MathemagiciansLibraryExplorationsDeep Dives
← Economics · Pre-University
EconomicsPre-University

Cognitive biases — how we really decide

PE3.1 — Consumer Theory › PE3.1.2 — Behavioural EconomicsLecture 2 of 4~2810 words

A number that should not matter

In an experiment reported in 1974, the psychologists Amos Tversky and Daniel Kahneman spun a wheel of fortune, numbered from 0 to 100, in front of each person taking part. Secretly, the wheel was fixed: it stopped only at 10 or at 65. Then they asked two questions. First: is the percentage of African countries in the United Nations higher or lower than the number on the wheel? Second: what is your best estimate of that percentage?

Everyone watched the wheel spin. Everyone knew its number was chance. So the number carried no information about the United Nations at all. Before you read on, make a prediction.

Predict, then reveal — does a random number change an estimate?

One group saw the wheel stop at 10. The other saw it stop at 65. What happened to their estimates? Choose one.

The figure shows the median estimates of the two groups in Tversky and Kahneman's experiment (Science, 1974).

0%10%20%30%40%50%60%70%25%Wheel showed 1045%Wheel showed 65the number on the wheelMedian estimate
Median estimates of the percentage of African countries in the United Nations. People who saw 10 gave 25%. People who saw 65 gave 45%. The dashed lines mark the number each group saw. Both groups moved away from their number, but not far enough.

People who saw 10 gave a median estimate of 25%. People who saw 65 gave a median estimate of 45%. A number from a spinning wheel moved the answers by about 20 percentage points.

The consumer of the last few lectures ignores irrelevant information. For her, the wheel cannot matter. Yet it did, and in a direction anyone could have predicted.

The last lecture showed that defaults and frames change choices, and it promised to open the box: to look at the biases that make them so powerful. This lecture takes four of them: anchoring, loss aversion, present bias and status quo bias. For each, it asks: what is the pattern, why does a sensible mind produce it, and what does it do to an economy?

Why a sensible mind makes predictable errors

In the 1950s, the economist and psychologist Herbert Simon argued that real decision-makers have bounded rationality. They try to choose well, but with limited information, time and attention. So they use heuristics: mental shortcuts, such as "start from a number you have and adjust it", or "if in doubt, keep what you have".

Heuristics are not stupid. Most of the time they give good answers quickly. In Thinking, Fast and Slow (2011), Kahneman described two modes of thinking: a fast, automatic "System 1", where shortcuts live, and a slow, effortful "System 2", which can check them. Checking takes effort, so it often does not happen. The two systems describe how thinking feels; they are not two parts of the brain.

A heuristic becomes a bias in situations where it fails in a predictable direction. This is the key word for an economist: systematic. Random mistakes are not a problem for economic theory. If some buyers overvalue a good and others undervalue it by the same amount, the errors cancel out, and market demand is where the rational model puts it. A systematic error does not cancel. If nearly everyone is pushed the same way, the whole market moves.

Stop and think

A shop puts a high "recommended price" next to every item. If it anchors every shopper upwards a little, does this change market demand, or only the choices of a few careless people?

Show the answer

It changes market demand. Each shopper is pushed in the same direction, so willingness to pay rises a little for many people at once. The demand curve for those items shifts to the right. A random error by a few careless shoppers would cancel out; a systematic push does not. That is why economists care about biases, and not only psychologists.

Anchoring: the first number sticks

The pattern. When people estimate an uncertain number, they start from a number already in front of them, the anchor, and adjust. The adjustment is usually too small. So the final answer stays close to the anchor, even when the anchor is irrelevant.

Why. Adjusting takes effort, and people stop as soon as the answer seems reasonable, which is at the edge closest to the anchor.

The evidence. The wheel is one case. Another comes from courts. In a study published in 2006, Birte Englich, Thomas Mussweiler and Fritz Strack gave experienced German judges a case of shoplifting. Before sentencing, each judge rolled dice loaded to show 3 or 9. Kahneman reports average sentences of about 8 months after a 9 and about 5 months after a 3. Experience did not protect the judges.

In the economy. Anchors are everywhere that prices are uncertain. In a negotiation, the first offer anchors everything that follows. A sign saying "Was $200, now $140" makes $140 look cheap by anchoring on $200. That is only fair to the buyer if the $200 was a real price. In the United States, the Federal Trade Commission's Guides Against Deceptive Pricing say that a former price shown in an advertisement must not be fictitious. Notice what this rule does. It does not ban anchors; it bans false ones.

