Tuesday, July 30, 2013

Poll Roundup and Seat Betting Watch July 30 - Includes History of Seat Transfer By Swing

 Note added 5 Aug: I see there's a lot of people coming here today, could be something to do with something that happened yesterday.  I watched it on ABC News 24 and thought it was the slowest form of motorsport with the longest pitstops I'd ever seen!

Yes, OK, the election's been called. There will be new material here (the usual weekly roundup and seat betting stuff plus some stuff on projecting results five weeks out) sometime overnight or, if things go slowly, tomorrow late morning.  

There could even be a Wirrah!

=======================================================================


2PP Aggregate (Tuesday 30 July): 50.2 TO COALITION (effectively +0.1 since last week - slight method change)
Individual Seat Betting: Labor favourites in 65 seats (+2, Banks and Reid)
Seat Total Market: Labor 68 seats (+2)

This is week five in a regular weekly series in the leadup to the federal election.  Week four was here and through it you can click back to the previous weeks.  Or just click the "betting" label at the bottom.  As stated before, the aim of this exercise is not to claim that seat betting markets have predictive value, but to test whether they do, and to see which of the markets and the aggregated polls see the ultimate outcome of the election first.



This Week's Polls

This week Galaxy released its first poll since the one it took just after the reinstatement of Kevin Rudd, and the result was 50-50.  Galaxy is a remarkably "underdispersed" (steady) poll for its sample size, and it is not quite clear what kind of magic it employs to achieve this, but anyway this result provided strong evidence that last week's Newspoll shift was mostly if not entirely random sample noise.  Of the weekly polls, Essential remained at 51:49 to the Coalition, while Morgan came in from 52-48 to 50.5-49.5 to Labor (by last election preferences.)  At the moment, Galaxy's 2PP results seem to be the least problematic of any Australian polling house - except that their stability seems too good to be true!

After nearly applying a house effect correction to Essential last week, I've now decided that the close-to-50-50 result from Morgan this week, together with Essential staying at 51-49, justify treating the two as close enough to cancelling each other out. So I am now "correcting" each poll by a point, solely for the purposes of stopping my aggregate bouncing around over the few hours each week in which it contains one but not the other. Essential has been on the one side of 50-50 for every poll it has released since the Rudd return, while Morgan MM has always been on the other.  My current aggregated result is 50.2% 2PP to the Coalition.  While there is a 0.1 point move to Labor in my aggregate this week, that is because it had a house effect for Morgan but not Essential last week.  Without that minor method difference the readings for the last three weeks are 50, 50.1, 50.2 for the Coalition.  If that sounds like progress for the Opposition, it probably isn't.

(By the way, while it may seem tempting to try to reverse-engineer the fortnightly Essential figures and work out what the weekly components were, it actually doesn't work because of rounding.  Julian King has a post which I found rather amusing here on the pitfalls of this little trap for beginners.)

There was some hope for the Greens in this week's polling with Morgan delivering a 1.5 point lift from to 10.5% and Essential showing a sudden jump from 7 to 9%, suggesting that this week's sample was probably a similar result to Morgan's.  There is stronger evidence now that Labor's move to the right on asylum seekers is having some effect in the party's favour, reversing a serious loss of oxygen for the party in the first three weeks of post-Rudd polling.  However none of that was visible in a northern and rural Tasmanian ReachTEL that delivered results so feeble for the party that, for the first time, I'm considering the Senate seat held by Peter Whish-Wilson to be at significant risk.

Leadership and Issue Polling

This week's leadership polling repeated the story from previous weeks.  Galaxy found Rudd to have a 51:35 lead as better Prime Minister over Tony Abbott, with Abbott +1 since Rudd's reinstatement - very similar to results from other companies in which preferred PM polling tends to favour the incumbent.  Galaxy also found a 46:38 lead for Malcolm Turnbull over Rudd, which is yet again the sort of lead that Opposition Leaders tend to have if their party is winning 55:45 or greater.  While this is slightly closer than Morgan's 52:38 on a similar question two weeks ago, the percentages of supporters of each party ratting to the other leader in a Turnbull-Rudd contest are nearly identical - 12% of Coalition supporters preferring Rudd to Turnbull, and a massive 28% of Labor supporters preferring Turnbull to Rudd. 

