Silver's main thrust is that the kind of statistical modelling the general public is often exposed to suffers from poorly understood raw data being run through overly idealized models. Nate Silver, again and again, using examples from weather, climate, earthquake, politics, and gambling, shows that many popular statistical techniques are seriously flawed. He believes that the Bayesian modelling technique, which incrementally changes probabilities based on prior data, is a superior approach that should be used more often.
I chose this book because I was impressed with Nate's political predictions on FiveThirtyEight.com. I came away a little less impressed after reading the book.