Loss aversion: losses weigh about twice as much

The pattern. People judge outcomes as gains or losses from a reference point, usually what they have now. And a loss hurts more than a gain of the same size pleases. Losing $100 feels worse than winning $100 feels good.

How much more? Tversky and Kahneman's estimate, from a 1992 study, was 2.25: a loss counts about 2.25 times as much as an equal gain. A 2024 meta-analysis of 607 estimates from 150 articles (Brown, Imai, Vieider and Camerer) found a mean of about 1.96, most likely between 1.8 and 2.1. So "about twice" is a fair average, though individuals differ.

The evidence. In a classroom experiment reported in 1990, Kahneman, Jack Knetsch and Richard Thaler gave coffee mugs to half the students at random. Owners were asked the lowest price at which they would sell. Non-owners were asked the highest price they would pay. Owners asked a median of $7.12; buyers offered a median of $2.87. That is about 2.5 times as much for the same mug. Ten minutes of ownership had turned giving up the mug into a loss. This is called the endowment effect.

In the economy. Loss aversion helps explain why money wages rarely fall. In a telephone survey reported in 1986, Kahneman, Knetsch and Thaler described a profitable company that cuts wages by 7% when there is no inflation: 62% of people called this unfair. When the same company gave a 5% rise while inflation was 12%, only 22% called it unfair. Both reduce what the wage can buy by about the same amount, but the first is coded as a loss and the second as a gain. So firms avoid money wage cuts, even in a recession: one reason wages are "sticky" downwards.

Present bias: now is special

The pattern. Everyone values money today more than money later. That alone is not a bias. The standard model, exponential discounting, says that each day of waiting reduces a reward's value by the same percentage. Under that model, a choice between two dates depends only on the gap between them, so your choice never changes as the dates come closer.

Real people behave differently. Here is a choice. On Monday, you are offered $100 on Friday or $110 on Saturday. Most people say: I will wait one extra day for $10 more. Now it is Friday morning, and you are offered the same choice again: $100 today or $110 tomorrow. Many people now take the $100. The gap is still one day. The amounts are the same. Only the distance from today has changed, and the choice has reversed.

This is present bias: the present gets extra weight compared with every future moment. A simple way to model it, used by David Laibson in 1997, multiplies every future amount by an extra factor below 1, often written β ("beta"). Today escapes the penalty; tomorrow and next month pay it equally. So the step from today to tomorrow feels much bigger than the step from Friday to Saturday seen from Monday. Try both models below.

Explore — $100 on one day or $110 the next

Choose a model. Then move the number of days until the $100 arrives (the $110 always comes one day later). Watch whether the choice changes as the day comes closer.

Model:

Days until the $100: 4

The table shows how much each option is worth to an invented person today. Each day of waiting costs 1%. In the present-biased model, every amount that is not paid today is also multiplied by 0.85.

Days until the $100Steady: $100 now-valueSteady: $110 now-valuePresent-biased: $100Present-biased: $110
0100.0108.9100.092.6
199.0107.884.291.6
298.0106.783.390.7
397.0105.782.589.8

With steady discounting, the $110 is worth more at every distance, so the person always waits. With present bias, the $110 wins whenever both payments are in the future, but the $100 wins when it is paid today: 100 against 92.6. The numbers are invented to make the pattern visible.

With steady discounting, nothing you do with the slider changes the choice. With present bias, the plan is always "wait", until the day arrives. That is the signature of present bias: plans and actions disagree. Economists call such preferences time-inconsistent.

In the economy. Stefano DellaVigna and Ulrike Malmendier studied 7 752 members of three health clubs in the United States (American Economic Review, 2006). Members who chose a monthly contract of over $70 went on average 4.3 times a month. That is more than $17 a visit, although a 10-visit pass at $10 a visit was available. When joining, members expected to go far more often than they did. The authors' leading explanation is overconfidence about future self-control: present bias, together with not knowing you have it. The same pattern explains why Save More Tomorrow, from the last lecture, worked: it asked people to commit future pay rises to saving.

Status quo bias: doing nothing wins

The pattern. People prefer the current state of affairs more than its merits justify. William Samuelson and Richard Zeckhauser named this status quo bias in 1988. Several causes work together: changing means giving something up, which loss aversion makes costly; comparing options takes effort; and a default can look like a recommendation.

The evidence. In the early 1990s, two neighbouring states in the United States, New Jersey and Pennsylvania, offered drivers the same kind of choice. A cheaper insurance policy limited the right to sue after an accident. A more expensive one kept the full right to sue. In New Jersey, the limited policy was the default. In Pennsylvania, the full policy was the default. Most drivers kept their state's default (Johnson, Hershey, Meszaros and Kunreuther, 1993, and later reports). About 80% in New Jersey kept the limited right. About 70–75% in Pennsylvania kept the full right. The drivers were similar. Their defaults were not.

