Saturday, June 18, 2016

Horse-Race Overview And Betting Roundup: Are Labor Too Long?

(Note: This post was actually released 23 June, not 18 June as for some strange reason stated!)

It's just about time to rule a line under the rolling poll roundup that I ran through the last several days and write something new about the overall horse-race picture.  A new rolling roundup will be started when there is a new national poll (most likely ReachTEL on Friday night).

The national polling data that we have on this election at the moment all dates from Sunday or earlier.  Only Newspoll is even dominated by data from last weekend.  The two-party picture is persistently close, and the Coalition's advantage in seat projections is not massive, but betting markets offer headline odds that imply it's almost a done deal.  Many people have suggested to me over the last few weeks that Labor are way too long, and that odds of $3 or $4 might be understandable, but surely not $6 and upwards as has been seen at times.

My own polling aggregate (which, I should restate, describes the state of play at a given time - it is not by itself a prediction) has found the Coalition to be in a winning position at all times since Malcolm Turnbull became Prime Minister.  The Coalition has sometimes slipped below 50% two-party-preferred, but has never yet fallen below 76 seats on my projection of what would happen in an election "held now" at the time.  However, it hasn't had a big lead for a long time.



Here are some possible reasons for doubt about the strength of the Coalition's favouritism based on current polling:

* The 2PP vote is relatively close.  Stranger things have happened than governments losing just just over one 2PP point in the last two weeks of an election campaign.

* The polling is less rich than at the 2010 and 2013 elections, with only four pollsters fully active in this campaign, one of which (Essential) is posting fortnightly-rolled data that is therefore not very fresh.

* Of the pollsters posting fresher data, only ReachTEL exists in the same format as at the last election, and even it has changed its preferencing method.  Ipsos is a new pollster for Australia at a federal election and Newspoll is neither the Newspoll nor the Galaxy of old.  The lack of continuity and the increasing global problem with polling representative samples create an increased chance that the polls may be wrong - in either direction.

* A surge by the Nick Xenophon Team (NXT) could potentially win several Coalition seats in South Australia even based on present polling, and has potential to increase further by election day.  As Mr Xenophon has specified a policy basis for negotiations in a hung parliament, it could be that he would back Labor in a hung parliament, if one eventuated.

* There is a particular issue with voters telling pollsters they will vote for NXT in seats which NXT is not contesting.  These voters might disproportionately vote for or preference Labor.

* There are mixed results on the question of respondent preferencing, with one pollster (ReachTEL) apparently finding evidence for a preference shift that may go beyond the issues created by NXT's entry (although other pollsters don't seem to be finding this.)

None of this is to argue that any of the above things are each likely to bite.   I can also argue that it is possible that the Coalition (following the 2004 model) will now pull away and win with as much as 52.5% (or maybe even more) of the two-party preferred vote.  Stranger things have happened than governments gaining bigtime at the post, as well.

The polls may have an overall skew to Labor as has often been the case at past elections, or there may be a repeat of the UK election polling failure based on oversampling of politically-engaged voters.  The NXT and respondent preferencing issues may fizz, or Greens preferences could even move to the Coalition.  The strategic argument for voting Coalition to avoid a hung parliament may catch on in the final week, or Labor's campaign about Medicare may simply run out of puff.  But none of that affects the question of whether Labor's chances of winning are actually much better than the betting markets say.  For instance, Sportsbet currently has Labor at only an implied 18% chance of forming government, though even that is up from 16% yesterday morning.

Looking at the historic polling-based projections I have, they continue to tell a similar story to what they have told all along.  Polling as of two weeks out explains about 60% of variation in the final 2PP, but even two weeks out there are some massive outliers.  The 1977 election is the biggest: even in the final weeks it seemed close, but Malcolm Fraser won very easily.  The two projections I have been following for 2016 both see the Coalition on average winning (with 50% and 50.6% 2PP) but still give an average error of about two points on that estimate.  If that error was normally distributed then they would give Labor chances of only 20% and 16% of winning from here, but it probably isn't, and on that basis it can be argued Labor's chances are better.

