An economic indicator tracks an underlying reality well for years, used passively as one input among many in analysis and forecasting. The moment that same indicator becomes an explicit policy target, with incentives or decisions tied directly to hitting it, its reliability as a measure of the underlying reality often starts to erode — not because the statistic was poorly designed, but because of a well-known dynamic named after the economist Charles Goodhart: when a measure becomes a target, it ceases to be a good measure.
Why this happens, mechanically
A metric is originally useful because it correlates reasonably well with some underlying reality that's harder to measure directly. As long as no one is specifically optimizing for the metric itself, that correlation tends to hold, because the paths that move the metric are largely the same paths that move the underlying reality. Once the metric becomes an explicit target — tied to incentives, policy decisions, or performance evaluation — people gain a reason to find any way to move the metric, including paths that move the metric without moving the underlying reality at all, and some of those paths are usually easier than actually changing the underlying reality. The correlation that made the metric useful in the first place starts breaking down specifically because of the new incentive to optimize it directly.
A classic economic example
A central bank targeting a specific monetary aggregate as an indicator of economic activity can find that the relationship between that aggregate and actual economic activity shifts once financial institutions and markets adapt their behavior specifically around the fact that the aggregate is being targeted — new financial instruments and practices emerge that move the targeted number without moving the broader economic conditions it was originally meant to proxy for. The aggregate that worked well as a passive indicator becomes progressively less informative precisely because of the attention and incentive now attached to it.
This doesn't require anyone acting in bad faith
Goodhart's Law doesn't depend on deliberate gaming or fraud, though it certainly permits that too — it operates just as reliably through good-faith optimization. An organization or individual sincerely trying to improve their standing on a targeted metric will naturally gravitate toward whichever available actions move the metric most efficiently, and some of those actions inevitably diverge from what actually improves the underlying reality the metric was meant to represent, simply because moving a number and improving reality are rarely perfectly identical tasks, and effort tends to flow toward whichever is more directly rewarded.
How to tell when this has already happened to a metric you rely on
Checking whether a targeted metric still moves together with independent, harder-to-game evidence of the underlying reality it was meant to track is the most direct diagnostic — if the metric continues improving while independent indicators of the underlying reality stagnate or diverge, that divergence is a strong signal Goodhart's Law has taken hold. This requires deliberately maintaining at least one measure that isn't itself a target, specifically so it retains its original diagnostic value as a check against the targeted metric's potential drift.
What this means for anyone designing or relying on targeted metrics
- Expect any metric that becomes an explicit target to gradually lose some of its original correlation with the underlying reality it was meant to track
- Maintain at least one independent, non-targeted indicator specifically to check whether a targeted metric has started to decouple from reality
- Be specifically suspicious of a targeted metric improving steadily while other, harder-to-move indicators of the same underlying reality stay flat
- Design incentive structures around multiple metrics where possible, since gaming several genuinely independent metrics simultaneously is considerably harder than gaming one
Goodhart's Law isn't an argument against using metrics or targets — it's a warning that a metric's usefulness has an expiration date once it becomes a target, and that expiration is worth actively checking for, not assuming away.