Decision Fatigue: The Effect Psychology Could Not Replicate
The Compass Team
November 6, 2026
The sixth product call of the day is where a founder stops auditing and starts approving.
The deck says 12% conversion. Last month's deck said 14%. You already asked why on the second call, got an answer you accepted, and moved on.
Now the same question is back in a different room with a different number, and you don't remember what you concluded. So you say "looks reasonable" and you move on.
Everyone has a name for this hour. Decision fatigue. A tank of willpower, full at breakfast, drained by dinner.
That tank is a metaphor carrying less evidence than the metaphor implies.
In 1998 Roy Baumeister and three colleagues ran the study that started it. People sat in a room with warm cookies. One group ate a cookie. Another group was told to eat the radishes and leave the cookies alone.
Then everyone attempted an unsolvable puzzle. The radish group quit sooner. Baumeister's reading: self-control behaves like a muscle, and resisting dessert spent some of it.
Twelve years later a meta-analysis of 198 published experiments put the effect at d = 0.62, medium to large (Hagger, Wood, Stiff and Chatzisarantis, Psychological Bulletin, 2010). That number went into books, conference stages, and a decade of productivity advice.
Then the replications arrived.
Carter, Kofler, Forster and McCullough reran the meta-analysis in 2015 with published and unpublished studies and a correction for small-study effects. The pattern was textbook inflation: the bigger the sample, the smaller the effect. Once corrected, they found "scant evidence that the depletion effect is distinguishable from zero."
In 2016 Hagger ran the Registered Replication Report himself. Twenty-three laboratories. More than 2,000 participants.
The same two-task design, preregistered. No reliable difference between the depleted group and the control. A multisite preregistered test published in Psychological Science in 2021 found nothing either.
The debate isn't closed in the way you'd want. Friese and colleagues, writing in Personality and Social Psychology Review in 2018, concluded that after several hundred published studies "the available evidence is inconclusive" for either side.
Lurquin and Miyake argued the field has a conceptual crisis underneath the replication one: across studies, self-control means inhibitory control, or emotional control, or any monitoring of behaviour at all. A construct defined six ways can't be tested once.
Still, nothing in that literature measures the thing a founder loses at 6pm.
Watch the call again. The judgement is intact. Nobody gets dumber at dinner.
What's gone is the record of the second call: the 14%, the reason the pricing test died, the assumption behind the hire. Without it the question has to be argued from scratch, and the argument costs the same at 6pm as it did at 9am.
What costs you the hour is re-deciding.
Re-deciding has one property a fuel tank doesn't: it is immune to rest. Nine hours of sleep leaves an unwritten decision exactly where you left it. You will spend Monday rebuilding Thursday.
The repair is small and unglamorous. Write the decision in the hour it's still open. Six lines: what we decided, what we assumed, what the number was supposed to do, what it did, what changed, what we would tell ourselves next time.
Then the sixth call becomes a read of those lines instead of a reconstruction. 12% against 14% answers itself, because the assumption sits on the page with a date on it.
The radish study stays in my head for a different reason now. It is the cleanest example I know of the most confident story winning by default.
The cookie story got the book deal. The failed replication got a footnote.
Decisions work the same way. The ones you write down survive the quarter. The ones you don't get retold, a little differently each time, until the retelling is the record.
Share this article