Decision fatigue: why the tenth choice is worse than the first

The claim, stated no stronger than the evidence allows

Deciding costs something. Work through a run of choices and the later ones are not made the way the early ones were — the usual drift is toward the default, or whatever needs no justification. How much they degrade is genuinely disputed. The financial consequence does not depend on the size.

That last sentence is the reason this article exists rather than the usual one. The popular telling of decision fatigue is much stronger than the research supports, and several of its most-quoted results have failed to reproduce. Repeating them would be easy and would make for a better story. It would also be one more confident retelling of something the evidence does not carry, with a citation attached to make it look settled.

So the piece is built in two halves. The first goes through what was actually found and where it broke, because a reader who has met the confident version deserves to know what happened to it. The second makes an argument about money that survives either verdict — it needs only that deliberating is effortful, which nobody disputes, and that when deliberating is expensive the option requiring no reason gets chosen more often.

The application is narrow and practical. A savings plan assembled from many small monthly choices is spending a resource it never listed anywhere. The allocation can be right and the plan still come apart, because what breaks is not the allocation — it is month seven asking for the decision again, in a week with no room for one.

What was found, and what happened to it afterwards

The founding experiments are Roy Baumeister, Ellen Bratslavsky, Mark Muraven and Dianne Tice, writing in the Journal of Personality and Social Psychology in 1998. People who had just performed an act of self-control did worse on an unrelated task that also demanded self-control. The authors proposed that a single resource is drawn on by both, and called running it down ego depletion.

Kathleen Vohs and colleagues, in the same journal in 2008, made the version that matters here. Their subjects were not resisting temptation but simply making choices, and the same downstream weakening followed. That is the narrower claim the phrase “decision fatigue” names: choosing is itself the expensive act, independent of whether anything was resisted.

Then the replication record arrived, and it is not kind. A preregistered replication run across many laboratories at once, reported by Martin Hagger and colleagues in Perspectives on Psychological Science in 2016, produced a pooled estimate small enough to be indistinguishable from zero. A preregistered multisite test published in 2021 found an effect in the reported direction but small enough that people continue to argue about whether it means anything outside a laboratory. Neither result says the original researchers did anything improper. Both say the size of the effect is nowhere near what the popular retelling claims.

The most-quoted piece of evidence has its own problem. Shai Danziger, Jonathan Levav and Liora Avnaim-Pesso reported in PNAS in 2011 that the proportion of favourable parole rulings declined across a sitting and recovered after a break. It is a wonderful story and it travels well. It also drew two serious objections: a 2011 letter in the same journal from Keren Weinshall-Margel and John Shapard showing that cases were not ordered randomly within a session, and a 2016 paper by Andreas Glöckner in Judgment and Decision Making arguing that the reported swing is far too large for the mechanism proposed. Which is why no proportion from that study appears anywhere in this article, and why any version you meet that leads with one is worth reading suspiciously.

The adjacent literature on having too many options went the same way. Sheena Iyengar and Mark Lepper reported in 2000, again in the Journal of Personality and Social Psychology, that a larger display of jams in a supermarket drew more interest and fewer purchases than a smaller one. A meta-analysis by Benjamin Scheibehenne, Rainer Greifeneder and Peter Todd in the Journal of Consumer Research in 2010 pooled the studies that followed and found a mean effect close to zero, with wide variation depending on the setting.

Here is the honest summary, and it is a judgement rather than a finding. What is left standing is not a battery that runs down on a schedule. It is something duller: a sequence of choices is not free, the cost is context-dependent and probably much smaller than advertised, and the failure — when it appears — has a consistent direction. That direction is the whole of the useful part.

Why the failure runs toward the default

Take the contested finding away entirely and the direction still follows, from something structural rather than psychological.

Choosing anything other than the existing arrangement requires a reason you could state if asked. Keeping the existing arrangement requires none. That asymmetry is fixed: it is a property of how the options are framed, not of anybody's stamina. So whenever deliberating becomes expensive — a heavy week, a difficult month, a decision taken at the end of a long day, or genuine fatigue if it exists — the option that costs nothing to justify gets relatively cheaper than the ones that do.

