Reasoning and decision-making

Question 1

Heuristics and biases (state)

Answer 1

• Availability
  
  • how easy it is to bring instance to mind
• Representativeness
  
  • based on similarity
• Anchoring
  
  • assimilation of numeric estimate towards another value

Question 2

Availability (heuristic)

Answer 2

Can be biased if experience doesn’t = true frequencies OR if ease to recall based on something other than frequency

Lichtenstein et al (1978) - estimate frequencies of causes of death

• overestimating low frequency events, underestimate high frequency events (e.g. heart disease)
• rare events receive disproportionate attention –> greater availability

Tversky & Kahneman (1973) –> recalled more famous names (more available)

• 12.3 famous recall vs 8.4 not famous /20
• conjunction fallacy: ease of recall bias: belief that conjunction of 2 events can have higher probability than either event individually

Tversky & Kahneman (1983) –> 4 page novel - number of -ing words vs number of -n- words

• 13.4 for -ing
• 4.7 for -n-
• even though -ing is a subset of -n-

Question 3

Representativeness (heuristic)

Answer 3

Base rate neglect = similarity-based judgements are insensitive to prior probabilities

Kahneman & Tversky (1973)

1. told profile pulled out of 70% engineer pool OR 70% lawyer pool

• probability of people saying he’s an engineer doesn’t change relative to pool
• based on prototype (similarity) instead

2. given profile, list of 9 academic subjects

• told to judge:
  
  • rank subjects by likelihood they specialise in it
  • rank how typical they are of each group
  • estimate proportion of all people that study it
• negative correlation for base rate estimation
  
  • % of all people AND likelihood they specialise
• positive correlation for representativeness
  
  • how typical + how likely

Question 4

Anchoring (heuristic)

Answer 4

DEMONSTRATING

• Tversky & Kahneman (1974) –> wheel of fortune (10 vs 65)
• Chapman & Johnson (1999) - last 2 digits of SSN = anchor %

EXPLAINING

• Anchor + adjust –> start estimation from anchor + adjust (effortful cognitive work)
  
  • BUT: you’d expect incentives would help people do better –> not the case
    
    • Epley & Gilovich (2005)
• Anchor value changes our perception of magnitude of other candidate values
  
  • Frederick & Mochon (2011)
• Externally presented anchors seen as a hint/suggestion even if uninformative
  
  • participation bias?

Question 5

Ecological rationality (state)

• • mistakes made

Answer 5

Evolved or based on experience interacting with world

• Natural frequency formats = better
• Gambler’s fallacy
• Hot hand fallacy
• Representativeness
• Memory constraints

Question 6

National frequency formats (ecological rationality)

Answer 6

Eddy (1982)

• medical professionals given probabilities of breast cancer given mammograms with FB rates
  
  • 95/100 gave the wrong answer
    
    • inverse fallacy: probability of cancer given a positive test = the probability of a positive test given cancer
    • only 8% got it right
  • people evolved to register counts not abstract percentages
  • NB: bayes theorem would get right answer
    
    • alter estimates based on new evidence + background information
• Hoffrage & Gigerenzer (1998): eg above in natural frequency formats
  
  • out of Z number of people, X have it and Y don’t
  • 46% get it right
• we are better at understanding natural frequencies - makes more sense to us
  
  • evolved to keep track of event frequencies by natural sampling (in this case, no need to take into account base-rate information)

Question 7

Fallacies:

• gambler’s fallacy
• hot-hand fallacy
• memory constraints

Answer 7

Gambler’s fallacy = when random process throws up a streak of the same outcome, there will be a correction for it

• Croson & Sundali (2005) - witnessed in Nevada casino when streak of 4+ the same, people likely to bet against it
• Tversky & Kahneman (1971) - people expect local sequences to have properties of large sequences:
  
  • in which case HTHTHT would be more likely than HHHHTTTT

Hot hand fallacy = infer streak is not representative of randomness - so streak will continue

• Gilovich et al (1985) - number of baskets scored in a row - believe if on a streak, they’ll make the next one
  
  • because streak is not representative of randomness, they think it is not random + that it will continue

Ayton & Fischer (2004) - roulette style game:

• probability of predicting same as last time decreases as streak increases (Gambler’s)
• confidence in predicting skill increases as streak of successful predictions increases (Hot hand)
• SO: people take into account previous experience, intentional human performance + positive recency
• people view random mechanical outcomes as sampling without replacement (where odds DO decrease - for Gambler’s)

A&F (2004) –> 2 possible outcomes with different alteration rates

• if low AR - more likely to say it was basketball shots
• if high AR - more likely to say it was coin toss
• consistent with sampling without replacement + skill improvement

