Saturday, March 17, 2018

2018 SA Election Late Polls And Other Comments

SA: Newspoll and ReachTEL 34-31 on primary votes to Liberal
No predicted winner - too close to call

On Saturday night I will be attempting to live-comment the SA state election and the federal Batman by-election at the same time starting from 6:30.  Really this shouldn't be too hard, since Batman is just one seat, so I hope it will be useful.  They will be on separate threads and I will be trying to give each about equal attention to start with, though if Batman can be called quickly I will wind it down and switch to focusing purely on South Australia.

A Few Quick Unpleasantries ...

Rant warning!

I have made an SA-election-related colour change to the boundary of my site, putting up SA-BEST colours as a personal protest against the Labor Party's decision to preference Cory Bernardi's Australian Conservatives third on their Upper House how-to-vote card (the Liberals of course are preferencing them second).  This should not be taken as an endorsement of SA-BEST as such (indeed I suggest South Australians vote how they like and ignore my foreign interference), but it is an endorsement of ripping the Labor upper-house how-to-vote card into tiny pieces of confetti and then posting it to SA ALP HQ with an appropriate message.

I am, as always, disgusted that a supposedly progressive party anywhere in Australia would preference the religious right in this manner.  It seems that some sections of Labor never learn and never stop playing these clueless whatever-it-takes games no matter how many times the religious minnows win them (Steven Fielding, Bob Day etc etc.)  The payoff is Conservatives preferences in Labor's marginal seat of Lee, and split tickets in Light and Newlands, but whether anyone minded to vote Conservative will be fooled by all this or even follow the card at all we will see.

Fortunately the Upper House preference recommendation is likely to be very weakly followed, based on experience of the 2016 Senate election (at which really only the Liberal Party had strong how-to-vote follow rates, and those parties that played these stupid hack games generally found nobody much followed their card). But for an even remotely progressive party - or even a centre party as SA Labor would probably claim to be - to endorse a religious right party like Bernardi's is wrong.

Righto, back to psephology ...

Final Polls

The Sky News ReachTEL has come out with primaries of Liberal 34 ALP 31 SA-BEST 16 Others 19. Frustratingly, the Others figure apparently includes "undecided" (as irregularly defined by ReachTEL), but the undecided rate isn't yet reported, and there is a 2PP figure stated (but apparently only for the undecided voters) - typical garbled and useless Sky poll reporting.   This ReachTEL might be taken as implying that SA-BEST are continuing to slide dramatically, but it's also notable that SA-BEST polled poorly in the Climate Council state ReachTEL at the end of January.  There after accounting for the undecided voters, the numbers were Liberal 36.4 Labor 28.4 SA-BEST 19.2 Other 15.9  -  much lower for SAB than others around the same time.

The Newspoll has also come out - at the same time so we can't accuse them of herding - and it also has a low result for SA-BEST, with Liberal 34 ALP 31 SA-BEST 17 Greens 8 (Others 10).  The Newspoll is adjusted for SA-BEST not running in certain seats (so it is compatible with a statewide vote of, say, 22) but it is unclear whether the ReachTEL is. At least we know that for Newspoll the undecideds have been removed.

My model for SA-BEST still has them able to win 2-3 seats off Newspoll's adjusted 17% (it has them ahead in Kavel, and Heysen and competitive in Mawson, Finniss and Giles) but as noted before these models are a little prone to overestimate third-party success.  And if the ReachTEL is not adjusted then the picture becomes even bleaker after aggregating the two (though I'll note the old Newspoll was much more accurate than ReachTEL in 2014.) On this polling the party is looking at a handful of seats at best, perhaps none - so not such a different outlook to One Nation in Queensland. It might not be so chaotic a night as expected.

Upper House

I'll be commenting on the Upper House results during the postcount and possibly on election night depending on whether there's enough counting progress to say anything useful.  11 seats (half the Upper House) are up for grabs so the quota is 8.33%.  The system is now similar to NSW's - optional preferential above the line, so the final seats will be falling on a lot less than a full quota.  The trap here is to remember that SA-BEST can poll Upper House votes even in seats where they're not contesting the Lower House (though perhaps that will deflate them slightly in those seats).  Based on the recent Newspoll a result along the lines of 3 Liberal, 3 Labor, 3 SA-BEST, 1 Green, 1 Conservatives would make sense, but on tonight's ReachTEL it might be more like 4 Liberal and 2 SA-BEST.  The seats being vacated are 4 Liberal, 4 Labor, Kelly Vincent (Dignity), 1 Green, 1 Conservative.  Under the new system it will be difficult for other parties to win (at least if their vote shares are anything like as low as last time), but if SA-BEST has indeed crashed to the high teens it might be possible for a micro-party (possibly Dignity) to get up from a few percent.

Lower House How-To-Votes

How-to-vote cards can be downloaded from the ECSA website, and Peter Brent has also prepared a very useful summary of who the major parties are preferencing out of the other major party and SA-BEST in each.  Labor are splitting their preferences with SA-BEST ahead of Liberal in 18 seats, Liberal ahead of SA-BEST in 17 and a split ticket in Hartley, where Nick Xenophon is the SA-BEST candidate.  The Liberals have preferenced SA-BEST over Labor in every seat except for preferencing Labor in Newland and Unley and a split ticket in Hartley.

