Thursday, April 16, 2015

Godless Wealthy Greens? Too Rich By Half

Advance Summary

1. A recent article claims that Greens voters are likely to be rich and to be atheists, agnostics or others who do not believe in God.

2. While the connection between Greens voting and irreligion might be sound, the connection between Greens voting and very high wealth is not.

3. Booth patterns in electorates with high Green votes tend to show that very rich booths within these electorates vote Liberal rather than Green.

4. The use of family income data for inner-city suburbs overestimates the proportion of very wealthy voters within them.

5. As income is also highly correlated with Liberal voting, it is much more likely that some high-income suburbs are attractive both to very rich voters and Greens voters, but for different reasons and without much overlap.

6. There is a tendency for higher-income voters to be more likely to vote Green overall but that tendency probably weakens among the very rich and is probably strongest among those with moderately high incomes.

7. Census data by electorate is an unreliable source of information about the reasons for minor party voting.

The Australian recently published an article by Mark Coultan entitled "Rich, godless find it easy voting Green" (Google the title if paywalled). The article - which made the front page - uses results of the recent NSW election, together with NSW Parliamentary electorate census data to show that electorates where the Greens did especially well tend to have high concentrations of:

1. atheists, agnostics, humanists, rationalists (pssst! Don't tell Chris Kenny) and people with "no religion"

2.families with weekly incomes over $3000

From this it is concluded that the Greens appeal to "the rich and godless".

The latter bit might well be true, though people with "no religion" and people who do not "believe in God" are not necessarily the same thing (in either direction).  There are many reasons why Green voters may be more likely to be atheists, including the correlation between certain religious beliefs and support for social issue positions opposed by the Greens.  I'd also expect higher nonbelief rates among tertiary-educated or enrolled voters (who also tend to vote Green).  In my view the link between higher education and voting Green is nowhere near as causative as John Kaye (in Coultan's article) might like us to think - it's also the case that people for whom making money quickly is a high priority are neither attracted to the party nor to collecting degrees, whatever their thinking skills.

I won't go into the atheism side too much more in this article, beyond observing that there are plenty of Christians who are Greens (including their federal leader!), and plenty of atheists who wouldn't vote Green in a fit.  But when it comes to the high-income aspect, this is a classic case of the so-called "ecological fallacy", of attributing the characteristics of a whole population to its individual members.

This is a very well-known problem in the US - for instance Andrew Gelman wrote a book  that covered such apparent paradoxes as this one: states with high average income have high Democrats support, but high-income voters are more likely to vote Republican. (The solution broadly is that high-income voters are pro-Democrat only if they live in already liberal areas, but in red states they tend to be more pro-Republican than the rest of their state, and by large margins).

In Australia we have nothing like the richness of exit polling data that the US has and answering things like "what percentage of people who belong to high-income families vote Green" directly is not too easy.  At least I haven't seen a crosstab of income against voting intention lately myself.  But we can infer from booth data that what is really happening is that the high-income voters who live in inner-city electorates, and the Greens voters who live in inner-city electorates, are mostly probably not the same people.

Example (From My Own Backyard!)

I happen to live in an excellent electorate for looking at this sort of thing, the Hobart-based electorate of Denison.  Denison used to be among the centres of the Green-voting universe until a chap called Wilkie (himself an ex-Green) came along and took away their marbles, reducing them to single figures at the 2013 House of Reps election.  But the Senate figures still show it to be one of the Greenest seats in the nation, exceeded only by Fraser (ACT) and a handful each of inner-city Sydney and Melbourne seats.

As I've often noted here, Hobart being a small place, Denison falls distinctly into two halves - the "working class" Glenorchy sector (which in a bigger city would be in an outer suburban electorate) and the wealthier Hobart City sector.  The Greens don't do much business in the former, but the latter routinely sees Green votes well into the thirties.  But if the Coultan theory was right we'd expect to see the Green vote continue to increase as we got into richer and richer suburbs.  It doesn't.  It peaks in the 40s and 50s in the professional/academic inner city suburbs and then crashes back to the 15-30 range in the city's richest areas like Battery Point and Lower Sandy Bay - still high, but not as high as you'd expect if there was such a strong link between being rich and voting Green.

