Tuesday, July 9, 2013

Poll Roundup and Seat Betting Watch - July 9-10

2PP Aggregate (Tuesday 9 July): 50.2 (-0.2 since last week) for Coalition 
Individual Seat Betting: Labor favourites in 61 seats (+3 - Page, Eden-Monaro, Moreton)
Seat Total Market: Labor 65 seats (+7)

Last week I introduced a Seat Betting Favourites Watch series in which I intend to monitor the performance of seat betting markets in predicting results for specific seats in an election that is difficult to model.  The reinstatement of Kevin Rudd as Prime Minister has resulted in a massive poll jump for a government that was clearly doomed under ex-PM Gillard, and nobody really knows whether the bounce will amplify before it declines.  It is a good test for whether the markets can out-predict the polls at a time when the polls have good reason to be suspect as predictors, or merely follow them.  I am personally sceptical of the reliability of betting as a predictive method.

This Week's Polls

This week we've had three national polls so far.  Essential Report showed no change on last-week's one-week-sample 52-48 to Coalition.  Newspoll was up one point for Labor to 50-50 and Morgan produced a generally dismissed headline figure of 54.5-45.5 to Labor.  As is sometimes the case Morgan's use of respondent-allocated preferences was creating an out-of-whack result, and when preferences were allocated following the more reliable last-election model, the result was 52.5-47.5 to Labor.

Taken together these polls are slightly better for Labor than the post-change polls of the last two weeks and my own rough aggregate has moved from 50.4 to 50.2 for the Coalition.  Mark the Ballot has Labor 50.5-49.5 ahead and a Bludgertrack update is expected tomorrow. (Update: Bludgettrack is at 50.5 for Labor as well.)

There are several points of note from the Newspoll:

* Kevin Rudd's approval of 43 is the highest approval rating for the Prime Minister since Julia Gillard polled 50 on 18-20 February 2011 (just before the carbon-tax-announcement-induced plunge). Rudd's net satisfaction rating (netsat) of +7 is also the best for the PM since 18-20 Feb 2011.

* The 50-50 2PP is the third time Labor has polled 50-50 since the possibly rogue 51-49 to Labor in March 2011. 

* Rudd's 53-31 lead as Preferred Prime Minister (for what little that indicator is worth) over Abbott is the highest for the incumbent since Gillard's identical lead on 18-20 February 2011 (The 2PP of that poll was also 50-50).  The last time the lead was higher was when Gillard led by 23 on 19-21 Nov 2010 (the 2PP of that poll was 52-48 to Labor).

* Between his first and second Newspolls as returned PM, Rudd's approval rating went up seven points while his disapproval was unchanged, for a netsat change of a rather healthy +7.  Looking at other examples of netsat change between first and second Newspolls where the question was asked for a new leader, Rudd v 2 scores well here, albeit off an unremarkable base.  In 14 previous party leadership changes, the average change from first to second Newspoll for the new leader has been just +0.3 points (approval +3.4, disapproval +3.1).  In only four previous cases has the new leader's disapproval not increased (and one of these was Crean, whose approval didn't increase either).  The best previous second-poll netsat changes were +9 for Latham (approval +9, disapproval level) and +9 for Peacock's second attempt (+5 and -4).  The worst such changes were -15 for Julia Gillard and -8 for Brendan Nelson.  In Gillard's case, the first Newspoll in which her approval was canvassed had a 2PP of 55-45 to Labor, which was inconsistent with other polling at the time.

As usual far too much has been made of Rudd's PPM lead, when the netsat results are far more important given that good Prime Ministerial netsats are often a driver of future 2PP change.  Preferred Prime Minister is a figure that massively advantages the incumbent.  I wonder if ABC-TV News would have made such a big deal of Abbott's PPM score of 31 being lower than Gillard's final such score had they been aware that in a 50-50 contest there is an average advantage of 16 points to the incumbent.   Rudd's 22 point lead is batting a little above that average, but it's not telling us anything we can't determine from the netsats and the 2PP.

In summary of the Newspoll, we're back about where we were at various points in the first six months after the 2010 election, save that Abbott's personal ratings are worse.   It is the sort of polling we were seeing before the "carbon tax backflip" came to dominate national politics.

Essential also released some attribute polling on which Rudd led Abbott on every positive while Abbott led Rudd on every negative, save for "erratic" on which they were tied.  On average across the listed attributes, Rudd outperformed Abbott by 10.6 points.  This contrasts with the April attribute comparison in which Abbott outperformed (or under-failed) Gillard on 11 of the 14 attributes and had an average lead of 5.8 points.  Among the more damning readings, Rudd was rated "a capable leader" by 58% while 42% rated Abbott thus, and 55% considered Abbott "narrow-minded" compared to only 34% for Rudd.  (The latter statistic is very similar to their respective Newspoll disapproval ratings.)

