SA election – how do we predict Nick Xenophon’s vote?


The recent Queensland election produced a dilemma for us electoral analysts: polls suggested One Nation would perform strongly, but the party had no recent history of contesting seats in most of Queensland at state elections.

We relied instead on the results of the 2016 federal Senate election, broken down by Queensland state electorate. This work was conducted by Alex Jago, who used data on which voters from each block voted at each polling place. He also used the entire preference dataset to distribute votes cast for minor candidates between Labor, the LNP, One Nation and the Greens.

The South Australian election has produced a similar dilemma. Nick Xenophon’s SA Best party has never contested a South Australian state election. Nick Xenophon previously ran for the state upper house in 1997 and 2006, polling quite highly in 2006, before running for the Senate with strong results at the 2007, 2013 and 2016 federal elections.

Thankfully Alex Jago performed a similar task for South Australia, distributing votes at a SA1 level between Labor, Liberal and the Nick Xenophon Team. He then gave it to me and I matched those SA1s to South Australian state electorates, to allow me to produce an estimate of the vote for the three biggest parties in each state seat. You can view this data here.

NXT polled well everywhere – the lowest NXT vote was 21.7% in Croydon, and the highest was 38.5% in Heysen. But there is a trend. The NXT vote was highest in the seats in the Adelaide Hills and to the south of Adelaide, as well as those in the north of the state, and it was lowest in centre and northern Adelaide.

Liked it? Take a second to support the Tally Room on Patreon!


  1. The numbers imply Xenophon is into the runoff in his 12 highest 3PP seats (if everything stays the same). Those seats include 10 seats the Liberals would have won by more than 13% 2PP in 2014… and 2 where it would be Labor by ~5%.

    Xenophon would also be into the runoff in a few seats outside of his top 12, usually when the Labor party is especially weak. Notably Hartley is his 20th best 3PP and he is starting in 3rd.

    I’m assuming both major parties will preferences SA Best to spite each other out of majority rule and therefore SA Best will get strong flows no matter who they are against in the runoff. Compared to PHON’s preference barrier in QLD this is a doddle for Xenophon to engineer a hung parliament. Hartley was certainly a foolish seat to pick for that strategy though.

  2. Great work Ben!

    Pseudonym: we use Senate data because you can reliably generate 3PPs across the entire state, regardless of whether NXT stood a lower house candidate there or whether they finished top-3.

    Senate is a good baseline, because Nick himself was at the top of the ticket; it should be relatively unaffected by other local candidate issues.

    If NXT stood a lower house candidate in all SA electorates and they all finished top-3, then yes, that would probably be better. But you’d have to collate that manually.

    The AEC releases a massive spreadsheet detailing every formal Senate preference ordering in the state or territory, by booth. (Aside: this easily allows vote coercion.) I can generate N-party-preferreds for any combination of tickets in a state (including their outgoing preference breakdowns to the other tickets). Such data isn’t available for the lower house.

  3. Ben:
    There is another serious contender.
    How about some analysis for Australian Conservatives please?


    Milton Wilde

  4. Family First polled 2.9% at the last federal election in South Australia and 4.4% at the last South Australian state election. Until we see any evidence that anyone is actually willing to vote for Cory Bernardi I don’t think they’re deserving of any particular focus, nor do I have any data to use for that analysis.

  5. Milton –

    If you consider the Conservatives as a continuation of Family First, they aren’t likely to win any lower house seats without massive changes in voter behaviour (which would overwhelm any precision in a model).

    In some seats they might well beat the Greens, but would still come a distant fourth, without much impact on the contest for second and first, unless they preference tactically (i.e. SABest > Lib > ALP in Labor seats).

    Best just to look at the 2014 Upper House results (by booth or district) for Family First, as a prediction of how the Conservatives *might* do.

  6. AlexJ would you be able to do the same analysis on lower house votes? I think AEC provides distribution of preferences by booth for lower house votes, all of which would be Lib/Lab/NXT in SA

  7. John –

    Having had a quick squiz at the distribution of preference spreadsheets, the federal booth collation should be pretty straightforward. I’ll see if I can find the time and motivation in the coming week.

    SA1 projection should likewise be fairly straightforward.

    Ben did the state-district collation, but if someone wants to link me the relevant dataset…

  8. I’m happy to send it to you Alex. I personally think the Senate is better because it eliminates candidacy effects which wouldn’t be the same for the state election.

    I also don’t think the distribution of preferences is that important, particularly in the house. You could just report the primary vote totals. The preference distribution is handy in the Senate because there are so many running but that’s not so true in the House.

  9. I’ve done the booth collation, but not the SA1 projection or state-district aggregation.

    What I *have* done is a comparison with the Senate booth-by-booth 3PPs.

    Statewide, I’ve found that Labor does about 2.2 percentage points better in the House and the Liberals about 0.6 points better (so NXT about 2.8 points worse).

    However, there’s quite a bit of variance by federal division. There’s a slight correlation with the Senate 3PP performance (in general, the lower the Senate NXT 3PP, the more their House 3PP underperforms their Senate 3PP).
    I imagine a reasonable part of that would be candidate effects.

  10. New poll out today for SA: 32% for SA Best, 29% Lib, 27% ALP, 6% Greens, 6% Other.

    I would split the Greens’ 6% as going 4% Labor 1% SAB 1% Lib, and mirror-similarly the Others’ 6% as going 1% Labor 1% SAB 4% Lib, for overall 3PPs of 32% ALP, 34% Lib, 34% SAB.

    The Reps 3PP I mentioned in the previous comment had NXT on 27% statewide, Labor 33% and Libs 36%.

    Of course, with SAB only running in half the state this time around, they’ll outperform their election-day statewide percentage.

    If today’s polling holds up and they target their seats ‘correctly’, they should make the top two in most of their seats, and go on to win.

  11. I’ve started doing some modelling on SA based off this (similar to what I did with Queensland). What I’ve noticed is that if you extrapolate from the Senate 3PPs to a notional 2PP by splitting the Xenophon votes 60:40 to Labor and the “none” votes 50:50, then what you get is under-dispersed by almost a third compared to the state-level 2PPs for the given seats that Antony has posted. In the seats where Labor got 60% 2PP last state election their 2PP from the Senate results comes out at 60%, but in the seats where they got around 25 the Senate conversion comes out around the high 30s.

    It is not obvious to me why this should be the case; all the obvious explanations I thought of and tested fell over.

  12. One possible factor in the above is that NXT votes come disproportionately in Lib electorates (but they don’t break more strongly to Liberals as preferences at least if 2016 federal is any guide). Another, I think, is that some seats might be uncompetitive at state level but more heavily campaigned federally because they formed part of a larger, closer electorate.

Comments are closed.