(Changes from pre-election/notional; Labor gained two seats from Liberal during 2022-6 term)
The 2026 South Australian lower house was remarkable in so many ways. It makes Queensland 1998 seem almost boring by comparison, except that Queensland 1998 was there first. Maybe all elections are going to be like this now and this soon will not seem so unusual but if that's so my colleagues and I are going to have a very busy time in the future!
All manner of curious things happened here. Finally, someone (Lou Nicholson in Finniss) won a state or federal seat from fourth on primaries; hooray we have lived to see it. Both majors missed the 2CP in Stuart and Mount Gambier in the first such cases since Nicklin 2001. The Liberal Opposition missed more 2CPs (29) than they made (18) and were outpolled by One Nation (unprecedented) but are still the Opposition. Worse than that they missed nine 3CPs as well and even managed to finish fifth in Port Adelaide and Black - Black being a seat they won at the previous election! And so on. It was obvious this was going to be a very messy election - a little while out I thought how on earth will we ever make a pendulum from THIS - but aspects of it were even more unique than I saw coming.
What this election was not about
This election was not about preferences. It takes its place alongside Victoria 2022 as an election where there has been a high volume of blaming preferential voting for a disproportionate result but preferences had virtually nothing to do with it! The primary vote leader won in 45 of the 47 electorates. Each major party had one seat where they led on primaries (Finniss for the Liberals and Kavel for Labor) but were beaten by an independent. With such low primary vote shares that is bound to happen somewhere.
Much of the nonsense about the impact of preferences came from the One Nation corner, and to be fair in the early stages of the postcount this was exacerbated by confusing wording (or lack thereof) around counting realignments on the ECSA website and by the ABC not quickly calling One Nation's wins. (And with good reason as Narungga which ended up extremely close was called by some sources on the night.)
But ultimately the reason why One Nation won 4/47 seats off their 22.9% is simply that they were only ahead on primaries in that many. They were almost run down by an incredible preference flow to the Liberals in Narungga but survived by 58 votes. Cory Bernardi, showing he is just as good at spreading nonsense about elections as he was at spreading it about human beings getting married, has claimed One Nation should be the opposition because "preferences got some of the uniparty people over the line". In fact only two of the five Liberal wins even had a Labor exclusion and in both of those the Liberal candidate was well ahead of One Nation at the time. This cannot be said for One Nation themselves in Hammond where they fell behind during the distribution and would in fact have lost had the Liberals followed the "uniparty" plot and recommended preferences to Labor. Indeed it's another shameless (or brainless) display by Bernardi to be pushing this "uniparty" nonsense when the fact is that the Liberals tried to help One Nation beat Labor via their how to vote cards and One Nation didn't return the favour.
Estimating 2PPs - is this even worth it?
This section has been rated Wonk Factor 5/5. Beware!
The ECSA is not going to put out an official 2PP and it looks like what we will get instead is a Labor vs Non-Labor figure estimated by Antony Green (who has done a tremendous amount of analysis very quickly here) at 55.7 to Labor. I decided I would have a go at estimating a 2PP and a Labor vs One Nation equivalent for this election anyway, partly because I wanted to have a look at Liberal vs One Nation competitiveness against Labor in certain seats and partly because, well, it was there and somebody has to climb the north face.
The method I used is the same method (which I call the "actual distribution method") that I used for Queensland 2017 among other cases:
- where one of the 2PP sides is excluded prior to the 2CP, follow the exclusion to that point and credit all votes received by each side to that point to their 2PP.
- for votes with other candidates at this point (One Nation for the classic 2PP and Liberals for the alternate, Greens and Independents in both cases) assign those votes on 2PP based on the total of all preferences that flowed from that source in other seats to each of the 2PP sides.
This means that I use partial preference distributions and not just the cases where a party is excluded at the final stage; this partly helps cancel out the tendency of 3CP preference flows to be weaker than the flow from an excluded party's actual voters because of votes coming in from other sources. Sometimes this can produce exaggerated flows in some seats - for instance if Greens preferences split 80-10-10 between Labor, Liberal and One Nation in a seat I will credit this as 88.9% for Labor vs each party, but a voter voting Greens-One Nation isn't that likely to go to Labor next. (A voter voting Greens-Liberal, incidentally, generally will go to Labor over One Nation).