A different way of doing it is to predict the share of seats the government will hold.  In this case the errors on the projection do appear to be more or less normally distributed, which is nice, but the model is also far more sensitive to whether you take the track record of the parties into account.  In this case the simple historic relationship between government 2PP two weeks out and percentage of seats won implies the government will hold about 78 seats, with a 70% chance of holding at least 74.  However the version that takes into account the Coalition's history of outperforming its polling at elections predicts that the Government will win 82.5 seats, with a 91% chance of holding at least 74.  So that model is saying that the markets have Labor too short.

We also know that the governments that did lose in the past were all clearly losing at this stage.  All were on 48% 2PP or worse.   The five governments with 2PPs between 48 and 50 all won, as did the four that were between 50 and 52.  So for a government polling above 50-50 to actually throw it away in the last two weeks would be unprecedented at federal level (there are cases at state level though).

There are two points against the idea that Labor is too long.  It's true that many of the uncertainties applying at this election were not present at past elections, but all elections have some kind of uncertainty about the quality of national 2PP polling.  Often these uncertainties turn out to have no effect.  However, so few Australian governments have lost that we do not know enough about all the possible circumstances under which they could lose.

We did see the ALP win two state elections at longer odds than this - Queensland and South Australia.  In the case of Queensland there was a preference shift of unprecedented size and of which there was no neutral polling evidence (because nobody asked), and in the case of South Australia the natural skew of the state's seats meant that a slight lift in the government's polling got it across the line with a miserable 2PP vote.  In both cases psephologists generally thought the Coalition would probably win, but did not think this nearly as strongly as the markets.  So there's plenty of precedent for markets to be overconfident.  I suspect that favourites in election betting markets may even be prone to have money thrown at them close to the day just because they are favourites, and because so many people believe the myth that markets are very reliable.

So I've outlined above some reasons why Labor might be considered much too long, but also some objective arguments for why they have very little chance at all, based on the state of polling as of last weekend.  I don't think there is an objective way to say which of these arguments are stronger.  Intuitively, I'd have more confidence in the stronger projections if the Coalition's campaign hadn't appeared so lacklustre.  Strong, stable, majority government that only the Coalition can provide?  You didn't deliver that with 90 seats, why on earth should we think you would do so with 80?

This week

We have seen some pretty strange stuff so far this week, without even mentioning "Fake Tradie" (who was of course a real tradie, though the ad still seemed terribly scripted.) Labor launched an audacious scare campaign - based on just the smallest grain of truth - over the alleged privatisation of Medicare.  Instead of treating it from the outset as paranoid, ludicrous and desperate rubbish, the government became sufficiently rattled to go on the defensive and play straight into Labor's hands with a "never ever" type pledge that just recalled similar broken promises in the past.

There may be further jitters - if anyone believes the results - following the publication of commissioned NSW Teachers Federation ReachTELs of six NSW Coalition seats showing an average swing to Labor of 7%, far above what other seat polls (commissioned or not) have generally found in that state or in such of those seats as have been neutrally polled.  These polls - taken as a whole - are so far off the scale that if they are accurate the only explanation could be a sudden large snap in voting intention (at least in NSW) following the Labor launch and what I am calling the Mediscare.  So far, this hasn't been backed in even by reports of insider mutterings from the parties, but if there had been a major shift would either side admit to it?

A tweet from Ben Eltham (New Matilda) today said that Labor sources had said their internal polling was stronger than they are publicly saying.  Are they fibbing to the media in general, or only to the left to boost morale?  We will need to see more national polls to see if the Coalition has actually slipped up this week.

Seat Betting

As usual, just noting that I track seat betting not to claim it is predictive nor to encourage it, but to test the theory that it is predictive.

Quite a lot has happened since last week's review of the seat betting markets.  This is how things line up at present.  A seat is considered close if both parties are at $3 or less on any market:

Loss (Coalition to Labor): Barton*, Paterson*, Petrie, Page
Close Loss (Coalition to Labor): Solomon, Dobell*,  Eden-Monaro,  Macarthur,  Capricornia, Burt,  Cowan
Loss (PUP to Coalition): Fairfax
Tied (Coalition vs NXT): Mayo (split markets)

ALP Close Holds: Batman (vs Grn)
Coalition Close Holds: Banks,  Gilmore, Robertson, Macquarie, Brisbane, Dickson, Forde, Leichhardt, Boothby (vs NXT), Grey (vs NXT), Barker (vs NXT), Lyons, Braddon, Deakin, Dunkley, La Trobe, Hasluck, Swan

In terms of overall favourites, the Coalition has gained Hasluck in the last week while Mayo has become a split market (various sites have NXT or the Liberals favourite or a tie), which is very little change.  I think Mayo may have been a cross-market favourite for NXT prior to claims of links between Rebekha Sharkie and an anti-immigration Facebook site, claims which may have resulted from a set-up rather than being actually valid.

