Sufficient Scruples

Bioethics, healthcare policy, and related issues.

January 13, 2010

How Can We Make It Clearer? When Will Anyone Notice?

by @ 4:33 PM. Filed under Access to Healthcare, General, Global/Community Health, Healthcare Politics

This is staggering:

Industrialized countries ranked by health expenditures vs. life expectancy - US is worst.

Health Expenditures vs. Life Expectancy

(See link for larger version.)

The graph is a ranking of industrialized countries by per-capita healthcare expenditures. The average figure is $2,986/year; Finland and Spain come in a little below that, and Australia and Sweden are a little above. Canada spends about $1,000 more per person per year than the major-nations average; free-wheeling Switzerland is about $1,400 above average.

See that lone, single red figure wa-a-a-a-a-y up in the left-hand corner? That’s the US. Our per-capita healthcare expenditures, at $7,290/year (!), are more than 240% of the average of all those nations together (actually, more than 260% of the average of all those nations other than the US, which comes to only $2,771/year if you leave the US out of it). And note that those expenditures, in the US, are for only 85% or less of the population; for every other country on the graph except Mexico, that figure covers every permanent resident of the country without exception.

Now look on the right-hand column: the same nations are ranked by average life expectancy. This is a crude, but still useful, indicator of what we’re getting for our healthcare costs. (Crude, because simple measures like sanitation and nutrition can contribute a great deal more to life expectancy than high-tech medical care. But the whole point is that better medical care produces longer and better lives, at the margin at least, and there is good evidence that this is true. So this is not a bad way to scale things out for quick and easy comprehension.) Implicitly, this graph establishes a relationship: assuming all things are equal, average healthcare expenditures should produce average life expectancies (which you could quantify as a numerical ratio, though that would be taking the thing rather too literally). In fact, that is almost exactly what the UK achieves ($2,992/year for about 79.1 years lifespan). About two-thirds of the countries on the list do better than that: their life expectancies, relative to average, are greater than would be expected given their healthcare costs relative to average. (This is indicated by the lines sloping up to the right on the graph. The slopes are not precisely indicative, because the right-side scale range does not match the left-side scale – the ranges should have been correllated better. But a positive or negative slope indicates an above- or below-average ratio, respectively.) A relatively small number of countries do worse.

Whether above or below average, the deviations tend to fall into a small range – note that most of the lines up, and down, are roughly parallel. All except one, of course: the US, as usual, is completely alone in its breathtakingly negative ratio of cost to life-expectancy benefit. That screaming red line plunging down the graph from off-the-charts high expenses to below-average benefits has no peer among any industrialized country: nowhere in the world does any country get such an incredibly below-average relative return for its healthcare expenses (and in fact below average in absolute terms compared to all other countries). The US, with per-capita healthcare expenses 260% highher than its peers, actually averages a total life expectancy almost 1.5% lower. (Only one other country, Denmark, manages to achieve above-average expenditures and below-average life expectancy; their expenditures are still less than half ours and their life expectancy is higher).

Note finally the width of the lines, which indicates average number of doctors’ visits provided per year by each country: the fat lines are 12 or more; the medium lines are 4 to 8; the US comes in at an average number of visits per year per person that rounds off most closely to . . . zero. (Note also that of the 4 countries that average effectively 0 visits per year, two of them are the only two on the graph that do not provide universal coverage.) Not only does every other country on this list except Mexico manage to provide universal healthcare coverage at vastly lower expense than the US, not only do 2/3 of them achieve greater life expectancies than the US, not only do 2/3 of them achieve an above-average ratio between relative expenditure and relative life expectancy, but over 80% of them provide an average of at least 4, and in some cases 12 or more, covered visits per person per year for their entire populations.

Note in passing, too, that the only other nation that can’t afford to provide universal health coverage is Mexico, which spends less than 30% of the average among these nations on healthcare and is still getting a vastly greater bang for its its bucks than is the US.

The utter, abject failure of the US’s profit-sucking healthcare morass is made as stark here as it has ever been. Basically, we’re spending over $4,300 per year for every covered person for nothing whatsoever, and giving up over a year of average life expectancy as our reward – while leaving tens of millions of people with no coverage whatsoever for most or all of their needs! It would be almost impossible to have a healthcare system worse than this, other than one with even less protections for patients than the US already has.

As Ezra Klein notes:

consider this: If we spent what Canada spends per person, our deficit problem would go away entirely. And Canada’s per-person average is in a country where everybody is fully covered and so has full access to care. America’s is in a country with 47 million uninsured, and so many people skimp on needed care. So the comparison is actually unfair to Canada. . . .

This is serious pitchforks-and-torches stuff, if only people really understood it. I continue to believe, however, that the improbable size of the disparity is a barrier to understanding. People just don’t believe these numbers. America may not be the best, but we’re not supposed to be the worst by such a large margin.

Oh, yes, we are. The system is designed to suck money out and deny care. It’s working perfectly. But why do we have a system designed to do that?


There are some problems with the above graph, which I somewhat glossed over in the original post. DanM alludes to them in his comment below. It’s just as well to clarify some of these points.

First, the graph is somewhat misleading because it seems to position life expectancy as a direct function of healthcare spending: a certain amount of money buys you a certain number of years of life, and the slope of the line from one axis to the other describes the mathematical relationship between them. That is the inevitable broad-brush interpretation of the data, that is true (the whole point is that there is a link between the two factors, otherwise there’d be no point graphing them – and indeed the relation is clearly non-random as even a casual inspection of the graph shows) – but the line-graph format makes it much too literal.

