Saturday, June 25, 2022

Two-Party Swing Decided This Election (Plus Pendulum)

We're now just over a month out from a remarkable House of Representatives election.  There's been a lot of attention on the seat gains by six teal independents and three Greens, and a lot of claims that the old two-party preferred model for elections is broken.  Not the case.  Labor won this election on classic two-party swing, largely because the Coalition's primary vote crashed and Labor's (modest as it was) didn't.  The teal gains were a major story of the election and are a big headache for the Coalition going forward, but they are not where the election was won and lost.  

There is a fair amount of nonsense from some fringe supporters of the losing side about Labor's low primary vote, with claims that it is wrong that a party not voted for by two-thirds of the country should govern.  The problem is that both sides had very low primary votes (the Coalition's being lower than, for instance, Labor's primary when it lost heavily under Mark Latham in 2004) and somebody has to win.  Labor was the clearly preferred choice between the two major parties, and would have won this election easily under any single-seat system, including optional preferential voting and first past the post, though in the latter case tactical voting would have given it a much higher (but much less sincere) primary vote.  Those complaining about Labor winning a majority off such a low primary vote should embrace proportional representation or shut up.  (I may write a detailed article about this sometime.)  

Incidentally, a widespread theory that Labor's low primary was a result of strategic voting for teal independents is also false.  There only were about 17 seats with new teal independents (depending on definition) and these had an average primary vote swing against Labor of 5.2%, meaning there was still a 0.2% primary vote swing against Labor on average in the rest.  Labor's failure to increase its primary vote despite the Coalition's primary vote losses would have had more to do with Labor losing votes to the Greens and One Nation (the latter purely on account of One Nation contesting nearly every seat).  

The national two-party preferred vote currently sits at 52.13 to Labor, a swing of 3.66%.  However, this isn't necessarily final, eg there appears to be an error in the 2PP count in North Sydney, which if fixed would put them up to 52.14.  There will probably be other minor corrections.  What is certainly final is the declared seat count.  Labor (77) gained ten seats from the Coalition but lost Griffith to the Greens and Fowler to independent Dai Le.  The Coalition (58) regained Hughes from defector Craig Kelly (UAP) but lost ten seats to Labor, two to the Greens and six to teal independents.  

In the two seats the Coalition lost to the Greens, they also lost the two-party preferred vote to Labor.  Bearing in mind that Labor to Greens and Greens to Labor preference flows are similar to each other, this means the Coalition lost these seats on two-party swing anyway, irrespective of which out of Labor and the Greens made the final two.  So two-party swing cost the Coalition twelve seats, alone putting it back to 64 seats and out of government without considering the six losses to teal independents.   The Coalition would still have had serious problems had no seats changed hands between the majors (could it possibly have governed with, say, 68 seats to Labor's 67 and the same crossbench?) but the crossbench losses were only by themselves the difference between winning and a mess, while the 2PP losses by themselves turned winning into losing.  

2PP swings largely result from changes in the major party primary vote, and this election was no different, though it will be one of the four elections in the last 40 years to see a moderate shift in preferencing behaviour by voters for established minor parties (cancelling out the moderate shift to the Coalition in 2019).  The overall components of the 2PP swing at this election were (on current figures) differences in primary vote swing between the major parties 2.49%, preference shifting in Labor's favour 0.99% and other factors such as changes in the minor party vote share breakdown or the impact of three-cornered contests 0.18%.

Labor won a majority despite a low primary vote because it is in the nature of our system that a moderate advantage on two-party preferred turns into a convincing advantage on seat share (convincing enough).  This was not understood by such commentators as Chris Uhlmann, who claimed (following a Newspoll with Labor on 36%) that "neither [major party] can form government in their own right from there." Such commentaries relied heavily on media tropes about major parties needing certain votes to win because that was what they had won from in the past, and ignored evidence (as published here, though even I would not have credited a Labor majority off 32.6% as likely) that those historic records were irrelevant.  Labor has won a majority with a primary vote 3.4 points lower than that from which Uhlmann maintained that Labor winning a majority was impossible.  

How Labor Beat The Bear Pendulum

On the classic pre-election pendulum, a swing of 3.66% was projected to net Labor eight 2PP gains, still easily enough to win the election in some form.  Historically that would be a rather meagre return for such a swing because there were unusually few Coalition seats on lower margins.  And given that four seats where Labor won the 2PP were gains for the crossbench, it would seem the swing should only have been good for 73 Labor seats.  But Labor won 77 anyway - how did they do this?

One reason for this is simply probability: a uniform swing of 3.66% would have resulted in six Labor wins by less than 3% 2PP and fourteen Coalition wins in the same range.  But seat swings are never uniform, and simply by treating them as able to vary randomly it would be expected that Labor would get more like eight wins in this range. (They in fact got nine, as well as two above this range, but missed one of the two below, Bass.)

