Kahneman describes how, while working with the Israeli Air Force, one of the instructors emphasized punishment over reward. The instructor stated that when he praised flight cadets for a good maneuver, they usually did worse the next time. Screaming into a cadet’s ear for bad execution, however, generally led to better performance.
The instructor’s assumptions not only misattribute causality (as Kahneman goes on to describe), but the instructor also places too much confidence and weight into his own actions as a reason why the cadets do better after being yelled at.
Kahneman notes that the instructor was right—but also very wrong! The instructor was inappropriately attaching causality between his actions and the cadets’ performances, ignoring the fact that a particularly good execution of a certain maneuver will likely be followed by a less well-executed maneuver, and vice-versa with a particularly bad execution.
The instructor’s perspective provides another aspect of our predilection for stories. We prefer to think that all events have causal explanations, despite the fact that some things simply occur due to randomness.
Kahneman writes that success = talent + luck. Kahneman explores this principle in looking at a golf tournament. A golfer who scores above average on day 1 can be assumed to be both more talented and luckier; a golfer who scores below average on day 1 is both less talented and less lucky. On the second day, the golfer who did well on day 1 is likely to be successful, but less successful than on the first day because his luck will likely not hold. The golfer who did poorly will probably be below average on day 2, but will improve because his streak of bad luck is unlikely to continue.
We often place more emphasis on talent than on luck in determining what makes someone successful. On any day, if a person does particularly well one can assume that their success was due at least in part to luck—but we have a very difficult time understanding this in practice and like to believe that their talent is the true cause of good performance.
This pattern is called regression to the mean. The more extreme the original score, the more regression we expect, because an extremely good score suggests a very lucky day. The same effect can be observed by looking at day 2 and then day 1, which should help convince people that regression does not have a causal explanation.
Regression to the mean is a difficult concept for many people to understand because regression has an explanation, but not necessarily a cause—and our System 1 processing is designed to look for causes and to make coherent sense of the world.
Regression effects are everywhere, and people often misattribute causes to explain them. Kahneman points to analysis of the Olympic ski jump, in which athletes jump twice. If athletes have a good first jump, commentators say they will have a worse second jump because they will feel pressure; if athletes have a bad first jump, commentators say that they have nothing to lose and will have a better second jump. The analyst has detected a principle of luck and chance and has assigned a causal story to it.
The Olympic ski jump provides another example of how regression to the mean works, and the commentator’s analysis also demonstrates some of the ways in which we will create explanations to provide a sense that there is a causal explanation for the athletes’ performances.
Kahneman goes on to discuss how regression can be measured between variables on different scales, using a “correlation coefficient.” The correlation coefficient is a measure of the relative weight of the factors they share. For example, the correlation between height and weight among adult American men is .41, meaning they share some factors. The correlation between family income and the last four digits of their phone number is 0, meaning these two qualities are unrelated.
By establishing correlation as related to regression, Kahneman can elaborate on how regression does not necessarily have a cause. Even though height and weight are correlated, one’s height does not cause one’s weight—they simply share factors.
Correlation and regression are different perspectives on the same concept. Whenever the correlation between two scores is imperfect, regression can be found, as in this example: highly intelligent women tend to marry men who are less intelligent than they are. People will readily explain this statement in terms of causality. But when faced with this statement, “the correlation between the intelligence scores of spouses is less than perfect,” people do not bat an eye, even though it means the same thing. It is a mathematical inevitability that highly intelligent women will be married to husbands who are on average less intelligent than they are.
This example, when stated both in terms of regression (the first sentence) and in terms of correlation (the second sentence) explains yet again how we like to look for causality to create coherent stories. People have a hard time understanding probability that does not have an explanation, and so they try to invent explanations as to why intelligent women might intentionally marry less intelligent men.
Kahneman makes up a headline: “Depressed children treated with an energy drink improve significantly over a three month period.” He writes that the fact that it reports is true, but it would be true without the energy drink as well. Depressed children are an extreme group, and they regress to the mean over time. This is why control groups are so vital in experiments.
Kahneman explains why it is important to understand that causation and correlation are not the same thing. It is important that the regression effect is not the only thing that accounts for certain outcomes in testing of medicine and other research scenarios.
In a final example, Kahneman adapts a question from Max Bazerman’s Judgment in Managerial Decision Making. The given circumstances are as follows: you are a sales forecaster for a department store chain. All stores are similar in size and merchandise, but their sales differ due to location, competition, and random factors. Overall sales are expected to increase by 10% across the board. It then asks the reader to complete a table, predicting how each store will do in the coming year. It is tempting simply to add 10% to each store’s sales, but one must also adjust for regression and add slightly more to the underperforming stores, and slightly less for the overperforming stores.
Lastly, Kahneman provides an example that might be more relevant outside the fields of psychology and research. Understanding the effects of regression can allow people to more accurately make predictions about the future, as is the case with this scenario in which the reader can attempt to make projections about the performance of different department stores.