Lecture 3: Heuristics and Biases PDF

While you wait: what does the term ‘mental shortcut’ make you think of?  Lecture 3: Heuristics and Biases 1. Heuristics and Biases Background 2. Representativeness & Availability 3. Anchoring and Adjustment & Framing Effects 4. Prospect Theory * Content notice: brief mention of cancer/illne...

While you wait: what does the term ‘mental shortcut’ make you think of?  Lecture 3: Heuristics and Biases 1. Heuristics and Biases Background 2. Representativeness & Availability 3. Anchoring and Adjustment & Framing Effects 4. Prospect Theory * Content notice: brief mention of cancer/illness on slide 21, towards the end * There’s a lot in this lecture so it may spill over into the second hour, marked for Practical 1 2  Lecture 3: Heuristics and Biases Intended learning outcomes: Know what a decision heuristic is and why people use heuristics Understand the major heuristics, including the representativeness, availability, and anchoring and adjustment heuristics, and the main experimental finding related to them Know what framing effects are and what the most important experimental findings are for them Have a basic understanding of prospect theory, including the biases that are built into it 3  Heuristics and Biases - Background A heuristic is a rough-and-ready procedure or rule of thumb for making a decision – Without an exhaustive comparison of all available options – Without any guarantee of an optimal or best result – The term was introduced by Herbert Simon (1957) in his book Models of Man – A consequence of bounded rationality: We simply do not have the cognitive capacities required for exhaustive examination of alternatives and computation of optimal decisions to apply expected utility theory – Satisficing (the first heuristic) 4  Historical Development H&B research programme, starting in the 1970s, led by psychologists Tversky & Kahneman (T&K) – Can generate biases and fallacies in judgment and decision From the late 1990s, German psychologist Gerd Gigerenzer and his group promoted fast and frugal heuristics – can be better than considered judgments and decisions (Caren Frosch lectures) 5  Representativeness (a) Typically used for judging What is the probability that object A belongs to class B? What is the probability that event A originates from process B? – Probabilities evaluated by the degree to which A resembles B Leads to serious errors: similarity or representativeness neglects relevant factors First, base-rate neglect – K&T (1973): People shown brief personality descriptions; assess the probability that it belonged to an engineer or a lawyer – “Dick is a 30 year old man. He is married with no children. A man of high ability and high motivation, he promises to be quite successful in his field. He is well liked by his colleagues.” 6  Representativeness (b) – Conveys no relevant info, so probability should = base rate, but people judged p =.5 regardless of base rate – One group told: sample contained 70 lawyers, 30 engineers – Other group: 30 lawyers, 70 engineers – Results: essentially the same probability judgments Second, sample size neglect – K&T (1972): Probability of a man over 6 feet tall judged the same value for samples of 1,000 men, 100 men, and 10 men – Also, probability of a maternity hospital having 60% or more boys on a single day judged the same (21%) for a large hospital (45 births per day) or a small hospital (15 per day) 7  Representativeness (c) Third, regression fallacy o K&T (1973): flying instructors noticed that praise given to a trainee pilot for an exceptionally smooth landing was usually followed by a rougher landing on the following attempt, and harsh criticism for an unusually rough landing was usually followed by a smoother landing on the following attempt o Concluded that praise was counter-productive and punishment highly effective; but any random deviation from the mean is likely to be followed by one closer to the mean o Think of rolling a die repeatedly: A high score (6 or 5) will usually be followed by a lower score, and a low score (1 or 2) by a higher score – regression toward the mean o Same regression fallacy occurs when police crack down on a sudden increase in drug-related crime (say) in a particular area 8  Availability (a) People often judge the probability of an event by the ease with which they can think of instances of it Can cause bias: availability is affected by other factors First, retrievability – K&T (1973): People saw a list of people in public life and judged whether the list contained more men than women – In some lists the men were more famous than the women; in others the women were more famous than the men – In each list, people judged that the gender with more famous names be more numerous 9  Availability (b) Second, effectiveness of search set – K&T (1973) asked people: Suppose we sample a word of three letters or more at random from an English text. Is it more likely that the word starts with k or that k is the third letter? – Much easier to search memory for words starting with k, so people judge this more likely, but a typical text has twice as many words with k the third letter Third, imaginability bias – T&K (1973) asked people to judge How many different committees of 2, 3, 4, …, 10 can be formed from 10 people? – Easier to imagine groups of 2 than of 8 -- easier to imagine disjoint sets -- so estimates decreased monotonically with size – Median estimates were 20 (groups of 8) and 70 (groups of 2), although they must be the same (actually 45) 10 FYI 11  Availability (c) Fourth, illusory correlation Term introduced by Chapman & Chapman (1967), who showed experienced clinicians and students information about a number of patients—diagnostic statements and a drawing of a person made by that patient Later the judges judged the frequency with which each diagnosis (e.g., paranoia) had been accompanied by various features of the drawing (e.g., peculiar eyes) They greatly overestimated the co-occurrence of familiar associates, such as these two, even when they were actually negatively correlated in the information that they viewed Interpreted by T&K as availability – ease of recalling examples that fit prior beliefs, causing bias in this case 12  Anchoring and Adjustment (a) We often make judgments by starting with an initial estimate and the adjusting it – Adjustment is typically insufficient – Slovic & Lichtenstein (1971) asked people to estimate the number of African countries in the United Nations by first indicating whether a randomly chosen number was