Tuesday, August 6, 2013

Poll Roundup and Seat Betting Watch: Five Weeks To Go (August 6)

2PP Aggregate (Tuesday 6 August): 50.8 TO COALITION (+0.7 since last week)
Individual Seat Betting: Labor favourites in 65 seats (no net change, Reid and Forde become tossups)
Seat Total Market: Labor 67 seats (-1)

In this issue:
* Bad weekend of polling for the Government
* Rudd just not that popular
* Strange views of Green voters on asylum seeker issues
* Seat poll of Melbourne is suspicious
* A small 2PP shift can mean a lot

This is week six in a regular weekly series in the leadup to the federal election.  Week five 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.

Finally, the election has been called, which means we now know how long it is to go until the election for the purposes of comparisons with how other governments have gone before.  Of course, we thought we knew that for most of the year, but it turned out we were wrong - but only by a week.

This Week's Polls

The polls so far this week have been a bit unfriendly for the Government.  Newspoll stayed at 52-48 to the Coalition (when its normal response following a two-point fall the previous fortnight would be a move back), ReachTEL went to 52-48 (+1), Essential stayed at 51-49 and Morgan came in a point (by last-election preferences) to 50.5-49.5 to Coalition.  So we have two polls moving to the Coalition, one remaining at a value that was over trend to the Coalition last time, and just Essential really going nowhere (which is hardly unusual for it).  The net impact is that my aggregate has gone to 50.8 to the Coalition, which is about where it was after the first few days of Rudd-return polling.  The Coalition would be very likely to win an election held right now, but further down I'll have a rough look at what such a lead five weeks out might mean predictively.  (The aggregate is up 0.7 points not 0.6 because late last week I back-included a Lonergan Research 50-50 from a few weeks previous, which has since fallen out.)

Newspoll delivered worse than just the headline with a nine-point drop in Kevin Rudd's net satisfaction to -9.  After two Newspolls Rudd's netsat progress looked good, after three mediocre, and now it is by historical standards poor.  Indeed only Gillard (-17) and Nelson (-19) shed more than nine points of net satisfaction from their first to their fourth Newspoll; almost all other leaders gained.  It's possible that it is harder for a recycled leader to build popularity.  The key point here is that at the moment, Kevin Rudd isn't as popular as those spruiking for his return might have predicted.  He's just less unpopular than Gillard was and Abbott is - and he's so far holding the two-party vote at a much higher level than Gillard.  It's amusing to note that Rudd's satisfaction rating and Preferred Prime Minister score are now only two and one points abovewhere they were when he was removed from the job the first time around.Meanwhile Tony Abbott managed to avoid becoming the first ever leader to poll completely identical figures four weeks running, by losing a single point of approval. His netsat of -22 is his worst since late February.

ReachTEL created a few ripples with a Better Prime Minister score showing Abbott slightly ahead (50.9:49.1) but, as previously noted, ReachTEL's preferred PM scores are in a different universe to those of other pollsters because they do not have a neutral option. More concerningly for Labor, the question "Which of the following parties do you trust most to effectively manage the economy?" saw the Coalition with a massive 60.7:39.3% lead, with the Coalition almost universally behind their side but significant numbers of Labor supporters preferring the Coalition. The suggestion is that other pollsters have only shown the parties roughly level on the issue because many voters who would prefer the Coalition if forced to choose, say they have no preference when not forced to decide.

Essential this week (download) told us that the Green vote is soft and uncertain, but that is nothing new. It also showed Rudd as preferred to Abbott on most of a list of indicators, with Abbott leading by single figures on economic questions, asylum seekers, security/war on terror and "managing population growth". The pollster wrote "When compared with a similar question about trust in the parties asked last month, Kevin Rudd rates considerably better that [sic] the Labor Party on management of the economy, political leadership and security and the war on terror". However the supposedly similar question included the option of the Greens in place of the option "No difference". While the Greens polled very little in their own right on these questions, voters unsure which they preferred of the Greens and Labor's policies may have answered "don't know" to the previous set. Thus I don't think the comparison is valid.