There is more, because Labor also secured a favourable distribution of swings.  In existing Coalition 2PP seats, the average 2PP swing to Labor was 4.29%, with no real relationship with the seat's existing margin.  But in existing Labor 2PP seats, not only was the average lower at 2.73%, but the more marginal the seat, the better Labor did.  The regression for average swings in existing Labor seats was roughly:

swing = 5.27% -0.28*pre-election margin

A third factor assisting Labor was the relatively high variation between 2PP seat swings - a standard deviation of 4.25%.  This made it more likely that seats that were projected as narrow Coalition retains based on uniform swing would fall.

Given the national swing, Labor would have been expected by my seat model (which takes into account random variation) to win the 2PP in 79 of the 145 2019-election major party seats, gaining about eleven and losing about one.  However because of the other advantages mentioned above, Labor in fact became the 2PP winner in twelve of these seats and lost no 2PPs in its own seats.  That is then 81 of those 145 that Labor could have won had it been the 2CP winner, but it was knocked out in third in three of those by the Greens, and lost the 2CP to Dai Le in Fowler, for a total of 77.  (Labor also made a meaningless 2PP gain in Mayo, and had two existing unoccupied 2PP seats in Melbourne and Clark, so overall Labor won the 2PP in 84 seats to 67.)  

Here's a colour-coded graph (click for a larger version):

The blue dotted line is the regression line for Coalition 2PP seats and the red dotted line is the regression line for Labor 2PP seats.  Seats are colour-coded by winning party: red for Labor, blue for Coalition, Green for you'll never guess, teal for "Voices Of" style independents, grey for other indies, orange for Centre Alliance and a sort of pinkish dark red colour for Bob Katter who is hiding behind a bunch of Coalition dots anyway.  

Left of the vertical axis in the top left quadrant are thirteen Labor 2PP gains (ten wins, two Green wins and Mayo).  Just right of the vertical axis in the bottom right quadrant would be the Coalition's 2PP gains but that part of the graph is empty - there weren't any.  (The red dots near to what would be the losing zone are Gilmore, Lyons and Lingiari.  The grey dot that marks the biggest 2PP swing against Labor is Fowler).

Here's the 2019 version.  In 2019 there was a uniform swing to the Coalition in Labor seats, but in the Coalition's own marginals, the Coalition had a swing to it, while in the Coalition's safe seats there was a swing to Labor.  2022 was nothing like as spectacular, but it is still clear that Labor benefited from a helpful swing distribution, getting higher swings in marginal seats and also second-tier Coalition seats than in their own safe seats.

2PP Swing Still A Good Predictor

Before the election I wrote this piece about how 2PP had continued to be a fine predictor of seat share swings even as the crossbench expanded and how a substantial 2PP win for Labor would mean that Labor won the election.  So how did this go with the crossbench jumping remarkably from six seats to sixteen?

The answer is it still went pretty well.  The relationship between 2PP swing and seat share changes is as healthy as ever, though the relationship between 2PP and overall seat share is beginning to struggle. Here's one of the graphs from the previous piece with the 2022 result super-imposed:

The red star is 2022.  It is almost exactly on the line of best fit (which hasn't been recalculated): the Coalition's share of the major party seats has fallen from 53.1% to 43% and that 10.1% seat share swing is almost exactly the 10.4% that was expected for the 2PP swing against it. Now here's the other one (this time I've used a blue star, because this one is colour-coded by party.)

The growing crossbench meant the government did finish up several seats short of the historic projection (again, regression line not recalculated), showing that this (slightly weaker anyway) relationship is weakening to a degree as the crossbench expands.  Even so, it's still not an enormous outlier.  I would expect governments to continue to underperform compared to the trend line on this one for the forseeable future, but still, this graph correctly predicts that for the 2PP recorded, the government wasn't going to survive.

If the major party primary vote continues to decline, the number of crossbench seats is likely to increase, and majority government will become harder and harder to win.  

Another thing to point out here concerns state breakdowns.  I often see claims that the national 2PP is not a reliable guide because there are uneven swings between states.  At this election one state (Western Australia) recorded the biggest difference in 2PP swings from the national 2PP swing in any mainland state since 1943 (though there was a very similar difference the opposite way in WA in 1972).  It makes little difference; unless a state is unusually packed with marginals then what goes up in one state comes down somewhere else.  Labor made three gains above national swing in WA but failed to make two below it in Tasmania.


I've decided to post my own provisional post-election pendulum here, mainly out of dissatisfaction with the way pendulums are being done by almost everyone, including even their inventor (note the positioning of Mayo and Warringah).  I routinely see pendulums with classic 2PP margins mixed in with close Labor vs Green and Coalition vs IND margins irrespective of whether those seats were close on a 2PP basis and even irrespective of whether there will in future even be any independent, let alone the same one, contesting.  This approach to pendulums, treating them as basically just a list of how close seats were, causes confusion and compromises the original purpose.  The purpose of the pendulum is to give a simple model of how many seats will fall for a given uniform 2PP swing.   Sometimes this model is erroneous because of systematically uneven swings, natural swing variation between seats, new non-classic winners or losers and personal vote effects.  However, it's not a bad starting point.  