too high or too low, and then estimating the actual number – Those who began from 10 gave a median estimate of 25 (too low), and those who began from 65 gave a median estimate of 45 (too high at that time) 13  Anchoring and Adjustment (a) We often make judgments by starting with an initial estimate and the adjusting it – T&K: High school students estimated, within 5 seconds, either 8x7x6x5x4x3x2x1 or 1x2x3x4x5x6x7x8 – Median estimates were 2,250 and 512 [actually 40,320] People anchor on first few steps and then adjust insufficiently 14  Anchoring and Adjustment (b) Conjunctions and disjunctions – Israeli psychologist Maya Bar-Hillel (1973) invited people to take one of the following bets (a) simple event (e.g., drawing a red marble from a bag containing 50% red marbles) (b) conjunction (e.g., drawing a red marble seven times in succession, with replacement, from a bag containing 90% red marbles) (c) disjunction (e.g., drawing a red marble at least once in seven successive tries, with replacement, from a bag containing 10% red marbles) 15  Anchoring and Adjustment (c) Majority preferred – (b) conjunction to (a) simple event, and – (a) simple event to (c) disjunction – The less likely option in both cases, if the actual probabilities are calculated – People overestimate probability of conjunctions and underestimate probability of disjunctions – T&K (1974) believe that they anchor on simple events (easiest to imagine) and then adjust insufficiently down for conjunctions and up for disjunctions 16  Anchoring and Adjustment (c) Conjunctions are less likely than they are judged to be, and disjunctions are more likely that judged – p(disjunction) > p(simple event) > p(conjunction) 17  Anchoring and Adjustment (d) Conjunction fallacy (T&K, 1983) Linda problem – participants told: – Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a student, she was deeply concerned with issues of discrimination and social justice, and also participated in antinuclear demonstrations. Which is more probable: – A: Linda is a bank teller – B: Linda is a bank teller and is active in the feminist movement – 85% chose B, but probability p x q is never higher than p, because it comes from multiplying numbers of 1 or less 18  So far: Representativeness – Base-rate neglect – Sample size neglect – Regression fallacy Availability – Retrievability – Effectiveness of search set – Imaginability bias – Illusory correlation Anchoring and adjustment – p(disjunction) > p(simple event) > p(conjunction) Conjunction fallacy 19  Framing Effects (a) A framing effect is an effect of the description, labelling, or presentation of a problem on responses to it – Classic Asian disease problem (T&K, 1981): Choose between two programs for combatting a disease expected to kill 600 people – In the gain frame, participants chose between A and B: (A): 200 people saved; (B): 1/3 probability of 600 people saved and 2/3 probability of no people saved 73% preferred A to B. – In the loss frame, participants they chose between: (A*): 400 people die; (B*): 1/3 probability that nobody dies and a 2/3 probability that 600 people die 78% preferred B* to A* but the versions differ only in framing – Certainty of saving life attracts, and certainly of death repels 20  Framing Effects (b) McNeil, Pauker, Sox, and Tversky (1982) showed that patients choosing between surgery and radiation for cancer are strongly affected by framing effects – In this case, whether the relevant statistics are framed in terms of survival or mortality – People chose options with 90% (one-month) survival significantly more (84%) than those with 10% mortality (50%) – This framing effect equally pronounced among experienced physicians and patients – Again, it is loss aversion that seems to explain the differences – Video 1:49 (Louise Cooper, financial journalist) 21  Prospect Theory (a) Formulated by T&K as an alternative to EU theory, designed to accommodate experimental findings: – Reference dependence: We judge outcomes as gains or losses relative to our current situation rather than their absolute values – Loss aversion: We attach greater weight to losses than to corresponding gains: T&K showed we feel losses about 1.2 times as intensely as equivalent gains – Probability weighting function: We overweight very small probabilities and underweight moderate and high probabilities – S-shaped value function: People show risk aversion for gains but risk seeking for losses; confusingly, they call it a value function rather than a utility function 22 Prospect Theory (b) S-shaped value function, concave (decreasing marginal utility) for gains, and also convex for losses Would be a straight line at 45 degrees if objective gains and losses perfectly matched their utility Half the utility (labelled VALUE) of gaining $100 is felt when gaining only $35, and gaining twice as much as $35 is not enough to double the utility to 100% Half the pain of losing $100 is felt when losing only $40 But slope for losses steeper: we feel losses more intensely than gains 23  Prospect Theory (c) T&K (1992) would you accept this gamble? – 50% chance to win $150 and 50% chance to lose $100 If your overall wealth were lower by $100? – Very few takers, and overall wealth makes no difference – Which would you choose: – A: Lose $100 with certainty – B: 50% chance to win $50 and 50% chance to lose $200 If your overall wealth were higher by $100? – Vast majority chose the gamble B – But the two versions are identical, with only the reference point changing (all values $100 lower) – framing effect 24  Prospect Theory (d) Endowment effect (Thaler, 1980) – Tendency to demand much more to give up an object than one is willing to pay to acquire it – People given either a lottery ticket or $2.00. Some time later, each offered the opportunity to trade the lottery ticket for money, or vice versa – Very few chose to switch – preferred what they had – People given mugs estimate their cash value higher than people given pens, and vice-versa – Loss aversion, relative to reference point Vast amount of data supports prospect theory and a slightly revised version called cumulative prospect theory; the best decision theory we have – see K&T (1983) and K (2003) on reading list video 1:33 25  Lecture 3: Heuristics and Biases 1. Heuristics and Biases Background 2. Representativeness 3. Availability 4. Anchoring and Adjustment p(disjunction) > p(simple event) > p(conjunction) Conjunction Fallacy (Linda problem) 5. Framing Effects Gain/loss frame (Asian disease problem) 6. Prospect Theory Endowment effect 26