Saturday, May 11, 2019

New South Wales 2019: Final Lower House Results And Poll Accuracy

Amid the gearing up for the federal election, the finalisation of the NSW election figures has been largely ignored.  For whatever reason it took a few weeks longer than last time to get to the final 2PP, which is necessary for assessing which polls performed the best among other things.  Though in this case, comparing polls is hardly worth the bother.

Vote Share, 2PP and Preference Change

The primary votes were 41.58% Coalition, 33.31% Labor, 9.57% Green and 15.54% Others (including 3.46% Shooters, Fishers and Farmers with the rest led by 1.53% Sustainable Australia, 1.52% Keep Sydney Open and 1.51% Animal Justice.) The 2PP was 52.02% to the Coalition, a 2.3% swing away from their 2015 result.

The overall two-party preference flow barely changed at all, with 33.0% of preferences flowing to Labor (-0.6%), 14.0% to Coalition (-0.8%) and 53.0% exhausting (+1.4%).  But overall what happened here was that the flow of Greens preferences strengthened, but this was cancelled out by the Greens' share of all non-major voting falling from just over 50% to just 38%.  In those seats in which Greens candidates were excluded, the preferences distributed at that stage flowed 51% to Labor with just 9% to Coalition and the rest exhausting.  The stronger flow in Lismore, 72%, was among the highest and was crucial to winning the seat, while East Hills would have been extremely close had Labor managed to match the state swing there.

Seat Swings

Very few seats changed hands at this election.  Labor won two seats from the Coalition, as did the Shooters Fishers and Farmers.  The SFF and an independent also retained seats they had won in mid-term by-elections. That's all!

The following is my usual messy chart showing the relationship of 2PP swings required (to win the 2PP vote, which doesn't always mean winning the seat) compared to 2PP swings obtained by Labor:

Colours by outcome: blue for Coalition, red for Labor, green for Greens, brown for Shooters and grey for independents.  Labor had six target seats on 3.2% or less, but the 2.3% swing netted only two of them (in the darkly outlined section), narrowly missing the very marginal East Hills where they failed to get any swing.  A version of the above chart showing separate trendlines for the Labor 2PP and Coalition 2PP seats from the previous election shows how Labor failed to collect seats:

In its own seats, Labor got bigger swings the closer the seat was at the previous election.  But in Coalition-held seats, not only was the average swing to Labor lower by about a point, but also it was especially low on average in the Coalition marginals.  Indeed, East Hills, which appeared to be a miracle save, was actually just on the trendline in this regard!  It seems the Coalition owes its majority to some very effective sandbagging, though they did overkill Monaro ever so slightly.

Part of the explanation is, yet again, that personal vote effect is a theory and a fact.  In Labor's six double sophomore seats (where they had a first term incumbent who had defeated an incumbent at the previous election) Labor got an average swing of 6.2%, 3.9 points above the state average.  But it's also the case that whatever Labor were trying to do to the Nats in their regional seats, it didn't work (except in Lismore where a very long-term Nats incumbent quit and Labor preselected former federal MP Janelle Saffin.)  Labor landed about two seats shy of where they should have for the swing they got, and even that swing was weak compared to how things had looked early in the campaign.

Poll Accuracy

In contrast to the heavily polled 2015 election this one was a joke in terms of the quantity of polling for Australia's most populated state.  By the final week only Galaxy was polling, firstly with a YouGov-Galaxy poll for the tabloids and then with the final Newspoll. A Sun-Herald uComms was seen two weeks before the election, and the last Essential over a month before.  Because these polls were taken so far out from the election, they were inaccurate on both the 2PP and the major party primaries, and especially inaccurate on the Others primaries which they had way too low.  But in the absence of polling by competing pollsters taken in the final week, doing a comparison table to show how far out they were seems a little bit pointless.

Allowing that Galaxy/Newspoll was the best pollster by default, the final Newspoll was still rather impressive.  It had the 2PP at 51% (actual 52.02%), Coalition at 41% (41.58%), Labor at 35% (33.31%), Greens at 10% (9.57%) and others at 14% (15.54%).  At the last election this would have been better than all the final polls except ReachTEL, Galaxy and the old Newspoll.  The final Newspoll was the only poll to have the Coalition actually ahead on the 2PP.

I saw six YouGov-Galaxy seat polls, with an average error of 3.65 points on the two-candidate result (implying a margin of error of about 7 points, which is pretty typical for seat polls).  However these six results included three bullseyes (within half a point) but also a 9-point miss in Ryde and a 7.6-point miss on the Shooters' win in Barwon (where Galaxy had Coalition ahead 51-49).  Probably the changes in state voting intention through the campaign had some bearing here, but Galaxy/Newspoll margins often seem to be clustered around 50-50 compared to the actual results.  Also in Lismore, Galaxy nailed the 2PP but did so off primaries that were nearly 5 points too low for the Nationals.  They got away with it because their preference flows between the Greens and Labor were much weaker than was actually the case, cancelling out the error on the primary vote.

Looking ahead, I could do a 2PP pendulum but see no reason to do so for now as there will be a redistribution before the next election.


  1. for non classic contests look at 2pp... Barwon based on alp/nat 2pp was 52/48 nats

    1. Yes, of the three seats won by Shooters, Barwon had a massive 2PP swing to Labor (hence the brown dot close to the heavy line), Orange had an above average swing and Murray had no swing.

      The main difference there is the behaviour of those Shooters preferences that didn't exhaust, which were fairly even in Orange, about 68% to Labor in Barwon and about 40% to Labor in Murray.

      This pattern occurs in seats where Shooters were excluded across the state too - the average split was very close to even between the two but in specific seats it could be quite strongly one way or the other.

  2. NSWEC actually make data available for every 2CP pairing, which lets you calculate the condorcet winner in each seat (if you want to go full wonk).

    It also confirms that the Nats would have won a Green vs Nat top 2 in Lismore, and only the Shooters were in a position to win Barwon and Murray (despite non affiliated "put Nats last" campaigns in the seats).

  3. Are we even sure any more that YouGov-FiftyAcres was off? I mean, their polls were a bit better for the Coalition, which was also the case at the election.

    1. Yes. They reached the 2PPs they did off absurdly strong preference flows to Coalition from very weak primaries. Their Coalition primary only twice exceeded 34 (in both cases being 36.) Whatever they were doing was very weird.

  4. Another aspect of the Clive Palmer media spend is that availability of media spots (tv, radio, print and billboards) would have dried up and alternatives would have become less visible or prime spots more expensive - this may have caused a dynamic change which favoured parties that did grassroots local campaigning - perhaps Labor were so over confident that the Kool-aid was drunk and they pressure relieved from the pedal....and local campaigning suffered or became less "genuine"....