Availability and Salience: What Comes to Mind Gets Overweighted

Availability answers how common something is with how easily an example comes to mind. Salience is the property of a display that made the example easy to bring back in the first place. Ease of recall tracks vividness, recency and repetition rather than frequency, so the error runs in one direction rather than scattering, and it can be several times the size of the truth.

Everything in this guide follows from a single mismatch, and it is worth sitting with before any of the names arrive. Consider what the mind actually does when it is asked how often something happens. The mind does not go out and count. It reaches back, brings up whatever arrives first, and reads the effort of that reaching as an answer. Two examples came quickly, so the thing must be common. Nothing came, so it must be rare. Recall is a retrieval, not a sample, and a retrieval is weighted by how the memory was formed rather than by how often the thing occurred. Weighting by how the memory formed is the whole engine. Every named effect below is one consequence of it.

What easy question does availability put in place of how often something happens?

A heuristic works by swapping a hard question for an easy one and answering the easy one instead. The hard question here is a frequency: out of everything in some class, how many were of this kind? Answering it properly means having a list, going down it and counting. The easy question that gets answered instead is about effort: how readily does an example of this kind come to mind? AvailabilityJudging how common something is by how easily an example comes to mind. is the name Tversky and Kahneman gave to that substitution in Cognitive Psychology in 1973, and their point was not that people are careless. Their point was that the substitution usually works, and a shortcut that keeps producing the right answer is one nobody notices running.

Take it out of finance first. Ask somebody how common road accidents are on the stretch outside their office and they will not consult a record. They will remember the one they saw last month, the ambulance, the crowd. Ask how common a quiet uneventful journey is on the same stretch, and nothing comes back at all. A journey where nothing happened leaves no handle to grip. The two questions have the same denominator and the same list of days. Only one of them has memories attached. The moment a class of events differs in how well it is remembered rather than in how often it happens, the substitution stops being harmless and starts producing a number that is wrong in a predictable direction.

Why is recall not a sample of the world?

A sample earns its authority from how it was drawn. Take 240 lines from a record with equal probability and the share of any kind of line in the sample is an honest estimate of its share in the record. RetrievalBringing something back from memory, as against counting it in the world. is drawn on a completely different rule. Retrieval brings back what was encoded strongly, rehearsed often, or happened recently, and none of those has any obligation to line up with how often the thing occurred. The selection is not random and it is not proportional. The selection is weighted, and the weights were set at the moment each memory was laid down.

Weighted retrieval is why thinking harder does not fix the answer. Effort is applied to the retrieval, and the retrieval is doing precisely what it always does, only more thoroughly. Straining to remember more instances of a memorable class brings back more memorable instances. The only repair that touches the actual problem is to stop remembering and start counting something that was written down at the time. Counting a written record is the one practical repair available.

What is Salience, and why is it a property of the display rather than the mind?

If ease of recall decides the answer, the next question is what decides ease of recall. Part of it sits inside the person, and part of it sits outside, in how the thing was presented. SalienceThe property of a display that makes something stand out and be remembered. is the outside part: the quality of standing out from its surroundings. Bordalo, Gennaioli and Shleifer developed salience into a formal treatment of choice in the Quarterly Journal of Economics in 2012, and their framing is useful because it locates the effect in the situation rather than in a person's character.

Consider any screen or printed list of things. Something on it is at the top and something is at the bottom. Something is set in larger type. Something carries a colour nothing else in the display carries. Something appears three times while its neighbours appear once. Something sits against a background that throws it forward. Each of those is a decision somebody made while building the display, and each one raises how well that item is recalled an hour later. Salience is not a fact about the mind at all: it is a fact about the arrangement in front of the person looking. The arrangement was chosen, and it could have been chosen differently.

A display raises later recall, and raising recall is not the same as misleading anybody. Whether a given presentation meets a duty owed to somebody is a conduct question, and conduct is examined separately on its own terms.

What actually makes something memorable, and does memorability track frequency?

Three properties do most of the work, and none of them is frequency. The first is vividnessHow concrete and sensory a memory is, which drives recall more than frequency does.: a concrete, specific, sensory event lays down a stronger trace than an abstract one. A named thing on a screen with a number beside it is vivid. A quarter in which nothing much moved is not. The second is recencyHow lately something happened, which raises recall independently of how often it happens.: what happened lately comes back more easily than what happened two years ago, whatever the two rates were. The third is repetition. Repetition is the only one of the three with any honest relationship to frequency at all, and even that relationship breaks the moment one event is shown three times.

Now put frequency on one axis and ease of recall on the other, and ask whether the points line up. They do not. A standing instruction that moves Rs 25,000/- on the same day every month is designed to be forgettable, so it is at once one of the most frequent things in an ordinary financial life and one of the least memorable. A single evening in which a television segment named one company and money moved that same night is one of the rarest and most memorable. Frequency and memorability are two different axes, and the estimate produced by availability follows the memorability axis while the true answer sits on the frequency axis. The distortion is not a small one at the edges. It is the whole of the error.

Why do memorability and frequency come apart in one direction rather than at random?

