether the attribute in question is positive or negative (leading to some misleading reporting that Turnbull was ahead on all fronts!)

ReachTEL also has some issue importance/party trust questions, the results of which are unsurprising.  The Coalition does well in its traditional strength areas, Labor does well in its, and economic management scores highest of the list of issues by some margin.  Likewise on party trust issues from Essential.

Seat Poll

The only new seat poll I have seen in the past week is a ReachTEL of Macarthur, which finds basically the same result (51-49 to Labor) as the Galaxy poll of the same seat.  I get a similar result by seat-specific last-election preferences, which may be a bit misleading because of the redistribution.  With so long to go and given the general issues with seat polls, all we can say is this seat is in play, which is what would be expected anyway.

Seat Betting

The usual disclaimer applies - this section is not about promoting gambling or claiming that markets are predictive, but rather tracking whether they are predictive or not.

At present these are the seats considered on the Sportsbet market to be changing hands or else reasonably close:

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

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

Here's the colour-coded tracker for seat betting odds:

(Red = ALP favourite, blue = Coalition favourite, grey = tie) (Updated with minor corrections).

The changes in the past week are minor.  Labor has become favourite in Hasluck again but Eden-Monaro has become tied. The markets sense the Coalition as being vulnerable in a very large number of seats but only likely to lose in a modest number.  My own seat model frequently shows the same thing - projecting about 14 total Coalition losses based on the sum of probabilities, but generally only having Labor favourite in about nine specific seats at a time.

For a national 2PP of 50-50 and with a 50% weighting for state polling, my national seat model currently has Labor favourite in Barton, Paterson, Dobell, Petrie, Capricornia, Solomon, Cowan, Hasluck and Burt, with Macarthur and Brisbane as tossups.  I may post some detailed sample output from it next week.  The model is blind to local factors that may be informing seat betting, unless they are reflected in a seat poll, and because of the lack of information on specific seats it will normally get about ten seats wrong nationwide even in its election-eve version.

(Edit: Crownbet also now has a full seat betting market, showing no difference in favourites above except that Swan is listed as a slight Coalition favourite, and relatively minor differences in the other odds.  In the past I've used multiple agencies for this tracking where possible, so that gives the Coalition 81.5 favourites this week. The colour tracker has been updated to reflect this.)

The various exact seat total markets I was tracking last week have all shown no movement since then.

Back next week with more, and the way things are going, it could well be more of the same.

Update Thurs am: Movement at the station - Greens now narrowly favourites in Batman on the Centrebet site but not yet on Crownbet, meaning that an arbitrage currently exists.

Update Thurs midday: That's gone now; Greens $1.50 on both markets.

Tuesday: Morgan Indi: strange indeed

Morgan has released a batch of aggregated seat results for the electorate of Indi taken from the last few years.

The stated voting intentions are:

* For the full term (sample 596), Liberal 37 National 7 Ind 19.5 Labor 24 Green 8 Other 4.5
* Since Turnbull became PM (sample presumably under 200), Liberal 42 National 10.5 Ind 16.5 Labor 16.5 Green 7 Other 7.5

(The primaries last time were Liberal 44.7 McGowan 31.2 Labor 11.7 Green 3.4 Other 8)

The release claims the 2PPs to be 51:49 and 58.5:41.5 to Coalition over Cathy McGowan respectively.  This only works if you assume 100% of Nationals supporters prefer the Liberals (in particular Sophie Mirabella) to McGowan.  In fact, the fact that they don't is one of the reasons McGowan won the seat in the first place. With a more reasonable assumption (say, 70:30 from the Nationals to Liberals) McGowan still wins 51:49 even off the rather sickly primary vote claimed.  Mirabella would still win on the Turnbull-era primaries (about 55:45) but their sample size will be so small (probably well below 200) that they can't be taken seriously.

The pollster has further muddied the waters by lumping voters who didn't give a party preference in with Others. They also claim that not reading out candidate names favours the major parties, but if so why do they say that the profile "shows a tight battle" when by their own admission it may all be a load of rubbish.

Morgan is a panel pollster and as such very prone to local pool selection errors..  It's likely that the re-emergence of the Nats will take some votes from McGowan since several of her votes were theirs to begin with.  But overall these primaries look really odd to me (especially the high ALP primary).  This is another Morgan release that is such a trainwreck of dubious and contradictory assumptions and calculations that it's little wonder that Morgan draws reactions like this from some readers:


  1. I think this model is way out for Dunkley. Libs almost giving up on the seat, as have recognised mistake of replacing a popular local 20-year member - Bruce Bilson - with an outsider that is still arguing to abolish weekend penalty rates (which unions have latched onto). Payout on Sportsbet has gone from 7.50 to 2.30 for Labor and still moving.

  2. Actually this article from two weeks ago doesn't discuss any model for Dunkley - just the seat betting markets (which still have it as a close hold as I write.)