A chart is doing the rounds purporting to show the number of administrators working in health care in the USA has grown much faster than the number of physicians – more than 3,000% growth from 1970 to 2009 for administrators (allegedly) compared to about 90% or so for physicians. I don’t much like the original chart so I’ve relegated it to the bottom of this post. It presumably dates from the time of the debates about the introduction of the Affordable Care Act (aka ‘Obamacare’). I find it very difficult to believe the 3,000% number, and suspect there is either deliberate definitional sleight of hand going on, or a genuine classification challenge. One obvious possibility is that some “administrator” classification has been cherry-picked that was very rarely present under that name in the 1970s, and much of the growth is movement from other differently-classified roles into that one.
A similar visualisation with Australian Labour Force Survey data
It did cross my mind that the problem was the visualisation method; in fact the tweet that brought this to my attention was from a researcher wondering what it would look like if it showed absolute numbers rather than cumulative growth. While I’m not really interested in the facts of “administrators” in the US health system, the broader data viz question sounded like something I should know about. So I had a look at Australian figures from the Australian Bureau of Statistics’ Labour Force Survey Quarterly Detail. Here is my own version of a chart showing cumulative growth in various occupations in an industry:
Actually, I think my chart is much better than the US original, not only because it uses an official and well-defined occupation classification, but because it has a go at showing absolute size as well. So we can see that while the total hours worked in the health care and human services industry by managers and professionals who aren’t health-specific (more on this below) have grown fast, the orange and grey dots are still small compared to the pink dots that represents health professionals.
The industry I’m looking at here is “Health Care and Social Assistance”, so some of those managers and other professionals (lawyers, accountants, statisticians, etc) are in social assistance rather than health, but this is as granular as we can get for an occupation and industry crosstabulation with this data without a custom request to access the microdata.
In fact, clearly one of the big stories from this chart is the thick blue dots and the rapidly rising blue line – community and personal care workers. The biggest occupations by far in that category in this industry are “Aged and Disabled Carers” and “Child Carers”, and the growth in importance in these roles (particularly the former) is one of the big news stories of the economy seen over a few decades’ perspective.
I have split the “Professionals” ANZSCO code (the lowest level published by industry in this dataset) into health and other, by getting employment hours for everyone in occupation codes 2500 to 2544 (“Health Professionals not further defined” to “Registered Nurses”). This is from a different cut of the data from the industry version and is only published for “all industries”. I adjust the all-industries health professionals number downwards by about 14%, based on the 2011 Census which tells us that there were 433,726 health professionals in total of which 373,609 worked in the health care and social assistance industry (see screenshot at bottom of post). For example, a mining company or sports team can hire a doctor or nurse. To avoid working on this thing all weekend, I’ve applied that single correction across all years of data.
Other ways of showing this data
What are the other obvious ways to visualise this data? Obviously, in absolute numbers as a stacked area chart:
… or as above but with position “fill” so we see changing proportions:
All three of these methods are completely valid.
The very first line chart – cumulative growth – is visually equivalent to converting labour hours to an index. It’s great for showing growth over time, and for many purposes would be suitable. For example, it nicely highlights that the number of labourers in the health care and social assistance industry has declined, and the fastest growing occupation types are managers and non-health professionals.
The second – absolute numbers – highlights the aggregate size and growth of labour in these occupations, while still allowing basic comparisons of changes.
The third – proportions – lets you see changes in the proportion of the workforce while still getting a snapshot overview (like a pie chart would, but for many times). In this case the change we see is the growth in community and personal service workers rather than health professionals.
Physicians / medical practitioners compared to other health professionals?
The original US chart had focused specifically on “physicians” and I’ve used a broader category of “Health Professionals”. This prompted me to do one last bit of analysis with this – to find out how much of the Australian health profession’s labour is by medical practitioners and whether this is changing. I was surprised to see that the proportion of all health professional labour done by medical practitioners of various sorts (there is no “physician” in the ANZSCO so I chose the combination of unit groups I thought was closest to this) has stayed pretty constant over the past 35 years:
That’s a very boring chart, but it’s boring for interesting reasons – a fairly steady composition of employment hours within the health professionals category, at least when divided between medical practitioners and others.
So what occupations are growing fast?
Finally, I was intrigued by the 3,000% cumulative growth in the one occupation in the original US chart. Could an occupation really grow this much in a few decades? Turns out we have a couple in the Australian data, but in our case I think these are genuine changes in employment patterns. Professional outdoor guides and ICT test engineers are two professions that we believe really have grown materially in the last few decades, partly due to changes in demand and workflow and partly due to specialisation and reclassification of other roles.
I like that plot because it gives a real snapshot (at least from one perspective) of how the economy has changed over 32 years.
Here’s today’s R code, all in one chunk. The most interesting thing here is the need to use the occupation detailed data (cube EQ08) to separate out the single digit occupation data that we get from the higher level industry by occupation data in cube EQ09.
Other supplementary material
Here’s the original image that prompted me to think about all this:
And here’s a screenshot from the ABS Census tablebuilder, for anyone curious about which industries other than Health Care and Social Assistance employ medical professionals: