Bernoulli's Fallacy

Really phenomenal. This is what I wanted The Theory That Would Not Die to be - that is, a book that explores why Bayesian statistics/Bayesian rationalism is meaningful and powerful. I guess that one was a pop stats history book, whereas this is aimed more towards practitioners.

That said, this book is written with such a lively, passionate tone, such joie de vivre and humor and brio and so on, that I often cackled with glee. He also made me get pumped again about being Bayesian. Someone at work recently asked me, “Are you frequentist or Bayesian?” My response: “Bayesian. But we all are, in our hearts.” I definitely believe that. Bayesian stats aligns with what people THINK many frequentist concepts (confidence intervals being a big one, p-values being another big one) are conveying.

But this book also tied things together in an unexpected way. Namely, I had never thought that the replication crisis which created such drama in so many fields (psychology, economics, medicine) was almost a deterministic result of using frequentist statistics. My take had always been a pragmatic/economic one: that is, that the incentive structures of academia and “closed science” (aka paywalled journals) had created a perverse system of p-hacking, etc. Clayton (who taught a course at Harvard's Extension School that became this book - I WISH I had taken that course!!!) argues that frequentism is just rotten to the core, and it's now pulling the rest of empirical science with it.

Very eye-opening, Pikachu shocked gawping, was that the American Statistical Association had been sounding this alarm as well - for years! I had no idea.

Anyway, I'm going to go tell everyone at work that we need to be Bayesian from now on!