Hmmm true. So let's see what Bayes' formula says : A = recovering drug addict B = suicide victim Bayes theorem P(B|A) = P(A|B) P(A) / P(B) First link google

Hmmm true. So let's see what Bayes' formula says : A = recovering drug addict B = suicide victim Bayes theorem P(B|A) = P(A|B) P(A) / P(B) First link google found with some numbers : http://www.ncbi.nlm.nih.gov/pubmed/1932152 http://www.ncbi.nlm.nih.gov/pubmed/1932152 P(B) is much smaller than P(A), assuming most recovering drug addicts do not actually commit suicide. Let's say 3/4 don't (this is meant to be a lower bound, since increasing this number will mean drugs are responsible for more suicides, given the rest of the calculation). Therefore P(A) / P(B) ~= 4. So this depends on what fraction of suicides drug addicts represent. If that fraction is above 25%, that means drugs lead to suicide. If that fraction is less, then they don't. According to the article that fraction is between 25% and 70% (in 25% of cases suicide victims were clearly addicts, in 75% of cases drug abuse was seen before the suicide). While this is not too rigorous (esp. the sources part), this seems to indicate that "scientifically" you'd clearly find drugs cause suicide. If the fraction of suicides were to be ~ 45%, that would mean drugs increase suicide risk about 4-fold (and thus the number of suicides in a given group). That does not sound like an unreasonable number to me. It also means that free availability of drugs, increasing the number of people exposed will increase suicides by a factor between 1 and 4, depending on how big the problem currently is. An effective banishment of drugs from our society would cut suicides by that number, between 25% and 70%, assumed to be about 45%. The obvious criticism of this would be that P(A) and P(B) are nothing but a rational estimate, given that you ONLY know the things on the right side of the equation. In practice we know more. It's still very likely to be a good approximation though. So good point about reversal of causality, but following through on that point yields more evidence for it to be true, not less.