This article seems relevant: http://lesswrong.com/lw/20i/even_if_you_have_a_nail_not_all_hammers_are_the/ http://lesswrong.com/lw/20i/even_if_you_have_a_nail_no

This article seems relevant: http://lesswrong.com/lw/20i/even_if_you_have_a_nail_not_all_hammers_are_the/ http://lesswrong.com/lw/20i/even_if_you_have_a_nail_not_all_... In summary, many of the studies that have shown scary correlations like "multivitamins increase mortality" are (according to the author) based on inappropriate (but easy to apply) statistical models. Scientists are too-often not trained well enough in how to evaluate the model they are using to see if it is appropriate for the task, and peer review boards often do not include statisticians. A simple summary of how this can apply to vitamin supplements: suppose you do a trial where you give different groups different doses of a vitamins and then monitor their health. Most vitamins are detrimental to have too little of, helpful to have the right amount of, and then increasingly detrimental to have too much of. The graph of their benefit/harm would be a curve starting at a low number, curving up to a high peak, and then dropping down to lower and lower numbers off to infinity. If you fit this graph to a line, you'll see a downward slope, giving an indication like "multivitamins increase mortality rates".