Precommitment: The Psychology of Binding Your Future Self

Precommitment works in almost exact proportion to how hard the constraint is to escape. That single finding runs through every study in this area, and it explains why most commitment devices people build for themselves accomplish so little. Ulysses did not resolve to resist the sirens. He had himself tied to the mast and gave the rope to somebody else.

Schelling and the bargaining problem

Thomas Schelling, who spent his career on strategic interaction between adversaries, turned the same apparatus on the self. His Richard T. Ely lecture, published in the American Economic Review in 1984 (volume 74, issue 2, pages 1 to 11) as "Self-Command in Practice, in Policy, and in a Theory of Rational Choice," reframed self-control as a negotiation between a present self and a future self with genuinely different preferences.

That changes what counts as a solution. If the future self wants the same things less intensely, the answer is motivation. If the future self wants different things, motivation is irrelevant, and the only workable move is to limit what that person can do.

Schelling catalogued the tactics people use: handing authority to somebody else, committing in advance, removing oneself from temptation, removing the resources that make defection possible, submitting to surveillance, and arranging penalties and delays. His example of a woman asking her obstetrician in advance to withhold anesthesia is the cleanest illustration, since the request is made by one person and must be honored against the wishes of the same person later.

Schelling's paper is an essay, not an experiment. It supplies the frame. The tests came afterwards, and they are uneven.

The deadline study, and the replication attached to it

The most-cited experimental work here is Dan Ariely and Klaus Wertenbroch's 2002 paper in Psychological Science (volume 13, issue 3, pages 219 to 224).

In their first study, 99 professionals in an MIT executive education course were split by section. Forty-eight received fixed, evenly spaced deadlines for three papers. Fifty-one set their own binding deadlines, enforced by a 1 percent grade penalty per day late, with no advantage to submitting early. Most chose early dates. So people do voluntarily accept costly constraints, which was the first question.

The grades answered the second. The fixed-deadline section averaged 88.76. The self-imposed section averaged 85.67. Freedom to schedule made performance worse, and the gap largely disappeared among students who happened to space their own deadlines evenly, which locates the problem in timing rather than in self-imposed rules feeling less binding.

Their second study, a proofreading task with 60 participants paid per error found and penalized per day late, produced the ordering usually quoted: evenly spaced imposed deadlines best, self-imposed second, a single end deadline worst.

Here is the part that belongs with every citation of this paper. Kyle Hyndman and Alberto Bisin ran a preregistered replication of that second study with 124 participants in 2025, after a near-identical 120-participant version in 2024, and it did not replicate. Deadline condition had a negligible effect on all three performance measures, and evenly spaced imposed deadlines did not stand apart. It is published in Psychological Science, the same journal as the original.

One result did survive, and it is the more interesting one. Over 93 percent of participants in the self-imposed condition voluntarily bound themselves. People reliably know they have a self-control problem. What failed to replicate is the claim that this particular fix does anything about it.

Where precommitment has held up: money and cigarettes

The economics field evidence is in better shape than the psychology lab evidence, which is an unusual arrangement worth noticing.

Richard Thaler and Shlomo Benartzi's Save More Tomorrow program, published in the Journal of Political Economy in 2004 (volume 112, pages 164 to 187), asked employees to commit in advance to directing part of future pay raises into retirement savings. In the first implementation at a midsize manufacturing company in 1998, 78 percent of those offered the plan joined, 80 percent of joiners were still in it through the fourth pay raise, and average saving rates rose from 3.5 percent to 13.6 percent over 40 months.

Look at why it works. The commitment is to money you do not have yet, so it costs nothing at signing. Escalation is automatic, and leaving requires an active step. It is precommitment fused to a default, which is the friction argument doing at least half the work, as it usually is when structure closes the distance between intending something and doing it.

Xavier Gine, Dean Karlan and Jonathan Zinman tested a harder version. Their CARES product, reported in American Economic Journal: Applied Economics in 2010 (volume 2, issue 4, pages 213 to 235), offered Philippine smokers a savings account: deposit your own money for six months, then take a urine test for nicotine and cotinine. Pass and it comes back. Fail and it is forfeited to charity.

Eleven percent of smokers offered the product took it up. Those randomly offered it were 3 percentage points more likely to pass at six months, and the effect persisted in surprise tests at twelve months. That pair of numbers is the honest shape of this literature: real, durable, small, and rarely accepted.

Heather Royer, Mark Stehr and Justin Sydnor found the same profile in exercise. In their 2015 field experiment at a Fortune 500 company, a four-week gym incentive alone produced only small lasting effects, while adding a self-funded commitment contract afterwards produced changes still detectable years later.

The 61 percent

Katherine Milkman, Julia Minson and Kevin Volpp's temptation bundling study, published in Management Science in 2014, restricted tempting audiobooks to a university gym. When it ended, 61 percent of participants chose to pay for continued gym-only access to the iPods, having lived inside the arrangement for weeks.

Set that beside the 93 percent who bound themselves in the Hyndman and Bisin replication, and beside the MIT students who chose early deadlines and still chose them badly, and one picture emerges: people are markedly better at predicting their future self-control failures than at preventing them. The full evidence trail on the bundling study, including the parts that faded, is covered separately.

What makes a commitment device work, ranked

1. The constraint has to be genuinely costly to escape. This is the whole finding. A rule you can revoke is a preference with paperwork.
2. Somebody other than you should hold the key. Schelling's first tactic, and the one most self-designed systems omit.
3. The cost should land at the moment of defection. A penalty arriving next quarter is not competing with a temptation arriving now.
4. Bind the default rather than the decision. Save More Tomorrow works because doing nothing produces the saving. Anything requiring a repeated act of will is not a commitment device.
5. Commit to money you do not yet have. Future income binds more easily than present income, because forgoing it does not feel like a loss.
6. Expect low take-up, including your own. Eleven percent accepted the smoking contract. Between a device you will decline and a weaker one you will accept, take the weaker one.
7. Do not bind yourself to a schedule you have not tested. The MIT students set their own deadlines badly. Knowing you need structure is not knowing which structure.

The honest limits

The most famous precommitment experiment in psychology has now failed a preregistered replication, and anyone citing it should say so. Take-up rates are low everywhere they are measured. The field effects are durable but modest, a few percentage points rather than transformations. And the mechanism is uncomfortable: it concedes your future self is not persuadable and treats that person as an obstacle to route around.

What it does not require is that you want anything more than you already do, which is why it belongs in the same family as making a goal specific enough to be checkable. Both replace an appeal to willingness with a change to the situation, and the situation is the part you can actually edit.