Sunday, September 15, 2013

Modelling the Tasmanian State Election off the Federal Election

Note Re Tas Senate Counting: An erroneous claim was published today (18 Sep) on Tasmanian Times that the Sex Party has won the last Tasmanian seat.  This is untrue (at this stage) - the count is still going, and the Sex Party would not win on current figures - the ABC Calculator is a rough model only.  Senate count intricacies can be followed in the post below.


Advance Summary

1. Claims have been made that the result of the federal election, if repeated at the state election, would produce a hung parliament with the Palmer United Party holding the balance of power.

2. These claims are incorrect.  If the federal election results were repeated at the state election, the result would be a Liberal majority government with 13 seats.

3. The claims are also irrelevant as the Liberal Party has been tracking at a much higher level in state polling than in federal polls taken by the same pollster at the same time, and it is necessary to adjust for this.

4. A model based off the federal election results but with appropriate correction for state/federal polling differences projects a Liberal majority government with 13-14 Liberal, 8-10 Labor and 2-3 Green seats.  It does not at this stage project any seats for other parties.

5.  Despite this, the Palmer United Party has laid a strong base in Tasmania with its result in the Tasmanian election.  If it can keep its act together up to the state poll then it is capable of increasing its vote share and winning seats.

6. A very large increase in PUP vote share would be necessary on current polling to prevent the Liberals winning majority government.  It seems that actually the party most threatened if PUP performs competitively could be the Greens.


Late counting from the federal election is still very much alive (Tas Senate , House of Reps) but many things are clear enough for useful conclusions to be drawn.  In the case of Tasmania, the performance of the Palmer United Party (PUP) has received a lot of attention as a possible portent for the state election.  PUP polled about 6% in the House of Representatives, including nearly ten percent in Braddon, and are currently running at over 6.5% in the Senate.

The theory goes that if Franklin's five seats split 2-2-1 (Liberal-Labor-Green) then PUP might win a seat in Braddon at the state election and that the Liberals might be deprived of a majority, assuming the Federal results repeated at the state election.

But Would PUP Win The Braddon Seat?

I'll look at that Braddon theory on that basis critically first.  The Federal election results are running at: Liberal 46.8%, ALP 37.7%, PUP 9.4%, Green 5%, Rise Up Australia 1%.  That's 2.81, 2.26. 0.56, 0.3 and 0.06 quotas respectively.  The idea is that PUP would win the fifth seat, but on those figures, they would not.  They would need a preference split of about 70:30 from Labor and the Greens if all voters expressed a preference, and even that could be a tallish order (as those who are following the Fairfax count would know).  But the real problem is that the Tasmanian House of Assembly has semi-optional preferencing.  A voter can stop once they have numbered five squares, and in practice about half the Labor vote and about a third of the Green vote does this.  So in practice the flow PUP would need to catch the Liberals would be above 87%; this wouldn't happen.   The result of the 2013 federal election results, repeated at the state election exactly, would be three Liberal and two Labor.  Ditto Bass, ditto Lyons.  If the 2013 federal election results were repeated exactly at the state election (assuming 2-2-1 in Denison, where the Wilkie factor complicates things) then the Liberals would still win 13 seats and a majority.  The claim that the results would produce a hung parliament is not true, and if the results were reproduced exactly at state level then the Palmer United Party would not actually win a seat.

Not only is the claim that the federal results point to a hung parliament if applied at state level not true; it is also not relevant.  It's been obvious both from results in 2010 and from what polling has been available since that Labor is tracking significantly worse at state than at federal level.  A part of this, at the most recent election, was incumbency: Labor had sitting members in four of the five state seats.  Another part is that the federal election saw a diluted backlash against issues at state level.  Some voters punished the party federally for such issues, some did not.  Voters who voted for the federal Liberals won't be voting for state Labor in any numbers, but some voters who voted for federal Labor will vote for the state Liberals - especially if majority government is clearly on the line.

Nonetheless, we have learned useful things from the federal election.

ReachTEL Accuracy

The first concerns the accuracy of ReachTEL polling.  ReachTEL polling was the only public source of detailed federal polling of Tasmanian seats, and has also become a major source of state polling (EMRS conducted state polls in February and May but none since.)

