Ego Depletion and the Limits of Resource Metaphors

There are moments when effort seems to have been used up.

You finish something demanding, turn to the next task and find that the concentration or willingness available earlier is no longer present in the same way. Resisting one impulse appears to make the next one harder. After making decisions for hours, another choice can feel disproportionately difficult.

The language that comes naturally is the language of depletion: drained, spent, running on empty.

The experience can be real. But the metaphor may not explain it.

For decades, one influential theory proposed that acts of self-control consume a limited internal resource. Use enough of it and less remains for the next act of control. This idea became known as ego depletion.

The theory was intuitive, productive and initially supported by a large experimental literature. Publication-bias analyses, preregistered studies and major multilaboratory replications have since changed the scientific picture.

Prior effort may sometimes affect what happens next. What has become much harder to defend is the claim that this occurs because a common store of willpower has been consumed.

What ego depletion originally meant

Ego depletion was not originally another name for feeling tired. It was a proposed account of how self-control works.

The theory was commonly tested through a sequential-task paradigm. Participants first completed a task assumed to require either more or less self-control. Both groups then completed a second task that was also thought to require control. If the group exposed to the more demanding initial task performed worse on the second task, the difference was interpreted as an ego-depletion effect.

The tasks varied widely. Participants might suppress an emotional response, resist tempting food, inhibit a habitual response, control their attention or persist on a frustrating problem. The later outcome could measure persistence, inhibition, accuracy, endurance, consumption or another behaviour interpreted as requiring self-control.

The important claim was that these different activities drew on something shared. An act of emotional regulation could therefore impair later persistence because both supposedly depended on the same limited resource.

This became known as the strength model of self-control. Willpower was compared with a muscle: it could become temporarily fatigued through use, recover after rest and perhaps become stronger through repeated exercise.

The comparison made a complex theory easy to understand. It also bundled together several propositions. Self-control might have limits. Earlier effort might affect later performance. Different tasks might depend on common processes. But none of those propositions necessarily shows that a resource is consumed when control is used.

Why the theory became persuasive

The foundational experiments reported cross-task aftereffects consistent with the theory. Participants who had regulated themselves during an initial task sometimes persisted less or performed worse afterward (Baumeister et al., 1998; Muraven et al., 1998).

The findings appeared across different forms of control. This created the impression that researchers were detecting a broad underlying mechanism rather than an effect confined to one task.

A major 2010 meta-analysis strengthened that impression. It combined 198 independent tests from 83 studies and reported a medium-to-large average effect of approximately (d = 0.62), alongside substantial heterogeneity across the included tests. The analysis also reported changes in effort, perceived difficulty, negative affect and subjective fatigue (Hagger et al., 2010).

The theory now appeared to connect laboratory findings with familiar experience. People felt drained after prolonged control, experiments seemed to reproduce the pattern and the published literature suggested that the effect was sizeable.

There was, however, a problem. A meta-analysis can summarise only the studies and results available to it. If positive findings are more likely to be published, or if small studies use flexible methods that increase the chance of finding something significant, combining them does not remove that distortion.

Under those conditions, a meta-analysis can consolidate the distortion rather than correct it.

When the evidential picture changed

Later analyses found signs that the ego-depletion literature was affected by small-study effects and selective reporting.

Small studies produce imprecise estimates. Some generate unusually large positive results through sampling variation. If those results are more likely to be published than null findings, the visible literature will make the effect appear larger and more consistent than it is.

The ego-depletion paradigm also allowed considerable variation. Researchers used different initial tasks, outcome measures, exclusion rules and analytic choices. Not every choice was unreasonable, but flexibility created many routes through which a positive finding could emerge.

Carter and colleagues conducted a more focused meta-analytic reassessment. After examining publication bias and small-study effects, they found little evidence for the depletion effect under the laboratory methods most frequently used (Carter et al., 2015).

Other analyses reached less absolute conclusions. Dang's updated meta-analysis retained a smaller positive estimate of approximately (g = 0.24) after trim-and-fill correction. Some task categories appeared more effective than others, although those differences required further confirmation (Dang, 2018).

The disagreement cannot be resolved by selecting a preferred correction method. Statistical adjustments for publication bias rely on assumptions, particularly when studies differ substantially in design. They cannot reveal the one true effect hidden behind a heterogeneous literature.

The responsible conclusion is that the original estimate became difficult to trust. If an effect remained, it was probably smaller, less uniform and more dependent on method than the early literature suggested.

What the major replications found

The most important change came from large studies designed and registered before researchers saw the results.

In 2016, Hagger and colleagues coordinated a preregistered replication across 23 laboratories. The project used a standardised initial control task followed by a second interference task. Across 2,141 participants in the primary analysis, the pooled estimate was (d = 0.04), 95% CI ([-0.07, 0.15]) (Hagger et al., 2016).

