Wednesday, November 18, 2020

Queensland 2020: Final Results And Poll Accuracy

 Queensland: ALP 52 LNP 34 KAP 3 GREEN 2 PHON 1 IND 1

2PP Estimate 53.13% to Labor (+1.9% from 2017)

Another Queensland election is over.  In 2017 I wrote that the 2017 election had been "one where a great many dramatic things could have happened, but virtually none of them did", and in some ways this one has been similar.  Nonetheless, the Queensland election has again thrown up more than its fair share of electoral curiosities.  

Historic patterns

This election yet again showed that state and federal politics are fundamentally different and that projecting state elections from federal elections (just because it's easy) is false consciousness.  The 2PP result was over eleven points different from the 2019 federal election in Queensland.

In the leadup to the election I was curious about whether not being in government federally should provide an ongoing boost to the Palaszczuk Government so I wrote this.  Based on the age of this Queensland government and the fact that Labor is in opposition federally, the average expected result was a net gain of 2.5 seats.  The actual result, after two very close seat wins and one close loss, was a four-seat gain, so very close to the historic expectation.  The government was helped, perhaps decisively, by the pro-incumbency mood during COVID-19, but had also had some wear and tear during the term.  By election day the government was polling very well in terms of personal approvals of Annastacia Palaszczuk and which party was best to handle the economy, and it seems these polls were telling us something the voting intention polls were not.

2PP and swing

There are many different approaches to estimating the 2PP for this election because of the incomplete information provided by the ECQ.  For the 2017 election the method I used was what I call the "actual distribution method" - attribute as many votes as possible to 2PP based on the actual distributions and then use the average preference flows when those other parties still holding votes were excluded in other seats to estimate preference flows for those parties.  I could do that again, but the ECQ has made it difficult by not including party names on the same page as the preference distributions.  

The method I have used instead is:

* Add the exact totals for the 85 seats that finished as 2PP contests.

* For the remaining eight seats, distribute primaries to 2PP using the sampled distributions, which were done by ECQ in 70 2PP divisions and posted by Antony Green.  For parties not included in Antony's tallies, I use the sampled distributions published by Poll Bludger.  

This gives a result of 53.13% to Labor (+1.9 swing from my 2017 estimate.) (1523953 votes to 1344371).  Antony Green's estimate is 53.2% to Labor (1524766-1343558, which comes out to 53.16)  In any case, Antony and I agree that there was a 1.9% swing, as Antony's estimate for 2017 was 51.3%. In 2020, it is unlikely the treatment of the non-classic seats matters too much as their overall effect on the 2PP is very small.

Few seats changed hands (again)

After the bloodbaths of 2012 and 2015 the next two Queensland elections have been quiet in seat transfer terms, and this one was the quietest since 1980.  Only nine seats changed hands on a notional basis in 2017, and this time only six did, excluding the LNP's recovery of Whitsunday from Jason Costigan after kicking him out of the party.  The LNP lost Pumicestone (0.8%), Caloundra (3.4%), Bundaberg (4.2%), Nicklin (5.3%) and, on a monster swing by the standards of a low-swing election, Hervey Bay (9.1%).  Labor lost South Brisbane (3.6% vs Green) to the Greens, but had an average 3.5% swing in its favour in its 14 most marginal two-party seats, with only one very small swing against.

The overall 2PP seat tallies are exactly as predicted by the simpler version of my 2PP model, given the 2PP vote that actually occurred.  Labor overperformed by one seat compared to the Mackerras pendulum method and underperformed by two seats compared to the regional swing version of my 2PP model (the difference in swings between the Gold/Sunshine coasts and the rest of the state was not as large as polling predicted, mainly because a predicted small swing against Labor outside the southern coasts did not occur).  However, seat losses were quite widely strewn across the pendulum, with the LNP losing its 3rd, 8th, 12th, 16th and 23rd (!) "safest" seats on paper out of 39.

