A broken promise usually gets explained one of two ways. Either the person never really meant it, or they meant it and lacked the willpower to follow through. Both explanations are about character, which is why both sting.
Two experiments, from different decades and different fields, put the failure somewhere else. Not in the will, and not in the sincerity, but at two specific points that have each been measured.
The first failure happens at the moment of promising
Buehler, Griffin and Ross asked 37 psychology students to say when they would finish their honors theses, and then waited to see.
The Artful Age
A weekly letter on aging well, family across generations, and the creative life after the kids leave home.
The average prediction was 33.9 days. The average reality was 55.5. Only 29.7 percent of them finished within the time they had given themselves.
The researchers had been cannier than that, though. They had also asked each student for a worst-case estimate: how long it would take if everything that could go wrong did. That figure averaged 48.6 days, and only 48.7 percent of students managed to beat it.
Read that twice. Fewer than half of them met the deadline they set while deliberately imagining disaster.
Why the error is more specific than optimism
The tempting reading is that people are simply optimists. The same set of studies rules that out rather neatly.
When observers were asked to predict how long these same people would take, they guessed later, not earlier. The bias attaches to predictions about yourself and reverses when you are looking at somebody else.
A think-aloud study shows the mechanism. Asked to make a prediction, participants spent 71 percent of their reasoning on plans for the task ahead, and 7 percent on their own past experience with similar tasks. Observers watching the same situation drew on past experience 48 percent of the time. The people actually doing the task: 2.4 percent.
That is the finding underneath the famous one. You know how long the last one took. You simply are not consulting it. You are consulting the plan, and the plan is always a clean run.
Their fourth study is the encouraging footnote: when people were instructed to connect relevant past experiences to the prediction, the optimistic bias disappeared.
The second failure happens at the moment of keeping
The other half of the question is what a promise actually does once it has been made, and Vanberg designed an experiment to find out.
Two explanations were on the table. Either people have a genuine preference for honoring their word, which creates an obligation to the particular person they gave it to, or promises work indirectly: they create expectations, and people simply do not like disappointing expectations.
Those two look identical in ordinary life, because the person you promised is also the person expecting something. So the design pulls them apart. Across 192 subjects and six sessions, people communicated, made promises, and then some pairs were randomly rematched. A person might find themselves still with the partner they had promised, or moved to a new recipient who was holding somebody else’s promise.
Expectations are equivalent either way. The only thing that changes is whether the person in front of you is the one you gave your word to.
Among those who stayed with their own partner, 73 percent of promisers chose the cooperative option. Among those rematched, 54 percent.
The gap supports the first explanation. What does the work is not the expectation floating in the room. It is the fact that you said it, to them.
What this suggests about the promises you make to yourself
Here the article has to be careful, because it would be easy to overclaim. Vanberg tested promises made to another person. He did not test promises made to oneself, and nothing here is a finding about those.
What can be said is narrower and still worth saying. Whatever it is that held those promises in place was directed at a specific other person, and it weakened measurably the moment that person was swapped out. A promise you make to yourself at eleven at night has no second party at all. It has the planning fallacy in full working order at the moment of making, and, on the evidence available, nothing equivalent standing behind it at the moment of keeping.
Where the evidence stops
Buehler and colleagues flag that self-reported completion times in one of their studies raise validity concerns, and that it stays uncertain how well people can use their own history to sharpen predictions outside the laboratory.
Vanberg notes that his results may reflect correlation with unobserved factors rather than pure causation. And it is worth saying plainly that a dictator game with randomly rematched strangers is not a promise between two people who will still know each other in a year. What it offers is a clean measurement of one mechanism, not a portrait of promising.
What survives
Two failure points, and neither is weak character.
At the making, you priced the job by imagining it rather than by remembering it, which is a thing you do about yourself and not about other people. At the keeping, if there was another person involved, something real was holding the promise up, and it was stronger than anybody’s expectations.
And if there was no other person, then the promise you broke was the one built with only half the apparatus. Which is worth knowing before you conclude anything about your character from having broken it again.