In sickness and in wealth

A Sydney Morning Herald article compares the top and bottom ten Sydney locations across five health outcomes and finds stark differences.  But does this format miss the point?

At the confluence of three trends — big data, online mapping and open government — a powerful tool has emerged: spatial analysis for health.  With a few clicks, anyone can visualise a huge range of demographic and health information by area.

High or very high psychological distress by statistical local area in Sydney metropolitan region (Kessler 10-item score on National Health Survey 2008)

High or very high psychological distress by statistical local area in Sydney metropolitan region (Kessler 10-item score on National Health Survey 2008) (Source: PHIDU Social Health Atlas 2012)

Recently I took up a new position with a genuinely statewide focus, and I’m finding spatial analysis an incredibly powerful tool.  It’s useful as a starting point for discussions around intervention targeting, as I can identify areas that are low on a particular healthy behaviour or high on some risk factor; I can also visualise different demographics, like the proportion of people in an area who don’t speak English at home, to guide in cultural tailoring of messages and resources.

There are some dangers in spatial analysis.  One involves the old saying ‘correlation does not imply causation’.  I might map two categories, such as the proportion of migrants in an area alongside the practice of a healthy behaviour.  If there are lots of migrants in most of the areas lowest on the health behaviour, can I conclude there’s ‘a problem’ that needs fixing by intervention in those groups?

Not so fast.  I’d want to check for other relationships first.  For instance, the Scanlon-Monash Index of Social Cohesion (Markus, 2012) found major areas of immigrant concentration are also among the poorest localities in their state — so the lower healthy behaviours might be driven (wholly or in part) by poverty instead.  This is where multi-level analysis is necessary — comparing or ‘triangulating’ area- or group-level trends in a model with data on individual behaviour and outcomes.

It also helps to have qualitative evidence from interviews, focus groups, ethnography and cultural studies, otherwise you’ll have no idea what the health risk or behaviour means in and to the groups and localities in question.  You probably also want some social theory — at a bare minimum an understanding of social stigma — so that you know what you’re doing when you label people in ‘x’ group/area as having a need/problem.

That’s where the SMH article bugs me.  It leads with the ‘rich areas have an alcohol problem’ — let’s not kick the poor councils while they’re down.  But the data tells the story, right?

Screen Shot 2013-01-06 at 10.04.19 PM

Not so fast.  All the time, I hear tables like this referred to as ‘data’.  Let’s be clear about this: it isn’t data.  Data is what you can analyse.  This is a summary and a selection.  I’m sure it’s factually correct, but the selection of the “Best 10” and “Worst 10” (check those labels) presents diabetes as a dichotomy — the results have been split between two poles and the middle has been silently dropped.

This presentation is the health equivalent of ‘the gap between rich and poor’ as a measure of income inequality.  It offers maximum drama — always helpful when you have an issue to advocate about.  It also facilitates social comparison, which we all love, after all.  If you’re living in that silent middle and reading the SMH over your terrible Sydney latte, who are you aligning yourself with, the “Best” or the “Worst” ten council areas?  Best, right?

Not so fast, middle Sydney.  Here’s a graph from the PHIDU Social Health Atlas 2012 comparing two maps: along the y (vertical) axis, the Index of Relative Social Disadvantage score for statistical local areas (smaller than councils but larger than suburbs) and on the x (horizontal) axis, percentage of residents with Type 2 diabetes.

Index of Relative Social Disadvantage x Type 2 Diabetes

Type 2 diabetes tracks with socioeconomic status in all the Sydney localities.  When I linked him a similar version of this graph, Dr Tim Senior, a Sydney GP who works in Campbelltown (“Worst”) with Aboriginal communities, put a succinct response: “gradient not cliff then.”

That’s exactly right, and it’s the opposite story from the one conveyed by the SMH infographic; socioeconomic status predicts type 2 diabetes prevalence in every Sydney council area.  Health inequalities are everybody’s business, but stigma — the cultural production of social divisions — distracts us from that, and that’s always the danger in league tables.


This is not an attack on the SMH — we are only able to have this discussion because they published this article.  Nor is it an attack on the journalist, SMH Health Editor Amy Corderoy (@AmyCorderoy), who was game to discuss this stuff late on a Sunday night on Twitter.  Whoever did the infographic, though: using bigger bars to indicate lower wealth? Yick.