ReachTEL polling of the three northern electorates (late July), the whole state (Aug 22) and Bass (Sep 3) painted a very consistent picture and correctly forecast the fate of all five Tasmanian House of Representatives seats.  ReachTEL data were especially valuable in the case of Lyons, which fell with the largest swing in conventional ALP-Coalition seats in the nation (14%).  Even with other polling evidence that the swing in Tasmania should be at least ten points, forecasting the fall of Lyons would have been difficult without ReachTEL data repeatedly saying the seat was falling, and saying so by a margin difficult to explain away by the pollster being Coalition-skewed.

Nonetheless, ReachTEL figures were repeatedly criticised as too harsh to Labor by ALP state secretary John Dowling, and it's clear he had a point.  During the campaign, ReachTEL had Bass at Labor 2PPs of 42 then 41 (it was eventually about 46), Braddon at 43 (eventually about 47.5), Lyons at 44 (eventually about 48.5) and Franklin at 50.6 (eventually about 55).   The figures for the given seats will move around in post-counting but the difference in all these seats was around 4.5 points in favour of the Coalition.

Looking at ReachTEL's results in other federal seats a similar pattern is apparent.  ReachTEL's results were very accurate (unlike those of some other pollsters) in the Queensland seats of Griffith and Forde, and they also did well in the Victorian seats of Deakin and Corangamite.  But in the NSW seats of Kingsford-Smith, McMahon, Blaxland and Bennelong, ReachTEL's figures were even further out from the election result than in Tasmania, and the first two seats were retained (McMahon comfortably).

ReachTEL weren't alone in this: extreme seat polling was also recorded by both JWS and Lonergan in both Forde and Lindsay, JWS in Lilley, Forde and McEwen (though the latter was a useful pointer that the seat, still in the balance, was capable of falling), Newspoll and Lonergan in Griffith, and probably Newspoll in some of its aggregate polling.  None of the "way gone" seats actually fell with margins exceeding 55:45.  It can always be argued that such seat polling (when taken a few weeks out) captures trouble spots that parties then bombard with largess, narrowing margins by the election. But in the case of Bass, when this kind of thing happens four days from the election, that argument is harder to sustain.  Indeed given that Labor's national position slumped over the campaign, seat polls taken a few weeks from the election should have been strong for Labor and not weak.

So although ReachTEL's national polling did not display much of a house effect at all in the campaign, it seems that for the Tasmanian federal seats it produced Coalition-leaning results, and it is reasonable to suspect that state polling by the same method does the same thing.   Since the state's other regular pollster (EMRS) also badly underestimates the Labor vote, this makes modelling the state Labor vote tricky, save to say the party is clearly losing.

Fortunately, we now have three matched federal/state ReachTEL samples for the three northern electorates this year, and two for Franklin and Denison (though the last are of little use thanks to the Wilkie factor). 

The Greens In All Sorts Of Bother

Although the Greens retained their Senate seat, this was an awful federal election for them in all other resects in Tasmania.  They lost 42% of their House of Reps vote in Franklin, 51% in Bass, 52% in Lyons, 58% in Braddon and 59% in Denison.  The latter figure was hugely affected by loss of votes to Andrew Wilkie; their Senate decline in Denison was "only" 32%.  Overall the Greens lost 44% of their Senate vote statewide.  Similar declines at state level would see the party reduced to about 13% of the state vote.

PUP Are A Factor in Tasmania

Palmer United polled 9.4% in Braddon (high-profile candidate Kevin Morgan had run in the Legislative Council election, but they did well in the Senate there too), 6.9% in Lyons, 6.1% in Franklin, 5.4% in Bass and 2.4% in Denison.  The Denison vote was also, of course, Wilkie-deflated; they got 4.3% there in the Senate.  These are good results for a party that has never contested in the state before.  It may well be that PUP prove to be a flash in the pan, but if they stick around and keep their act together, a well-resourced campaign in the state has potential for similar results.  It's possible, as Will Hodgman suggests, that voters will not take PUP seriously when majority government is on the line.  My own view of the PUP surge federally is that the party boomed when voters realised the election was over as a contest and that they no longer had to take the possibility of another hung parliament seriously.  But PUP are well resourced and there are plenty of Tasmanians who are fed up with all of Labor, Liberal and Green and looking for any even halfway-sane fourth-party alternative.

