- Most epidemiology papers I’ve seen do too many tests, which has the same effect as throwing away data. However, the criticisms of epidemiology I’ve seen are often too harsh (e.g., epidemiology is worthless). The next time someone criticizes epidemiology, point them to these three papers on prenatal pesticide exposure and IQ.
- positive reinforcement training improves chimp welfare
- Rajiv Mehta introduces Tonic, an app to help you remember and track recurring events, such as taking medicine. Video.
- The STAR*D scandal (STAR*D was a large study of the effectiveness of antidepressant drugs).
- The statin scandal. “The years 2008 and 2009 [were] very disappointing for cholesterol experts and the cholesterol drug industry.” Via Steve Hsu. Statins are expensive and have serious side effects — and it is now clear they are close to worthless. My half-stick of butter/day is looking better and better. (Link fixed.)
Quick question, Seth, as I’m not sure if you’re still following the earlier butter threads:
Is the half-stick of pastured butter now your only daiily fat/oil supplementation? Or are you also still supplementing with 2T flax oil? (Or if not flax, what else?)
Thank you.
Joe, in addition to the butter I still take a flaxseed supplement, now use ground flaxseed (66 g/day) rather than flaxseed oil. I grind the flaxseeds myself using a blender and eat it with yogurt. Whole flaxseeds are much easier to store than flaxseed oil, which must be kept cool. I suspect flaxseed oil goes bad faster, too.
Thank you.
Can you explain what you mean by this?
“Most epidemiology papers I’ve seen do too many tests, which has the same effect as throwing away data.”
By too many tests I mean too many statistical tests — tests of hypotheses. So they get too many false positives. They don’t correct for the number of tests done. They do too many tests in two ways: 1. They divide the data into many pieces (e.g., men and women) and test each piece separately. 2. They fail to combine across measures (combine correlated measures into one). For example, testing diastolic and systolic blood pressure separately, even though they are correlated. This has the effect of adding noise (false positives) — thus decreasing the signal to noise ratio. When you throw away data, you decrease the signal to noise ratio.
interesting stuff re:epidemiologists doing too many tests, enjoyed that