Saturday, June 15, 2013

Federal Labor Getting Smashed All Over Tasmania

 UPDATE 27/6: With the replacement of Julia Gillard by Kevin Rudd the polling in this article becomes (unfortunately) largely obsolete and we will have to wait for fresh polling to see how Labor is travelling in the Tasmanian seats under the new leader.  My view as hinted at before is that because the factors affecting Labor's vote in Tasmania are in large part local, it's possible the contribution of the Ruddstoration to Labor's fortunes will be smaller than elsewhere.

My brief take on the likely prospects under Rudd (without any specific polling):

Bass and Braddon: Labor is still struggling in these seats but now should have some practical chance of holding instead of none.

Franklin: I favour Labor to retain. 

Lyons: The work I did on ReachTEL's record in the Queensland election makes me a bit more cautious about trying to project the seat based on polling under an old leader that showed an apparently extreme swing.  I think this seat would have been lost under Gillard, and still may fall, but new polling is needed before the risk can again be declared severe.

Denison: Andrew Wilkie remains entitled to strong favouritism assuming no adverse preferencing decisions, but if the Labor primary now lifts then the risk to him from adverse preferencing decisions by other parties will increase.  By no means over.

---------------------------------------------------------------------------------------------------------------------
 Federal Labor Getting Smashed All Over Tasmania

...being the sequel to "Federal Labor Getting Smashed In Bass".

The monster Mercury-commissioned ReachTEL, which I believe to be the largest opinion poll in Tasmanian history, is out!  I saw the results on Friday afternoon but under embargo conditions.  The survey includes federal and state polling for each electorate and statewide.  Today I am releasing my federal comments and tomorrow my analysis of the state results.




Advance Summary: Federal

A federal ReachTEL poll of all Tasmanian electorates shows a massive swing against the Gillard Government in Tasmania.  Based on the poll results:

1. The Liberal Party is easily winning the Labor-held seats of Bass, Braddon and Lyons.

2. The Labor-held seat of Franklin is too close to call and is winnable for the Coalition.

3. Andrew Wilkie could lose Denison only if several other parties preference against him, and not necessarily even then.  Otherwise he is safe.

Statewide Federal Patterns

The statewide federal poll, with the 6.8% undecided redistributed proportionally shows the Liberal Party on 48.8% (+15.2 since last election), the ALP on 28.2 (-15.8), the Greens on 11.3 (-5.5) and Others on 11.8 (+5.2).  Most of the increase in the Others vote occurs in Denison, which alone accounts for 7.7 points of the statewide Others vote.

A state 2PP figure is not given, and no 2PP figure is given for Denison.  However, overall the poll points to a state 2PP swing of around 16 points.  This is higher than estimates previously available, which were typically (and based on very limited data) in the 10-14 point range.  However, those estimates were largely made at times when Labor was doing less badly in federal polling than now. The polling was all taken on Thursday night and it is plausible that Labor's failed attempt to convincingly exploit the tawdry menu saga had some small impact on the results.

The following table shows responses statewide to the question of how voters rate the federal government's performance:

This table is important because it gives us a handle on the preferences of Undecided voters.  Undecided voters are more than five times as likely to think the government is performing poorly than to think it is doing well.  Others voters have a mild tendency to think the government is performing poorly and Labor and Green voters think the government is performing well by a ratio of over 3:1.  Liberal voters think the government is performing badly by a ratio of 36:1.  The impression is that Undecided voters are leaning towards the Liberal view and could break towards the Liberal Party.  Therefore, in distributing the "undecided" voters evenly, I might be being generous to Labor and the Greens. 

We don't know anything here about the views of Others voters who are not Wilkie voters and hence it's hard to say if they would take any edge off the swing.  But there are not many of them anyway.  There are hints in the state poll results (released tomorrow) that those Others who are not Wilkie voters are Liberal-leaning.  They may be supporters of right wing minor parties (KAP and PUP and so on) or there may be some voters who are displeased with the government but have other reservations about the Liberal Party.

On to the electorates:

Bass (2010: ALP +6.7%)

This is one we already have a poll for (see Federal Labor Getting Smashed In Bass).