The study did not reproduce the effect expected from the earlier literature.

The choice of tasks became a source of dispute. A replication can directly test only its selected procedure, and critics questioned whether that procedure adequately represented ego depletion. This limits how broadly the result can be generalised, but it does not erase the near-zero estimate for the tested procedure.

A later preregistered project led by Dang used a different task combination. Participants first completed a Stroop task with differing control demands and then performed an antisaccade task. Across 12 laboratories and 1,775 participants, the project detected a small sequential-task aftereffect under its specified Stroop–antisaccade procedure: (d = 0.10). After excluding participants considered likely to have responded randomly, the estimate increased to approximately (d = 0.16) (Dang et al., 2021).

This showed why a universal-null conclusion would also be premature. A small aftereffect could appear under some conditions. But its magnitude was far below the early estimate, and the post-exclusion estimate depended on decisions made about which observations to retain.

A third major project, led by Vohs and colleagues, recruited 3,531 participants across 36 laboratories. It used two procedures intended to represent paradigmatic tests of ego depletion. After the preregistered exclusions were applied, the confirmatory estimate was (d = 0.06) and was not statistically significant. Its confirmatory Bayesian analysis favoured the point null over the informed alternative specified by the researchers. An exploratory analysis retaining the full sample produced (d = 0.08), with the Bayesian comparison offering approximately equal support to the two hypotheses (Vohs et al., 2021).

Across these three projects, the primary estimates clustered between approximately (d = 0.04) and (d = 0.10)—near zero to small, and far below the early pooled estimate.

The replication record is therefore incompatible with a large, automatic and task-general depletion effect. It remains compatible with small aftereffects under some procedures, effects that depend on particular measurements or several different processes being grouped under the same label.

An effect is not its mechanism

Suppose an experiment finds that people who complete one demanding task perform slightly worse on another. Four different questions still need answers.

First: did the groups differ?

Second: does that difference recur reliably?

Third: what process caused it?

Fourth: did that process involve consuming a finite resource?

A sequential-task experiment primarily answers the first question. With replication, it may contribute to the second. It does not answer the third or fourth automatically.

The initial task may change many things. It may be boring, frustrating or tiring. It may redirect attention, alter the value of rewards, change expectations or make continued effort feel less worthwhile. Participants may infer what the experiment expects from them. They may learn a strategy that helps or interferes with the next task.

Resource depletion is one possible explanation among several.

This distinction is easy to miss because the phenomenon and theory came to share a name. A poorer second-task result was called an ego-depletion effect, while ego depletion also referred to the proposed loss of self-control strength. The result therefore seemed to contain its own explanation.

A performance decrement is an observation. Depletion is an interpretation of that observation.

The problem with an unnamed resource

Resource explanations become scientifically useful when the resource has properties that can be specified and measured.

What is consumed? How can its available quantity be measured independently of the behaviour it is supposed to explain? By what process does control reduce it? What restores it? Which result would distinguish resource loss from changed motivation or allocation?

In much of the sequential-task literature, the resource was not measured independently; its depletion was inferred from poorer subsequent performance. The presumed depletion was then used to explain that performance.

This creates a risk of circularity.

Calling something a resource can still be useful at a descriptive level. Attention is limited. People cannot execute every demanding process simultaneously. Time and metabolic energy are finite. But a limit on simultaneous processing is not the same as a store that is consumed by one task and therefore unavailable afterward.

A system can allocate processing capacity to one operation instead of another without using up that capacity. Likewise, a person may redirect control because another action has become more valuable, because continued effort has become more aversive or because its expected benefit no longer justifies its cost.

A 2025 experiment tested one straightforward implication of progressive depletion: impairment should increase as time on a control-demanding task accumulates. It did not find the predicted pattern, although that result remains specific to its numerical-Stroop and global–local procedure (De Luca et al., 2025).

Reduced capacity and changed allocation can both lower performance. Only the former requires something to have been consumed.

Why glucose did not solve the problem

One attempt to identify the self-control resource focused on glucose.

The proposal was that acts of self-control consumed enough glucose to reduce the energy available for later control. Some studies linked blood-glucose changes with self-control performance and reported that consuming glucose improved subsequent results.

This appeared to give the resource metaphor a biological substrate. Willpower was no longer merely like fuel. Glucose appeared to be the fuel.

The subsequent evidence did not establish that ordinary acts of self-control consume enough glucose to explain later performance decrements. Physiological analyses questioned whether the required consumption pattern was plausible, while evidential reassessments found mixed results, weak evidential value and indications of selective reporting (Kurzban, 2010; Vadillo et al., 2016).