BUT: memory constraints

• Hahn + Warren (2009) - looked at mathematical properties of fair coin under realistic conditions
  
  • there is a higher likelihood of high AR sequence for short/finite sequences
    
    • these are the ones people see + remember
  • so it is rational for people to expect T after long streak of H

Question 8

Why do people make mistakes on reasoning (about syllogisms OR propositional reasoning)

Answer 8

• Heuristics
• Comprehension
• Mental models
• Framing and experience

Question 9

Syllogistic + propositional resoning

• what are they
• common mistakes

Answer 9

syllogisms + propositional reasoning –> 2 kinds of deductive reasoning

Syllogism = 2 premises then conclusion involving quantifiers: all, no, some, some…not

• All A are B, All B are C –> so All A are C
  
  • 88% correct (Robert & Sykes, 2005)
• All B are A, all B are C –> so some A are C
  
  • 8% correct (Robert & Sykes, 2005)

Propositional reasoning: about propositions containing conditionals (if, and, not, or):

• Schroyens et al (2001): ‘if A then B’ rule
  
  • modus ponens (97% correct) = it’s A..so it’s B
  • modus tollens (74%) = it’s not B..so not A
  • affirmation of the consequent (64% commit it) = if B then A
  • denial of the antecedent (56% commit it) = if not A then not B

Wason (1968) - 4 card selection task (D, K, 3, 7) –> rule = if D then 3 on other side

• only 1/34 turned over right cards to test rule (D+7)

Oaksford & Chater (1994) - rule = if p then q

• p=89%; not-p=62%; q=62%; not-q=25%

initial theory = confirmation bias - to show rule is true

Question 10

Heuristics

• syllogisms
• propositional reasoning

Answer 10

SYLLOGISMS

• Atmosphere theory = people match ‘mood’ of premises to nood of conclusion
  
  • quantity - can be universal (all/no) or particular (some, some not)
  • quality - can be affirmative (all,some) or negative (no, some…not)
• Begg + Denny (1969) - 64 syllogisms, 4 possible conclusions (45/65 have no valid conclusion - how did people respond to them)
  
  • when both premises were positive - 79% conclusions endorsed were positive
  • when at least one premise was negative - 73% chosen conclusions were negative
  • when both premises universal - 77% conclusions chosen were universal
  • when at least one premise was particular - 90% chosen conclusions were particular
  • BUT: doesn’t explain why sometimes people correctly endorse that there is no valid conclusion

PROPOSITIONAL REASONING

• Evans & Lynch (1973) 4 card task –> S, 9, G, 4: ‘if S then not-9’
  
  • if confirmation bias - turn s and 4
  • people turn S and 9
  • Matching heuristic - turn cards mensioned in rule

Question 11

Comprehension:

• syllogisms
• propositional reasoning

Answer 11

SYLLOGISMS:

• errors in syllogistic reasoning partly reflect use of language in formal logic vs every day use
  
  • all A are B –> A=B
  • some –> not all (but in logic can be all too)
• Ceraso & Provitera (1971) –> giving explanations/clarifying premises reduces error rates
  
  • 58% with traditional format
  • 94% correct with modified version

PROPOSITIONAL:

• Wagner-Egger (2007): bidirectional interpretation of ‘if’ (if A then B = if B then A) - common pattern of misunderstanding
  
  • not an error or bias in selection task - just don’t understand rules
• Gebauer & Laming (1997) - the illogicality is only apparent, if people understand it they do fine
• BUT: doesn’t explain why people do worse on modus tollens than modus ponens

Question 12

Mental models

• syllogisms
• propositional reasoning

Answer 12

SYLLOGISMS:

• Laird (2005):
  
  • construct mental model of world implied by premises (comprehension)
  • make a composite model + draw conclusion (description)
  • validate by searching for alternative models and checking they don’t contradict conclusion (validation)
  • more models = more likely conclusion right
• Copeland & Radvansky (2004):
  
  • more possible models = less accurate + slower
  • better working memory = more accurate + faster
• Newstead et al (1999) - no correlation between no. of models considered + accuracy
  
  • people usually just construct 1 - multiple models are harder

PROPOSITIONAL

• initial model created relates to items mentioned in the rule
  
  • explains why modus ponena easier than modus tollens
  • can avoid affirmation of the consequent + denial of the antecedent if put in cognitive effort to flesh out models

Question 13

Framing and experience

• syllogisms
• propositional reasoning

Answer 13

SYLLOGISMS:

• Evans et al (1983) –> if valid more likely to endorse it; if believable more likely to accept it
  
