What Happens if the Electoral Realignment in 2019 Continues?

Electoral shifts at the 2019 federal election

(if you’re familiar with the pattern of shifts in the electoral map at the 2019 Australian federal election, click here to skip to the modelling)

At the 2019 Australian federal election, the nationwide swing on the two-party-preferred (2pp) vote was fairly small, with a net shift of just 1.17% from the Australian Labor Party to the governing Liberal/National Coalition. The pattern on first-preference (aka “primary”) votes was similar, with few parties seeing net swings of greater than 1% in their primary vote.

These nationwide figures, however, mask the large amount of diversity in electorate-by-electorate shifts on a two-party basis. In Warringah, the election of independent Zali Steggall saw the Coalition margin over Labor on the 2pp cut down by 9%, while Capricornia swung towards its Coalition incumbent by an eye-watering 11.7%. This turned Warringah – a historically safe seat for the Coalition – into a marginal seat on a Labor/Coalition basis, while historically-marginal Capricornia was converted into a Coalition safe seat in a single election.

It is often useful to look at these changes in terms of how they shift the two-party lean of each electorate i.e. how much higher or lower the Coalition/Labor two-party-preferred is in an electorate versus the rest of the nation. On that basis, these are the electorates which shifted the most towards the Coalition:

And the electorates which shifted the most towards Labor on a 2-party lean basis:

Do note that two-party lean can mask the source of the shifts – swings in some electorates (e.g. Warringah, Capricornia) were driven by primarily major-party-to-minor-party/independent shifts while others saw primarily swings from one major party to the other (e.g. Higgins, Lindsay).

That being said, some patterns stand out – inner metro electorates (particularly high-income inner metro electorates) tended to shift towards Labor, while rural and provinicial electorates shifted towards the Coalition. Research by others have also found that these shifts line up with the worldwide trend of tertiary-educated voters toward the left and non-tertiary-educated voters toward the right.

Overall, these shifts led to an unfavourable electoral pendulum for Labor, with the 2pp needed for Labor to win majority on a uniform swing going up from 50.7% to 52.4%. The recent redistribution has made the pendulum slightly more balanced, with the 2pp needed for a Labor majority going down to 51.8%, and the government remaining slightly favoured on the uniform-swing pendulum.^1x The calculation of the same statistic for the Coalition depends on how you treat Hughes, where Liberal Craig Kelly defected to Palmer’s UAP and is recontesting his electorate.

If you assume the Liberals fail to retake it off him, the 2pp needed for a Coalition majority on uniform swing is 51.7%.

If you assume the Liberals do retake Hughes (which appears more likely than not at this time), the Coalition starts with a majority and would need a 2pp of 51.2% to avoid losing it on a uniform swing.

Do note that this statistic does not account for the possibility of crossbench challengers taking seats off either the Coalition or Labor, or the converse (but usually much smaller) probability of the major parties retaking crossbench-held seats.

Trapped in the Electoral Valley of the Left

Labor’s pendulum woes mirror the challenge faced by the Hillary Clinton campaign at the 2016 US Presidential election. Clinton’s voter coalition contained fewer non-college-educated white voters but a greater proportion of college-educated white voters and Hispanic/Asian-American voters, leading to an unusually high probability of an Electoral College split favouring Trump.

The key to that was the Democrats’ new gains coming primarily in states which were historically very Republican-leaning (e.g. Texas, Georgia), whereas their losses were concentrated in states which were only somewhat more Democrat-leaning than the rest of the nation; resulting in their new voters not being sufficient to gain them any new states while the loss of their old voters was sufficient to flip several states from the Democrat to the Republican.

Something similar happened to Labor in 2019; of the table above showing swings to Labor, just three electorates listed had a Coalition 2-party vote of less than 60% (Canberra, Bendigo and Macnamara – all three Labor-held to begin with). Of the seats with the highest swings to the Coalition, only four (Hunter, Kennedy, Bass, Indi) had a 2-party margin of greater than 54%, with Kennedy being held by someone outside of Labor/the Coalition and Bass tossing its Labor member out anyway.^2x Indi also elected an Independent, although it wasn’t clear at the time whether it would do so given the previous Independent incumbent’s retirement.

Hence, you can see why there are calls for Labor to court its historical voter base in order to regain a majority, and why the Coalition Prime Minister would be wandering around Australia in high-vis wear. However, none of this necessarily implies that pursuing reversion in historical marginals would be a better strategy for Labor than pursuing realignment in the electorates which moved to it last time.

