Book Review: The Art of Statistics by David Spiegelhalter

I recently finished reading “The Art of Statistics” by David Spiegelhalter.  That’s *Sir* David Spiegelhalter.  One can be knighted in the UK for one’s contributions to statistics, and Professor Spiegelhalter achieved that distinction.

The depressing part of the book was the inclusion of the author’s own work on the case of British doctor Harold Shipman, who murdered hundreds of his patients, mostly women, before being caught and convicted.  The author was part of a team tasked by the UK government to determine if the serial killer-doctor could have been identified earlier using statistics.  The team put a control chart on physicians’ patient-death-rate, but such a crude algorithm would take about 8 years before raising a flag.  I wondered whether patient-death metadata would be a better fingerprint.  Harold Shipman’s victims all died in the early afternoon during regular business hours.  I wondered whether time-of-day-at-death, coupled with crude death rate, might have made a decent algorithm.

Metadata can be surprisingly useful.  Apple’s algorithm for ‘finding’ your CD label after you add it to your music collection depends entirely on metadata.  It turns out that if you use the number and length of tracks on a CD, you get a fairly unique “fingerprint” identifying the CD label.  If there is any doubt, the user might be asked to make a selection from only a few choices.

The chapter on Bayesian statistics piqued my interest on the subject, and I plan on delving into it further. The author expressed that there are three battling schools of thought within statistics: Fisher followers who use p-values, Neyman-Pearson followers who calculate statistical power, and Bayes people who distinguish between epistemic uncertainty and probability. I think that all three schools tackle different aspects of the same field.  Rather than being at odds (pun) with one another, they are simply complementary tools. The book included statistical analyses of data on statin use, breast cancer survival, hospital success rates, number of sex partners in the UK, Higgs boson, whether some skeleton was Richard III, and other more delightful subjects.

The author included a discussion of good and bad uses of statistics. The good examples were British, and the bad examples were American. (Smile.) I will forgive the knight the slight in appreciation of the British humor he speckles throughout the book.  It was a good read despite its (few) flaws. The main flaw, in my humble opinion, was the relative lack of math in a math book.

I enjoyed the book because I love statistics, analysis, and math. Overall, “The Art of Statistics” was a delightful read.