William Bowe (paywalled) has noted that the SA Greens have for some reason preferenced the Liberals ahead of SA-BEST in 15 of the latter's 36 contested seats, however the SA Green vote is low and Green how-to-votes struggle to alter the flow of Green preferences by even 10%, so the impact of this should be small.  What I have noticed is that in the seats that I assessed as competitive for SA-BEST (modelling approach described loosely here - and I'm counting anything where that model says SA-B is within 3% of making the final two and within 3% of winning the two-candidate race), SA-BEST nearly always has the preference of the party likely to be excluded first.  The Labor preference decision has mostly hit SAB in seats where it is not competitive anyway.  So if SA-BEST could get the statewide low-20s percent result of the recent Newspoll then the party could conceivably win as many as a dozen seats.  The vibe of the thing in commentary in the commentariat is overwhelmingly that this won't be happening.

It isn't really clear what the SA-BEST candidate for Newland, Rajini Vasan, has done wrong to deserve being put last by both the majors, as searching online I could not find evidence of anything controversial involving her.  However, the ballot order may have something to do with it:

Both parties have adopted designs that allow their voters to easily "donkey" most of the ballot, presumably to minimise the risk of voter error (although SA has a partial savings provision in that regard).  These kinds of semi-donkey how to vote card orders are common, and in my view they provide a rather strong argument for using rotating ballot papers.  Genuine donkey voting in single-member electorates with party names shown on the ballot is typically a fraction of 1%, but the proportion of voters following major party HTVs is well into double figures (even in SA where the major party vote will be so low).  So, if a major party preference decision between the other major and a third party contender is based on ballot order, that's a potentially massive random advantage.

(A cheaper but in several ways less practical alternative to rotating ballots would be to require parties to submit their HTV order prior to the draw for positions.)

Seat Polls Vs State Polls

This is another election where the summary position from seat polls is different from the picture of the statewide polling.  With the exception of the seat of Taylor (an ALP vacancy) the SA-BEST vote in seat polls has been generally worse than would be expected after adjusting the NXT Senate vote to match the Newspoll.  In cases it is much worse, such as Mawson where the party should be in the low 30s, not the 20% of the recent Galaxy.

It is tempting to aggregate the seat polls and use that aggregate to cover for the paucity of state polls, but past experience should make us very wary of doing this.  At the 2013 federal election, for instance, seat polls displayed large skews to the Coalition compared to the national polls, but the national polls were right and the seat polls were wrong.

So what might happen?

It looks like SA-BEST are going to flop, and we may even get an outright winner, but even with these late polls I can't tell which way this is going to go.  The polls make it too close to call.

On the primary figures given, it really comes down to the SA-BEST preferences.  It baffles me that pollsters haven't even attempted a state 2PP with the evidence increasing that SA-BEST are in third place and will be excluded in the bulk of seats.  It would be necessary to use respondent preferencing, which is unreliable, but SA-BEST have open how-to-votes everywhere, so perhaps for them respondent preferencing would actually work.

If the SA-BEST preferences break as per the 2016 federal election, then on the current Newspoll primaries Labor should win about 51.5% 2PP (maybe 51% if we assume Conservative preferences are a bit worse than Family First) and on that basis would be likely to hold government.  For 51% 2PP, Labor wins the 2PP on average in about 25-26 seats and should at least be able to form minority government after dropping Florey and, say, one to SA-BEST.

If the SA-BEST preferences break more like 50-50, then that's a Labor 2PP around 49.5-50.  And that's the zone where either side can win with more or less equal chances, though a hung parliament becomes rather likely.

If the SA-BEST preferences break to the Liberals, then Labor should lose.

There is a fascinating Newspoll question which asks voters which way Nick Xenophon should jump if he gets the balance of power.  He ought to worry about actually getting any seats first, but 52% say he should support the Liberals while only 28% (lower even than the Labor primary - what?) say he should support Labor.  That might be taken as an indication that preference flows will be carrying a time's-up factor, but even still it is a strange result.  Another it's-time result comes from ReachTEL which says "After 16 years of Labor in office SA needs a change of government" (59% agree 27% disagree).  The problem with this question is that it presents one side of the argument in isolation, which is not how voters think about elections.  If the question was "The South Australian Liberals are not ready to govern yet" that might also get a positive response.

I should again mention a few of the big polling picture stories that I always follow.

Firstly, my pet federal drag theory says that governments that are opposite to the party in power in Canberra only lose about 15% of the time.

Secondly, however, incumbency effects (both state per the same link and federal) say that governments that have been in power longer are more likely to lose, especially these days.   Somewhat cautiously, I previously assessed the SA government's historic chance of survival on this basis at 50-50.

Thirdly, there's the history that unpopular State Premiers are toast. Jay Weatherill's last two Newspolls have had netsats of -21 and -20.  Only one state Premier has ever won after recording that or worse at any stage of a term, and that was much further out from the election than this.  Weatherill's small lead over Steven Marshall in a two-way question (5 points), also puts him in an uncomfortable position (see historic graph here) though David Bartlett (Tas) was the one winner from a very similar position.  However, when we look at the polling of those unpopular state Premiers who made it to elections, it was usually obvious that they were going to lose.  If a Premier is unpopular but his party is competitive, then isn't this just like the 1993 federal election?

