Ethical experiment: "natural" experiment. Say, WWII: GI's were
given free cigerettes. More got addicted than those that
didn't go. But the two groups are very different.
Handle each confounder individually: Age, sex, genetics,
income, education, exercise, nutrition, etc.
But, you also have to handle all the confounders at the same
time!
Compliers/Adherers vs non-compliers
In a randomized clinical trial, clofibrate didn't effect heart
disease. (21% vs 20%).
But many didn't take the drug
Of the 80% assigned to take the drug that took it, 15% died.
VS 25% of those that didn't comply.
Sounds like the drug is effective?
Compare to placebo-non-compliers: 15% vs 15%.
In clinical trials, it is important that placebo is as "tasty" as the
treatment, so difference in compliance aren't created.
Pellagra occurs in households, in southern Europe. Then in
southern US. In homes that have more fly's than average.
Spread from spain to italy to US. Sound like an infectious
disease? No, it was niacin.
Jews and Moslems have circumcision and lower cervical cancer.
Casual? No: current theory is it is a STD.
Almost any relationship between two variables can be explained
by some "story." That a reasonable story can be found doesn't
provie support of the theory.
Sex bias
Berkeley admits about twice as many men as women. Is this discrimination?
What cofounders
More men apply: After correcting for that it is 45% vs 35%