William Samuelson and Richard Zeckhauser documented the resulting preference in the Journal of Risk and Uncertainty in 1988, and gave it the name status quo bias: an option chosen more often once it is labelled as the existing state than when the same option carries no such label. Nothing about the option changed. Only its position in the frame did.

This is where the practical point sits, and it is worth stating as flatly as possible. The interesting question is never how depleted anyone is. It is what the default does while no decision is being made. Call it the named failure mode of this whole topic: whoever set the default gets the outcome — and a default nobody set deliberately still produces one.

Consider the two ends of that. Money that is meant to be invested each month but is moved by hand has a default of staying in the savings account, so a skipped decision is a decision to hold cash. A portfolio that drifts has a default of whatever weights the market has produced, so a skipped rebalance hands the allocation to the asset that ran hardest. Neither of those is a choice anyone made. Both are outcomes somebody now owns.

Note what this argument does not require. It does not require that people get tired in any measurable way, and it makes no claim about how any particular reader behaves. It needs only that some decisions cost more to make than others, and that the cheapest one is always available.

A plan has a decision cost, and almost nobody prices it

Two plans can have the same expected return, the same charges and the same asset mix, and be entirely different products — because one demands four decisions a year and the other demands fifty.

That count is a real property of a plan, and it appears on no factsheet. Every document a household is given describes what the money does. None describes what the arrangement asks of the person, which is the part that determines whether it survives a bad quarter at work. So the reframe worth taking from this article is to score a plan by decisions, alongside everything else already being scored.

Set the common arrangements out with the default written down next to each. The last two columns are the ones that matter.

Read the fourth column down the page and the pattern is unmissable. Every row where the answer is “nobody” is a row where an outcome is being produced month after month by an arrangement that was never chosen. The rows with an owner are not better because the person was more disciplined. They are better because the decision was made once, in advance, when there was room to make it.

The idle-cash row deserves singling out, because it is the one that hides. Money waiting for a decision looks like prudence and behaves like a permanent allocation to the savings rate. It is also the row that most often carries a second label — “this is for the house” — which is mental accounting doing the work of a plan. The two mechanisms stack, and the money sits there through both.

A missed instalment is not a random sample of instalments

Now the arithmetic, because the size of this is easy to underrate and easy to check. Every input below is illustrative and stated so it can be redone with different ones.

Take ₹10,000 a month moved by hand into an equity scheme, for ten years. That is 120 separate decisions, and ₹12 lakh if every one of them happens. Suppose two months in every twelve get missed — not through any decision to stop, just through the month going the way months go. The amount invested is ₹10 lakh rather than ₹12 lakh, and if the missed months fall evenly across the decade the final corpus is short by close to the same one-sixth — evenly spread skips remove instalments of average vintage, so what is lost is roughly in proportion. Where the skips bunch early the shortfall is larger than a sixth, since each instalment compounds on its own from the day it is made and the early ones had the longest to do it.

That is the mechanical part, and it is the smaller half of the problem. The larger half is that the missed months are not drawn at random.

An instalment that requires a decision can be skipped. An instalment that requires none cannot. So the months that go missing are, by construction, the months in which the transfer became a judgement call rather than a chore — and something has to make it a judgement call. A week with nothing left in it will do. So will a falling market, which converts “move the money” into “is this the right week to move the money”, a question the same transfer never asks in a rising one. Whether that particular pattern holds in any given household is an empirical matter this article is not measuring, and it is offered as a reading rather than a result. The mechanism behind it is not in dispute: what a falling market does to the felt cost of an outgoing payment is the subject of loss aversion, and the arithmetic of stopping during a drawdown is worked through in SIPs during market crashes.

Put the two halves together and the shape of the cost is clear. A plan that asks for a decision every month is not merely losing some instalments. It is losing a non-random selection of them, weighted toward the months when deciding was hardest — and the price of a missed instalment is not the instalment. It is the instalment plus everything it would have earned over the years still to run.

Automation, and the thing it costs you

The evidence in this area that has held up best is not from a laboratory. It is from payroll records, and it points at defaults rather than at willpower.

