3.1. The regression fallacy

Someone is guilty of the regression fallacy when they propose an unnecessary

causal explanation for what is, in fact, an instance of regression to

the mean (Gilovich 1991). Kahneman and Tversky (1973) reported on the

classic example of the regression fallacy. Israeli air force instructors had

been urged (presumably on the basis of psychological results) to use positive

(rather than negative) reinforcement in pilot training. However, these

instructors claimed to know that punishment is more effective in training

than praise. The reason? Punishment was typically followed by improved

performance, while praise was typically followed by a worse performance.

However, the instructors failed to control for a regression effect: Wherever

performance tends toward a mean, a poor outcome (one that is deep in

the left tail of a normal distribution) is far more likely to be followed by an

outcome that is better than an outcome that is even worse. And an outstanding

outcome is far more likely to be followed by a worse one. When

regression was controlled for in the Israeli air force example, reward did

mold behavior more effectively than punishment.

There is an important lesson to be drawn for those of us who raise

children. Many parents and coaches believe that rewards inhibit performance

while punishment enhances it. In fact, we can predict that those

who have the most experience with children should believe this falsehood

most strongly. After all, they do have observations to support their view:

After rewarding a very good performance, the next performance really is

likely to be worse; and after punishing a very bad performance, the next

performance really is likely to be better. These falsehoods not only have the

‘‘observations’’ on their side, but they also have ‘‘theory.’’ It is very easy to

come up with plausible narratives to ‘‘explain’’ the observations. Punishment

improves performance because it concentrates the mind; and reward

diminishes performance because it leads to overconfidence and selfcongratulation

and so ultimately to a loss of effort and concentration.

As we have seen, such stories come easy. But the failure of parents to call

into question this hypothesis—to consider the control by asking whether

they know what would have happened to the performance without the

intervention—has undoubtedly led to needless unhappiness for countless

children and regrettable frustration for parents.

Someone is guilty of the regression fallacy when they propose an unnecessary

causal explanation for what is, in fact, an instance of regression to

the mean (Gilovich 1991). Kahneman and Tversky (1973) reported on the

classic example of the regression fallacy. Israeli air force instructors had

been urged (presumably on the basis of psychological results) to use positive

(rather than negative) reinforcement in pilot training. However, these

instructors claimed to know that punishment is more effective in training

than praise. The reason? Punishment was typically followed by improved

performance, while praise was typically followed by a worse performance.

However, the instructors failed to control for a regression effect: Wherever

performance tends toward a mean, a poor outcome (one that is deep in

the left tail of a normal distribution) is far more likely to be followed by an

outcome that is better than an outcome that is even worse. And an outstanding

outcome is far more likely to be followed by a worse one. When

regression was controlled for in the Israeli air force example, reward did

mold behavior more effectively than punishment.

There is an important lesson to be drawn for those of us who raise

children. Many parents and coaches believe that rewards inhibit performance

while punishment enhances it. In fact, we can predict that those

who have the most experience with children should believe this falsehood

most strongly. After all, they do have observations to support their view:

After rewarding a very good performance, the next performance really is

likely to be worse; and after punishing a very bad performance, the next

performance really is likely to be better. These falsehoods not only have the

‘‘observations’’ on their side, but they also have ‘‘theory.’’ It is very easy to

come up with plausible narratives to ‘‘explain’’ the observations. Punishment

improves performance because it concentrates the mind; and reward

diminishes performance because it leads to overconfidence and selfcongratulation

and so ultimately to a loss of effort and concentration.

As we have seen, such stories come easy. But the failure of parents to call

into question this hypothesis—to consider the control by asking whether

they know what would have happened to the performance without the

intervention—has undoubtedly led to needless unhappiness for countless

children and regrettable frustration for parents.