The final two-party preferred result is 50.36% to the Coalition to 49.64% to the ALP. This represented a clear success for the last-election method of preference prediction, which would have predicted a 2PP of 50.53% for the Coalition based on the primaries actually cast. The respondent-preferences method (on average across polls using it) expected a shift in preferencing large enough to shift the 2PP result by at least 0.6 points (ie Labor would have won the 2PP). This continues the superior track record of last-election preferences, and I will continue to treat respondent preferences with caution.
The major source of what error there was in preference prediction was the appearance of the Nick Xenophon Team as a lower house party. Although NXT polled only 1.85% nationally, these split 60.2% to Labor, increasing Labor's 2PP by 0.26 points compared to treating NXT as like "others" from the previous election. This was partly counterbalanced by Green preferences being all of 1% (oooh!) better for the Coalition than in 2013. Overall both Greens and the generic Others preferenced very similarly to in 2013.
These are the preference flow figures I'll probably be referring to a lot in the next few years:
* Greens to Coalition: 18.1%
* Others including NXT and PHON to Coalition: 50.8
* NXT to Coalition: 39.8
* PHON to Coalition: 50.5
* Others excluding NXT and PHON to Coalition: 52.9
* Coalition loss from three-cornered contests: 0.13 points
This was the weakest ever flow of One Nation preferences to the Coalition. At previous elections they have generally flowed in the 54-56% range.
As for the remaining 0.27 points of error in my polling aggregate's prediction of the national 2PP (I had it at 50.8%), this is a modest error well inside what could easily occur just by sample noise in the final polls. But I suspect the major cause of it was that I continued to apply a 0.3-point global house effect correction based on apparent (but far from clearcut) evidence that the polls overall had been leaning slightly to Labor. However in the final week, the difference between Newspoll and Essential on the one hand, and the remaining polls on the other, suddenly more or less disappeared.
The net 2PP swing was 3.13%. As a uniform swing this would have resulted in fifteen Labor gains. Labor actually gained sixteen seats, more than cancelling out the impact of personal vote effects with unusually large swings in target marginals in Tasmania, NSW and Queensland. Seats lost to Labor that were on margins above the swing were Macarthur, Bass, Macquarie, Cowan, Burt (new seat without sitting member), Herbert and Longman. Seats won by the Coalition that were on margins below the swing were Robertson, Banks, Capricornia, Petrie, Chisholm (gained) and Page. These were all Coalition sophomore effect seats (mostly double sophomore), except Chisholm where the sitting member had retired.
There were a massive seventeen non-classic seats (seats where the final contest wasn't Labor vs Coalition) at this election:
ALP vs Green (3): Grayndler, Batman, Wills
Coalition vs Green (3): Warringah, Higgins, Melbourne
Ind vs Coalition (3): Indi, Cowper, New Englan
Ind vs Labor (1): Denison
NXT vs Coalition (3): Barker, Mayo, Grey
ALP vs NXT (1): Port Adelaide
KAP vs Coalition (1): Kennedy
Coalition vs One Nation (1): Maranoa
National vs Liberal (1): Murray
Melbourne had been Coalition vs Labor at the previous election. At this election Labor won the 2PP in the electorate with 66.6% of the 2PP vote, but finished outside the top two after preferences. I am not aware of any previous federal case of this.
As explained last time it is possible to use the various statistics to derive 2PP Labor-vs-Coalition splits for voters who put the "non-classic" candidate ahead of both the majors. In rare cases it is possible to find the exact 2PP split for voters who voted 1 for the non-classic candidate. After the 2010 election there was a very large amount of debate about which side's voters had supported various crossbenchers but since then it has been less topical. But for the record, here's a table of these figures:
The table also shows what percentage of the votes the figures are based on were #1 votes for the non-classic contender. This is especially relevant in Warringah, where the votes the Greens had when they were excluded were almost as much from other parties as their own.
As usual, voters who preferred the Greens over both major parties in non-classic seats strongly preferred Labor over the Coalition. Denison voters for Andrew Wilkie also did so, by about ten points more than last time. (The table above omits the exact split for Denison for 2013 which was 61.7%). We've already seen that NXT voters tend to prefer Labor over Coalition by about 60:40, while Katter voters in Kennedy and One Nation voters in Maranoa both leaned to the Coalition.
For Indi, there is a very rare situation: one of the major parties (Labor) finished fourth. Therefore it is not possible to calculate how many of Cathy McGowan's voters preferred Labor to Liberal by the same method. @sorceror43 has estimated 72%. I am still exploring some other methods of deriving an exact figure here but they will probably end in failure.
Strongest preference flows
As in 2013, the list of the strongest known preference flows in particular seats is dominated by Green to Labor flows. The top ten I can find are:
1. 92.54%, Duncan McGauchie (Liberal) to Damian Drum (National), Murray, Vic
2. 90.48% Jim Casey (Green) to Anthony Albanese (Labor), Grayndler, NSW
3. 90.28% Terry Morgan (Green) to Susan Templeman (Labor), Macquarie, NSW
4. 89.53% Sylvie Ellsmore (Green) to Tanya Plibersek (Labor), Sydney, NSW
5. 89.38% Carly Saeedi (Green) to Andrew Leigh (Labor), Fenner, ACT
6. 89.17% Terrill Riley-Gibson (Green) to Ross Hart (Labor), Bass, Tas
7. 89.15% Rosemary Glaisher (Green) to Lisa Chesters (Labor), Bendigo, Vic
8. 87.98% Tamara Ryan (Green) to Mike Kelly (Labor), Eden-Monaro, NSW
9. 87.86% Sophie Guy (Green) to Kate Ellis (Labor), Adelaide, SA
10. 87.73% Hugh McKinnon (Green) to Jenny Macklin (Labor), Jagajaga, Vic
The figure in bold for Grayndler is the flow from votes that placed Casey ahead of both the majors, not just from Casey's own votes. The flow from Casey's #1 votes to Albanese is probably a bit higher.