It is quite strange then to see something as mundane as the lack of ability for the party to gather in one place being argued (even by Peter Hartcher) as a conclusive obstacle to a Turnbull return if Parliament is not recalled.  What, for want of the hire cost for a venue for a few hours, the airfare costs to get everyone together and a bit of schedule disruption, the party would be prepared to lose the election it has fought so hard to win, should its polling turn to the point where it strongly appeared that Abbott would lose?

Essential also released preferred Liberal leader polling but it was marred by very high Someone Else/Don't Know ratings compared to Morgan's.  Possibly this is partly down to a difference in question form, but clearly not entirely.  (Essential asks "Which of the following do you think would make the best leader of the Liberal Party?" but Morgan asks "“If you were a Liberal or National Party voter and helping to choose the Coalition leader for the next Federal Election, who would you prefer?")  Anyway, Turnbull 37 led Abbott 17% and Hockey 10%, a slightly less emphatic result than Morgan, and with Abbott leading Turnbull 36-35 among Liberal voters.  

Galaxy (most full breakdowns here) has found that Rudd/Labor leads Abbott/Liberal on the handling of asylum seekers 40-38, on climate change 45-31, and that Rudd leads Abbott on the best vision for the future 46-38, and on being more likely to keep election promises 37-32.  The two leaders were tied on economic management 41-41. Simon Benson tried to spin that as a negative for Labor, when in fact the past history of economic questions shows that the Coalition as a party and even sometimes its leaders (see Mumble here and here) have led on them even at times when Labor was governing and a million miles ahead on 2PP.

I enjoy extracting breakdowns for non-major-party voters (Green/Ind/Other) from Galaxy results, and here are those I've been able to work out (these are approximate only).  They tend to have 50+% uncommitted rates:

* On the handling of asylum seekers, Rudd/ALP leads Abbott/Coalition 30-11.
* On climate change, Rudd/ALP leads Abbott/Coalition 46-8.
* On vision for the future and likelihood of keeping election promises, Rudd scores 43 and 32 respectively, while Abbott's score is so low (probably a few points only) that he is lost in the rounding on both questions!

Essential also covered asylum seeker policy with the PNG announcement receiving a big thumbs up, 61-28.  A 75-16 response from Labor supporters has had some asking whether the party retains any social conscience, but I think that some Labor supporters are so determined to keep Tony Abbott out of the Lodge that they will approve of anything that looks like doing the job.  Coalition supporters also approved the plan, 58-32, and even Greens supporters (32-62) weren't universally against.  And no, those Greens supporters supporting the plan are not necessarily soft voters.  The party's history of a strong stance on asylum seeker issues is relatively recent and there are still many within the party's base who prefer that Australia refuse to accept lots of new arrivals for environmental reasons.

A poll question on which party had the best asylum seeker policy had the Coalition ahead 26-25 but with a huge none/don't know response.  The very high none/don't know response compared with other polls makes the question of limited use, but it's worth noting that this question gives the Greens as an option.  Any such question that gives the Greens as an option (as Newspoll did last week but Galaxy did not) will tend to record worse results for Labor than one that does not, because Greens supporters are more likely to pick their own party rather than Labor.  Given the choice of Labor or the Coalition only, Greens supporters tend to pick Labor's policy over the Coalition's.  For this reason, the differences between the Galaxy and Newspoll results on this question were probably largely down to differences in the range of answer options available. That said, in this Essential poll question Greens supporters were the least likely to be confident their party has it right - another reason not to expect the issue to drive too great or lasting an increase in Greens polling.