As a sanity check I also considered the following: as we have seen with Hunter, parties do better on preferences from a given source in seats where they perform better, so it could be that One Nation and the Liberals would each do worse on each others' preferences in the seats where they don't receive them than the seats where they do. Therefore, for instance for the classic 2PP, I graphed the ratio (Labor/(Labor+Liberal) at the point where one major was excluded) vs Labor's share of One Nation preferences, and then modelled what that should look like across every seat. In both cases the modelled flow from one conservative party to the other came out between 1 and 2 points weaker than the flow I had already obtained, which is not significant and is about what I would expect based on batched preference flows being weaker than the flows of voters who voted 1 for a party.
Where I have the most reservation about my numbers is that there was a curiously weak flow from excluded independents to Labor vs Liberal. This is partly driven by the exclusion of ex-Liberal Nick McBride whose preferences flowed strongly to the Liberals. But two of the eight independents for whom there are no 2PP preference flows were also ex-Liberals, and the weak pattern extends further than just McBride. I'll adjust my numbers if any more light is shed on these; for instance in Flinders there should eventually be a classic 2PP.
Overall my model has the following preference flows:
* To Labor vs Liberal: One Nation 27.2%, Greens 85.9%, IND 50.8%
* To Labor vs One Nation: Liberal 33.9%, Greens 84.8%, IND 55.9%
The estimate I've obtained to date is that despite the slightly lower primary vote, the Liberals lose the 2PP vs Labor very slightly less heavily than One Nation loses the equivalent figure: I get Labor at 57.89 vs the Liberals, and 58.19 vs One Nation.
I also did (extremely rubbery) 2PP and shadow 2PP estimates for every seat using the above flows for candidates who outlasted the relevant exclusion points. On these estimates, the win/loss outcome for Labor vs the Liberals is the same in every seat except for Heysen (49.3 vs Liberal, 56.6 vs One Nation) and Bragg (41.5 vs 56.6). One Nation outperforms the Liberals in 24 seats but nearly all of them are either safe for Labor in either case or seats Labor lost to both parties. Four Labor seats that are not marginal vs the Liberals are marginal vs One Nation: the northern Adelaide set of Elizabeth, King, Light and Taylor. If One Nation sticks around Labor may be vulnerable in these seats next time around, but Elizabeth and Light were vacant so new Labor MPs should pick up personal votes.
Two-party swing strikes again
People are always complaining about two-party-preferred being useless and surely this election would be a standard for that measure if ever there was one, but not quite so fast. Again, the two-party swing (which I've modelled as 3.3%) told much of the story of this election. It would have been expected from this swing that Labor would have gained about four seats from the Liberals; they in fact gained five. Below that swing level Labor won Morialta and Unley but Josh Teague somehow retained Heysen, and Ngadjuri was a dud for Labor but fell to One Nation anyway. Slightly above the swing level, Labor won Hartley, Morphett and Colton. So they outperformed the pendulum, but not by much, especially as three of the five seats they won became vacant. On my estimates they may have also won the 2PP in Kavel (or not, see Addendum below), but that was in effect a retain for independents.
For all that there was a striking pattern that Antony has also observed regarding Labor getting great swings in marginals and having swings against it in its own safe seats. Antony has graphed it in terms of pseudoswings (eg ALP/Lib to ALP/ON) but in terms of estimated 2PP swings the pattern (coloured by seat winner) is just as stark:
The suggestion here is there could have been a "hybrid vigour" issue for Labor in seat share terms based on One Nation running strongly in some areas and the Liberals in others, except that Labor more than drowned it out by doing so well in the marginals. Victoria might be a different story ...
I've decided not to do a mixed post-election pendulum; Ben Raue has one here and I will decide what SA pendulums come to my site and the circumstances in which they come much closer to the next election. What is worth noting is that there are hardly any classic two-way marginals left because of the smashing inflicted on the Liberals in most of their marginals - only Heysen (Lib 0.6%), Morphett (ALP 0.7%) and Hartley (ALP 4.7%) - all the others that I model as 2PP marginals are occupied by someone else. On the other scale Labor vs One Nation marginals are Hammond (ON 4.9%), Light (ALP 1.6%), Taylor (ALP 4.2%), Elizabeth (ALP 4.5%), King (ALP 5.2%). Four years is an eternity in the history of One Nation being competitive or not so we will see in 2030 whether any of this matters. Ben also has lots of excellent detail about 3CP marginality of which there is plenty (most notably Heysen, the latest in a national string of very close Liberal/ALP/Green postcounts). Also see Ben Messenger's Blusky thread with diagrams.
