by Nate Silver Print book
Forecasters, how to be less wrong with Bayes   (2012-12-25)
Silver has the knack for turning to a good story to illustrate his points, so the book is entertaining as it guides the reader to some profound points about predicting the future. The ideas and the delivery are both solid and well done.
Nate Silver's basic thesis is that, especially for forecasting, Bayesian statistical models should be used rather than Fisherian or frequentist models. R. A. Fisher campaigned against Bayesian statistics, because with Bayes one introduces ones biases into the model explicitly. Fisher wanted to make statistics objective, without the biases of the researcher intruding. But the problem is that, especially in forecasting, the researcher builds biases into the model regardless, so, if the biases are explicit, then at least you cannot fool yourself into thinking they are not there.
As far as the structure of the book, you can generally skip a chapter if the topic does not interest you. If you read the last six pages or so of each chapter, you will get the chapter's take away message that is relevant to Silver's overall thesis. To me, the important chapters for the thesis are chapters 5 (overfitting data) and chapter 8 (on Bayes). I found the chapters on economic and political forecasting very good. I skipped the chapters on baseball and poker, and I don't regret that. I think you could easily skip the chapter on chess, unless you are partial to chess.
Was this review helpful to you?