My modelling provides no basis for tipping a winner in this election (it comes out on average with 21-22 seats for each major party and a crossbench that might do anything), and I think that predicting someone to win when there's no basis for having the foggiest is pretty useless.  (If the prediction is wrong it hasn't helped the public, and if the prediction is correct then it looks like you knew what you were doing when you didn't, which is worse.)  The last SA election was a genuine surprise but there were always reasons for thinking it was closer than the betting markets said.  This one appears to be very close to a genuine tossup - either an ancient and unpopular government continues or a struggling Opposition wins while the same party is in power federally - so it should be a fascinating night to see which way it falls!


  1. Is there any information available on how many people actually follow how to vote cards? Personally I've never even been offered one but looking like something that you wouldn't take home to meet mum probably has a lot to do with that.

    1. For lower house elections major party HTVs are usually followed by about 40% of each party's voters, significant minor parties get around 10% and micros get practically nothing because they don't have enough people to hand out and are usually attracting protest votes or votes for the candidate because the voter knows them anyway. For the Senate election it was somewhat weaker than all that (especially for Labor which had a strict follow rate of about 10%.)

  2. one sentence need completing... "At least we know that for Newspoll the undecideds have been"

    1. Ta, a common gremlin species in this place and fixed.

  3. Poor old Cottesloe. Personally I am hoping something huge has been going on there while no one was looking, but I doubt it.

    Looking forward to the dual coverage tonight!

  4. Voted in Newland. The Liberal candidate 'should' win and seems to be a very pleasant individual. The sitting Labor MP will not get some of the preferences, that he would be expecting. Only the second time I have ever preferenced a Lib. Kenyon is probably the most right-wing Catholic, Labor MP ever to hold a state seat. Even wanted to vote against state recognition of de facto relationships for same-sex couples. Was good to be able to vote above the line in the Legislative Council and only number the parties of one's choice for the first time. That was easy, only 3 parties numbered. Agree with your views concerning Labor and their Cory deal. The old 'Family First' types were out handing cards en masse as per usual. Hope they bomb big-time.

  5. Rough preferences alp/lib 50/50

  6. I've just had a run through the 2017 Queensland election results, substituting PHON for SAB and calculating how many seats they would have won, given the preference flows expected in the SA election. On that basis I get PHON having won 13 seats and narrowly missing another 10. For example Burnett was never considered a chance for ONP. The result was LNP/PHON/ALP/GRN = 42/27/25/6. As it turned out Green preferences put the ALP ahead of PHON and the final distribution was between LNP and ALP. However on the flows likely to be seen tomorrow night in SA, this seat would likely have been a PHON (SAB) win as green preferences would not have been sufficient to propel the ALP ahead if PHON were subsitituted for SAB, then ALP preferences would have given the seat to PHON, rather than the LNP.

    There are half the number of seats in SA, so this would equate to 6.5 seats for PHON was their positioning centrist like the SAB, rather than far-right.

    There are also more factors to consider:
    1. PHON had 18% (adjusted for uncontested seats), whilst the Newspoll figure for SAB would be approx 20%
    2. The effective 3PP figure for SAB will be higher still (comparative to PHON) as they will receive a higher portion of minor party (esp Green) preferences than Phon did. Instead of say 10% of Green preferences for PHON, SAB will probably get 30-40%. So lets add another 2% to the effective SAB statewide vote.
    3. SAB is running in 77% of seats, whilst PHON ran in only 66%. This means additional potential for SAB to win seats.
    4. Tactical voting is likely to be much higher in SA as the ideological differences between the main parties and SAB are much less. Therefore Liberal voters in the northern suburbs, say a seat like Enfield, might switch to SAB as they realise its the best chance to defeat John Rau. This will propel SAB ahead of the Liberals into second, which would not show up in the models, as they don't account for tactical voting. The same is likely to happen in the rural seats surrounding Adelaide and in places like Davenport, where everyone knows the ALP has no chance, so left-leaning voters might plump for SAB, even if they identify (in the polls) as Greens or ALP supporters.

    I've indicated, SAB would have won 6.5 seats, based on Phon's Qld vote. Add another seat to adjust for point 3 above makes 7.5. Whats an effective extra 4% of the statewide vote (points 1 & 2) worth? Perhaps the additional 8-10 seats that PHON (SAB) narrowly lost on my substitution exercise. (halved to 5 because there are half as many seats in SA). That would put SAB on 12 SA seats, based on PHON's Queensland showing, certainly a lot better than what PHON managed. And this is before taking tactical voting into account.

  7. I have declined a comment just received (not from any of the above posters) because I don't want to be sued for defamation, and also for an unnecessary excess of exclamation marks and capital letters.

    I can only accept or reject comments in full so when in doubt will reject comments entirely.

  8. I put the Bernadi lune train last, ignoring my ALP HTV. I agree with you.