Measurement theory

What has to be true of a procedure before its output counts as a measurement, and what the resulting number means. The corpus holds one account, mari-measurement-across-sciences, and it is a strong one: an attempt at a single concept system covering thermometers and reading-comprehension tests alike.

Why the spoke has this page

Every other subject here is mathematics. This one is not, and the page says so at the top rather than burying it. Measurement theory sits below the mathematics rather than inside it: statistics starts with numbers and reasons backwards to the mechanism that produced them, and measurement theory asks where those numbers came from and under what conditions they refer to anything.

The spoke’s boundary rule — a text teaching a body of theory from first principles routes here, an application of that theory routes to the spoke owning the claim — files it here, since no spoke in the hub owns measurement as a subject and three of them borrow it. Whether the rule should have filed it here is open question 7 in synthesis.

The four necessary conditions

mari-measurement-across-sciences proposes that a measurement is (i) an empirical process, (ii) designed on purpose, (iii) whose input is a property of an object, and (iv) whose output is information in the form of values of that property. Necessary, not sufficient — which is the book’s structure: the uncontroversial conditions first, then several hundred pages establishing what has to be added.

Two consequences follow, and both cut against how measurement is usually described:

Measurement is not the same as quantification. Treating it as such hides the empirical half of the process. It also raises the question the book spends chapter 6 on — whether a non-quantitative property can be measured, and under what conditions.
Measurement does not transmit a preexisting true value. A model is always involved; a result is a claim built through that model, not a reading taken off the world.

Error, uncertainty, validity

Three ways of accounting for a measurement that is not perfect. The split runs across the whole hub.

Error / true-value approach: a true value exists and the result differs from it by an error. Uncertainty approach: the result is the set of values compatible with the measurand given what is known, made comparable across people and places by traceability to conventional units. The physical sciences moved from the first to the second.

Validity is the human sciences’ answer to the same problem, and it asks something different: not how far off the number is, but whether the procedure measures the intended construct at all. The book traces early validity, construct validity, argument-based and causal accounts, and argues the two traditions are addressing one question in mutual ignorance.

Scales, values, and quantities

The other durable piece is the analysis of scale types and of what a value of a property is. Quantities get a constructive treatment: additivity, reference quantities, scale transformations between alternative references — then the cases that break additivity, temperature and reading-comprehension ability, which is where the physical and psychosocial examples stop looking different.

This is the layer that classical statistics assumes silently. A t-test does not ask whether its input is on an interval scale; the analysis that decides it is has already happened, somewhere, off the page.

Where it sits under this spoke

Directly beneath mathematical-statistics. Inference starts once a number exists; measurement theory governs whether the number should exist. Also adjacent to ../psychology-wiki (psychometrics and the validity tradition are half of this subject’s material — the runner-up spoke in routing) and to ../ai-governance-wiki and ../agentic-tooling-wiki, where “measure the model’s capability” is a live problem and the measurand is rarely defined before the benchmark is run.

mari-measurement-across-sciences · mathematical-statistics · probability-theory · mathematics · synthesis