Per Capita

Two bags of apples sit on a table. The first holds a thousand apples, twenty of them poisoned. The second holds ten apples, two of them poisoned. One apple has to be eaten, and the eater picks the bag it comes from.

The pull is toward the second bag, because two is a smaller number than twenty. The second bag poisons one eater in five and the first poisons one in fifty, so the choice is ten times worse than the alternative and it feels safer the whole time.

That is the entire concept. Per capita means per head — the count divided by the population it came from — and it exists because a raw total tells you how much of something there is while a rate tells you what it means for anyone inside it. The twenty poisoned apples are real and they are not the number governing the decision.

The same substitution runs through everything measured across groups of different sizes. A country with larger total output is not thereby a country whose citizens are better off; that question is answered by output per person, and the two rankings come apart.

Where it fails in practice

The characteristic error is comparing raw counts drawn from populations of very different sizes and treating the larger count as the larger problem. A group four times the size of another will generate more of nearly everything, including things nobody wants, and saying so establishes only that it is larger. The comparison people actually intend — is this more common here than there — requires the division, and the division frequently reverses the answer the raw counts suggested.

When somebody declines to do it, there are three explanations, and only one of them is respectable.

One: the abstraction is one layer too many

Per capita requires holding two numbers at once, dividing one by the other, and interpreting a result that has no direct referent in the world. That is one layer of abstraction more than reading a total, and one layer is enough to lose people. This is not a moral failing and it is worth ruling out first, because the two explanations that follow are only visible once it has been eliminated.

Two: the total supports the preferred conclusion

Someone who can perform the division declines to, because the divided figure undercuts the point being made. This subdivides. In the deliberate case the person knows the rate reverses the claim and presents the total anyway. In the motivated case the raw number confirms something already believed, and no impulse to check arises — which is a real cognitive failure and a milder one, since the machinery was available and simply never engaged.

The test that separates this from the first case is whether the same person uses rates when rates favour them. Someone who cites per capita where it helps and calls it misleading where it does not is not confused. Confusion is indifferent to which side it lands on, and this is not.

Three: the rate leaves other variables uncontrolled

This is the only legitimate objection, and it is legitimate because it is true: a rate tells you how common something is and says nothing about why. Comparing crime rates without accounting for poverty genuinely does leave the interesting question unanswered, and someone raising that is raising something real.

It has a counterfeit, and the two are easy to tell apart. The honest version says a further variable is needed to understand the pattern. The counterfeit says the rate is worthless because it does not explain everything — which is a demand no statistic in any field has ever satisfied, and which conveniently removes the number from consideration.

The discriminating test is whether the person actually performs the control they are asking for. Someone who names a confound, applies it, presents what came out, and explains why that variable is the relevant one is doing analysis. Someone who names a confound and stops has produced a reason to disregard a figure, which is a different activity wearing the same vocabulary.

What the tool is for

The rate is the beginning of an argument rather than the end of one. It establishes that something is more or less common in one population than another, which is the fact any causal claim has to start from and is not itself a causal claim.

What it protects against is narrower and more useful: being moved by a number that was never about the question. The twenty poisoned apples were real, verifiable, and irrelevant to the only decision on the table.

Links

• Validity and Truth — why a true number can license nothing, and how a premise differs from evidence for a practice.
• Racial Egalitarianism - The Disparity Inference — where group comparisons go once the rates are established.
• Mass Immigration - Cohesion — the aggregate-output defense, which depends on this distinction being missed.
• Decision Making — the general question of what a figure entitles anyone to conclude.
• Moral Language as Leverage — the same selective-application test, used on a moral vocabulary rather than on a rate.

Open questions

Where a confound is named honestly but the data to control for it does not exist, what should be concluded from the uncontrolled rate in the meantime?

Sources

Why Is Per Capita Impossible For Them?, 2026-05-04 — https://www.youtube.com/watch?v=h-zvwRdRzuA. Supplied the apples case, the three explanations, and the selective-use test.