TL;DR: For two decades, the most repeated idea in self-improvement was that willpower behaves like a muscle: it tires as you use it and recovers when you rest. The source was a genuine experiment. In 1998, participants asked to eat radishes while resisting fresh chocolate cookies subsequently gave up faster on an unsolvable puzzle. A 2010 meta-analysis of 83 studies found the effect significant, and the model became settled knowledge. Then psychology’s replication era began, and the same experiment was rerun with hypotheses declared in advance and sample sizes an order of magnitude larger. A 2016 registered replication across 23 laboratories and 2,141 participants found d = 0.04, with a confidence interval that included zero. A 2021 test across 36 laboratories and 3,531 participants found d = 0.06 and reported the data were four times more likely under the null hypothesis. The empirical floor beneath the muscle metaphor is gone. What remains is a better question: if willpower is not a reservoir, what is the tiredness telling you, and how should a day be built?
Where the metaphor came from
The idea began with a four-experiment paper published in the Journal of Personality and Social Psychology in 1998. In the first experiment, participants entered a room containing freshly baked chocolate cookies and a bowl of radishes. One group was told to eat the cookies; the other was told to eat only radishes and leave the cookies untouched. Both groups were then given a puzzle that had no solution. The radish group quit markedly sooner.
The interpretation was that resisting the cookies had drawn down a shared internal resource, and that the same resource was needed to persist at the puzzle. Three further experiments in the same paper reported comparable results with different tasks. The effect was named ego depletion.
Its intuitive appeal was enormous. Everyone remembers a bad decision made late in a tiring day, and the model offered a tidy explanation. A popular book in 2011 carried the idea well beyond academia, and “willpower muscle” entered ordinary speech.
The evidence over a decade
The 2010 meta-analysis pooled 83 studies and concluded that the ego-depletion effect was significant. That was the model at its strongest.
Then methodological standards changed. Researchers began declaring hypotheses and analysis plans before collecting data, and running the same protocol simultaneously across dozens of laboratories rather than one. Ego depletion was among the first major findings put through the new process.
Table 1. CEOtudent evidence audit: a quarter-century of the “willpower is a muscle” claim. Each row reports the sample and result as stated in the primary publication.
| Year | Study | Method | Sample | Reported result |
|---|---|---|---|---|
| 1998 | Baumeister, Bratslavsky, Muraven and Tice, Journal of Personality and Social Psychology, volume 74, pages 1252-1265 | Four experiments in one laboratory, including the radish and cookie study | Single laboratory | Effect found; origin of the muscle model |
| 2010 | Hagger, Wood, Stiff and Chatzisarantis, Psychological Bulletin, volume 136, pages 495-525 | Meta-analysis of 83 studies | 83 studies | Significant effect reported; model became consensus |
| 2016 | Hagger and colleagues, Perspectives on Psychological Science, volume 11, pages 546-573 | Preregistered multilab registered replication report | 23 laboratories, 2,141 participants | d = 0.04; 95 percent confidence interval [-0.07 to 0.15] includes zero |
| 2021 | Vohs and colleagues, Psychological Science, volume 32, pages 1566-1581 | Preregistered multisite paradigmatic test | 36 laboratories, 3,531 participants | d = 0.06; data four times more likely under the null hypothesis |
Read the bottom two rows together and a figure emerges that is stronger than either study alone.
Table 2. CEOtudent derived total: the scale of the preregistered evidence. Summed from the final two rows of Table 1.
| Measure | Value |
|---|---|
| Laboratories in preregistered tests | 59 |
| Total participants in preregistered tests | 5,672 |
| Effect size range reported across both | d = 0.04 to d = 0.06 |
| Direction of the effect | Indistinguishable from zero |
Fifty-nine laboratories and 5,672 participants ran the experiment with their hypotheses declared in advance and did not find the effect. That is not the same as an inconclusive result. It means the effect is not present at the size claimed, and the difference between those two statements matters.
Why it took so long
A finding that appears in 83 studies and vanishes across 59 laboratories looks like a contradiction. It is not, and understanding why is the durable lesson here.
Before preregistration, researchers could decide which analysis to report after seeing the data. Studies with significant results were far more likely to be published, while null results stayed in the file drawer. Combine those two mechanisms and an effect that is small or absent in reality can look large and consistent in the published literature. The 2016 replication was itself launched on that basis: earlier meta-analyses had questioned the size and existence of the effect and identified instances of possible bias.
So the 83 studies were not lying. They were showing how a published literature gets assembled.