Personal votes: oddities, but the theory holds

This election was a strange one from the perspective of personal votes in seat results, because the LNP lost all their seats where a sitting member was retiring, but also the only two seats which they gained in 2017 and where a new sitting member was recontesting.  However, on overall averages the election provided extremely strong support for 2PP personal votes (which I should remind the sceptics are a theory and a fact).  

By my calculations in six "double sophomore" seats (where the party has a new incumbent who defeated the opposite party's incumbent at the previous election) the party holding the seat on average outperformed its 2PP state swing by 2.0%.  In eleven "single sophomore" seats (where the sitting member was first elected in 2017, either because the seat was new or replacing a retiring member) the party on average outperformed its state swing by 0.6%.  In five cases where a sitting MP who had been re-elected at the previous election retired, the party on average underperformed its state swing by 3.3%. It can be argued that LNP retirements happened to be focused in an at-risk area of the coast, but speculation that the LNP would lose marginals on the Gold Coast where it had no retirements came to nothing. There was also one seat (Currumbin) that was almost a vacancy because the new incumbent had had only half a year to establish herself, and the LNP underperformed by 0.9% in this seat.  (ABC estimates have been used for 2PP swings for seats that had no official 2PP in 2017.)

It was just the LNP's bad luck that in their sole "double sophomore" seat of Bundaberg, the 2017 result had been artificially bolstered by One Nation preferencing them and polling a very high primary vote, and therefore the seat was more at risk than it appeared.  

There were some very strong cases of new MPs building a personal vote in divisions created in 2017, such as the LNP's Sam O'Connor in Bonney (+8.4%) and Labor's Charis Mullen in Jordan (ABC estimate +7.1%).

Almost no seats were decided by preferences

This was also a section last time, but even more so this time! There were only two seats where the primary vote leader did not win on preferences, these being Nicklin (where the LNP started 3.51% ahead and ended up 0.28% behind Labor) and Mirani (where Labor started 0.31% behind Stephen Andrew (PHON) but Andrew got 59% 2PP after an 83.4% two-candidate preference flow from the LNP).  The LNP also very nearly overtook Labor in Bundaberg.  

Of course, there were, as usual, seats where an even split of preferences from a specific party would have altered the result had it happened, but in the case of Greens to Labor it was never likely to.  It's worth mentioning that the UAP (Anna Palmer) preference flow in Currumbin was not among those cases - the LNP would still have won the seat.  The LNP would however have lost Currumbin had both the UAP and Tracy Takacs-Thorne's preferences done so.  (Takacs-Thorne was notionally an independent but in fact a candidate for the unregistered Australia One party, a right-wing conspiracy theory outfit.) 

I get that the LNP won four seats that it would not have won had the split from One Nation been even: Currumbin, Coomera, Burleigh and Glass House (these are based on the ECQ preference samples, not the distributions).  I get that Labor would not have won with an even 2PP split of Greens preferences in any of Bundaberg, Nicklin, Barron River or Cooper.  (In Cooper, despite Labor's massive final margin, Labor was actually overtaken during the preference count prior to the exclusion of the Greens).  Had optional preferential voting been used, the only seat that I think would have had a different result is Nicklin.  

The overall preference flow was about 55% to Labor, up from about 51% in 2017, with the increase being mainly caused by the One Nation vote collapsing.  Greens preferences appear stronger than last time in the ECQ estimates, but this is probably only because the 2017 Green preferencing rates were extracted from the preference distributions and included votes received by the Greens from other candidates.  Katters Australian Party preferences were more pro-LNP this time.  

The One Nation collapse meant that One Nation finished second in only one seat (Bundamba) compared to 20 last time.  In Bundamba, One Nation overtook the LNP by getting more preferences from the Animal Justice Party and Greens, especially the former.  I attribute this at least partly to gender preferencing: the AJP, Greens and One Nation candidates in Bundamba all being female while the LNP candidate was male.  