Some experiments also reported benefits when glucose was present in the mouth without being swallowed. Because it had not yet replenished a systemic energy supply, the finding was difficult to explain as restoration of consumed fuel. Reward signalling or motivational change offered alternative interpretations, although neither was uniquely established by the result.

None of this means glucose is irrelevant to cognition. The brain requires energy, and abnormal glucose conditions can affect performance. Consuming or detecting glucose may also alter behaviour through several pathways.

But glucose affecting performance is not the same as self-control consuming glucose. Improvement following glucose exposure is not proof that willpower fuel was restored.

What may change instead

If self-control does not operate like a tank that steadily empties, why might one demanding task affect the next?

One alternative focuses on motivation and attention.

After sustained control, attention may shift away from signals associated with obligation and toward rewarding alternatives. Continued regulation may become less attractive. A person may retain the capacity to perform while becoming less willing to allocate control to the current task (Inzlicht & Schmeichel, 2012).

A second account focuses on opportunity costs. Control-related processes cannot be directed toward every possible action simultaneously. Allocating them to one task prevents their use elsewhere. As alternative actions become more valuable, maintaining the current allocation carries a growing cost. Subjective effort may partly reflect this comparison and encourage reallocation (Kurzban et al., 2013).

In this view, effort does not necessarily signal that a resource is disappearing. It may signal that continuing the present action has become less worthwhile relative to other possibilities.

Fatigue and boredom may also contribute. A repetitive task can become unpleasant or difficult to sustain. Participants may change strategies, disengage or attend to something else. Earlier experience can also alter expectations about the next task and how much effort seems justified.

These explanations are not interchangeable. Motivation-and-attention models, opportunity-cost accounts, fatigue processes and learning mechanisms make different claims. Several might operate within the same episode, but their possible coexistence does not establish one unified replacement theory.

The evidence warrants considering explanations beyond automatic resource consumption. It does not establish one alternative as the universal mechanism.

Beliefs can matter without making the evidence simple

Beliefs about willpower offer another possible influence.

Influential experiments suggested that people who believed willpower was limited showed stronger depletion-like patterns, while those who viewed it as less limited were less affected (Job et al., 2010).

There are plausible reasons for such an effect. Beliefs can shape expectations, effort, interpretation and persistence. If people expect earlier exertion to leave them unable to continue, that expectation may influence how they respond to later difficulty.

Later preregistered evidence has not consistently reproduced the original moderation. In the Dang multilaboratory project, support varied across outcomes and participant-exclusion decisions. Carruth and colleagues' preregistered replication did not reproduce the predicted interaction between depletion condition and willpower mindset (Carruth et al., 2023).

Beliefs may matter in some self-control episodes. They have not been established as a universal switch that turns depletion on or off.

This does not make the experience imaginary. Expectations, valuations and motivational changes are real psychological processes. The unresolved question is when and how reliably they affect later performance.

What survives the controversy

The early literature made ego depletion appear larger and more general than the strongest later evidence supports. Major preregistered collaborations have instead produced estimates near zero or in the small range, with results varying across procedures and analytic decisions.

Small sequential-task aftereffects may occur under some conditions. It remains unclear whether they form one coherent phenomenon or whether the same label has been applied to several distinct processes.

Most importantly, an aftereffect does not establish a finite-resource mechanism. The evidence does not show that ordinary acts of self-control steadily consume a common store of willpower. Glucose has not been validated as that store. Motivation, attention, valuation, fatigue, learning and opportunity costs provide plausible alternatives, but no single account has been shown to explain every result.

The strongest defensible conclusion is therefore narrower than either "willpower is fuel" or "ego depletion never happens":

Prior exertion can sometimes alter later self-control performance, but the effect is smaller, less general and more method-dependent than the original strength model implied. The evidence does not establish that self-control consumes a finite internal resource.

Metaphors guide attention—and can mislead it

Metaphors are not merely decorative. They shape the questions we ask.

If willpower is fuel, failure suggests an empty tank. Recovery suggests refilling it. Effort becomes evidence of consumption, and poorer performance becomes evidence that little remains.

A different metaphor would direct attention elsewhere. Control might be understood as allocation, negotiation or changing priority. The same performance decline would then raise different questions: What became more costly? What alternative became more valuable? Where did attention move? What changed in the expected value of continuing?

None of these metaphors should substitute for mechanism. Their value lies in the hypotheses they make visible—and the assumptions they expose.

A person can feel drained without possessing a measurable store of willpower that has run out. The experience deserves explanation, but it does not provide that explanation by itself.

Feeling as though a resource has been used up tells us that control became different. It does not tell us what changed.