Grayndler, Sydney, Bendigo, Macquarie and Fenner (then Fraser) all made this list last time as well. However, Batman, which may well have actually had the strongest Green-Labor flow in 2013 (counting only #1 votes), falls well down the list. The weakening of Green preference flow to Labor in this seat was probably a result of Green-voter displeasure with David Feeney's performance.
The weakest Green-to-Labor preference flows for Coalition-vs-Labor seats were 56.7% in Fowler, 63.2% in Parkes and 63.7% in Calare.
Among two-candidate-preferred preference flows from excluded candidates to one of the final two in their seat, 45 of the top 50 were Greens to Labor. The remainder were Labor and the Greens to McGowan in Indi, Socialist Alliance to Labor in Sydney, Greens to NXT in Mayo and Labor to Green in Higgins.
Winning from behind: the strange last laugh of PUP
Sixteen seats were won by the candidate who finished second on primaries - Batman by Labor against the Greens, Mayo by NXT against the Liberals, and all of these by Labor against the Coalition: Herbert, Hindmarsh, Cowan, Longman, Melbourne Ports, Griffith, Macquarie, Braddon, Lyons, Perth, Richmond, Moreton, Adelaide and Jagajaga.
Of the fourteen Labor wins, in ten cases Labor had a large lead if Green preferences were factored in. In these ten seats the Coalition would have needed at least 64.6% of fourth-party preferences to win (and in two the share of fourth-party preferences required was mathematically impossible.) I mention this because Senator Abetz has tried to blame Tasmanian seat losses on - among very many other things - the Recreational Fishers Party.
While the RFP's preferences did flow fairly strongly to Labor, the Coalition needed a mid-60s share of non-Green preferences in both Braddon and Lyons. They didn't even hit their target off the Christian Democrat vote in Lyons, and barely did so in Braddon, so there was no way the Coalition were going to hold those seats whatever the Recreational Fishers voters did. The Recreational Fishers could not have caused the Coalition to get 88.4% of their Braddon preferences or 86.6% of their preferences in Lyons (the share of them needed to win) even had they wanted to!
The four seats where Green preferences weren't the single clear deciding factor in Labor's favour were:
* Melbourne Ports, where the Coalition would have won had it got 58.2% of the fourth-party vote, but the minor candidates were mostly left-wing.
* Longman, where the Coalition needed only 47.8% of the fourth-party vote but was stymied by One Nation (43.5% to Coalition) preferencing Labor. It is worth noting here that even had the One Nation vote split 50-50, Labor would still have won off the preferences of the Greens and the minor candidates. However the actual flow of One Nation preferences left the LNP needing a 56.7% split off the mostly left-wing minor candidates; Wyatt Roy actually got 44.5%.
* Hindmarsh, where the Coalition needed only 42.1% of the fourth-party vote but the fourth-party vote was mostly NXT, meaning that Green and NXT preferences were about equally responsible for Labor's win.
* Herbert. The Coalition needed 48.3% of the fourth-party vote and got almost exactly that from One Nation (even though this was a shade below the national One Nation to Coalition share - had One Nation preferences split 50-50, the Coalition would have retained the seat). However they failed to hit the mark on the preferences of each of Katters Australian Party, Glenn Lazarus Team and, in a ridiculous irony, Palmer United Party.
As noted last time Clive Palmer made some ridiculous claims about PUP preferences deciding the 2013 election. PUP polled over 700,000 votes in 2013 but there was not a single seat where a 50:50 split of PUP preferences would have led to a different outcome, and there were only ten seats (seven won by the Coalition and three by Labor) where any split of PUP preferences (however unlikely) could have changed the winner. The impact of PUP on the 2013 lower house election was basically just this: they won one seat.
In 2016 the Palmer United Party ran just a single lower house candidate, Martin Brewster, who happened to run for the seat of Herbert which then happened to be the closest in the nation. Mr Brewster polled 315 votes, which is 0.04% of PUP's national vote from 2013. And yet, these preferences split 198-117 to Labor, causing Labor to gain 81 votes, which exceeded the winning margin of 37. It might thus be (a little cheekily) argued that 315 Palmer United Party voters had as much impact on the 2016 outcome (one seat) as 709,035 PUP voters did in 2013.
On the other side of the coin, the Coalition won two seats where they were behind after considering Green preferences. In Forde, the LNP needed 51.4% of Others preferences to win but these preferences were dominated by Family First and Rise Up Australia. In Gilmore, the Liberals needed 57% but the only fourth party candidate was a Christian Democrat, so they did it easily.
Other sections may be added to this piece if I notice anything else worth adding, or if there are any interesting requests that are practical to answer here.
And see also Antony Green.