Measurement Theory
The conditions under which numbers can legitimately represent attributes — and what it means for a measurement claim to be valid rather than merely numeric.
Overview
It is possible to assign numbers to almost anything. Measurement theory asks when doing so is meaningful — when the numbers preserve the structure of the attribute they represent, and when they merely create a false impression of precision.
Measurement theory asks the prior question to psychometrics: under what conditions do numbers legitimately represent attributes? The answer is not obvious. You can assign numbers to exam papers by marking them in order from worst to best — but the resulting rankings do not support arithmetic. You cannot say a student who scored 80 knows twice as much as a student who scored 40, or that the gap between 40 and 50 is the same as the gap between 70 and 80. The question is what mathematical operations any given measurement procedure actually licenses.
S.S. Stevens' 1946 paper "On the Theory of Scales of Measurement" established the standard taxonomy: nominal scales (categories with no order), ordinal scales (ordered but without equal intervals), interval scales (equal intervals, no true zero — temperature in Celsius), and ratio scales (equal intervals with a true zero — weight, time). Different mathematical operations are licensed at each level. Ordinal scales permit comparison; interval scales permit addition and subtraction; ratio scales permit multiplication and division. Much of the controversy in psychometrics turns on whether psychological scales achieve interval or only ordinal status.
The representational theory of measurement, developed in Krantz, Luce, Suppes and Tversky's Foundations of Measurement (1971–90, three volumes), provides a more rigorous foundation: measurement is valid only when the numerical assignment preserves the relational structure of the empirical attribute. This requires demonstrating that the empirical system — the objects and their relations — satisfies the mathematical axioms needed for the numbers to represent it faithfully. This is an existence proof problem, not merely a correlation problem: you must show that the empirical structure has the right formal properties before assigning numbers to it.
Joel Michell's critical work (1997) argues that most psychological measurement fails this test. Psychometrics has systematically assumed quantitative structure — that its attributes are genuinely measurable in a mathematically strong sense — without ever testing whether that structure exists. Most psychological scales, Michell argues, are ordinal at best. Denny Borsboom's Measuring the Mind (2005) extends this into a constructive proposal: what would it mean to genuinely measure a psychological attribute? What theoretical and empirical work is required before we are entitled to treat observed scores as measurements in any serious sense?
Key Texts
Foundational works in this research tradition.
Nominal, ordinal, interval, ratio scales; what mathematical operations each licenses. The conceptual scaffolding for 80 years of measurement discourse. The key distinction for AI-derived inference: ordinal rankings tell you who is higher, not by how much; interval scales permit the arithmetic most scoring systems assume.
The representational theory of measurement: numerical assignments are valid only when they preserve the relational structure of the empirical attribute. Three volumes covering additive, geometric, and probabilistic measurement. The rigorous foundation that psychometrics aspires to and rarely achieves.
The sceptical argument: psychology has assumed quantitative structure without testing for it. Most psychological scales are ordinal at best. A challenge to the routine treatment of Likert scales, composite scores, and IQ as interval measurements — with consequences for any AI system that acts on their values.
What it would mean to genuinely measure a psychological attribute: the theoretical requirements, the empirical tests, and what most contemporary psychometrics falls short of. A constructive extension of Michell's critique; the starting point for rigorous measurement programme design.
Related Research
Connected areas of inquiry.