I’ve been reading The Drunkard’s Walk: How Randomness Rules Our Lives, by Leonard Mlodinow. No, no, don’t click away yet. The ideas in this book are important. For anyone who has ever tried to change herself, or to teach something to somebody else, the most important is on page 9. Its called “regression toward the mean.”
Here’s the short version.
When we do something well its usually a combination of skill and luck. Ditto the reverse. When we perform badly its lack-of-skill and/or lack-of-luck. Examples–SATs, golf shots, investments, job interviews, sales calls, first dates (do people still date?), Yorkshire pudding. Whenever luck is involved there will be variations in performance–we will do better sometimes than others. These variations will tend to cluster around their true value, or average value. If you have a really good day on the golf course, chances are your next day will be worse. Conversely, if you have a really bad day, then the next game you play will most likely be an improvement. In both cases, the scores are tending to move in the direction of your true or average score. This means you can’t really tell very much from one or two observations. Its only by watching and recording outcomes consistently, time after time that you can discern whether your golf game, investments, or cooking is getting worse, getting better or staying about the same.
The problem is that our brains have difficulty holding on to this concept. We are hard-wired to make associations (to find patterns). We naturally tend to discount the role of chance and to see patterns where they don’t actually exist.
Mlodinow tells this story to illustrate. In the 1960′s Daniel Kahneman was teaching behavioral psychology to Israeli flight instructors to help them improve their teaching. A basic principle is that rewards(praise) work better than punishment (scolding). A flight instructor disagreed with this, saying when he yells at a student,they do better next time but when he praises a student that student does worse the next time. Therefore, the flight instructor concluded, scolding was more effective than praise.
What was going on? Chances are that a performance worthy of a good verbal thrashing is way below average and, because of regression toward the mean, it is highly likely that a better performance will follow. The student does better next time but its because of random variation, not because of the the instructor’s tirade. Conversely, that excellent performance that the instructor praised was also partly a matter of luck and, statistically speaking, the most likely outcome next time will be a worse performance but that too had nothing at all to do with the instructor’s praise. (By the way, Kahneman devoted much of his career to studying how humans consistently misinterpret random events. In 2002 he received the Nobel prize in economics along with Tversky for his work.)
Still seem a little “out there?” I’m going to follow up with a few examples in my next post.