But the close-seat picture has seen a return of the imbalance of close seats that has been a feature for most of the campaign, Here's the colour-coded tracker.



Key to colours:

Dark blue - Coalition favoured in all markets
Light blue - Coalition favoured in some markets, tied in others
Grey - seat tied, or different favourites in different markets
Orange - Labor favoured in some markets, tied in others
Red - Labor favoured in all markets
Dark green - Green favoured in all markets

While no Labor-vs-Coalition seats are "close", a number are almost in the close seat pile.  These include:

Greenway, where Labor would have lost in 2013 but for a hopeless Coalition candidate (so the seat might be considered notionally Liberal).
Adelaide and Griffith, where the inner-city appeal of Malcolm Turnbull may have Labor in some trouble.
Chisholm, where Labor is affected by the retirement of the sitting member and also the Country Firefighting Authority stoush that is damaging its chances in Victoria
Richmond, for no reason obvious to me.

This graph which I prepared before the last election shows the history of Opposition seat losses to governments by size of swing:


(The red dot just right of the -4 and on the zero line is actually two red dots now).  The graph shows that for a swing of up to 3% to the Opposition, Oppositions have still lost at least one seat somewhere in seven out of ten cases.  For swings of around 3.5 to 5.5% they have done so about half the time.  So the chance that Labor loses at least one seat somewhere is substantial, but there has been scarcely any polling in the at-risk Labor seats.

The Windsor and Oakeshott challenges in New England and Cowper are also lurking just outside the close-seat radar.  Also, Kennedy has closed a bit since I gave Katter's then-$1.01 odds a blast a few days ago.  The counter-arguments are the general anti-major-party mood, that Labor are preferencing Katter, and the LNP challenger Jonathan Pavetto is young, low-profile and not just openly gay but also vocal on gay rights issues given Katter's past form on them.  However if the electorate does finally decide it is time for Katter to put the gun away then even these things might not matter.

The Sportsbet exact seat total market now has a median of 81 or 82 Coalition seats, but a weighted average of 78-79.  The former figure is now more closely in line with Sportsbet's expected 51.5% 2PP to the Coalition.  The William Hill banded seat total market has 81-85 seats the favourite range, but with a weighted average of 78.  The William Hill exact seat total market has 81, 82 and 83 Coalition seats joint favourites but with a weighted average of 77.6.  The weighted averages are a bit misleading because they don't factor in longshot bias, so it might be said that what is expected is really closer to the median.

There is an apparent discrepancy in the Sportsbet market for NXT vs Greens to win the most Lower House seats.  The Greens are still $1.75 vs $2 favourites here, but in the individual NXT and Greens seat totals markets NXT are expected to on average win more seats.  Meanwhile the market for the number of crossbenchers still expects five, possibly because rising perceptions of NXT's chances are slightly counterbalanced by doubts about Katter and McGowan, or perhaps just because some of these markets haven't been updated yet.

I'll start a new rolling poll roundup tomorrow night when ReachTEL comes out.

10 comments:

  1. I too am a little surprised that there's been no polling in Labor marginals. I agree that at the very least Greenway might be a contest: without the hopeless Lib candidate last time, it would probably be a Liberal seat with a margin of 2-3% going into 2016. A couple of other seats such as McEwen, Chisholm, Griffith(!), Werriwa (!!) have been sometimes mentioned by "Labor insiders" as being of some concern.

    Also, a number of polls have shown Labor finishing in third place in South Australia, behind Liberals (1st) and Xenophon (2nd). There's been alot of talk about Mayo, Boothby etc, but surely some Labor seats would be a risk of being lost if their primary vote was so low? Adelaide, Makin and Kingston would be a couple of seats where NXT would likely poll quite well.

