Sunday, November 30, 2014

Victorian Election Wrap: Federal Drag Strikes Again

Expected seat tally: ALP 47 L-NP 38 Green 1 Ind 1 (1 unclear)
Unclear seat: Prahran (Lib vs ALP vs Green)
Seats in minor doubt: Morwell (Nat leading ALP), South Barwon (Lib leading ALP), Bentleigh, Frankston (ALP leading Lib), Brunswick (ALP leading Green by large margin), Melbourne (Green leading ALP), Shepparton (Ind leading Nat by large margin)

Note: Following this article more serious doubts developed about Melbourne because of changes in vote totals (counting errors most likely).   The seat is now close but if 2010 patterns of non-booth voting are repeated the Greens will still win.  I've also put Shepparton in the "minor doubt" category because of the unusual nature of the campaign and the slim possibility the Independent candidate (now leading with 53.5%) will do very poorly on non ordinary votes.

I have a separate thread on Prahran.


Daniel Andrews has led the Australian Labor Party to a narrow majority win in the Victorian state election, defeating the Liberal-National Coalition led by Denis Napthine after a single four year-term.
As of the close of counting on election night the ALP appears, from a base of 45 seats, to have gained all four sandbelt seats (Carrum, Mordialloc, Bentleigh and Frankston) but lost Ripon to the Coalition and probably Melbourne to the Greens.  The Coalition has almost certainly lost Shepparton to independent and family lawyer Suzanna Sheed (who announced her intention to run a mere four weeks ago!) and Prahran is a fascinating mess with all three parties seriously in the race.  With the huge number of outstanding early votes, some seat leads in the 1-2% range could still be overturned, but this won't be easy, and at the close of election night Prahran is the one seat that is in very serious doubt.  I'll have a post-count thread up for the Legislative Assembly seats later today.

(Note also that I have placed the seat of Brunswick in the minor doubt category despite Labor having a seemingly crushing lead.  The reason is spectacular differences between the booth-plus-postal vote and the early-plus-absent-plus provisional vote in 2010, that Michael McCarthy has drawn my attention to, and now written a post about.  If these were repeated in 2014 with the same swing as on votes so far, the Greens would probably just win the seat.  I'm expecting they will not be repeated to the same degree, but the seat may still close up a lot.  Curiously these differences were much less pronounced in Melbourne and Richmond, but there was enough in them to suggest the current closeness of Melbourne may be a bit artificial.)

Predictable Result

Before I get into all my usual fine technical details I want to explain the big picture of why this ALP win is not a surprising result.  It is unsurprising not just because the polls showed strong Labor leads for so long (and so steadily even when the lead did narrow) but also because it was very much the result to be expected based on electoral history.

The main reservation people had about predicting this result in advance was a view that first-term state governments don't often lose.  As Nate Silver tells us, hedgehogs (predictors who think one big thing is more important than anything else) make worse predictors than foxes (who think that lots of little and medium-sized things are relevant.)  In this election, the big hedgehog idea was that first-term governments almost never lose.   In fact, first-term state governments lose often enough (22% in the past sixty years nationwide prior to this election) that that was a far from convincing argument, especially as lacking a majority and changing leaders increase the chances of defeat.  But a far bigger factor than first-term status was federal-state drag, explored here in July in Victorian Liberals, Going, Going ... and more definitively in What Kills State Governments: Age Or Canberra.

Being of the same party as the party in power in Canberra is a massive weight around the neck of any incumbent state government, especially if that federal government is unpopular.  The article showed that a very simple linear model using just the age of a state government and whether the same party is in power federally predicted, without any polling input at all, that the government would lose about six seats (it has obliged).  Given that it had no majority to begin with, electoral history was telling us that when a government that barely won the last election goes to an election with the same party in power federally, it should expect defeat even if it is a first-term government.

Since that the federal government is unpopular as well, it can be argued that the Liberals did well to be merely beaten and not thrashed.  Their first-term status contributed to that, as perhaps did risky ALP policy tactics, and as in my view did the strong Victorian economy and the better-than-expected leadership of Denis Napthine.  But on the other side we find high unemployment and severe internal party turbulence.  The Victorian Liberals need to review their candidate screening practices after the enormously damaging Shaw debacle and two embarrassing disendorsements during the current campaign.

In the SMH Mark Kenny writes that normally "Voters understand the delineation between state and federal governments and are loath to waste one trip to the ballot box pointlessly ventilating grievances about the other".  My research shows the absolute opposite:  it is normal and systemic for voters to take the federal government into account when voting at state level, and it is probably also far from pointless.  Voters want their state government to be willing to stand up to the federal government and counteract its mistakes.

You can still see the single-term theory being overrated after the event.  Mention is made of how Steve Bracks went from a scraped victory to an easy win next time around (and other state examples of the same include Beattie, Carr and Rann).  But Bracks (and the others) did it with the other side in power federally.  Another problem with a lot of the over-focus on first-term status is that people thought it was a big thing that there had been no other one-term government in Victoria since 1955.  But in that time there had been seven in other Australian states.  Victorian electoral history since 1955 had contained just four new governments, and a sample size of four is not enough to draw conclusions on.

