Why Praise Seems to Backfire and Scolding Seems to Work

A veteran flight instructor once told Daniel Kahneman that his praise made cadets worse and his yelling made them better. Kahneman recounts in Thinking, Fast and Slow (2011) that this objection, which at first seemed to completely undermine what he was teaching, gave him the most satisfying experience of discovery of his career.

The flight instructor’s objection

Kahneman was training instructors from the Israeli Air Force on the psychology of effective training. His message was the standard one in behavioral research at the time: rewards for improvement tend to be more effective than punishments for mistakes, an idea backed by decades of studies with pigeons, rats and people.

At the end of his talk, one of the most veteran instructors raised his hand with a specific objection, based on his own experience:

• When he praised a cadet for executing an aerobatic maneuver well, on the next attempt the cadet usually did worse.
• When, by contrast, he yelled at a cadet over the headset after a poor execution, on the next attempt the cadet usually improved.

For the instructor, the conclusion was obvious: the theory was wrong, punishment worked, and praise didn’t.

Kahneman describes that moment as a revelation: the instructor was entirely right about what he’d observed, and at the same time completely wrong about the explanation he gave for it.

What was really happening: regression to the mean

What the instructor had detected, without knowing it, was a statistical phenomenon called regression to the mean.

When a cadet’s performance on a maneuver depends partly on skill and partly on chance (a momentary streak of good or bad luck), an especially good result tends to be followed, on the next attempt, by a somewhat worse one, and an especially bad result tends to be followed by a somewhat better one.

Not because there’s some compensating force at work, but simply because the extreme, by definition, is unlikely to happen twice in a row.

The instructor only praised a cadet when their execution had been better than usual, so it was statistically expected that the next attempt would be somewhat worse, entirely regardless of the praise.

And he only yelled when the execution had been especially bad, so it was expected that the next attempt would improve, also regardless of the scolding.

The instructor had confused a random fluctuation with a cause-and-effect relationship.

The improvised demonstration with two coins

Aware that a purely statistical explanation wouldn’t convince the room, Kahneman turned to an on-the-spot physical demonstration: he drew a target on the floor with chalk and asked each officer present to toss two coins in a row at it, with their back to the target and without looking.

The pattern that showed up on the board was exactly the same one the instructor had described with the aerobatic maneuvers:

• Most of those who landed closer to the center on the first toss did worse on the second.
• Most of those who missed by more on the first toss improved on the second.

And in this case it was obvious that neither praise nor scolding had anything to do with it, because no one was praising or yelling at anyone between one toss and the next: the only possible explanation was regression to the mean.

A phenomenon with more than a century of history

Kahneman didn’t discover regression to the mean. Francis Galton described it in 1886, comparing the heights of parents and children: the children of very tall parents were tall, but on average a little less so than their parents, and the children of very short parents were short, but a little less so. Galton called it “regression towards mediocrity.” Today it’s understood as a statistical consequence of any imperfect correlation between two measures, with no need for any force pushing toward the center.

What Kahneman and Amos Tversky added was the psychological part. In their 1973 paper on the psychology of prediction they showed that people’s intuitions ignore regression when they make predictions, and in their famous 1974 paper on heuristics and biases they already used the case of the flight instructors: when regression shows up, people invent a cause for it. That’s the instructor’s error: the statistical phenomenon belongs to Galton; the mistaken causal explanation is universal.

A pattern that also shows up in golf

Kahneman illustrates the same principle with a different example: a high-level golf tournament where the average first-day score is 72 strokes.

A player who shoots 66 on the first day (far better than average) probably has real talent, but it’s also reasonable to think that on that day they had, on top of that, a better-than-usual run of luck.

The same applies, in reverse, to a player who shoots 77.

The most reasonable prediction for the second day isn’t that both will repeat their first-day result exactly, but that both will move somewhat closer to the average: the one who did very well will probably keep doing well, but a little worse than the first time, because their run of luck is unlikely to repeat with the same intensity; and the one who did poorly will probably improve, not because they’ve “learned their lesson,” but because their bad run isn’t likely to continue either.

Why this matters beyond sports or flying

Kahneman draws an uncomfortable conclusion from this episode about how feedback works in everyday life: since we tend to praise others when their behavior pleases us and reprimand them when it doesn’t, and since other people’s performance naturally fluctuates around its true average, we end up, in his own words:

statistically punished for being nice and rewarded for being nasty.

Not because scolding genuinely works better than praise, but because the very pattern of when we praise and when we reprimand means regression to the mean almost always works in the scolder’s favor.

This same kind of error, attributing a cause to what is actually a random fluctuation, often shows up in how the performance of a team or a person at work gets interpreted. That’s relevant to any leadership style that relies on positive or negative reinforcement as a management tool, and to anyone with a more analytical profile when interpreting one-off swings in their own or someone else’s performance.

If you’d like to identify your own leadership pattern in situations like these, the leadership style quiz sums it up in a few minutes.

Sources

• Galton, F. (1886). Regression towards mediocrity in hereditary stature. The Journal of the Anthropological Institute of Great Britain and Ireland, 15, 246-263. https://doi.org/10.2307/2841583
• Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux.
• Kahneman, D. and Tversky, A. (1973). On the psychology of prediction. Psychological Review, 80(4), 237-251. https://doi.org/10.1037/h0034747
• Tversky, A. and Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131. https://doi.org/10.1126/science.185.4157.1124