Frank Hansen writes:
Life expectancy is an outcome with possibly many confounding factors like genes or lifestyle, other than cost.
I used the 2007 oecd data on system resources to construct a health care system score. Here is the graph of that score against per capita cost:
I’d prefer to just put in the country names without the dots, but that’s a style issue. Here is the writeup of what Hansen did.
I don’t really have anything to add here, just following up on my earlier entry on the topic.
Would it be possible to subdivide the US data somewhat? I've heard claims that you can segregate the US into two populations: one that has white-collar jobs that come with good health care, and one which is often uncovered or which must pay for its own health care. I think the criterion I was told about to distinguish the two was simply race, but most likely it's a proxy for something more subtle. In any case, the claim was that the first population has a life expectancy in line with other developed nations, while the second population more closely resembled that in a third-world country.
I suppose even if deaggregated data on life expectancy were available it would be a big challenge to evaluate the health-care costs of the two groups separately.
What if inovation were added as a variable?
Don't know where you would find the data you'd need to do that. Census might have it. Elizabeth Warren's analysis of bankruptcy risk and the middle class seems like it must have used those kinds of data.
"the claim was that the first population has a life expectancy in line with other developed nations, while the second population more closely resembled that in a third-world country"
Amartya Sen in his book "Development as Freedom" covers this issue in page 22 (follow the below link). He has got the sources of his data below the charts showing life expectancy.
http://books.google.com/books?id=Qm8HtpFHYecC&lpg…
Hope this helps. Otherwise its a great book on thinking about development issues. Ethics meets Efficiency.
The problem with the PCA approach is, and this might explain why "official" rankings differ from Hansens, is that the PCA does not necessarily weigh the data in a meaningful way (total hospital beds maybe more important than numbers of mammographs). Or am I wrong about this?
And thanks To Frank Hansen for laying out the anlysis. Great learning resource for newbies like me!
I agree that weights from PCA may not yield the "best"[whatever that means] linear function of the resource measures. However, because the resource measures have different scales the data have been standarized (mean=0, std dev = 1, or equivalently cor=True in the princomp(.) function call). So 1 hospital bed does not carry the same weight as say 1 MRI machine.
You can still still claim that the resulting weights are not optimal, but then you get into defining what is optimal from a policy or social welfare point of view.
For me the appeal of PCA in this case is avoiding arguing about what is optimal. And PCA could do better than you might think since presumably if a given country has too few MRI machines compared to hospital beds, then resources may adjust depending on the needs/perceptions of the country.
So PCA is a reasonable first iteration, possibly not too far from "optimal."