Essential also released yet more asylum seeker issue polling, showing voters as slightly more likely overall to think Kevin Rudd's response is "about right" but less likely to have any view of Tony Abbott's response. Coalition supporters are three times more likely (33%) to think Rudd has the balance right than Labor supporters are to welcome Abbott's approach.  Labor voters who don't like Abbott's policies are most likely to find them too harsh (38%) but 21% find them too soft.  Coalition voters are the other way around re Rudd, 35-18.  Labor dissenters re their own leader's policy are more likely to find it too harsh (19-12), while Coalition supporters who disagree with Abbott find his approach too soft (16-5).  Both leaders seem to be seen by their own parties as slightly more centrist than the parties' support bases on this issue.

Finally, there is some light amusement in Greens supporters returning a stronger Too Harsh: Too Soft breakdown for Rudd (49:13) than for Abbott (42:21).  Unfortunately the sample size is too small for the differences in these readings to be statistically significant, but at least it seems that Greens voters as polled by Essential don't see Labor as clearly softer.  It might surprise some people that any Greens supporter would think Rudd or Abbott were too soft, but prior to Howard a form of environmental xenophobia was much commoner in the party.  It should also be remembered that some people vote Green on particular issues rather than in agreement with the party's entire platform.

Seat Poll Of The Week ... Not!

Most reports of seat polling this week concerned bulk results of Coalition and Labor internal seat polling, which were said to show the Coalition performing strongly in Labor marginals, with apparent agreement that Labor would probably lose the election right now.  But one poll that was published in detail was of the seat of Melbourne (held by Adam Bandt).  It was conducted from 24-29 July, and was reported as showing Bandt winning 66-34.  
The fine print, however, said that that was by 2010 election preferences. Since the big deal with Melbourne is the assumption that the Liberals will preference Labor on how-to-vote cards, that huge margin is irrelevant.  Galaxy did provide a 56:44 breakdown in the event of the Liberals doing so.  It's no surprise that a Green with a 48% primary would win by that sort of margin.  
However another warning sign is that the poll was commissioned by the party that benefits from its release (the Greens), and yet another is that the poll questions have not yet been published verbatim by either the pollster or the client.
Perhaps there is nothing in all this, but at the exact time of the poll, Crikey reported that Galaxy were in the field in Melbourne asking a set of questions, albeit not exactly matching those for which results are now published.  People who were polled have reported the voting intention question as listing candidate names but not parties.  Even if these reports are in error (which they may well be, see below), or the poll mentioned by Crikey wasn't by Galaxy, or Galaxy for some bizarre reason did two different polls of Melbourne at once, it all doesn't matter.  The publication of all questions asked, verbatim, should be a minimum standard for party-commissioned polls to be taken seriously.  Adam Bandt may yet retain Melbourne but this seat poll is not yet sufficient evidence and it is hereby awarded the wirrah.
Wirrah Award For Fishy Polling
[Update: After ReachTEL's Nick Adams claimed on Poll Bludger that "Galaxy didn’t offer party names in that poll and just gave candidate names.", David Briggs gave a response that commenced "This assertion is factually incorrect"; the remainder of  Briggs' reply attacked the accuracy and quality of ReachTEL polls generally without further comment on the Melbourne poll.  As the poll questions list attributed to Galaxy that lacks candidate names differs from the results published by Galaxy, I am inclined to accept Briggs' response and accept that party names were given.  However, the questions as asked should be published.]

 How Much Does A Small 2PP Difference Mean?

Because we're now a known, and small, time from the election, it is possible to compare polling with previous elections at the same stage and see how predictive polling is.

For these purposes (lacking full past data sets for all pollsters) I've had a look at the elections back to 1987 using Newspoll 2PPs.  I've used a method of Newspoll rolling averages that's a bit more sophisticated than the one I usually use - indeed it's exactly the same 5,3,2,1 weighting I use in my aggregate but on twice as slow a time scale.  In some years, that still doesn't allow a lot of data.  For the Rudd-Gillard transition in 2010 including the Rudd polls seems incorrect and including only the Gillard ones produces a result I know is well out of whack with what other pollsters were getting, so I've set the five-week-out 2PP at 53:47 to ALP.

