Regression to the Mean Neglect

A simple explanation

Regression to the mean neglect is the chronic failure of intuition to expect that extreme results — the very good day, the very bad week, the unusually high score, the catastrophically low one — will tend to be followed by results closer to the underlying average. The Threat System treats each outlier as informative about the state of the situation, and updates its model of how things are now accordingly, when most of the time the outlier was simply random variation around a stable mean.

The bias is not the response to outliers, which can be appropriate. The bias is the durability of the response: treating a single extreme result as a new baseline rather than as a point on a distribution that will, statistically, drift back toward the centre next time.

An everyday example

A coach watches a player have an exceptional game and praises them publicly. The next game, the player performs closer to their season average — worse than the exceptional game. The coach feels privately that praise made the player complacent. Later in the season, a different player has a poor game and the coach delivers a sharp correction. The next game, that player performs closer to their season average — better than the poor game. The coach concludes that criticism works and praise does not.

The coach is reading a statistical pattern as a causal one. Both players regressed toward their own mean. The coach's interventions did nothing. But the felt experience is so clean — I praised them and they got worse, I criticised them and they got better — that the conclusion installs itself with the full force of evidence, and the coach's behaviour around feedback adjusts in a direction the data does not actually support.

Why do extreme good or bad results never quite repeat?

Because most performance is a mix of stable underlying ability and random variation. The very best result you produced in a domain was likely a happy combination of skill and a tailwind of small lucky factors that lined up. The next result will sample skill again but a different draw of luck. Statistically, the second result will be closer to your average than the first. The same is true of the very worst result.

The Threat System, however, was not designed to think in terms of distributions. It was designed to detect signal in a present that matters now. When an extreme result arrives, the System flags it as informative — and the conscious mind, taking the flag seriously, builds a story around it. The story is almost always wrong about what the next data point will look like.

The behavioral loop

A loop that builds stories on noise:

Outlier event — a result far from your usual average arrives.
Salience spike — the System flags the result as informative and routes attention toward explaining it.
Causal attribution — a story is constructed about why the extreme result happened — talent, effort, decline, breakthrough, breakdown.
Action update — behaviour adjusts based on the story: more of what produced the high, less of what produced the low, or vice versa.
Regression arrives — the next result is closer to the underlying mean, statistically.
Confirmation read — the change is attributed to the action update rather than to regression.
Story hardening — the causal account installs as durable belief, and similar outliers in the future are processed through the same frame.
Calibration drift — over years, beliefs about what causes what become misaligned from what actually causes what, with regression doing structural work the mind has never noticed.

Emotional drivers

Four feelings, often in stack:

An ambient hunger for causal explanation that the System rewards as understanding.
A pleasure in the felt clarity of a clean cause-effect story, which makes statistical explanations feel unsatisfying.
An identity satisfaction in being able to attribute outcomes to deliberate action, often quite strong.
A faint shame around extreme low outcomes that the bias converts into stable narratives about the self.

What your nervous system does

Extreme results produce strong autonomic activation — joy and pride on the high end, shame and disappointment on the low end. The body files the extremes with high emotional charge, which makes them disproportionately retrievable later. Regression-toward-the-mean results, sitting closer to the average, produce muted autonomic responses and are filed with less charge.

This asymmetric encoding means that the data the mind retrieves easily — the extremes — is exactly the data least representative of the underlying process, while the data that would correct for the bias — the unremarkable regressed results — is harder to bring back. The body's memory system is structured to mislead the conscious mind about distributions.

The DojoWell interpretation

Regression to the mean neglect is a clear case of a Threat System heuristic that maps poorly onto statistical reality. The original ask — detect when the underlying situation has actually changed — is honest and important. The substitute — treat each extreme observation as evidence of a changed underlying situation — is what produces the cost.

The deposit register shows occasional wins: sometimes the extreme result really does reflect a real change, and the responsive System catches it early. The residue register shows the larger pattern: praise withheld because the praised performance regressed and looked like complacency; criticism doubled because criticism appeared to work when it was statistics that worked; medical interventions credited with effects that were really regression; investment strategies adopted on streaks that would have ended without intervention.

The density signature is false_progress because the stories built on regression feel like understanding. The System counts each completed causal explanation as evidence of mastery, while the residue accumulates in the misattributed actions, the punished outliers, and the self-narrative that swings with the variance of a noisy series. The equation looks lucid from inside the loop. The cost is the stability the actual underlying process would have offered if you had stopped treating its noise as signal.

How do I tell a real trend from a lucky streak?

You require the trend to demonstrate itself across enough data points that the variance would have averaged out. The System wants to commit to a story on a single observation; the work is to make it earn the story.

Three moves:

Wait for the third data point. A single extreme result is uninformative about trend. Two consecutive results in the same direction are weak evidence. Three or more begin to suggest the mean itself has shifted.
Track your historical variance. Knowing roughly how much your results swing under stable conditions makes it possible to ask whether a fresh result is really outside that envelope or just at its edge.
Suspect dramatic causal stories about single events. The more vivid and clean the story is, the more likely it is being told to explain a regression that was going to happen anyway.

Practical steps

Keep a baseline log. A record of your typical results across several domains. The log makes extreme observations visible as departures rather than as new baselines.
Delay action on outliers. Resist the urge to change strategy after a single extreme result, good or bad. Let the next two or three observations contribute.
Be careful with feedback timing. When praising or correcting based on performance, recognise that the next observation is likely to regress regardless of the feedback. The feedback is for the long run, not the next data point.
Audit one strong recent causal story. Find a moment when you concluded X caused Y based on one extreme event. Ask whether regression is sufficient to explain the pattern.
Tell smaller stories. Replace I have become a different person with I had a good month. Replace the team is collapsing with they had a bad week. The smaller story is usually closer to the truth.

Reflection questions

Which domain of your life have you built the largest causal story on the smallest sample of extreme results?
Where has regression to the mean been quietly doing the work you attributed to your intervention?
What recent feedback decisions did you make based on one outlier rather than on a trend?
How would your self-narrative feel if you allowed your worst and best months to be statistical events rather than verdicts about who you are?

Frequently Asked Questions

Why is regression to the mean so hard to see intuitively?

Because intuition is built around causes and stories, while regression is a property of distributions. The mind looks for the why of any observation, and finds one — usually the most salient action that preceded the result. Statistical regression has no agent and no narrative, so it is difficult for the system to credit. Kahneman called this one of the most important blind spots in intuitive judgement.

How does this affect feedback at work?

Significantly. Managers who reward extreme good performance often see regression and conclude that reward causes complacency. Managers who criticise extreme poor performance often see regression and conclude that criticism works. Both conclusions are wrong on the data and lead to feedback strategies miscalibrated against what actually moves performance. The fix is to evaluate feedback effects across enough observations that variance averages out.

Is this why fad treatments and interventions look effective at first?

Often, yes. People typically try new interventions when their condition is at an extreme — pain is high, performance is low, mood is bad. From an extreme, the next observation tends to regress toward the mean regardless of what was done. The intervention gets credit for a change that was statistically going to happen. This is a major driver of the appearance of effectiveness in untested treatments.

How is this different from the gambler's fallacy?

Gambler's fallacy is the false belief that a string of one outcome makes the opposite outcome more likely on the next trial in an independent process — red is due after five blacks. Regression to the mean neglect is the failure to expect extreme observations to drift back toward the mean in series that have an underlying central tendency. They are related family members but mechanically distinct.

How does this connect to Meaning Density?

Regression neglect is a false_progress signature on the threat-interpretation register. Each clean causal story built on an outlier feels like understanding, which deposits a real sense of mastery. The residue accumulates in the misattributed interventions, the feedback strategies built on noise, and the self-narrative that swings with the variance of a noisy life. The equation runs in the black on storytelling and in the red on accuracy, and the second register is where the larger truths about what is actually happening get lost.