Why on earth do we see, for instance, the seat of Northcote included on Victorian state pendulums as if it is at all meaningful or useful to put a seat that was Labor vs Green 1.7% but Labor vs Coalition 33.2% in the middle of a pack of Labor/Coalition marginals?  Yes, it's a seat Labor could very easily lose, but the axis on which it would be lost is completely different to the one that will determine surrounding seats on the pendulum.  And also, identifying seats Labor might lose to the Greens is a complicated exercise involving both 2CP and 3CP seats - at this federal election Griffith just appeared on pendulums as ALP 2.9%, suggesting no risk of it falling to anyone in an election with a substantial  swing to Labor.  

Throwing Labor-Greens seats and seats with independents in with classic 2PP marginals leads to confused commentary about how many seats will fall on given swings and I've seen this far too often in media reporting.  I end up making my own versions before each election because the versions available online are mostly  unusable.  I think 2PP pendulums should be noting significant 2CP and 3CP results, but should not let themselves be governed by them.

These are the rules for my own pendulum that follows:

* If the seat was won by a major party, it is ranked by 2PP, whether the major parties were the final two or not.

* If the seat was won by a major party, but finished as a non-classic seat (one or the other major party did not make the 2CP) then the 2CP margin is noted.

* If the seat was not won by a major party, then it is excluded from the 2PP section of the pendulum, the reason being that that the 2PP can only again determine the winner if the seat winner falls out of the top two.  In the case of a crossbencher elected at the previous federal election as such, this hasn't happened since three cases of it happening in 1998, and it hasn't happened at state level since 2013 (indeed there hasn't been a loss by a crossbencher elected at the previous election as such in their own seat since 2014). However the 2PP is still noted.  

* I also note the 3CP swing required to change the result (this applies the same way as 2PP swing, ie how many voters would need to move from one candidate to the other) in cases where the 3CP loser would have won the seat had they made the top two.  And I note it in some other cases where the 3CP was relevant to the seat's marginality:

- S3CP (stands for survival three candidate preferred) - refers to the swing-terms 3CP margin for the winner over another candidate who would have narrowly lost the 2CP had they reached it

- L3CP (stands for losing three candidate preferred) - refers to the swing-terms 3CP margin for the 2CP loser over another candidate who would have narrowly lost the 2CP had they reached it.  It may seem speculative to note such seats at all, but I want to keep an eye on potential future Ryans and Brisbanes.  

There were no cases this election where a candidate who would have won the 2CP was narrowly squeezed out of the 3CP by a candidate who went on to lose.  

My cutoff for inclusion of 3CPs is my estimate being at or under the AEC's marginal seat limit (6%).  I should note here the unusual case of Forrest.  The 3CP Labor over Green margin in Forrest is 5.7%, and Forrest is now 2PP-marginal (oh strange new world ...) but I do not believe Forrest was 2CP-marginal between the Liberals and the Greens given the latter's low primary.

Here we go then!  This pendulum is provisional pending final results (a corrected version will be edited in when all results are final), and I have unofficially anticipated the 2PP correction for North Sydney.  

One thing that is interesting here (and another mirror image of 2019) is the asymmetry in marginals.  Labor has twelve seats under 5% compared to the Coalition's nineteen.  If this survives the next redistribution intact, it will make life even more difficult for the Opposition, which unless it can retake some seats from the teals or Greens would need a 54.15% 2PP to retake majority government with a uniform swing (51.2 to become the largest party.)  If the Opposition retakes all its crossbench losses, it needs 51.2% for a majority.  

I will have other comments about the results - especially polling accuracy - when the party preferencing breakdowns are known and all figures are absolutely final.  

1 comment:

  1. Interestingly, if you "reset" the results to a zero result (50.0 2PP) by shifting everything 2.14 toward the Coalition, Labor loses five seats, but from that point forward you have 20 marginal Coalition seats and 21 marginal Labor seats. But of course the Coalition is outnumbered in total seats 72 to 63 (or maybe 64 if you assume the same swing flips Curtin back), so the proportion is very similar.

    I don't know whether this is the result of intentional districting choices (a kind of "antimander" by the line drawers) or just a coincidence; I just thought it was salutary.


The comment system is unreliable. If you cannot submit comments you can email me a comment (via email link in profile) - email must be entitled: Comment for publication, followed by the name of the article you wish to comment on. Comments are accepted in full or not at all. Comments will be published under the name the email is sent from unless an alias is clearly requested and stated. If you submit a comment which is not accepted within a few days you can also email me and I will check if it has been received.