How to Measure Anything

Douglas W. Hubbard treats measurement as a way to reduce uncertainty enough to inform a decision. The publisher's third-edition excerpt introduces the approach. This is a useful starting point when a team calls something important but “impossible to measure.”

Define the decision before collecting data

In this original example, a library is considering a new checkout screen. “It will be easier to use” is too vague to evaluate. The immediate decision is whether to spend four hours installing it. One relevant uncertainty is the time it might save per checkout.

Suppose there are 1,000 checkouts a month. Your rough estimate is a saving of 5–20 seconds each. That would mean about 1.4–5.6 hours saved per month. The calculation does not prove a benefit; it makes the uncertain assumption visible. It also ignores training time and mistakes, which may change the decision.

Choose a measurement that could change your answer

Try observing a small set of representative checkouts with each screen, including difficult cases. Record completion time, abandoned attempts, and staff help. If only experienced staff use the new screen while newcomers use the old one, the comparison will mix the interface change with user experience.

Before the trial, decide what finding would favor keeping the old screen. Perhaps the time saving is tiny or errors become more frequent. If no plausible result could change your decision, ask whether the measurement is serving a real choice or just decorating a choice already made.

Do not turn a proxy into the goal

A faster checkout is not automatically a better service. A visitor may need an explanation, an accessible display, or help correcting a mistake. Record those outcomes too. A single number can be convenient while leaving out what matters most.

Choose this book for structuring practical measurement questions. An estimate with a range remains an estimate, and a small convenience sample is not representative merely because it contains numbers. Naked Statistics is a useful companion for examining samples and comparisons.