Probability distributions differ in how much influence extreme outcomes have.
Some variables stay within narrow limits. Others can produce outcomes far larger than anything previously observed.
The difference matters because a rare outcome may be a minor exception, or it may determine the total result.
This is Nassim Nicholas Taleb’s distinction between Mediocristan and Extremistan.
The Power of Mediocristan and Extremistan
In Mediocristan, no single observation changes the total very much.
Height gives the intuition. If two people have a combined height of 4.1 metres, the likely explanation is two unusually tall people. It is not one ordinary person and one person four metres tall. Human height has practical limits, so extreme individuals do not overwhelm the whole group.
In Extremistan, one observation can dominate the total.
If two people have a combined wealth of $36 million, the likely explanation may be one person with almost all of it and another with very little. A single person can change the total more than thousands of ordinary people.
In statistics, these are roughly the difference between light-tailed and heavy-tailed distributions. A light-tailed distribution makes extreme values rapidly less likely. A heavy-tailed distribution gives extreme values enough probability to matter.
Why This Matters
In Mediocristan, more observations usually make the average more reliable. The ordinary cases tell us something useful about the whole.
In Extremistan, ordinary cases may tell us very little. The most important event may not have happened yet. When it does, it can change the average, the total, and our explanation of what happened.
Taleb calls this the preasymptotic problem. Statistical laws often describe what happens with an enormous or infinite number of observations. Real life gives us finite samples. In a heavy-tailed world, the average may settle very slowly, and the next observation may matter more than all the previous ones.
Heavy-tailed does not mean that extreme events happen constantly. It means that a model treating them as negligible gives us the wrong picture of reality.
The Practical Principle
Before trusting an average or a pattern, ask:
- What are the limits? Can one observation be vastly larger than the others?
- What drives the total? Many ordinary events, or a few exceptional ones?
- Has the important event happened yet? A calm sample may simply be incomplete.
- Which world are you in? Mediocristan or Extremistan?
- What would change your conclusion? Identify the observation that could rewrite the story.
The fundamental insight is that the average is useful only when the world is not being governed by the extremes.
References
- Taleb, N. N. (2020). Statistical Consequences of Fat Tails: Real World Preasymptotics, Epistemology, and Applications. STEM Academic Press. Chapter 3.
- National Institute of Standards and Technology. Histogram Interpretation: Symmetric, Non-Normal, Long Tailed. NIST/SEMATECH e-Handbook of Statistical Methods.