Eggplant.place

TL;DR: a more practical companion to Expert Political Judgment, useful for forecasts in politics, economics, and other areas of human activity, if not quite as compelling as its predecessor. 3 1/2 stars. Superforecasting is a book that presents the results of Philip Tetlock's research on prediction and avoiding the biases that afflict even experts' forecasts within their domain of expertise. The book is something of a sequel to Expert Political Judgment, which shows that, on average, the political forecasts made by experts don't fare much better than random, when one takes into account both accuracy and calibration (that is, events which were attributed high probabilities occur with higher frequency, and the converse for low-probability events). This dismal scenario is alleviated by the discovery that some experts had a significantly better performance than that, and that those experts shared certain traits. That silver lining led to the creation of the Good Judgment Project led by the author, which studied the behavior and results of those overperforming predictors. The book describes some of the findings of that study: despite the (practical?) impossibility of perfect prediction, some heuristics and habits can improve the accuracy of forecasts in significant and persistent ways. The author claims, supported by the project's evidence, that testability plays a central role on that improvement: if there are metrics that allow the verification of how right or how wrong a given prediction (or predictor) was, then the higher level of scrutiny will push forecasters towards "good judgment" as described by the Good Judgment Project. This programme is open to some critiques that the author acknowledges but -- at least not in my opinion -- does not fully address, especially about the relevance and scope of the kind of predictions which have results that can be easily metrified. Yet, as the author claims, those doubts don't mean that a programme for improving forecast, even if somewhat narrow on its outputs, would not be beneficial for the public debate on the relevant subject matters. While, in general, I think that Expert Political Judgment is a more interesting book, Superforecasting has a much wider and more general scope. As such, it is a good read for those interested in practical forecast at whatever level or topic.