Power is not persuasion—it’s the feedback mechanisms that bend outcomes over time.
Previous · Part 5
The Black Swan: How Rare Shocks Reshape the System Itself
Next · Part 7
Incentives Over Intentions: Why Behaviour Follows Rewards
This builds on Part 5: The Black Swan: How Rare Shocks Reshape the System Itself
Continue with Part 7: Incentives Over Intentions: Why Behaviour Follows Rewards
Regression to the Mean
There’s a special kind of disappointment in watching something wildly abnormal settle down. Markets drop. Talent disappears. Systems “mysteriously” level out. And yet the pattern is older than any headline: extremes tend to normalise over time.
What “regression to the mean” really means
Regression to the mean is the tendency for unusually high outcomes to move down later, and unusually low outcomes to move up later—even if nothing fundamental has changed.
The key detail: this can happen when you’re not dealing with true reversals. You might simply be observing a signal contaminated by noise. The extreme value you saw was partly “real” and partly “luck.” In the next round, the luck is less likely to repeat at the same magnitude.
The two engines behind normalisation
1) Noise: the extreme includes “unrepeatable” help
Suppose performance scores reflect a person’s underlying ability plus randomness. When someone scores at the top of the distribution, that top score likely contains more favorable noise than usual. On the next measurement, noise still exists—so the score often declines toward the person’s baseline.
This applies far beyond sports and investing. Promotions, viral attention, outsized profits, sudden failures—many “extremes” are partly sampling effects.
2) Constraints: the world has limits, and limits create rebounds
Even without randomness, real systems have boundaries. Growth can’t be infinite. Losses can’t go below zero for long. Capacity, policy, ecology, and human behavior all impose ceilings and floors.
When you’re near a boundary, you tend to get “bounce-back” dynamics. Extremes aren’t stable because the system architecture doesn’t allow them to persist at full intensity.
Why this matters for power and systems
In the larger logic of systems, regression to the mean is a quiet counterforce to grand explanations.
When we see an extreme, we instinctively reach for narratives: “This rise proves the strategy works.” “This collapse proves the system is broken.” Regression says: hold your certainty. Even if narratives contain truth, the timing can fool you.
The result: overconfidence at both ends
- After extreme success, we often over-credit ourselves and under-credit luck.
- After extreme failure, we often over-blame ourselves and under-credit temporary conditions.
Either way, you’re likely to forecast poorly—because you’re treating a noisy extreme as if it were the mean.
A simple way to “see” it in your own life
Think about moods, discipline, and momentum. When you have a breakthrough week, the next week often feels harder—not because you “lost the plan,” but because breakthrough weeks are partly driven by energy spikes, timing, and favorable conditions.
Same for crises. The most painful month often isn’t the final state; it’s a snapshot pulled by stress, scarcity, and other temporary forces. When conditions change, outcomes move—sometimes back toward baseline, sometimes toward a new one. Regression is the baseline pull.
“Normalisation over time” isn’t calm. It’s structured.
The world doesn’t necessarily smooth out because reality is kind. It smooths because measurement is imperfect, and systems have constraints.
When regression is strongest (and when it isn’t)
It’s weaker when:
- You genuinely changed the system’s underlying parameters (new strategy, new incentives, new technology).
- The extreme is produced by structural shift rather than random fluctuation.
- Measurements are precise and noise is low.
In other words: regression is a property of how you measure and how stable the generator is.
Decision-quality: how to respond to an extreme
When you encounter an unusually high or low outcome, treat it like a clue—not a verdict.
Diagram: why extremes often drift back
Diagram: Random variation leads to Observed outcome; Ceilings/floors & limited capacity leads to Observed outcome; Observed outcome leads to Next observation; Next observation leads to Likely drift toward typical range.
Diagram: Random variation leads to Observed outcome; Ceilings/floors & limited capacity leads to Observed outcome; Observed outcome leads to Next observation; Next observation leads to Likely drift toward typical range.
A note on “normalisation” versus “improvement”
Regression to the mean can look like recovery, but it doesn’t guarantee progress. If you want to distinguish them, you’re looking for evidence that the baseline changed—not just the sample.
Look for:
- consistent improvement across multiple measures
- persistence of gains beyond typical fluctuation
- changes in inputs (process, incentives, skill, structure)
Practical takeaway checklist
What recent “extreme” outcome in your life or work do you interpret too confidently? If it were partly luck, what would you expect to happen next—and what evidence would change your mind?
If this resonates, see how to apply it to your own work with the interactive Dispatch agent.
Be first to like this dispatch
