Saturday, 24 October 2020

Backdated - Results, Queensland 2020

Predictions from the Queensland 2020 state election had a decent success rate of 93.7%, or 89 of 95. 89 out of 95 is also what I would have got by simply following the polling to state there would be no change (except that Whitsunday would return to the LNP after a 2019 expulsion of incumbent Jason Costigan). So my analysis didn't hurt my predictions, but probably weren't worth the effort either. Interestingly,both approaches would have miscalled the same four seats in their six: Caloundra, Hervey Bay, Nicklin and South Brisbane.

But this blog is misleadingly called the Infographinomicon, not the Wordy-spiel-riddled-with-numbers-apedia, so here's the coloured table:

ELECTORATE PREDICTION RESULT
Algester ALP ALP
Aspley LNP ALP
Bancroft ALP ALP
Barron River ALP ALP
Bonney LNP LNP
Broadwater LNP LNP
Buderim LNP LNP
Bulimba ALP ALP
Bundaberg ALP ALP
Bundamba ALP ALP
Burdekin LNP LNP
Burleigh LNP LNP
Burnett LNP LNP
Cairns ALP ALP
Callide LNP LNP
Caloundra LNP ALP
Capalaba ALP ALP
Chatsworth LNP LNP
Clayfield LNP LNP
Condamine LNP LNP
Cook ALP ALP
Coomera LNP LNP
Cooper ALP ALP
Currumbin LNP LNP
Everton LNP LNP
Ferny Grove ALP ALP
Gaven LNP ALP
Gladstone ALP ALP
Glass House LNP LNP
Greenslopes ALP ALP
Gregory LNP LNP
Gympie LNP LNP
Hervey Bay LNP ALP
Hill KAP KAP
Hinchinbrook KAP KAP
Inala ALP ALP
Ipswich ALP ALP
Ipswich West ALP ALP
Jordan ALP ALP
Kawana LNP LNP
Keppel ALP ALP
Kurwongbah ALP ALP
Lockyer LNP LNP
Logan ALP ALP
Lytton ALP ALP
Macalister ALP ALP
Mackay ALP ALP
Maiwar GRN GRN
Mansfield ALP ALP
Maroochydore LNP LNP
Maryborough ALP ALP
McConnel ALP ALP
Mermaid Beach LNP LNP
Miller ALP ALP
Mirani ONP ONP
Moggill LNP LNP
Morayfield ALP ALP
Mount Ommaney ALP ALP
Mudgeeraba LNP LNP
Mulgrave ALP ALP
Mundingburra ALP ALP
Murrumba ALP ALP
Nanango LNP LNP
Nicklin LNP ALP
Ninderry LNP LNP
Noosa IND IND
Nudgee ALP ALP
Oodgeroo LNP LNP
Pine Rivers ALP ALP
Pumicestone ALP ALP
Redcliffe ALP ALP
Redlands ALP ALP
Rockhampton ALP ALP
Sandgate ALP ALP
Scenic Rim LNP LNP
South Brisbane ALP GRN
Southern Downs LNP LNP
Southport LNP LNP
Springwood ALP ALP
Stafford ALP ALP
Stretton ALP ALP
Surfers Paradise LNP LNP
Theodore LNP LNP
Thuringowa ALP ALP
Toohey ALP ALP
Toowoomba North LNP LNP
Toowoomba South LNP LNP
Townsville ALP ALP
Traeger KAP KAP
Warrego LNP LNP
Waterford ALP ALP
Whitsunday LNP LNP
Woodridge ALP ALP

The six incorrect predictions were Aspley, Caloundra, Gaven, Hervey Bay, Nicklin and Brisbane South. In Aspley and Gaven the ALP incumbent held out against the LNP, and in Caloundra, Hervey Bay and Nicklin the ALP defeated the LNP, meaning that the result was stronger than expected for Labor. In Brisbane South the Greens defeated the ALP incumbent. At a superficial level it may be tempting to generalise this into a left-wing shift, but I don't see that as meaningful in Australian politics.