Brigitte Madrian and Dennis Shea studied a single employer in the Quarterly Journal of Economics in 2001, after it switched its retirement plan from “sign up if you want in” to “you are in unless you opt out”. Participation rose sharply. More telling, the contribution rate and the fund selection clustered on whatever the plan had nominated as its defaults — people enrolled without choosing largely did not choose afterwards either. Sheena Iyengar, Gur Huberman and Wei Jiang, writing in a 2004 volume on pension design, reported the companion direction: as the number of funds on offer rose, participation fell.

Richard Thaler and Shlomo Benartzi took the next step in the Journal of Political Economy in 2004 with a programme that has the mechanism in its title, “Save More Tomorrow”. Employees committed in advance to raising their contribution rate, with the increases tied to future pay rises. The decision was moved to a moment when it was cheap to make, and the money was moved before it was ever seen.

India already runs both structures, and the reason they work is the same in both cases. A SIP executes on a standing instruction to the bank, so the monthly instalment is not a decision — skipping it requires an act, which reverses the asymmetry from the previous section entirely. Contributions to the provident fund are deducted at payroll, so the money is never in an account from which a decision could divert it. The National Pension System nominates a lifecycle option for subscribers who do not pick one. Each of these is the same manoeuvre: settle it once, in advance, and let the arrangement be the thing that persists.

And now the cost, because there is a real one and it is usually left out. Automation removes the monthly decision, and it removes the monthly attention along with it. A mandate running into a scheme that stopped matching its purpose four years ago runs exactly as faithfully as one that still fits. The same property that stops you skipping a good instalment stops you noticing a bad arrangement, and an allocation automated at the wrong weights is an error that now compounds without supervision.

So the trade is not deciding less. It is trading twelve decisions for one — a scheduled review, on a date chosen in advance, at which the arrangement is examined rather than the instalment. A household that automates the contributions and never books the review has not solved the problem. It has replaced a plan that decays visibly with one that decays quietly, which is worse in exactly one respect: nothing tells you.

Four checks with answers you can look up

None of these needs any position on whether ego depletion is real. Each one has an answer that is a number or a date, not a self-assessment.

1. Count the decisions your current arrangements demand in a year. Transfers made by hand, allocations reconsidered, cash awaiting a destination, the rebalance that keeps not happening. That count is the first honest measure of whether a plan can survive a bad quarter.
2. Write down each default — what happens if you do nothing at all for six months. Every arrangement has one, whether or not anybody chose it. The ones where the answer is “the money stays in the savings account” are the expensive ones, and they are also the ones that feel like nothing is wrong.
3. Check when the decision falls. A monthly transfer scheduled for the day after payday competes with nothing. The same transfer left until the end of the month competes with everything, and the order in which choices arrive is something you can set rather than something that happens to you.
4. Book the review before automating. An automated contribution without a scheduled examination is not a finished plan, it is an unsupervised one. A date in a calendar converts the recurring decision into a single one, which is the whole point of the exercise — goal-based investing gives that review something to check against.

What none of this settles is what to conclude at the review. Two households running the same four checks will reach different answers about the same standing instruction, because one is nine months from a stated goal and the other is nineteen years from one. The checks guarantee only that the decision was made deliberately once rather than avoided repeatedly. That is the entire contribution, and it is worth more than it sounds.

Where deciding stops and arithmetic starts

Everything above turns on the same property: these arrangements persist in the absence of a decision, not because a bad one was made. Which means the useful move at a scheduled review is arithmetic rather than resolve.

The question a review has to answer is narrow. What did the contributions that actually went in produce, measured from the money that went in — not from a peak, not from an entry price, and not from what a comparable scheme did over a period chosen after the fact.

FNOTrader's Mutual Funds app runs a contribution schedule against the full published record of daily per-unit prices — the net asset value, or NAV — kept by AMFI, the mutual fund industry body, around 34 million rows of it. It reports invested against value, the worst peak-to-trough fall along the way, and the return measure built for money arriving on irregular dates, XIRR. Past performance is a record of what happened, not an indication of what will happen.

What that does for the mechanism in this article is narrow and worth stating precisely. A review with a computed answer takes one decision, once a year, and hands the other eleven months back to the standing instruction. FNOTrader is not a SEBI-registered investment adviser and does not give investment advice. The broader set of mechanisms this article sits inside is mapped in behavioural biases in investing, and the account of how a falling market changes the felt cost of acting is in emotional investing.

Common questions

Continue reading

More in Behavioural Finance · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article