# Book Review: Seven Pillars of Statistical Wisdom

The Bottom Line: 35 stars

Seven Pillars of Statistical Wisdom by Stephen Stigler purports to share seven of the most central insights in the history of statistics, what they mean, and why they were revolutionary when first introduced. The book sometimes reaches this lofty ambition, but often sags under the weight of that challenge. Despite the unevenness, it’s definitely worth a read for statisticians and data scientists alike.1

Stigler sets forth the Seven Pillars (Aggregation, Information Measurement, Likelihood, Intercomparison, Regression, Experimental Design, and Residual) as deep truths in modern statistical thinking. And indeed, much of Stigler’s discussion on these topics is fascinating, but the book unfortunately sometimes gets caught up on historical details rather than focusing on the importance or meaning of the pillars.

The book is at its best when it focuses in the scientific and philosophical implications of the pillars. The Information Measurement’s exploration of how statistics flaunt the obvious expectation that the 10th observation matters as much as the 1st by gaining information at the rate of $\sqrt(n)$ is enlightening. Similarly, the section on regression includes the most thoughtful and thought-provoking discussion of regression to the mean that I’ve ever seen. I’m still caught up on it.

Unfortunately, other sections suffer from too much focus on the trees and lose steam in the face of overwhelming historical detail. The aggregation section was particularly guilty on this front. I am the (relatively rare) type of person who finds the paradox of throwing away data to learn something when taking a mean to be fascinating, but Stigler spent all too little time examining this issue in favor of explanations of how to read aggregation tables from old books.

And it never really becomes clear what some other sections, especially Intercomparison and Residual, are really about. Stigler sometimes seems to want to define by example, but a clearer approach to defining pillars definitely could’ve been helpful in the more technical sections.

Unlike some other books on the history of statistics, Seven Pillars was definitely written with a goal in mind – one it often meets. It’s too bad that it’s dragged down by some sections that need a little more clarifying and editing.

1. The book makes frequent use of pretty technical terminology without much definition. This book is probably well-within the grasp of someone who’s taken a statistics or econometrics course that included a section on likelihood estimation, but is probably a little over the head of statistical novices. [return]