Essential also asked voters to state whether the asylum seeker issue was the most important or one of the most important issues, with 35% rating it one of these.  But only 14% had it in their top three from a list last week.  I suggest that asking respondents to rate a single issue in isolation leads to an exaggerated response and that this question should therefore be ignored.

All up, it is very clear that in current polling Labor has at least neutralised the asylum seeker issue.  It remains to be seen what requirements the electorate will have for it staying that way, and whether perceptions will drastically alter depending on whether the rate of boats falls by election time, or in response to asylum-seeker incidents.  I don't buy the claim that Labor will lose the election if the boats are not stopped, especially given that the Coalition has not been expecting to stop them for years.

The Market Responds: Galaxy Cancels Out Newspoll

Betting market responses following the Galaxy poll have pretty much reversed the excessive impact of last week's Newspoll, with Labor back on the headline rate from the high $3 range last week to $3.10 on one market and $2.90 on the other as I write.

Sportsbet reinstated betting across the full range of seats today and so I am back to being able to use two sets of odds as the basis for the obligatory chart:


The colour-coding again:

Medium blue: A seat in which the Coalition is favoured in all betting markets.
Pale blue: Coalition favoured in some markets, level in others.
Grey (none this week): All markets tied or both parties ahead in some markets and behind in others.
Orange: Labor favoured in some markets, level in others.
White: Labor favoured in all markets.

Bold shows a seat that has changed in colour in the last week.  There are three of these - Banks and Reid have crossed the floor (though Sportingbet has them tied) while Robertson remains in the Coalition's pile on average but is a tossup according to Sportsbet.

The market now expects Labor to lose nine specific seats (four in NSW, three in Victoria and two in Tasmania) and to gain Melbourne from the Greens and Brisbane from the LNP, for a net result of 65 seats.  Seat total betting gives the party a slightly more optimistic 68 seats assuming that Melbourne is won.

On to the close seats picture.  The number of seats where Labor are close favourites (defined as opponent under $3 in any market) is up two to fifteen this week, but that's including the two seats that have crossed from the Coalition.  The number of Coalition-favoured close seats in the markets is eleven, an effective gain of one (that one being Greenway).  Labor remains at $3 or worse in five seats that it currently holds: Bass, Dobell and the Victorian trio of Deakin, LaTrobe and Corangamite.

A strange addition to the close seats list for unfamiliar interstate audiences will be the Tasmanian seat of Lyons (Labor +12.3%).  Labor lengthened in every Tasmanian seat following a ReachTEL poll finding it still behind in Bass, Braddon and Lyons (see Tas Federal BaByLon Still Falling?).  In the case of Franklin (which wasn't polled), this reaction doesn't make sense to me, because Franklin was only lineball in the June ReachTEL, and if it has received the same Rudd-boost seen in the other three seats then that should make it likely to be held.

The markets are still not that close to buying the remarkable idea that a seat held by a 20-year sitting member on a 12.3% margin will fall even in an election with next to no national swing. It is the sort of thing that probably needs stronger evidence than just a couple of poll findings (the fresh one reasonably close) by a poll that has little previous testing against election results in the state.  At the moment the markets give the Liberals about a 35% chance of winning the seat.  Furthermore, the poll was taken on Thursday evening when news of the forest-deal cash-splash would still have been filtering through and being digested, so it's possible Labor's situation has already improved or will do so.  However, the deep distrust of the Commonwealth-backed state forestry "peace deal" in rural communities and timber towns in this electorate should not be underestimated.  When the best that ALP state secretary John Dowling can come up with to attempt to quash the picture painted by these polls is that "feedback people are getting out in the field" suggests "It's going to be line ball, Lyons does remain interesting.

Something else that's interesting is this:


This is the Sportingbet market for Labor's best 2PP state.  Naturally, Tasmania has drifted a bit following the ReachTEL, but it might seem strange that the markets still think Tasmania will poll Labor's highest 2PP when polls are showing Tasmania is a trouble zone for the party.

