2 ms·
Big datasets matching health outcomes to blood sample spectroscopy seems like a good candidate for identifying chemicals which detrimentally affect human health
by londons_explore 6y ago
Big datasets matching health outcomes to blood sample spectroscopy seems like a good candidate for identifying chemicals which detrimentally affect human health by small amounts.
Obviously you won't be able to separate cause and effect, but in my view if "X chemical in people's blood leads to a higher likelihood of dropping out of school", thats a good enough reason to restrict the use of X, even if it sometimes turns out there is no causal effect.
- robocat 6y agoIt is a statistical nightmare to do what you suggest. 1. False correlations: https://en.wikipedia.org/wiki/Data_dredging https://en.wikipedia.org/wiki/Data_dredging 2. True correlations: https://en.wikipedia.org/wiki/Confounding https://en.wikipedia.org/wiki/Confounding 3. The number of normal compounds in our body from cell processes or digestion is large. Measuring tiny amounts of unknown compounds is extremely hard in that environment. A better way is to proscribe more in-depth testing of compounds on animals. Removing all other factors gives statistical power to observations.