Farewell 2018 - Vic Review

Final, we have the Victorian election to review. 88 Legislative Assembly seats and 40 Legislative Council seats were up for election: the full Victorian parliament.

Legislative Assembly

74 seats, or 84.1%, were predicted correctly.

Seat	Prediction	Result
Albert Park	ALP	ALP
Altona	ALP	ALP
Bass	LIB	ALP
Bayswater	LIB	ALP
Bellarine	ALP	ALP
Benambra	LIB	LIB
Bendigo East	ALP	ALP
Bendigo West	ALP	ALP
Bentleigh	ALP	ALP
Box Hill	LIB	ALP
Brighton	LIB	LIB
Broadmeadows	ALP	ALP
Brunswick	ALP	GRN
Bulleen	LIB	LIB
Bundoora	ALP	ALP
Buninyong	ALP	ALP
Burwood	ALP	ALP
Carrum	ALP	ALP
Caulfield	LIB	LIB
Clarinda	ALP	ALP
Cranbourne	ALP	ALP
Croydon	LIB	LIB
Dandenong	ALP	ALP
Eildon	ALP	LIB
Eltham	ALP	ALP
Essendon	ALP	ALP
Euroa	NAT	NAT
Evelyn	LIB	LIB
Ferntree Gully	LIB	LIB
Footscray	ALP	ALP
Forest Hill	LIB	LIB
Frankston	ALP	ALP
Geelong	ALP	ALP
Gembrook	LIB	LIB
Gippsland East	NAT	NAT
Gippsland South	NAT	NAT
Hastings	LIB	LIB
Hawthorn	LIB	ALP
Ivanhoe	ALP	ALP
Kew	LIB	LIB
Keysborough	ALP	ALP
Kororoit	ALP	ALP
Lara	ALP	ALP
Lowan	NAT	NAT
Macedon	ALP	ALP
Malvern	LIB	LIB
Melbourne	ALP	GRN
Melton	ALP	ALP
Mildura	NAT	IND
Mill Park	ALP	ALP
Monbulk	ALP	ALP
Mordialloc	ALP	ALP
Mornington	LIB	LIB
Morwell	ALP	IND
Mount Waverley	LIB	ALP
Mulgrave	ALP	ALP
Murray Plains	NAT	NAT
Narracan	LIB	LIB
Narre Warren North	ALP	ALP
Narre Warren South	ALP	ALP
Nepean	LIB	ALP
Niddrie	ALP	ALP
Northcote	GRN	ALP
Oakleigh	ALP	ALP
Ovens Valley	NAT	NAT
Pascoe Vale	ALP	ALP
Polwarth	LIB	LIB
Prahran	GRN	GRN
Preston	ALP	ALP
Richmond	ALP	ALP
Ringwood	LIB	ALP
Ripon	ALP	LIB
Rowville	LIB	LIB
Sandringham	LIB	LIB
Shepparton	IND	IND
South Barwon	ALP	ALP
South-West Coast	LIB	LIB
St Albans	ALP	ALP
Sunbury	ALP	ALP
Sydenham	ALP	ALP
Tarneit	ALP	ALP
Thomastown	ALP	ALP
Warrandyte	LIB	LIB
Wendouree	ALP	ALP
Werribee	ALP	ALP
Williamstown	ALP	ALP
Yan Yean	ALP	ALP
Yuroke	ALP	ALP

The 14 incorrect predictions were:

Seat	Prediction	Result
Bass	LIB	ALP
Bayswater	LIB	ALP
Box Hill	LIB	ALP
Brunswick	ALP	GRN
Eildon	ALP	LIB
Hawthorn	LIB	ALP
Melbourne	ALP	GRN
Mildura	NAT	IND
Morwell	ALP	IND
Mount Waverley	LIB	ALP
Nepean	LIB	ALP
Northcote	GRN	ALP
Ringwood	LIB	ALP
Ripon	ALP	LIB

Mildura and Morwell saw independents elected. Because it is not feasible to look into the prospects of every independent running in every seat, I accepted long ago that upsets of this nature are unpredictable in practical terms.

I had also noted at the foot of my predictions that, for unempirical reasons, I expected the greens to pick up a few seats not officially predicted, and singled out Melbourne as likely to be retained. I do not consider Melbourne or Brunswick to be surprises.

Eildon and Ripon were retained by Liberal incumbents against the swing and remain very narrow as a result. ABC has Ripon held by the Liberal party with 50.0% of the final count, meaning there were only a handful of ballots between the major parties in the end. Eildon is a little less precarious, but with 52.4% of the vote we came very close to being right in both cases, and I think these are acceptable mistakes.

Northcote was held by the Greens, but only as the result of a by-election, the Liberals did not contest. I do not think the lack of Liberal contest in that by-election makes that data unreliable, as traditional liberal voters would have allocated their votes exactly as their preferences in a contested election. However, the by-election may have been atypical for the focus it provided on local issues. The Greens certainly still poll well here, but despite the ABC indicating a swing in their favour (compared to the general election, not the by-election) they did finish second. I am in two minds as to whether this is a reason to be more cautious with by-election data and incumbents or just part of the usual statistical hurley-burley.

The remaining seven seats (Bass, Bayswater, Box Hill, Hawthorn, Mount Waverley, Nepean and Ringwood) are evidence that I underestimated the swing to the ALP in this election. It is true that this swing was far higher than most predicted, but being wrong en mass is little consolation. Previous work on predicting the size and distribution of swings is required, though past efforts have provided little encouragement. Nevertheless, these are seven seats out of 88, and a mid-80% accuracy overall is both typical and acceptable.

Legislative Council

The LegCo results were more chaotic by far.

As with all proportional, preferential votes, the final member from each division is always unpredictable, so the districts with scores of 4/5 are perfectly acceptable to me despite the patchwork of colours in the infographic above.

The total prediction score of 30/40 or 75% would be disappointing in a race of single-member seats like the Legislative Assembly, but this inherently difficult fifth-member prediction means a score above 80% would be outstanding. The remaining 20% are inevitably written off in almost every case as tossups for the same reason seats won by independents are not considered particularly concerning in the lower houses.

The areas of concern, then, are the Southern Metropolitan district and both the Northern and Western regional districts.

In South Metro, the ALP gained from the Greens, and the Reason Party (formerly the Australian Sex Party) took the scraps from both sides to earn the 5th seat. This is the first time a party outside of the major trio (ALP, LIB and GRN) have won a seat in South Metro and though they did get a mention in the prediction, this is unexpected based on the earlier polling. Similarly, for the Greens to consistently poll 90% of a quota historically would seem to guarantee them a seat, but this year they were closer to 77%. Both errors here demonstrate that historical data makes for poor predictions of the future, though the absence of alternative numbers is an issue we will struggle to overcome.

In the regional districts, the Liberals lost ground to several minor parties too--a cause of errors in the 5th seat of other districts also (Metro East and Metro South East in particular). Again, this is the errors of past data being used predictively. It is evident the Liberals' support from previous years was not present in 2018. One predicted shooters seat in Vic West was also lost as part of the general uncertainty and minor-party chaos of the 5th post, but I'm more impressed that I called them correctly in Vic East (and got a full 5/5 in that district).

The lessons of the Legislative Council predictions then are that tossups should be used liberally if this blog still handed those out, and that past data is poor predictively but more or less all that was available.