Berkson's Paradox
Two things with nothing whatever to do with each other can look strongly related — purely because of who ended up in the study.
194615 min read
Israeli-American computer science and philosophy of causation
1936
Computer scientist at UCLA and winner of the 2011 Turing Award for his work on probabilistic and causal reasoning in artificial intelligence. He developed Bayesian networks and then the do-calculus, which turns causal questions into formal ones. His treatment of Simpson's paradox — that no amount of data can tell you which table to trust, only a causal model can — is the standard modern resolution.
Two things with nothing whatever to do with each other can look strongly related — purely because of who ended up in the study.
194615 min read
A treatment does better for men, and better for women, and worse for people. None of the arithmetic is wrong.
1903 (Yule), 1951 (Simpson), 1972 (named)20 min read