21 November 2012

Receiver Operating Characteristics

Cafe Hayek | Don Boudreaux | Testing Jonathan Haidt

I happened across this paragraph from the site breastcancer.org – not a political or ideological site:
Biopsy is usually a simple procedure. In the United States, only about 20% of women who have biopsies turn out to have cancer. By contrast, in Sweden, where cost accounting is much stricter and only the most suspicious lesions are biopsied, 80% of biopsies turn out to be cancerous (malignant).
Notice how this account can easily be read as evidence (1) against the alleged superiority of heavy government involvement in financing health care, and (2) for the alleged superiority of heavy government involvement in financing health care.
Sweden and the US sit on different points along the biopsy ROC curve. That is all you can conclude from this extract.

File under: "benefits to studying Machine Learning that have nothing to do with technology."


  1. Sweden and the US also sit at different points on the national health care system curve. Sweden has had socialized medicine for quite a while, and the progressives there have realized that they have to control costs. In the US, the progressives are still looking forward to a future where everything gets paid for, and the funding comes from the same account that produces free ponies forever. The Swedish progressives are harder and nastier than the naive, romantic US progressives.

    And, it is interesting to talk to people here (France) about what is covered and what is not, and who is covered, and who is not(e.g., a pregnant woman who is only covered in her country of nationality, but not in her country of residence). The stories could fill a couple of weeks of wailing on NPR about inadequate medical care.

  2. I completely agree. But (1) we can't tell that from the extract above, we just know it as background, and (2) that's one cause rather than effect of being at different places on the ROC. My point is just that the statistic cited above doesn't tell us, by itself, which system is better. I believe the US system is, but if all we know is the true positive rates of biopsies we don't have extra evidence one way or the other for normative claims.