If the gap between the two axes were random noise, availability would be an imprecise shortcut rather than a bias, and it would be nothing to teach. Estimates would sometimes overshoot and sometimes undershoot, and the errors would cancel across enough judgements. Cancelling is not what happens, and the reason is structural. Events that are dramatic, recent, personally involving or repeatedly shown are systematically overweighted, and events that are quiet, old, routine and unshown are systematically underweighted, and no class of events is quiet and dramatic at once.

So the direction of the error is decided in advance by what kind of event a class contains. The property that makes a shock rare and the property that makes it memorable are the same property, so a rare shock is overweighted every single time it is estimated by recall, by everybody who estimates it that way. A bias is the error that is left over when a substitution fails in one direction rather than scattering, and availability qualifies because the memorable classes and the frequent classes are not the same classes. A reader who thinks the errors cancel will never bother with the repair, so the point deserves care.

What is the Focusing Illusion, and why does whatever is examined closely seem to matter more?

There is a close relative of availability that runs during the looking rather than afterwards. The focusing illusionWhatever is being attended to seeming more important than it is. is the finding that nothing in life is as important as it seems while it is being thought about. Schkade and Kahneman set it out in Psychological Science in 1998, comparing how satisfied people in two regions said they were with how satisfied each group predicted the other would be. The question itself pushed attention onto the one dimension on which the two regions clearly differ, and that dimension then swelled to fill the judgement. Both sides expected a difference the measurement did not support.

The focusing illusion differs from availability without being unrelated to it. Availability is about what returns from memory when a question is asked later. The focusing illusion is about what is currently under the spotlight, right now, being inflated by the fact that it is under the spotlight. The two meet at the point where a thing heavily attended to becomes a thing easily recalled. Salience travels down exactly that road.

The practical form of this is unglamorous and constant. One holding opened, with three screens of writing read about it, will feel like the position that decides the outcome, whatever share of the total it actually is. Suvarna Chemicals Limited was Rs 4,00,000/- of a Rs 13,00,000/- cost, or 30.8 per cent, and on the evening of 19 February it was one hundred per cent of what was being thought about. The weight a thing carries in a judgement is inflated by the attention currently pointed at it, so the act of examining something carefully is itself a distortion, and the repair is to look at the same list of items in the same way rather than to look harder at one.

Availability Bias vs Representativeness Heuristic: which easy question is each one answering?

These two get merged more often than any other pair in this subject, and the merge is understandable, because both are substitutions and both produce a confident number where a frequency was wanted. The difference is which easy question each one answers. Availability answers a question about effort: how readily does an example arrive? Representativeness answers a question about similarity: how closely does this case match the type I have in my head? Kahneman and Tversky set out representativeness in Cognitive Psychology in 1972, one year before the availability paper, and they are deliberately different mechanisms rather than two names for one.

Take a single example and run both over it. Running both shortcuts over one case is the only way to feel the difference. A person is thinking about a company whose sales have grown quickly for three quarters. Judging it likely to keep growing because it fits a picture of what a fast growing company looks like is representativeness: the work is being done by resemblance to a type. Judging it likely to keep growing because three cases of exactly this pattern come to mind immediately is availability: the work is being done by ease of retrieval. Neither of the two ever asks how often companies with three fast quarters carry on, and that frequency is what would have produced the right answer. The shared silence about the base rateHow often something actually happens, independent of how well it is remembered. is why the two get confused.

Attention Bias vs Availability Bias: what reaches a person, and what comes back afterwards?

Attention is a filter on arrival. Attention decides what reaches a person out of everything that was there to be noticed, and a limited attention account is about the size of that filter and what gets through it. Availability is a filter on return. Availability decides what comes back out of everything that once got through. The two sit at different points in time on the same event, and the ordinary way of talking about both, that something grabbed the attention and stuck in the mind, welds them into one sentence and hides the difference.

Keep them apart by watching a single event twice. A newsletter arrives listing forty items, and a person reads all forty. Every one of them passed attention. Ask that person a month later how many of the forty were of a certain kind and thirty eight of them are simply gone, so they never enter the estimate at all: seen and not recalled. Now take one item from a different day, shown once, in large type, on an evening when money moved because of it. The single item passed attention exactly once and comes back every time anything nearby is asked about: recalled far out of proportion to its share of what arrived. Attention and availability can run in opposite directions on the very same event. Opposite directions on one event are the plainest proof that the two are separate mechanisms and not one.

The practical consequence is about timing, and it is the reason this distinction earns its space. An attention problem is repaired before deciding, by changing what gets put in front of somebody or what they choose to look at. An availability problem is repaired at the moment of estimating, by counting a record instead of consulting a memory. A correction aimed at the wrong one of the two does nothing whatsoever: widening the list of what gets looked at does not repair an estimate distorted by recall, and counting a written record does not help with something that never reached the person in the first place.

What happens to the estimate when the same events are made more memorable?

Here is the one frequency this case actually locks down, and it is worth being precise about where it comes from. The Palash decision log is an invented record of 240 decisions taken by 60 investors over eight quarters. Against that log sits an eligible list of things that could be bought, and in a given week 11.0 per cent of that list is mentioned anywhere at all. The 11.0 per cent is the true rate. The rate is a counted number, it comes from a record, and it does not move for anything that follows.