Apart from a significant increase in the Green vote (and a similar drop in the Others vote) this poll is basically the same as the previous Bass ReachTEL.  Nothing to see here except that nothing has changed.

Braddon (2010: ALP +7.5%)

For some time there has been a disparity in betting between Bass and Braddon, for no better reason than that Bass had had a poll released and Braddon had not.  (Some people might say that Sid Sidebottom is a better regarded local member than Geoff Lyons but this, if true, would have been already factored into the previous result.)  Braddon is on only a slightly larger margin, and the sitting Bass MHR has "sophomore effect" on his side, the state Labor government is copping it in both, so there was no basis for the difference.  And indeed the poll result for Braddon is much the same:

Any questions?  No, I didn't think there would be.  Okay, the Others vote is a bit high and it is possible to speculate endlessly about whether these are KAPpers and/or PUPpers and/or something else entirely, but it really doesn't matter.  Moving right along then ...

Lyons(ALP +12.3)

This is the one that's going to hurt.  Dick Adams has held this seat for twenty years and been said to be in danger (usually incorrectly) at most of the elections in that time, but just when he'd beefed up the margin to what should be the "seat for life" level, along comes a swing of enormous proportions.

The poll shows a swing of a staggering 21%.  I'm not yet convinced it's quite that bad for Labor in this seat, but however you slice it with your 4.2%-margin-of-error knife, this, the only published seat poll for Lyons thus far shows Labor not just behind, but in an uncompetitive position. I've been pointing out for some time that despite the margin in Lyons being larger than that in Franklin, Lyons could well be the more at-risk of the two seats.  But I did not expect to be seeing anything quite like this.

Franklin (ALP 10.8%)

The last of the four ALP-held seats is the only one this particular poll gives them a hope of retaining.

If you see any claims that this poll shows that the Liberals "would win" Franklin, they are nonsense, as the difference between the two candidates is well within the poll's margin of error.  All the same, did you know that if a correct sample of this size is 51:49 to one party, the chance of that party actually being ahead in the electorate is 68%?  I didn't either until I checked it!  There are other sources of uncertainty of course, such as the true intentions or behaviour of Other/Undecided voters, the day the poll was taken on, any slight lean in the polling, and changes between now and the election, so we shouldn't expect to see Black at $1.50 tomorrow unless there are some very silly punters out there. What the poll points to is that the seat is very much in play and may go either way.  Given that the seat only marginally prefers Julia Gillard as PM over Tony Abbott (50.7:49.3) and given that PPM is an indicator that greatly advantages incumbents, it seems that Labor may even be punching above its weight in still being competitive here.

Denison (Ind 1.2% vs ALP)

Another one that would have to sting for Labor, unless they have a very big bundle of internal polls showing it all to be nonsense (or more likely, saying the same thing, so they're used to it by now). They've replaced the hapless Jonathan Jackson with a good candidate who is ideally suited to the left-leaning electorate and speaks as if she's been in politics for years (you can see the footage of a rather dull "debate" involving her, Wilkie and Greens candidate Anna Reynolds here).   Meanwhile the Liberals are messing about with still no candidate announced three months out (they had one, then they didn't - failed due diligence is the scuttlebutt I heard) and are widely accused of running dead to help Wilkie retain the seat.  And yet who is in second place?

The Liberals! Ouch.

Now, I don't believe for a moment that the Green vote in a once heartland seat has fallen into single figures, not even with the triple whammy of the party's general decline, state disillusionment over the peace deal and continuing gouging of their vote by Andrew Wilkie.  But this poll clearly underlines that the Greens are running fourth and cannot win, not that that is news to anyone.   It confirms that Wilkie will make the final two easily, but it does suggest genuine uncertainty about who else will make it with him.