In the economy. Status quo bias makes defaults powerful, and it slows switching between banks, phone contracts and energy suppliers. That gives existing firms some market power over loyal customers.

Real case: paying teachers in advance

The policy. In the 2010–11 school year, Roland Fryer, Steven Levitt, John List and Sally Sadoff ran a field experiment in nine schools in Chicago Heights, in the United States. Teachers were offered a bonus based on their students' maths results. Some teachers were in a "gain" group: they would be paid at the end of the year. Others were in a "loss" group: they were paid about $4 000 at the start of the year. They signed a promise to return all or part of it if their students did not improve enough. Both groups earned the same final bonus for the same results.

The prediction. The rational model predicts no difference, because both contracts pay the same for the same results. Loss aversion predicts that the loss group works harder.

What happened. In the first year, students of teachers in the loss group improved their maths scores by about 0.2 to 0.4 standard deviations. The authors compare this to raising teacher quality by more than one standard deviation. The gain group's effects were smaller and not statistically significant (NBER Working Paper 18237, 2012). A study in a high-tech factory in China found the same direction, on a smaller scale (Hossain and List, Management Science, 2012). Teams worked about 1% harder when a bonus was framed as a loss.

The verdict. Loss framing can change behaviour, and sometimes cheaply. But the evidence needs care. When the Chicago Heights study ran a second year, the effects were concentrated in the first year, and the second wave showed no effect (Fryer and others, published version 2022). Other studies of loss-framed pay find little effect. There is also an ethical question: asking workers to hand back money already paid can feel harsh. Loss aversion is robust in experiments, but turning it into policy is harder.

Worked examples

Worked Example 1 — Turning down a good bet

Kofi is offered a coin toss: heads he wins $120, tails he loses $100. (a) Should a person who cares only about expected money accept? (b) Suppose Kofi weighs a loss twice as heavily as a gain. Does he accept? (c) What is the smallest win that would make him accept?

(a) Expected value = 0.5 × 120 − 0.5 × 100 = 60 − 50 = $10. It is positive, so yes.

(b) Felt value = 0.5 × 120 − 0.5 × 2 × 100 = 60 − 100 = −40. It is negative, so he refuses a bet that pays $10 on average.

(c) He accepts when 0.5 × W ≥ 0.5 × 2 × 100, so when W ≥ $200. With loss aversion of about 2, the possible win must be about twice the possible loss.

Check. At W = 200: 0.5 × 200 − 0.5 × 200 = 0 ✓. This model uses straight-line values; the next lecture adds the curvature of prospect theory.

Worked Example 2 — The gym plan that fails

Mei plans her week on Monday (invented numbers). A gym visit on Saturday costs her effort worth 6 units on the day. It gives health benefits worth 8 units, which arrive later. She is present-biased with β = 0.7, and has no other discounting. (a) On Monday, does she plan to go? (b) On Saturday, does she go? (c) How large must β be for her to go?

(a) From Monday, both the cost and the benefit are in the future, so both are multiplied by 0.7: 0.7 × (−6 + 8) = 0.7 × 2 = 1.4. It is positive, so she plans to go.

(b) On Saturday, the cost is now, and escapes the penalty. The benefit is still in the future: −6 + 0.7 × 8 = −6 + 5.6 = −0.4. It is negative, so she stays home.

(c) She goes when −6 + β × 8 ≥ 0, so when β ≥ 6 ÷ 8 = 0.75.

Reading it. On Monday, Mei sincerely means to go, which is why she buys a membership. A firm can profit from the gap between her plan and her action.

Check. With β = 0.75: −6 + 0.75 × 8 = 0 ✓.

Trap 1 — "Biased" does not mean "random" or "stupid"

A bias is a predictable direction of error from a shortcut that usually works. Do not write "people are irrational" as an explanation. Name the bias, say which way it pushes, and say why the shortcut fails in this case.

Trap 2 — Loss aversion is not risk aversion

Risk aversion is a dislike of uncertainty, judged on the final outcome. Loss aversion is about the reference point: the same $100 counts more as a loss than as a gain. The mugs had no risk at all, yet owners and buyers valued them differently. The next lecture shows that loss aversion can even make people take risks to avoid a loss.

Trap 3 — Impatience is not present bias

Preferring money sooner is normal, and a steady discount rate is consistent. Present bias is the reversal: the plan says wait, and the moment says now. In an answer, show the change of choice as the date comes closer. Do not just say that people prefer the present.

