The Signal and the Noise: Why So Many Predictions Fail—But Some Don't
(excerpts) An excellent popular (easy to read) overview of a variety of statistical topics, with a good focus on not fooling yourself with overfitting. Silver, somewhat like Meehl, is a subjective Bayesian decision-theorist in fundamental outlook and approach to analysis, but a methodological pluralist, which makes some of his work a little confusing: he is judging things by how they approximate a proper fully Bayesian decision analysis (as is necessary for betting and other applications of forecasting), but this is not always explicit and he can’t compare the implemented methods with the gold standard (because they’re too difficult to implement, which is why he falls back to more conventional methods). And some of the technical aspects are a little weak (the Hume discussion comes to mind), but what do you expect, Silver’s a busy guy.