  • belief bias = people don’t reason independently of experience or framing of problem
• Klaver et al (2000) –> base rates kick in when uncertain of conclusion - pushes reasoning after
  
  • dual process framework
  • if conclusion believable = only construct 1 model (consistent - verification
  • if conclusion unbelievable = look for inconsistent models (tto falsify)

PROPOSITIONAL

• Giggs & Cox (1982) - beer, coke, 16, 25
  
  • ‘if beer then over 19’
  • if abstract - 0% correct
  • if in context - 73% correct
  • due to prior experience with rule
• BUT: Cosmides (1989) - even if no experience, context helps (due to social contract violation sensitivity)
  
  • “If a man eats cassava root, then he must have a tattoo on his face” – still chance
  • If give extra information – that cassava root eating is privilege of married men (married men have the tattoo) then performance increases to 75%
• BUT: Manktelow & Over (1990)
  
  • “If you clear up spilt blood, then you must wear rubber gloves” (75% correct)
  • but this is not a cost-benefit situation or social contract
    
    • due to relevance/expected utility
• Girotto et al (2001) –> if going to X, must be vaccinated
  
  • 62% correct –> going to X, not immunised (rule-violation sensitivity)
  • if boss worried they made a mistake - don’t actually need it
    
    • 71% check going to X and immunised
    • information relevant to problem

Question 14

Kinds of decision making

Answer 14

• Riskless multi-attribute choice
  
  • 2+ options differing on 2+ attributes, no probability
• Intertemporal choice
  
  • choosing between options available at different points in time
• Decisions under uncertainty
  
  • don’t know in advance what the outcomes are (don’t know the probabilities)
• Decisions under risk
  
  • 1+ of the possible outcomes are probabilistic (know the probability)
  • THESE ARE THE ONES WE LOOK AT

Question 15

Rational choice theory:

• expected value
  
  • BUT issues
• expected utility theory
  
  • BUT issues

Answer 15

EXPECTED VALUE

The EV of an option = sum of each possible outcome weighted by its probability

BUT: K+T (1979) - people risk averse for gains

• £3000 for sure over 80% chance of £4000

EXPECTED UTILITY

Transform actual value to subjective value

• explains risk aversion –> decreasing utility for higher rewards, less sensitive to increasingly large gains

BUT:

• people make decisions with respect to a starting point
• preference reversal –> risky when losses not gains –> EU only looks at end states (K+T, 1979)
• people weigh losses + gains differnetly - framing of the question alters choice (T+K, 1981)

Question 16

Prospect theory

• explaining actual (seemingly irrational) behaviour of people
• the value function
• the decision weights function

Answer 16

K+T (1979):

• reference-dependence –> outcomes considered as gains/losses with respect to reference point
• risk attitudes –> risk averse for gains (decreasing sensitivity - concave); risk-seeking for losses (decreasing sensitivity - convex)
• Loss aversion - loss curve steeper than gain curve
  
  • ‘losses loom larger than gains’
• EVIDENCE:
  
  • Endowment effect (Knetsch, 1989): give mug or hot chocolate (reference point) - keep or swap
    
    • loss feels worse than gains even if equally desirable at the start
• Decision weights function –> maps objective probabilities onto decision weights (W x V)
  
  • small probabilities are overweightted
  • large probabilities are underweighted
  • steeper around 0+1 (near certainty) - more significant –> certainty effect
• Editing - when considering options, people round small amounts + combine probabilities with the same outcomes

Question 17

Probability

• when people don’t act rationally
  
  • against expected utility theory

Answer 17

Allais’ paradox (K+T, 1979)

• people treat 0%–> 1% chance differently from 66%–>67%
• starting point makes a difference –> violates rationality + EU

Certainty effect (K+T, 1981)

• change in probability has a greater effect if going from point of certainty than midrange
  
  • making one option certain distorts probability judgement

Non-linear probabilities (Gonzalez & Wu, 1999)

• 5–>10% more significant than 30–>35%
• 65–>70% less significant than 90–>95%
• probabilities treated differently even if just near certainty

Question 18

Problems with prospect theory

Answer 18

Limited scope

• valuation vs choice (Lichtenstein & Slovic, 1971) - p-bet (high prob.), $-bet (high value)
  
  • 93% of people who chose the p-bet sell $-bet for a higher amound (want p but $ more valuable)
  • only 17% showed reversal in other direction

Attraction effect (asymmetrical dominance

• Ariely (2009) - online ($59.99), online + print ($125), print ($125)
  
