Wednesday, December 29, 2010

Snake Oil

The folks at Information is Beautiful have assembled a nice dataset consisting of measurements of the effectiveness and popularity of 129 different dietary supplements.  The graph they produced is appropriately beautiful and interactive, although I don't find it particularly useful for addressing the main question -- is there some relationship between effectiveness and popularity?  That is, are people using the useful supplements or are they just popping snake oil?

I downloaded their data and produced the following scatterplot.  I've highlighted some of the key outliers.
Some interesting lessons: First, the relationship between effectiveness and popularity is positive and statistically significant at the .001 level.  That's encouraging!  But some of the outliers are rather interesting.  Vitamins A and C are apparently way overhyped.  St. John's Wort, which is reputedly useful for mild to moderate depression, is under-utilized, especially when you consider the size of the industry devoted to treating depression.  No one really uses anti-oxidants, and that's a good thing, it seems.  Green tea, folic acid, fish oil, and Vitamin D (my personal supplement of choice) are the real winners here.

(h/t Harris Masket)

1 comment:

Andrew Oh-Willeke said...

The evidence for effectiveness scale is pretty far off. This is a literature that I follow pretty closely in the scientific journals. Saint John's Wort should be near zero. Vitamin C is grossly underrated. Fish oil should be in the middle -- it is not at all effective for many of the benefits it is marketed for (e.g. dementia prevention) but does have positive anti-inflammatory aspects. Given that so many identified data points are wrong, I am sure that many other data points are wrong. The effectiveness scale is not based on any of the good meta-analysis papers conducts to date.

Also, the statistical significance is illusory and is coming mostly from lumping data in arbitrary categories that aren't closely scaled rather than treating them as individual points. And, in the scheme, any level of inclusiveness of powerfully influenced by selectivity in the data set.