Minor parties: incumbents v the rest

Something I have noticed before is that once crossbenchers get elected they tend to do very well in subsequent elections even if their party doesn't do so well.  Thus it was again in this election:

* Stephen Andrew (PHON, Mirani), primary vote swing -0.4%, 2CP swing 4.2%

* Michael Berkman (Greens, Maiwar), primary vote swing 13.6%, 2CP swing 4.7%

* Sandy Bolton (IND, Noosa), primary vote swing 12.5%, 2CP swing 4.3%

* Nick Dametto (KAP, Hinchinbrook), primary vote swing 21.6%, 2CP swing 7.2%

Only Stephen Andrew had a primary vote swing against him, and a very small one at an election where his party lost almost half its statewide vote despite running in more seats.  (Andrew, incidentally, became only the second One Nation MP, out of 33 One Nation MPs who have served at state and federal levels, to ever complete a term and be re-elected under the One Nation banner.  The first, twice, was Rosa Lee Long.)

This election saw a decline in the overall minor party and independent vote, from 30.9% in 2017 to 25.4%.  The loss was essentially One Nation's, but none of the others did anything special overall.  The Greens had a small overall swing against them, but this was probably just because there were more parties running candidates.  Katters Australian Party did nothing special outside the footprint of the federal seat of Kennedy despite the One Nation collapse, and a much talked-up independent, Claire Richardson in Oodgeroo, couldn't get into the top two.

The Greens cut the gap to second after preferences in McConnel from 5.8% last time to 2.8% this time, but that was mainly because the LNP performed so poorly there.  They had a remarkable 9% swing to them in Cooper off the retirement of Labor's Kate Jones, and briefly threatened to win that seat, but fell 4% short in the end.  

Their sole pickup was the widely expected defeat of Jackie Trad in South Brisbane.  This has produced more gnashing and wailing from certain members of the ALP faithful (especially federally) than if Labor had won that seat but lost the election.  However it is crocodile tears mixed with cluelessness; Labor didn't reinstate Trad to Cabinet, left her to fend for herself during the campaign, and are looking for someone to deflect blame to.  As to whether Trad would have won had the LNP preferenced her, in 2017 she got 62% of the LNP distribution (which included votes from other parties).  A repeat of that flow in 2020 would have got Trad home 51-49, but any flow below 57.7% would have seen Trad defeated (it was actually 36.2% thanks to the LNP's preferencing decision.)  It's debatable whether the controversy surrounding Trad - coupled with the LNP's votes at exclusion including a slightly larger share of votes from other parties - would have reduced her share of LNP preferences by 4.3% or not, but at least plausible that it would have, so the question of whether the LNP preferencing decision caused Trad to lose is inconclusive.  It certainly robbed her of any chance she might have had.

There is no evidence of a general swing against minor parties caused by COVID-19 in other elections or polling, so the poor One Nation result seems to have been their own doing.  

Early in the count there was a lot of nonsense about One Nation votes moving to Labor, because there was a large primary vote swing against One Nation, a large swing to Labor and very little swing to the LNP.   In the end this appearance moderated anyway, with Labor finishing on a 4.1% primary vote swing to the LNP's 2.2%.  This doesn't mean the votes leaving One Nation (6.6%) went mostly to Labor.  More likely, about 4 points (net) moved from One Nation to the LNP, about two points moved from One Nation to Labor, and about 2 points moved from the LNP to Labor.  In some seats, Labor got a swing despite One Nation getting a positive swing because they had not run in 2017.  Overall, the One Nation preference flow based on actual two-party splits was about the same as the flow for One Nation exclusions in 2017.  That suggests the actual preference flow from One Nation voters to the LNP reduced slightly (because exclusion flows are watered down compared to 2PP splits), and that could have been explained by the party issuing open how to vote cards instead of tending more to preference the LNP.  

Polling: state

State polling at this election (which saw the first test of the new Newspoll/YouGov methods) was reasonably good, but not brilliant:

The four campaign-period polls published all underestimated Labor on a 2PP basis, but by an average of only 1.5 points, the overall miss being typical of state election polling.  However, on the primary votes, every poll underestimated Labor and Others and overestimated One Nation and the Greens.  The overestimate of One Nation is especially surprising as the party contested nearly every seat, removing a common source of polling error in the past.  Shy Hansonites?  Doesn't look like it!  

