What the Mean Measures

"Virtue is the mean between two vices — not the arithmetic mean, but the proportional one." — Aristotle, Nicomachean Ethics, II, 6.

The word "mean" is confused in two senses at once. For Aristotle the mean is the right measure, found by weighing excess and defect in the particular case. For Gauss the mean is the arithmetic operation: add and divide. The two means sound similar and are completely different. Corporate analytics for the last thirty years has systematically conflated them and treated them as one.

The Gaussian mean is a description of a distribution. It prescribes nothing. If the average height in a country is 170 cm, it does not follow that 170 is the right height. The Aristotelian mean is a prescription: the right measure a reasonable person chooses for a particular case. The first mean obliges no one; the second obliges only if we have understood the context. Without context they do not meet.

The tyranny of the average

Modern KPIs operate by Gaussian logic but are interpreted by Aristotelian. The metric "average response time — 12 minutes" automatically becomes the norm: anything slower is bad, anything faster is good. The mean says nothing about what should be. It says what was. Between those two verbs lies the whole gap between description and norm.

The substitution is especially dangerous in managing people. The KPI "average productivity of the team" creates pressure toward the mean: those above are not rewarded for the gap; those below are punished. A year later the team converges to the mean — but yesterday's mean, with no relation to what should hold today. That is the tyranny of the average: a norm derived from a distribution and confused with a norm derived from reflection.

Why A/B tests do not answer the main question

An A/B test is a tool for comparing two means. It answers: what worked better on average across the sample in period X? It does not answer: what ought to work? These two questions are of different kinds. The first is descriptive, the second is normative.

In the classic example: an A/B test showed the red button brought 4% more clicks. The decision — install the red one. But the red button drove loyal users away after six months — invisible in the test because the test measured the average over two weeks. A mind trained only on the Gaussian mean is systematically blind to what lies outside its frame.

The Aristotelian mean is not arithmetic

Aristotle warned directly: the ethical mean is not the arithmetic mean. If for one man the right portion is two loaves and for another (the wrestler Milo) it is ten, then "the average six" is wrong for both. The right measure depends on context, organism, purpose. Arithmetic does not find it.

This does not mean numbers are useless. It means the number is the input to reflection, not its output. A strategist who looks at a dashboard and takes a decision on the basis of the mean makes the same mistake as the man who ate six loaves because of "the average."

The mean is a window into the distribution, not a verdict. A good analyst shows variance, median, tails — and only then the mean. A bad analyst shows the mean alone and considers it a complete report.

Where the conflation is allowed

One context in which the Gaussian mean really carries normative weight is quality control in mass production. If a part must be 10 mm ± 0.1, then the lot mean of 10 mm is both norm and description at once. But that is the rare case: the mean works as a norm only when we already set a range, and the mean merely confirms that we hit it. Without a range, the mean is not a norm.

Tails as strategy

In real economies almost all important phenomena live in the tails of the distribution, not at the centre. A venture portfolio: one or two investments return the whole fund, the rest near zero or below. The book market: 1% of titles drive 90% of sales. Technology products: a pair of viral features determines everything, the rest is background. The mean in these cases is not merely useless — it misleads, because there is no "average" deal that even remotely describes how the game works.

Taleb insists: the difference between Mediocristan (the world where the mean works) and Extremistan (the world of tails) is fundamental. In Mediocristan, strategy means improving the average: process optimisation, margin economy. In Extremistan, strategy means maximising odds of hitting the tail: experiments, optionality, readiness for large losses for the sake of large wins. Whoever applies the strategy of one world in the other systematically loses.

What to do

When you see a mean in a report, ask three questions. What is the variance? (If high, the mean is uninformative.) What is the median? (If far from the mean, the distribution is skewed.) What is the tail? (Where the anomalies live — often where the business actually is.) If a dashboard shows only means, it systematically misleads. Aristotle would have said: the mean cannot be found by arithmetic; it is found by reflection, and arithmetic is only an aid.