Following the schedule of our Reading Calendar
This chapter comes closely tied to the previous one. In chapter 17 we were told about the phenomenon of regression to the mean and how easily we ignore it in our intuitive processes and how difficult it is to grasp it when we reflect about it. Kahneman goes a little too far away, for my taste, talking about it as a kind of obscure, incomprehensible phenomenon that eludes even the greatest of the minds. At least I would have liked more examples and elaboration on that.
Now we are shown here how not taking into account regressions to the mean results in biased predictions. The chapter begins with an overall introduction to the prediction business and makes the distinction between analytic prediction (system 2) and intuitive ones (system 1). And shows how inside the intuitive wagon there are also those predictions that are based in system 1 performing quickly and effectively tasks that system 2 has taught him (expert intuitions) and other predictions were system 1 does as he pleases (everyday predictions). At that point of the chapter, that is page 1, I was very excited about the prospect of a global analysis of the prediction world and I have to admit that I was a little disappointed by the chapter centering only into the regression issue.
The author gives a procedure to correct our intuitive predictions taking into account the regression to the mean. As in the previous chapter everything gets too long and complicated. He simply means: “Calculate the amount of luck that happened in the past and consider that in the future such luck will not be there”. And that’s it.
And I have being growing uneasy as I was reading the chapter for several reasons, all of them intuitive as always, because these are deep questions that need a lot of reflection.
First, his procedure to correct predictions is a kind of blending the colors of past observations and generating a gray vision of the future. On average closer to reality but dull like death. There is no room in this system for extreme predictions. This cannot be good. He addresses this critique later in the chapter.
Second, he also comments on that, accuracy in the prediction is not always the main goal. Statisticians laugh at people who play lottery because the expected value of it is negative. Playing lottery is very rational because the harm of losing 1 dollar, in practical terms, equals zero and so no matter how low the probabilities of gaining 5 million may be, and what the expected value may say, the rational thing is to play lottery if you have fun expending the time to do it.
Third, there are anchoring effects. Some of them hidden. Past grades may be, lets imagine, 100% result of luck. But if future grades are decided by teachers aware of past grades they may create a spurious causality there were there was none. And if you are a puritan statistician you may end up over correcting and yielding predictions far worse than those of the average Joe. The world is populated by “regression to the mean non correctors”. So, since the world you are living in is not only the result of the laws of physics but also the acts of people, sometimes it may happen that your predictions about the future may be wrong because it is influenced by people who is not correcting anything.
I cannot provide a rational elaboration of point 3. I simply smell “rational fool”.