    ReplyDelete
    Replies
    1. Yes. If the highest level of NXT state polling so far (the 29% in Newspoll) came true, NXT could win pretty much any seat in the state, no matter who it is held by. That said, Kingston, which seems to be the Labor seat that would be first to fall, has a controversial NXT candidate.

      Delete
  2. NXT vs Green has now swapped on Sportsbet. Yesterday their line betting tightened from 12.5 to 11.5 seats. Using 6 crossbenchers, that measurement implies a 78/66/6 result.

    ReplyDelete
    Replies
    1. Possibly another case of my site impacting on betting markets. :) Yes I saw the line betting had come down but the odds on it are not currently equal.

      Delete
  3. One other superficially interesting thing is that so many polls have been commissioned by Labor or left-leaning affiliates, but there's been almost none from the Liberals. Perhaps they haven't bothered because the "neutral" seat polling has been good for them, but it is kind of interesting....

    ReplyDelete
    Replies
    1. I find this to be a common thing in general though at the last election there was that series of JWS aggregated marginals polls commissioned by a group with strong Liberal links. This election there was one big batch of claimed Liberal internals very early on that was denied, and very little either from them or commissioned from the right since.

      Delete
  4. Kevin, you may be interested in this, which I published on Facebook tonight (knowing you're not a Facebooker):

    I haven't been registered in Denison for over two years, and my mobile phone billing address was moved when I moved to a new electorate. Despite that, I was phoned on that phone tonight by the ReachTEL poll system and asked about my voting intentions and feelings in Denison.

    What does that say about their underlying data and the reliability of their Poll results?

    ReplyDelete
  5. This issue of incorrect electorate assignments in ReachTEL polling has been commonly reported at this election. The company say that it is more prominent simply because of the immense volume of calls they are placing (at one stage they made 2 million calls in a few days - note that that number is calls, not successful interviews) and say that the error rate in electorate ID is below 1%. How they audit that error rate I'm not sure. Issues like yours may be explained by an out-of-date database but there have been other cases that were less easy to figure out.

    I'm interested that someone's still polling Denison! Wonder who that is and if the results will be published.

    ReplyDelete
  6. My musings.
    150 seats available. Need 76 to govern.
    Assume the following will retain their seats:
    Cathy McGowan (IND) in Indi
    Bob Katter in Kennedy
    Andrew Wilkie (IND) in Denison
    Adam Brandt (GRN) in Melbourne

    Also assume that Rebekha Sharkie (XNT) will win Mayo

    This leaves 145 seats to contest between ALP and Coalition.
    Following redistribution the ALP has a notional 57 seats.
    Assuming that is has a 50-50 chance of losing Batman to the Greens let us say it starts on 56 seats.

    It is unlikely to lose any other seats. It needs thus to gain 17 seats to have more seats than the coalition and be odds on to form government. I think it will win the following 8 seats from the coalition:
    Petrie QLD
    Capricornia QLD
    Solomon NT
    Eden-Monaro NSW
    Page NSW
    Macarthur NSW
    Cowan WA
    Burt WA

    It must then also win at least 8 others. Below are the seats it has a chance of winning (I've excluded Lindsay). I think it may win about one in three of them – so of the 18 seats below they may win about 6.
    Lyons TAS
    Bass TAS
    Braddon TAS
    Deakin VIC
    LaTrobe VIC
    Dunkley VIC
    Banks NSW
    Robertson NSW
    Reid NSW
    Gilmore NSW
    Macquarie NSW
    Brisbane QLD
    Forde QLD
    Longman QLD
    Dawson QLD
    Hindmarsh SA
    Hasluck WA
    Swan WA
    This would leave the ALP on 70 seats to the Coalition’s 74
    However the coalition has perhaps a 50-50 change of being beaten by XNT in Grey and Barker.
    So I’m guestimating

    Coalition 73
    ALP 70
    GRN 2
    XNT 2
    Others 3 (Katter, McGorie and Wylie)
    This would be a difficult environment where the XNT would decide who governed.

    ReplyDelete
  7. Kevin, in view of the Brexit result I went looking for bookie info and found this - http://www.zerohedge.com/news/2016-06-22/something-strange-emerges-when-looking-behind-brexit-bookie-odds
    This could easily explain Labor/LNP odds here too, but only time will tell.

    ReplyDelete

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. 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.