Second, the scales of the axes are distorted. There is most obviously the fact that neither the expenditures axis, on the left, nor the lifespan axis, on the right, start at zero. The actual spread between high and low values on both axes is thus exaggerated, especially for lifespan. Also, the data ranges shown for each bear a very different relationship to the total range for data of each type: the top and bottom entries for healthcare expenditures span about 90% of the value of the top end of the scale, or about 80% if you exclude the US; the top and bottom entries for life expectancy span only about 12% of that range. If the two axes were scaled similarly, the right-hand values would all cluster into a tight knot and the blue lines would converge from high and low on the left into that small range, diminishing the impression of a clear correlation between the two values which is created by spreading the lifespan values out so much.

In addition, setting the average values of the two scales at the same vertical level is an arbitrary decision that reinforces the implicit message that the two are correlated. (A ratio between healthcare spending and lifespan that matches the dollars/years ratio of those average values will be a horizontal line at any level on the graph – thus those countries doing better on a dollars/years basis will have lines that slope up, and others will have lines that slope down.) Again, this is not unreasonable as a way of displaying this data, but it requires as an organizing assumption that the implicit correlation illustrated by the graph is in fact true – which puts the cart before the horse.

Finally, as Dan notes, there are other factors influencing lifespan, and implying that it is a direct function of healthcare expenditure, as this graph seems to do, is much too crude.

Nate Silver, brilliant statistical interpreter at “538“, recasts the same data in this fashion (click graph for larger version):

Healthcare Expenditure vs. Life Expectancy Scatterplot

Healthcare Expenditure vs. Life Expectancy Scatterplot

This graph is much fairer in certain ways. By removing the horizontal lines, it removes the visual implication of a direct mathematical function linking the two data sets. By graphing the data as a scatterplot on two orthogonal axes, it allows the viewer to draw their own conclusions without dictating a relationship in the design of the graph. Silver also takes the obvious steps of scaling the axes fairly and accurately, starting a zero for each.

However, this graph also supports the basic point made in the original version: there is an obvious trendline through the data set, and the US is an extreme outlier that falls insanely far below that trend. (To see how far, hold a ruler against your screen, paralleling the slope roughly marked out from the origin through the data cluster running up to the right – about where Canada falls out. Continue that line up to the right until it is directly above the red “USA” below. It should run off the graph up to somewhere in the third paragraph above the graph. That’s where the US should be, given what we spend (on only a fraction of our population). If you want it in numerical terms (and again taking the implied correlation rather too literally), US citizens who actually have access to healthcare should live more than 193 years, on average, if we were spending that money as effectively as most other countries do. From the reverse perspective, given the below-average life expectancy we get for our healthcare dollars, we could spend at least $4,000 per person per year less than we do¬† and still achieve our current quality outcome, if we were merely as efficient in our expenditures as, say, Denmark. That $4,000 – more than the average amount other industrialized nations spend per person in total – is the amount we are throwing away on our for-profit healthcare system, for no benefit whatsoever to ourselves.

It must be acknowledged that that correlation has not been subjected to statistical analysis, but the basic point is that the original graph, though its designers made some questionable choices, was not as bad as all that.

Hat Tip: to Andrew Gelman at Columbia, who did the original re-analysis from which Silver took his own version.

2 Responses to “How Can We Make It Clearer? When Will Anyone Notice?”

  1. Dan M. Says:

    So, given that sanitation and water quality are strong correlates with life expectancy, if we are actually above average on those (we are, right?), that means our mediocre life expectancy is actually craptacular.

    However, I think “crude” is even too generous an assessment of life expectancy as a metric. It doesn’t account for things like lifestyle choices, genetic stock, immigration, emigration, or dietary preferences.

    Arguably, our corn subsidies reduce our life expectancy more than or bad long-term care. (Yes, that’s a wild-ass guess with no stats, but it seems prima facie possible to me.) On the other hand, life expectancy does capture our shitty neo-natal care. On the grasping hand, it’s driven down by high-tech facilities that allow more would-be miscarriages to be brought to life birth.

    Really, it’s a crappy, crappy stat. Oh, and that’s one of the most irritating formats for a graph I’ve seen. If the goal is to show how we fail to match some otherwise present correlation, then just use a freaking 2-D graph! In fact, graphing of that bad quality makes me suspicious of the intentions of the grapher.

    (By the way, it’s a good thing you post so rarely, otherwise I’d have to install and learn to use one of those “rodents of substantial size” readers, and I fear and detest web technology.)

    (No, wait, I’m lying. I’m just lazy.)

  2. Dan M. Says:

    That’s a vast improvement,, so that the graph at least means something. But I think it still doesn’t mean much at all. One of Nate’s commenters covers at least the most obvious problems, but even that doesn’t cover the much more problematic claim that there’s “a strong correlation”. First, Why are we even using a linear scale in dollars? Why are the dollars not adjusted for local buying power? Why is there no discussion of the fact that there clearly exists two separate populations in the graph (not counting the US), one for all the developed countries, which seeps to actually have a pretty weak correlation, and one for developing countries, which actually is much more obviously correlated?

    The new graph is a huge improvement. It’s now only kinda crappy. But thanks much for the update.

Leave a Reply

Logged in as . Logout »

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

CommentLuv badge


Sufficient Scruples:



January 2010
« Dec   Apr »

Links & Feeds:

RSS 2.0

Comments RSS 2.0


Follow KTKeith on Twitter


Powered by WordPress

Get Firefox!

Theme copyright © 2002–2014Mike Little.

Ask the Ethicist!


White Papers:

Bioethics Links: