Goodhart's Law and Metrics That Get Gamed

What this topic is

Goodhart's law is the observation that a statistical relationship you rely on for measurement tends to break once you start using it for control. Charles Goodhart made the point about monetary aggregates in 1975; the version everyone quotes ("when a measure becomes a target, it ceases to be a good measure") is Marilyn Strathern's 1997 compression of it. The mechanism is not mysterious. Every metric is a proxy for something you actually care about but cannot observe directly, and the proxy tracks the real thing only under the pattern of behaviour that existed when you validated it. Setting a target changes that behaviour, which is the entire point of setting it, and the relationship you were relying on was never guaranteed to survive the change.

Why it matters

Almost every management system in existence is built on proxies: a support number standing in for whether customers are looked after, a pipeline number standing in for future revenue, an engagement number standing in for whether a product is worth having. The moment any of those enters a bonus plan, a board pack or a performance review, it starts to drift away from the thing it was chosen to represent. And it drifts silently, because the number itself keeps improving. Organisations rarely discover the problem from the metric. They discover it from a customer, a regulator, or a resignation.

Who should learn it

Anyone who sets targets, runs a dashboard, writes objectives, or has to explain to a board why a number that has been green for six quarters is attached to a business that feels worse than it used to.

What you will understand

• The mechanism: why a proxy validated under one behaviour pattern stops holding under another
• The four distinct ways a target breaks a measure, and which repair fits which one
• How to instrument a target so gaming shows up in weeks rather than in a regulatory filing
• When a number should be a diagnostic nobody is judged on, and how to keep it that way

Prerequisites

• Managing Managers and the Second-line Problem
• One-to-ones and the Manager's Information Loop

Common misconception

"Goodhart's law is about dishonest people, so hire better and it goes away." Two of the four ways this failure happens require no intent at all. If a measure is a noisy proxy, selecting hard on the measure gets you the noise as well as the signal, and no one has done anything but their job. If the correlation you relied on was produced by a common cause rather than by the measure driving the outcome, then pushing the measure moves nothing real, and again nobody has cheated. The version involving deliberate manipulation is the most visible and the least interesting. A second misconception is that the answer is to stop measuring. An organisation without measures is not an organisation without incentives. It is one where the incentives are set by proximity to whoever decides, which is worse and much harder to audit.