Well, it's better to be in the lead at this stage than not (and this pattern is stronger in pre-1987 Morgan data, such as it is); the party that was leading on this rolling average measure with five weeks to go in the Newspoll era won the 2PP and the election five times out of nine, the election but not the 2PP once, and the 2PP but not the election once.  But in terms of tipping the final margin you'd have been 40% closer on average by just ignoring the magnitude of the lead and predicting a 50-50 2PP.  And in 2001, 2004 and 2007 the final result differed from the Newspoll rolling average this far out by 3.6, 4.1 and 4 points respectively.  (And no, making the model more sensitive to the most recent poll five weeks out and less to the previous month and a half doesn't really help. The big shifts were still mostly yet to happen.)  (Note added 27 Aug: In the case of 2004 this was because Newspoll used respondent-allocated preferences.  With this corrected the error drops to 2.3 points.)

A simple statistical model considering the relationship between polling five weeks out and the final election result might say that the likely average result from here is about 50.4 to the Coalition and that the Coalition is highly likely to win the 2PP (say, about 65%).  However, two of the three previous cases that most mess up such models (1993 and 2004) did involve governments that were also trailing slightly at this stage, and won handsomely.

Though the move is not enormous I take this week's polling shift as bad news for the Government's objective chances of winning.  There are probably three main scenarios after the plateau phase of the last few weeks: (i) the Rudd bounce subsides steadily as the election approaches, leading to a pretty easy Coalition win (ii) events or campaign strategy success help Labor to a markedly higher 2PP than now, and they win (iii) the bounce was more like a paradigm shift and voting intention will stay close with only random moves one way or the other.  This week's move could just be an instance of (iii), but even if so the Coalition would have an increased chance of staying ahead.  It's also consistent with (i), and indeed makes (i) appear more likely to be true.  So, there are now stronger objective reasons to agree that Labor is the underdog.

Headline Haywire Again, But Seat Market Barely Moves

The usual late-week rally for the Coalition on betting markets (if Labor's polling early in the week was good) continued into the new week with the release of these favourable polls, especially the much-watched Newspoll.  Labor blew out from the $3.50-$3.70 range to around $4 following Newspoll.

The Centrebet/Sportingbet seat total market saw significant moves in favour of the tighter outcome range, a natural response to the election timetable being defined.  However, the implied total of ALP wins has only dropped by one seat to 67.  

On to the individual seat markets.  I was going to use the limited range of TAB seats as a third data site but have found their web service so unmanageable I've decided not to bother. Here's this week's 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: 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 four of these - Robertson is no longer tied on one market, Forde and Reid now each have one party ahead on one market and the other on the other, and Longman is now tied in one market.  Labor is credited with one and a half gains from the Coalition (and also Melbourne, though that's tightened) and the Coalition with nine and a half gains from Labor.  That would leave Labor with 65 seats.  If you see something saying the markets think they're actually on for four or five less, and that they have zero-point-something-small chance of winning, see the bottom update last week for a refutation.

Looking at the close seats, there are now two tossups, one held by each side.  On top of that there are fifteen (effectively up one) seats where Labor is narrowly favoured (inside $3 on one site or the other) and ten (effectively steady) seats where that applies to the Coalition.

All up, this has been the first week in which the seat markets showed only minor movement but the polls moved towards what the seat markets are implying.

PS: An interesting market just started on Sportsbet is a "Queensland Special" entitled "No. of LNP seats Labor take".  Fisher is not counted as an LNP seat for these purposes.  Currently, four is the favourite at $3.50, with an average of about 4.5 and a predictably high rake of over 30%.  That assessment seems optimistic compared to the remaining assessments implied by the markets and I'll keep an eye on how this plays out.

There is also a market on the performance of Craig Thomson, with 2-4% the favourite at present (where's Bob Ellis?)

Update (Thursday 8 August):  The announcement of former Queensland Premier Peter Beattie as Labor's candidate for the seat of Forde resulted in Labor becoming a very short-priced favourite immediately, currently at around $1.20 (ie not even what I am calling "close").  There have also been substantial moves in the seat of Dobell, where the completely humdrum news that Labor has managed to find an acceptable candidate (together with the party claiming internal polling showing no swing at all) has knocked a whole dollar off Labor's price, moving Dobell into the close seat category (in my view, not before time).  The Coalition remains favourite in the seat.

