Thinking, Fast and Slow

Daniel Kahneman's book Thinking, Fast and Slow has a great ending, a fine middle, and a weak beginning. The "Conclusions" chapter summarizes his key points quite nicely in terms of "two fictitious characters, two species, and two selves":

Kahneman's flaws? The book is slow and often feels padded. The evidence cited from psychological experiments is often weak, possibly wrong. Statistics are skimpy. And every chapter ends with silly little make-believe quotes. Is that a deliberate device to deceive the System 1 anecdote-prone part of the reader's brain? If so, it's a disappointing trick that doesn't work well and reduces credibility.

Happier tidbits from the book? There are many. One of the best is in Chapter 17 ("Regression to the Mean"):

There's also the excellent rule for correcting one's intuitive predictions in Chapter 18 ("Taming Intuitive Predictions"). To forecast something in a specific case — an individual person's height, a particular company's growth next year, the score on day 2 of a golf tournament by the leader on day 1, whatever — follow these four steps:

  1. begin with an estimate of the baseline — the general average over the appropriate population of the quantity to be predicted, without taking into account any information about the individual involved
  2. estimate the intuitive best answer for the specific case based on additional known information for that individual
  3. estimate the correlation between the known information about the specific case and the quantity to be predicted — a fraction ranging between 0 (perfectly uncorrelated) to 1 (perfectly correlated)
  4. starting with the baseline value (1), move toward the intuitive answer (2) — but only by the fraction given by the correlation (3)

And there's the solid, valuable argument in Chapter 21 ("Intuitions vs. Formulas") in favor of ridiculously simple rules of thumb. Based on Robyn Dawes's article, "The Robust Beauty of Improper Linear Models in Decision Making", Kahneman presents the formula for predicting marital stability:

frequency of lovemaking minus frequency of quarrels

Positive numbers mean a healthy marriage. A similar rough formula gives the Apgar Score to assess newborn baby health. Expert human judgment, based on intuitive appraisal of a multitude of factors, tends to be much worse than the results of simple additive formulas, as Philip Tetlock's Expert Political Judgment showed in other realms.

Thinking, Fast and Slow is a best-seller (talent + luck) that appeals (System 1) but also persuades (System 2). Good reading.

(cf Predictably Irrational (2011-02-22), ...) - ^z - 2013-10-24