Now ask somebody to produce that same number from memory rather than from the record. The estimate they give rises with how memorable the mentions were, not with how many there were. Work the illustrative scale used here in both directions before trusting it. The estimate is the true 11.0 per cent plus 39 points multiplied by how memorable the class is, on a scale running from nothing to one. At nothing the estimate is 11.0 plus zero, giving 11.0 per cent, exactly right. At half it is 11.0 plus 19.5, giving 30.5 per cent. At the top it is 11.0 plus 39.0, giving 50.0 per cent, and 50.0 divided by 11.0 is 4.5. The estimate can run to more than four times the truth without one single extra event having occurred, purely because the events that did occur were easier to bring back.

Two notes on that table, because a reader checking it will find both. The three quarter row works out to exactly 40.25 before rounding, a tie at one decimal place, and it is shown rounded up to 40.3 rather than down. And the true rate of 11.0 per cent does not appear as a row at all. The true rate is not a row: it is the same number at every line, and holding it still is the entire point of the exercise.

One more line from the same log shows the effect landing on something a person could check about themselves. Of the 96 buys in the log, 41 followed a media mention within three days. Divide 41 by 96 and the answer is 42.7 per cent. Set that beside the 11.0 per cent of the list mentioned at all in a week and the two numbers are doing very different jobs: one is a counted rate about the world, the other is a counted rate about the decisions. A buy made the evening after a mention is exactly the kind of buy that comes back easily, and an ordinary buy on an ordinary Tuesday leaves nothing to retrieve. Ask any of the 60 investors to recall what share of their own buying followed a mention and the number they give will run above 42.7 per cent.

Where is the shortcut correct, and what is the only repair when it is not?

An account that only lists errors leaves a reader unable to use the idea, so the correct case deserves stating without hedging. Ease of recall is an excellent estimate of frequency whenever memorability is flat across the classes being compared. Asked how often it rained last month against how often it was clear, recall answers well. Both are ordinary, both are unremarkable, and neither has been dressed up in larger type by anybody. The substitution is not a defect. The substitution is a fast method with a stated condition, and the condition is that nothing has made one class of events easier to remember than another.

The condition fails whenever one class is dramatic, recent, personally involving or repeatedly displayed and its comparison class is none of those. A failed condition is the moment to stop estimating from memory. And the repair, the only one that works, is to count something that was written down at the time rather than to consult a recollection. The repair therefore has a precondition worth knowing in advance: it needs a record to exist. If nobody wrote anything down, the estimate cannot be rescued by thinking harder, and the honest move is to name it as a recollection and give it the weight a recollection deserves.

How does an adviser, or a person deciding alone, actually use this?

Devika Rao, the adviser at the invented Palash Advisory Services Private Limited, uses it in one narrow and unglamorous way: she refuses to let any frequency claim into a conversation from memory. When a client says that this keeps happening, or that everybody is doing this now, or that the last three times were all like this, she treats the sentence as a report about ease of recall and not as a report about the world. The question back is always the same. Which record exists to count that? If the answer is the log, the answer gets counted. If there is no record anywhere, the claim stays in the conversation but stops being a number.

For a person deciding alone, with no adviser and no committee, the same method works and costs an evening. Whatever was about to be estimated from memory was written down somewhere: the bank entries, the statement of transactions, the messages sent on the day. The record gets counted, dull entries that leave no trace included, and the counted number goes somewhere it will be seen again. The value of a written record is not that it is more detailed than a memory but that it weights every line equally. Equal weight on every line is the one thing a memory can never do. A household keeping a single sheet of what actually happened has bought itself the only repair on offer here.

Overselling that method would do the reader no favours, so here is one warning. Twenty of the sixty investors in the log adopted a written checklist on 4 November, and across the following four quarters they recorded a written reason on 34 of 41 decisions, being 82.9 per cent, against 19 of 63, being 30.2 per cent, among the other forty. The checklist produced a measured change in how much got written down. Sixty people over eight quarters could not support a claim about outcomes, so the change is not evidence of one, and no claim about returns follows from it.

What does availability not explain?

Three limits, stated plainly, because a mechanism that explains everything explains nothing. Availability does not explain what a person chose to look at in the first place. Choosing what to look at is a separate matter, settled before anything is recalled. Availability does not explain the story somebody builds around a set of events afterwards. A story is about the shape given to what is remembered rather than about which items return. And it says nothing whatsoever about what a memorable event does to a price. A price is the sum of what a great many people did, and the subject throughout has been one person producing one estimate.

There is also a limit on the case itself. The Palash decision log records 240 decisions by 60 investors over eight quarters, and every number taken from it is illustrative rather than measured. One log is a worked example, never evidence that any effect is a certain size, and the 39 point scale used in the control above is an illustrative device rather than a measured constant. The arithmetic shows the shape of the error, not its magnitude in any real setting.

Devika Rao, Palash Advisory Services Private Limited, the Palash decision log, 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.