Rather, what happens in the Hobart area of Denison is that there is a high income by Tasmanian standards, and there are a lot of Green voters.  But the very rich voter booths and the very Green voter booths are not the same.  The very rich booths vote strongly Liberal.

We saw much the same in Prahran, won by the Greens in the Victorian election in an amazing three-way cliffhanger against Labor and the Liberals.  Prahran has a lot of wealth in it and a lot of Greens voters in it, but voting across Prahran is hugely stratified by booth.

The picture in New South Wales

In fact while there is a correlation in NSW between median income in an electorate and voting Green, there is also such a correlation between median income and voting Liberal, and that correlation explains a far greater proportion of the voting in high-income electorates.  See the graphs posted by Luke Mansillo on Twitter here:

Party vote vs median income by electorate

You can click on the graphic to bring up and scroll through income patterns by Green, Labor, Coalition and Green + Labor vote.

I would take the cubic lines of best fit with caution in some cases, but the important things to note are these:

* While Green voting is high in the highest median-income electorates, it is usually not above the high teens even in these.

* It does not appear that the very highest median-income electorates are the Greenest electorates on average, although it's hard to say because there are five electorates with varying incomes and extremely high Green votes.  (Update: a linear best-fit does show Green vote increasing with median income, but with only 9% of variation explained, which suggests a very noisy relationship.)

* There is much more diversity in Coalition vote levels among low-income suburbs.  This is because rural electorates dominated by the Nationals often contain income-poor farmers who may nonetheless be asset-rich.  In these areas, income is an inexact measure of wealth.

* The very high-income electorates (all the top ten by median income) have a majority Coalition primary vote.

So while it's supposedly easy for rich voters to vote Green, voters in electorates with high median incomes actually find it much easier to vote Liberal.

Ah, but what about Balmain?

Hang on, a reader of the Australian's piece might say.  Did not Coultan show that Balmain, where the Liberals polled only a quarter of the vote and came third, is loaded, rating 2 in the state for families with incomes above $3000?

Well yes, but one problem is that as an indicator of the general economic behaviour of an electorate, families with incomes above $3000 is a shaky one.  There is no electorate where these families are close to half the population.  The indicator peaks in North Shore with 38.3%, then Balmain with 35.6%.  So while we know that 35.6% of Balmain is very high-income and that 37.4% of it voted Green, we have no idea from the census data if there is even one person in Balmain who is a high-income Green voter.  Realistically, there would be some, but more likely the Greens voters in Balmain come more disproportionately from the second-tier "professional"/middle-class income bracket - people who are earning fairly high incomes but are not exactly super-rich (especially since they are doing so in expensive areas).

Also, as an indicator of the number of rich people in an electorate, family income falls over in the very high Green voting inner-city electorates of Newtown and Balmain.  It does so because these electorates (and Sydney, which would have a huge Green vote but for having a huge Greenwich vote) have among the highest proportions of lone-person households (often young inner-city apartment-dwellers).  In the leafier Liberal suburbs, a family with an income above $3000 will more often include two or more high earners rather than just one.  That is why Balmain appears further down the list by median income.

And then there is the issue that a lot of the low-income people in inner cities are young people without big houses and families to look after, living in rented rooms in share houses and so on - these occur at the tail end of the electorate's income, below the median line, and are never going to be captured in any indicator of how many families have incomes above $3000. Youth and heavy public transport usage also correlate with Green voting.  The lowish-income young voter living without family in the middle of a city is far more likely to vote Green than one living in the outer suburbs.