As for the result, if preference distributions fall along normal ideological lines then Wilkie will be safe.  But if there is a gang-up against the Independent by other parties, perhaps not.  The most difficult scenario for Wilkie on the above figures is if the Labor vote picks up and the Liberals drop back before polling day, the Greens preference Labor and enough Greens follow the card (not that that many do) to put Labor in second.  Now at this point, if the Liberals preferenced Labor as well it's just possible that the generally slavish following of HTV cards by Liberal voters would put Labor over the line.  (It is harder for the Liberals to win on Labor preferences, since it is hard to see them getting much joy from the Greens)

I'm doubtful it's that easy for Labor even given those assumptions.  If the parties, for whatever reason, try to stitch up the incumbent it will give him a lot of publicity with which to attack their deals and increase his vote further, and furthermore party supporters might refuse to follow the cards (especially given Tasmania's tradition of cross-party-lines voting in state elections).  He would have real chances to just get home even with all parties targeting him.  But it is worth keeping an eye on as evidence that the retention of Denison, while likely, is not yet a sure thing.

Overall, this Denison result suggests that at the moment Denison left-wingers are quite annoyed with Labor and the Greens, and are not inclined to vote for a good candidate from what they see as a bad party.  And in the apparent slump in the Labor primary here (even Jonathan Jackson got 35) we can see the signs of a government that isn't cutting through to either of the very different voter bases north and south of the so-called "flannelette curtain". (For those unfamiliar with that local term replete with lashings of dated anti-bogan snobbery, it refers to the divide between inner-suburban Hobart and working-class Glenorchy, that in 2010 perfectly demarcated booths won on 2PP by Wilkie over Labor and booths won by Labor over Wilkie.)  To the northern suburbanites this is a government that has sold out to the Greens to the detriment of job prospects, and to the proverbial latte-sippers  it is not socially progressive enough.

As for the Greens, while I don't believe their vote is quite that low, it does show that their rather negative tactics of continually attacking Wilkie over minutiae of how many times he has spoken on bills, said positive things about Tony Abbott, and so on are not accomplishing anything, just as personal attacks on Jim Wilkinson by forces linked to the Greens did nothing in the LegCo poll for Nelson.

More of this was on display at the Denison Debate (linked above) with Reynolds unsoundly claiming that Wilkie would be required in a hung parliament to give confidence to one side or the other so that a party could tell the Governor-General it had confidence and supply.  The answer, of course, is that confidence is determined on the floor of the parliament and that the Governor-General will allow for it to be tested if uncertain - but why are we even canvassing the question in the first place?

It's obvious enough that a hung parliament is an extremely unlikely outcome of the upcoming election - what Denison voters really need to know is who will deliver the best for a left-leaning electorate in a right-wing parliament.  I'm doubtful Labor and the Greens will win back many votes from Wilkie while they continue playing their politics in the fairylands of hung parliaments and Labor majorities.

The poll result is not greatly different to a previous ReachTEL conducted almost a year ago.

Senate Implications

These are covered in an update in my Senate Race article.  In short, as dire as this polling is for Labor and the Greens, it doesn't change the Senate projection picture greatly.

Leadership Implications

Yes, I know, the leadership thing was supposed to be dead following the non-challenge in March.  But there is certainly, again, a lot of talk about it, although not yet a lot of verifiable action.  And it's very likely this poll especially will add fuel to the fires on that count.

This poll provides strong evidence that if Julia Gillard remains Prime Minister, Labor is very likely to lose at least three Tasmanian seats and possibly the lot.  It is consistent with (and more important than, because internal polls are less trustworthy) recent reports that "Party figures anticipate the loss of all seats in Tasmania".   All the ifs and buts that can be raised against this (or any other) poll count for little because the seat margins are so wide. Even leadership change is probably not a big enough if/but to counter the size of apparent Coalition leads in Bass and Braddon.  If a switch to Kevin Rudd (or another leader who was well received following the switch) went reasonably smoothly would secure Franklin and might save Lyons.  That said, concerning the latter, I am sceptical that the Rudd bounces shown in hypothetical polling would last at full strength all the way to polling day, and a lot of the issues affecting Labor's standing in Tasmania are not leadership-driven.

Footnote: Does ReachTEL Have A House Effect?

The dire nature of this polling for Labor has raised questions about whether the poll can be entirely trusted.  The following are some points I think are relevant in considering whether ReachTEL has any kind of "house effect" that might affect the results:

* Newspoll and ReachTEL tracking in Victoria has been very similar.