It is possible that this government would have lost even with a Labor government in power federally, but I believe it would have got back in despite the Shaw shenanigans.  We'll never know.

Note: none of this is to say that the Napthine government was going to lose no matter what just because of the Abbott government's existence.  We can imagine a world in which there were no Geoff Shaw issues, no unemployment problems and a perfect public transport network and easily argue that the government would have been re-elected anyway.  Some state governments have had swings to them despite being of the same party as the federal government.  The other factors all contribute, but federal-state drag is probably the biggest one.

Results And Modelling

At various stages in the campaign my seat model showed from 47 to 49 Labor seats (including Melbourne) and 41 Coalition seats (including Shepparton lost to an independent).  The final model was 47 to Labor off a 2PP of 51.7% and primaries of L-NP 41 ALP 38.6 Green 12.1 Other 8.3.  (The second-to-last version, which in my experience is usually better than the final one, was 48 to Labor off 52.6% with very similar primaries.)

Currently the ABC has primaries of L-NP 41.7 ALP 38.8 Green 11.2 Other 8.3.  (The VEC has different figures because it is, for some presumably insane reason, including informals in the primary figures count). It will be a long time before we know the full primary totals and a very long time before we know the final 2PP.  I currently estimate the 2PP, based on the swing in the 82 "classic" 2PP seats, as 51.9% to Labor (a swing of 3.5%) but this could well rise into the mid-52s, given that many postal votes have been counted but remaining early, provisional and absent votes have not.

It looks like my 2PP estimate was on the slightly cautious side for Labor compared to the reality, but there won't be a huge deal in it.  At this stage it looks like a slight underestimate of the Coalition primary by the pollsters has been more than compensated for by a stronger flow of Green/Others preferences to Labor.  Something similar happened in the federal election but in that case these two forces exactly cancelled each other out.

All up the model I used appears to have predicted all bar either one or two ALP-vs-Coalition seat contests correctly, failing in Bentleigh (which it had as just about a tossup anyway) and perhaps Prahran.   The non-classic results in Melbourne and Shepparton were outside its scope.  The lesson of the model's sole more or less definite miss appears to be either that seat polls are crap, or that even Galaxy had to release an outlier sometime before the heat death of the universe.  On the other hand the model successfully picked the loss of Ripon, and did so by use of historic data on sitting member effects in rural seats.   I will review the model more carefully once all the data are in.

It is looking likely that seat betting odds will have been very successful at this election (unlike the federal poll), and indeed this time around they seem to have outperformed my model by winning the one seat that it differed from them on.

Poll Accuracy

It is too early to review the accuracy of the polls definitely.  However it is looking like a repeat of the federal election in that most pollsters got the right 2PP but for slightly the wrong reasons - the primary gap from the Coalition to Labor looks larger than expected, but preferences are cancelling that out.  Currently the Liberals have a primary lead of 2.9 points with ReachTEL (1.4 points) closest to that pin.  Ipsos may have the most accurate Liberal primary vote if the ABC's projections of the primary vote are accurate, but at the cost of parking four points of Labor support with the Greens.  Newspoll had the closest final Greens vote.

The final Morgan and Ipsos both overestimated the primary gap between the parties, but the other pollsters all slightly underestimated it.  So while my aggregate downweighted these two pollsters because they were a new pollster for Australia in one case and a more or less new method with suspicious properties in the other, their data (as downweighted) did make a useful contribution.

Concerning the Green vote, for much of this election part of the site header here read "May argue that it is easier for a camel to pass through the eye of a needle backwards than for a pollster to underestimate the Green vote." It's a very constant theme in Australian polling.  In this election it appears that all 17 polls published in the last six weeks overestimated the Greens, by amounts varying from about a point (final Newspoll and October Essential) to at least eight points (the two worst Morgan SMSs).  I corrected the Green vote using data on overestimation from the last federal election, but the problem was apparently even more severe at this one.  It is possible the Green vote will get up to 12% based on strong Green performance on absent and early votes though, so it might be only 15 out of 17 were over.

All up it doesn't look like there will be a single clear winner of polling honours, but the endless procession of 52:48s were on the money and the very erratic behaviour of Morgan SMS was not.  If the 2PP ends up under 52.5 then the closest final 2PPs will have been produced by Newspoll, ReachTel, Galaxy and respondent-allocated Ipsos, but if it goes over 52.5 then the somewhat maligned 794-vote Essential sample from 7-24 November will claim the prize.  It will be a long time before the final 2PP figure is available.

Generally, this was another good election for opinion polling, and another bad election for those who want to argue that you can't trust polls.

Upper House

I have not had time to look at the Legislative Council closely but in the meantime you can see William Bowe's coverage.  I have not decided whether to put up post-count threads for the Legislative Council yet but it looks like the new government will inherit a Senate-style mess, needing the votes of Greens and a messy crossbench of unclear composition to pass legislation after tiddler parties scored a combined 18% of the vote.  There is a very long way to go in those counts indeed based on the percentage of the vote counted and the impact of below the line voting, and ABC Calculator projections should not be trusted without very close scrutiny of the cutups and figures concerning below-the-line voting rates (which we don't have yet).

No comments:

Post a Comment