It's actually not that odd at all.  The basis for it is that Tasmania had a 60% 2PP to Labor last time, so a swing can be large enough to knock off the northern seats on their margins of 6.7 and 7.5%, and even threaten Lyons on 12.3%, yet the state 2PP can still be above 50.  This is especially so as the swing in Tasmania is not uniformly distributed.  It just happens that the party is at risk of getting a still-good 2PP in the state but being sorely dudded by geography in seat terms.

A Brief History of Government Seat Losses - And What This Means Re The Markets

Warning: this bit climbs up to about Wonk Factor Three. 

The market expectation of the number of Labor seat losses is shrinking, week after week.  But assuming Labor is not going to make a lot of gains, this raises the question: how low can Labor's seat losses go.

Sitting governments virtually always lose seats to their oppositions, even if they obtain a swing in their favour.  Only the newly-formed Nationalists under Billy Hughes in 1917 (an artificial case),  the Nationalists again under Stanley Bruce in 1922, the Liberals under Menzies in 1955 and 1963 and the Liberals under Harold Holt in 1966 have retained office without losing a single seat to their opponents.

This raises the question: with the 2PP at around 50-50, how many seats would the Rudd Government expect to lose?

This is a graph that shows all election results from 1940 onwards and compares the percentage of government-held seats from the previous election (or notionally allocated during expansions) lost to the Opposition (not the crossbench or friendly fire) with the 2PP swing to or against the government.



It's not that far off linear for cases of a swing against the government, with a big outlier for 1969 in which the government's massive win at the previous election doubtless helped it to cushion the damage.  But the graph as a whole is a shallow curve (any suggestions on what equation might model it and the two to follow welcome.)  Historically if there is very little swing either way, a government will lose some seats, but probably around 5% (say, three or four), and unlikely to be more than 10%.  For a swing around or exceeding 2% to the government, a government loses very few if any seats; if the swing exceeds 2% to the Opposition then at least 10% of seats have historically been lost.

Here is the same graph for the history of seat losses by Oppositions to Governments:


It is fairly common for Oppositions to not lose any seats to Governments at all, and this has happened eight times since 1940, including 1951 in which the swing was close to zero (but Labor had performed badly in seat terms relative to 2PP at the previous poll).  For a zero swing it would be normal for an Opposition to lose about 4-5% of its seats, and for a swing above 1.5% to the Government, the Government should win at least 10% of Opposition seats.  We don't know what a 2PP swing to a sitting Government of between 0 and 1.5% does because there hasn't yet been one!

And most importantly, the net outcomes (in terms of seats won by Governments from Oppositions minus seats won by Oppositions from Governments):

It's rather close to linear, but perhaps not quite.  On average, the relatively few governments that have had swings in their favour don't seem to have had got quite the bang for buck that Oppositions have. 

If we treat it as linear the equation is:

proportion of seats transferred = 0.0455*percent swing to government -0.0141

...with a surprisingly good 85% of variation explained.  For a zero swing that means an average expected one-seat loss, and the net three-seat gain from the Coalition probably required for majority government doesn't kick in (on average) until around the 51.5% 2PP mark.  I think the actual (slightly non-linear) curve, whatever it is, would have a very similar zero swing result.

I had a look at a multiple linear regression including a term for performance at the previous election too (on the grounds that winning lots of seats the election before helps a government to withstand swings) and didn't find that much in it (the predictiveness increase was negligible). But it's possible that the  government should be docked about one expected seat for a given swing for its poor 2010 performance.  The tiny swing to the Coalition in current aggregated polling is worth another seat at present.

So far, it's been tempting to assume that the market odds imply a net belief (as an average of all those throwing money about) that Labor's polling will go down by a point or two and that the current bounce won't last to election day.  But actually, the seat total market is now extremely close to the expected result for Labor based on historical relationships between swings and net Government seat outcomes.  If the seat total market hasn't quite converged with the polling just yet, there's actually not that much in it.

