Thursday, December 25, 2014

Victoria: Final Results, Poll Performance And 2PP Post-Election Pendulum

With results well and truly finalised here's a final wrap-up of various matters from the Victorian election.  This article includes a 2PP pendulum.  I've decided to publish one as a complement to Antony's 2PP pendulum here and I've done so for a reason.  In coverage of the Queensland election it's been common to see estimates of seat gains that are read off non-2PP pendulums, without considering who the seat contests are between.  I can't stop people wrongly concluding that a smallish swing to the Liberals will win them Richmond and Brunswick, but at least it won't be my fault.

I already covered a number of big-picture issues about the Victorian result in my day-after wrap.  This article covers all those aspects that depend on the final results.

Final vote share results

The final primaries for the Lower House were Coalition 41.99%, Labor 38.10, Green 11.48, Others 8.43.  By 2013 election preferences, this would have been a 2PP result of only about 50.7% to Labor, but the 2PP result ended up being 52%. (Officially, 51.99% - that 0.01 will matter a lot to some of those who betted on the 2PP range).

Many pollsters (including the final Newspoll, ReachTEL and Galaxy) got the 2PP more or less spot on (52) by last-election preferences in their final poll.  However as with the 2013 federal election this was a case of errors cancelling out: they had the Coalition primary too low but the preference flow to Labor strengthened markedly.  In all Labor went from getting about 64% of all third-party preferences to about 69.5%. (Ipsos respondents said they would gave Labor about 75%).

Seat results, swings and personal votes

There was some silly stuff about personal votes and supposed invalidity of the sophomore effect theory on Twitter on the night.  The silly stuff seemed to be based on misrepresentations of what the theory actually predicts (though it wasn't coherent enough to tell), but any case for it weakened further in post-counting anyway. The state swing from Coalition to Labor was 3.57 points and based on the uniform-swing pendulum model, Labor would have been expected to win the 2PP in every seat up to Forest Hill, for a total of 50 2PP wins.

In fact Labor came up two seats short of that, missing out in Forest Hill (double sophomore Coalition seat) and losing Ripon (ALP sitting member retired).  A modest difference, and they also missed Prahran by a whisker, but another case of the uniform-swing pendulum model falling slightly short in seat predictions, apparently because of personal votes.

The average swing in the Coalition's double-sophomore seats (seats their sitting member won from Labor's sitting member in 2010) was 1.4% lower than the average total, which equates to about 2% lower once the impact of the redistribution is considered (some seats were not covered fully by personal votes).  I'd previously calculated the long-term average at 2.3% (though that includes some elections with redistributions too) so all up the impact of personal votes looks to have been slightly lower at this election than on average.

The swing to Labor was actually non-uniform, as can be seen on this graph showing 2PP swings to Labor compared to swings required for Labor to win the seat:

A few seats of interest have been marked: R - Ripon, P - Prahran, M - Morwell, S - Shepparton, Mi - Mildura, FH - Forest Hill.  The diagonal line is not a line of best fit; it marks the boundary between Labor and Coalition 2PP wins.

Overall there was a weak corellation between the seat's existing 2PP to Coalition and the size of the swing to Labor, which applies even if the outliers Morwell, Shepparton and Mildura are removed.  This meant that while swings to the Coalition happened in eleven seats, seven of them were wasted in reasonably safe Labor seats, and only one (Forest Hill) made any difference.

The question of what 2PP figure Labor needed to win was widely discussed in the leadup to the election. The pendulum implied Labor would win with about 49.5% while Peter Brent and I both estimated, based on personal vote issues,that something like 50.5% was a more likely break-even point.  A figure of 49.9% is now being quoted based on the fact that if one deducts 2.1 points from the swing to Labor in every seat, then Labor wins the 2PP in 45 seats (including Melbourne won from them by the Greens.)

