Ego Depletion

What it is

In 1998, at Case Western Reserve University, a laboratory baked chocolate chip cookies in the room before the participants arrived. Two dishes sat on the table: a stack of cookies with chocolates beside them, and a bowl of red and white radishes. A third of the sixty-seven students were told they could eat only the radishes. The experimenter left and watched through a one-way mirror to check that they did. Everyone was then given geometry problems that had no solutions.

The radish group gave up after 8.35 minutes. The students allowed the cookies lasted 18.90, and those who skipped the food stage lasted 19.40.

Roy Baumeister, Ellen Bratslavsky, Mark Muraven and Dianne Tice named the thing the radish group had spent ego depletion, and published it in the Journal of Personality and Social Psychology. The model is intuitive to the point of feeling obvious: self-control behaves like a muscle, it tires with use, and the tiredness carries across unrelated tasks. Two claims travelled from there into general circulation. First, that a day of small acts of restraint leaves you measurably weaker by evening. Second, in Baumeister and John Tierney's Willpower (2011), that the resource is glucose and a sugary drink restores it.

The evidence base for both was a 2010 meta-analysis by Martin Hagger and colleagues, covering close to 200 experiments, which put the effect at d = 0.62. That is a medium-to-large effect, and it is large enough to organise advice around.

Then the methods changed. In 2014 Evan Carter and Michael McCullough reanalysed Hagger's own dataset using techniques that detect and correct for small-study effects, and published the result in Frontiers in Psychology. They found very strong signals of publication bias and concluded that once corrected, the true effect could be zero. No new data was collected. The same numbers, better methods, a different answer.

In 2016 Hagger ran the test on his own finding. A registered replication across 23 laboratories, preregistered, standardised protocol, 2,141 participants, returned d = 0.04, with a confidence interval running from −0.07 to 0.15. In 2021 Kathleen Vohs and colleagues ran the largest version yet, 36 laboratories and 3,531 participants, and got d = 0.06; their confirmatory Bayesian analysis found the data four times more likely under the null than under the effect.

One result cuts the other way, and most popular coverage leaves it out. Also in 2021, Junhua Dang and colleagues ran a 12-laboratory replication with 1,775 participants and found a small but statistically significant effect, d = 0.10, rising to 0.16 after excluding participants who appeared to be answering at random. So the accurate summary is not that ego depletion is zero. It is that the effect sits somewhere between zero and 0.16 against an original estimate of 0.62, which is a different and less satisfying sentence than the one usually written.

In effect

The glucose claim, which is the part that reached the public

Of everything built on ego depletion, the sugary drink travelled furthest. It is the reason someone once told you to eat before a difficult conversation, and it appears in Willpower as practical advice. It has not held up. Miguel Vadillo, Natalie Gold and Magda Osman ran a p-curve analysis of the glucose literature in Psychological Science in 2016 and found its evidential value weak. Separate experiments testing glucose drinks and mouth rinses against depletion found nothing; Lange and Eggert's 2014 paper in Appetite is titled, with some feeling, "Sweet delusion". There is also a physiological objection that never went away, which is that the quantity of sugar involved is far too small to change how a brain works.

Decision fatigue, and everything downstream of a number

The commercial afterlife of this finding is larger than the finding. Decision fatigue slides in consultancy decks, productivity systems organised around spending your willpower early in the day, hiring and scheduling advice about when to put the hard conversation, and a general background assumption that a day of small choices leaves a person measurably worse at the next one: all of it needed the effect to be roughly the size the 2010 meta-analysis said it was. At d = 0.10 there is nothing to schedule around. The advice did not become wrong so much as it stopped having a mechanism, and almost nobody who repeats it has noticed the difference.

Baumeister has not conceded, and the objection is real

In February 2022, Baumeister and Tice published a piece in the Scholarly Journal of Psychology and Behavioral Sciences under the title "Ego Depletion is the Best Replicated Finding in All of Social Psychology". They set out four criteria for calling an effect well replicated, and on the hardest of them, multi-site replication, they point to Dang's twelve-lab study as a clear success. In 2022 Baumeister and Tice published methodological objections to both replications, one concerning the fatigue check in the Hagger protocol and one concerning the sample used in the Vohs analysis; the replication reports set out their methods in full, and both positions are on the record.

Grant the whole of that and what remains is a real effect of about a tenth of a standard deviation under some conditions. That is a defensible scientific position and a useless practical one. It is also worth noticing what the argument does not reach: it says nothing about willpower mindset, the belief-based theory brought in to replace the resource model, which rests on separate evidence and is in worse shape.

What it does not say

It does not say that self-control is fake, or that people do not get worse at resisting things over a long day. Something happens; the question is what, and how big. Michael Inzlicht, Brandon Schmeichel and Neil Macrae proposed in 2014 that what looks like exhaustion is a motivated shift in priorities, attention and motivation moving away from the tasks you have to do and toward the ones you want to do, with nothing consumed. That account fits the same data and had almost nothing built on it, which may be because there is no drink to sell with it.

It does not license the phrase "ego depletion has been debunked", which appears constantly and overshoots. Three preregistered multi-laboratory tests do not agree with each other, one of them found a small significant effect, and the model's originators are still arguing. "Debunked" is a cleaner word than the evidence supports.

It does not transfer to the belief version by default. A reader who learns that the resource model failed and concludes that the belief model must therefore be right has made the same move the field made, and the evidence for the second theory is thinner than for the first.

And it does not tell you what to do instead. The practical conclusion that survives is narrower and duller: prefer interventions whose value does not depend on which theory of self-control turns out to be correct. Precommitment passes that test because removing a trigger never asks self-control to do anything. Telling people their willpower is unlimited does not.

---