Book Review: The Signal and the Noise

I just finished reading Nate Silver's The Signal and the Noise: Why So Many Predictions Fail — but Some Don't. I enjoyed it overall, probably the best basic introduction to Bayesian reasoning I've seen. Silver runs through a huge variety of fields that make predictions--earthquake predictions, election predictions, baseball player performance predictions, stock market predictions, weather predictions, on and on. It makes for a fun read, and is a great example of how basic math/statistics form the basis of every meaningful field of study.

I do have to complain about a few glaring typos and things.

• The very fitrst chess diagram he shows is wrong. Really, only 3 moves into the game and you can't get it right?

[caption id="attachment_76" align="alignnone" width="257"] This get's labeled "Position after Kasperov's [playing white] 3rd move in game 1" Go ahead, try to make a chess board look like that in 3 moves.[/caption]

• Quoting Richard Rood, "At NASA, I finally realized that the definition of rocket science is using relatively simple psychics to solve complex problems" [emphasis added]. As someone who has to explain the difference between astronomers and astrologers, this struck me as funny.
• Silver keeps using the word "reflecting" when he means absorption and re-emission. This is actually a really important distinction in the whole global warming biz. Light that reflects off earth's surface (or cloud tops) heads right back out to space, it doesn't do any heating. Light that gets absorbed and then re-emitted in the infrared is what gets trapped by greenhouse gases. Maybe I'm wrong, but given the previous two typos, my Baysian prior is strongly set to think this is an error in the book. I also think he gets a little too cute trying to keep some skeptical street cred when looking at the climate models.
• Silver spends a lot of time comparing earthquake predictions to terrorist attacks, and makes a big deal out of both of them having power-law distributions. As an astronomer, I see a lot of power-laws and have become a bit jaded. It's become a bit of a joke that you can plot just about anything on a log-log scale and have it look like a meaningful relation. Anyway, making predictions from extrapolated power-laws sounded like a dumb idea before I read the book, and it still sounds like a bad idea.
• I was surprised he didn't include a chapter on technology predictions like the Y2k bug predictions or if/when Moore's Law will end.

Some highlights include:

• Great chapter on stock market predictions. Take away message is that pretty much the only people who beat the market consistently have insider info.
• Step-by-step example of how to useBayesian reasoning to calculate the probability your spouse is cheating on you if you discover someone else's underwear in your dresser. Seriously.
• "some dufus has to reboot al qaeda's servers"
• Most "for-profit" weather forecasters have a "wet bias". They intentionally over-predict rain.
• He talks a lot about Philip E. Tetlock's Expert Political Judgment, which is also an awesome book, so go read that first.
• The book includes a lot of details about Silver's career path--going from a boring office job, to baseball analytics, to on-line poker, to election forecasting at 538. Safe to say, no one would have predicted that one.

Overall rating: Good book, I probably should have waited for the paperback to come out though.