The YouGov stable polls were more accurate than Morgan, but unluckily the second-last Newspoll was a little more accurate than the final one.

A reasonable read of these polls was that Labor were very likely to govern and very slightly more likely than not to do so in majority.  As such, despite not being entirely accurate, the polls did predict the overall result correctly, and seat projections off them only underestimated Labor by a few seats.  As usual the "inevitable hung parliament" narrative misread the polling and underestimated the impact of a relatively small average polling error on the chance of majority government.

Polling: local

YouGov issued seat polls of Mundingburra, South Brisbane, Mansfield and Pumicestone, with sample sizes of just over 400 voters per poll.  The South Brisbane and Pumicestone polls were remarkably accurate, being only 0.8% and 1.3% out on the two-candidate result, with negligible errors in the primary vote.  The other two were not so good - Mundingburra had Labor 4.4% too low on 2PP and hence incorrectly had Labor losing the seat very narrowly (it had Labor 3.2 points low on primary vote and One Nation 6.3 points too high).  The Mansfield poll had the right winner but a 6.3 point miss on the margin, mainly by having Labor's primary 5.1 points too low and the LNP's 6.6 points too high.  Overall an average error of 3.4% on the 2CP vote across the four polls.  This is no worse (and in the case of 2019, better) than considerably larger federal seat polls, but given that those seat polls were themselves so far out, I don't think YouGov has found the magic bullet for completely fixing seat polling just yet.  

A combined YouGov poll of Currumbin, Redlands and Mansfield was partially released two months before the election and at that time had a 2% swing to Labor in these seats.  This (off a combined sample of 600 votes) was very portentous indeed as the actual swing in these seats was 2.9%.

Four seat polls by Omnipoll (established by staff of the old pre-2015 Newspoll) were reported in mid-September in the Australian.  These polls, part of a commissioned survey for a coal industry client, were extremely inaccurate, in a way very unlikely in my view to be explained by vote-shifting in the meantime.  The seats covered were Ipswich, Keppel,  Thuringowa and Mackay.  These seat polls overestimated the LNP's eventual primary in every seat and by an average 8.5%.  They underestimated Labor in every seat by an average 8.6%.  They underestimated One Nation in every seat by an average 6.5%.  They overestimated the Greens in every seat by an average 2.8%.  They underestimated KAP in Thuringowa by 9%.  The only remotely useful thing these seat polls did was suggest that the One Nation slump might be more severe than the state polls were suggesting, which was true, but not that true.  

Greens seat polls by Lonergan of their three big seats (Maiwar, South Brisbane and McConnel) were reported, but not in full detail, so it was unclear whether these polls overestimated the vote for parties other than Labor, LNP and Green, or had an undecided component included.  The 2CP result released for Maiwar (56-44) was correct.  Overall these polls on average and on a three-party basis overestimated the Greens by about a point and underestimated Labor by 2.5 points.  The primary votes in Maiwar and South Brisbane were very accurate on this basis, but the McConnel poll overestimated the Greens by 3.5 points on a three-party basis and underestimated Labor by 5.3, as a result wrongly putting Labor in third when they in fact led on primaries.  

Long way back for the LNP

The LNP went backwards both in seat terms and swing terms for the second election in a row, a highly unusual feat for oppositions.  The LNP has now won one and a half of the last twelve Queensland elections - one and a half out of five while Labor was in office federally and zero out of seven while the Coalition were.  The LNP has serious problems in Queensland state elections.  It struggles to attract quality leadership (partially excepting Campbell Newman who was an excellent campaigner in Opposition but a divisive Premier).  Its internal culture suffers from an excess of vested interests, domination by backroom boys (the gender reference isn't an accident) and also some remarkably immature social media personnel.  As a female leader Deb Frecklington had the opportunity to present a different style of leadership and to improve the party, but instead chose to attack another woman over her choice of clothes and her failure to have children.  The LNP now needs a uniform swing of 5.7% to take majority government at the next election.

1 comment: