One of the most historically accurate predictors of the way people vote in Australian elections has always been the notion of “class”. At its most reductionist, working class voters generally support the ALP, owners of capital and employers generally support the Liberal Party and in the post-war years, rural Australia has generally supported anti-Labor candidates of various hues.

To a large extent that basic pattern still holds, but class structures are evolving to the point where many of the relationships we used to see are no longer as predictive as they once were.

For instance, manual labourers used to be the backbone of the Labor Party voting stock (that and the public service). The more manual labourers that lived in an electorate, generally, the larger was the ALP vote. Yet if we look at the electorate level results of the last election and compare the ALP primary vote with the proportion of people in the last Census that described their occupation as “Labourer”, the relationship has pretty much broken down:


Historically, we would have expected to see that scatter plot producing a pattern that started from the bottom left and ended in the top right of the chart – signifying a strong positive correlation between the two. Now, however, it’s almost a random cloud showing no correlation between manual labourers and the ALP primary vote at the electorate level.

There’s all sorts of reasons for that, from the changing socio-economic mix of our urban electorates through to the changing nature of our economy itself.

Yet, despite historical relationships like this no longer turning up with electorate level data, there are some proxies for class that still dominate the electorate level data.

The biggest predictor of the Labor Party primary vote at the last election wasn’t the usual suspects. It wasn’t the proportion of the electorate that came from a non-English speaking background, it wasn’t the proportion of the electorate that received family tax benefits, it wasn’t the proportion of the electorate that worked for the public service.

The biggest predictor was an inverse relationship, and one that explained 55% of the variation in the Labor Party vote – it was the proportion of the electorate that classified themselves in the Census as working as a “Manager”. The higher the proportion of managers in an electorate, the lower was the ALP primary vote.

If we run a scatter and regression line of the two for all 150 electorates, this is what we get:


For every 1% increase in the proportion of the electorate that worked as “Managers”, the ALP vote decreased, on average, by 2.3%.

As you would expect, the Coalition experienced the opposite effect:


Although the strength of the effect was slightly weaker, with the proportion of the electorate working as “Managers” only explaining 37% of the variation of the Coalition primary vote, and where a 1% increase on the proportion of managers working in an electorate, the Coalition primary vote increased, on average, by 1.8%

The other interesting thing that pops up is where the Greens come into all this.

Why are the Greens strongest in the inner cities? What is the cause of Greens voters living in the inner cities?

One thing that may help explain it is occupation. If we use the census data to sum together the people in each electorate that work in the three categories of Arts & Recreation Services, Information Media & Telecommunications and Education – and then scatter that against the Greens primary vote across 150 electorates, this is what we get:


The proportion of people working in these three industry sectors explains 51% of the variation in the Green’s primary vote. What really strikes me here, and makes me think it’s more a causative result than a proxie for “something else”, is the way that the linear relationship holds strongly across the entire spectrum of the Greens vote.

If we do the same thing again, but this time run a locally weighted polynomial regression (Loess curve) through the scatter as well, it will give us a non-linear trend line of the correlation between these industry employment types and the Greens vote as we move across the Greens vote spectrum from 2% through to nearly 23%.


The black line is our linear regression line, the red line is our loess curve – and I’ve adjusted the left axis scale here to visually maximise the difference between our two trend lines.

The loess curve sits very close to the linear regression line across the entire spectrum of the Green vote, telling us that the linear relationship isn’t being skewed or distorted by some block of either Greens voters or industry workers. If the red line started diverging massively from the black line, it would tell us that the correlation isn’t consistent across the whole spectrum as well as letting us know through what values the inconsistency exists.

But here there is only a very small divergence at around the 10% Greens vote mark, but then only maxing out at around 1% of the proportion of these industry workers before heading back to match the linear trend.

This is why I think there’s some strong causative aspect to this relationship – the consistency across all electorates and across all values is so strong, from the lowest Greens voting electorates, through the bulk of the middle Greens voting electorates and into the top Greens voting electorates.

If there wasn’t a causative aspect to this, but just, say, some generic correlation between the inner cities and the people that work in these particular industries, then we would expect to see the linear nature of the relationship break down as we moved away from the high Green voting seats and into the middle and low voting Green seats – we would expect to see the linear relationship breakdown as we move away from the inner city clusters of high Greens voting electorates.

But we don’t – the linearity of the relationship remains consistent.

So what does this mean in practice?

For starters, a large part of the “cultural” aspect of Greens voters living together appears to come as a result of hard cultural issues such as their employment, rather than because of soft, vacuous stereotypes like the cultural pursuit of the perfect latte – the sort of nonsense promulgated by pop-demographers like Bernard Salt (a story of which I should tell you about one day – a question I asked him at the recent Qld population summit resulting in some horseshit about “Metropolitan Chauvinism” and latte drinkers living in West End. Ugh!).

People working in the arts, education, media and technology industries are more likely to vote Green, and as a result of the distribution of workplaces for these industries having a higher density in the inner suburbs, the people living within close proximity to their workplaces naturally leads to the inner cities having higher levels of Greens voters.

As the density of these industries reduces, so to does the Greens vote.

In one respect, it’s just a modern evolution of class based electoral analysis – whereas the Labor vote used to correlate strongly with manufacturing and low skill, labour intensive industries – today, the Green vote correlates with new skilled services industries like arts, education, media and technology.

So it’s not something particularly new, it’s just a relatively new type of something that we’ve long seen at play in the Australian electorate.