While seat betting markets haven't moved that much this week, there was a bit of a surprise when Centrebet/Sportingbet blew out its headline rate to $4.80 for Labor, even creating a temporary technical arbitrage with another market, although no other bookmaker has Labor much above $4.  (At the time of writing, arbitrage also exists between Sportsbet and Sportingbet in the seat of Reid, on the assumption that a major party wins the seat.)  It was amusing that this blowout came at around the same time as a scathing SMH article attacking election betting as a low-money novelty designed to drum up publicity for the bookies' more mainstream endeavours, and to suck the media into providing it. Since I saw the original of this article it has had eight sentences of responses, additions and corrections added.

The object of my comments on betting on this site is neither to encourage nor condemn election betting (on either side of the counter) but to follow developments in it and see what relationship, if any, they have to the final result.  If, as the SMH article asserts, election-betting is largely a publicity gimmick, then it should follow that its projections (at least at this stage) are more likely to be wrong.  However, should the market predictions prove wrong, then that doesn't prove the markets were not serious ones.

Meanwhile, a ReachTEL of the seat of Griffith has shown a slightly underwhelming result for Prime Minister Rudd in his electorate, but given the slight Coalition lead apparent in national ReachTELs, the main thing to see here is the lack of evidence of a swing to the party in this area, consistent with other signs that the Queensland-raft-of-seats scenario is struggling.

Another ReachTEL of Melbourne, this time an internal Labor poll that has been released, contradicts the Greens' internal polling by claiming to show Labor ahead on primaries, with numbers that would result in Labor winning the seat easily if the Liberals preference Labor ahead of the Greens as expected.  The figures from this ReachTEL are far more in line with what should be expected given the national downturn in the Green vote, but it is always possible that Melbourne has held out against that. So we have two sets of internal polling that more or less completely contradict each other, not an uncommon situation. 

ReachTEL also received some questionable publicity in a questionable article by The Age's Tim Colebatch.  Colebatch reported that he disconnected a ReachTEL call when asked to choose which approach to asylum seekers he preferred out of Kevin Rudd's and Tony Abbott's.  Colebatch writes:

" An Essential poll asking the same question the same weekend, but allowing a wider choice of answers, found 49 per cent of voters think neither big party has the best policy on asylum seekers. That's a lot of people for a pollster to hang up on."

In fact, that 49 points represents 6 points who think the Greens have the best policy, 28% who think none of Labor, Liberal or Green have the best policy, and 15% who don't know.  And it's a higher total than for other pollsters.  But the real core question is: whatever the proportion who really dislike both major parties' policies on asylum seekers, do these voters actually tend to hang up when forced to choose.

I asked ReachTEL about this and their answer was that about 3% of interviewees fail to complete the phone call on average (as a result of which their entire response is discarded) but that the highest dropout rate was actually at the first question, voting intention.  It doesn't seem like it's a big issue, and indeed if it was, it should produce a large two-party skew in those ReachTEL surveys that include attitude questions on contentious issues.  (There's no evidence that it does.)  So it seems like the main demographic who answer the first few questions of a ReachTEL then hang up when forced to choose are journalists who do it so they can write articles about doing it.  There is just no evidence that it is common behaviour.

ReachTEL's James Stewart justifies forcing in the Age article as follows: "We don't allow people to say they're undecided [..] Our philosophy is that when you walk into a polling booth, there's no 'don't know' box on the ballot paper.''  Indeed not, but there is the option of voting informal, and more importantly, a person doesn't have to have a view about the parties' stances on every issue in order to vote.  So really it's only the voting intention question for which it is necessary to exclude firmly undecided responses, which most pollsters do anyway.

All the same, I'm pleased that ReachTEL does force responses on issue questions and Preferred Prime Minister, because it provides us with data not available otherwise about what may be in the most tentative thought patterns of the irritatingly high proportion of "undecideds" in a lot of issues polling.  So long as it is not causing substantial and party-skewed dropout rates, it seems like a good idea to me - so long as not everyone does it.

Update 5 pm: Not surprisingly, that $4.80 didn't last long; it's back to $4.50 now.  And the Reid arbitrage is gone with the Coalition now favourites on both markets there.

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.