  • 84% chose O+P
  • if P removed, 68% chose O
  • violates core assuption of rational choice theory –> independence from irrelevant alternatives
• Wedell (1991) - preference reversal if decoy more similar - attracted to option that dominates decoy

Empirical problem –> differing probabilities change risk-taking behaviour

• Weber & Chapman (2005) - high probability - people risk averse, low probability - different vanishes - more risky
  
  • prospect theory would expect same difference (between % choosing low/high stakes) no matter the probability

Purely descriptive –> lacks mechanism to explain how people arrive at decision

• POSSIBLE RESPONSE: decision by sampling

Question 19

Decision by sampling (mechanism of prospect theory)

Answer 19

Stewart et al (2006)

• when people given a value/probability, they turn it into subjective value by comparing it to other values available in memory
  
  • cognitive mechanism of decision making
• Looked at bank credits –> put small amounts in more frequently (easy to map objective –> subjective value based on freq. of encounter)
• Looked at bank debits –> spend small amounts more frequently (spending small amounts = more frequent than saving small amounts)
• put numerical value to everyday english phrases used for probabilities - what people thought they were objectively
  
  • overweighting of small probabilities
  • underweighting of large probabilitites
  • steep around 1 + ) –> same as decision weight function

Question 20

Basic emotions

• what are they?
• views
• yes/no?

Answer 20

Universal emotions - culturally-ubiquitous

• Darwin (1872) - categorical idea:
  
  • anger/fear/sadness/disgust/enjoyment (across species)
• Ekman (1992) –> basic emotions have rapid onset, brief duration, unbidden occurrence, distinctive signs, physiological correlates

Dimensional view:

• Russel & Barrett (1998) : core affect –> arousal + valence (pleasant/unpleasant) dimensions (high to low)

NOT universal (Gendron et al., 2018)

• depends on culture + definition of emotion
• 1975-2008 –> moderate-strong evidence for universal expression of emotions
• 2008-2018 –> no strong, 2 moderate, 9 weak evidence

Question 21

Physiology of emotions

• for and against specific physiology

Answer 21

James-Lange view (James 1884; Lange 1885):

• stimulus –> percept –> physiology –> emotion
• emotion = product of somatic/physiological change

NO:

• Cannon (1927) - people without peripheral inputs still experience emotion
  
  • peripheral arousal doesn’t create emotion
  • peripheral states no sufficiently differentiated per emotion
• Siegel et al (2018) –> metaanalysis - prediction of emotion from physiology = 31-32% correct
  
  • almost the same as if angry were guessed for all

Question 22

Role of cognition in emotion

Answer 22

Schachter (1964) the effect of somatic arousal depends on its attribution - how it’s interpreted given the social context

Schachter & Singer (1962) –> physiological arousal provides raw ingredients, cognition defines the emotion

• given ‘suproxin’ (adrenaline) + told side effects –> attribute arousal to the pill
  
  • stooge not that effective (euphoric or angry)
• given suproxin + not told side effects –> attribute arousal to mood as result of stooge (euphoric –> positive mood; angry –> negative mood)

Zajonc et al (1980): rejects role of congition in emotion

• previously-encountered stimuli elicit more positive affect than novel do, even if no conscious awareness of having seen it

Scherer (1984) –> appraisal theorist –> cognitive appraisals underlying emotion need not be conscious

• various appraisal dimension shape emotion: certainty, control, responsibility of others, attention

Question 23

Common ground of contemporary views of emotion

Answer 23

Scherer & Moors (2019) - review:

• multi-level appraisals –> cognitive components - evaluation of memories/event/stimuli
• physiology - physical responses in body
• action tendencies - propensity to behave in certain ways
• motor expression - facial expression, voice tone, body language, gestures
• component integration-experienced feeling –> subjectively what it’s like to feel an emotion
• no one-to-one mapping - all integrated

Question 24

Amygdala lesions and emotion

Answer 24

• Blanchard & Blanchard (1972)
  
  • reduced fear conditioning - decreased capacity to learn to be afraid of something
• Calder et al (1996)
  
  • failure to recognise fear from face photos (mean = 4/10 vs control = 8.6/10)
• Adolphs et al (1997)
  
  • decreased memory of emotional components of narrative (usually strongest)
• Hamann et al (1999)
  
  • better encoding of emotional info = more excitation in amygdala

Question 25

ventromedial prefrontal cortex damage and emotion

Answer 25

• Damasio et al (1990)
  
  • no increased SCR for emotional stimuli with social significance
• Koenigs et al (2007)
  
  • more likely to overcome an emotional response in moral dilemma - easy util.
• Anderson et al (2006)
  
  • increased irritability + frustration, decreased emotionality - blunted affect
• Bechara et al (1994) - EVR
  
