Limited Attention: Buying What You Happened to Notice

Attention is the input every decision needs and nobody has enough of. Attention does not change what a person prefers. Attention changes what ever gets considered. Considering comes earlier than comparing, and it matters more. Something that never reaches notice is not rejected on its merits. The unnoticed thing was never a candidate at all, and a filter on candidates is a quieter and much larger effect than any preference.

Almost everything ever written about deciding well starts at the same moment: the options are on the table, and now a choice has to be made between them. Weighing this against that, scoring them, ranking them, picking. Comparing is a useful thing to teach and it is not where most of the damage happens. The options got onto the table by some process, and that process is doing more work than the comparison everybody argues about. Every individual bias arrives only after that earlier step has already run. The earlier step therefore comes first.

The reason the filter goes first is structural rather than rhetorical. A preference can be argued with. Two things can be put beside each other, a person can be shown that one of them was scored wrongly, and a mind can change on the spot. An option that never reached a person offers nothing to argue about, and so cannot be argued with. The filter is never experienced as running. A shortlist is experienced instead, and it feels like it arrived on its own.

Why is attention the scarcest input to a decision?

AttentionThe limited capacity to take in and hold information while deciding. feels free. Money is obviously limited, time is obviously limited, and both can be watched running out. Attention is the one input that is just as limited and has no meter on it, and having no meter is exactly why nobody budgets for it. Daniel Kahneman set this out formally in Attention and Effort in 1973, treating attention not as a switch that is on or off but as a limited pool that gets allocated across whatever is competing for it. More to one thing means less to something else, whether or not the trade was noticed being made.

Take it out of finance first. A person walks into a vegetable market with forty stalls in it. The shopper will not compare forty stalls. The shopper walks down one side, looks properly at four or five, and buys. Ask them afterwards why they did not buy from the stall at the far end and they will not say it was worse. The answer will be that they did not get to it. The thirty-five stalls they never reached were not judged and found wanting. None of the thirty-five was ever in the running. The shopper walked out feeling they had chosen carefully, and within the five stalls they did reach, they had.

Herbert Simon made the same point about decisions generally in the Quarterly Journal of Economics in 1955, arguing that a real decider works with a small, manageable slice of the world rather than the whole of it. The whole of it will not fit. Scarcity of attention is not a flaw in a particular person; it is a fact about any decider with a finite head and a world that is larger than it. The difference between people is not how much attention they have. The difference is what gets in front of them, and who decided that.

Scarcity of attention matters more the larger the world of options is. A person choosing between two schools for a child can genuinely look at both. A person choosing where to put a monthly saving is facing something in the thousands, and no amount of effort closes that gap. So the interesting question stops being how carefully somebody compared, and becomes what reached them in the first place.

What is a consideration set, and what decides what gets into it?

The consideration setThe options actually on the table when a choice is made. is the short list of options that are actually on the table when a choice gets made. The consideration set sits between two other things. Behind it is the candidate setEverything that could in principle have been chosen at that moment., meaning everything that could in principle have been chosen at that moment. In front of it is the single thing that was chosen. Three stages, and the arithmetic of a decision is usually taught as though the first two did not exist.

Think about a household choosing a caterer for a wedding. There are perhaps two hundred caterers who could do the job. The household will consider four. The four arrived because a cousin mentioned one, a hall recommended two, and one had a board outside a shop the household walks past. Nobody in that household believes they surveyed the market. Rejecting requires having looked, so the household does not experience the other one hundred and ninety-six as rejected either. The four were selected by a process that involved no comparison at all, and the comparison everybody remembers making happened entirely inside that four.

What is Attention Bias, and how is it not the same thing as a preference?

Attention biasA filter that decides what is considered, before anything is compared. is the name for a systematic tilt in what reaches the consideration set. A tilt in how the options are scored once they are there would be a preference. Attention bias is instead a tilt in which ones arrive at all, and a tilt in which ones arrive is a filterAnything that removes options before they are judged on their merits. rather than a ranking. The difference between those two words is the whole of the subject.

One clean test tells a filter and a preference apart, and the test is worth holding on to. Everything a person values is held completely fixed: their tolerance for risk, their view of the future and their sense of what a good holding looks like all stay as they were. Only what happens to reach them in a given week changes, and what they buy is then observed. No preference was touched, so if the choice moves, a preference did not cause the move. The set the preference was applied to is what moved, and a moving set is attention bias in one experiment.

Attention bias is hard to spot from the inside. Asked why they bought a thing, somebody will give a genuine, coherent answer about its merits. The answer is usually true. The answer is simply about stage three of the four stages above, and a question about stage two would have caught the effect. Nobody can report on a filter they did not watch operating, and this is not a failure of honesty. The failure is one of access.

Attention bias is worth defining by what it is not. Attention bias is not a claim that noticed things are worse than unnoticed things. Nor is it a claim about anybody in particular. Cutting thousands down to a handful is a job that has to be done by something, as the section below on when the shortcut is correct sets out, so attention bias is not a criticism of shortcuts in general.

What does Attention-Driven Buying look like in one line of a log?

Attention-Driven Buying is the name for what happens when notice does the selecting: the option that reached a person becomes the option that gets bought, without any step in between that examined the alternatives. Brad Barber and Terrance Odean introduced it in finance in All That Glitters, published in the Review of Financial Studies in 2008, where they set out both the buying pattern and the asymmetry between buying and selling that is taken apart below.

The Palash decision log, an invented record of 240 decisions taken by 60 investors over eight quarters, carries one line that shows the mechanism with unusual clarity. On 19 February a television segment named Suvarna Chemicals Limited. The same evening, Meera Sundaram added Rs 1,00,000/- to the holding, taking its cost from Rs 3,00,000/- to Rs 4,00,000/- and her total cost from Rs 12,00,000/- to Rs 13,00,000/-. She already held it, so this was not a discovery. She simply put more money into the one of her four holdings that had been in front of her that evening.

Notice how little this needs. No claim that the segment was wrong, no claim that Suvarna Chemicals Limited was a poor thing to hold, and no claim about what happened to its price afterwards. The mechanism is complete once it is observed that money moved on a day when the only thing that changed was what was in front of her. One dated line is enough to show the mechanism for that reason, and a hundred lines would be needed to measure it.

How large is the effect once it is measured against the base rate?

One evening is an anecdote. A claim rather than a story becomes possible because the same log records what happened across all 240 decisions, and a share only means something when it is set beside the share to be expected anyway. The share to be expected anyway is the base rate of noticeHow much of the available list gets mentioned at all in a period., and skipping it is the single commonest way this whole subject gets reported badly.

Start with the composition. The 240 logged decisions break into 96 buys, 84 sells, 36 switches and 24 pauses of a standing instruction, and those four sum back to 240 with nothing left over. Of the 96 buys, 41 followed a media mention within three days. Work that out rather than reading it: 41 divided by 96 is 42.7 per cent.

Now the number that turns this into a measurement. Across the same period, 11.0 per cent of the eligible list was mentioned at all in a given week. If notice had nothing to do with buying, buys would be drawn from the mentioned part of the list roughly in proportion to its size, so about 11.0 per cent of 96. Working that out gives 10.6 buys, call it eleven. Forty-one is what actually happened. The finding is not that 42.7 per cent is a large number; it is that 42.7 per cent sits nearly four times above the 11.0 per cent that proportion alone would give.

What does the whole range between no effect and total capture look like?

The 42.7 per cent reads better as a point on a scale than as a fact on its own. Put the strength of the effect on one axis, running from zero, where notice does no selecting whatever, to one, where nothing unmentioned is ever bought. At zero strength the mentioned part still supplies buys in proportion to its own size, so the share of buys coming from the mentioned part of the list runs from 11.0 per cent up to 100 per cent in a straight line.

The floor of that scale is 11.0 per cent rather than zero, and the control below is built around that floor. A reader who expects zero at zero has quietly assumed that the mentioned part of the list would otherwise be bought by nobody. A base rate does not mean that. The answer is worth predicting before the control is moved.

Why does the buying side show this and the selling side barely at all?

Here is where most readers form the wrong idea, and it is worth slowing down. The natural explanation for a buying effect is enthusiasm: buying is exciting, selling is dull, so notice bites on the exciting side. The enthusiasm explanation is available, it is intuitive, and it is not what is happening. The real reason is arithmetic about set sizes, and it has nothing to do with mood.

When Meera Sundaram buys, her candidate set is everything she could possibly buy. Her buying candidate set runs to thousands of things, and no filter she can apply will ever consider more than a handful of them. Almost all of the removing had to be done by something, so a filter that admits only what reached her notice is doing almost all of the selecting. When she sells, her candidate set is the four holdings she already has. She does not need reminding that she holds them; they are on her own statement. A filter has nearly nothing left to remove. AsymmetryAn effect that runs strongly in one direction and weakly in the other. here is a property of the two sets, not of the person standing in front of them.

Barber and Odean framed the asymmetry this way in the 2008 paper, and framing it structurally is what makes it testable rather than merely plausible. The size of the set being filtered decides how much a filter matters, so the same person with the same attention in the same week shows the effect strongly in one direction and barely at all in the other. Nothing about enthusiasm predicts that. Set size predicts it exactly.

