Regression to the Mean

Numbers bounce back. After an unusually brilliant game, terrible pain day, or record sales week, the next result is often closer to ordinary. Part of the extreme was the stable thing you care about: skill, health, demand. Part was luck, a bad day, a lucky week, or measurement noise. The noise is unlikely to return at the same size, so the next reading moves inward. It can still be good or bad. It is just less extreme.

This is regression to the mean psychology in practice: the story you insert between those two readings. Someone is praised after a brilliant landing and the next one is more ordinary, so praise looks like it spoiled them. Someone is scolded, treated, coached, or apologized to after a terrible result and the next one is less terrible, so the intervention gets the credit. The rebound was already the likely next chapter.

The clearest regression to the mean examples appear wherever a measure wobbles: test scores, pain, revenue, sports, mood, crime rates, customer complaints. Very tall parents tend to have children who are still tall and less extremely tall. A shop's best month is a weak promise that the next month will match it. The more room noise had to push the first number out, the more room the second number has to fall back.

Practice, medicine, coaching, and apologies can all be real causes. The bug is reading one rebound after one extreme as the receipt. If you only intervene when things are unusually bad, improvement will often follow you around even when you did not cause it. If you only reward what is unusually good, a drop will often follow the reward.

• "I praised the great landing and the next one was worse. Praise backfires."
• "I yelled after the worst game and they improved. Yelling works."
• "The new supplement fixed it. I started it at the worst week, and then I improved."
• "The fund's best year was followed by a dull one. The strategy died."
• "Employee of the month got lazy. Look at the ordinary month after."
• "We changed the software after the outage from hell, and the next month was calmer. The tool saved us."
• "The apology fixed the friendship. Things were at their worst, and then they weren't."

The flight line: instructors praise an unusually smooth landing and criticize a rough one. The next landing is typically less smooth after the praise and less rough after the criticism. A teaching method gets the credit for a bounce that repeated practice would often have produced alone.

The dashboard: a team changes process in the month after a record failure. The following month is merely bad. The postmortem celebrates the process. Nobody asks what the month after a record failure usually looks like.

The slump: a strong performer has a terrible fortnight, gets a stern review, and returns to their usual level. The review is filed as the cure. The fortnight is never treated as noise around a stable level.

The remedy: a cold, a pain flare, or a stretch of insomnia is treated at the peak. It eases. The easing is real. The receipt is not, until you know how often that peak eases when you wait, or when you do something else.

Punishment gets a better reputation than it deserves, and reward gets a worse one. You are most tempted to punish at the bottom and to praise at the top, which is exactly where regression will flatter the punishment and embarrass the praise. People and teams then get more harshness and less recognition for reasons that feel empirical.

Programs, drugs, coaches, and tools get adopted or killed on a single bounce. Money follows the story. So do careers: someone is branded complacent after a normal month that followed a brilliant one, or branded saved after a normal month that followed a disaster.

The quieter cost is that you stop learning what actually works. If every extreme is "explained" by the thing you did next, the comparison that would teach you never gets built.

Extremes demand a story. A number that far from usual feels like it must have a cause you can name, and the thing you did in the gap is the cause sitting closest to hand. The mind also expects the next result to be as extreme as the one that impressed it. A milder sequel then looks like a change in the world, not like noise going home.

Intervention is aimed at extremes on purpose. You do not overhaul a process in an average month, and you do not start a remedy on an average day. That habit guarantees you will often be standing next to a rebound, ready to take credit or blame.

Galton's 1886 paper on hereditary stature documented the statistical shape and gave it the name that stuck: regression towards mediocrity. Children of very tall parents tended to be tall, and closer to the population's ordinary height than their parents were. The same inward pull showed up from the short side. He was describing a measurement pattern, not a moral slide toward being average as a person.

Tversky and Kahneman's 1974 Science paper used that pattern as a warning about causal stories. Experienced flight instructors had noticed that praise after an exceptionally smooth landing was typically followed by a poorer next try, and harsh criticism after a rough landing was typically followed by a better one. They concluded that verbal reward hurts learning and verbal punishment helps it. The paper's point is that the conclusion does not follow. After an extreme performance, a less extreme one is the usual sequel even if the instructor says nothing. Because praise is handed out at the top and criticism at the bottom, regression alone will often make punishment look effective and reward look harmful.

• You changed something only because the last result was extreme.
• The "effect" is a move back toward this person or this team's usual level.
• You have one before-and-after, and no similar case where you did nothing.
• Praise is on trial because a peak was followed by a normal day.
• A remedy started at the worst moment is getting credit for the improvement that followed.

Assume an extreme number is a mix of signal and noise until a comparison says otherwise.

• Before you explain the sequel, ask what usually comes after a result this far out, even when nobody intervenes.
• Compare the person with their own typical level, not only with the spike or the crash.
• If you must act at the extreme, decide in advance what would count as more than a bounce: a change that holds, or a change larger than the rebound in cases you did not touch.
• Do not rewrite a reward policy from one drop after one prize, or a punishment policy from one rise after one scolding.
• When a cold, a chart, or a slump improves after a new fix, write down the other reasons a peak eases. Then see if the fix still has a job.

Does this mean feedback and treatment do nothing?

No. It means one rebound is a weak receipt. Feedback, practice, and treatment can move the usual level. You see that move by comparing, by looking across many cases, or by checking whether the change is still there after the bounce would have faded. A single extreme-then-ordinary pair cannot carry the proof.

Is this the gambler's fallacy?

No. The gambler's fallacy expects a coin or a roulette wheel to "balance" after a streak, as if chance owed you the opposite. A coin has no usual skill to return to. Regression is about a noisy measure drifting back toward a person, a team, or a population's ordinary level. You can make one mistake at a casino and the other on a sales chart.

Against regression to the mean, the rebound is the suspect, not the verdict.

Praise backfires

The miracle month

The spoiled star

Practice this pattern in the Reframing App - capture the trigger, label it (like Regression to the Mean), check evidence, and write a more balanced thought.