In fact had Labor won with that 2PP they would have been extremely lucky, since they would have won four seats by less than 1%, including one by very little, without losing any close contests.  The idea that if the state swing had been 2.1 points less, the swing in individual seats would have been 2.1 points less in every case is completely artificial, so the final results don't really show that 49.9% would have been enough - only that it could have been in theory, which was never in dispute anyway.  (The break-even figures Peter and I offered were not figures that Labor needed to win, but figures we thought would give it a 50-50 chance.)

I do, however, suspect that based on the slightly smaller impact of personal votes in this election, and possibly on other issues like the tendency of the sandbelt seats to move as a block, that my pre-election break-even estimate for Labor was a few tenths on the high side, and that 50.2% or so should have been enough.

2PP Pendulum

The following is a two-party preferred post-election pendulum for the 2014 Victorian election. If a seat was actually non-classic (ie the final candidates were not ALP vs L-NP) then the 2CP for the real final two candidates is shown in brackets.

Some comments:

1. The Greens finished second in the seat of Preston but the Liberal preferences weren't distributed because Labor had a majority over the Liberal and Green candidates.  I have estimated the Liberal to Labor preference flow based on the average of the four seats in which Liberal preferences were thus distributed.

2. The seat of Prahran breaks any pendulum it is placed on, because it is so closely three-cornered that primary vote swings between any of the three parties and any of the others can alter the result without any two-candidate swing between the other.  While the seat is ultra-marginal on a Labor-Liberal basis, a 2PP swing of exactly 0.03% would just result in Labor winning the 2PP but still being excluded in third place on primaries (a 2PP swing of about twice that would put them second and win them the seat if the Green vote did not change).

An additional issue with Prahran is that while the primary votes and the Liberal vs Green result have been fully rechecked and even recounted, the Liberal-Labor 2PP result has not. The changes in primary vote during the recount suggest that the Liberal-Labor 2PP was even closer than the 25-vote official margin, but Greens scrutineers believed this wasn't actually the case and the margin would in fact have blown out if rechecked further.

3. It looks like Labor's majority is very fragile in that a uniform swing of 0.8% would result in it losing three seats and depending on Green or Independent support to pass bills.  In reality there would be personal vote effects in the three closest seats so they are not as marginal as they appear.  I'll wait to see if there is another redistribution before estimating the target swing figures for (i) Coalition majority government (ii) Labor minority government.  Long way til the next election after all.

4. Already it will be difficult for Labor to pick up large seat numbers next time around without a large swing to them.  Of the three closest seats, one (Morwell) will probably swing back next time, while two others will have personal votes for new incumbents.

Pollster performance
At the 2013 election I compiled a set of pollster ratings taking into account final polls, 2PP tracking and seat poll performance.  At this election there were very few seat polls released, and most that were released were commissioned.  For instance:

* Galaxy polls of Bentleigh and Buninyong had the Coalition primary two points higher than actual and the 2PP 2.8 and 2.4 points better for the Coalition than actual.  This meant that the Coalition was shown as leading in Bentleigh, which was won narrowly by Labor.

* Essential polls of Mordialloc, Frankston and Bellarine for the ALP-aligned Trades Hall Council correctly projected all three as Labor wins and were rather accurate by seat-poll standards.  The 2PPs for Mordialloc and Bellarine were bullseyes while that for Frankston was three points too generous to Labor.

* Lonergan polls of Melbourne and Richmond for the Greens correctly projected the Greens' win of Melbourne (margin just over half a point over actual) but missed by nearly six points in Richmond, wrongly projecting the Greens winning the seat.

The following are my ratings for the final polls using the same method as in my federal election assessment:

Especially close results in blue.  The lower the RMSQ score - which is heavily weighted towards an accurate 2PP score, however calculated - the better.  There was no spectacularly good final poll (none of them beat the aggregate, despite it having the 2PP 0.3 points too low) but most pollsters did pretty well. 

I didn't keep detailed enough records of all changes in my aggregated 2PP to measure tracking through the campaign, and it is hard to measure it when most pollsters produced just two or three polls each.  But if I'd done so I think Newspoll's overall result would be a bit weaker than its final poll result, while Morgan's disparity compared with the other polls would increase based on its extreme volatility and terribly inaccurate Green vote estimates.  Overall I would extend Adrian Beaumont's assessment that "there was no clear winner from the final Victorian polls, but Morgan is a clear loser" and say that this applies to the whole election campaign and lead-up. It's odd that Morgan's purely SMS state poll is displaying these issues because its national SMS/face-to-face hybrid polls have been rather stable.

Upper House
The Legislative Council result was a microcosm of the 2013 Senate contest (which was run using a very similar system).  Micro-parties proliferated with the Coalition losing five seats - two to Shooters and Fishers, one each to the DLP, the Sex Party and Vote 1 Local Jobs.  Labor also lost two to the Greens.

While many of the group-ticket preference flows seemed logical (right-to-right and left-to-left) that's oversimplifying things and there was one especially illogical preference flow.  The Liberal Democrats and Democratic Labour Party might be both considered "right", but apart from a certain degree of anti-Green sentiment they have very little in common ideologically.  The LDP supports economic and social liberties and the DLP is not too keen on either.  

In Western Metro the Liberal Democrats polled 5.5%, again as a result of benefiting from being placed to the left of the Liberal on the ballot paper.  The DLP polled 2.6% but moved ahead of the LDP on above-the-line votes from a collection of mostly right-wing micros.  At this point, with Labor, the DLP and the Liberals competing for one vacancy, the Liberal Democrat preferences went to the DLP - which makes no sense at all given that those LDP voters who were not just confused Liberal voters would mostly have preferred the Liberals to the DLP.  Had the Liberal Democrat votes gone to the Liberals, the DLP would have been excluded, most likely electing the Liberals.

An aberrant result that has caught the eye of many in the pseph community was in Northern Victoria, where vote-value distortions caused by the woeful Inclusive Gregory system diddled the Country Alliance out of a seat actually won by Labor.  As the Country Alliance were only in the mix because of group ticket voting it was an amusing (though not for the Country Alliance) case of multiple problems with the electoral system cancelling each other out.  

The Upper House share of votes won by micro-parties roughly doubled to 19.6% at this election.  In Adrian Beaumont's review, he suggests that if Victoria used a NSW-style optional preferencing system and did not use group voting tickets then none of the micro-parties would have won.  I think that's likely true based on the votes as cast.  However, the use of an optional preferencing system with such high seat quotas (a key difference to NSW) would have caused many of the micro-parties to merge or at least run merged tickets, and as a result some might have won under such a system.  

The Victorian results do further underline the difficulty of devising a truly fair upper house system that is broken into electorates that return small numbers of members each.  A system in which a low-effort vote (1 above the line) flows as a preference to others makes it too easy for under-scrutinised micros to be elected on the preferences of voters who have no idea where their preference is going.  A system in which the low-effort vote exhausts, however, which I've argued to be the least of the available evils, makes it rather hard, because preferences only flow between micros where the voter makes an effort, and a lot of voters don't.  Any system that eliminates the effort differential (eg by scrapping above-the-line altogether and requiring a set number of boxes) leads to increased voter confusion and informal voting, and greatly increased data entry costs.  

The best solution at state level is to use the NSW system (minus its unnecessary random sampling) but to have members elected with relatively small percentage quotas, as NSW does. (NSW uses a statewide system but elects only half at each election.) However, the Senate does not currently have that option, and many states would need constitutional change to adopt it at state level.

With that, it's time to bring my Victorian coverage to a close and I will soon be moving on (probably not in quite so much detail) to NSW and, depending on when it is held, Queensland.  (There is some risk of me missing the Queensland election if it is called early.)

1 comment:

  1. Kevin, Thanks for your erudite and well reasoned contribution to the greater good.