In the most major event since reclaiming the leadership, Kevin Rudd has announced he intends to push massive reforms to Labor's leadership election system that would make it very hard for an incumbent Labor PM or even Opposition Leader to be removed. Probably these ideas will be very well received at first but it remains to be seen how they go when it comes to the task of getting them endorsed as they are analysed and criticised. 

 Next week, Nielsen enters the post-spill fray. 

Under Or Over?

It was useful for modelling purposes that Newspoll came in at 50-50, because the Morgan and Essential readings were at opposite extremes of the post-Rudd polling so far, and both came from pollsters that are under a bit more of a cloud than some.

A big issue in trying to aggregate polls at the moment is how to deal with Morgan Multi-Mode, which is a very regular poll with a huge sample size but also some rather suspicious properties.  The Pottinger model downweights it by a factor of five because it is too "overdispersed" (ie bouncy) for its sample size.  My rough model does something similar by treating it as no more useful than a Newspoll, Nielsen or Galaxy with much smaller sample sizes (and indeed removing one completely when the next comes along).  It is not only the overdispersion that is an issue with MMM but also that there are still no public details that establish the proportional breakdown of the different polling methods use and whether this is stable,  and any possible variation between the results of the different methods over the full time that MMM has been running.

Essential has the opposite problem - under-dispersion.  While all other polls in the field so far have shown a massive surge for Labor following the return to Kevin Rudd, Essential moved only three points.  Over the past year the poll didn't buy the full extent of the late-2012 Gillard recovery, then didn't buy the full extent of the early 2013 crash, and now it seems to be slow in crediting the Rudd return revival, if it is even going to fully credit it at all.  In an earlier article about Essential I suggested that its subsampling could be causing the apparent problems with this poll.  It's since occurred to me that a poll that uses a panel (as both Essential and (partly) Morgan do) is very vulnerable to either underdispersal or overdispersal should its panel just happen to be a more or less volatile sample of the Australian electorate than average (in some way that scaling does not cater for).

If a panel poll has more rusted-on voters and less swinging voters than average, it will not be as bouncy as it should be, and it will underestimate the direction of changes in public support.  On the other hand if a panel has more swinging voters and less rusted-ons, it is more likely to over-amplify swings when they happen.  It's especially a concern for aggregators that in the third week of Rudd return polling, only one pollster has showed a Labor lead, and the same pollster has done so twice.

The Market Responds

I don't think there was anything that amazing about this week's polling.  Since bounces usually last longer than a week it was most likely Labor's position would either improve or go nowhere, and it seems it improved slightly.  But there has been substantial market movement which seemed to follow hot on the heels of the Newspoll result.

The headline rate for Labor has been slashed from around $4 to a little above $3.  Here's the seat total betting at Sportsbet:

Last week 30-40 seats was at $5 (which was silly as noted at the time) and 41-50 seats was at $3.50, while the new favourite, 71-80, was at $4.50.  On average this market is now expecting about 65 Labor seats (+7).  I'm modelling "under 30" as having an average of 25 and "over 80" as having an average of 85, but perhaps the latter assumption should be changed, though it would have to be changed substantially to make a real difference.

Turning to the individual seats, here's the same chart from last week, updated.  Bold shows that the colour for the seat has changed (though in three cases this is not a change in favourite; it just means there is no longer a tie on one exchange):

The colour-coding:

Medium blue: A seat in which the Coalition is favoured in all betting markets.
Pale blue (none this week): 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.

Again, the starting point is 75 Coalition, 73 Labor and 2 Others.  The seat betting markets now have Labor favourite in 61 seats (+3), with 12 expected losses.  While this is a relatively minor change, Labor has also firmed in many seats where favouritism has not changed, so the situation last week in which Labor was supposed to be much more "exposed" to close seats is over.  On Sportsbet, Labor's opponents are now inside $3 in nine seats where Labor is favourite, including Melbourne (down from 13 last time, in spite of three seats crossing over from a Coalition lead) and the ALP is now inside $3 in eight seats where the Coalition is favourite (up two, again in spite of three seats crossing).  Seats where the ALP is now inside $3 but trailing are its own seats of Braddon (Tas), Reid (NSW) and Lingiari (NT), and the Coalition's Queensland pack of Brisbane, Dawson, Longman, Herbert and Forde.  The market and/or the bookies are starting to take the possibility of a big swing to Labor in Queensland more seriously, and if this continues, Labor will become favourites in Coalition-held seats for the first time since seat betting opened. 

