Monday, July 13, 2015

The Signal and the Noise by Nate Silver

I've written about management fads before, and read about them too. They're often silly yet pernicious, as knowledgeable individuals give themselves to a cycle that they've seen and experienced before. Prior knowledge of the world is thrown out of the window because this time it's different. The Signal and the Noise is a sort of anti-management fad book. Anyone remotely interested in technology has surely heard of Big Data, but I had yet to read a convincing account of what it is and isn't. Analyzing large chunks of data with incredible speed is great, but it can't change a few of the basic facts that surround data analysis.

And Nate Silver is more or less the go-to guy for anything related to data analysis. He runs the website fivethirtyeight, which gained some notoriety for correctly predicting all of the US presidential elections, in all fifty states. He's young, and he definitely isn't an academic. Overall, The signal and the Noise is well researched, well written and insightful. It's sincere (mostly) and doesn't preach; all great qualities for a non-fiction book. Silver guides us through the basics of data analysis and prediction, and goes into detail in a few well-selected areas such as poker, baseball, economics and weather. Baseball might not interest non-Americans as much, and it treads familiar subject material for readers of Michael Lewis's Moneyball, but Silver's prose and insights are valuable all the same.

If there is one thing that I personally learned from this book, it's that in order to accurately predict something you need to both understand the phenomenon in question and possess accurate data. Earthquake data is somewhat plentiful, but earthquakes themselves are poorly understood (relative to the weather, for example). Therefore earthquakes are almost impossible to predict, with any reasonable accuracy. In somewhat the same way, social phenomena are difficult to predict, even with Big Data; the whys and hows need to be understood first. Don't trust forecasters who only use data to support claims, these cases often mistake the noise for the signal.

While, I did mostly enjoy Silver's thoughts, I did sometimes question his commitment to Bayesian statistics. Now, I'm no professor of math, but I've taken my fair share of statistics classes and from what I can tell, these methods do provide accurate insights, when used reasonably well. The Bayesian theorem is useful, but let's not pretend that it is a jack-of-all-trades solution for statistics and, well, life.

No comments:

Post a Comment