What would the electoral map look like if Labor managed to reverse some Coalition gains in historically-marginal rural electorates? Or if the Coalition continued to lose ground in “blue-ribbon” seats while firming up its newfound support in the regions?

Modelling reversion vs realignment in the electorates

To analyse this, I look at the Coalition two-party lean in each electoral division, calculated by (Coalition 2pp in electorate – Coalition 2pp nationwide), and compared the two-party lean in 2019 to the two-party lean in 2016, with adjustments for redistributions where necessary (thanks to Ben Raue of the Tally Room for that data).

I then model the change in two-party lean from 2016 to 2019 using demographic data from the 2016 Census, with the demographics-predicted shift combined with the actual shift in two-party lean. The reasoning for this is to smooth out anomalies in electorates where unusual candidates likely confounded the 2pp (e.g. the disendorsed Liberal in Lyons) – full methodology below (and data available here):

Before going any further, I’d like to note the ecological fallacy. Analysis at the electorate-level – as we’re doing here – can only tell us what kinds of electorates shifted one way or the other, and not which kinds of voters. It is entirely possible that, for example, Labor made gains among low-income voters in high-income electorates while the Coalition makes gains among high-income voters in low-income electorates. Hence, any results from this kind of analysis cannot be used to make inferences about individual-level voting behaviour without individual-level polling.

With the modelled shifts in two-party lean, I have simulated the following scenarios:

1. Reversion: modelled electorate 2-party leans changes in 2019 reverse e.g. Warringah returns to being 10.7% to the right of the nation, Capricornia moves back to being a marginal etc. Note that some particularly large shifts may not fully reverse due to a combination of the demographic model and a large-shift limiter I have in place – I will note below if those have any impact.
2. Partial reversion: only half of the modelled 2019 change in electorate 2-party lean is reversed. e.g. Dickson, which shifted towards the LNP by 2% (on 2pp lean basis), shifts back towards Labor by 1%.
3. No change: as it says on the tin.
4. Partial realignment: modelled seat-by-seat shifts in 2019 are repeated, but only by half. e.g. Higgins, which shifted towards Labor by 7% (on 2pp lean basis), shifts to Labor by another ~3%.
5. Continued realignment: modelled electorate lean shifts in 2019 are repeated in full. Warringah becomes a Labor-leaning seat on 2pp basis, Capricornia becomes an ultra-safe LNP seat etc. See above note on reversion scenario about large shifts.

Using a probabilistic seat model of the kind described here and here, and assuming that all crossbench incumbents hold their seats and no new crossbenchers are elected (not realistic but for purposes of this analysis should have minimal effect),^3x Most of the prominent crossbench challengers are running in safe seats on a Coalition-vs-Labor basis. In any case, unless the crossbencher’s chance of winning is strongly correlated to the two-party contest, the effect would be to simply shift all votes-to-seats curves up or down.

My view is that there is some correlation, but it is very weak. For example, if Labor does really well, it could boost the crossbench challengers in Coalition seats (because they now have more Labor voters’ preferences), but Labor could also do so well that they knock the challenger down to 3rd and lose against the Coalition candidate.

Another possible failure mode is that the crossbench challengers’ odds are dependent on whether the electorate realigns or reverts; but that’s very complicated to model accurately and hence I’ve left it out of the model. here are the probabilistic seat totals for Labor if there is either partial reversion or partial realignment:

And for the full reversion and full realignment scenarios:

Some general trends:

As far as winning seats alongside a 2pp majority goes, Labor’s current electoral map is the worst of both worlds. It has neither fully realigned seats like Higgins and Kooyong to the point where it has high probabilities of winning them alongside the popular vote, nor is it competitive enough in old marginals like Flynn and Petrie to snag many of them.

Reversion appears to be a uniform improvement over Labor’s current map, while realignment produces a bigger seat haul for Labor if it wins the popular vote but may end up a disaster if Labor falls short overall. Additionally, there isn’t much difference between reversion versus realignment if Labor wins the popular vote; the expected difference at 50-50 2pp is just 0.2 of a seat.

