A History of Data Visualization and Graphic Communication

38 practical, and more effectively promoted, than their Bayesian counterparts. I am also not certain about the extent to which the replication crisis is due to the use of the p-value. Although it is certainly a contributing factor, I personally believe the root cause is that the data analyst is usually fully invested in the outcome, and has every incentive imaginable to obtain the most flattering result. All in all, I believe a book as belligerent and instantly controversial as Bernoulli’s Fallacy deserves an appendix in which some of the claims are debated with other statisticians (cf. Berger & Wolpert, 1988). My final misgiving is that the author occasionally gives Ed Jaynes a little too much credit. For instance, the notion that all probability statements are conditional on prior knowledge is found in Keynes (1921), and both Jeffreys and Lindley consistently conditioned on “H.” (for “history”) or “K” (for “knowledge”) before Jaynes. Similarly, the idea that Bayesian inference is a logic of partial beliefs predates Jaynes—it goes back at least to De Morgan (1847/2003), Ramsey (1926), and de Finetti (1974). Despite these minor misgivings, this book comes highly recommended. Bernoulli’s Fallacy elegantly connects the past to the present in an attempt to dismantle the reigning statistical orthodoxy. Buy this book, and give it to your students so they may learn about Bayesian inference and the history of statistics; give it to your colleagues working in the empirical sciences so they will understand that the frequentist emperor is scantily dressed; give it to your frequentist friends as a provocation. Or read it yourself, so you will be prompted to think more deeply about the foundations of statistical inference.