What is apparent, however, is that the individual seat markets are not buying the picture from state polling.  They are still expecting losses in NSW, although the Bludgertrack aggregation of state polling samples does not support these losses, and most of them are not verified by post-Rudd seat polling.  Nor is the seat total market apparently buying the picture of a slightly friendly national distribution of votes for the ALP.   Or, to the extent that markets are factoring these things in, there is a small degree of remaining pessimism about Labor's vote staying high - perhaps because the markets are still not absolutely certain Tony Abbott will not be replaced by Malcolm Turnbull. 

We need to keep in mind that "the market" does not think one particular thing.  It's a composite of opinions, some of them informed and some of them wildly wrong (in either direction). 


There was going to be more waffle on the subject of the very confused signals of polling-vs-result history from this position, and how little we can really know about the upcoming result based on polling right now, but I think this one is long enough as it is.  I'll save the rest for next week - if we are still in 50-50 land then.

A sobering note for Coalition fans: you'd better hope this is a bounce that's going to wash out of the system by polling day, because otherwise ... governments in this sort of aggregated 2PP position this close to a poll have always won.  Even if most of those governments were yours.

UPDATE (31 July): Forde Crosses (Sort-of):  A second Coalition seat has crossed at Sportsbet with Labor now $1.75 to the LNP's $1.95 in Forde.  At this stage Forde remains an LNP favourite on Sportingbet, $1.68 to $2.05.  That discrepancy shouldn't last. It's actually the first seat that has been split in this way that I've noticed, and so Labor's favouritism tally (counting ties as a half) rises again to 65.5.

Update 2 August:  Simon Jackman's Seat Betting Model

Simon Jackman has just published his latest The Swing article entitled "Betting markets give Labor almost no chance of winning election".

The article computes implied probabilities for the betting markets for each seat and goes on to conclude based on random simulation: "Some 61 or 62 seats for Labor are the single most likely scenarios, given the current state of the betting markets, with only a seven in 10,000 chance that Labor wins 76 or more seats."

But firstly, probabilities in given seats are not within a million miles of independent.  If an informed punter thinks there's a 1/4 chance of Labor winning the seat of Woopwoop and a 1/5 chance of Labor winning the slightly posher seat of Upper Woopwoop, that doesn't mean the punter thinks the chance of Labor winning both is 1/20.   It's far more likely they think Labor will probably lose Woopwoop unless the state or national 2PP improves by a certain amount, which is a realistic but not terribly likely probability.  If that 2PP improvement does occur then the chance of Labor winning Upper Woopwoop goes up sharply.  Indeed, the punters would probably reason that if Upper Woopwoop is won, the swing has likely carried Woopwoop as well, making the real chance of winning both something like, say, 1/6 or 1/7.  It is for this reason that the first election bookmakers in Australia to offer electorate multis got stung to blazes by those quick enough to jump on the offerings, and so far no other bookmakers have been seen making the same generous blunder since.

Some models examining this sort of thing take due account of non-independence (which is a tricky business), but when I see such a tight curve with such steep dropoffs only half a dozen seats each side of the mean, I figure that it either isn't one of them or if it is, it's not doing it adequately. 

But there's another issue:

"The results of this analysis may reflect nothing other than the seat-by-seat betting markets simply flying below the radar for now, and hence somewhat uninformative as to the actual election outcomes. There is no way that Labor could be polling as well as it is nationally but wind up with only 62 seats. Labor will do better than that, if its TPP vote winds up in the 49-50% range."

Now it does seem (as I have argued above) that seat betting markets are flying a little below the radar of normal conclusions from the 2PP on either a national or state basis, and there are a number of possible reasons for this.  But let's look at the seat of Brisbane.  Brisbane is "50.3% at Centrebet, 47.8% at Sportsbet".  The latter figure is indeed what you get by averaging out the probabilities of the given parties in a normal fashion based on odds of 1.75 ALP, 1.85 LNP, 26 PUP, 41 Green, 51 Other.