(Removing the large-shift limiter has no effect on the partial scenarios, but it does make full reversion and full realignment somewhat better for Labor on a votes-to-seats basis; they gain an extra 0.9 seats under reversion and 0.7 seats under realignment)

The same votes-to-seats plots for the Coalition under partial reversion/realignment:

And full reversion/realignment:

The picture is broadly the inverse of the Labor plots – reversion would uniformly hurt the Coalition on a 2-party votes-to-seats basis, while realignment would boost them when they win the popular vote and hurt them if they lose it. As noted above, none of this accounts for minor party and independent challengers; many of the latter who are challenging in historically-Liberal electorates which shifted left in 2019. If many of these independent challengers get up under realignment but not reversion, they could make realignment look worse for the Coalition.

Comparing tipping-point statistics under reversion and realignment

Another way of comparing skews in the electoral map is by determining both sides’ 2pp needed for a 50% chance at a majority, and the predicted median electorate’s 2pp when the popular vote is tied; I refer to the former as majority tipping-point and the latter as tipping-point electorate 2pp. If the pendulum shifts against a party, it will need a higher 2pp for a majority than its opponent and the median electorate will usually (but not always – see footnote) shift away from it.^4x In theory, the map could shift against a party without any change in the median electorate. For example, a redistribution could change all Coalition seats with 2-party margins of between 3% to 6% such that all of those seats become Coalition 53.1% (with no Labor seats touched, and all excess L/NC votes moved into very safe L/NC electorates).

This would have no effect on the median electorate (currently Mayo, L/NC 52.54%), but it would make the map more difficult for the Coalition as they would have less buffer against Labor in those electorates.

A redistribution could also make Labor marginals safer to achieve a similar effect, by making it harder for the Coalition to pick up Labor marginals to offset random losses in its own seats.

Tipping-point statistics for the Coalition under each scenario:

By these metrics, the reversion scenarios is somewhat more favourable to the Coalition than the realignment scenarios. While the differences in majority tipping-point 2pp are quite small, the Coalition tipping-point electorate 2pp goes up from 50.2% to 51.7% in the partial realignment vs partial reversion scenario and from 50% to 51.2% in the full realignment vs full reversion scenario.

It should be noted that these metrics are influenced quite a bit by the large-lean-shift limiter. As above, I have included a limiter to limit lean-shifts to a maximum of 10%; if this limiter is removed then the Coalition tipping-point electorate 2pp goes down to 49.9% for full reversion and 49.6% for full realignment. Hence, the outlook for the Coalition under full reversion is a lot worse if you assume massive lean shifts of the kind seen last election can be reversed in a single election.

The same metrics for Labor under each scenario:

On the 2019 map, and assuming no new crossbenchers, Labor currently needs a slightly higher 2pp than the Coalition for a majority in most scenarios. It’s also worth reiterating that the current map is worse than either a reverted electoral map or a realigned electoral map for Labor’s odds of getting a majority, leaving Labor stuck between its historical and new voter bases.

As with the Coalition estimates, realignment is slightly better than reversion in terms of improving Labor’s odds of a majority; however this depends on whether one assumes massive lean shifts of the kind seen last election can be reversed/continue in the same electorates. The Labor tipping-point electorate 2pp would go up to 50.1% for full reversion and 50.4% for full realignment if this assumption is made, while the majority tipping-points for Labor would go down to 50.8% and 50.9% for full reversion and full realignment.

(I’ve analysed how accurate these kinds of probabilistic seat models are in the piece below:)

So what should each major party do going forward?

The Coalition? Not much, really. The status quo is fairly favourable for the Coalition, with the tipping-point metrics being broadly in their favour despite their relatively close win in 2019 on seat and vote terms. Hence, the Coalition would probably want to challenge Labor wherever they think Labor is most likely to retake voters and prevent any reversion or further realignment in the electorates’ 2-party leans, although this is conditional on there being relatively little difference between the danger of crossbench challengers winning Coalition 2pp seats under realignment versus reversion.

(This is of course referring to the electoral map and not the overall contest; if Labor wins by 53% or 54% on the 2pp as some polls currently suggest, the electoral map is unlikely to deny Labor a majority irrespective of reversion or realignment.)

Labor, on the other hand, is stuck between the collapse of its old voter base and the incomplete rise of a new Labor voter base. While obviously winning both would be preferable, if it has to pick, Labor’s broad strategy should depend on whether they think they can achieve very large lean shifts in electorates like Capricornia, Dawson, and Hunter. If they can, either reversion or realignment would work in their favour vis-a-vis the current electoral map. If they can’t, pursuing further realignment to flip electorates like Higgins, Ryan and Kooyong would improve its odds of a majority (assuming it wins the popular vote) whereas reversion may be somewhat better at saving the furniture in the event of a popular vote defeat.