"The Signal and the Noise"

The new book by Nate Silver about the statistics involved in making predictions from gambling and stock markets to weather and climate to terrorism was quite good. The perspective that each topic adds is not trivial and the story is layered enough that it binds.

Take the weather chapter for example. The discussion of human centered forecasting based on accurate predictive models and the relative success of predictions over time was well grounded. Of course there are always nuances missed and you just have to get over them. But the idea that weather forecasts have been successful is because we have paired human cognition with technology advances. The heuristics of forecasting the weather are understood, albeit arguably on an intuitive or experiential level. Nevertheless, there are gains in understanding our successes and failures quite rapidly because of the ability to test hypotheses by continually collecting data and turning it into information, and updating our understanding. The key tenet of Bayes theorem. Use what we know as a starting point, collect data, and update that understanding because of data.

This is unlike many fields including terrorism where the signal is only evident after the attack. Mental models do not exist. You have psychology, lack of signals, deceptive acts, and then blips on the radar that seem disconnected to anything you knew, until you know what happened that is. Then you can work backwards to understand the signals and the noise. This isn't very productive when you are trying to make predictions!

The failure of imagination theme (e.g. out of sample, rare events) runs throughout the book as does Bayesian thinking. The hidden element to the average reader may very well be the notion of ensembles (easily recognizable to modelers). Its treated in many contexts including crowdsourcing, averaging and bias correction. The human element discussions that compose things like the stock market and rational behavior are the most foreign to me and yet were the most perspective broadening topics I read.

The ideas presented don't lead to Bayesian thinking, because, well the book stays anchored to it. Use the facts to update your view of the world, even acknowledging that bringing your biases to the table is helpful, but only if you are willing to update them.

I wont offer any more in the spoiling dept. suffice it to say the section involving Israel and terrorism is the clearest example of choosing your battles carefully, accepting (probably the wrong word choice here) sacrifice to avoid significant catastrophe. This is not to say it can't or won't happen, but that the probabilities are made low by making good probabilistic forecasts about cost and loss, while at the same time allowing pressure to be released from the unstable system. Understanding what goes in and what comes out allows the system to maintain predictability, at least to the level you understand such systems.

But never forget that your knowledge is imperfect at best and that fact constantly requires you to update your hypotheses.