Tom Chivers, Everything Is Predictable: How Bayesian Statistics Explain Our World (Book Review)

By jgn on Friday, July 5, 2024 in Reading, Reviews, and Technology

I'm not sure why, but I've recently seen a lot of mentions of Bayesian statistics in the things I read (esp. the use of the word "priors"): So I decided to read a book about it. There are two sides to this book. One is a critique of "frequentist" statistics (i.e., mainstream statistics that compute a p-value -- this is the statistics you learned in college). That's quite good, and there's an entertaining chapter on how many studies nowadays can't be reproduced, in part because scientists are prowling their data for subsets that support decent p-values); change the data just a bit, and you struggle to find the same likelihoods.

The other side is a kind of "theory of everything" for Bayesian ways of thinking. This actually seems a little less successful -- but I might have to read the book again. In short, the book memorably argues that everything is ultimately prediction. There's a line at the end that says "beliefs are predictions." The author would likely be the first to admit -- on Bayesian grounds -- that anyone who believes that beliefs are not predictions would be entirely unswayed by this book.

Originally reviewed on Goodreads on 5 July 2024.