There are some good bits in here. I love the subtitle especially: "The real challenge isn’t being right but knowing how wrong you might be." Knowing when not to provide an answer is hard. A big part of my job is communicating statistical findings and giving a good non-answer is much harder than giving a good answer, both technically speaking and socially speaking.
One thing I'll add that drives me nuts is the fetishization of bayesian reasoning I see some times here on HN. There are times that bayesian reasoning is helpful and times that it isn't. Specifically, when you don't trust your model, bayes rule can mislead you badly (frequently when it comes to missing/counterfactual data). It's just a tool. There are others. It makes me crazy when it's someone's only hammer, so everything starts to look like a nail. Sometimes, more appropriate tools leave you without an answer.
Apparently that's not something we're willing to live with.
Thinking Fast and Slow left me with a feeling of despair about the human inability to reason effectively about statistics.
I like to tell people that charts work better for asking questions than answering them. Once people know you look for answers there, the data changes. More so than they do for question asking (people will try to smooth the data to avoid awkward questions).
Maybe I'm just missing the joke here, but "Bayesian reasoning" is hardly needed to realize that if the sun did explode, the $50 you'd lose in the bet is worthless anyway.
That's not the joke. The point is that the sun exploding is extremely rare, so you can essentially conclude that the trial is a false positive. If you just look at the p-value you ignore p(B| not A) (false positive), and conclude that the sun exploded (no actual statistician would make that claim in this example, but they sure do in more subtle situations).
One thing I'll add that drives me nuts is the fetishization of bayesian reasoning I see some times here on HN. There are times that bayesian reasoning is helpful and times that it isn't. Specifically, when you don't trust your model, bayes rule can mislead you badly (frequently when it comes to missing/counterfactual data). It's just a tool. There are others. It makes me crazy when it's someone's only hammer, so everything starts to look like a nail. Sometimes, more appropriate tools leave you without an answer.
Apparently that's not something we're willing to live with.