(For those who haven't done this, this is how it goes:

Implied ALP probability = (1/1.75)/(1/1.75+1/1.85+1/26+1/41+1/51) =0.5714/1.1944 = 0.478)

But look at the PUP odds.  Supposedly, the market at Sportsbet thinks the Palmer United Party has a 3.2% chance of winning the seat of Brisbane, and that the chance some non-major-party candidate will win it overall is 6.9%.  Now, checking the PUP odds in other seats, it turns out PUP is at $26 almost all over the place, supposedly giving it a few percent chance in each seat.  If we follow this logic, we would think the Sportsbet market is expecting PUP to win at least four seats nationwide, with a scatter of seats for the Greens and indies and so on too - well, one would hardly expect the ALP to be winning a majority with that sort of crossbench!

The reality is that the punters do not think PUP has even a few percent chance of winning the seat of Brisbane.  Nor do the betting company.   They've just whacked PUP out at a token but fairly safe $26 all over the place in case anyone (probably a one-eyed PUP supporter) wants to bet on it in a given seat.  For all the company knows, PUP might decide to go all out on a specific seat with a massive spend campaign and, if very lucky, jag it.  So they're hardly about to go offering $300 in seats where they don't know PUP's campaign plan, lest they get fleeced senseless by someone who does.  But the real attitude of punters to PUP chances can be seen on the markets on how many individual seats PUP will win - the heavy favourite being zero

I suspect that the reason for the discrepancy between my article (showing nine expected Labor losses) and Jackman's (showing eleven) boils down to the same thing - that taking the minor party odds too literally drags Labor below 50% in seats where Labor is the market favourite.  If these are going to be counted and listed as "Labor losses" then Coalition-held seats where the Coalition has fallen below 50% should be treated likewise.  But Brisbane, which surely has a Coalition implied win chance below 50%, is just described as "well short of being considered a Labor pickup."

So all up, the markets are indeed pessimistic of Labor's chances compared to current polling.  But not that pessimistic!

4 comments:

  1. Turnbull isn't and won't be a viable option for the Libs, even though it is 'the line' being put. Even if they do change, that isn't who they will change to.

    If I was looking to fit that to an equation I'd be looking for something that apporaches +/- 100% asymptotically as you head each direction.

    You could fit an cotangent function, or a hyperbolic tangent function - both will give you a near linear portion near the centre with the appropriate trend to 'all out all change' at either end.



    ReplyDelete
  2. I was going to leave a comment saying that Forde re-opened yesterday at Sportsbet with Labor slight favourites, but you'd already updated by the time I finished reading!

    I also noticed the change in the Franklin odds, with the Coalition shortening from $3.20 to $2.70 at Centrebet/Sportingbet. At least as a mainlander, I think I can rationalise it. The story would go something like: Tassie's swinging hard against Labor federally in large part because of state issues; in the last state ReachTEL poll, Labor was getting trounced in Franklin as well as Lyons/Bass/Braddon; perhaps therefore the close Franklin federal ReachTEL result was a rogue (rather than Lyons), and under Gillard, the Coalition was leading 54-46 (say); post-Rudd-bounce it's close to 50-50.

    It might only take one or two bettors thinking along these lines to move a seat market (I have no idea what the volume really is).

    It's not a story that would have convinced me to bet on the Liberals at $3.20, but it's enough to stop me backing Labor in Franklin even at a generous-looking $1.40.

    ReplyDelete
  3. For what it's worth David, the markets are very sensitive at the moment.

    The TAB and Sportingbet in particular are proceeding with extreme caution, and it doesn't take much to move prices there in electorate betting.

    ReplyDelete
    Replies
    1. I've only just seen that TAB partial seat market and will include it in next week's calculations. Couple of differences (i) it still has the Coalition favourite in Banks (ii) it has Deakin as a much closer seat than the other two.

      Delete