Monday, July 07, 2008

I recently finished reading Leonard Mlodinow's book The Drunkard's Walk: How Randomness Rules Our Lives, which I'd like to discuss. In the name of full disclosure, in case this may have colored my reaction to the book, I want to note that Dr. Mlodinow contacted me, offering a free review copy of the book, and I accepted.

Perhaps the foremost task of hot-hand analysts is trying to determine whether streaky sports performances simply reflect random variation on established generating processes (i.e., like a tossed coin coming up heads several times in a row due to chance) or something beyond chance. The relevance of The Drunkard's Walk is thus clear, and in fact, the book cites hot-hand research (pp. 178-179).

I actually found The Drunkard's Walk to be like two books, the first reviewing the subleties of probability calculation and the second, discussing how people can be mislead by random processes, consistent with the book's title.

The first major part of the book, on probability calculation, matches closely with how I teach probability in my introductory graduate statistics course. I have boiled probability down into one simple question: "How many possible ways can something turn out?" and Mlodinow presents the subject matter in much the same way (Chapter 3).

Then, after shifting to the second major topic, how randomness can mislead, the book finishes strongly, in my opinion. Concepts covered during this latter part of the book include hindsight bias, the arbitrariness of starting and ending points when measuring someone's success, and the distinction between a specific, named person experiencing some type of coincidence or unusual accomplishment and someone, somewhere doing the same (here's an additional example of the latter, beyond what's in the book).

On a somewhat negative note, the central message of The Drunkard's Walk is not exactly new. The book Fooled by Randomness, written several years ago by Nassim Nicholas Taleb and which I reviewed in the Journal of Behavioral Decision Making, plowed similar ground, and many of the ideas in that book had some history.(Parenthetically, Taleb has a follow-up book in the stores, entitled The Black Swan.)

There was only one assertion by Mlodinow that raised questions of accuracy in my mind. Specifically, on page 138, he stated that the "normal" (bell-shaped) curve is "the most widespread manner in which data have been found to be distributed." I am not prepared to say that this statement is flat-out wrong, but it does contradict the evidence of which I'm aware. The textbook I use in my introductory statistics class (King & Minium, 2008, Statistical Reasoning in the Behavioral Sciences, 5th ed.) cites a 1989 Psychological Bulletin article by Theodore Micceri, who:

“…examined the distributions of 440 measures of achievement and psychological traits… Nearly 70% included samples of 1,000 or more. Only 19 of the 440 distributions were found to approximate the normal curve” (p. 101).

Mlodinow has a lively writing style. If, like me, you've enjoyed the recent genre of relatively non-technical books on mathematics and statistics for a general, educated audience, the The Drunkard's Walk is for you; however, if you haven't yet read either of Taleb's books, that might be a better place to start.

No comments: