Before you look up the price: write down the last two digits of your Social Security number, then bid on this bottle of wine
Try a small experiment before reading on. Write down the last two digits of your Social Security number (or your phone number), treat them as a dollar price, and ask yourself: would you buy this bottle of red wine at that price? Then write down the most you would actually pay.
Think that number couldn't possibly matter? Ariely, Loewenstein & Prelec (2003) had 55 MIT Sloan MBA students do exactly this: first write the last two digits of their Social Security number (SSN) and answer 'would you buy at this price,' then submit real, binding maximum bids for 6 products (a cordless keyboard, a cordless trackball, wine, Belgian chocolates, design books — average retail around $70), using the Becker-DeGroot-Marschak mechanism, where transactions really happen. The result: people whose SSN's last two digits were above the median bid 57%–107% more than those below it; by quintiles, the top quintile typically bid about 3 times the bottom quintile — $56 vs $16 for the same cordless keyboard. The correlation between SSN and bids ran 0.32–0.52, significant for all six product categories. An anchor doesn't have to come from someone else: an irrelevant number you merely 'brought to mind' can price goods for you — and real money on the line doesn't stop it.
The granddaddy of this trick is Tversky & Kahneman's (1974) wheel-of-fortune experiment: a wheel marked 0–100 (rigged to stop only at 10 or 65) was spun in front of subjects, who were first asked whether the percentage of African nations among UN members was higher or lower than that number, then asked for their own estimate. The group that drew 10 gave a median estimate of 25%; the group that drew 65 said 45% — a number generated at random, in plain sight, with zero relevance to the question, steered the judgment. Paying subjects for accuracy did not weaken the effect.
'That's just clueless students'? Northcraft & Neale (1987) had professional real-estate agents (n=47) and business students (n=54) tour the same house in person (appraised at $135,000 a year earlier); the only difference was the listing price in the packet, set at $119,900 / $129,900 / $139,900 / $149,900. The experts' average valuation was pulled from $114,204 all the way to $128,754 — anchored about as strongly as the amateurs. The stinger came in the debrief: only 19% of the experts admitted considering the listing price (vs 37% of the amateurs), and just 8% ranked it among their top three considerations. Wilson et al. (1996) delivered the final blow: explicitly warning people to guard against the number's influence did not eliminate anchoring either.
Two ski trips, same weekend: the $100 one is less fun, the $50 one more fun — which do you go on?
Next question. You hold two non-refundable, non-exchangeable ski-trip tickets for the same weekend: a $100 trip to Michigan and a $50 trip to Wisconsin — and you know clearly that you would enjoy Wisconsin more. Which do you go on?
Arkes & Blumer (1985) gave this to 61 students: 33 of them (54%) chose the more expensive but less enjoyable Michigan trip — 'I already paid $100' beat 'which weekend would actually be better spent.' This reasoning has a name: the Concorde fallacy, coined not in economics but in animal behavior — Dawkins & Carlisle's 1976 paper in Nature, mocking the British and French governments for pouring money into the Concorde supersonic jet, known to be commercially hopeless, on the grounds that 'too much has already been invested.' Rational decisions should compare only future costs and benefits: the money is sunk either way; the only thing still up for choosing is 'which weekend is more fun.'
Hypotheticals not convincing enough? They also ran a real-money field experiment with the Ohio University theater: season tickets were randomly sold to the first 60 customers at three prices — full price $15, $2 off ($13), or $7 off ($8). Random pricing created a natural control group: all three groups held exactly the same tickets. First half of the season: the full-price group attended an average of 4.11 plays, while the two discount groups attended only 3.32 and 3.29 — money already paid and unrecoverable shouldn't affect 'do I go tonight,' yet those who paid more clearly worked harder to 'get their money's worth.' And a detail that usually gets overlooked: in the second half of the season the difference disappeared — the psychological pressure of sunk costs decays over time.
An epidemic threatening 600 lives: are 'save 200 for sure' and '400 die for sure' the same plan?
Last question. An epidemic is expected to kill 600 people. Program A: 200 people will be saved for sure. Program B: a 1/3 chance all 600 are saved, a 2/3 chance no one is. Which do you pick? — In Tversky & Kahneman's (1981) 'lives saved' frame (N=152), 72% picked the certain A and 28% picked B.
A different group (N=155) saw the same epidemic reworded: Program C: 400 people will die for sure. Program D: a 1/3 chance nobody dies and a 2/3 chance all 600 die. This time only 22% took the certain C, and 78% switched to gambling on D. Yet A≡C and B≡D are mathematically identical — the same numbers, rephrased, flipped the majority outright: a gains frame induces risk aversion; a losses frame induces risk seeking. Don't count on professional training as a shield: McNeil et al. (1982) presented physicians the same surgery/radiation data framed as 'survival rates' vs 'mortality rates,' and the doctors' treatment choices swung dramatically too. The counter: before any big choice, restate the options in the opposite frame (saving 200 = letting 400 die) — if your answers to the two versions disagree, the frame is steering you.
Why do frames hit so hard? Because our mental pricing of losses and gains is asymmetric. Tversky & Kahneman (1992), fitting cumulative prospect theory's value function v(x)=x^α (x≥0), v(x)=−λ(−x)^β (x<0) to pricing data from 25 subjects, found median parameters α=β=0.88 and λ=2.25: for the same amount, a loss weighs about 2.25 times as much psychologically as a gain — the pain of losing $1 roughly equals the pleasure of gaining $2.25.
But careful: loss aversion ≠ risk aversion everywhere. In the original 1979 prospect-theory data (the reflection effect), on the gains side 80% chose 'a sure 3,000' over (4,000, p=0.8); on the losses side 92% (N=95) preferred to gamble on (−4,000, p=0.8) rather than accept a sure loss of 3,000. In the domain of losses, people are risk seeking — the psychological root of the gambler doubling down to 'win it back' and of investors clinging to their losing stocks.
Loss aversion also has an everyday incarnation: the endowment effect. Kahneman, Knetsch & Thaler (1990) randomly handed Cornell mugs (priced $6.00 at the campus bookstore) to half the students, then opened a market between 'mug owners' and 'non-owners': sellers' median asking price was $5.25, while buyers' median offer was only $2.25–2.75 — the moment the mug lands in your hands, your selling price becomes more than double the buying price. Theory predicted about 11 trades per round; only 1–4 actually happened (V/V*≈0.20), and four repeated rounds produced no 'learning.' The most elegant part came in a follow-up adding a third group, the 'choosers': they picked between a mug and various cash amounts — objectively in exactly the seller's position. Result: sellers' median valuation $7.12, choosers' $3.12, buyers' $2.87. Choosers ≈ buyers, far below sellers — ruling out income effects: the gap comes not from 'having a mug or not' but from 'to own is to be endowed; to give up is to lose.' Same counter-move as before: when valuing your holdings, old stuff, or old plans, ask 'if I didn't have it now, what would I pay to get it?'
自测 · 学完检查一下
想真正动手做题、记进度、攒连胜？到互动课里练。
Tversky & Kahneman (1974) spun a wheel marked 0–100 in front of subjects (rigged to stop only at 10 or 65), first asked whether the percentage of African nations among UN members was higher or lower than that number, then asked for an estimate. What were the two groups' median estimates?
答案：The group that drew 10 gave a median estimate of 25% and the group that drew 65 gave 45% — a number generated at random in plain sight, with no relevance to the question, steered the judgment
The median estimates were 25% (anchor 10) and 45% (anchor 65). The anchor was a 'pure random number' generated publicly, right before the subjects' eyes, with nothing to do with the UN — yet it steered the judgment. The key detail: paying subjects for accuracy did not weaken the anchoring effect, so the last option gets it exactly backwards. Anchoring doesn't require a plausible-looking number — a transparently random one works just as well. (Source: Tversky & Kahneman 1974, Judgment under Uncertainty: Heuristics and Biases, Science 185(4157):1124-1131)
Ariely, Loewenstein & Prelec (2003) had 55 MIT Sloan MBA students write down the last two digits of their Social Security number, treat them as a dollar price and answer 'would you buy at this price,' then submit real, binding maximum bids for 6 products including a cordless keyboard and wine (Becker-DeGroot-Marschak mechanism — transactions really happened). What was the result?
答案：People whose SSN's last two digits were above the median bid 57%–107% more; by quintiles, the top quintile typically bid about 3 times the bottom — $56 vs $16 for the same cordless keyboard
Those whose SSN's last two digits were above the median bid 57%–107% more than those below; the top quintile typically bid about 3 times the bottom quintile — $56 vs $16 for the cordless keyboard; correlations between SSN and bids ran 0.32–0.52, significant for all six product categories. Two counterintuitive points: (1) an anchor doesn't have to be given by someone else — an irrelevant number you merely 'brought to mind' can price goods for you; (2) the BDM mechanism made bids real and binding, and real money didn't stop the anchoring. (Source: Ariely, Loewenstein & Prelec 2003, 'Coherent Arbitrariness', Quarterly Journal of Economics 118(1):73-106)
True or False: Professional real-estate agents are valuation experts who deal with listing prices every day, so when appraising a house on site they won't be anchored by the listing price printed in the packet.
答案：False
False. Northcraft & Neale (1987) had professional real-estate agents (n=47) and business students (n=54) tour the same house in person (appraised at $135,000 a year earlier), varying only the listing price in the packet across four levels from $119,900 to $149,900: the experts' average valuation was pulled from $114,204 to $128,754 — anchored about as strongly as the amateurs. Yet only 19% of the experts admitted considering the listing price (vs 37% of the amateurs), and just 8% ranked it among their top three considerations. Wilson et al. (1996) further showed that an explicit advance warning does not eliminate anchoring. Neither expertise nor knowing the bias's name grants immunity — experts are just less likely to admit being anchored. (Sources: Northcraft & Neale 1987, OBHDP 39(1):84-97; Wilson et al. 1996, JEP: General 125(4):387-402)
Arkes & Blumer (1985) worked with the Ohio University theater to randomly sell season tickets to the first 60 customers at three prices: full price $15, $2 off ($13), or $7 off ($8) — all three groups held identical tickets. How many plays did the groups attend in the first half of the season?
答案：The full-price group averaged 4.11 plays while the two discount groups attended only 3.32 and 3.29 — those who paid more worked harder to 'get their money's worth'; but the difference vanished in the second half: sunk-cost pressure decays over time
The full-price group averaged 4.11 plays; the $2-off group 3.32 and the $7-off group 3.29. Random pricing created a natural control group — all three groups held identical tickets, and the only difference was the unrecoverable money already paid; those who paid more clearly worked harder to 'get their money's worth.' The 'gap stayed stable' option fails on its second half: in the second half of the season the differences disappeared — the psychological pressure of sunk costs decays over time. That's the most commonly overlooked detail, and it means a cooling-off period is a natural ally against sunk costs. (Source: Arkes & Blumer 1985, The Psychology of Sunk Cost, OBHDP 35(1):124-140)
Where does the term 'Concorde fallacy' come from, and what reasoning does it mock?
答案：From ethologists Dawkins & Carlisle's 1976 Nature paper — mocking the British and French governments for continuing to fund the commercially hopeless Concorde jet because 'too much has already been invested': letting past investment dictate future behavior
The term 'Concorde fallacy' was coined by ethologists Dawkins & Carlisle in a 1976 Nature paper (on parental investment and mate desertion) — using the British and French governments' continued funding of the commercially hopeless Concorde supersonic jet, justified by 'too much has already been invested,' to mock that reasoning. The principle it exposes: rational decisions should compare only future costs and benefits — sunk investments get no vote. That the term originates in animal behavior rather than economics is exactly the trivia this concept gets tested on. (Source: Dawkins & Carlisle 1976, Parental investment, mate desertion and a fallacy, Nature 262:131-133)
Arkes & Blumer (1985) asked 61 students to imagine holding two non-refundable ski-trip tickets for the same weekend — a $100 trip to Michigan and a $50 trip to Wisconsin — while being told explicitly that they would enjoy Wisconsin more. What did most choose?
答案：33 students (54%) chose the more expensive but less enjoyable Michigan trip — 'I already paid $100' overrode 'Wisconsin is more fun'
33 of the 61 students (54%) chose the $100 Michigan trip — despite being told explicitly they would enjoy Wisconsin more. The money for both tickets is sunk and unrecoverable whichever trip they take; the only future payoff still up for choosing is 'which weekend is more fun,' so the rational answer is Wisconsin. But 'more expensive' beat 'more fun': most people paid tribute to money already down the drain. This is the laboratory version of the Concorde fallacy — past investment hijacking a future choice. (Source: Arkes & Blumer 1985, The Psychology of Sunk Cost, OBHDP 35(1):124-140)
The Asian disease problem (Tversky & Kahneman, 1981): an epidemic is expected to kill 600 people. In the 'lives saved' frame (N=152), 72% chose 'save 200 for sure' over a 1/3 chance of saving all 600. When the same numbers were recast in a 'deaths' frame (400 die for sure vs a 1/3 chance nobody dies), what happened?
答案：Only 22% still chose the certain program and 78% switched to the gamble — a gains frame induces risk aversion, a losses frame induces risk seeking, and the majority flipped outright
In the 'deaths' frame (N=155), only 22% chose the certain Program C ('400 die for sure') while 78% switched to Program D's gamble (a 1/3 chance nobody dies); in the 'lives saved' frame (N=152), 72% had chosen the certain Program A ('save 200 for sure'). A≡C and B≡D are mathematically identical, yet the gains frame induced risk aversion and the losses frame induced risk seeking — the majority flipped from 72% to 78% on the opposite side. McNeil et al. (1982) presented physicians the same surgery/radiation data framed as survival vs mortality rates, and the doctors' treatment choices swung dramatically too — professional training is no shield against framing. (Sources: Tversky & Kahneman 1981, Science 211(4481):453-458; McNeil et al. 1982, NEJM 306:1259-1262)
Cumulative prospect theory (Tversky & Kahneman, 1992) fitted a value function to pricing data from 25 subjects and obtained a median loss-aversion coefficient λ ≈ 2.25. What does this number mean?
答案：For the same amount, a loss weighs about 2.25 times as much psychologically as a gain — the pain of losing $1 roughly equals the pleasure of gaining $2.25
λ ≈ 2.25 is the amplification factor on the loss side of the value function: v(x)=x^α (x≥0), v(x)=−λ(−x)^β (x<0), with median fitted parameters α=β=0.88 and λ=2.25 from 25 subjects' pricing data — for the same amount, a loss carries about 2.25 times the psychological weight of a gain: 'the pain of losing $1 ≈ the pleasure of gaining $2.25.' It describes the asymmetric weighting of losses versus gains ('losses loom larger than gains,' an idea originating in 1979 prospect theory) — not probability overestimation, not an insurance premium, and not memory duration. (Source: Tversky & Kahneman 1992, Advances in Prospect Theory, Journal of Risk and Uncertainty 5(4):297-323)
True or False: Loss aversion does not mean risk aversion everywhere — in the original prospect-theory data, 80% chose 'a sure 3,000' over (4,000, p=0.8) on the gains side, yet 92% (N=95) preferred to gamble on (−4,000, p=0.8) rather than accept a sure loss of 3,000: in the domain of losses, people are risk seeking.
答案：True
True. This is the reflection effect: on the gains side, 80% were risk averse (taking a sure 3,000 over an 80% chance of 4,000); on the losses side, 92% (N=95) were risk seeking instead (preferring an 80% chance of losing 4,000 over a sure loss of 3,000). Loss aversion (λ ≈ 2.25) describes the asymmetric weighting of losses versus gains — it does not mean 'more conservative when facing losses.' Quite the opposite: to escape a certain loss, people take bigger risks — the gambler doubling down to win it back and the investor clinging to losing stocks both trace back to this. (Sources: Kahneman & Tversky 1979, Prospect Theory, Econometrica 47(2):263-291; Tversky & Kahneman 1992, JRU 5(4):297-323)
In the mug experiment (Kahneman, Knetsch & Thaler, 1990), sellers' median asking price was $5.25 while buyers' median offer was only $2.25–2.75. A follow-up added a third group, the 'choosers,' who picked between a mug and various cash amounts — objectively in exactly the seller's position. What did the choosers' valuations show?
答案：Choosers' median valuation was $3.12 — close to buyers' ($2.87) and far below sellers' ($7.12) — ruling out income effects and proving the gap comes from loss aversion: 'to own is to be endowed, to give up is to lose'
Choosers picked between a mug and various cash amounts, objectively in exactly the seller's position (both end up 'keeping the mug or walking away with money'), yet their median valuation of $3.12 sat close to the buyers' $2.87 and far below the sellers' $7.12 — if the gap came from income effects or disagreement about the mug's worth, choosers should have matched sellers. This design ruled out income effects and proved that the 'selling price ≈ more than double the buying price' gap comes from 'to own is to be endowed; to give up is to lose.' Supporting details: for the $6.00 mug, theory predicted about 11 trades per round but only 1–4 happened (V/V*≈0.20), with no learning across 4 repeated rounds. (Source: Kahneman, Knetsch & Thaler 1990, Experimental Tests of the Endowment Effect and the Coase Theorem, JPE 98(6):1325-1348)