Try it yourself

Work each one on paper first. Open a hint only when you are stuck.

  1. Name the bias most clearly at work in each case, and explain in one sentence. (a) A homeowner will not sell below the price she paid, although prices in her area have fallen. (b) A student plans to revise every evening, and each evening watches videos instead. (c) A worker keeps the pension fund she was given on her first day for twenty years. (d) A shop shows a television at $499 next to a crossed-out "$899".

    Hint 1

    Ask what the reference point or the first number is in each case, and whether time plays a part.

    Hint 2

    One case is about a plan that fails when the moment comes. One is about doing nothing.

    Full answer

    (a) Loss aversion: the price she paid is her reference point, and selling below it is coded as a loss. (b) Present bias: from the morning the plan looks good, but in the evening the cost is now and the benefit is later. (c) Status quo bias: keeping the default needs no decision. (d) Anchoring: the $899 is the first number and makes $499 look cheap; it is only fair if $899 was a real former price. Check: each answer names the direction of the push, not only the label ✓.

  2. Aisha is offered a coin toss that loses $40 on tails. Her losses weigh 2.25 times as much as gains. (a) What is the smallest win on heads that she would accept? (b) What would it be using the meta-analysis mean of about 1.96?

    Hint 1

    She accepts when 0.5 × win ≥ 0.5 × λ × 40.

    Hint 2

    The halves cancel, so the win must be at least λ × 40.

    Full answer

    (a) 2.25 × 40 = $90. (b) 1.96 × 40 ≈ $78.40. In both cases she refuses many bets with a positive expected value, such as win $70, lose $40. Check: 0.5 × 90 − 0.5 × 2.25 × 40 = 45 − 45 = 0 ✓.

  3. Daniel has β = 0.8 and no other discounting. (a) Today he is offered $100 now or $112 in one week. Which does he choose? (b) A month earlier, he was asked about the same two payments. Which did he plan to choose? (c) What is the smallest later amount that would make him wait today?

    Hint 1

    Only amounts paid today escape the factor 0.8.

    Hint 2

    A month earlier, both payments are in the future, so both are multiplied by 0.8.

    Full answer

    (a) $100 now is worth 100; $112 later is worth 0.8 × 112 = 89.6. He takes the $100. (b) A month earlier: 0.8 × 100 = 80 against 0.8 × 112 = 89.6. He planned to wait. His choice reverses, so his preferences are time-inconsistent. (c) He waits today if 0.8 × X ≥ 100, so X ≥ $125. Check: 0.8 × 125 = 100 ✓.

  4. In Singapore, life insurers give a 14-day "free-look" period: a buyer can cancel a new policy within 14 days of receiving the documents and get the premiums back, less some costs such as medical fees. (a) Which biases from this lecture might lead someone to buy a policy they later regret? (b) Which bias could stop the free-look period from working well? (c) Suggest one change to the design, and give one argument a critic might make against such rules.

    Hint 1

    Think about the sales meeting: who names the first number, and what feels urgent?

    Hint 2

    What does a buyer need to do to cancel, and what happens if they do nothing?

    Full answer

    (a) Anchoring, if the salesperson's first figure for the cover or premium sets the frame; present bias, if a "decide today" offer makes the present feel decisive; loss aversion, if the sale stresses what the family could lose. (b) Status quo bias: cancelling needs action, and doing nothing keeps the policy. So many buyers who would choose differently after reflection may never cancel. (c) One option is a short reminder near the end of the period that asks the buyer to confirm, so the reflective choice is prompted. A critic might say that such rules add costs that insurers pass on in premiums, that they treat adults as unable to decide, or that regulators have biases of their own. A good answer weighs these against the evidence on how many buyers regret. Check: the answer uses at least two biases and one criticism ✓.

The bigger idea

The rational consumer was meant to predict markets, on the argument that individual errors cancel out. When errors are systematic, they do not cancel; they add up, and they can be used. A firm can set the anchor, write the default and design a contract that the present-biased self signs. Some economists call the result a behavioural market failure, and use it to support rules such as cooling-off periods and honest-price rules.

There is a strong case on the other side too. Markets can teach. At a sports-card show, John List found that beginners showed a strong endowment effect, but experienced dealers showed little or none (Quarterly Journal of Economics, 2003). And the people who write rules have biases of their own, so a regulator can fail as well as a market. So the economist's question is not "are people rational?". It is "where, and by how much, do predictable errors change outcomes, and who can best correct them?" The next lecture builds loss aversion into a full theory: prospect theory.

Key concepts in this lesson