Choice Overload: Why More Options Produce Worse Decisions

Choice overload is the finding that past a point, adding options makes people less likely to decide at all, and less content with whatever they do pick. Deciding less often and liking the result less are separate findings, not one. The mechanism underneath both is arithmetic. The work of comparing grows as the square of the number on offer, and the attention available to do it does not grow at all.

What is choice overload, and is more choice not obviously better?

Picture the cooking oil shelf in a large store. Forty containers, eleven brands, four sizes each, three of them on some sort of offer, and every label carrying a different claim about what it is good for. Now picture the stall at the end of the street, where the same shopper is offered three. Ask which shopper walks out holding oil. The store has more of everything, so most people guess the store. Watch the two of them for ten minutes and the answer inverts. The stall shopper buys. The store shopper reads four labels, puts two containers down, picks up a third, decides to look it up later, and leaves with the rest of the shopping done and no oil in the bag.

Nothing went wrong there. Nobody was careless, nobody was rushed, and the store was not being unfair to anybody. The shopper simply met a menuThe set of options actually put in front of somebody who has to choose. larger than the attention available to work through it, and the result is regular enough to have a name and a measurement behind it. Choice overload is the finding that past a point, adding options stops helping and starts obstructing, and the point arrives sooner than almost anybody expects.

The finding is due to Sheena Iyengar and Mark Lepper, in their paper When Choice is Demotivating, published in the Journal of Personality and Social Psychology in 2000. Their setting was ordinary in the same way the oil shelf is ordinary: a display of things to taste, offered in two sizes of assortment, with the purchase recorded afterwards. The larger display attracted more people to stop and look. The smaller display produced far more purchases. Attracting a crowd and producing a purchase are not in tension, and holding both results at once is most of the work.

The objection that arrives first is a good one and deserves to be taken seriously. A larger set contains the smaller set inside it, so more choice is obviously better, the argument goes. Everything that was on offer before is still on offer. Nobody is being forced to look at the extra items. If the best thing for a given chooser was the third container, it is still there among the forty, so the largest achievable outcome can only have improved or stayed the same. The argument is valid. The argument also rests on an assumption that is doing all of the work, and the assumption is that comparing is free.

Comparison is not free. Comparing costs attention, and attention is the one input in this whole picture that does not scale. A chooser facing forty containers has exactly the same working memory as a chooser facing three, and roughly the same number of minutes to spend, and no way of buying more of either. The extra options arrive as work rather than as opportunity, and work is subtracted from the same fixed budget the choosing itself has to come out of. Once that is in view, the rest is arithmetic.

What are the two effects, and why are they two rather than one?

Almost everybody who has heard of this finding has heard of half of it. The famous half is that people faced with too much on offer choose less often. The unfamous half is that among the people who did choose, satisfactionHow somebody feels about the choice they made, which is a different thing from how good that choice actually was. with the choice falls as the number of options rises. Choosing less often and being less content are two separate results with two separate mechanisms, and merging them into one slogan about too much choice quietly throws away the more useful of the two.

Take the first effect on its own terms. Faced with a set too large to work through, a chooser does not usually pick badly. A chooser postpones. In the language of the research this is deferralPostponing or declining a decision rather than making it. Deferral is a choice with no visible result, and it is easy to miss., and its distinguishing feature is that it leaves no trace. The person who buys the wrong oil comes home with the wrong oil, and somebody can eventually notice. The person who buys no oil comes home with nothing, and there is nothing to notice at all. Sheena Iyengar, Gur Huberman and Wei Jiang carried the same effect into a setting with real money attached in How Much Choice is Too Much, published in 2004, where the thing on offer was participation in a retirement arrangement and the size of the assortment was the number of investment options in it. Participation fell as the number of options rose.

The second effect is stranger and, for the person assembling the options, more uncomfortable. Satisfaction among those who did decide also falls as the assortment grows. The claim needs care. Falling satisfaction does not mean those people chose worse. The same choice feels worse to the person who made it, and the option itself can be entirely sound. Fewer decisions and worse satisfaction are two findings with two mechanisms, and an account that runs them together has lost the one that surprises people.

The invented Palash decision log carries a small illustration of the first effect. In the log, 30 people were shown 214 options and 11 of them chose, being 36.7 per cent. Another 30 were shown 7 options and 21 of them chose, being 70.0 per cent. The gap is 33.3 percentage points, and the smaller assortment produced 1.9 times as many decisions. One invented cohort is not evidence that anything works; it is a way of holding the shape of the finding still long enough to be examined.

Why does the work of comparing grow as the square?

Here is where the whole thing becomes countable, and countable is much better than memorable. The unit of work in choosing is not the option. The unit is the pairwise comparisonWeighing one option directly against one other. No smaller piece of comparing exists.: one thing weighed directly against one other thing. Better and worse are relations between items rather than properties of them, so a set cannot be ranked by looking at each item once in isolation.

Counting the comparisons in a set of size n gives n times n minus 1, all over 2. The reasoning is short. Setting each of the n options against each of the other n minus 1 gives n times n minus 1 pairings. Every pairing has now been counted twice over, once from each side, and halving the product removes the duplication. Nothing in that argument is about finance, or about shopping, or about people: it is a fact about sets, and it holds wherever anybody has to rank things.

The series is the teaching. Now work it.

Every count in that table is a ceiling rather than a description. A ceiling is what weighing every pair exhaustively would cost, not what any person actually performs. The qualification matters enough to be taken up on its own once the arithmetic is in place.

The last column is the one that hurts. Read it rather than the third. At two options, the chooser does half a comparison per option. At fifty, twenty four and a half. The chooser has not merely been given more to do. Each individual item on offer has itself become more expensive to consider, and every one of them has to be held against everything else. Work of that shape is what quadraticGrowing in proportion to the square of a quantity, so doubling the quantity roughly quadruples the result. means: doubling the options does not double the work, it roughly quadruples it.

Why does the collapse feel sudden rather than gradual?

People who describe a large assortment as overwhelming almost never manage to say which option was the problem, and they are not being vague. There genuinely is no such option. The difficulty does not live in any item; it lives in the relations between items, and relations are invisible when things are looked at one at a time.

Here is the arithmetic of that experience. Go from a set of n to a set of n plus 1, and the number of comparisons rises by exactly n. Adding the fifth option to four costs four new comparisons. Adding the eleventh to ten costs ten. Adding the twenty first to twenty costs twenty. Adding the fifty first to fifty costs fifty. Each additional option is added one at a time and costs more than the one before it, so the difficulty accelerates while the thing being added looks identical every time.

The arithmetic explains why the experience changes abruptly. Nobody feels the difficulty of a set rising smoothly. The person doing the adding is counting options, and the burden is counting pairs. Somewhere between the two counts, the chooser crosses from a set they can hold in their head to one they cannot. The last option added looked exactly like the first, so the crossing arrives without warning.

What happens between four options and fifty?

Set the two ends of that table beside each other and the point stops being abstract. Four options require 6 comparisons at most. Fifty options require 1,225 at most. Divide 50 by 4 and the options have multiplied by 12.5. Divide 1,225 by 6 and the comparisons have multiplied by 204.2, to one decimal place.

The absolute version is the one people remember. Say the same thing that way. Going from four options to fifty adds 46 options and 1,219 comparisons. Nobody ever decided to add one thousand two hundred and nineteen units of difficulty to a decision; they decided to add forty six things somebody might like, and the difficulty came along uninvited. Both of those counts are ceilings on the work rather than accounts of what anybody does, and the difference between a ceiling and a description is taken up under what the figure of 1,225 describes. Even so, the ceiling decides whether a set can be worked through at all, and its shape is what matters. The gap between the two multiples is what makes the size of a set a conduct question rather than a question of taste. The person assembling the set and the person facing it are not looking at the same object.

What does the figure of 1,225 actually describe?

The comparison count is an upper bound on the work of choosing. An upper bound is what an exhaustive weighing would cost. No human being facing fifty options performs 1,225 comparisons, and asserting otherwise would describe a machine rather than a person.

Real choosers eliminate in groups. Facing fifty containers of oil, a shopper does not begin comparing. A shopper begins discarding: everything above a certain size, everything in an unfamiliar brand, everything without the one property they came in for. Forty of the fifty leave the picture in a few seconds, on a rule the shopper could probably not articulate, and the actual weighing happens among the survivors. Ten survivors is 45 comparisons, and 45 is a decision somebody can genuinely make.

The elimination is itself a decision, taken quickly, on one attribute, and recorded nowhere. That matters more than it sounds. The chooser who discarded forty options never compared them against anything and cannot say why they went, so if the best option for them was among the forty, nothing in the process would ever surface it. The upper bound is not describing the work done. The bound describes the work avoided, and the avoiding is where the quality of the outcome is actually decided.

What do people do when comparing stops being feasible?

If comparing every pair is out of reach and eliminating has already thrown most of the set away, something still has to determine the final answer. One rule that reliably appears is the even splitDividing equally across whatever options happen to be present, rather than in proportion to anything about them.: dividing equally across whatever is in front of the chooser.

The even split shows up everywhere once it is looked for. A household with four children and one box of sweets divides by four. A wedding budget with five heads of expenditure and no strong view about any of them gets a fifth each and is adjusted later when somebody complains. The rule requires no comparison at all, only that the options can be counted. Requiring nothing is precisely its appeal.

The invented case opens exactly that way. On 4 January, Meera Sundaram opens a holding of Rs 12,00,000/- as four positions of Rs 3,00,000/- each. Divide Rs 3,00,000/- by Rs 12,00,000/- and each position is 25.0 per cent, one quarter, one divided by four. Four options, four equal shares, no weighing performed. The evenness did not last: on 19 February a television segment named Suvarna Chemicals Limited and Rs 1,00,000/- went into it the same evening, taking that position to Rs 4,00,000/- and the total cost to Rs 13,00,000/-. At that point the shares are 30.8 per cent for Suvarna Chemicals Limited and 23.1 per cent for each of the other three. The even split has gone, and nobody decided the new allocation either.

The even split is not laziness, and reading it as laziness is the mistake worth avoiding. It is the sensible answer to a comparison problem that has stopped being solvable in the time available. Where the options genuinely cannot be ranked, dividing equally is the one rule that treats that ignorance honestly. The rule declines to express a preference nobody holds. The cost is that the result depends entirely on what happened to be on the list. Putting three of one kind and one of another in front of somebody makes the even split deliver seventy five per cent of one kind, and nobody chose that.

Why does satisfaction fall even for somebody who chose well?

Now the second effect, the one that repays the closest attention. Among people who did decide, and decided sensibly, contentment with the decision falls as the set they chose from grows. The option is fine. The feeling about the option is worse. Two mechanisms have been proposed. Both are worth carrying, and they behave differently.

The first is that a larger set makes the forgone alternativeAn option that was available and not taken. A larger set produces more of them, and makes each one easier to picture. vivid. A chooser picking from three has turned down two, and can hold both of them in mind without effort. A chooser picking from fifty has turned down forty nine, several of which had something the chosen one lacked. Each of those becomes an easy thing to imagine, and imagining it is what does the damage. The comparison is no longer against nothing. The comparison is against a reference pointThe level a result is judged against. Gains and losses are felt relative to it rather than in absolute terms. assembled out of the best feature of each rejected option, and no single real thing can beat a composite made from the best parts of everything else. Daniel Kahneman and Amos Tversky set out in Econometrica in 1979, in the paper introducing prospect theory, that outcomes are felt as gains and losses against a reference point rather than as final positions, and this is that structure applied to an option list.

The second mechanism is different and does not need any imagining at all. A larger set raises the standard. Offered three options, a chooser expects one of them to be adequate. Offered fifty, a chooser expects one of them to be right. With fifty on the table, the correct answer seems likely to be somewhere among them. An adequate outcome then reads as a failure of searching rather than as a decent result, and the chooser blames the search. The same outcome is scored against a harder standard purely because the set was larger, and nothing about the outcome itself has changed at all.

Where does adding options genuinely help?

The finding has limits, and naming them is not balance for the sake of balance. Left unnamed, the limits turn the finding into a slogan, and the slogan is wrong. There are conditions under which a larger set is straightforwardly better for the chooser, and they can be named exactly rather than gestured at.

The first is that the preference is already specific. Somebody who walks in wanting one particular property, and knows it before arriving, is not comparing anything. Such a chooser is filtering, and filtering grows in a straight line rather than as the square: fifty options means fifty checks, not the 1,225 comparisons an exhaustive weighing would take. For that person the fiftieth option is a genuine extra chance of a match and costs almost nothing to consider.

The second is that the options are sorted on one dimension. If everything on the list is arranged along a single axis and the chooser knows which end of it they want, the set collapses to a position on a line. Ordering is what does this. A sorted list of fifty is easier than an unsorted list of eight. The sentence sounds strange until sorting is seen for what it does: the comparisons have been made in advance and handed to the chooser for free.

The third is that eliminating is cheap. Where most of the set can be discarded quickly on one visible attribute, a large list behaves like a small one. The chooser never faces the large one. The honest version of the upper bound sits here. The count of 1,225 only bites where the options resist being thrown out in groups, and options resist when they differ on several attributes at once, with no single sweep able to remove them.

Put the three together and the risk is not size, it is the combination of size with vague preferences, no ordering and options that all look defensible. That combination is worth naming because it is common, and because each of its three parts can be worked on separately by whoever is assembling the list.

What does the decision log record about this, and what does it not?

The invented Palash decision log covers 240 decisions taken by 60 investors over eight quarters. The log records what was decided, and on 84 of the 240 it records a written reason, being 35.0 per cent. The decisions that were not taken appear nowhere in it, and structurally cannot.

A measurement problem sits underneath the first effect. Deferral produces no row. A log of decisions is a log of the people who got through, and the people the assortment defeated are simply absent from it, indistinguishable from people who were never there. The two small comparisons in the log, 11 of 30 choosing from 214 options against 21 of 30 choosing from 7, only exist because somebody counted the people who were offered the choice rather than the people who took one. Counting the people offered the choice is the whole trick, and outside a comparison somebody set up on purpose, almost nobody does it.

The log also records that over the eight quarters, 9 investors complained in writing and 14 left without saying anything. Albert Hirschman, in Exit, Voice, and Loyalty in 1970, separated exactly those two responses and pointed out that the silent one carries no information back to whoever might fix the problem. A set of options too large to work through produces the silent kind. Nobody writes in to say they found the list difficult. The people the list defeated just do not come back.

What duty does this create for whoever assembles the options?

Devika Rao, the invented adviser at Palash Advisory Services Private Limited, faces a question here that has no comfortable answer. Give somebody four options and she has done the narrowing herself, on their behalf, using judgement they cannot see. Give them forty and she has done no narrowing. Doing none looks neutral and is not: the burden she declined to carry has landed on somebody with less time and less practice at carrying it.

Both moves are choices about the shape of the decision. Shaping the decision is what choice architectureThe arrangement in which options are presented. No arrangement is neutral. Something has to be first, and something has to be left out. means, in the sense Richard Thaler and Cass Sunstein gave it in Nudge in 2008: there is no arrangement that is not an arrangement. An order has to be chosen. No list is complete, and something has to be omitted. Presenting everything is not the absence of a decision; it is a decision to move the work to the other side of the table.

The duty this creates is a duty to know which of the two is being done and to be able to say so, not a duty to arrive at any particular number. The arithmetic gives no number: it gives a shape, and the point at which the shape becomes unmanageable depends on how specific the preferences are, whether the list is sorted, and how cheaply things can be discarded. Specificity, ordering and ease of discarding are facts about a situation, and no general figure can stand in for them.

For a person deciding alone, with no adviser and no committee, the same three levers are available. The first is to make the preference specific before opening the list. Filtering then replaces comparing. The second is to sort on the one dimension that actually matters to that person, and the list collapses to a position rather than a set. The third is to do the eliminating on purpose, in writing. The sweep was going to happen anyway, and writing it down puts it on a chosen rule rather than on whichever attribute happened to be visible first. None of that reduces the number of options. All of it reduces the comparisons.

What the arithmetic does not settle

The arithmetic settles nothing about how many options belong in a set. No number of options is given to anybody, for any purpose, in any situation, and there is no number hidden in the working that a careful reader could extract. The arithmetic explains why the size of a set is a question worth asking. Answering the question would require knowing the chooser, the options and the setting, and the answer would be a recommendation rather than an explanation.

The arithmetic also offers no view on any holding. Meera Sundaram opening four positions on 4 January is an invented illustration of what an even split looks like, and four is not being presented as a right number of anything, nor as a wrong one. The finding here is about the relationship between the size of a set and the work of choosing from it, and every application of it to a real decision is a separate judgement.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log and Suvarna Chemicals Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.