What can a person do about a filter they cannot feel operating?

The honest starting point is that the usual advice does not work here. Trying harder to notice more is not a thing attention can do on command. There is no volume control, and a person who resolves to be more observant this week has changed nothing about the size of the world. The world is still larger than they are. Kahneman's 1973 treatment of attention as a limited pool implies exactly that: the allocation can be moved around, but the total cannot be raised.

The only move available is structural: deciding what to look at before looking. A written list settled in advance, an order in which things get examined, a rule that a decision waits until at least two options nobody put in front of the decider have been added to the set. None of these require noticing more. Each of them changes the set itself, and the set was always beyond the reach of effort.

The Palash log has one measurement close to this, and it needs stating carefully. Twenty of the 60 investors adopted a written checklist on 4 November. Across the four quarters that followed they recorded a written reason on 34 of 41 decisions, being 82.9 per cent, against 19 of 63 for the other forty, being 30.2 per cent. The comparison measures whether a reason got written down, and it measures nothing else at all. Eight quarters and 60 investors could not support a claim about returns, so no difference in returns follows from these two shares.

When is filtering by what reached notice the right thing to do?

The alternative is examining thousands, and nobody has ever done that and nobody ever will. So something has to cut thousands of possibilities down to a handful. Notice is a free filter. Notice costs nothing to apply, it runs whether or not anybody asks it to, and it delivers a workable shortlist in no time at all. Judged as a filter rather than as a judgement, it is doing a job that must be done by something.

The conditions under which it is a reasonable filter can be stated plainly. First, the cost of examining any one option is high relative to how much better the best option is than an average one. Second, what reaches notice is not systematically related to whether the option is any good, so the filter is behaving like a cheap random sample rather than a biased one. Under those two conditions a shortlist drawn from whatever happened to be seen is a sensible economy, and calling it an error would be a mistake.

The second condition fails whenever what reaches notice is being chosen by somebody with an interest in that choice being made, and the trouble starts precisely there. Then the filter stops resembling a random sample and starts resembling a selection made on the decider's behalf by a party whose objective is a different one. A filter goes wrong in exactly that way, whoever is doing the choosing and whatever is being chosen.

How does an adviser use this without telling anybody what to hold?

Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, does not use any of this to form a view about a holding. She uses it to ask a different question in the meeting, and the question is about stage two rather than stage three. When a client arrives having decided, the usual question is why this one. Her question is what else was on the list, and where did these three come from.

The two questions get very different answers. The first produces a reasoned case, usually genuine and telling her nothing about the filter. The second produces either a source she can look at, such as a screen or a list somebody keeps, or a shrug. A shrug is the finding. A shrug does not mean the choice is bad; it means the choice was made inside a set that assembled itself, and that fact is now on the table where it can be discussed.

For a person deciding alone, with no adviser and no committee, the same move works with a pen. Before deciding, write down the three or four things being compared and, beside each, one line saying how it came to be on the list. The exercise takes two minutes and its whole value is that it makes stage two visible. Stage two is the one stage nobody can recall without being asked. If every line says the same thing, the filter has been found.

Two habits follow from this in practice, and both are about the set rather than about anybody's judgement. One is keeping a standing list that gets updated on a schedule rather than in response to events, so the set exists before the moment of deciding rather than being assembled at it. Decisions taken within a few days of something reaching notice are the population where the filter has most room to have operated, so the other habit is a rule that every such decision gets its written reason recorded. In the log, 71 of 240 decisions, being 29.6 per cent, were taken within 48 hours of a news item, and a written reason existed on only 84 of 240, being 35.0 per cent, so the overlap between those two groups is where an adviser would look first.

What else decides which options reach a person?

Attention is one of several things that decide which options reach a person, and only one of them is what arrives from outside and reaches notice. Memory supplies a second route and a display supplies a third: what comes back when a person tries to think of options, and what makes one item on a screen stand out against the others around it, are both set out under availability and salience. The distinction is genuinely useful: one is about what got through to a person, the other is about what can be retrieved or what a display emphasises.

Attention bias also stops well short of the aggregate. A great deal of interesting work asks what happens when many people are subject to the same filter in the same week, and that question belongs to market behaviour. How one decision got made and what happens to a price are two different questions, and the second one cannot be answered from a single decider. Every individual bias is likewise a statement about a decision rather than about a price.

Meera Sundaram, Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, the Palash 100 index, the Vindhya index scheme, the Nilgiri mid-cap scheme, Suvarna Chemicals Limited and Kesari Logistics Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.