* Two Seven News Federal ReachTELs with large sample sizes have both produced results that were more Coalition-friendly than the average of other polling at the time.  A federal ReachTEL on 12 April was 57:43 and one on 3 May was 58:42.  At the same time the Mark the Ballot aggregate for other pollsters (without reference to prior election results) had the Coalition around the mid-55s and low-55s respectively (just looking at the graphs) suggesting that in those two polls ReachTEL was about two points more Coalition-favourable than other pollsters. 

* A ReachTEL of the Legislative Council seat of Nelson proved an extremely accurate portent of the result, allowing for vagaries involving some candidates not being technically of a given party in what was a very party-polarised campaign on a "shadow" basis.

* Two ReachTEL state polls I have seen from last year (one commissioned by the Liberal Party and publicly released, one commissioned by an activist group that is no friend of the Liberals) had similar patterns to my corrected interpretation of EMRS state polling.

It is too early to conclude ReachTEL federal results have any skew to the Coalition based on a mere two national polls, but even if it is assumed there is a skew of about two points in the current results it changes very little.  Labor still heavily loses Bass, Braddon and Lyons and Franklin remains too close to call (but favouritism would then switch to Labor).

Given that past Tasmanian federal polling has mainly been conducted by EMRS which has displayed massive undecided rates and rather large house effects (such as a persistent lean of around 4 points to the Greens in headline rates, and largely missing the Wilkie surge) this is very likely as good as we are going to get, at least until very close to the election.

Update (19 June): The Market Isn't Buying It

I have been keeping a vague eye on seat betting movements in the various seats since the poll was released.  Prices that follow are rough averages. In Bass the Coalition has plunged to around $1.06 and in Braddon to around $1.27.  Still there is a discrepancy between the two that seems to have not much logic behind it beyond that Bass has now had two published polls to Braddon's one.  In Denison Andrew Wilkie has firmed to about $1.22.  But in Lyons and Franklin, poll results showing Labor to be heavily losing and dicey respectively have left the party still warm favourite ($1.50 and $1.42 respectively).

There are a few things that may be going on here.  Firstly, it's an uncertain environment to bet money in; if you throw money on the Liberals in Franklin thinking $2.45 is a fairly good deal, and Kevin Rudd is installed and lifts Labor's vote by five points nationally, you'll be feeling pretty sorry about that.  So possibly these odds will change if this fortnight passes without an ALP leadership change.

Secondly, in the case of Lyons, it's always possible that this is an extreme sample and that insider betting is occurring on the basis of unpublished samples that show it to be closer.  The ReachTEL poll may be by far the largest source of public polling but there may still be distrust in robopolling and some belief that it is yet to earn its stripes in Australia. 

Thirdly, it may be that there are punters out there who just don't believe that seats on 12.3% margins fall and are not inclined to believe that Lyons is in great danger even on the basis of a 500-ish sample showing a 59-41 deficit.  If so I would suggest looking at the history books: for instance, in 1975 a statewide swing of 11.6% in Tasmania claimed the seat of Franklin (12.9% margin) with a 14.7% swing in that seat.  And that was even without the benefits of an unpopular state government, indeed during a time of state-level Labor dominance.

Another possibility is that seat betting volumes are still very low and the prices have been slow updating following the poll.

Currently, national seat total betting is projecting a Labor seat tally of around 50.  This is very similar to the totals of current polling-based modelling (eg Bludgertrack 47 seats, Mark the Ballot 51).  Basically the seat total market is projecting, on average, that Labor's polling is not going to improve, or that if it does improve, the party will disproportionately lose seats on the basis of the distribution of swings.

The individual seat markets also don't project any improvement; Labor is currently favourite (cross-agency average) in 49 of them.   But the seat markets currently think Labor will lose everything on a 2PP (vs Coalition) margin of below 5.2, and hold everything on a margin above 7.5.  There is a crossover zone of just 13 seats in the 5.2-7.5 range of which Labor is currently favourite in eight (some very narrowly.).

If the average swing nationwide really is over 5% then some seats on margins well above 7.5%  are certain to fall.  The difficulty is picking which ones in advance, and it is especially acute because there are a lot of seats where the party is polling badly.  Among those above 7.5%, not only Lyons but also Blaxland (twice), Isaacs, Chifley and Franklin have all been the subject of public polling that has shown the incumbent trailing.  None of these, however, have been nearly as heavy as the Lyons sample.