The State Election Modelled Off the Federal Election

Even assuming that ReachTEL polls in Tasmania have had a skew in favour of the Coalition (as the federal results suggest they have), it is possible to use those polls and the federal result to model the state result.  This is done by assuming that ReachTEL polls asked at the same time to the same respondent have the same skew (it is not even necessary to judge what that skew is) and that therefore if the state government has been tracking X points below the federal government in such polling, then it would likely perform X points below in the state election.  So for each matching result I have just found the difference between the state and federal outcomes, as polled by ReachTEL for the parties, to model what an election might look like assuming the PUP vote is about the same as in the federal election.

Here then is that model.  I've weighted the August figures at 1.5 times the others, because they are more recent and taken closer to the federal election that is used as a base (Click for larger version).

The model projects a state result of 49.2% Liberal, 28.7% Labor, 13.4% Green and 8.7% Other (mostly PUP).  This compares to my most recent three-party aggregate which had 55.3% Liberal, 29.2% Labor and 15.5% Green on a three-party-preferred basis only.  There are two things going on here - drift from all parties to Other (because Others are not in the aggregate), and drift from Liberal to Labor (because the federal election result says that the existing polling favoured the Liberals.)

What does this model say?  Firstly it says that even with PUP polling the same vote as at the federal election, the Liberals still win the election outright easily.  The Liberals comfortably win 13 seats and a 14th in Franklin is line-ball.  The Liberals are well clear of their quotas for their minimal 13 seats and can even afford to drop below three quotas in some electorates without missing the third seat.  Labor wins eight to ten, the Greens two to three, and nobody else wins anything even if it is assumed that all the Others votes go to Palmer United.

With those exact votes, the profiles of Lara Giddings and David O' Byrne would win two seats for Labor in Franklin, but it's quite possible focus on the issue of majority government would result in the Liberals doing better than the model implies.  It's also important to consider that in Franklin, candidate factors were important in the federal election - Bernadette Black had a gaffe-prone campaign and received a much smaller swing than the northern Liberals, while Julie Collins was recontesting as a Cabinet minister.  That affects the Franklin projection in Labor's favour.

Secondly the model doesn't project any exotic four-seat batches for the Liberals in Bass, Braddon or Lyons (unlike some of the ReachTEL and EMRS state polling).  This is no surprise, because if we assume ReachTEL is indeed four or so points generous to the Liberals then those situations also disappear from modelling based on those polls. 

Thirdly, the model is dire for the Greens.  They retain Denison, Franklin and possibly Bass and lose Lyons and Braddon.  It may seem hard to believe that a model could project a vote of only nine percent for the Greens in Lyons at a state election (compared to the 13 in my aggregate), and it's a figure that may have a bit of polling error influencing it, or that may also suggest that the Green vote in Lyons is soft and at risk of being picked off by PUP or other fourth parties.  

Indeed, it could be not the Liberals but the Greens who would suffer most in seat terms from the emergence of PUP at state level, because PUP voters would effectively split the third-party vote, but many of them would exhaust their votes at the end of the PUP ticket and not go on to preference the Greens.

Although this model does not show PUP winning a single seat (as was the fate of the Tasmania First party in 1998, which polled 5.1% statewide including 9.9% in Lyons, and yet won nothing) it is far from the end of the story.  If the party can impress the voters over the next six months then it has the potential to build on its vote share from the federal poll and win seats at the state election.  Whether it does so or subsides and has little impact remains to be seen, but it does seem that the party is serious and will be giving the state election a shake.  

As a result of this modelling exercise and the Greens' dire vote in Tasmania at the federal poll, I have docked the Greens their fourth seat in my Tasmanian projection.  The model returns to 14-8-3; I will give consideration to whether Palmer United should be projected with any seats if/when we start to see some state polling that supports giving them some.  At this stage there is no evidence to justify doing so, but that could change.

1 comment:

  1. Good interview on TAS Mornings today. I can't see there being final results until next Monday, as the postal votes are still open until last mail Friday, after which those votes will still need fresh scrutiny and the BTL system entry down at the state office.


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.