Sunday, July 3, 2016

2016 Federal Postcount: Grey and Other NXT Seats

Preferences are being reallocated to correct two-candidate preferred
Ignore raw totals during this process 
Outlook: Liberal retain

This thread follows postcounting in the seats where the Nick Xenophon Team might be in some kind of contention in South Australia, other than Mayo which they have already won.  The big one is Grey but some people are saying there is still life in Boothby and there are a few others worth mentioning.

Grey is a case where the AEC had the "wrong" pair of candidates on the night and these cases always create great confusion because of the process involved in fixing this.

For starters I will just cover Grey, the vast seat that includes Whyalla and also the South Australian outback, and add others later.  Troubles with the Arrium steelworks propelled Grey from one of NXT's weakest areas to a potential win with a startling ReachTEL having it 54:46 ahead.  That proved to be not too bad as far as seatpolls for always volatile third-party attempts go and for now Grey is in limbo.  Another area of interest is that should Andrea Broadfoot defeat Rowan Ramsey, her election might in theory be challenged over her decision to not resign a council seat - a grey (pun intended) area. One would not think either party in a hung parliament would be so stupid as to challenge, but it's always possible some random person might.  (Opinions vary as to what the courts would do in this untested area, and the Labor Party has been requiring all its candidates to quit local council seats for many years.)

The preference counts that were posted on the ABC website on the night were never real numbers, just their back-of-envelope estimates on preference flow.  What is going on now is a reallocation process in which booths are thrown as Liberal vs NXT.  The two-party preferred will bounce around like crazy based on which booths are thrown at what time. At present, Broadfoot is shown as leading 56.5:43.5 causing certain Fairfax sites to call the seat an NXT win.  Ignore this; it is totally misleading. Once enough booths are in the count it is possible to use regressions to fairly reliably estimate where the count might finish up.  Another difficulty is that Grey has many remote booths, not all of which are counted yet.

On the primaries as I wrote this article, Ramsey (Lib) had 41.6%, Broadfoot (NXT) 28.48%, Martin (ALP) 21.99%, Kaminski (FF) 4.18%, Marsh (Green) 2.46% and Gourlay (Ind) 1.28%.  On these numbers Broadfoot needs 71.9% of all preferences to go to her, but that figure may increase when more booths are added to the count.

The early signs are not promising for NXT since the first four booths (the fifth shown has a zero vote count) have had these results:

Port Augusta Central: 64.7% flow (primaries NXT 32.96 Lib 31.47, ALP 25.37 rest 10.2)
Port Augusta East: 68.7% flow (primaries NXT 34.36 Lib 25.29 ALP 32.76 rest 7.6)
Port Augusta PPVC: 68.4% flow (primaries NXT 31.64, Lib 38.14, ALP 23.08 rest 7.14)
Stirling North: 67.5% flow (primaries NXT 38.87, Lib 29.32, ALP 24.21 rest 7.6)

I have had a scrutineering comment suggesting similar flows in a Port Pirie booth, and also suggesting that a factor in the weak preference flow could have been that the pro-Liberal preference recommendation on the Labor split ticket was easier to follow (donkey order).

More booths will be needed to identify what the preference flow to NXT correlates to, but an early suggestion from a whole four data points (don't bank on this!) is that it might correlate to the proportion of the remaining votes that are for Labor.  If that is so then the flow to Broadfoot should stay at about 66% of preferences and she'll probably miss out with about 48.2% 2PP (in which case Antony Green's intuition was right on the money.)  However further data points may refute this idea very quickly and suggest a new one!

You can see some modelling of the Mayo vote by poster sykesie in comments which suggests that if the same preference patterns seen in Mayo apply to Grey then the seat could be very close.

The other point that is important here is that there may well be regional variation in the strength of the preference share.  It may be exceptionally strong for NXT in Whyalla, for example (even given the huge Labor vote there).  A wider range of booths are required.  But if NXT are not getting the flow in bad booths for the Liberals, the question is why they will get it anywhere.

(The flow is a bit weaker than I thought it might be, by the way.  Despite the open ticketing by the major parties, I thought it was a fair chance NXT candidates could get 70% flows.)

Nick Xenophon said on ABC TV news tonight that Broadfoot was half a percent behind.  I don't know whether he based this on a scrutineering projection or not, but let's keep in mind that this is a person who said that his team winning a Reps seat was a one in a million chance.  I posted the Pratchett quote at the time and it was right again.  One in a million chances happen nine times out of ten ... especially in Mayo.

I'll add more comments with more data and say something about some other SA seats (Barker, Boothby and Sturt at least) when time permits.

Update Tuesday 6:30: Many more booths have been reallocated in Grey causing Rowan Ramsey to open up a large lead.  No Whyalla booths (at least those called "Whyalla") are yet included.  I haven't updated my projection attempt yet but I expect that any reasonable projection off the booths counted is going to show Ramsey hanging on.

Update Tuesday 9 pm: I've done some projections from the 32 booths processed so far.  Using the model I suggested at first (share of the Labor vote) I get a 64% share for Broadfoot with 20% of variation explained - this rises to 67% with 33% of variation explained if I kick out four obvious outliers.  If I use the share of the Xenophon vote the projection is 66% with about 38% explained, but when I kick out the outliers the model becomes less powerful not more.  If I use both together with the outliers removed the projection is 68.5% of variation with 34% explained (with outliers included 67% with 42% explained).  The best-case projected final 2PP for Broadfoot in these models is 48.8% 2PP and that is before any further damage that might occur in postcounting.  The suggestion from the other booths is that when we get into the Whyalla booths (where nearly all the preferences are Labor) the preference flow will not crack 72% let alone whatever it by then needs to beat the flow in the rest of the electorate.  I think it is extremely very likely that Ramsey has retained Grey but I would like to see one Whyalla booth to put it away.

Update Tuesday 11:15 pm: Something very interesting has turned up in the other NXT seats: far from suffering damage on postals, NXT so far does better than the Liberals on them (albeit not massively).  Although I still expect Ramsey to win I've downgraded Grey to "very likely".  I know the ABC has called the seat but I am very cautious about these things and I have not.

Wednesday 11:00 am: An early batch of postals in Grey hasn't followed the above script (which given the different geography of Grey should not come as any huge surprise) and has split 57:43 to Ramsey, pushing the required preference flow up to 73%.  Even without seeing the Whyalla booths I now think this is too hard for NXT.

Thursday 7 July: Ramsey is ahead 52:48 with only a small number of booths to add, pretty much as predicted.  I am no longer following this seat.

Barker, Boothby and Sturt

There has been some thought that NXT might have chances in the post-count in some other Liberal-held electorates if they can get into second place (or in one case they already are).

We have seen that the flow of all preferences in Mayo, a seat where preferences are overwhelmingly from Labor or the Greens and the sitting member was scandal-affected, is 73% to NXT.  In Grey so far it is 66% to NXT.

In Barker, NXT are second.  In Boothby they can move into second if they gain 4.41% on the combined 10.71% preferences of other parties.  In Sturt, NXT need to gain 2.07% on the combined 11.96% preferences of other parties (again mostly Greens), which seems more plausible but still won't necessarily happen.

The problem in all these cases is that even if the NXT candidates are second, they need a very strong flow of all preferences to win.  In Sturt it is 85%, in Barker 85.5%, in Boothby 78.6%.  These are current figures and as postcount votes are added it's quite likely these numbers will increase.  They are already a lot higher than the known flows in Mayo and Grey.

My view is that in Sturt and Barker the mountain is way too steep.  In Boothby the preference flow required couldn't be completely ruled out, but it will probably get harder in the post-count and I'm very doubtful NXT will make it into second there anyway.

Scrutineering figures on the Green split in Boothby would be useful but it seems unlikely NXT can get any of these.

For completeness I note that NXT are trailing the Liberals by 68 votes for second in Port Adelaide and may move into second there on minor party preferences.  If they do so they need a mere 97.4% of all preferences to win.

Update (Tuesday midnight) Booth realignment has started in Barker.  It's the same process as Grey and the 2PP early on is probably generous to NXT.  They're getting 64%, 67% and 82% in the only booths of any size done thus far.


  1. It should be noted that that Port Augusta was very strong for NXT on primary votes and were some of the best polling areas

    Can the preferences in Mayo (there is some data) provide any further information on this?

    1. That's a good point, I'll have a look at Mayo.

  2. Hi Kevin,

    Here is something I posted yesterday afternoon on pollbludger based on a model similar to the one the first poster above suggested. This was before the handful of 2PP counts in Grey were published. The initial 2PP count in Grey does not look strong enough for NXT, but we would need a more geographically diverse seat of booths to be more confident of this.

    From my pollbludger post:

    I was looking at trying to predict the result in Grey, and I got thinking about creating a regression model for Mayo as the ALP/GRN/others preference flow is known. This is similar to models which I created previously (which worked very well) for predicting Denision in 2010 and Indi in 2013. Then after creating the Mayo model, assuming the same preference flow in Grey (which is probably reasonable within a degree of uncertainty), the final result could be predicted.

    When I created the model for Mayo using all polling place data and applying to the total primaries I got the following:
    NXT 2PP = 0.534*(ALP+GRN primary) -0.313*(LIB primary) +0.783*(NXT primary) + 28.2. This gave a predicted 2PP NXT vote in Mayo of 55.8% versus an actual 2PP of 55.7%, with a standard error of +- 0.835%. (R2=0.971)

    Now for the fun part; lets try and apply the same model to Grey:
    When this is done, the NXT 2PP figure in Grey comes out as 50.5% .

    I also played around with a model having the ALP and Green as separate variables; this only very slightly improved the stats but came out with a NXT 2PP value of 49.9% …

    Conclusion: NXT are still very much in the race for Grey.

    Incidentally if the same model (first one above) is applied to Boothby, assuming NXT scramble into second place, I get a NXT 2PP value of 48.4%. For academic interest, Barker and Sturt were 46.3% and 45.9% respectively.

    Conclusion 2: Outside chance that Boothby could still be in play.

    Cheers, Sykesie.

  3. This comment has been removed by the author.

  4. This comment has been removed by a blog administrator.

  5. Looks like Grey is in the bag for the Coalition. Booths being added quickly now.

    Do you think the Labor party running open tickets was a tactical blunder?

  6. If the Coalition wins by one seat it won't look terribly bright.

  7. NXT actually has moved into second now on primaries, and I assume would make the 2CP cut on preferences.