No truth stands alone.

Which areas drove the swing to Labor in WA 2021?

With the final distributions of preferences having come in, we have the 2-party-preferred estimates for every electoral district, and they show the sheer magnitude of the swing towards Labor. No district swung by less than 5% towards Labor, with 4 districts with swings towards Labor of greater than 20%. Here, we look at what variables can explain the variations in the swing throughout the state.

But first, two caveats.

Caveat 1: You might also know this as the ecological fallacy. There’s a family of sub-fallacies under that broad umbrella, but broadly speaking, it’s defined as the error of assuming that correlations between large groups also apply to individuals. A commonly cited example from American politics is how the wealthiest states tend to vote Democrat (think California, New York etc) but the wealthiest voters vote Republican. Just because areas with a certain characteristic vote in a certain way (e.g. wealthier electorates swinging to Labor) doesn’t necessarily mean that wealthier voters swung to Labor; it could be that wealthier voters stayed roughly the same while poorer voters in wealthy electorates swung more to Labor.

Additionally, when the data can be broken down into further sub-groups, we have to be wary of Simpson’s paradox, which is basically where a trend you might see in individual sub-groups (e.g. QLD electorates, VIC electorates) either disappears or reverses itself when you combine the individual sub-groups. For example, if you look at the 2pp swing vs % recent migrants chart in this Guardian article, an alternative explanation might just be that Queensland has relatively few migrants and a big swing to the Coaliton; it’s not apparent that the recent migrants = less Coalition swing relationship holds in, say, Victoria (where the correlation is either nil or recent migrants = more Coalition swing) or NSW (pretty much no correlation at all).

Caveat 2: At the same time, you often hear the ecological fallacy being invoked to dismiss any and all demographic analyses. While people often do go too far in claiming that some correlation between a demographic and a party’s vote share means that they’ve won/lost/need to gain ground with that demographic, if the only data we have is a correlation between demographic X and vote share A, then without any other data to the contrary, our assumption should be that there is more likely than not to be some correlation between X and A.

In the example above, the only way we know that that is an example of the ecological fallacy is because of exit polls, voter file matches and other data external to the demographic analysis which show wealthier voters prefer Republicans. Without that additional data, we would not know that wealthier voters prefer Republicans even though wealthier states vote Democrat (and if you’re going to say “we know it through…”; that is still data external to the demographic analysis).

Here’s an example of a % demographic vs vote share by state correlation (taken from the US):

Demographic vs Democratic share of two-party vote in 2020
R2 = 0.89, p-value < 0.0001

Ecological fallacy? Well, the demographic in question here is “two-party percentage of registered voters who self-identify as Democrats”, so you tell me.

The takeaway here is not that the ecological fallacy is wrong, or that demographic analyses are always useless; rather the point I’m trying to make is that demographic analyses tell us what kinds of areas voted for certain candidates, but more evidence (or theory, which is usually codified evidence anyway) separate from electorate-level demographic analyses are needed to demonstrate what kinds of individuals voted for certain candidates.

With that out of the way, let’s get to the analysis.

Incumbency was a strong factor in explaining variations in the swing

Using Antony Green’s 2pp estimates, I calculated the deviation from the statewide 2pp swing (-14.1%, or 14.1% swing to Labor) for each district, and then classified them by the incumbency status of their MPs. Incumbency matters due to the theory of personal vote effects, which refers to the small slice of voters who would only vote for e.g. their Labor MP and not a generic Labor candidate. My classifications as follows:

  • Incumbent recontesting (Labor or Liberal/National) – we would usually expect less swing against the incumbent’s party (or more swing to the incumbent, depending on the statewide environment) thanks to the personal vote.
  • Party gain at the last election (also known as sophomore surge) – since all contests in WA involve one of the major parties, a party gaining a seat at the last election means they flipped it out of the hands of “the other side”. This effectively removes the personal vote of “the other side” (here, it’s usually the Liberals/Nationals, thanks to Labor’s landslide win at the last state election) while instating the personal vote of the incoming party MP.
  • Incumbent retiring – we would usually expect less swing to the incumbent’s party/more swing against the incumbent’s party when their MP retires, due to loss of that personal vote.

Below are the average deviations from uniform swing for each category of MP (negative values = more swing to Labor, positive = less swing to Labor; italicised values are not statistically significant at p = 0.05):

Incumbent recontesting1.9%1.1%
Party gain (new incumbent)-3.2%6.6%*
Retiring incumbent4.3%-0.9%

* : The only seat I have as a Liberal gain is Darling Range, where the Liberal MP gained it off Labor at a by-election, so this isn’t so much an average as just Darling Range. Treat with caution.

Most of the figures in those are roughly in-line with what we would expect – Labor did significantly better in seats it had just gained at the last election while doing worse in seats where its incumbents retired. Though there aren’t many seats on the Liberal/National side of the pendulum (just n = 1 for Lib/Nat gains and n = 3 for Lib/Nat retirements), we can also tentatively conclude that the same appears to be true for conservative side of politics.

Interestingly Labor did somewhat worse in seats with incumbents of either side; if I had to hypothesise why my guess would be that they didn’t actively do worse (in the sense that people weren’t actively voting against their Labor incumbents), but rather the massive swing in the unusually large number of seats they’d just gained at the 2017 election (n = 18) outweighed the swing in incumbent-held seats, making those seem smaller by comparison (Labor did gain in those seats as well, after all).

Electorates with a greater share of seniors swung less to Labor

Demographic data for the following analyses courtesy of the Western Australian Parliamentary Library’s Community Profile page: The variables I present here are statistically significant even when accounting for incumbency effects and other demographics in the same category.

This demographic was a bit of a surprise; prior to the election there was a significant amount of speculation that the seniors community would swing hard to Labor based off Queensland Labor’s wins in retiree-heavy electorates like Hervey Bay the previous year. However, it seems like senior-heavy electorates actually swung less to Labor than their younger counterparts:

% of the population 60+ y/o vs 2pp swing
R2 = 0.2, p-value < 0.001

The 10 electorates with the highest 60+ year-old share of their population are as follows (swing to Labor in brackets):

  1. Dawesville (14.7%) – Liberal leader Zak Kirkup’s seat, roughly swung in-line with rest of state.
  2. Moore (11.0%) – safe Nationals seat, swing somewhat under the statewide swing. A common theme at this election is ultra-safe seats swinging less than similar but more marginal counterparts.
  3. Warren-Blackwood (14.1%) – usually a safe Nationals seat but narrowly won by Labor on a swing in-line with the rest of the state.
  4. Kalamunda (9.5%) – a Liberal-leaning marginal seat where the swing was significantly under the rest of the state.
  5. Rockingham (14.2%) – the Labor Premier’s own ultra-safe seat, got a swing to him roughly in-line with the rest of the state.
  6. Mandurah (7.5%) – a Labor-leaning seat where the swing was well under the rest of the state. The MP here still won 75.2% of the 2pp; Mandurah has historically been a fairly inelastic seat (the swing there tends to be lower than the statewide swing) so that might be the most likely explanation.
  7. Albany (7.8%) – a marginal seat which Labor has successfully held onto despite one-vote one-value reforms (you may also have heard of them by their other name, democracy) thanks to their popular local MP Peter Watson. Watson retired at this election, so part of this is likely the removal of his personal vote.
  8. Central Wheatbelt (11.5%) – ultra-safe National seat held by the Nationals’ leader, Mia Davies.
  9. Cottesloe (6.6%) – ultra-safe Liberal seat which recorded the second lowest swing to Labor of all electorates; the MP who won it at the 2017 state election was the then-Premier Colin Barnett who then resigned after his party lost.
  10. Vasse (10.3%) – the safest Liberal seat prior to the 2021 state election.

Electorates where a greater share of people lived in fully-owned housing also swung less to Labor

A large chunk of this is because of the very strong correlation between % of people who live in fully-owned housing and the 60+ years old demographic:

% of the population 60+ y/o vs % of the population in fully-owned housing
R2 = 0.78, p-value < 0.0001

Once you control for the % of the population that’s over 60 years old, there’s not much correlation left between housing type and the deviation from uniform swing.

Electorates where a greater share of the population work in clerical and administrative jobs swung more to Labor

I’m honestly unsure why this is the case. I originally thought this was just an artifact of the fact that most such seats are suburban seats which Labor just won at the last state election, but controlling for incumbency doesn’t affect the correlation much (and controlling for age + language spoken at home + incumbency actually makes it stronger).

% of population in Clerical/Admin work vs 2pp swing
R2 = 0.12, p-value < 0.01

The top 10 electorates with the highest share of their population in Clerical/Administrative jobs are (2pp swing to Labor in brackets):

  1. Kingsley (15.7%) – inner-city Perth seat that Labor flipped at the 2017 state election; the slightly larger than average swing here is likely in part due to sophomore surge.
  2. Landsdale (15.8%) – very similar to Kingsley (inner metro Perth seat) except Labor held onto this seat’s predecessor (Girrawheen) in the 2013 state election, so the swing here can’t be explained through sophomore surge.
  3. Forrestfield (16.1%) – similar to Kingsley; a metropolitan Perth seat where the Labor candidate just won at the last election.
  4. Jandakot (19.2%) – very similar to Kingsley and Forrestfield; metro Perth seat where the Labor candidate narrowly won at the last election.
  5. Cockburn (12.4%) – fairly safe Labor seat in Perth with a smaller swing to Labor than average; part of this can probably be explained by the fact that the Labor incumbent (Fran Logan) retired at this election.
  6. Swan Hills (15.0%) – see above comments about Forrestfield; although I will note that Labor’s win here in 2017 was very large compared to some of the other seats on this list (64.5%).
  7. West Swan (15.7%) – fairly safe Labor seat in Perth whose member has held it since it was created in 2008.
  8. Wanneroo (19.8%) – a very elastic swing seat on the outer edges of Perth whose Labor member just won it at the 2017 state election.
  9. Darling Range (7.5%) – historically a Liberal-leaning seat which flipped to Labor in 2017, but whose member resigned after a scandal around his credentials, resulting in a by-election where the Liberals retook the seat. Part of the reason the swing here is relatively low can probably be attributed to that scandal.
  10. Hillarys (19.3%) – see above comments about inner-city Perth seats that Labor just won at the 2017 election.

Based off those electorates, another way of explaining this could also be “metropolitan Perth seats, especially those that Labor just flipped in 2017, swung more towards Labor”, with % of population in Clerical/Administrative work functioning as a useful proxy for metropolitan Perth-ness. However, this doesn’t seem to square with the fact that there was no correlation between the % of population working as Professionals and the 2pp swing (with analysis suggesting that whatever correlation existed was actually reversed – i.e. the Labor 2pp swing actually dropped as % of population in Professionals increased, though again I stress this is not significant); if anything I would think that % of population in Professional work would correlate better with metropolitan Perth-ness.

Electorates where a greater share of the population spoke a language other than English at home swung harder towards Labor

This one is somewhat weaker than the others but still significant – electorates where a greater share of the population spoke a language other than English (as compared to those who reported exclusively speaking English) swung harder towards Labor:

% of population speaking a language other than English vs 2pp swing
R2 = 0.08, p-value < 0.05

The top ten electorates with the highest share of the population speaking a language other than English (2pp swing to Labor in brackets):

  1. Mirrabooka (10.4%) – inner-city Perth seat which is usually safely Labor; the sitting MP retired before the 2021 election which helps to explain why the swing here is lower than the statewide swing.
  2. Cannington (12.9%) – another inner-city Perth seat which is safely Labor, although the sitting MP here didn’t retire.
  3. Balcatta (17.7%) – a Labor-leaning inner-city Perth seat where Labor just won at the last election; part of the increased swing to Labor here is likely explainable by sophomore surge.
  4. Morley (16.2%) – another inner-city Perth seat where Labor just won at the last election; unlike Balcatta though Morley is more Liberal-leaning having been won by the Liberals from 2008 to 2017.
  5. Victoria Park (11.1%) – an inner-city Perth seat which is very safe for Labor; the lower-than-statewide swing here is likely due to the incumbent (Ben Wyatt) retiring.
  6. Riverton (13.2%) – inner-city Perth seat which is Liberal-leaning but whose incumbent (Mike Nahan) retired at the 2021 state election. Interestingly the swing to Labor here was slightly below average despite the loss of his personal vote for the Liberals.
  7. Southern River (25.3%) – the largest swing to Labor in any seat; this is a highly elastic swing metro Perth seat where the swings tend to be bigger than average.
  8. Thornlie (15.1%) – metro Perth seat which is usually safe for the Labor party; swing here is slightly above the statewide swing.
  9. Jandakot (19.2%) – covered above; a metro Perth seat won at the 2017 election and where the swing was bigger than the statewide swing.
  10. Belmont (17.8%) – metro Perth seat which leans towards Labor but was won by a Liberal in 2013. Labor won it back in 2017 and part of the above-statewide swing here is likely due to sophomore surge.

From the looks of it, it seems like metropolitan Perth seats which are either marginal or lean Labor swung the hardest to Labor, especially those which were newly won by a sitting Labor member in 2017. On the other side, safe conservative seats (Moore, Central Wheatbelt, Cottesloe, Vasse) swung less to Labor despite having significant senior populations; contrary to expectations before the election it seems like seats with a large share of the 60+ years-old cohort did not swing hard to Labor as expected.

Data for all of these analyses can be found here.

Add Your Comment

* Indicates Required Field

Your email address will not be published.



This site uses Akismet to reduce spam. Learn how your comment data is processed.

× Home Politics Projects

You appear to be using an outdated browser, for which this site is not optimised.

For your security, we strongly recommend you download a newer browser.

Although any of the latest browsers will do, we suggest the latest version of Firefox.

Download Firefox here

(Click on the button in the top-right to close this reminder)