That is a general reading habit, not a one-off caution, and we trace the same mechanism through a number that reached peer review with no measurement behind it at all in our audit of the 35,000-decisions claim.
What survives the collapse
The feeling of making worse decisions in the evening is real. What changes is the explanation.
The muscle model said a reservoir drains and rest refills it. If that were right, the behavioural implication would be to conserve: spend no willpower on trivial choices, wait while the tank refills. Millions of people organised their days around exactly that logic.
The preregistered evidence removes its foundation. A more defensible replacement is that the problem is not a depleting resource but an accumulating load. Decisions leave residue across a day: deferred matters, unclosed loops, unresolved conversations. The reason judgement degrades by evening may have less to do with an empty tank than with the number of open files you are carrying at once.
That distinction changes the prescription. A reservoir model tells you to wait. A load model tells you to close things.
The practical replacement: budget decisions, not willpower
Convert repeats into rules. Any decision you make more than once a week, with the same inputs and the same acceptable answer, is a candidate. Decide it once and stop re-deciding. What you save is not willpower; it is attention residue.
Choose speed on reversible decisions. Where you could undo the choice within a week at modest cost, delay costs more than error. Deliberation here is waste, not virtue.
Put irreversible decisions on the calendar. For most people the monthly count of genuinely irreversible decisions fits on one hand. Those should not be made in corridors, on phones, or between two other things. That was the muscle model’s correct instinct; the mistaken part was treating it as a question of reserves.
Treat tiredness as a prompt, not a measurement. The preregistered evidence suggests subjective fatigue is not a reliable indicator of decision quality. What is reliable is the record: which decisions you later reversed, which approvals you gave without reading. That record is your own data and needs nobody’s meta-analysis. Our decision journal protocol covers how to keep one.
The CEO and the student
The student may read this as a loss. A model that was believed to work for years is gone, and no single-sentence rule has replaced it. But watching a model collapse is the most valuable kind of learning available: seeing how an idea was built, how it settled, and how it was tested by better methods leaves a skill that outlives the idea itself.
For the CEO the conclusion is more direct. Good management is not holding the correct list of beliefs; it is updating beliefs when the evidence moves. The willpower muscle was plausible, popular and widespread for twenty years. Then it was examined more rigorously and it did not hold. The same will happen to several things you currently treat as settled.
Learn like a student: follow the source, ask about the method, accept “we do not know.” Lead like a CEO: stop budgeting a resource you cannot measure, and spend your scarce deliberate attention on the small number of decisions that are genuinely expensive.
FAQ
Does this mean willpower does not exist?
No. What the preregistered studies failed to find is a much narrower claim: a measurable drop in performance on a second self-control task immediately following a first one. Individual differences in self-control, the strength of habits, and the translation of intention into behaviour are separate research areas and are not affected by these results.
Was the 1998 radish experiment flawed?
There is no allegation of misconduct. The issue is not that one study was wrong but that its result did not generalise. Small-sample single-laboratory findings failing to survive preregistered multilab replication is a common pattern in psychology.
Does eating sugar restore willpower?
That claim was an extension of the muscle model and weakened along with it. Since preregistered replications did not find the underlying effect, interventions proposed to counteract that effect lose their rationale.
Then why do I make worse decisions in the evening?
Probably not because a reservoir has drained, but because unclosed work and the number of options in play accumulate through the day. The two explanations imply different remedies: one tells you to wait, the other to close loops and set rules.
Why was this article updated?
When this page was first published, willpower functioning as a muscle was treated as established. In the intervening years the same claim was tested with far larger samples and stricter methods and was not supported. A published text should be updated when the evidence beneath it changes.
Sources
Baumeister, R. F., Bratslavsky, E., Muraven, M. and Tice, D. M. Ego depletion: is the active self a limited resource? Journal of Personality and Social Psychology, 1998, volume 74, pages 1252-1265.
Hagger, M. S., Wood, C., Stiff, C. and Chatzisarantis, N. L. D. Ego depletion and the strength model of self-control: a meta-analysis. Psychological Bulletin, 2010, volume 136, pages 495-525.
Hagger, M. S. and colleagues. A multilab preregistered replication of the ego-depletion effect. Perspectives on Psychological Science, 2016, volume 11, pages 546-573.
Vohs, K. D., Schmeichel, B. J. and colleagues. A multisite preregistered paradigmatic test of the ego-depletion effect. Psychological Science, 2021, volume 32, pages 1566-1581.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