In 2020 the polling was not indicating any radical shift in polling sentiment. Three assumptions were made from this polling: anything considered fairly safe or better (6%+) would be unchanged, that One Nation's decline in polling would prevent them gaining seats and that a breakaway Liberal candidate in Whitsunday would lose his seat back to his party. For the most part these were good assumptions--state wide 2PP swing to Labor was +1.9 and One Nation's first preference was down 6.6 percentage points. However, Hervey Bay had been won in 2017 with a margin of 9.1 and saw an 11.1 pp swing in 2020. That is a weird outlier that wasn't predicted.

Hervey Bay (and Caloundra)

Hervey Bay covers Frasier Island and the mainland city of Hervey Bay. It was created in 1992. It was first held by Labor for six years, then by One Nation/One Nation Queensland/City Country Alliance for a term during the height of ONP's popularity. Labor recovered the seat in 2001 and held it for 8 years before being defeated by the LNP who kept if for the next decade. This was the situation as it stood in 2020.

The seat was clearly not immune to changing hands, but that hardly sets it apart from many others that behaved predictably.

The incumbent, Edward "Ted" Sorensen, was retiring in 2020. This was likely a significant factor--two other LNP candidates also retired in 2020, one in Pumicestone which was predicted to fall to Labor on this basis and the other in Caloundra which, like Hervey Bay) saw an unpredicted ALP take-over. The retirement of Mark McArdle was not factored into this call. Labor also had three retiring members (Kate Jones, Anthony Lynham and Coralee O'Rourke in the seats of Cooper, Stafford and Mundingburra respectively) but with the swing favouring Labor these all held (Cooper and Stafford were fairly safe, and Mundingburra was assessed as a Labor seat with a single historical outlier.

The lesson here for future (2022+) predictions is to not ignore the importance of incumbency.

Nicklin

With Caloundra explained on similar grounds to Hervey Bay, it remains to be determined what, if anything, can be taken as the oversight in Nicklin, which is the only other seat where Labor seized power. Nicklin was called on the basis that Labor had never won there. This isn't entirely irrelevant--going on history alone allowed us to correctly call 8 LNP and 11 ALP seats, and it's hard to find a seat with a more definitive history than Nicklin. It's also worth noting that Nicklin was won by the second-narrowest margin of 0.14 percentage points (second only to Bundaberg's 0.01).

On this ground I'm happy to treat this as a narrow outlier--no election will ever be perfectly predictable, after all.

Aspley and Gaven

Aspley, Redlands and Gaven were the last three seats predicted, a reliable sign I was running out of any meaningful data to work from. These were essentially tossups, with Aspley and Redlands called one for LNP and one for the ALP as insurance against a unforeseen landslide either way, and Gaven basically decided on a coin flip.

Both of these would have been right to call according to the original poll+pendulum prediction, which was suggesting no change, again emphasising the importance of incumbency in the absence of any other data. (Though, caveat: where electoral history is solid, this can be seen to be more important than the current member).

South Brisbane

South Brisbane was decided much earlier than Aspley or Gaven, and was confidently but wrongly called for ALP on the basis of a strong Labor history (which is likely in any seat the Greens win for the first time) and probably the unspoken recognition that most Australian voters thing in two-party terms.

I'm not sure there is enough data to test whether these are bad assumptions on my part, or South Brisbane is another outlier, but going forward it is important to not dismiss the possibility of a Greens win in a new seat.

New Zealand, Old Mistakes

NZ was the first in our rapid-fire electoral predictions, with Queensland going to the polls next weekend and the US Presidential in the following week. My apologies, also, to the ACT who I mistakenly thought was in November but occured simultaneously with New Zealand; still, they've got Hare-Clarke STV across multi-member electorates so the territory's got their ACT together.