The biggest victory ever?
Labor shouldn't let this one go to their heads as there are warning signs in these results, well if you stare at them for long enough there might be. But one theme I go on about a lot on here is federal drag, the theory and fact that when a state government is of the same party as the federal government, it tends to go backwards (and the older it is the more it does so). Yet here a federally dragged government that had been in office throughout the previous term has beaten its previous election result by 7/47 seats (14.9%). And this has never happened, in any state, ever. It's not quite the seat share record for an at-election change (because the 7 seats includes the two the government, also remarkably, picked up in by-elections during the term, so if ignoring those it's beaten by the Playford government's 5/39 coming off an indie rebellion in SA 1941) but something has gone very right for SA Labor and very wrong for its main opponent here.
Despite this, it could have been worse for the conservative forces, with widespread predictions that either the Liberals or One Nation might actually get zero or one seats. (I saw at least half a dozen forecasts of a One Nation seat zero by people who were pretty well informed about elections but may in some cases have let their dislike of the party lead them astray). With nine seats between them, the two parties did even do slightly better than my read of the polls as pointing to a few seats each if correct, but that was partly because the election was actually just a little bit closer than the polls suggested. The major party gap was slightly lower, and so was the Greens primary, and the preference flow from independents may have been weaker - all pointing to a lower 2PP swing than expected. It was also because independents on the whole did not have such a great election. Fraser Ellis finally lost (perhaps because voters realised a vote for him could be a vote for having to vote again) and Lou Nicholson was the only gain in a seat that didn't have an outgoing independent. Other touted indies flopped even if they had done well in 2022. Credit by the way to Australian Election Forecasts for a remarkably close to exact seat tally forecast.
And finally ... the polls!
This was an excellent election for Australian polling under unusual and difficult circumstances. Four pollsters using online panel polling polled at least twice each (Newspoll, YouGov, DemosAU and Fox&Hedgehog) and these collectively did very well on the overall shape of the result. Non-online offerings such as an SMS poll by Roy Morgan and an experimental AI voice automated poll by Resolve were a bit more excitable on the size of the One Nation vote, whether because of the method differences or otherwise.
Three of the six pollsters decided not to release two-party figures (the three who did were YouGov, Fox&Hedgehog and Roy Morgan; these also all issued ALP vs One Nation figures). I've decided it is not appropriate to formally measure 2PP accuracy for an election where the utility of 2PP is so heavily debated by analysts with many not even estimating it, and where there is no single method of estimating it, so for this election I am only using primary votes. Here is my usual table minus the 2PPs (note: I am using the ABC's figures as there is an issue with the ECSA total numbers):
Ave is simply the average of the raw errors and RMSQ is root mean square error, which punishes big misses on one party more than several small differences. For the record I am using Ave as the lead method this time but the order is the same in both cases. And if I was to include the 2PPs for those producing them then YouGov with 59-41 for both is closest to my estimate (all who bothered had 59-41 for ALP vs One Nation, Fox&Hedgehog had 60-40 for the classic 2PP and Morgan had 61-39). Such has been the accuracy of Newspoll in recent years that this is the first election since Victoria 2018 where there have been more pollsters above it than below it, but this is a tribute to just how well the others did. Extremely close between the top three but YouGov has taken home the Easter chocolates.
Addendum April 11: Antony Green has estimated 2PPs for eight seats that are missing them; these figures for Labor with my method's estimates are shown below (Green/Bonham):
Chaffey 31.4/31.4
Finniss 42.9/41.1
Flinders 25.9/30.9
Kavel 49.2/52.4
Mackillop 32.5/34.1
Mount Gambier 44.4/46.4
Narungga 35.4/38.4
Stuart 39.4/40.3
These may be based off pulled 2CP counts and if so will be more accurate than my estimates. There is however a large discrepancy in the 2PP vote totals for Stuart and a smaller one for Mackillop. If I include these numbers as is my 2PP estimate for Labor reduces to 57.72 (perhaps slightly further accounting for the Stuart undercount). However as this is a skewed sample of the seats that do not have exact 2PPs (skewed towards seats where Labor polled poorly and was eliminated) it is better not to do this so 57.89 remains my overall estimate for now. Antony has also provided an ALP/ON estimate of 79.7 vs my 78.8; if this is included then ALP vs ON changes to 58.21; again, it's better not to do this.