  • normal reasoning ability but poor real world decision making

Question 26

Amygdala + vmPFC damage and decision making

• IGT

Answer 26

Bechara et al (1994) - Iowa Gambing Task

• A + B - good decks (in long run), C + D - bad decks (in long run)
• controls learn to avoid C+D, patients do not –> controls ‘feel’ risky decks

Bechara et al (1996) - IGT + SCR

• when punished SCR increases a lot, when reward it increases a bit
• controls - high SCR before turning cards (anticipatory SCR), patients didn’t

Bechara et al (1999) - SCR before + after card

• controls - SCR after card (loss+reward), SCR before card (higher for risky)
• amygdala –> not much SCR before OR after card
  
  • associates reward-punishment outcomes with stimulis
• vmPFC –> SCR after card, not much before
  
  • summons association at point of decision-making

Bechara et al (1997) - what are people conscious of?

• pre-hunch stage:
  
  • controls start generating SCR for bad decks (but not explicitly aware); patients don’t
• hunch stage
  
  • controls have sense that C+D = risky, no knowledge or SCR for patients
• conceptual period
  
  • controls could explicitly say what was going on (70% reached this), 50% patients did but still didn’t choose risky less often
• concluded
  
  • nonconscious biases guide behaviour in controls before conscious knowledge does
  • physiological arousal guides behaviour

explanation = somatic marker hypothesis

Question 27

Somatic marker hypothesis

Answer 27

situation –> elicits emotion –> amygdala learns association –> vmPFC retrieves association –> vmPFC activates automic nervous system –> causes physiological arousal –> has biasing effect on reasoning –> affects decision making

Question 28

Problems with somatic marker hypothesis

Answer 28

1. We may not need somatic cues

• Heims et al (2005) - patients with pure autonomic failure
  
  • physiology doesn’t change in respons to stress
  • perform better than controls on IGT
  • so lack of somatic signal doesn’t stop decision making

2. Somatic cues may not signal outcomes

• Tomb et al (2002) - in IGT, bad decks had higher variability (big loss outcomes)
  
  • if A+B are better AND have higher variability, the SCR increases more than for the bad decks
  • so maybe physiological response encodes the variability of outcomes (not goodness)
• REPLY: Damasio et al (2002)- in Tomb’s version the anticipatory SCR of good decks does indicate goodness
  
  • BUT then same signals for goodness AND badness

3. no need to posit unconscious knowledge

• Maia & McClelland (2004) - B (1997):
  
  • questions too vague, good/bad decks badly defined
  • explicit knowledge is better than expected (18/20 knew from q1)
  • behavioural data falls behind explicit knowledge
    
    • still explore other decks even if they know
  • they base knowledge on the cards thyey’ve experienced, not experimenters knowledge of good/bad decks

4. alternative explanations of patient data

• Fellows & Farah (2005) - vmPFC have problems with reversal learning
  
  • in IGT first few trials with no loss (bad decks)
  • vmPFC learn association between risky decks + winning - this is hard to reverse
  • so..shuffled trials (started from 9)
    
    • vmPFC performed much better
• Rolls et al (1994) - vmPFC patients struggle with reversal learning
• Dunn, Dalgleish & Lawrence (2006)
  
  • reversal learning could underpin poor IGT performance
  • vmPFC patients may not care enough about negative eoutcomes to actively avoid them (apathy –> see Barrash et al 2000)

Question 29

Intuitive Reasoning Task

• to replace IGT

Answer 29

Dunn et al (2010): choose deck, predict colour when turned over (50/50)

• decks are rigged - A+B have 60% chance of winning, C+D have 40% (not about experimenter knowledge or order)
• amount at stake same for A+C and B+D (so not about variability)
• addresses: reversal learning, deconfounds magnitude + goodness
• re-ran M+M’s 2004 conscious knowledge probing
  
  • found they were NOT consciously aware
    
    • supports somatic marker hypothesis
• results:
  
  • people choose better decks (A+B) more often
  • SCR increases for C+D prior to choice (anticipatory SCR)
  • HR decreases before choosing good decks (A+B)
  • physiological response strongly predicts patient data

Question 30

Why cost-benefit analysis is wrong

• behavioural economics
• heuristics
• emotion

Answer 30

Behavioural economics

• humans are fallible - don’t necessarily think rationally (Thaler & Sunstein, 2008)

Heuristic decision makinb

• in risky situations - use rules of thumb - instinctive (Tversky & Kahneman, 1974)

Emotional impulses

• overwhelm rational cognitions
• especially if short time-frame
• (Bensons & Sams, 2013)