The Electorates

The hasty formulation of my NZ predictions also introduced an unforced error: my list of electorates was out of date, building on last election's results without taking redistribution into account. Several seats I predicted did not exist this year, and several new seats were created. Fortunately, several of these new seats were broadly comparable to previous, abolished seats: Give or take some small areas Dunedin North was replaced by Dunedin, Helensville was replaced by Kaipara ki Mahurangi, Manukau East became Panmure-Ōtāhuhu, Clutha-Southland dropped the Clutha- to become Southland, Dunedin South is now Taieri and Rodney became Whangaparāoa.

These are marked below with an asterisk, but I have carried over my predictions.

ELECTORATE
INCUMBENT PREDICTION RESULT
Auckland Central  National Labour Green
Banks Peninsula New Seat None Labour
Bay of Plenty  National National National
Botany  National National National
Christchurch Central  Labour Labour Labour
Christchurch East  Labour Labour Labour
Coromandel  National National National
Dunedin  New Seat Labour* Labour
East Coast  National Labour Labour
East Coast Bays  National National National
Epsom  ACT ACT ACT
Hamilton East  National Labour Labour
Hamilton West  National Labour Labour
Hauraki-Waikato  Labour Labour Labour
Hutt South  National Labour Labour
Ikaroa-Rāwhiti  Labour Labour Labour
Ilam  National National Labour
Invercargill  National Labour National
Kaikōura  National National National
Kaipara ki Mahurangi New Seat National* National
Kelston  Labour Labour Labour
Mana  Labour Labour Labour
Māngere  Labour Labour Labour
Manurewa  Labour Labour Labour
Maungakiekie  National Labour National
Mt Albert  Labour Labour Labour
Mt Roskill  Labour Labour Labour
Napier  Labour Labour Labour
Nelson  National Labour Labour
New Lynn  Labour Labour Labour
New Plymouth  National Labour Labour
North Shore  National National National
Northcote  National Labour Labour
Northland  National National National
Ōhariu  Labour Labour Labour
Ōtaki  National Labour Labour
Pakuranga  National National National
Palmerston North  Labour Labour Labour
Panmure-Ōtāhuhu New Seat Labour* Labour
Papakura  National Labour National
Port Waikato New Seat None National
Rangitata  National Labour Labour
Rangitīkei  National National National
Remutaka  Labour Labour Labour
Rongotai  Labour Labour Labour
Rotorua  National Labour National
Selwyn  National National National
Southland New Seat National* National
Taieri New Seat Labour* Labour
Takanini New Seat None Labour
Tāmaki  National National National
Tāmaki Makaurau  Labour Labour Labour
Taranaki-King Country  National National National
Taupō  National National National
Tauranga  National National National
Te Atatū  Labour Labour Labour
Te Tai Hauāuru  Labour Labour Labour
Te Tai Tokerau  Labour Labour Labour
Te Tai Tonga  Labour Labour Labour
Tukituki  National Labour Labour
Upper Harbour  National National Labour
Waiariki  Labour Labour Maori
Waikato  National National National
Waimakariri  National National National
Wairarapa  National Labour Labour
Waitaki  National National National
Wellington Central  Labour Labour Labour
West Coast-Tasman  Labour Labour Labour
Whanganui  National Labour Labour
Whangaparāoa New Seat National* National
Whangarei  National National National
Wigram  Labour Labour Labour

Banks Peninsula, Port Waikato and Takanini are made up of the remnants of several former electorates and no prediction can be honestly recorded against these. Excluding these three seats there are 69 seats of which 61 we predicted rightly: 88.4%. Including these three as failed predictions, success falls to 84.7%.

Normally these wouldn't be great results, but first-past-the-post is somewhat more volatile and unpredictable if the system can sustain more than two credible parties (see: New Zealand, United Kingdom; compare: United States of America).