Update (June 25): More On Possible House Effects

The recent JWS Research poll projects a Tasmanian swing of 10.6% (MOE about 3.3%) in the seats of Bass, Braddon and Franklin combined.  The recent ReachTEL with a combined sample in these seats of over 1500 found an average swing in these seats of 16.4% (MOE about 1.8%) and an even larger swing in Lyons (a staggering figure of 21.3% - though I don't think anyone really believes it is quite that big).  If it is assumed that the sampling in the three Tasmanian seats by JWS was roughly even (which is not explicitly stated) then there is a statistically significant difference between the small JWS sample and the larger ReachTEL one.  It's notable in this context that findings concerning the swing in Tasmania prior to the ReachTEL giant were generally up to six points lower than the ReachTEL result, and with betting markets continuing to react coolly to the projection of Dick Adams losing Lyons by a whopping nine points it is timely to examine the past record of ReachTEL seat polls in more detail.

I thought I'd have a look at ReachTEL's primary vote performance in the Queensland state election 2012, using the company's last published poll in each seat.  I should caution (and will use bold to do so) that this is not a valid way to prove a house effect exists.  There are two reasons such an exercise will be inconclusive - firstly, most of the seat polls were taken well out from the election and hence comparisons do not allow for changes in the state vote picture over that time.  Secondly, parties make strategic decisions to go hard in particular seats and go home in others, and these decisions can in turn be influenced by polling.  If a party has polling showing that it is winning a seat 65-35, it may be inclined to channel resources elsewhere.  (One would think this would also apply to the losing side, but that doesn't necessarily follow.)  Anyway here's the spreadsheet for those seats:


The figure "moe<" refers to the margin of error for a 50% result.  The margin of error for a lower result is smaller, but given that so many seats are being discussed at once, not too much should be made of the Greens polling 3 points over the poll in a seat where their margin of error was 2.4 (or whatever).  Exceptionally good results compared to the poll are shown in blue and exceptionally low in red.  There are a few minor irregularities here - KAP had no polled reading in one seat but contested it, and was polled in one seat but did not contest it.  There was also a seat with no Others candidate. 

However, the overall pattern is clear - except for the one seat poll taken just prior to the election (which was also the largest) the polls tended to understate the Labor vote and the KAP vote, and overstate the LNP's, compared to the final result.  This is the reverse of the pattern for Newspoll in the same period.  Some of the seat polls were very well replicated, but there was a big discrepancy in Ferny Grove, with the actual primary vote between LNP and ALP seventeen points closer off a sample of 370 than the ReachTEL seat poll.  The equivalent to that discrepancy in Ferny Grove, considering sample size issues, would be Dick Adams closing the gap by 14 points, which all other things being equal would result in a 48:52 loss instead of a 41:59 one. That's a movement outside the maximum MOE of the current poll, but the use of scaling in polling means that notional margin of error figures are often too conservative anyway - especially when dealing with robopolls.

My view is that Labor will probably lose Lyons if they go to the election with a national 2PP of 45 or worse (which I currently think is about even money assuming Gillard is not replaced), but I expect it to be a lot closer than the ReachTEL sample if so.  The analysis I've done of the Queensland results makes me a bit more doubtful of the predictive robustness of the single ReachTEL result, both on the grounds of possible house effect and because of the wide variation between some of the admittedly smaller Queensland samples and finer arguments.  However, Lyons was looking a bit shaky to me even before the ReachTEL, and nine points is nine points - it's a big gap to explain away. 

Footnote: I received some feedback on the issue of some of the Queensland seat polls being older than three months and hence not strictly a valid comparison for a poll taken three months out.  I don't think this is a big deal because the average age of all the polls included is around three months and some are actually very much younger, and there was not much change in the state 2PP over this time.  However, if we consider only those polls younger than three months, which is just the top five, the two-party gap drops considerably (ALP vote average +1.9 LNP +0.7, for a gap of 1.2 points compared to 5) but there remain rather large differences between the polled vote and the outcome in two of the five seats. 

No comments:

Post a Comment