Decision fatigue

Decision fatigue is the idea that choosing costs something, that the cost accumulates, and that by evening you are worse at deciding than you were at nine — which is why you order the same thing, buy the thing at the till, and say yes to something you would have declined that morning.

The experience is familiar enough that the explanation went unquestioned for twenty years. This page exists because the explanation is in serious trouble, and almost every article on the topic still repeats it as settled.

Our position: the tiredness is real and the theory people attached to it is not. That is an uncomfortable thing for a page like this to say, since the clean version makes better advice. But the clean version is built on two studies that did not hold, and passing them along would be the cheap thing to do.

Roy Baumeister's 1998 paper is the origin. Participants who resisted a plate of cookies gave up sooner on an unsolvable puzzle afterwards than participants who had been allowed to eat them. The interpretation: self-control draws on a single limited resource, one act of it depletes the supply, and the next act suffers. The term was ego depletion.

It was a generative idea. Hundreds of studies followed, a meta-analysis in 2010 reported a medium-sized effect, and the model went out into the world as the muscle metaphor: willpower tires like a muscle and recovers with rest and glucose.

Then the field started checking.

The multi-lab replications came back null. Hagger and Chatzisarantis ran a preregistered replication across twenty-three laboratories in 2016 and found an effect indistinguishable from zero. In 2021 Vohs, Schmeichel and a large consortium ran a bigger and more sympathetic test — thirty-six labs, over three and a half thousand participants, with the original authors involved in the design so nobody could say the paradigm had been misapplied. The confirmatory result was again non-significant, and their Bayesian analysis found the data about four times more likely under no effect than under the effect the literature had predicted.

The glucose version did worse. The idea that self-control burns blood sugar, and that a sugary drink restores it, was the most physiologically concrete part of the model and the most widely repeated. Vadillo, Gold and Osman's 2016 p-curve analysis found the supporting literature had little evidential value, consistent with publication and reporting bias. The brain's energy use does not vary nearly enough between an easy task and a hard one for the story to work in the first place.

The famous judges study has a boring alternative explanation. Danziger's 2011 paper — favourable parole rulings falling from around sixty-five percent to near zero across a session and resetting after a food break — is the single most-cited illustration of decision fatigue anywhere. Two things are wrong with using it that way. Weinshall-Margel and Shapard pointed out in a reply that the cases were not randomly ordered: prisoners with legal representation were heard first within each session, and ordering was influenced by other administrative factors. Glöckner then showed by simulation that the dramatic slope can be produced as a statistical artefact, because favourable rulings take longer to deliver than unfavourable ones, so a judge running out of time before a break is mechanically more likely to take the quick option. Neither of these proves the judges were not tired. Both mean the study cannot carry the weight put on it.

Friese and colleagues' 2019 review is the fairest summary of where this landed: not "ego depletion is definitively false", but that after two decades the evidence does not establish that it is real in the form claimed, and the confident textbook version should not be taught.

Several things, and they are worth more than the tank metaphor was.

Fatigue changes what you want to do, not what you can do. The leading replacement account treats the effect as motivational rather than resource-based: after sustained effort, the appeal of continuing drops and the appeal of something easier rises. Which predicts something the tank model does not — that a big enough incentive, or a genuinely interesting alternative, restores performance immediately. It does, reliably, in exactly the experiments where a depleted tank should not have refilled.

Choosing is genuinely more expensive than executing. Deliberating among options draws on the same limited attention that everything else does, and switching between decision frames carries the cost described in attention residue. A day of forty small decisions is harder than a day of four large ones, without any tank being involved.

The time-of-day effect is real and is not about willpower at all. Performance sags in the mid-afternoon on a circadian schedule, in people who have decided nothing all day. That is the afternoon slump, and it explains a fair amount of what gets filed under decision fatigue.

And genuine depletion exists — it just runs on a longer clock. What people mean when they say they have nothing left by Thursday is much closer to burnout and emotional exhaustion than to an afternoon of choices. That literature has held up well.

The practical advice that came out of the tank model mostly still works, which is why it survived the theory collapsing. But the reasons change, and the reasons matter when the advice stops working.

Front-load the decisions that are actually hard. Not because you have a finite daily allowance, but because the afternoon has a physiological dip in it and because the day accumulates residue.

Reduce the number of decisions, not the number of tasks. Standing choices — the same breakfast, a default meeting length, a rule about what you say yes to — remove deliberation, which is the expensive part.

Be suspicious of the evening version of yourself, without moralising about it. By ten at night the pull toward the easy option is strong. That is a preference shift, not a moral collapse, and the design answer is to have decided earlier rather than to try harder later. The entry on revenge bedtime procrastination is where that specific version goes.

And when a claim about willpower comes with a number, check its age. A surprising amount of what circulates in this area is downstream of two experiments that did not survive being run again. We would rather tell you that than hand you a tidier page.