Banks Peninsula, Port Waikato and Takanini (as noted) were incorrect predictions purely (and technically) by virtue of not being considered. The other eight incorrect predictions are Auckland Central and Waiariki where smaller parties unexpectedly triumphed (Greens and Maori respectively); Ilam and Upper Harbour where Labour won safe National seats; and Invercargill, Maungakiekie, Papakura and Rotorua where Labour failed to unseat vultnerable Nationals despite a strong swing to them.

Auckland Central and Waiariki both had targetted polling that was upset by the results last weekend. That being said, targeted polling like this is normally a good sign the race is worth watching.

Ilam is an unexpected win for Labour. Since it was created in 96 it has been won by the Nationals; in 2014 Nationals had over 50% of the total votes and more than a 10,000 vote lead over Labour; in 2017 Labour didn't even come second. Upper Harbour likewise has always been a National seat, though admittedly much younger, and in 2017 nationals had an almost 2:1 lead over Labour and over 50% of the votes. To see these seats fall must have been a shock for the Nationals and the cherry-on-top for Labour in a genuinely historic win.

Invercargill and Papakura were bold predictions in hindsight. Papakura has been National since creation in 2008, as was its precedessor 1978-2003. Invercargill, although it has a long and mixed history, has been National for a decade and a half. Nationals won with over 50% of the vote in 2017 in both seats. Labour was definitely closer here last election than in Ilam or Upper Harbour, but these seats are sticky. Looking at these facts would give reason to pause on the more general, nation-wide predictive methods but the simple fact is that such individual races are too time consuming to study at this level in each seat. Invercargill was called hastily. Papakura was simply a mistake.

Maungakiekie and Rotorua, on the other hand, should probably have flipped. They have a strong National vote in recent history, but they were close in 2017 and the swing should have brought them back to Labour. These seats are two of those abberations determined by the campaign on the ground despite national trends.

The List

The prediction for the final composition of the NZ parliament was:

Labour -- 61 seats
National -- 41 seats
ACT -- 10 Seats
Greens -- 8 seats
 
This was alsmost blown apart by Waiariki, where the election of a Maori party candidate made the party eligable for several seats. As it happened the national vote was so low that no additional Maori seats were won.

The actual result was therefore:
 
Labour -- 64 seats
National -- 35 seats
ACT -- 10 Seats
Greens -- 10 seats
Maori -- 1 Seat

Compared to the prediction this saw worse peformance from National (▼6) and improved results for Labour (▲3), Greens (▲2) and Maori (▲1) parties. ACT performed as expected.

Quantifying this success is dificult; it would be oversimplistic to say that 6 seats predicted for Nationals were won by other parties, and therefore 114 seats (95%). The Mixed-Member Proportional representation used in NZ doesn't really work that way; even small changes in the voting number can cascade through the quota calculations and change the order that seats are won, technically making a number of unpredicted appointments that will tend to balance out over the count.

Instead, it is best to measure this result qualitatively; the prediction was that Labour would, for the first time for any party under this system in NZ, win a majority government. This turned out to be correct.
 
Another useful metric, if numebrs are your thing as much as they are mine, is to look at the difference between seats predicted and won for each party as a percentage of the prediction; since the final composition is based on national proportions this allows a rough measure of predicted % support against actual % support.

For example, Labour won 64 seats but was predicted to win only 61. This is a difference of ▲3 and 3/61 x 100% = 4.9%, or a 4.9% error, or a (100% - 4.9% =) 95.1% accuracy

Results

Electoral predictions:
61 correct/69 predictions = 88.4%
61 correct/72 possible seats = 84.7% (Three seats mistakenly not predicted)

List predictions:
Labour to form NZ's first majority government under MMP: correct
Labour ▲3 on a predicted 61 seats: 95.1% accuracy
National ▼6 on a predicted 41 seats: 85.4% accuracy
ACT correct prediction of 10 seats: 100% accuracy
Greens ▲2 on a predicted 8: 75.0% accuracy
Maori 1 Seat won when none predicted: 0% accuracy