Bayesian data analysis, which Andrew Gelman has pioneered, is about taking one’s beliefs into account when doing data analysis. When I wrote The Shangri-La Diet, I was being a kind of Bayesian: I realized that the facts I had gathered so far did not establish the diet as any sort of panacea. Based on the facts in the book, it was hard to say how widely helpful the diet would turn out to be. I wrote the book anyway because the facts I had gathered so far were so surprising, so inconsistent with what almost everyone said about how to lose weight. From a Bayesian point of view — taking prior beliefs into account — they were impressive. If conventional views were right, no one should lose weight following SLD. But several people had. Some of them, such as Tim Beneke, had lost a lot of weight. To complain that there was no clinical trial, no certainty, was to miss the point that the book includes data that should have been impossible.
My first reaction to [SLD] was, of course, that it was one of the stupidest things I’d ever seen. Then I started reading the forums on the creator’s (Seth Roberts) site, and then I did some Googling. And would you believe that, in the absence of anything that I would call scientific evidence, this thing seems to work for most people that try it. . . . Five days ago I honestly believed the Shangri-La Diet to be hooey — interesting hooey, maybe, but still hooey. . . I decided I’d try it for myself and report on the results. I want to make it really clear that I approached this diet with a very healthy dose of skepticism. You should also understand that I’m a staunch advocate of the “eat right and exercise, stupid” philosophy of weight loss. I have never followed a prepackaged diet strategy. Having said all that: it works. I do not know why or how it works, but it works.