Measurement as Uncertainty Reduction

Definition

A measurement is "a quantitatively expressed reduction of uncertainty based on one or more observations." It does not need to be exact, and it does not need to eliminate uncertainty — it only needs to narrow it. The thing observed doesn't even need to be a conventional quantity: a yes/no question ("will we win the lawsuit?") counts as a measurable object as long as your confidence about the answer can be expressed as a probability. This borrows directly from Claude Shannon's information theory, where information is defined as the amount of uncertainty ("entropy") a signal removes from a receiver who already knows something.

In the Book

Hubbard opens by arguing that "intangibles" don't really exist — things labeled immeasurable (management effectiveness, public image, the risk of a project failing) are simply things nobody has yet applied an observation to. In Chapter 3 he traces two academic sources that popular usage garbles: Shannon's 1948 "A Mathematical Theory of Communication," which frames information as uncertainty reduction relative to a receiver's prior state, and psychologist Stanley Smith Stevens's 1946 taxonomy of nominal, ordinal, interval, and ratio scales, which shows measurement doesn't require homogeneous, addable units — a nominal "yes/no" or an ordinal ranking can still be a measurement. Hubbard also stresses the "Bayesian" reading of probability this implies: probability describes the state of the observer's knowledge, not necessarily an objective feature of the system, so measuring the average weight of salmon in a river changes your uncertainty about their weight without changing the fish. He gives the corporate-dashboard critique as a foil: displaying numbers with no defined decision or threshold for action isn't measurement in his sense, because it doesn't demonstrably improve anyone's decisions — the test of a real measurement is that estimates and decisions actually get better on average.

Why It Matters

Redefining measurement this way dissolves the "immeasurable" excuse: instead of asking whether something can be measured with certainty, you ask whether any observation would narrow your current uncertainty at all, and by how much. That reframing turns a binary (measurable / not measurable) into a continuous, economic question — how much uncertainty reduction is a given observation worth — which is what makes rigorous measurement possible for soft, ambiguous, or "intangible" targets in any domain, not just physical science.