An old article by Possum also looks at still more trends that may be underlying the inner-city Greens vote (again based just on overall electorate stats but with rather stronger correlations than The Australian's). Certain industries attractive to Greens voters tend to have inner-city workplaces, causing those voters to be more likely to move near those workplaces whether those voters are typical of their electorate's income or not.

The data published by The Australian also show that their links break down at the other end, too.  The electorates of Ballina and Lismore are low-income electorates, partly because of their rural Nats-voting base but also because of their alternative-lifestyle attractions.  Yet these electorates have high Green votes for the latter reason and have had for a long time.  Ah, so Green voters are rich atheists then, except when they're low-income dole-bludging crystal-gazing hippies.  As some Greens have observed, the new stereotype makes a change.

The Australian's analysis is, then, a cocktail of unsound inferences.  It conflates family and individual income in exactly the electorates where it is most important not to do this.  It mistakes correlation for causation and it is awash with fallacies of composition.  To go from very wealthy electorates with high Green votes to the idea that Green voters are very wealthy is a bit like saying that if an electorate has high ratio of female to male voters and also a high percentage of footballers that it must also have a lot of female footballers.  The Greens find a lot of support among the "upper middle class" (hence the old stereotype of their voters as "doctor's wives") but their support is diverse by income and seems to taper off among the seriously rich.

Big Data Minus Clue Fairy Equals Zero

This isn't the first time the Australian has been publishing this sort of flawed census-aggregate analysis lately, and it's far from alone in doing so.  Frequent opinion writer John Black has made an art form out of trying to pin the tail on the demographic donkey in explaining which minutely categorised walks of life are (supposedly) prone to vote in certain ways.  This led to a predictive model for the NSW election that was a spectacular failure; more on that one a few days into next year (ho ho ho).

Really, we all shouldn't be playing this game, and the reason we are playing it is that compared to the US where you can find exit poll crosstabs of virtually everything, we have extremely primitive exit polling and not enough readily-obtainable public voting-intention data matched directly to attributes like religion or income.

If The Australian wants to get serious about big-data reporting of electoral behaviour it needs to realise that the first requirement of big data is that you know what you're doing with it. Commissioning studies that properly answer the questions you want to ask - instead of rushing out pieces that derive answers from aggregate data grossly unsuited to that purpose because your readers feel the fabric of the universe is threatened when the Greens win three (3) seats - would be the way to go.

I should make it clear that in writing this article I am not seeking by any means to boost the Greens, but I will defend anyone, left or right, against rubbish psephology.  I have readers who love the Greens and others who can't stand them, and my own relation to the party has been rather double-edged.  However even those who strongly oppose the Greens and everything they stand for should agree that you can't fight what you don't understand.

This article has shown that to try to beat the Greens by dismantling Bertrand Russell on the doorsteps of Vaucluse would be completely misconceived, and that those seeking to tackle the party's real strengths will need to look elsewhere - and harder.

Addendum: Francis Markham, PhD student at ANU, has tweeted something that was exactly what I was hoping someone out there might have: an assessment of the proportion of each party's voters that are in each income bracket, taken from the "noisy" Australian Election Study.

There is a sharp uptick for the Greens in the highest income bracket at the far right, but there are also upticks for Liberal and Labor, simply because that bracket is the largest bracket (hence the uptick for the purple total).  The graphs show that by House of Reps votes, the very highest-income voters are disproportionately both Liberal and Green compared to those party's overall shares of the vote.  But, since the Greens' overall share of the vote is small, it is still the case that the Liberals have far greater appeal to this income bracket than the Greens.  It's not so much the case that rich people are especially likely to vote Green, as that they don't vote Labor.

While there's a lot of noise and a surprising difference between Reps and Senate voting, it's in the high income but not quite super-rich brackets where the Greens vote seems to run proportionally furthest ahead of the total in the Reps (which is what I'd expect).

Note added 23 April: William Bowe in yesterday's Crikey subscriber email notes that 71% of Greens supporters describe themselves as "middle class".

No comments:

Post a Comment

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.