The seat of Dobell is one that interests me.  Following the Craig Thomson scandals, it is one where Labor is thought to be in great trouble, and for this reason it has resisted the move back to the party in betting, and is now Labor's "safest" seat being flagged as a loss outside Tasmania.  It is true that Labor has the disadvantage of losing the personal vote of a sitting member who performed extremely well at the last election.  But how much that should rub off on the subsequent candidate is not so clear.  In Robertson at the last election, a problematic (albeit less so than Thomson) sitting member was booted and this was expected to make the seat a write-off; in the end Labor retained it and even picked up a small swing.  It would be very useful to see some fresh polling in Dobell. 

At the moment, these moves suggest the markets last week did not know any better than the polling, were confused, and are reacting to poll evidence (while still being sceptical that the contest will stay as close as polling indicates). If anything significant happens in either polling or betting over the next few days that I notice, I'll update this article.

PS: I've just noticed the Sportingbet "Correct Election Result" market which has more gradation than the Total Labor Seats one previously mentioned.  It comes out to a projection of about 66 Labor seats assuming a crossbench of two.

Update (10 July): Lingiari (NT) moves to light blue - it is now a tossup on one exchange, but still a Coalition gain on the other.

There have been large moves towards Labor today on the headline markets with prices coming down to the $2.75-$3 range.  The implied seat market mentioned in the PS above now projects about 67 Labor seats.

However, I really should watch what I wish for.  When I said "It would be very useful to see some fresh polling in Dobell" I meant real polling, not a poll "conducted by university students and paid for by author Bob Ellis" (see here) and supposedly showing Craig Thomson winning off the back of some extremely dubious preferencing assumptions (why on earth should the Greens or their voters preference Thomson ahead of Labor?)  Aargh!

AMR Update (11 July): A relatively rare player, AMR Australia has released a poll showing the ALP leading 51-49, but that if Malcolm Turnbull was Liberal leader, the Coalition would lead 57-43.  The results of this weighted online sample of 1107 voters (a similar method to Essential but without the rolling averages) show Labor with a primary rather higher than what most other pollsters are getting, 42%, but this is counterbalanced by a rather miserly 7% for the Greens. A similar poll in March, which I overlooked at the time, showed the Coalition leading 56-44 (around what others were getting) but a 51-49 lead for Labor under Rudd (also in the ballpark)

As I've noted before, "counterfactual" polling is unreliable because it is hard for voters to anticipate how they will vote in circumstances that are different to those they currently find themselves in.  Nonetheless, the counterfactual polling for a Gillard to Rudd change has so far scrubbed up rather well as a predictor of the polling immediately upon Rudd's return.  So I see no reason to doubt that if the Coalition did switch to Turnbull they too would receive at least a large short-term bounce for getting rid of their unpopular leader.  The view that the Coalition might yet replace Abbott with Turnbull is still a bit "fringe" given the closeness of the election and the unlikeliness of Abbott reaching a clearly lost position in that time, to say nothing of the difficulties of arguing against leadership instability while also demonstrating it. But there is just a little bit of muttering out there about the Abbott leadership, and if you're absolutely certain Abbott will make it (and right) you can now pick up a 6% return in a few months.

Although AMR polls rarely, I don't have any particular reason to doubt that its results are broadly valid and therefore I have included it in my aggregate.  The result is that my aggregate has for the first time reached 50-50.  

Today's Bludgertrack update sees Labor ahead 50.5:49.5 and projected to win 77 seats, the main feature here being that the Queensland boom remains strong in most state breakdowns.  Bludgertrack now projects an average loss of only one seat in Tasmania, but the state is difficult to model with much of the data coming from small and mostly unbelievable (at least to me) Morgan samples, for which there is very little pre-spill data.  I would still expect seat polling to be showing Labor behind in Bass and Braddon, although while the national 2PP remains around 50-50, Franklin and Lyons should be retained.  Hopefully a significant body of non-Morgan Rudd-return Tasmanian polling will appear soon.

Update 13/7: Yesterday (thanks to David Barry for letting me know when it happened), Brisbane reached parity on Sportingbet/Centrebet, though not yet on Sportsbet.  It is the first Coalition seat where the Coalition has not been favourite on both exchanges.

No comments:

Post a Comment

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.