Thinking tools · 4 of 6
Incentives explain more than intentions
Predict behavior from who gets paid for which outcome.
In the 1990s the British government set a target: nobody should wait more than four hours in an accident and emergency department. Waits above four hours duly collapsed. Some hospitals had achieved this by keeping arriving patients in the ambulance outside, because the clock started when they came through the door. Nobody involved wanted patients in car parks. The clock started at the door, so the patients waited at the door.
The rule
To predict what someone will do, find out what they get paid for and what they get punished for. That will beat knowing what they say they want, what they sincerely believe, and what they are like as a person, most of the time and by a lot.
This is not a claim that people are cynical. The uncomfortable part is the opposite: people are very good at believing the thing that pays them. The surgeon declining the hard case has a genuine clinical reason ready. It is not a lie. It arrived because it was useful, and it would not have arrived if the number were not published.
A factory that does exactly what it is told
Below is a nail factory with a hundred hours of labor and a choice of what length to make. It has no motives, no cynicism, and no capacity to cheat. It is arithmetic picking the largest number in a list.
Pay it by the nail and it makes two thousand four hundred things too short to fasten anything. Pay it by the ton and it makes seven hundred and fifty crowbars. Both plans were met in full.
The four-hour target and the ambulance queues are documented in UK National Audit Office and parliamentary committee reports from the 2000s, which are public. The nail factory is a model, and the figures come from it. There is a test in the repository confirming that widening what counts as a usable nail far enough makes "as many as possible" the correct instruction, which is the limit on the whole argument.
ConventionGoodhart's law: when a measure becomes a target, it stops being a good measure. The version worth carrying is the mechanism rather than the slogan. A metric is a proxy that used to correlate with the thing you want. Paying on the proxy makes people push on the proxy directly, which breaks the correlation that made it worth measuring. The metric was not wrong. Paying on it is what made it wrong.
Where to look
The useful version of this is a checklist rather than a worldview. When you want to predict what an organization or a person will do, find four things.
- What is counted. Not what is stated as the goal. What appears on the scorecard, the dashboard, or the commission statement.
- Who is paid on it.The person doing the work, their manager, and their manager's manager often have different numbers, which is why the levels disagree.
- Over what horizon. A bonus paid this year on a decision that pays off in five is a bonus on the decision looking good now.
- Who carries the downside.If the upside is yours and the downside is somebody else's, expect more risk than anybody would choose for themselves. Most of 2008 is in that sentence.
What to do with this
Before you accept advice, ask how the person giving it gets paid. Before you predict what a company will do, find out what its managers are bonused on. Before you set a target for anybody, including yourself, ask what the laziest possible route to a good score looks like, because that route will be found.
And apply it to yourself, which is the hardest direction. If you get paid for hours, you will find reasons the work takes longer. If you get paid for shipping, you will find reasons the testing was probably enough. Knowing which way your own incentives push is not the same as escaping them, and it is most of what you can do about it.
Test yourself
01A hospital publishes surgeon-by-surgeon mortality rates so patients can choose well. What happens next?
Surgeons start declining the hardest cases. A patient who is very likely to die is a patient who will damage a published number, and the safest way to protect the number is to send them somewhere else.
Everybody in this story is behaving reasonably. The hospital wanted transparency, the surgeon wants to keep operating, and the patient who needed the operation most is the one who cannot get it. The fix is not to hide the numbers. It is to adjust them for how sick the patient was, which is harder, more expensive, and the only version that works.
02Your financial adviser is paid a commission by the fund company whose products they recommend. What should you assume?
That the recommendations lean toward the products that pay, and that the adviser believes their own advice. Those are not in tension. People are extremely good at finding sincere reasons for conclusions that happen to pay them, and that is what makes the arrangement dangerous rather than merely dishonest.
The question to ask is never whether they are a good person. It is how they get paid, and the answer changes what you do with what they tell you. A fee you pay directly and a commission somebody else pays produce different advice from the same adviser.
03When is the incentive not the explanation?
When people do not know how the incentive works, when the effect is too small or too delayed to notice, or when something stronger is pulling the other way: professional norms, being watched by peers, or simply not wanting to be the kind of person who does that. Incentives are the strongest single predictor and they are not the only one, and treating them as the only one produces its own bad forecasts.
A tutor that knows this lesson. It asks before it explains, and it will not tell you what to do with your own money.
Educational material, not investment or policy advice. Figures are cited where they come from a filing or a statistical series, and labelled as illustrative where they do not.